Ethical Issues in the Design and Conduct of Rcts

Health Technology Assessment 1998; Vol. 2: No. 15 Review

Ethical issues in the design and conduct of randomised controlled trials SJL Edwards RJ Lilford DA Braunholtz JC Jackson J Hewison J Thornton

Health Technology Assessment NHS R&D HTA Programme HTA Standing Group on Health Technology Chair: Professor Sir Miles Irving, Professor of Surgery, University of Manchester, Hope Hospital, Salford †

Dr Sheila Adam, Professor Howard Glennester, Mr Hugh Ross, Department of Health Professor of Social Science & Chief Executive, The United Bristol Professor Martin Buxton, Administration, London School of Healthcare NHS Trust † Professor of Economics, Brunel University † Economics & Political Science Professor Ian Russell, Professor Angela Coulter, Mr John H James, Department of Health, Sciences & Director, King’s Fund, London Chief Executive, Kensington, Chelsea & Clinical Evaluation, University of York Westminster Health Authority Professor Anthony Culyer, Professor Trevor Sheldon, Deputy Vice-Chancellor, University of York Professor Richard Lilford, Director, NHS Centre for Reviews & † Dissemination, University of York † Dr Peter Doyle, Regional Director, R&D, West Midlands Executive Director, Zeneca Ltd, Professor Michael Maisey, Professor Mike Smith, ACOST Committee on Medical Research Professor of Radiological Sciences, Director, The Research School † & Health UMDS, London of Medicine, University of Leeds Professor John Farndon, Dr Jeremy Metters, Dr Charles Swan, Professor of Surgery, University of Bristol † Deputy Chief Medical Officer, Consultant Gastroenterologist, † North Staffordshire Royal Infirmary Professor Charles Florey, Department of Health Department of Epidemiology & Mrs Gloria Oates, Dr John Tripp, Public Health, Ninewells Hospital & Chief Executive, Oldham NHS Trust Department of Child Health, Royal Devon & Exeter Healthcare NHS Trust † Medical School, University of Dundee † Dr George Poste, Professor John Gabbay, Chief Science & Technology Officer, Professor Tom Walley, Director, Wessex Institute for Health SmithKline Beecham † Department of Pharmacological Therapeutics, University of Liverpool † Research & Development † Professor Michael Rawlins, Professor Sir John Grimley Evans, Wolfson Unit of Clinical Pharmacology, Dr Julie Woodin, Department of Geriatric Medicine, University of Newcastle-upon-Tyne Chief Executive, Nottingham Health Authority † Radcliffe Infirmary, Oxford † Professor Martin Roland, Dr Tony Hope, Professor of General Practice, The Medical School, University of Oxford † University of Manchester † Current members

HTA Commissioning Board Chair: Professor Charles Florey, Department of Epidemiology & Public Health, Ninewells Hospital & Medical School, University of Dundee †

Professor Ian Russell, Professor Sir Miles Irving Dr Tim Peters, Department of Health, Sciences & (Programme Director), Professor of Reader in Medical Statistics, Department of Clinical Evaluation, University of York * Surgery, University of Manchester, Social Medicine, University of Bristol † † Dr Doug Altman, Hope Hospital, Salford Professor David Sackett, Director of ICRF/NHS Centre for Professor Alison Kitson, Centre for Evidence Based Medicine, † Statistics in Medicine, Oxford Director, Royal College of Oxford Mr Peter Bower, Nursing Institute † Professor Martin Severs, Independent Health Advisor, Professor Martin Knapp, Professor in Elderly Health Care, Newcastle-upon-Tyne † † Director, Personal Social Services Portsmouth University Ms Christine Clark, Research Unit, London School of Dr David Spiegelhalter, Honorary Research Pharmacist, Economics & Political Science MRC Biostatistics Unit, Institute of † Hope Hospital, Salford Public Health, Cambridge Professor David Cohen, Dr Donna Lamping, Senior Lecturer, Department of Public Dr Ala Szczepura, Professor of Health Economics, Health, London School of Hygiene & Director, Centre for Health Services Studies, University of Glamorgan Tropical Medicine † University of Warwick † Mr Barrie Dowdeswell, Chief Executive, Royal Victoria Infirmary, Professor Theresa Marteau, Professor Graham Watt, Department of General Practice, Newcastle-upon-Tyne Director, Psychology & Genetics Woodside Health Centre, Glasgow † Professor Martin Eccles, Research Group, UMDS, London Professor of Clinical Effectiveness, Professor Alan Maynard, Professor David Williams, Department of Clinical Engineering, University of Newcastle-upon-Tyne † Professor of Economics, University of York † University of Liverpool Dr Mike Gill, Professor Sally McIntyre, Director of Public Health and Health Policy, Dr Mark Williams, MRC Medical Sociology Unit, Glasgow Brent & Harrow Health Authority † Public Health Physician, Bristol Dr Jenny Hewison, Professor Jon Nicholl, Dr Jeremy Wyatt, Senior Lecturer, Department of Psychology, Director, Medical Care Research Unit, Senior Fellow, Health and Public Policy, † University of Leeds † University of Sheffield School of Public Policy, University College, † Dr Michael Horlington, Professor Gillian Parker, London Head of Corporate Licensing, Smith & Nuffield Professor of Community Care, * Previous Chair Nephew Group Research Centre University of Leicester † † Current members HTA

How to obtain copies of this and other HTA Programme reports. An electronic version of this publication, in Adobe Acrobat format, is available for downloading free of charge for personal use from the HTA website (http://www.hta.ac.uk). A fully searchable CD-ROM is also available (see below). Printed copies of HTA monographs cost £20 each (post and packing free in the UK) to both public and private sector purchasers from our Despatch Agents. Non-UK purchasers will have to pay a small fee for post and packing. For European countries the cost is £2 per monograph and for the rest of the world £3 per monograph. You can order HTA monographs from our Despatch Agents: – fax (with credit card or official purchase order) – post (with credit card or official purchase order or cheque) – phone during office hours (credit card only). Additionally the HTA website allows you either to pay securely by credit card or to print out your order and then post or fax it.

Contact details are as follows: HTA Despatch Email: [email protected] c/o Direct Mail Works Ltd Tel: 02392 492 000 4 Oakwood Business Centre Fax: 02392 478 555 Downley, HAVANT PO9 2NP, UK Fax from outside the UK: +44 2392 478 555 NHS libraries can subscribe free of charge. Public libraries can subscribe at a very reduced cost of £100 for each volume (normally comprising 30–40 titles). The commercial subscription rate is £300 per volume. Please see our website for details. Subscriptions can only be purchased for the current or forthcoming volume.

Payment methods Paying by cheque If you pay by cheque, the cheque must be in pounds sterling, made payable to Direct Mail Works Ltd and drawn on a bank with a UK address. Paying by credit card The following cards are accepted by phone, fax, post or via the website ordering pages: Delta, Eurocard, Mastercard, Solo, Switch and Visa. We advise against sending credit card details in a plain email. Paying by official purchase order You can post or fax these, but they must be from public bodies (i.e. NHS or universities) within the UK. We cannot at present accept purchase orders from commercial companies or from outside the UK.

How do I get a copy of HTA on CD? Please use the form on the HTA website (www.hta.ac.uk/htacd.htm). Or contact Direct Mail Works (see contact details above) by email, post, fax or phone. HTA on CD is currently free of charge worldwide.

The website also provides information about the HTA Programme and lists the membership of the various committees.

Ethical issues in the design and conduct of randomised controlled trials

SJL Edwards1 RJ Lilford1 DA Braunholtz1 JC Jackson2 J Hewison3 J Thornton4

1 Department of Public Health and Medicine, University of Birmingham, UK 2 Department of Philosophy, University of Leeds, UK 3 Department of Psychology, University of Leeds, UK 4 Department of Epidemiology, University of Leeds, UK

Published December 1998

This report should be referenced as follows:

Edwards SJL, Lilford RJ, Braunholtz DA, Jackson JC, Hewison J,Thornton J. Ethical issues in the design and conduct of randomised controlled trials. Health Technol Assessment 1998; 2(15).

Health Technology Assessment is indexed in Index Medicus/MEDLINE and Excerpta Medica/ EMBASE. Copies of the Executive Summaries are available from the NCCHTA web site (see overleaf). NHS R&D HTA Programme

he overall aim of the NHS R&D Health Technology Assessment (HTA) programme is to T ensure that high-quality research information on the costs, effectiveness and broader impact of health technologies is produced in the most efficient way for those who use, manage and work in the NHS. Research is undertaken in those areas where the evidence will lead to the greatest benefits to patients, either through improved patient outcomes or the most efficient use of NHS resources. The Standing Group on Health Technology advises on national priorities for health technology assessment. Six advisory panels assist the Standing Group in identifying and prioritising projects. These priorities are then considered by the HTA Commissioning Board supported by the National Coordinating Centre for HTA (NCCHTA). This report is one of a series covering acute care, diagnostics and imaging, methodology, pharmaceuticals, population screening, and primary and community care. It was identified as a priority by the Methodology Panel and funded as project number 93/41/02. The views expressed in this publication are those of the authors and not necessarily those of the Standing Group, the Commissioning Board, the Panel members or the Department of Health. The editors wish to emphasise that funding and publication of this research by the NHS should not be taken as implicit support for the recommendations for policy contained herein. In particular, policy options in the area of screening will, in England, be considered by the National Screening Committee. This Committee, chaired by the Chief Medical Officer, will take into account the views expressed here, further available evidence and other relevant considerations. Reviews in Health Technology Assessment are termed ‘systematic’ when the account of the search, appraisal and synthesis methods (to minimise biases and random errors) would, in theory, permit the replication of the review by others.

Series Editors: Andrew Stevens, Ruairidh Milne and Ken Stein Editorial Assistant: Melanie Corris

The editors have tried to ensure the accuracy of this report but cannot accept responsibility for any errors or omissions. They would like to thank the referees for their constructive comments on the draft document.

ISSN 1366-5278

Enquiries relating to copyright should be addressed to the NCCHTA (see address given below).

Published by Core Research, Alton, on behalf of the NCCHTA. Printed on acid-free paper in the UK by The Basingstoke Press, Basingstoke.

Copies of this report can be obtained from:

The National Coordinating Centre for Health Technology Assessment, Mailpoint 728, Boldrewood, University of Southampton, Southampton, SO16 7PX, UK. Fax: +44 (0) 1703 595 639 Email: [email protected] http://www.soton.ac.uk/~hta Health Technology Assessment 1998; Vol. 2: No. 15

Contents

Glossary and list of abbreviations ...... i What are the physical effects of participating in trials? ...... 29 Executive summary ...... v What are the psychological effects of participating in trials? ...... 35 1 Introduction ...... 1 What is the ‘best’ method for obtaining 2 Research questions ...... 5 informed consent? ...... 36 Degrees of uncertainty and the ethics of What is the quality of informed consent conducting RCTs ...... 5 in practice? ...... 41 What are the physical effects of participating How do patients, the public and healthcare in trials? ...... 5 professionals view RCTs? ...... 44 What are the psychological effects of participating in trials? ...... 5 5 Recommendations ...... 53 What is the ‘best’ method for obtaining informed consent? ...... 5 6 Questions for future research ...... 55 What is the quality of informed consent in practice? ...... 5 Acknowledgements ...... 57 How do patients, the public and healthcare professionals view RCTs?...... 5 References ...... 59 3 Review methods ...... 7 Literature search strategy and retrieval ...... 7 Tables ...... 77 Ethics literature ...... 7 What are the physical effects of participating Appendix 1 Electronic search strategies ..... 123 in trials? ...... 8 What are the psychological effects of Appendix 2 Members of the External participating in trials? ...... 9 Advisory Group ...... 125 What is the ‘best’ method for obtaining informed consent? ...... 9 Appendix 3 Bayesian and frequentist What is the quality of informed consent statistics in clinical trials ...... 127 in practice? ...... 9 How do patients, the public and healthcare professionals view RCTs? ...... 10 Health Technology Assessment reports published to date ...... 129 4 Review results ...... 11 Degrees of uncertainty and the ethics Health Technology Assessment of RCTs ...... 11 panel membership ...... 131

Health Technology Assessment 1998; Vol. 2: No. 15

Glossary and list of abbreviations

Technical terms and abbreviations are used throughout this report. The meaning is usually clear from the context but a glossary is provided for the non-specialist reader. In some cases usage differs in the literature but the term has a constant meaning throughout this review.

Glossary Altruism Benevolent concern for the interests Clinical trial A prospective controlled study and welfare of other persons, and is typically involving patients, i.e. one in which an contrasted with egoism. The possibility of intervention is allocated and patients are genuine altruism has been debated – followed up. apparently altruistic individuals may Cluster trial An interventional design which themselves benefit. attempts to make inferences about individuals Autonomy A person’s capacity for self- but where the intervention is allocated to determination. Respect for autonomy is often clusters of individuals. contrasted with Paternalism. Crossover designs An interventional study in Bayesian statistics Based on the concept of which more than one treatment is allocated probability as subjective degree of belief and is sequentially to at least some patients, often with named after Reverend Thomas Bayes a ‘wash-out’ period between active treatments. (1702–1761), whose theorem is interpreted as Data monitoring committee A group of the probability of A given the data are people, typically independent from the proportional to the probability of A, multiplied trial, which decides if and when to stop the by the probability of the data, given A. trial early. Beneficence The moral principle of doing Decision analysis A formal mathematical good. One of the four principles in medical approach, based on (typically Bayesian) ethics proposed by Beauchamp and Childress. probability theory, to making decisions Blinding The term is used in trials to refer to under uncertainty. keeping individuals ignorant of the treatment Deontology Strictly, the study of ‘duty’ or allocation. Individuals include any combi- ‘obligation’ from the Greek ‘deon’ meaning nation of physicians, patients, observers and ‘must’. This may include theories according statisticians. A double-blind trial is one where to which the rightness of an action is not both the physician and the patient are blind. exclusively determined by the value of the consequences. Case–control study A type of observational analytic epidemiological investigation in which Duty Having a duty to do something subjects are selected on the basis of whether implies that one ought to do it. The converse they do (cases) or do not (controls) have a statement is not valid – there are things one particular disease under study. The groups are ought to do without having a duty to do them. then compared with respect to the proportion Kant distinguished between ‘perfect’ and having a history of an exposure or ‘imperfect’ duties. The former can be express- characteristic of interest. ed by the Categorical imperative. A perfect duty is absolute, and an imperfect duty is Categorical imperative An ought-statement desirable but not necessary. that is independent of any condition that a certain end is desired by the agent. According Epistemology Strictly, the study of knowledge. to Kant, an action has genuine moral worth Philosophers have focused was on what is it to only if the ought-statement that motivates the know something and knowledge has been agent is categorical. defined as justified true belief. continued i Glossary and list of abbreviations

continued Equipoise Various forms of equipoise have is that the ideal individual should be self- been proposed. Equipoise refers to a state of determining, making decisions independently. regarding two treatments as an equal bet in Induction Inference from a finite number prospect. Collective equipoise is where the of particular cases to a further case or profession at large is equally balanced, while generalisation. individual equipoise refers to the individual. Interim analysis Statistical analysis, performed Equivocation Semantic ambiguity of meaning before the trial has ended, the results of which where a word(s) sometimes means one thing are typically made available to a Data and, at others, something different. monitoring committee alone. Ethics The reflective inquiry into how people should think or behave with a view to Justice A property of a political system. formulating norms of conduct and evaluation Often called an ‘imperfect duty’. Justice can of character. Contrast with Morality. Applied be retributive or distributive. The latter aims ethics is the study of what standards are applied at allocating resources or tasks fairly amongst to actual situations. a given population. Expected utility The expected utility of an Liberalism A set of ideas in social and political action is the probability of an outcome times its thought which underlines the importance of value (utility), summed over all possible individual rights. The role of the state is outcomes. primarily to protect these rights. Feedback trial This is a trial in which any Medical research Systematic investigation to preliminary findings are publicly available. establish facts and reach conclusions. Contrast See also Open trial. with Medical practice. Frequentist statistics Based on a concept of Meta-analysis The statistical analysis of data probability that is objective (typically relative) from more than one study of the same frequency. Many of the methods employed are intervention (or association) in an attempt to based on hypothesis tests. summarise the current state of knowledge. Hazard ratio A ratio of two hazards. A hazard Morality Beliefs about what is good or bad, is an instantaneous risk. right or wrong. See also Ethics. Helsinki Declaration World Medical Association n of 1 trial A crossover trial involving only Declaration of Helsinki – recommendations one person where one or more treatments guiding doctors in biomedical research involving are evaluated for their effectiveness for the human subjects. These were adopted by the individual in question. The order of exposure 18th World Medical Assembly, Helsinki, Finland, might be random. 1964, and revised by the World Medical Non-maleficence The moral principle of not Assembly in Tokyo, Japan, 1975, in Venice, doing harm. One of the four principles in medi- Italy, in, 1983 and in Hong Kong in, 1989. cal ethics proposed by Beauchamp and Childress. Hippocratic Oath An oath sworn by Null hypothesis The proposition to be tested Hipprocrates of Cos who was an influential statistically, that the experimental intervention Greek physician of the fifth century BC. This has ‘no effect’. oath has formed the ethical basis of medical practice ever since. Nuremberg Code In 1947 an International tribunal declared the Nuremberg Code the Hypothetical imperative An ought-statement standard by which a group of doctors in Nazi which describes an action necessary to achieve Germany should be judged. a certain end that is desired by the agent. Such an imperative is thus conditional to the partic- Obligation Having an obligation to do ular end in question and is to be distinguished something implies that one ought to do it. from the Categorical imperative. Often used synonymously with Duty. Individual Anything regarded as a single unit, Odds ratio A ratio of two odds. Odds are the especially a person. One view of individualism ratio of a probability and its complement. ii Health Technology Assessment 1998; Vol. 2: No. 15

Open trial We use this to mean that the treat- A right can be to be free from interference or ment assignment is known by the physician allied to an external duty. and patient. Contrast with a blinded trial Relative risk Ratio of risks. The risk of (poor (see Blinding). outcome) given exposure over the risk of Paternalism Any situation in which decisions (poor outcome) given non-exposure. are taken on behalf of a competent person. Contrast with an autonomous choice Sequential trial A trial design in which the (see Autonomy). results are continuously monitored and the trial terminated according to a predefined Placebo A chemically inert substance which ‘stopping rule’. has a psychologically suggestive effect and is used in place of an active drug. Also used as a Social contract theory A moral theory which control in a clinical trial to determine whether emphasises the need for social cohesion and improvement and/or side-effects can be preservation. Individuals ‘agree’ to relinquish attributed to the active substance. some of their natural liberty in a civil society for the sake of collective advantage. Pluralism A philosophical point of view that an inquiry into morality should draw on more Society An organised collection of people with than one theory, as no single theory can fully shared values. explain it. Trust Confidence in the reliability of Posterior probability distribution What a another. The primary quality of the rational Bayesian observer should believe, doctor–patient relationship. given his/her prior beliefs, the data and Uncertainty In common parlance, uncertainty the model. is simply the opposite of certainty. Compare Prior probability distribution What an observer with Equipoise. believes prior to seeing additional data. Utilitarianism A theory of morality which Probabilistic dominance This situation arises assigns moral significance only to the conse- in decision analysis where there are only two quences of action. An action is right if it outcomes, e.g. life or death. produces good consequences (typically this is happiness for the greatest number). Some Probability The meta-physics of chance or the utilitarians include rules-of-thumb to get state of being probable. There are two main round problems the problem of being able theories of probability: frequentist and Bayesian. to say with certainty what the right course Protocol A formal and explicit treatment of action is in advance. regimen for care. A clinical trial will have two Virtue ethics The moral theory that or more such protocols along with a descrip- emphasises a person’s character, rather than tion of the research design or method, eligi- actions or guiding principles. A virtue is often bility requirements and the proposed method seen as the ‘golden mean’ between the vices of analysis. It can also be used to describe the of excess and deficiency. comprehensive written document detailing all procedures to be followed in a trial, though we Zelen’s design A randomised controlled trial do not use it in this sense. design in which a novel treatment is compared with standard treatment. Randomisation Randomised controlled trial An intervention occurs, and then consent is elicited from study in which treatments are allocated to participants. Typically, consent is only sought patients on a random basis, thereby avoiding from the study, not control, group in order to any uneven distribution of patients with bolster recruitment or avoid distress at not known, or unknown, risk factors between receiving a preferred therapy. treatment arms The glossary was compiled with reference to the Right A moral or legal power belonging to Penguin Dictionary of Philosophy (Mautner, 1997) an individual or group. Although it may be and Penguin Dictionary of Psychology (Reber, 1985), permissible for a person to exercise a right, and Epidemiology in Medicine by Hennekens and it may not be the morally ‘right’ thing to do. Buring (1987). continued iii Glossary and list of abbreviations

List of abbreviations ACAS Asymptomatic Carotid HRT hormone replacement therapy Atherosclerosis Study* IDU injection drug user* ALL acute lymphoblastic leukaemia IRB Institutional Review Board AMIS Aspirin Myocardial Infarction Study* * ISIS-2 Second International Study of ARC AIDS-related complex Infarct Survival BDI Beck Depression Inventory* LATE Late Assessment of Thrombolytic BHAT Beta-blocker Heart Attack Trial* Efficacy* BIDS Bath Information Data Services MRC Medical Research Council CAPRIE Clopidogrel versus Aspirin in * NHMRC National Health and Medical Patients at Risk for Ischemic Events Research Council, Canberra* CCSG Children’s Cancer Study Group* OR odds ratio CCT controlled clinical trial* PEFR peak expiratory flow rate* CI credible interval RCT randomised controlled trial COMS Canadian collaborative Ocular Melanoma Study* RR risk ratio CSS Cancer Surveillance System* SCI Science Citation Indices * DFS disease-free survival SOLVD Studies of Left Ventricular * DMC data monitoring committee Dysfunction ECMO extra-corporeal membrane SSCI Social Science Citation Indices oxygenation STAI State–Trait Anxiety Inventory FDA (US) Food and Drugs Administration TPN total parenteral nutrition* GUSTO Global Utilisation of Streptokinase and Tissue Plasminogen Activator UKALL UK Acute Lymphoblastic * for Occluded Coronary Arteries* Leukaemia HMRI Hospital Medical Records Institute* HR hazard ratio * Used only in tables

iv Health Technology Assessment 1998; Vol. 2: No. 15

Executive summary

Objectives in trials, and look to Kantian ethics for support. Kantians would object, however, if the investigators • To review ethical arguments put forward in use patients merely as the means to societal ends the literature which bear on randomised and, given voluntary consent, this is not the case. controlled trials (RCTs), focusing particularly At any rate, patients are not required to sacrifice on uncertainty as an underpinning issue. themselves (whether voluntarily or not) for the • To review empirical data (from comparative, benefit of society if we endorse the uncertainty observational and qualitative studies) which may principle, or, less ambiguously, equipoise, whereby be relevant to the ethics of conducting trials. each (or all) comparator treatments are an ‘equal bet’ in prospect. When equipoise applies, patients do not lose out prospectively, in order to benefit Methods others. Given equipoise, a trial should be acceptable to both utilitarians and Kantians, and A review of the literature was conducted. hence ethical in its use of patients. If known or The aims were achieved by completing the potential side-effects of the comparator treatments following tasks: are unequal, then a trial should be acceptable to both parties provided the expected utilities are • Ethics: equivalent, that is, equipoise is only a ‘null’ prior – development of an intellectual framework to belief if there are no trade-offs to be made. structure the ethics literature Although there are possibly valid objections to – creation of a database containing references the use of RCTs in particular disciplines or cases relating to the ethics of conducting clinical (e.g. if the offer of trial entry will make the patient trials, including any methodological papers very upset), such arguments do not make the which have implications for medical ethics RCT necessarily unethical. Further argument – classification of the articles, according to the concerns the idea of uncertainty as a moral basis type of information contained within (e.g. for trials as well as the significance of different, ‘under-powered’ trials versus replication less ambiguous, constructs, i.e. collective versus of trials) individual equipoise. – summary of the ethical arguments and commentary on those arguments. The empirical evidence on the whole was seen to • Empirical data: support the view that RCTs are justified in clinical – classification of studies according to topic practice, contingent, however, on the existence of and study design patient equipoise (informed consent). Indeed, – creation of a database containing trials themselves may have a beneficial effect on empirical studies relevant to the ethics patients’ outcome both in terms of physical of conducting trials prognosis (at least, when an effective treatment is – abstraction and quality assessment of already available) and psychological experience, relevant empirical data perhaps due to increased levels of care that are – summary of the primary data. unintended. Any such benefit should be incidental to routine care and not used as an incentive to increase recruitment rate, lest the principle laid Results down in the Helsinki Declaration (that non- participation in a study will not intentionally RCTs affect the standard of care) be violated. The main reason for using the RCT design is a scientific one: society is likely to suffer as a direct Informed consent result of avoiding such high quality evidence. It is evident that patients are unlikely to understand The most widely cited assaults on the RCT are all the information which is given to them by what- claims that patients necessarily sacrifice themselves ever means during consent consultations. Patients for the benefit of future patients by participating have particular problems grasping abstract, as v Executive summary

opposed to concrete, information. Consequently, moral basis for trials. It should not be used to fully informed consent for all patients at least is disguise such existing data as may affect patient an unobtainable ideal. There are three possible preferences, even when such data are responses to this problem: (1) declare all trials insufficient to engender ‘certainty’. unethical, unless the participants are themselves • Patients must be given as much information medical experts (or as expert as anyone else in the as they need to bring their values into play. relevant field), (2) abandon the requirement of • Patients are least alarmed and understand the informed consent, but rely on other safeguards issues most clearly when they have encountered such as ethics committees to protect individual the concept of comparative trials before. participants, or (3) retain the spirit of informed • Practitioners should pay particular attention consent, taking all practical measures to maximise to explaining abstract ideas (especially that patient understanding, but still rely on ethics of randomisation). committees as a further level of protection. The • Small trials of existing therapies are not authors favour the third option. While difficulties necessarily unethical provided that they with communication are regrettable and should be are in equipoise. reduced as much as possible, some failure would • Clinical trials should start early in the life appear inevitable. The need to advance medical of a new treatment. understanding is important, but communication • The idea that patients in trials do better than difficulties remain ethically critical if a patient average, even when the trial produces a negative decides to participate on the basis that he/she will result, may be true. If the effect is real, it would benefit therapeutically in a way incommensurate seem to come from enhanced attention to detail with clinical prior belief. It is therefore important inherent in following the trial protocol for both that the patient understands that equipoise exists, control and experimental groups. It should not, and so has realistic expectations. however, be used as an inducement to accept randomisation since the Helsinki accord requires that the intention should be to Conclusions provide the ‘best’ care for all patients.

• The caring professions must articulate clear, ethical justification for trials if public confidence Recommendations for research is to be retained. • Patients should not lose out in prospect by Areas in the ethics of RCTs which need to be taking part in a trial. further analysed, include: • Given treatments which are generally available, patients do not lose out in prospect when prior • ethical issues in the design and conduct of estimates of effectiveness and values interact to cluster trials produce equal expected utilities. • ethical issues in interim analysis • When treatments are not generally available, • the conduct and constitution of ethics patients do not lose out by participating in trials committees. when the expected utility of the new treatment is at least as high as that of standard treatment. There are a number of empirical questions which • The term ‘uncertainty’ prevaricates on prior also need to be addressed, and these are detailed in probabilities and values, making it an inadequate the main report.

vi Health Technology Assessment 1998; Vol. 2: No. 15

Chapter 1 Introduction

uring this century, the medical profession has (well-designed) research is justified, and this report D acquired a growing capacity to distinguish is concerned with the need to advance medical between what they know and what they do not, knowledge while protecting patients in research thereby changing the professional emphasis from from harm at the same time (and protecting a demanding public trust without sound epistemo- patient’s rights). Many ideas have been articulated logical basis to an ‘evidence-based’ approach. and rehearsed, but there is still considerable This is typically contrasted with practice based on diversity of opinion over what constitutes justifi- intuition, though the two approaches may lead to cation for trials. If the necessary and sufficient the same ‘answer’ of course; it is just that scientific moral conditions for conducting RCTs are clearly knowledge provides explicit justification for belief. set out, then patient and public confidence would Clinicians can therefore talk more frankly about be reinforced, and clinicians would be encouraged their uncertainty and the randomised controlled to offer participation in trials as an option in health trial (RCT) has played an enormous role in pro- care. To address this debate, we have conducted a ducing valid data upon which practice can rely. The review of the topic, and, in so doing, we have clinical trial is an experiment which most closely identified a key underpinning issue: whether resembles that in the natural sciences, since it is uncertainty over relative treatment effectiveness based on the idea of an intervention, rather than is sufficient justification for randomisation. A simple observation. Carefully designed, conducted theoretical framework based around this issue was and analysed trials typically provide the strongest superimposed on the literature in order to make epistemological clues to the existence and nature of sense of the various arguments and assertions a cause–effect relationship. Observational studies contained within. The arguments that we lift from may produce false clues due to various forms of bias the literature, and our discussion of these, relate (mostly confounders), and/or chance. The RCT has to utilitarian and Kantian traditions in moral the unique advantage of using randomisation as a philosophy predominately. There are variations of method of determining patient allocation to treat- both utilitarian and Kantian theories. For present ment, which eliminates selection bias if correctly purposes, let us understand the utilitarians to hold executed. Any imbalance between trial arms with that one’s ultimate duty is to maximise utility by respect to other variables (that might confound the producing happiness of the greatest number of analysis) can only then arise by chance. Bias in the people – all other duties being derived from this. observation of outcomes or execution of therapy can Let us understand Kantians to hold that one should be minimised by the use of blind or double-blind always treat people with respect – never treating procedures. Having eliminated bias, only chance them merely as the means to other people’s ends. (random error) remains a source of false clues to The putative tension between the individual and the ‘true’ effect of a given intervention; this risk can society is brought to the fore because both be reduced by increasing sample size. Drawing utilitarians and Kantians see the need to balance general conclusions from trial populations to ones duties to, and rights of, the individual and society. with different characteristics inevitably involves extrapolating beyond the data. To provide reassur- There are other traditions in philosophy besides ance that the data are externally valid, the trial utilitarian and Kantian. Recently, there has been investigator should describe the patients entered, a revival of interest in the virtues. But the focus so that practitioners can make a judgement about on virtues concerns most directly character rather the applicability of results to particular patients than principles or rules. Although character is (or groups of people). broadly supportive of apt choices, rules and principles are still needed to determine aptness, The main reason for using the RCT design is as one cannot simply read the relevant rules from therefore a scientific one: properly conducted the virtues. Social contract theories provide RCTs produce valid data from which society another alternative to utilitarian and Kantian can benefit. Research enables advances in accounts of duty. Contractarians explain general medical practice, and hence the alleviation of obligations in terms of our shared needs – our suffering. However, not all scientifically sound need to live in peace and the rules we need to 1 Introduction

live by if we are to live in peace. However, both (1) there is no preference between treatments utilitarians and Kantians may incorporate social since they have identical side-effects and A is as contract theory into their own. In any case, which- likely to be better than B and vice versa, or (2) ever theory one draws on, the basic issue that there is some preference in the sense that one concerns us here is how to handle competing of the treatments is more likely to be superior, claims of the participants involved in trials on the but this is uncertain; treatment A has not been one hand and of society at large on the other: what ‘proven’ to be more effective than B. Indeed, trade-offs are ethically defensible and why? Gifford (1986) observes that most arguments which use uncertainty as justification for RCTs Clinical research and RCTs in particular have often equivocate on the term ‘know’, sometimes mean- been accused of resting exclusively on utilitarian ing one thing and sometimes something else. This grounds and of using people as ‘guinea-pigs’ by is not simply for want of definition as uncertainty sacrificing the individual patient for the good of has a precise, yet ambiguous, meaning in common future patients: this represents a ‘lose/win’ situation parlance, that is, the opposite of certainty. If there for participants and future patients, respectively. is ‘absolute’ uncertainty, the decision-makers are However, since the formulation of the Hippocratic ‘agnostic’ or in ‘equipoise’, that is, the mean Oath, the medical profession has embraced the benefits in prospect from both treatments are notion of physician–patient trust as the primary equal, then the prospects for the patient are the moral requirement for medical practice and has same, whether or not a trial is conducted (here sought to do its ‘best’ for each individual patient. there is a separate argument about whether doctors The importance of trust in the doctor–patient can be absolutely uncertain). For those familiar relationship can be derived from either Kantian or with Bayesian thinking, this situation would arise utilitarian philosophies, and is widely discussed in where the median of a symmetrical prior distri- the context of rationing of scarce resources or bution was a null result given no side-effects. situations where a doctor may have a conflict of interests (e.g. clinical research). Under the Hippo- We also need to clarify the meaning of ‘best’. Two cratic Oath, every patient is entitled to receive the situations apply: firstly, suppose treatments A and B ‘best’ quality treatment available, and should not (two treatments are discussed here for conveni- stand to lose so that others might gain, that is, ence) have equal side-effects in prospect, and the patients may not be used merely as the means to major outcomes are binary (e.g. live or die). In this an end, however admirable that end might seem. situation, A is preferred to B, provided it is more In some circumstances, a doctor may have to put effective (a situation we refer to as ‘probabilistic society or other patients before the current patient – dominance’ because values do not enter into the for example, a doctor may be called away from a equation). Equipoise exists if the expected patient to attend an emergency (to focus on the effectiveness of A and B are equal. case which is most in need of immediate medical attention) or may deliberately avoid using an However, we must acknowledge that treatments antibiotic which the profession wishes to reserve might have unequal side-effects a priori. A, for for the sickest cases, so as not to build up resistant example, might be mutilating surgery, while B is strains. However, these are cases where a scarce a smaller operation. In order for the treatments resource is involved so treatments given to different to be equally desirable for the patient (i.e. for patients are not independent. It is a matter of more equipoise to exist), A must be better than B in serious debate whether a doctor can morally prospect to the extent that the difference in withhold access to an otherwise freely available effectiveness will be traded-off against its greater treatment in order to encourage compliance in a side-effects. In decision analytic language, the trial thereby improving future care. If this is held to expected utilities of A and B should be the same. be unacceptable, it may still be morally possible to This could be called the point of effective equi- conduct a clinical trial if the physician does not poise to indicate that equipoise occurs when ‘know’ which treatment is ‘best’ – this (possibly) something other than a null result (on the main represents a no-lose/win situation for participants outcome measure) is perceived as the most likely and future patients, respectively. However, many result in prospect. If the expected difference in authors point out that there are problems here are outcome equals, but does not exceed, the trade- with the words ‘know’ and ‘best’. off that an individual patient would require, then the patient is in individual effective equipoise. Dealing first with ‘knowing’. Consider first two Potential participants are not disadvantaged if treatments (A and B) with equivalent (or no) equipoise applies (unless a trial has intrinsic 2 side-effects. ‘Knowing’ can have two meanings: negative side-effects as a result of psychological Health Technology Assessment 1998; Vol. 2: No. 15

distress associated with the offer of randomisation). Although we have homed in on the ethics of Since values are personal, equipoise is primarily a conducting RCTs, it must be made clear that this property of the patient. It is therefore not surpris- is not a self-contained topic within ethics, and we ing that many authors argue for non-paternalistic must not lose sight of the broader context. consultations that allow patients to bring their Methodological issues have a bearing on ethics, values into play. This can only take place through and not only are there different designs from detailed discussion with the patients themselves. which researchers might choose but there are In this way, doctors can respect a patient’s also different phases of evaluation which culminate autonomy in the spirit of Kantian ethics, do their in the Phase 3 RCT. Commonly, Phases 1 and 2 best for a patient, and still recruit those patients of the evaluation of a new technology are designed who are equipoised into the study. All this assumes to determine toxicity and dosage, and are not that the patient is competent and there is an randomised. However, some have argued for individual decision to be had. We discuss later randomising from the ‘first’ patient, perhaps most special situations where this is not the case, fervently Thomas Chalmers, so that the process of such as trials involving young children. randomisation gets off the ground while equipoise is still common, subsuming Phases 1 and 2 as a Much has been written of collective equipoise result. Additionally, there are alternative study which is said to occur when the scientific com- designs in analytic medicine or, more specifically, munity as a whole is split over treatment prefer- health technology assessment (designs that seek ence. This is an interesting point to which we to establish a cause–effect relationship between return in the discussion section. exposure to a technology and outcome), but detailed discussion of these falls outside our We also consider the situation that might arise current remit, except for deciding when such when a new and unusual treatment that is not alternatives are morally desirable. Furthermore, widely available is compared with an existing even though formal medical experimentation sits treatment. We discuss the argument that equipoise alongside a tradition of routine practice, the two between such treatments is not required, since a processes are interrelated and sometimes even patient’s best interests are not sacrificed by trial converge, as in the case of n = 1 trials. Even more entry, provided the new treatment is perceived broadly, ethical issues in routine practice are a priori as being at least as good as the standard. relevant to those in RCTs and medical experi- However, there is an interesting issue of whether it mentation generally, for example the doctor– is necessary to inform control patients that they are patient relationship in routine practice impacts on involved in an experiment – we present arguments the ethics of conducting trials. on both sides of this debate. The bibliography resulting from our literature We approach the ethics of conducting RCTs search provides a resource for researchers and from the perspectives of ethics committees, of the philosophers working in the field of RCTs, and the clinicians who must decide whether or not to offer review itself provides a concise overview of issues trial entry to their patients (where individual prior and arguments. We review the philosophical and patient equipoise must be considered), and material in the style of the late medieval academic of data monitoring committees (DMCs) who must philosophers who used what was called the quaestio decide whether to stop a trial earlier than planned. format for discussing and debating issues. This Since further work needs to be done on combining format requires an author to review or document statistical and ethical concepts in the context of a sometimes long series of arguments on both interim analysis, we introduce the last of these sides of a question before providing his/her own perspectives, those of DMCs, with a view to opinion in reply. We have provided a glossary of pursuing it in more detail later. terms at the start of this review.

Health Technology Assessment 1998; Vol. 2: No. 15

Chapter 2 Research questions

Degrees of uncertainty and What is the ‘best’ method for the ethics of conducting obtaining informed consent? RCTs It would be nice to think that comprehension, The subject of the review was the ethics of emotional well-being, and recruitment rates are conducting RCTs and, as the review unfolded, positively correlated in relation to informed the primary focus became the issue of uncer- consent. However, a liberal patient-centred tainty; we sought to understand how the con- perspective would put the need to maximise cept of uncertainty, and degrees of uncertainty patients’ understanding ahead of maximising in particular, affect the ethics of conducting recruitment, should better informed patients RCTs. In so doing, we reviewed arguments for exhibit a higher refusal rate. In order to answer and against RCTs in general, and in particular the empirical question, we sought primary data disciplines or cases, adding authorial comment from comparative studies of different procedures where appropriate. for offering entry into trials on understanding, anxiety, attitudes and recruitment rates. What are the physical effects of participating in trials? What is the quality of informed consent in practice? We sought primary data on how participating in trials might affect a patient’s physical out- It order to answer this question, we sought primary come. In particular, we sought to identify data on how much patients, who were offered entry whether patients could benefit from partici- into real trials, understood of the various aspects of pating in trials, even in the absence of notable the trial process. In effect, this could be seen as an treatment effects. audit of communication with patients. Differences in levels of understanding, according to trial and type of information overlap with the previous question. What are the psychological effects of participating in trials? How do patients, the public and healthcare professionals view RCTs? Again, we sought primary data on how trials per se might affect outcome, but this time we It is important to identify how the RCT is were interested in psychological effects. If perceived by patients who participate in them, patients are distressed by the experience of the public in general, and healthcare professionals being offered trial entry and subsequent who might enrol their patients into such studies. participation, then we need to understand If clinicians are reluctant to recruit patients, then what it is about RCTs which is frightening, and a trial cannot get off the ground, however good its consider options to remedy this. On the other design. If clinicians are keen to become involved hand, if patients are happy to participate and in research, then we need to know whether the thrive on the level of care offered by involve- patients themselves view trials as important and for ment in trials, then we can be more comfortable what reasons. In particular, we sought data on what within current practice. motivates patients to participate in clinical trials.

Health Technology Assessment 1998; Vol. 2: No. 15

Chapter 3 Review methods

Literature search strategy best to approach patients, simply made reference and retrieval to ethics and so were not recorded twice. We carried out a search of the literature relating The empirical articles were then classified to the ethics of conducting clinical trials, including according to the research question posed and study empirical studies which might inform such discus- design, while the ethics literature was managed sion. We used five different methods to ascertain with the help of a well-structured framework the relevant articles: an electronic literature survey (see below). Some topics were abandoned due to (conducted on the three databases, MEDLINE, overlap with other projects commissioned by the PsycLIT and Bath Information Data Services HTA methodology programme, for example the (BIDS) Science and Social Science Citation ethical issue of informed consent was discussed in Indices), hand searches, personal contacts, an detail by the Liverpool project (93/41/04), and original collection of the second author, and empirical data on recruitment rates to trials were ‘explosion’ of articles obtained by the above sought by the Edinburgh project (93/43/02). methods. The electronic surveys were performed with in-house algorithms, and were limited to the Some studies covered more than one empirical English language (see appendix 1). In this way, topic, for example Aaronson et al. (1996) not a total yield of 812 references was achieved. only evaluated consent formats used in trials but examined participants’ understanding of their trial The electronic survey provided a yield of 183 such (audit of communication). Likewise, some ethics articles (see appendix 1). The MEDLINE search articles dealt with more than one issue, for example was repeated seven months into the project as the Collins et al. (1992) addressed both uncertainty database is reindexed on an annual basis. We identi- as justification for an RCT and under-powered fied another 20 relevant articles in this way. It became studies. Such articles were recorded under all the the habit of the investigator to make searches of jour- relevant topic headings on the database of articles. nals (volumes ascertained as above) when working in the library, and a further 32 articles of interest were A collaborative group was set up to discuss the found. Also, 88 additional articles (seven were pub- methods and progress of the review at regular lished in a foreign language) were ascertained intervals throughout the project (a list of group through a search of the database at Edinburgh (HTA members is given in appendix 2). The quality of the project No. 93/43/02), and 59 though other contacts articles was assessed independently by two authors in the field. This was a useful source of unpublished as indicated below. Any disparity in our findings material (n = 8). One hundred and twenty-nine was resolved by group discussion. articles had been accumulated by the second author over the years, and a further 328 articles were ascertained by ‘exploding’ the articles, that is, by examin- Ethics literature ing the references used by the authors of articles ascertained by one or more of the above methods. A total of 636 articles contained ethical issues relating to RCTs directly. Our next task was to narrow Our next task was to create a database of articles down, further subdivide, and record articles on our which contained references, classified according database, according to our intellectual framework. to type of information, that is, ethics versus empir- Four hundred and forty-nine articles addressed ical data. There were 176 articles which contained degrees of uncertainty and the ethics of conducting data, and 636 which dealt with the ethics of con- RCTs. The arguments in these articles related to ducting clinical trials directly. Four articles contain- how degrees of uncertainty affect the obligations of ed both, and were recorded under both headings healthcare professionals to society as a whole and to on the database (Alderson, 1996; Blum et al., 1987; the individual patients under the ambit of their care. Johnson et al., 1991; Lilford, 1994). Many of the The former discussion brings in considerations such empirical articles, which were concerned with the as statistical power, replication of results and the quality of informed consent in practice and/or how need for research to inform practice. The latter 7 Review methods

includes detailed analysis of the obligations of published in The Lancet (Edwards et al., 1997). Work doctors to patients, degrees of knowing (including on the ethics of conducting cluster trials and issues Bayesian conceptions of probability), notions of surrounding early stopping is in progress. values (and hence decision analysis), and related issues, such as use of placebos. It was necessary to Articles published between 1994 and 1996 were commission translation of one of the studies reviewed independently by SE and JJ so that the (Gracia, 1993) from the original Spanish. We degree of agreement over the articles’ content and were able to get a translation of a German article quality could be documented. Articles preceding from the authors (Ernst and Resch, 1995). this date were reviewed by the first author only.

We identify issues of current ethical concern and offer a summary of arguments. Although we report What are the physical effects of all references, it is not necessarily possible to derive participating in trials? a consensus view, nor should readers imbue the number of references behind a particular argument There were 17 articles which examined the effects as a mark of its validity. Most of the arguments do of participating in clinical trials on the patient’s not take the form of syllogisms; some simply make physical condition. One of the articles was a review bold assertions while others are based on analogy, in its own right which looked at the more general such that it would be difficult to perform a check issue of the use of treatment protocols (both within for internal consistency (which is a key feature of and outside clinical trials) and treatment centres on argumentative quality). Besides, internal consistency the outcome of care (Stiller, 1994). Another article of itself does not guarantee that the conclusion is was primarily methodological and did not report true, since it may be built on false premises. As a sufficient data for analysis of a possible trial effect result, a critique of the arguments is limited as far as (Olschewski et al., 1992). Thus, we are left with 15 possible so as to give a concise description of what is articles containing analysable primary data concern- out there, thanks to our intellectual framework, the ing the effects of being in a trial on mortality and/or derivation of which is discussed in the Introduction. physical morbidity (Tables 1 and 2)*. Eight of the More in-depth philosophical analyses of papers are 15 studies are included in Stiller’s review. being prepared for specific topics in light of our own appreciation of the subject (see the discussion of Included studies compared randomised patients review results, page 11). Indeed, the first of these in trials with at least one non-trial control group deals with the ethics of conducting ‘under-powered’ (Figure 1). The control groups were either patients studies and is included since it is complete and is who were offered entry in a trial but declined or

Condition X

Not offered trial entry Offered trial entry

Non-trial controls Non-trial controls Declined Accepted (concurrent) (historial) Non-trial controls Trial patient (refusers)

Randomised Randomised (experimental) (trial control)

FIGURE 1 Generation of controls.The figure illustrates how the control groups for studying a possible trial effect were generated by different authors.The non-trial controls were either patients who were offered entry in a trial but declined or patients who had the same medical condition as the trial patients, but who had not been offered trial entry.The latter patients were either treated concurrently with the trial patients or they constituted historical controls. Of those who accepted trial entry, patients were assigned to the trial control or experimental arm on a random basis and, in some cases, the trial and non-trial controls were compared

8 * For clarity, the tables are collected together at the end of this report (see page 77) Health Technology Assessment 1998; Vol. 2: No. 15

patients who had the same medical condition as the studies were found which provided comparative trial patients, but who had not been offered trial data on the effect of trial participation (Table 5). entry. The latter were either treated concurrently with the trial patients or they were historical con- Studies were included if they sought an effect trols. In each case, we noted whether the author of a trial on patients’ psychological state using had compared the trial and non-trial control concurrent controls and repeated measures groups for disparity in the incidence of prognostic (see Figure 2). variables and, if so, noted whether a statistical adjustment was made to allow for any difference Quality assessment was carried out by SE according found. We also noted the results of any comparison to an agreed checklist (Table 6). of outcome across groups within the RCT. In the case of an experimental and control treatment, this allowed us to compare outcomes across trial and What is the ‘best’ method for non-trial controls; if trials per se have a therapeutic obtaining informed consent? effect, then an improvement in outcome among trial patients (compared to non-trial controls) We reviewed comparative studies of different should not be confined to those in the methods for obtaining informed consent (Tables 7a ‘experimental’ arm (Figure 2). and 7b). There were 14 such studies, and all except three (Dal-re, 1991; Levene et al., 1996; McLean, 1980) were RCTs; that is, they were RCTs of the effects of different methods of offering entry to trials. It was necessary to commission translation AB C of one of the studies (Dal-re, 1991) from the original Spanish.

The outcome measures varied from study to study, Enrolment but included any one or number of the following: recruitment rates, attitudes to trials, anxiety and understanding. Our search also uncovered three reviews of empirical literature (not included in the DE F above figures) which included RCTs of different consent methods to clinical research in general (but including clinical trials), namely King (1986), Meisel and Roth (1983), and Silva and Sorrell (1988). However, the search strategies of these Trial study Trial control articles were not given, and the yield was much lower than that which we obtained (the most recent FIGURE 2 Example study design to determine the physical and review by Silva and Sorrell (1988) reviewed only psychological effects of trial participation.Tests should be taken at one relevant study, and King (1986) reviewed only time points A–F. The baseline result can be represented by A + B two); we reviewed the cited studies ourselves. versus C, and a positive effect of participation could be represented by D + E versus F or E versus F Quality assessment was carried out by SE, and a random sample (one in two) was reassessed inde- Quality assessment of all the studies was carried pendently by JT. Studies were assessed according to out by SE and independently by JT using an agreed a quality of data checklist (Table 8). Any disparity in quality of data checklist (Table 3). Data sets from our findings was resolved by group discussion. these studies were then categorised (Table 4a) and listed (Table 4b) by DB with a view to plotting the hazard ratios (HRs) for individual trials and strata What is the quality of informed within studies. consent in practice? Twenty-four studies examined participants’ know- What are the psychological effects ledge and understanding of their trial (audit of of participating in trials? communication), of which one used qualitative study methods (Snowdon et al., 1997). Our search We review studies which examine the psychological also uncovered three reviews of empirical literature effect of participating in clinical trials. Only three (not included in the above figures) which included 9 Review methods

knowledge and understanding of clinical trials, views on clinical trials (King, 1986; Meisel and namely King (1986), Meisel and Roth (1983), and Roth, 1983; Schain, 1994), but only one was a Silva and Sorrell (1988). However, the search systematic review which gave only limited inform- strategies of these articles were not given and the ation about the quality of evidence of the compo- yield was much lower than that which we obtained site studies (Meisel and Roth, 1983). In all three (the most recent review by Silva and Sorrell (1988) cases, the yield was much lower than we obtained, reviewed only one relevant study, and King (1986) even in the systematic review (only three studies reviewed only one). We reviewed the cited studies reviewed by Meisel and Roth (1983) were relevant), ourselves. It was necessary to commission translation and the most recent review (only six studies of two studies from the original Dutch (Oddens reviewed by Schain (1994) were relevant). It was et al., 1992) and French (Gallet et al., 1994). necessary to commission the translation of one study (Dal-re, 1993) from the original Spanish. Studies were included in the review if their design consisted of questionnaires/interviews to assess In summary, studies were included if they used patient comprehension of real trials, or questionnaires and/or interviews to elicit the views transcriptions of consent discussions. of the public, potential participants, participants and physicians on ‘clinical trials’. Sometimes the terms Quality assessment was carried out by SE, and a ‘clinical trial’ or ‘randomised controlled trial’ were random sample of papers (one in four) was assess- explicitly included in supplied questionnaires and ed independently by JH. Studies were assessed sometimes not, in which case we had to rely on what according to a quality of data checklist (Table 9) the investigators themselves reported. For example, Any disparity in our findings was resolved by misunderstanding of ‘clinical trials’ during a pilot discussion with RJL. study led to the investigators interchanging this with the more general term ‘research’ in the main study (Gerard et al., 1995). Since ‘research’ and How do patients, the public ‘clinical trial’ are not synonymous, this may lead to and healthcare professionals difficulties comparing the results with other studies view RCTs? which did use the words ‘clinical trial’. However, it is difficult to know how serious this is, methodo- There were 58 studies of views of patients, the logically speaking, as, in many cases, what respon- public and healthcare professionals on clinical dents understood by ‘clinical trial’ was not trials. Eight of these were qualitative studies, documented anyway (Table 10). which are not included in the results tables although we do summarise their main findings Abstraction of results and quality assessment for (Elbourne, 1987; Garcia, 1987; Madden, 1994; all studies were carried out independently by SE Marsden, et al., unpublished; Roberson, 1994; and JH. Quality assessment was difficult, in many Snowdon et al., 1997; Twomey, 1994; Wynne, 1989). cases, because insufficient information was given There were 51 quantitative studies (which provided by the authors. The dimensions of methodological data on the frequency with which particular views rigor used are listed in Table 11; we had originally were expressed) which were grouped into studies planned a more comprehensive review of methodo- of patients and the public (in both real and logical issues, but found that some items had to hypothetical clinical trial scenarios, n = 34), and be dropped as they were seldom mentioned by those of investigators and physicians (n = 21). One authors, while others failed to gain acceptance study performed both qualitative (focus groups) from the collaborative group set up to discuss the and quantitative (structured questionnaires) methods and progress of the review at two intervals methods to ascertain their data (Marsden et al., during the project, (a list of group members is unpublished), while another used both real and given in appendix 2). Any disparity in the findings hypothetical scenarios (Bevan et al., 1993). Our of SE and JH was recorded and resolved by search also identified three reviews which included group discussion.

10 Health Technology Assessment 1998; Vol. 2: No. 15

Chapter 4 Review results

Degrees of uncertainty and the known confounding variables, thereby reducing bias more effectively. The smaller the study, the ethics of RCTs more likely a confounding variable is to distribute Obligations to society of physicians itself unequally, and the greater the argument for and ethics committees ‘minimisation’ (see ‘under-powered’ studies, on Societal benefits page 12). However, ‘minimisation’ does not tackle The epistemological argument for RCTs. Forty- the problem of unknown confounders linked to one articles sought to defend the RCT by using different clinicians or patient variables which are an epistemological argument: a well-designed and not or cannot be enumerated but which never-the- conducted RCT is the most reliable method of less impact on the clinician’s psyche and skew obtaining epidemiological data and, because valid allocation. Burkhardt and Kienle (1983) further evidence is needed to inform widespread practice, pointed out that the RCT is inevitably flawed in the RCT can be justified on utilitarian grounds that the sample is inevitably unrepresentative (Baum, 1983; Byar, 1979; Byar, et al., 1976; Cavan, of the population. However, this could be said 1981; Chalmers and Chalmers, 1994; Chalmers of any alternative research method and, while et al., 1972; Cowan, 1981; Editorial, 1992; Eichler, randomisation is not a perfect solution, it is 1995; Engelhardt, 1988; Ernst and Resch, 1996; perhaps the least bad scientific option for Gilbert et al., 1977; Gillon, 1994; Fox, 1960; Haines, evaluations (Uberla, 1981). 1979; Harrison, 1986; Hill, 1958, 1963; Kardinal, 1994; Lasagna, 1970; Marquis, 1983; Mike, 1988, When second best is best. Ninety-nine articles 1989a, 1993; Miller, 1985, 1993; MRC, 1991; recognised that a well-designed and conducted Passamani, 1991; Raju, 1992; Roy, 1986a; Russell, RCT is good science, but observed that good 1989; Schafer, 1982; Shaw and Chalmers, 1970; science is not necessarily good ethics and/or Silverman, 1985; Smith et al., 1993; Stephenson, pointed out problems with RCTs in particular 1996; Sutton-Tyrrell et al., 1991; Tukey, 1977; disciplines or situations (Abraham, 1941; Altman, Uberla, 1981; Visscher, 1970; Zelen, 1979). One of 1980, 1983; Ambroz, 1978; Anderson, 1980; Angell, the references placed the professional duty not to 1984; Anonymous, 1979; Baum, 1995; Beauchamp harm above that to benefit, and so distinguished and Childress, 1989; Berkowitz, 1995; Bleich, 1995; between the need for scientific evidence to benefit Block and Elahoff, 1979; Bonchek, 1979; Burkhardt society (collective beneficence) and to prevent and Kienle, 1978, 1983; Buyse, 1991; Byer, 1983; possible harm from unproven therapies (collective Canner, 1970; Caplan, 1984; Cassel, 1985; Challah non-maleficence), claiming that an epistemological and Mays, 1986; Cornfield, 1966; Cutler et al., 1966; argument for RCTs supports the latter, not the Day, 1969; De Deyn, 1995; Donner, 1977; Dudley, former goal (Mike, 1989a). Three of the 41 refer- 1983, 1986; Elander and Hermon, 1995; Ernst and ences stated simply that utilitarian justification is Resch, 1996; Feinstein, 1973; Feenberg, 1992; not sufficient; that is, societal benefit is necessary Fleming, 1982; Fried, 1974; Freireich and Gehan, but insufficient to justify the RCT (Fox, 1960; Roy, 1979; Freund, 1970; Ganz, 1989, 1990; Gehan and 1986b; Marquis, 1983). However, they did not take Freireich, 1974; Giertz, 1980; Gifford, 1986, 1995; the argument further. Gilbert, 1995; Glasziou, 1995; Gross, 1993; Grunkemeier and Starr, 1992; Hellman, 1979; Two additional articles attempted to show that Hellman and Hellman, 1991; Henry and Hill, 1995; the RCT was not the most scientifically reliable Hilden et al., 1987; Hilden and Habbema, 1990; method after all, and argued that the random- Kadane, 1986; Kassirer, 1983; Kardinal, 1994; Katz, isation process does not eliminate systematic 1993; Kaufman, 1993; Kleijnen et al., in press; RJ error as is commonly assumed (Taves, 1974; Levine, 1991; Levine and Lebacqz, 1979; Lockwood Urbach, 1993). and Anscombe, 1983; Love, 1975; Lurie, 1994; Mackillop, 1986; McPherson, 1994; Marquis, 1983; Taves (1974) proposed a refinement, called Minogue et al., 1995; Morris and Brown, 1995; ‘minimisation’, to the randomisation process, Neuebauer et al., 1991; Nicholson, 1986; Palmer, which matches study and control groups for 1995; Passamani, 1991; Pollock, 1989; Pringle and 11 Review results

Churchill, 1995; Raju, 1992; Rudicel and Esdaile, (9) Research in general practice, given the 1985; Ruse, 1988; Rutstein, 1970; Schafer, 1982, importance of the physician–patient 1985; Schwartz et al., 1980; Shuster et al., 1985; relationship (Pringle and Churchill, 1995). Simon, 1977, 1994; Spiegelhalter et al., 1994; (10) Research using placebos, where crossover Spodick, 1973, 1982; Steinberg, 1991; Stirrat et al., designs may be preferable, so that all patients 1992; Takaro et al., 1976; Tauer, 1994; Toynbee, have access to active therapy (Elander and 1996; Trochim and Cappelleri, 1992; Van der Hermon, 1995). Linden, 1980; Vere, 1981; Visscher, 1970; (11) Research involving surgery due to various Wallenstein et al., 1980; Weinstein, 1974; technical difficulties (Anderson, 1980; Zelen, 1969). Berkowitz, 1995; Byer, 1983; Bonchek, 1979; Grunkemeier and Starr, 1992; Love, 1975; Thirty-one of the 99 articles argued that, in Van der Linden, 1980; Wallenstein et al., specific cases, second best could be best, that is, 1980); this could also be seen as part of the where alternative non-randomised designs such first stipulation above as surgical skill must as retrospective studies are morally or be taken into account. practicably preferable: (12) When the disease is rare, and recruitment slow, case–control studies may be preferable (1) When equipoise does not apply (Fleming, (Baum, 1994; Byar, 1979; Gehan and 1982; Kadane, 1986; Levine et al., 1991; Ruse, Freireich, 1974; Spodick, 1982). (See 1988; Shuster et al., 1985). We might infer also ‘under-powered’ studies, below.) that those arguing that RCTs should be undertaken only when equipoise applies Three articles pointed out that, historically, some would also find alternative designs preferable medical breakthroughs did not require an RCT when equipoise does not apply, at least when to produce valid evidence, such as the effectiveness the treatments are widely available. (See also of penicillin (Abraham, 1941; Kaufman, 1993; page 22 for a definition of equipoise and a Nicholson, 1986), but this leaves open the question discussion of situations where equipoise does of what to do about potential moderate, but not apply.) worthwhile, effects (see stipulation (3) above). (2) When the primary purpose of study is the estimation of toxicity and effectiveness, Five further articles suggested that there is a as in Phase 1 and 2 trials (Gehan and continuous scale of reliability and that too much Freireich, 1974). emphasis is put on the RCT (Botros, 1990; Dudley, (3) When large differences between treatments 1983, 1986; Reemtsma, 1986), while another author are expected (Ernst and Resch, 1996; Gehan suggested pooling both historical and prospective and Freireich, 1974; McPherson, 1994; data wherever possible (Pocock, 1975). Palmer, 1995; Passamani, 1991; Raju, 1992). (4) When proposed treatment can be compared Cappelleri and Trochim (1995) and Trochim and to standard treatment from a recent previous Cappelleri (1992) proposed a compromise position trial (Gehan and Freireich, 1974); this could and suggested that, when it is unethical or imprac- be seen as part of the first stipulation above, tical to randomise all patients, a cut-off-based since the earlier trial might dispel equipoise. design is ethically acceptable, according to which (See also page 13, on the scientific need the new treatment is given to the most sick, control for replication.) treatment(s) is given to the least sick, and patients (5) When a disease, if left untreated, is lethal who are moderately sick are randomised. However, and for which there is no known effective it could be argued that the most sick are the most treatment (Byar, 1990; Cutler et al., 1966; vulnerable and, since their competence may be Ernst and Resch, 1996; Minogue et al., 1995; brought into question due to possible desperation, Palmer, 1995; Rutstein, 1970). they should be ‘protected’ from investigative (6) When a current treatment can produce procedures, though this might then depend on a virtually assured cure, that is, when the goal how novel the treatments are. (See also page 15, is to reduce toxicity at the possible expense on the risk associated with new therapies.) of effectiveness (Shuster et al., 1985). (7) When there are repetitive emergencies ‘Under-powered’ studies. Six references stated associated with a condition (Spodick, 1982). categorically that conducting ‘under-powered’ (8) When the patient could act as his/her own studies is necessarily unethical on the grounds that, control, for example typical diseases in paired since they are unlikely to produce clear-cut answers, 12 organs (Spodick, 1982). society might be misled with potentially devastating Health Technology Assessment 1998; Vol. 2: No. 15

consequences (Altman, 1980, 1983; Anonymous, exposure to a technology and outcome or 1979; Freund, 1970; Gracia, 1993; Mike, 1989b; chance, since bias has been eliminated in theory. Newell, 1978; Shuster et al., 1985). This argument Other factors can be taken into account when is supported by a further 15 articles, stipulating evaluating the role of chance, including biological that it is statistically necessary for random errors credibility and consistency with other studies; in in measured effects of treatments to be small in the case of rare diseases, results from ‘under- comparison with the size of therapeutic effect powered’ trials (which may be more prone to sought (Ambroz et al., 1978; Brown, 1980, 1987; errors than adequately powered trials) might be Chalmers and Sinclair, 1985; Clayton, 1982; Collins interpreted alongside case–control studies, et al., 1992; Fetter, 1989; Freiman et al., 1978; Hall, for example. See also the scientific need for 1982; Kaufman, 1993; McPherson, 1982; Suther- replication, below. land, 1994; Van der Linden, 1980; Yusuf et al., 1984; Zelen, 1983). Indeed, Feinstein (1973) saw limited The scientific need for replication. Thirty-five value in randomisation in small trials since the references looked at confirmatory trials, and all results would be potentially misleading anyway – except Morris and Brown (1995) saw a need for avoiding error though selection bias is of little repeating trials given certain conditions. The value if error results through imprecision. exception described an anecdotal case where enough evidence was acquired through Phase 1 The use of meta-analysis to increase the statistical and 2 trials of Tarcolimus to make the start power of small trials was proposed, in an ethical of a multicentre RCT unethical (Morris and context, by ten authors (Anello and Fleiss, 1995; Brown, 1995). Barnard, 1990; Chalmers and Lau, 1993, 1996; Collins et al., 1987; Dagan, 1992; Horwitz, 1995; Thirteen references put forward an epistemological L’abbe et al., 1987; Light, 1987; Report to the argument and were concerned that trial results Health Services and Public Review Board, 1993). should be ‘confirmed’ by replication, even if This argument (see the discussion section, on page a previous result was ‘statistically significant’ XX) is predicated on the reliability of meta-analysis (Banta and Thacker, 1990; Buyse, 1991; Byar et al., as a method (Antman et al., 1992; Cappelleri et al., 1976; Fleming, 1982; Gehan, 1982; Henry and 1996; Detsky et al., 1992; Fortin et al., 1995; Gilbaldi, Hill, 1995; Nicholson, 1986; Senn, 1991a; Simon, 1993; Hasselbald et al., 1995; Lau and Chalmers, 1982, 1994; Spiegelhalter et al., 1994; Stamer and 1995; Lau et al., 1995; Peto, 1987; Stewart and Lee, 1982; Zelen, 1982). In this way, generalisations Parmer, 1993). might derive inferential support from the process of induction which proceeds from particular cases Eight references recommended that, given the to generalisations. Inductive support which takes choice between a small trial and no trial, the small the form of multiplicity of the same observation is trial is preferable and/or that Bayesian designs are called ‘enumerative’ induction. A different type of most appropriate in such cases (Pocock and Hughes, induction, called ‘eliminative’ induction, is based 1990; Fayers and Machin, 1995; Goodman and on variety, not multiplicity, of instances. Seven Berlin, 1994; Powell-Tuck et al., 1986; Rahimtoola, references were concerned to enhance generalis- 1985;), especially in rare diseases (Freedman, 1989; ability of results by repeating trials on different Lilford et al., 1995). However, there are situations samples, for example HIV vaccine trials in develop- where worthwhile gains are small in relation to the ing countries where there is considerable strain precision achievable by all envisaged trials such that variation (Burkhardt and Kienle, 1978, 1983; power remains ethically critical (Barnard, 1990; Ellenberg et al., 1990; Glasziou, 1995; Lantos, Buyse, 1991; Collins et al., 1987; Lilford, 1990; 1994a; Lurie et al., 1994; Pringle and Churchill, Lilford and Thornton, 1992; Lilford and Johnson, 1995). However, Senn (1991a) points out that 1990; Mathews, 1988; Pocock and Hughes, 1990). repeating a trial on a different sample will not Even then, a trial is not necessarily unethical since, necessarily improve generalisability (if the same given equipoise, the patient does not lose out in factors lead people to participate, for instance). prospect and a more precise estimate (though not a (See also page 16, on the risk that research will ‘definitive’ answer) is obtained by going ahead with not inform practice). We point out that a variety of a trial than by eschewing randomisation altogether. study designs might be used to ‘confirm’ results by (Lilford, 1990; Lilford and Thornton, 1992; Lilford way of eliminative induction (see also page 12, on and Johnson, 1990). under-powered studies). A third type of inductive inference is concerned with subjective degree of A valid statistical association from an RCT can belief (or Bayesianism) and is able to avoid taking be attributed to a causal relationship between sides on the multiplicity versus variety debate 13 Review results

above. Some theorists reject inductive inference what is needed is education and not just more trials altogether on the basis that its logic is invalid. Senn (see page 16, on the risks that research will not (1991a) examined Karl Popper’s philosophy of inform practice.) If it is expected that replication science which rejects induction as a process of will be necessary, it might be more appropriate to logical inference, but, at the same time, recognises conduct independent trials concurrently (wherever its psychological necessity. According to Popper, a possible), thereby circumventing the problem scientific theory can only be falsified by deduction (Levine, 1988, 1991). from observations that refute the hypothesis. Theories can never be confirmed, only ‘corrob- Seven references showed how repeating a trial orated’ as attempts to refute them fail – a prob- might be justified when (1) the previous evidence abilistic solution to the problem of induction is to was in some way flawed (Hellman, 1979), (2) take a subjectivist position in respect of probability previous evidence was from an ‘explanatory’ trial and use Bayesian methods (see appendix 3). which is inadequate to demonstrate effectiveness such that a subsequent ‘management’ trial might Nine references were concerned to balance the be justified (Willan, 1994), (3) evidence from need for equipoise with the putative scientific previous work was based on surrogate markers as requirement to replicate results, given the possible opposed to hard end-points which patients would tension between the alleged need to repeat a trial, recognise and value (Nowak, 1994), or (4) if once evidence has been made available, and the previous trials had been stopped early with requirement of upholding the moral principles imprecise results (Ashby and Machin, 1993; of beneficence and non-maleficence implicit in Fleming, 1982; George et al., 1994, Tannock accepting the need for equipoise (Byar, 1990; and Boyer, 1992). In all cases, equipoise might Royall, 1991) – see uncertainty as justification for or might not have been dispelled by the previous RCTs (page 19). Thus, it could be argued that it evidence. (See also page 16, on the risks that is unethical to embark on a new trial without first research will not inform practice.) preparing a systematic review (or meta-analysis) of the relevant existing trials (Chalmers and Lau, One article suggested that RCTs are justified, 1993, 1996). If the necessary review of previous even when there is evidence from previous studies yields an overall odds ratio (OR) close to 1, research, and highlighted the problems of using then (any) pre-existing equipoise is likely to be old treatments in new situations, arguing that maintained, especially if confidence limits are wide, effects cannot simply be inferred from data and a repeated trial would not violate the moral obtained in similar, but different, settings (Tukey, principle to do the best for patients, in prospect. 1977). On the other hand, Gehan and Frieireich Three of the nine articles suggested that recruit- (1974) would object to randomised designs when ment is acceptable when it is carried out by new ‘bits’ are simply bolted onto old treatments. individual physicians who are equipoised despite This may, however, be a case for equivalence trials, existing evidence sufficient to convince all but especially if the costs are unequal, although others the most sceptical (Collins et al., 1992; ISIS-2 have pointed out that, given equipoise, all trials are Collaborative Group, 1988; Simes, 1991). There ‘equivalence’, and the issues at stake are (1) size was no reference to the possibility that physicians of effect which would change practice, and hence may still be equipoised because they are unaware which the trial should measure, and (2) whether of a previous result, or that they might previously a null result, or some other result, is most likely in have believed a treatment to be inferior, moving prospect. (See also page 23, when equipoise does into equipoise only after seeing results favouring not apply.) More data might be needed to show it. (See also page 16, on risks that research will that a treatment justifies its financial cost, even in not inform practice.) The British clot-buster trial, the face of a positive result, e.g. the Extra-corporeal the Second International Study of Infarct Survival Membrane Oxygenation (ECMO) trial. In that (ISIS-2), was conducted in precisely such circum- case, the new treatment may be made available only stances (ISIS-2 Collaborative Group, 1988). An in the trial on the basis of collective equipoise (see interesting situation arises when the cognoscenti also page 22, on when equipoise does not apply). believe A is better than B, but most practitioners do not. Assuming that the practitioners had differ- Early stopping to make beneficial treatments ent (more sceptical) starting beliefs, repetition widely available. Seventy articles discussed the ethics would be entirely justified. If, however, the data are of stopping trials early when there is some evidence so strong as to shift almost any reasonable prior of efficacy (Anscombe, 1963; Armitage, 1963, 1975; belief, or if ‘slow adapters’ are slow because they Ashby and Machin, 1993; Baum et al., 1994; 14 are unaware of results, or, for other reasons, then Bellisant et al., 1994; Berry and Ho, 1988; Berry Health Technology Assessment 1998; Vol. 2: No. 15

et al., 1992; Bulpitt, 1983; Chalmers, 1979; Chalmers (3) Artificial heterogeneity of results may appear and Lau, 1996; Canner, 1979; Crowley et al., 1994; in subsequent overviews (Hughes et al., 1992). de Groot and Kennedy, 1995; Fleming, 1993; (4) Practitioners may not be persuaded Fleming and DeMets, 1993; Fleming and Waterlet, (Freedman and Spiegelhalter, 1989; Liberati, 1989; Freedman and Spiegelhalter, 1989; Freedman 1994; Prescott, 1979); this overlaps with point et al., 1994; Geller, 1987; Geller and Pocock, 1987; (2) above, but is not identical as more George et al., 1994; Goldman, 1987; Goodyear and precision may be needed to drive change Levine, 1987; Green et al., 1987; Greenhouse, 1992; in some contexts than others. Hamilton, 1983; Hetjan et al., 1992; Hilden and (5) The secondary aims of a trial may be Habbema, 1990; Hill, 1962; Hill and Sancho- compromised (Crowley et al., 1994; Fleming, Garnier, 1979; Hughes, 1993a, b; Hughes et al., 1982; Green et al., 1987; Greenhouse, 1992; 1992; Klimt and Canner, 1979; Laupacis et al., Hilden and Habbema, 1990; Hughes et al., 1991; Lewis, 1993; Liberati, 1994; Lockwood 1992; Laupacis et al., 1991; Lewis, 1993). and Anscombe, 1983; Lurie, 1994; Machin, 1992; (6) Some outcome measures are not obtained McPherson, 1982; Meier, 1979; Meinert, 1979; quickly (Raju, 1992). We point out that Parmar and Machin, 1993; Peto et al., 1976; Pfeffer, longer-term follow-up may compensate for 1993; Pignon and Arrigada, 1994; Pocock, 1977, adverse effects on short-term outcomes. 1979, 1993; Pocock and Hughes, 1989; Pollock, (7) Some people may be denied restricted 1989, 1993; Prescott, 1979; Propert et al., 1991; Raju, treatments (i.e. when treatments are not 1992; Rubinstein, 1982; Ruse, 1988; Seibert et al., expected to be available outside a trial) – 1993; Simberkoff et al., 1993; Simon, 1994; Smith randomisation might be the only route et al., 1993; Spiegelhalter et al., 1993, 1994; Stamler, to a preferred therapy (Lockwood and 1979; Sutton-Tyrrell et al., 1991; Whitehead, 1994; Anscombe, 1983). Working Group on Arrhythmias of the European Society of Cardiology, 1994). All authors except Given the moral advantages of early stopping in Ashby and Machin (1993), Berry (1988), Freedman certain circumstances, five articles argued that a and Spiegelhalter (1989), George et al. (1994), predefined stopping rule should be followed and Hilden and Habbema (1990), Hill (1962), Hughes that recruitment should be halted when the pre- et al. (1992), Laupacis (1991) Lewis (1993), Liberati specified boundary had been crossed (Bellisant, (1994), Machin (1992), Peto et al. (1976), Pignon 1994; de Groot and Kennedy, 1995; Lurie et al., and Arrigada (1994), Pocock and Hughes (1989), 1994; Whitehead, 1994; Working Group on Pocock (1993), Pollock (1989), Prescott (1979), Arrhythmias of the European Society of Cardiology, Raju (1992), Rockhold (1993) and Working 1994). Fourteen articles, however, argued against Group on Arrhythmias of the European Society the sole use of stopping rules, and suggested that of Cardiology (1994) believed that it is always they are merely guidelines and not strict rules unethical to continue a trial once efficacy has been (Ashby and Machin, 1993; Fleming, 1982; Fleming established and/or that fixed-sample trials with no and DeMets, 1993; Freedman and Spiegelhalter, mechanism for early stopping are unethical. 1989; Goldman, 1987; Goodyear and Levine, 1987; Lewis, 1993; Liberati, 1994; Ruse, 1988; Simon, Chalmers and Lau (1996) argued for early 1994), and/or that data outside the trial should be stopping when a trend conforms with the results taken into account (Ashby and Machin, 1993; of a previous meta-analysis. Baum et al., 1994; Parmar and Machin, 1993; Spiegelhalter et al., 1994). We also point out that, Problems associated with early stopping are: by incorporating a frequentist stopping rule, early termination may introduce bias as well as impre- (1) Failure to establish long-term safety (Berry, cision. (See also early stopping when there is some 1988; Rockhold, 1993), but Stamler (1979) evidence of toxicity, on page 24.) observes that no trial can ever answer every question about potential toxicity. However, one reference asserted that a researcher (2) Reduced statistical precision (Ashby and has the right (if not the obligation) to finish a trial Machin, 1993; George et al., 1994; Hill, 1962; to statistical significance, without being forced to Lewis, 1993; Machin, 1992; Peto et al., 1976; make premature conclusions (Ruse, 1988). (See Pignon and Arrigada, 1994; Pocock and also page 24, on the censure of interim results.) Hughes, 1989), though Pocock and Hughes (1989) suggest that statistical adjustment, for Risks to society example Bayesian ‘shrinkage’, could be made Risks associated with new therapies. Nine for this. references were in favour of restricting new 15 Review results

treatments to trials in order to establish definitive entry. Buyse (1991), Challah and Mays (1986), results before equipoise is lost (Dupont, 1991; Chalmers (1975a, 1990), De Deyn (1995), Goodare Freedman, 1992; Freedman et al., 1989; Freston, and Smith (1995) and Tauer (1994) simply com- 1986; Pfeffer, 1993; Reidenberg, 1987; Silverman, plained of the existence of the double standard, 1985), or to hedge in respect of new therapies while Hutton (1996) made explicit the uncertainty (Chalmers and Chalmers, 1994; Lantos, 1995) – over treatments used in routine practice apparently we would argue, however, that, if equipoise applies, given in the patient’s best interests. The literature expected benefit is sufficient to offset any risk such did not address the possibility that different that randomisation does not reduce the risk associ- standards may be ethically justified such that they ated with such technologies, though restricting are not ‘double’. It could be argued that a defence them to trials might serve to reduce the risk to for different standards can be made in that RCTs society as a whole. One reference, however, distin- carry a potential, if not actual, conflict between the guished between general and desperate sufferers, two roles of physician (as scientist and as therapist). and asserted that such restriction was only justified That said, a doctor may harbour any number of in the former population (Ruse, 1988). Six further interests which may conflict with the interests of references were concerned to make new treatments his/her patient. Allied to this, the default assump- widely available early (especially when patients are tion for patients may be that they are not in a trial, desperate), claiming that patients have an over- and patients’ perceptions of the detail which is riding ‘consumer’ right (Ellenberg et al., 1992; warranted may hinge on whether or not they Fennberg, 1992; Illich, 1976; Institute of Medical are in the default situation. Ethics, 1992; Levine, 1991; Young, 1993). Three references observed that double standards The desire to contain risk of harm from society at exist within trials: (1) there is no regulation of large and to protect the ‘window of opportunity’ n of 1 trials (Levine, 1991), and (2) there is less (equipoise) can be distinguished from restrictions stringent regulation of pharmacological than due to scarce resources such that equipoise is surgical trials, and, while there are intrinsic not a necessary condition for trial entry (Challah differences, there remains the potential for a and Mays, 1986; Freund, 1970; Hill, 1963; conflict of interests in both cases (Cornfield, Markman, 1992). 1966; Spodick, 1973).

In both cases, but particularly in the latter, Risks that research will not inform practice. Three concerns about distributive justice should articles presented evidence to suggest that RCTs prevail (see also page 23, on when equipoise do have a measurable influence on general does not apply.) medical practice patterns: on treatment of myocardial infarction (Friedman et al., 1983; Lamas Risks associated with non-participation in trials. et al., 1992), and the practice of radiotherapists Twenty-three articles addressed the ‘double and medical oncologists (Stephens and Gibbons, standards’ issue where there are different, more 1993). The authors, however, do recognise that stringent, standards which apply to RCTs than there is sometimes a time lag between production to innovative practice. This implies that there is of compelling research evidence and change greater ‘care’ associated with trial entry than with in practice. routine practice such that non-trial patients incur an implicit risk by being treated routinely. Eight Seven references provide examples of medical articles expressed a need to resolve this double practice which has followed research only after standard by embracing one universally applicable considerable delay; this is arguably a case for standard: Kleijnen et al. (unpublished) stated that repeating trials when such replication is likely to there should be a single standard across the board convince others. Ernst and Resch (1996) described and that this should be determined by the society the famous case of scurvy: it took the British Navy in question (Collins et al., 1992). An editorial 50 years before the regimen of citrus fruit was (1995) regarded the regulations in normal practice adopted after the publication of trial evidence as adequate for RCTs also; whereas Lantos (1994b) of its benefit in 1743 by Lind, itself not the first and Chalmers and Silverman (1987) expressed evidence of therapeutic or prophylactic benefit. a need to tighten the non-trial procedures in prac- Since then, the quality of evidence has increased, tice so that they are on a par with RCT regulations. but not the speed with which it filters down to Segelov et al. (1992) proposed that the decision practice. Avorn et al. (1982) reviewed comparative not to participate in an approved trial should be studies of propoxiphene analgesics versus aspirin 16 subject to the same ethical requirements as trial versus placebo, and found no treatment effect, Health Technology Assessment 1998; Vol. 2: No. 15

yet 20% of a sample of physicians working in the formulate health insurance and reimbursement field were under the misconception that aspirin policy. Indeed, evidence will impact differently was the superior therapy. Chalmers (1974) and according to what prior beliefs are held. Of course, Chalmers and Sinclair (1985) noted that despite as evidence accumulates, so everyone’s beliefs evidence from six controlled studies to suggest should, at least in theory, converge on the truth. that stilboestrol had no effect on preventing spontaneous abortion, a large number of pregnant In order to change practice in light of RCT women continued to receive the drug after 15 years findings, it has been hypothesised that a meta- or more. Elliott (1992) and Fisher and Fisher analysis of trials might be more persuasive than (1996) described the continued use of antidepres- a single large trial (Stewart, 1992). However, con- sants in children despite evidence to suggest that ducting an audit of practice, while known to raise the treatment has no advantage over placebo, while medical standards, may also improve awareness Lamas et al. (1992) showed that, although ISIS-2 of research findings. As a result, Liberati (1994) and other randomised trial results supporting the sees the relationship between trials and practice use of aspirin for myocardial infarction were ‘clear- as very complex, and Avorn and Soumerai (1983) cut’, as many as 29% of patients studied were not suggest that that commercial tools for disseminat- taking aspirin long after the results had been ing scientific information are most effective, i.e. published. Clinicians would be behaving ethically if using the principles of behavioural science, market they recruited while equipoise remained, but they research and communications theory. Boissel would not appear to be justified in continuing to (1989) concludes that research into evaluating the randomise their patients without disclosing a loss of effectiveness of dissemination is urgently needed equipoise, simply in order to persuade others. This and that an appropriate method for evaluating really would be merely using patients as means to the impact of RCTs is yet to be designed. societal ends and would thus violate the injunction of Kant, though it would be acceptable to utili- Dual obligations of physicians to tarians if the greatest happiness for the greatest society and to the individual patient number were, in fact, served thereby. The legacy of Hippocrates: the therapeutic obligation Pelligrino and Thomasma (1981) state that Sixty-four articles addressed the ancient debate individual physicians have the right to bypass over whether or not the interests of the individual research results and that they are justified in should take precedence over those of society. Four relying on their own individual judgement as of the 64 articles suggested that tension between opposed to scientific consensus. It is important, society and the individual could be reduced by the given the premise that autonomy is crucial, that use of adaptive designs or a mixture of frequentist they are familiar with research findings, whether and Bayesian statistics (Clayton, 1982; Palmer, or not they act accordingly, and education and 1993; Palmer and Rosenberger, 1998; Roy, 1986a). dissemination should be as pervasive as possible. Adaptive designs, in this context, alter the randomisation ratio to favour the treatment which appears The importance of disseminating results is widely ‘ahead’ at any one time as the trial progresses. recognised (Antman et al., 1992; Boissel, 1989). However, adaptive designs are not a perfect Indeed, having looked at the empirical evidence, solution to the problem since they do not maximise Boissel (1989) concludes that, at present, doctors individual expected utility which would be achieved do not even know RCT findings and, consequently, by opting for the therapy currently showing the do not use RCTs as a basis for determining patient most promise (Kadane, 1996; Royall, 1991; Simon, care, but instead use technologies which have 1977). This must be so, since the expected utilities not been adequately tested with RCTs. What is (which, given equipoise, were equivalent at the more, important results are sometimes not even start) are sensitive to any alterations in probability. published in standard textbooks and review (See also page 23, on when equipoise does recommendations (Antman et al., 1992). One not apply.) article pointed out that clinical trials are carried out for a variety of audiences and, while a result Thirty-four of the 64 articles expressed unwavering may be persuasive for one audience, it may not be support for the absolute primacy of the ‘therapeutic for another (Brown, 1980). Possible audiences are obligation’; that is, a physician always has a duty to trial investigators, investigator peers, practising do his/her ‘best’ (see the introduction, page 1) for specialists, general physicians, licensing agency each individual patient (Anonymous, 1979; Barry (e.g. the US Food and Drugs Administration (FDA), and Molyneux, 1992; Beecher, 1966; Blumgart, 1970; or the consumer, that is, patients and those who Byer, 1983; Engelhardt, 1988; Fried, 1974; Giertz, 17 Review results

1980; Gillon, 1994; Goodare and Smith, 1995; for the Protection of Human Subjects of Hellman, 1979; Hill, 1963; Kairys, 1996; Kardinal, Biomedical and Behavioral Research, 1978). 1994; Katz, 1993; Lantos, 1994b; Levine, 1992; None of these were explicit about accepting some Marquis, 1983; Miller, 1993; Moore and Papp, 1996; loss of expected utility in order to honour the said Nyapadi, 1995; Pfeffer, 1993; Pringle and Churchill, contract. All references examining the ‘social 1995; Schafer, 1982; Schwartz et al., 1980; Scoles and contract’ except Hilden (1990) acknowledged that Silagy, 1993; Shaw and Chalmers, 1970; Smith, 1992; not all individuals will have benefited and so would Thornton, 1992; Toynbee, 1996; Veatch, 1981; Vere, not carry such an obligation and/or that the ‘social 1981; Visscher, 1970, 1990). In addition, one contract’ should be an optional transaction and reference suggested that society does not have a should not be rule governed. Alternative ideas of right to pursue the Melioristic goal of medical the ‘social contract’ appeal to the following four progress (the attempt to make the world a better different ideas: place by human effort) and that medical experimentation objectifies the patient (Jonas, 1969). (1) Justice, but then any debt should be discharged to past not future patients Three of the 64 articles had a more utilitarian (Gilbert, 1995; Marquis, 1983). outlook and so viewed valid RCTs as ethical, at least (2) A duty to rescue according to which the where society is likely to benefit (Baum, 1995; therapeutic obligation should be overridden Dworkin, 1987; Kennedy, 1988). However, two if research may prevent or remove important references made the distinction between rule- and harms to others while imposing no more than act-utilitarianism, claiming that the former permits a minor compromise to the individual. The the use of rules/rights in order to main-tain trust therapeutic obligation is still regarded as the for the greater ‘good’ (Dworkin, 1987; Kennedy, primary obligation which must be satisfied, 1988). This means that respecting patient auton- but not always fully satisfied (as with the omy might promote the greatest good for the Rawlsian ‘lexical priority’ dictate), before greatest number in the long run. However, one a secondary obligation can be discharged article argued that the rule-utilitarian doctrine (Ackerman, 1994; National Commission for allows too many exceptions, that is, an act contrary the Protection of Human Subjects of Bio- to the rule is permissible if the consequences on medical and Behavioral Research, 1978). that particular occasion are good, but this then (3) If an individual acts only on that maxim destroys the distinction between acts and rules he/she could at the same time will that it (Nicholson, 1986). However, it could still be argued should become a universal law (the Kantian that the consequences of a general infraction of requirement that one’s maxims be univer- the rules would be disastrous to patients and the salisable), and if he/she does not will that public view of trials and of the behaviour of clinical everyone should deny consent to RCTs, then professionals. Better still, the rules themselves he/she should consent to trial entry (Marquis, could be amended so that the exception becomes 1983). However, while we may not wish to part of the rule (Lyons, 1965) and the inclusion of make universal a policy of never helping deontic notions of respect for autonomy may serve others, we may not wish to commit everyone to increase social utility without need for rules-of- to helping others in just the same way (an thumb. Two references raised the interesting imperfect duty – see the glossary). It is not situation of an epidemic during which such clear why there should be a duty to participate individualism might be an unaffordable luxury in an RCT even if we do not wish everyone to (Fethe, 1993; Jonas, 1969), despite flying in the be selfish. A maxim in refusing consent might face of Kant’s injunction. Four references thought be: let people participate in trials, but only if that strict individualism (i.e. individual rights they informedly consent to. This is univer- should always trump social needs) should be salisable and does not commit everyone to relaxed in the context of trials (Editorial, 1983; take part just in case they do not want to. Elander and Hermon, 1995; Gifford, 1986, 1995), (4) The Lockean idea that we are better off with while 11 articles addressed the idea of a ‘social an institution that conducts RCTs than with- contract’ which might impose a duty on the indi- out. Cases (3) and (4) do not, however, imply vidual to contribute to medical research due to that patients could be conscripted without their having benefited from previous research (or against) consent (Marquis, 1983), and, (Ackerman, 1994; Caplan, 1984; Gilbert, 1995; since consent should unmask any lack of Gilbert et al., 1977; Hilden and Habbema, 1990; equipoise, only altruistic patients would Lesser, 1989; Levine, 1981; Mackillop, 1986; consent given a preference and freely 18 Marquis, 1983; Meier, 1979; National Commission available trial treatments. Health Technology Assessment 1998; Vol. 2: No. 15

Three references argued against the idea of patient has nothing to lose, in prospect, by trial patients or healthy volunteers having a social participation (Alderson, 1996; Angell, 1984; obligation to participate in research on the basis Anonymous, 1979; Bartlett and Cornell, 1991; that any such obligation would invalidate informed Baum, 1994; Baum et al., 1994; Beecher, 1966; consent (Fethe, 1993; Jonas, 1969; Lesser, 1989). Berkowitz, 1995; Block and Elahoff, 1979; Boissel, 1989; Botros, 1990; Bracken, 1987; Bradley, 1993; It could be argued that, rather than a patient Brewin and Bradley, 1989; Burkhardt and Kienle, having or not having an obligation to some 1978; Buyse, 1991; Byer, 1983; Capron, 1978; Chard external entity, that is, society, he/she may see and Lilford, 1998; Challah and Mays, 1986; participation as a process of self-realisation (follow- Chalmers, 1968, 1975b, 1982; Chalmers et al., 1972; ing the Hegelian tradition), but recruitment to Chipman, 1993; Clayton, 1982; Collins et al., 1992; conventional trials still rests on the valid consent Coulehan et al., 1985; Cowan, 1981; Curran, 1979; of the individual in question, such that any lack Cutler et al., 1966; de Groot and Kennedy, 1995; of equipoise is transparent. Dudley, 1983; Eichler, 1995; Emrich and Sedrank, 1996; Fetter et al., 1989; Freedman, 1987, 1992; In light of the above constellation of arguments, we Freedman and Spiegelhalter, 1992; Freedman et al., would say that there are no compelling arguments 1984, 1994; Feinstein, 1973; Ferner and Neumann, to support a clinician entering patients in trials for 1992; Fried, 1974; Freireich and Gehan, 1979; societal gain against the individual’s best interests. Ganz, 1989, 1990; Gehan and Freireich, 1974; Arguments that patients have a duty to do so are Gifford, 1986, 1995; Gilbert et al., 1977; Gillett, only a little stronger – acts of altruism can only be 1994; Gillon, 1994; Goodyear and Levine, 1987; anticipated when a patient has little to lose and Gore, 1995; Grady, 1991; Gross, 1993; Harrington should not be expected in life or death decisions. et al., 1994; Hellman, 1979; Hellman and Hellman, This is, however, predicated on the informed 1991; Hilden and Habbema, 1990; Hill, 1963; consent of the individual in question. Hollenberg et al., 1980; Hutchins and Eckes, 1996; Hutton, 1995; Ingelfinger, 1972; Institute of Is the RCT at odds with the Medical Ethics Working Party, 1992; Johnson et al., therapeutic obligation? 1991; Kadane and Seidenfeld, 1986, 1996; Kassirer, Seventeen of the 34 articles which expressed 1983; Katz, 1981; Kaufman, 1993; Kennedy, 1988; unwavering support for the therapeutic obligation Korn and Baumind, 1991; Lantos, 1994; Lebacqz, further stated that RCTs could be seen as an 1977; Levine, 1988, 1991; Levine and Lebacqz, adjunct to good individual patient care, rather 1979; Lilford, 1987, 1988, 1994; Lilford and than as a threat to it (Anonymous, 1979; Blumgart, Jackson, 1995; Lilford et al., 1991; Lockwood and 1970; Engelhardt, 1988; Fennberg, 1992; Gillon, Anscombe, 1983; Machin, 1994; Marquis, 1983; 1994; Goodare and Smith, 1995; Hill, 1963; Mike, 1989a; Miller and Appelgate, 1993; Moore Kardinal, 1994; Lantos, 1994a; Levine, 1992; and Papp, 1996; Neugebauer et al., 1991; Nichol- Marquis, 1983; Nyapadi, 1995; Pfeffer, 1993; son, 1986; Oakley, 1990; Oettinger and Berger, Schafer, 1985; Scoles and Silagy, 1993; Smith, 1992; 1989; Passamani, 1991; Pfeffer, 1993; Peto et al., Visscher, 1970). However, none of these articles 1976; Pollock, 1989, 1993; Raju et al., 1992; Report considered cases when equipoise does not apply. to the Health Services and Public Health Review Board, 1993; Royall, 1991; Rudicel and Esdaile, Conversely, ten of the 34 articles thought that 1985; Sacks et al., 1982; Schafer, 1982, 1985; RCTs necessarily violate patient rights to the best Schaffner, 1996; Sedrank, 1996; Senn, 1993; Shaw treatment available and see the individual as and Chalmers, 1970; Shimm and Spece, 1983; ‘sacrificing’ him/herself for the benefit of future Shinebourne, 1984; Spodick, 1982; Steinberg, patients (Beauchamp and Childress, 1989; Giertz, 1991; Stephenson, 1996; Stirrat et al., 1992; 1980; Hellman, 1979; Pringle and Churchill, 1995; Tannsjo, 1994; Taves, 1974; Thornton, 1992; Schafer, 1982; Schwartz et al., 1980; Thornton, Thornton and Baum, 1995; Toynbee, 1996; Van 1985; Toynbee, 1996; Vere, 1981), particularly der Linden, 1980; Veatch, 1981; Weinstein, 1974; in surgical trials, according to Byer (1983). Wennberg, 1990; Weymuller, 1996).

Uncertainty as justification for RCTs All references except 24 expressed the idea of Uncertainty and substantial uncertainty. In an clinical uncertainty in terms of the concept of attempt to resolve the putative tension between equipoise. Equipoise only arises when treatments societal and individual interests, 129 articles are an equal bet in prospect. Equipoise is more than addressed the notion that, if there exists uncer- indifference between treatments as it captures the tainty over treatment effects, then the individual uncertainty over which treatment is best, as opposed 19 Review results

to confidence that they are the same. The Eight of the 129 articles specifically attacked the exceptions failed to recognise that, since medicine polarisation of knowledge (know or not know) and is an inexact science, further constraints are neces- hence directly or indirectly invoked the concept sary otherwise any RCT would be justified, irre- of degrees of knowledge and hence of equipoise spective of current evidence (Anonymous, 1979; (Botros, 1990; Gifford, 1986, 1995; Hellman, 1979; Barry and Molyneux, 1992; Bracken, 1987; Burk- Hellman and Hellman, 1991; Lilford and Jackson, hardt and Kienle, 1983; Buyse, 1991; Capron, 1978; 1995; Lockwood and Anscombe, 1983; Royall, Challah and Mays, 1986; Chalmers, 1968; Clayton, 1991). Seventeen articles argued that the concept 1982; Collins et al., 1992; Curran, 1979; Dudley, of equipoise was seldom achievable in practice 1983; Gillon, 1994; Hutchins, 1996; Lantos, 1995; and/or was merely an ethical construct and so Mike, 1989a, 1993; Miller, 1993; Moore and Papp, was inevitably elusive in practice (Byer, 1983; 1996; Oakley, 1990; Peto et al., 1976; Pringle and Challah and Mays, 1986; Cutler et al., 1966; Churchill, 1995; Raju, 1992; Schafer, 1982). The Dudley, 1983, 1986; Lockwood and Anscombe, remainder discussed different forms of ‘equipoise’ 1983; Nicholson, 1986; Schafer, 1982; Stephenson, which put varying constraints on the idea of clinical 1996; Weinstein, 1974), especially in surgery uncertainty. One reference discussed what the (Berkowitz, 1995; Buyse, 1991; Gross, 1993; rationale might be for mounting a trial when Oettinger and Berger, 1989; Pollock, 1989, 1993; equipoise exists, i.e. there must be some ‘reason’ for Schafer, 1982). While six articles proposed that running a particular trial rather than to do nothing RCTs are necessary despite this objection (Buyse, in such circumstances. However, equipoise is not 1991; Challah and Mays, 1986; Cutler et al., 1966; necessarily the same as thinking that a null result is Gross, 1993; Schafer, 1982), and perhaps putting most likely, because, although improved outcome as more emphasis on informed consent in order to the main comparison may be likely in prospect, its compensate (Stephenson, 1996), others thought likely extent may be in doubt, so that it is unclear that the RCT was unethical on this basis (Berkowitz, whether the ‘costs’ are warranted. In this case, 1995; Pollock, 1989), or sometimes simply unneces- equipoise may apply and a trial may then release a sary (Dudley, 1983, 1986; Oettinger and Berger, ‘locked’ decision (Chalmers, 1968; Gifford, 1995). 1989). The remainder simply mooted this point There is also some evidence to suggest that trial (Buyse, 1991; Byer, 1983; Nicholson, 1986; Pollock, patients fare better than those treated in routine 1993; Weinstein, 1974). practice (Stiller, 1989; Thornton and Baum, 1996), but this should not be offered as an inducement to Collective and individual equipoise. Nineteen patients, since it conflicts with the Helsinki require- articles regarded the existence of collective ment to do the ‘best’ for all patients, whether in a equipoise as sufficient justification for a trial, that trial or not. Interestingly, another article produced is, a trial is ethical if experts in general, rather than evidence to suggest that innovations which were the particular clinician (or clinician–patient pair) brought to the stage of RCTs were ‘successful’ only are equipoised (Alderson, 1996; Baum et al., 1994; half of the time and they were ‘highly preferred’ Chipman, 1993; Collins et al., 1992; de Groot and only one-eighth of the time (Gilbert et al., 1977). Kennedy, 1995; Eichler, 1995; Emrich and Sedrank, This result gives substance to the claim that there 1996; Freedman, 1987; Gillett, 1994; Johnson and is, as a general rule, little basis for choosing between Lilford, 1991; Kadane, 1986, 1996; Passamani, 1991; a standard and a new therapy prior to a trial, and Pfeffer, 1993; Schaffner, 1996; Sedrank, 1996; would provide the rationale for clinicians to remain Shimm and Spece, 1983; Tannsjo, 1994; Weymuller, in equipoise in the face of theoretical equivocal 1996). This willingness to discredit individual evidence. However, this relates to the basis for prior ‘hunches’ in favour of the collective equipoise is beliefs, not the ethics of what to do given the buttressed by evidence which indicates that physicians’ or patients’ prior beliefs. innovative therapies that are brought to the stage of an RCT are ‘successful’ only half of the time Eight of the 129 articles stated explicitly that (Gilbert et al., 1977) and that people who hazard, RCTs are necessarily unethical, even when equi- even educated, ‘guesses’ are frequently much poise applies, because allocating treatment by further off the mark than they expect to be (Albert lot is contrary to a patient’s need for the advice and Raiffa, 1969). Johnson et al. (1991) argue that, of his/her physician by diluting the salutary effect while collective equipoise must apply within certain of the physician him/herself appearing to know limits (preference ratio 80:20), a physician has a what to do, even in the face of uncertainty (Mike, duty, if holding a personal preference, to disclose 1989; Schafer, 1982) or equipoise (Gilbert, 1995; (or, at least, offer to disclose) this preference. Harrington, 1994; Hellman, 1979; Kassirer, 1983, However, four of the 19 articles found the idea of 20 Thornton, 1994; Toynbee, 1996). collective equipoise difficult to conceptualise in Health Technology Assessment 1998; Vol. 2: No. 15

Bayesian terms, given degrees of expertise among Ten articles examined the implications of the medical community and the difficulty in aggre- research on children. Six references believed it gating individual ‘priors’ to form a collective unethical to confer greater than ‘minimal risk’ ‘prior’ (Gifford, 1986, 1995; Hilden and Habbema, of harm on children irrespective of possible bene- 1990; Machin, 1994). Kadane (1986, 1996), fits (Berglund, 1995; Kopelman, 1981, 1989; Munir, Schaffner (1996) and Sedrank (1996) invoke a 1992; Nicholson, 1986; Tauer, 1994), whereas concept of collective equipoise to determine the Dworkin (1987), Freedman et al. (1993), Kauffman appropriate allocation of trial subjects to (1994) and the MRC (1991) believed that greater treatments – a treatment is acceptable to ethics than minimum risk was acceptable when it is offset committees if at least one expert prefers it to by the expected benefits. However, there is a variety the alternative(s). (See also patient equipoise, of standards, namely the National Health and on page 22.) Medical Research Council, Canberra (NHMRC), Medical Research Council (MRC), British Thirty-three articles discussed what size of risk Paediatric Association, Royal College of Physicians, is morally acceptable for trial participants from and the US Department of Health, Education and the perspective of an ethics committee. All Welfare (Berglund, 1995). The most often cited except Elks (1993) were concerned to establish interpretation of ‘minimal risk’ was that risk of the appropriate level of risk, regardless of the harm should not be greater than that ordinarily risks patients themselves are willing to take, while encountered in daily life or during routine medical the exception suggested that individuals should practice (MRC, 1991). However, in an attempt to always be able to decide for themselves without identify what exactly is meant by ‘everyday’ risks, any paternal protection (Elks, 1993) – see also Freedman (1993) and Kopelman (1989) point out patient equipoise, on page 22. Tannsjo (1994) that the risks incurred by everyday life vary across distinguishes two different philosophical views: societies (or, indeed, classes within societies) such the rational decision analysis, which simply com- that some trials might be acceptable in one place bines utilities and probabilities, and an ‘insurance’ (e.g. the developing world), which would not be analysis, which encapsulates the view that some acceptable elsewhere (e.g. the industrialised West). risks are simply not worth taking, whatever the However, both references argue for ‘trans-cultural’ expected benefits. The latter, known formally as ethics which take into account the norms of the minimax method, attempts to minimise maxi- different societies (or groups). In addition, some mum expected loss. We would argue that the references sought to quantify ‘minimal’ risk. rational (or Bayesian) analysis can incorporate Nicholson (1986) observed that, by US standards, an ‘insurance’ policy and that there is a small minimum risk of death is < 1 per million, a probability of severe outcome associated with minimum risk of major complication is < 10 per everyday action, so that, by minimising it, most million, and minimum risk of minor complication action would be ruled out anyway. Furthermore, is < 1 per 1000, and, by UK standards, minimum the values assigned to outcome are an integral risk of death is 1–100 per million, minimum risk of part of the concept of equipoise (Levine and major complication is 10–1000 per million, and Lebacqz, 1979) – see above. minimum risk of minor complication is 1–100 per 1000. However, Freedman (1993) stated that such Berglund (1995), Byar et al. (1990), Cassel (1985), decisions should not be derived quantitatively but Feenberg (1992), Gillet (1994), Goldberg and rather on a categorical basis, since risks associated McGough (1991), Hutchins and Eccles (1996), with new experiences (rather than with what Kauffman (1994), Lebacqz (1983), and Levine actually happens in everyday life) are, by definition, and Lebacqz (1979) state that probable benefits not fully known. We do not think that this analysis must outweigh the risks and that the greater the is applicable to therapeutic research where expected benefits, the greater the acceptable risk equipoise implies that the risks and benefits of harm, thereby advocating a decision analytic balance out in prospect. approach to determining collective equipoise when there are trade-offs. However, they do not More than minimum risk was also considered provide any definitive standards, and Nicholson to be inappropriate for ‘incompetent’ patients in (1986) worries that participants in trials where general (Ackermann, 1994; MRC, 1991; Prentice they can expect considerable benefit may be less et al., 1993). protected than in other trials which offer only small (or no) expected benefit, such as Phase 1 trials. Gillet (1994), Mariner (1990) and Tauer (1994) (See also page 16 on the risks associated with also discussed the level of risk acceptable for non-participation in trials.) healthy volunteers where any trade-off between 21 Review results

medical benefits and risk is questionable, even equipoise, ten articles argued that the individual when fully informed consent is obtained. Here, physician must also be in equipoise for an RCT to we are talking about non-therapeutic research in be ethical (Angell, 1984; Fried, 1974; Gehan and the sense that no clinical benefit is anticipated. Freireich, 1974; Gore, 1994; Hellman and Hellman, However, it could be argued that altruism could be 1991; Hill, 1963; Korn and Baumind, 1991; Lilford factored into the decision analysis, so that and Jackson, 1995; Peto et al., 1976; Van der participation is not ipso facto irrational. Linden, 1980).

Bruera (1994) examined the ethical basis of Three references argued for a ‘range of equi- research in palliative care where expected benefits valence’ from a statistical standpoint or ‘equiphase’ are limited. Here, the maximum possible loss from a psychological perspective such that small might be small whereas benefit could be long- preferences (within set limits) do not impact on lasting and hence large, but, despite this, he the decision to offer trial entry (Freedman and did not believe it ethical to impose greater than Spiegelhalter, 1992; Freedman et al., 1994; a ‘minimum risk’ because of the greater vulner- Chard and Lilford, 1998). ability of such patients. The reverse argument seems more plausible to us; that is, if there is so Four references raised the objection that equipoise little to lose and so much to gain, a new inter- could not account for individual patient prefer- vention, even if risky and/or relatively untested, ences or utilities (Angell, 1984; Fried, 1974; might seem attractive to a rational, but Kennedy, 1988; Weinstein, 1974), However, 34 desperate, person. references explicitly included patient preferences within their concept of equipoise. Twenty-eight Hammett and Dubler (1990) suggested that articles thought that this should take the form of ‘minimum’ risk should be applied to research in utilities (as exemplified in decision analysis) result- prisons given the potential coercion implicit in the ing in ‘effective equipoise’ or the use of preference environment from the institution, while Lasagna trials – where people choose treatment A, treatment (1970) supported research in general, given that B, or a trial of A versus B, but where outcomes are participation may of itself be beneficial in breaking collected from all eligible patients (Baum, 1994; the monotony of prison life. Botros, 1990; Bradley, 1993; Brewin and Bradley, 1989; Chard and Lilford, 1998; Dudley, 1986; One article suggested that individual physicians have Emrich and Sedrank, 1996; Ganz, 1989, 1990; a duty to exclude individual patients if they are at Hilden and Habbema, 1990; Institute of Medical risk from research and not rely solely on the eligi- Ethics Working Party, 1992; Kadane, 1996; Kassirer, bility criteria of the trial (Weijer and Fuks, 1994). 1983; Katz, 1981; Lebacqz, 1983; Lebacqz and Levine, 1977; Lilford and Jackson, 1995; Lilford It seems to us that this discussion of acceptable and Thornton, 1992; Oettinger and Berger, 1989; level of risk is germane only to non-controlled Schaffner, 1996; Silverman and Altman, 1996; studies (typically Phase 1 and 2 trials), but not to Stirrat et al., 1992, Sutherland et al., 1994; Veatch, prospective comparative studies (typically Phase 3 1981; Weinstein, 1974; Wennberg, 1990; William- trials), for example RCTs which are designed to son, 1996). However, one additional reference compare two (or more) treatments or treatment suggested that data from preference trials are with a placebo. Here, it is the balance of cost/ uninterpretable (Kadane, 1996), because the benefit that it important. Considerable risks from people who accept randomisation may be syste- a treatment are justified if considerable gains are matically different from those to whom the result equally likely in prospect – hence the need for may be applied. Presumably this problem increases decision analytic approach, at least to provide a as the plausibility of a link between preference and theoretical underpinning to the debate. We believe outcome rises. We offer the observation that where this approach is applicable to both competent this is plausible, ‘feedback’ trials or many trials adults who can give consent and to more vulner- (where preferences can change according to able groups whose proxies may give consent. accumulating prior evidence), offer advantages At any rate, by the time a Phase 3 trial is approved, because it is then possible to test the hypothesis that then the level of risk will have been set by preced- effect size is stable across the factors that may lead ing Phase 1 and 2 trials. (See also randomisation to expressed preference (or lack of preference). from the ‘first’ patient, on page 25.) Four references regarded informed consent as Patient equipoise. Once a trial has been sanctioned sufficient evidence of patient involvement in the 22 by an ethics committee on the basis of collective decision-making process (Alderson, 1996; Health Technology Assessment 1998; Vol. 2: No. 15

Coulehan et al., 1985; Curran, 1979; Miller and This technique involves assigning a treatment to Appelgate, 1985). the clinician, according to the clinician’s preference. The patient’s treating clinician is thus out of When equipoise does not apply. Twenty-six equipoise, though a third-party clinician who is in articles were concerned with the ethics of equipoise randomises and subsequently refers conducting an RCT when the individual physician patients to the treating clinicians. However, this is not equipoised. Bartlett and Cornell (1991), implies that allocation to treatment group takes Beecher (1966), Challah and Mays (1986), Freund place before the patient has seen a responsible (1970), Hill (1963), Lilford and Jackson (1995) and surgeon/physician and this is rarely the case except Markman (1992) thought that individual equipoise in a healthcare system in which patients have no is not always a necessary condition for the ethicality choice of where and by whom they are treated of the RCT or, more specifically, in circumstances (Gross, 1993). Also, such a study may not answer where a preferred experimental drug is restricted, the relevant scientific question, especially if i.e. available only within a trial. As Lilford and surgeons using a new experimental technique Jackson (1995) point out, this has intuitive appeal are few and highly (self-) selected. (See also when resources are scarce. It may be inappropriate randomising the first patient, on page 25.) to ask individuals to be the principal gate-keepers of medical developments, thereby making it accept- Four articles examined the ethics of randomis- able to restrict resources in this way – randomisation to units other than the individual patient ation may thus act as a justifiable form of distri- or physician, i.e. cluster randomisation, and butive justice. For example, there may be doubt as acknowledged that this may entail larger patient to whether or not putative benefits are sufficient to sample sizes (Fetter et al., 1989; Kramer and Sha- outweigh the societal costs of a new treatment, as in piro, 1984; Pocock, 1985; Pringle and Churchill, the ECMO trial. In such cases, an RCT may still be 1995). Randomisation of clusters is scientifically ethical, in spite of patient equipoise, if the central desirable when there is a risk of ‘contamination’ authority proscribes the new treatment outside a between arms of the trial. Pocock (1985) further trial. Put another way, the RCT in these circum- discussed randomisation in unequal ratios to stances provides the best hope of getting the groups when standard treatment effects are well preferred therapy. (See also the risks associated known or ethical concerns warrant more subjects with new therapies, on page 15.) receiving a new treatment (see above). Discussion in the literature surrounding the rationale for, and What should an individual clinician do when a the ethical implications of, conducting cluster preferred treatment is not restricted but when randomised trials is very thin on the ground and individual equipoise is seldom present? Gore this has prompted further discussion elsewhere (1995), Peto et al. (1976), Sposto and Krailo (1987) (Edwards et al., in preparation), though we pursue suggested that randomisation in unequal ratios it here in the discussion section. might be ethically acceptable. The ratio is set at the start of the trial, in accordance with individual n of 1 trials preferences, and remains constant throughout There were 11 articles which examined n of 1 trials, (even though preferences may not). Another of which ten saw them as intrinsically ethical when method is the adaptive design which might start there is doubt over the generalisability of results with the ratio 50:50 (on the basis of widespread from large trials or when there are no data avail- equipoise), and then change according to any able (Eick and Kofoed, 1994; Guyatt and Jaeschke, evolving treatment difference with accumulating 1990; Guyatt et al., 1986, 1990; Jaeschke et al., 1990, data. This is said to minimise the tension between 1991; Johnannessen, 1991; Johnannessen et al., individual and collective ethics during the course 1991; Larson et al., 1993; Spiegelhalter, 1988). of the trial. (See also page 17, on the legacy of The exception stated that the widespread use Hippocrates, and group randomisation, below.) of n of 1 trials could not be advocated as they However, Lilford and Jackson (1995) argue that are largely uncontrolled by ethics committees altering the randomisation ratio does not avoid (Mahon et al., 1995). reduced expected utilities for individual trial patients (see above). Placebo-controlled trials and alternative treatments Another idea to cope with different prior beliefs Eighteen references stipulated that there must be is the use of physician-centred randomisation no better alternative outside a trial for equipoise to (Korn and Baumind, 1991), especially in surgery be valid, that is, so that the patient does not lose (Rudicel and Esdaile, 1985; Van der Linden, 1980). out, in prospect (Anderson and Baden, 1971; Block 23 Review results

and Elahoff, 1979; Brown, 1980; Byar, 1990; Cavan, Waterlet, 1989; Geller and Pocock, 1987; Goldman, 1981; Clarke, 1993; Cleophas, 1995; Coulehan et al., 1987; Hughes, 1993b; Klimt and Canner, 1979; 1985; Fetter et al., 1989; Levine, 1985, 1991; Mar- Ruse, 1988; Seibert and Clark, 1993; Sutton-Tyrrell quis, 1983; Mike, 1989b; Moore and Papp, 1996; et al., 1991; Working Group on Arrhythmias of the Rothman, 1996; Rothman and Michels, 1994; Shaw European Society of Cardiology, 1994), or an and Chalmers, 1970; Spiro, 1986). This implies that improbability of achieving a conclusive outcome placebos should be used only when there is no (Ashby and Machin, 1993; Klimt and Canner, 1979; ‘standard’ effective treatment already available, Meier, 1979; Working Group on Arrhythmias of notwithstanding arguments that ‘absolute’ efficacy the European Society of Cardiology, 1994). While from placebo-controlled trials is scientifically more Fleming and Waterlet (1989) proposed that the valuable than ‘relative’ efficacy. Four references same criteria for early stopping should be used maintained that the placebo-controlled trial is for both positive and negative trials, Berry (1988) ethical even in the face of existing therapies since it thought that asymmetrical stopping boundaries yields the most reliable data (Collier, 1995; Glasser which pick up toxicity earlier than efficacy were et al., 1991; Katz, 1981; Wilhelmsen, 1979), while more appropriate. another asserted that placebo-controlled trials are scientifically most valuable but argued that the Censure of interim analysis problem of withholding (possible) effective Twenty-nine articles acknowledged that equipoise treatment can be circumvented by the use of might be dispelled during the course of the trial, dose–response studies wherein the control patients should the results be made available, thereby losing are given a very low dose of the experimental the ethical basis for further recruitment. We point treatment which is claimed to be tantamount to out that an individual physician, having lost his/her placebo in terms of effectiveness and side-effects equipoise cannot then justify recruiting a fixed (Freston, 1986). However, this argument is sample by claiming that he/she is still ‘uncertain’, fallacious because, if a very low dose of a potentially where ‘uncertainty’ is defined as anything short of effective treatment is, indeed, tantamount to statistical significance from the RCT in question – receiving a placebo in terms of therapeutic benefit, this would make the definition of uncertainty a then we are still guilty of withholding effective circular one. Such justification is only valid if, and treatment (standard or otherwise), only this time only if, we endorse the premise that knowledge is we are withholding a dose sufficient to produce dichotomised into certain or uncertain and that the any benefit. We would argue that placebo- only route to certain knowledge is via the RCT in controlled trials are only more valuable scientific- which a clinician is engaged. ally (in that they would ask the relevant clinical question) if it subsequently transpired that the As a result, it is widely accepted that interim data standard treatment is de facto worse than nothing should be censured. Indeed, 17 of the above at all. There is no intrinsic scientific reason why articles urged for censure of preliminary results placebo-controlled trials should be more such that physicians would remain ignorant of the ‘reliable’, than trials of new versus standard. findings until the trial had stopped (Burkhardt and Kienle, 1978; Byar, 1990; Chalmers, 1991; Chalmers Early stopping in relation to trial participants et al., 1972; Clayton, 1982; Editorial, 1983; Geller Data accumulates during a trial and, if it favours and Pocock, 1987; Hellman, 1979; Marquis, 1983; one of the comparator treatments, then, at a Peto et al., 1976; Pocock, 1977, 1989; Pollock, certain point, it may be deemed inappropriate to 1993; Prestifilippo, 1993; Royall, 1991; Shaw continue randomising patients. There is a balance and Chalmers, 1970; Shimm and Spece, 1983). here between obligations to trial participants and Eleven articles perceived a duty, on the part of the society at large (see page 17, on the legacy of physician, to discover any information which might Hippocrates – the therapeutic obligation). When make a difference to his/her decision to recruit a trial is stopped ‘early’, the beneficial treatment patients (Clayton, 1982; Freedman et al., 1984; may be made widely available, so that patients who Freedman, 1987; Gifford, 1986, 1995; Hutton, would otherwise have been allocated to the trial 1995; Kadane, 1996; Kennedy, 1988; Lantos, 1994a; control treatment would be provided with the Mike et al., 1993; Schaffner, 1996), but we point out superior therapy (Ruse, 1988). However, the result that this implies that the physician is free to seek may be less persuasive than it would otherwise out such information as may be available – it would be since it is of a lower statistical precision. Other exclude results revealed only to a DMC. Schafer possible reasons for stopping early are: the exist- (1985) suggested that patients could consent to ence of a harmful side-effect (Ashby and Machin, having preliminary evidence withheld from them. 24 1993; Berry, 1988; Bulpitt, 1983; Fleming and However, we point out that ‘voluntary ignorance’ Health Technology Assessment 1998; Vol. 2: No. 15

draws on Gerald Dworkin’s work where ‘consent’ Discussion would become an autonomous expression of a This review of the literature on the ethics of second-order preference and this overrides a less conducting RCTs has provided a summary of the important first-order preference (to obtain all various arguments and assertions surrounding the details about the intervention before making a issue of uncertainty as justification for trials. We can decision). However, in this context, the patient see that, once put into context, some arguments could only be said to ‘assent’, not consent, to are unsustainable while others generate healthy ignorance of preliminary data, for open disclosure controversy. We have tried to provide a concise (in Dworkin’s language, the first-order preference) account of the topic, but have elucidated points is not an option (unless the trial design is a ‘feed- where appropriate. back’ one). The only choice to be had for the patient is to enter a trial on the understanding Moral theory and RCTs that preliminary information will be withheld The RCT is not always scientifically necessary or or to refuse entry altogether. even desirable, but is most useful when expected treatment effects are small, yet worthwhile, due Another option is to conduct a ‘feedback’ trial to its unique capacity to minimise potential bias. where preliminary data (though not treatment A common criticism, however, is that any RCT rests assignment) is disclosed (or a series of small fixed- exclusively on utilitarian justification and is blind size trials) and hope that, while some individual to the interests of current patients who participate clinician–patient pairs are moving out of equipoise in them as a result. However, utilitarianism is not (as a consequence of seeing the results to date), the only theory from which RCTs can be defended. others will move into equipoise (Lilford and Jack- Kantians recognise that individuals have duties to son, 1995). That said, feedback trials may be society. Inviting individuals to make sacrifices from problematic if recruitment is thwarted by their use. good motives need not be exploitative. The usual reaction against patients being ‘used’ is that they Randomisation from the ‘first’ patient are invited to participate and enrolled only with Seven articles insisted that the ‘first patient’ should informed consent. Hence Kantians should have be randomised in order to exploit the ‘window of no difficulty with the ethics of RCTs as such. But opportunity’ where equipoise applies (Chalmers, RCTs using proxies to consent are another matter, 1968, 1975a, 1976, 1982; Chalmers et al., 1972; unless the proxy has been chosen by the patient. Boissel, 1989; Ingelfinger, 1972). On the other hand, nine articles saw some preliminary evidence Uncertainty as justification as necessary to test the feasibility of, and provide In this context, ‘equipoise’ is more meaningful the rationale for, a large trial (Cowan, 1981; than uncertainty (Lilford and Jackson, 1995). Freireich and Gehan, 1979; Hollenberg et al., 1980; ‘Uncertainty’ is ambiguous in two respects: firstly, Levine, 1991; Schafer, 1985; Spicker et al., 1988; knowledge comes in degrees and therefore uncer- Steinberg, 1991) or thought that randomising the tainty, used as the opposite of certainty in common first patient was not feasible due to the number of parlance, includes many possibilities (a concept experimental treatments being offered (Feinstein, represented mathematically by a probability 1973) or because the first patient is difficult to distribution); secondly, in circumstances where a define (Schafer, 1985). Twelve articles were known side-effect of treatment must be traded-off concerned that the evolutionary nature of a new against possible benefits, uncertainty may relate, operation was incompatible with randomising the not only to the prior probabilities of the benefits, first patient because self-perception of skill is an but also to how they are valued. Uncertainty important factor in determining equipoise therefore means different things depending on (Anderson, 1980; Bonchek, 1979; Brown, 1980; context and its use amounts to, what philosophers Byar, 1983; Gross, 1993; Leading article, 1985; call, equivocation. Equipoise, on the other hand, Neugebauer et al., 1991; Pollock, 1993; Rudicel implies that expected size and probability of and Esdaile, 1985; Shinebourne, 1984; Stirrat et al., improvement balance the size and probability 1992; Takaro et al., 1976), while two references of side-effects (perceived risks) of comparator thought that the ‘learning curve’ could be treatments; in decision analytic language, the incorporated into the RCT (Report to the Health expected utilities of the comparator treatments Services and Public Health Review Board, 1993; are equal and the participant does not lose out in Spodick, 1982) or that higher quality informed prospect. It has been argued that, while equipoise consent is appropriate during the early days of has a precise and unambiguous logical meaning, such trials. (See also page 23, when equipoise the concept is elusive in practice since its measure- does not apply.) ment is inevitably imprecise. However, it is our 25 Review results

opinion that it provides a clear goal to aim at in whether a clinician should be bound by individual contrast to the ambiguous term ‘uncertainty’. or collective equipoise. Once the premises are Indeed, it has been proposed that, psychologically, accepted that (a) equipoise is morally important, though not mathematically speaking, we function and (b) equipoise turns not just on probabilities, on the basis of categories rather than points such but individual values as well, and (c) values are that it would be ethical to recruit a patient on the likely to vary from patient to patient, the issue basis of ‘equiphase’ (Chard and Lilford, 1998); becomes one of whose prior probabilistic belief that is, where the decision is robust to small should be disclosed. This question can be decon- changes in expected utilities. We would argue structed into whether a clinician should give a that a doctor would be happy to be randomised personal prior probability estimate, a corporate him/herself. prior probability one, or both, though we recognise that corporate beliefs are seldom measured and Further argument concerns the moral significance might fluctuate in any case. of different types of equipoise – collective and individual. Freedman (1987) has argued that Under-powered trials equipoise among relevant clinicians (collective Recognition of the central role of equipoise has equipoise) is sufficient to make a trial ethical, even some radical implications. Firstly, if equipoise is if the individual clinician has a preference. Others accepted as the moral basis for trials, then ‘under- point out that patients may wish to know what their powered’ trials are not necessarily unethical in care-giver thinks or to make up their own mind on their ‘use of subjects’, although they may offer the basis of such evidence as may be available. In poor value for transferable resources. Detailed the usual situation where trade-offs are concerned discussion on this topic can be found elsewhere (see above), equipoise turns on values as much as (Edwards et al., 1997). prior probabilities, and so it cannot be determined outside the consulting room. Equipoise is primarily Deciding when to stop a property of the (competent) patient, rather than Just as there are moral complications associated of the doctor. As individual (patient) equipoise is with recruiting too few patients in under-powered the ethical basis of most trials, the invitation to trials, so there are with recruiting too many. participate (informed consent procedure) is Some trials are approved simply in the hope of therefore critical – see below. persuading practitioners to take up ‘proven’ effective treatments on the basis of greater statis- One can speak meaningfully of collective tical precision. However, in the case of ISIS-2, equipoise only in a situation of probabilistic previous results had not even been published in dominance (i.e. a treatment has no side-effects), standard texts, thereby cultivating a culture of because the only valued outcome of relevance is ignorance (Altman and Gardner, 1992). While that which is being measured in the trial, for emphasis must be placed on dissemination and example if more lives are saved with treatment A, education through collaborative effort (e.g. it is to B and vice versa. In the perhaps more usual Cochrane Collaboration and York Centre for situation where more than one outcome is at stake Reviews and Dissemination), there remains the and where some knowledge about the effect of problem of when investigators should stop a trial. treatment (i.e. where side-effects are involved), we Beneficial treatments need to be made widely argue that it is not meaningful to speak of collective available as soon as possible and it could be argued equipoise, unless the patients are not ‘competent’ that trials should be stopped at the earliest oppor- and an ‘average’ value set must be assumed. As tunity even before the ‘full’ sample has been soon as more values come into the frame (and the accrued. Trials might be stopped for other reasons patients are competent), it is reasonable to think in such as evidence of toxicity or a prediction that a terms of collective prior belief about probability, conclusive result will not be forthcoming. The but not collective equipoise for the latter would question of when to stop a trial gives birth to a assume that we know the patient’s values even number of other questions, for example deciding before she has been asked. The notions of collec- which stopping rules should be used, and if and tive equipoise/prior belief throw up two interesting how external data should be incorporated. A funda- issues which we discuss with reference to the liter- mental problem remains however: any trial has the ature: (1) what to do if the ‘vote’ is not split 50:50, potential to destroy the equipoise that it relies on but 80:20, say. We discuss the single paper which to be ethical. Furthermore, enthusiasm to stop has sought public opinion on how much ‘expert’ trials early must be tempered by an awareness that consensus is needed to make a trial ethical from the medical community may remain unconvinced 26 the point of view of an ethics committee, and (2) in light of potentially misleading data and may call Health Technology Assessment 1998; Vol. 2: No. 15

for replication if equipoise persists. Commonly, it is Replication the DMC that must decide if and when to halt (or It is important that the research question has at least recommend halting) recruitment. Referring not already been answered and ethics committees physicians are kept ignorant of any preliminary should avail themselves of a relevant systematic data as they are bound by the therapeutic oblig- review to ensure that the protocol in question is ation, whereas the DMC is at liberty to take account sufficiently original. Research that has been carried of societal interests in addition without violating out before may not be interesting even scientific- any obligations to the individual participant. ally, though conversely some would argue for However, it could be argued that this solution is too repeated attempts at refuting hypotheses in the contrived and that simply pretending that data do Popperian tradition. By using Bayesian statistics, not exist in order to maintain equipoise may only however, issues of repetition only arise where there be acceptable if there is no better alternative, that is some doubt over a previous result. is, if it is the least worst option. ‘Feedback’ trials (where preliminary data, though not treatment The null hypothesis assignment, is disclosed) are one such alternative, The null hypothesis would seem a poor basis for but there has been little practical experience with statistical tests in (the frequent) circumstances these hitherto. Indeed, many see the prima facie where trade-offs must be made due to side-effects advantages of another design, called ‘adaptive’ or ‘costs’. Under these circumstances, participants designs, according to which the ratio of random- in the trial would have had prior expectation of isation changes during the course of the trial along benefit sufficient to compensate for the expected with how the preliminary results look. However, costs, if they were entered ethically. The most adaptive designs are not a perfect solution to the interesting hypothesis is therefore whether or problem since they do not necessarily maximise not the data are compatible with this expected individual expected utility – this would be achieved benefit – the null hypothesis is not then the simply by opting for the therapy showing the most default position. promise of the moment. Controversially, we favour feedback trials because: (1) we do not think that it Informed consent is fair either on DMC members or the public to The invitation to participate, or (informed) weigh societal interests against those of future consent, is fundamental to the ethics of conducting potential participants behind closed doors and trials. Numerous authors point out that, for con- (2) it is inappropriate to dichotomise the decision sent to be valid, it must be competent, voluntary to continue or terminate the trial. Again, the (not only technically but also in spirit), and all decision is personal, as it turns on individual relevant information must be divulged. The values and how those interact with probabilities requirement that consent be informed applies to (Thornton and Lilford, 1996). Consequently, research subjects, but not whenever consent is people with a value system that may have made sought, say in routine practice. This tradition runs them ineligible for trial entry at the outset may counter to the Hippocratic one which is essentially wish to participate on the basis of interim data paternalistic. Paternalistic choices are ones that and vice versa. override otherwise autonomous decisions (whether or not information is disclosed), whereas a proxy Placebo-controlled trials decision is made in the absence of competence. It is unethical to conduct placebo-controlled trials However, it is now widely accepted that a (compe- in the face of a known advantageous treatment tent) person is in the best position to know where (Rothman, 1987), because equipoise cannot be his/her personal best interests lie and, in any case, present. We would argue that this only applies for that his/her autonomy should be respected. Since treatments that would otherwise be freely available. patient equipoise turns on personal values, its If a technology were not generally available, say presence or absence can only be determined because the local health service does not have through detailed discussion. Some authors add that enough money, then randomisation may offer the clinical investigators have a potential conflict of best opportunity to get the preferred therapy. interests (between the individual and society) and, There is still the issue, however, of what to do once by eliciting the patient’s own values, the risk of this the trial has come to term. A particular issue arises becoming a real conflict is minimised – ‘full’ when a trial is funded from a richer country to be disclosure acts as a safeguard against overzealous hosted by a poorer one. Here, the above logic holds scientists. Many authors have argued that the goal good, with the proviso that investigators ensure of disclosing all relevant information is imperfectly such research is not perceived as exploitation realised in practice, but we still have to decide what (see below). ought to be done. There are nevertheless many 27 Review results

situations where it may be argued that the goal consent typically would only be elicited from of full disclosure should be relaxed. individuals in the intervention group.

When consent cannot be obtained. One such Behavioural interventions and consent. The goal special circumstance arises when the potential of full disclosure may sometimes undermine the participant is not competent because he/she is objective of a trial, whether randomised at the unconscious due to cognitive impairment. In these level of the individual or cluster. When partic- circumstances, a surrogate (patient, partner, etc.) ipants cannot be blinded, knowledge of the is the best source of values for we might (or insofar treatment comparison can introduce bias through as we might) reasonably assume parity of interests differential changes in the behaviour or attitudes and he/she may have insight into what the patient of the treatment groups. This situation often would have wanted to boot. The patient is not occurs with education interventions which are thereby excluded from the opportunity to help typically evaluated by clusters to avoid the others. However, in the absence of a surrogate or aforementioned problem of contamination. in an emergency, some have argued that random- Knowledge of the intervention amongst control isation should be avoided (Kapp, 1994), notwith- subjects can lead them to seek the intervention standing the large societal costs of doing so. outside the trial, reducing the difference observed Others, including the MRC, the Royal College of between groups, or resulting in demoralisation, Physicians, and the FDA, have argued that, since which is likely to depress outcomes for the controls normal practice must proceed on the basis of and result in an exaggerated difference between ‘assumed’ preferences (i.e. those of the ‘average’ groups. A randomised consent design can be used patient) or ‘implied consent’, randomisation may in these situations, although it has been criticised. be ethical, but only if clinicians are equipoised with Here, randomisation is carried out prior to seeking respect to their best guess as to where the patient’s consent and consent is usually sought only for the best interests lie. In some cases, it is possible to treatment to which an individual is allocated, that elicit consent in advance of a situation which may is, without disclosing the treatment comparison. arise some time in the future. For example, a Another alternative is to elicit consent to take pregnant woman may give consent for scenarios part as a control without fully divulging the that may arise in labour or the early neonatal nature of the intervention. period. In addition, the presence of a proxy may serve as a further safeguard. Can consent be needlessly cruel? A somewhat different situation arises when it is the fear of Cluster RCTs and consent. The best interests of the causing distress, rather than logistics which may individual may be served by trial participation even inhibit free disclosure. Indeed, on occasion, there in the absence of competent consent where the may be a trade-off between beneficence and unit of randomisation is the group, for example, a autonomy (see below). In the West, the obligation school or general practice, instead of the indi- to respect autonomy is afforded paramount vidual. If the intervention is likely to be considered importance and doctors are increasingly opting for risky by the community, it may be appropriate to ‘full’ disclosure on the grounds that patients widen such surrogate consent. A decision not to cannot choose between options without knowing seek individual consent may also be taken for what these entail. It has been cogently argued by logistical reasons when an intervention is delivered Dworkin (1988), however, that the option to to a defined community or geographic area. In relegate the decision remains a valid one; that is, such circumstances, a surrogate in the form of a the patient may ask the doctor to choose on cluster ‘guardian’ (a head teacher or general his/her behalf. practitioner) is responsible for determining what is in the best interests of the group and can consent Where equipoise need not apply to participation and randomisation. Not all cluster There may be circumstances where equipoise need trials preclude individual consent to the inter- not apply for a trial to be ethical. When a preferred vention in question – we distinguish between trials experimental treatment is restricted and is only conducted out of pragmatically necessary (for available within a trial, then the only route to the example, fluoridation of water) and those con- preferred treatment is through randomisation. ducted for scientific reasons (for example, to avoid This is intuitively appealing when resources are contamination) where the unit of randomisation is scarce, but not when access is not limited by a the cluster, yet the intervention is implemented at central authority. Of course, there are wider issues the level of the individual. The latter category concerning distributive justice and, given a specific 28 includes many educational interventions where healthcare need, randomisation may indeed be a Health Technology Assessment 1998; Vol. 2: No. 15

fair form of triage. Even without resource limit- trial when published and regulatory checks on the ations, it may be inappropriate to ask individual trial process should be made throughout. practitioners to be the principal gate-keepers of medical innovations and it may therefore be ethical Conclusion for a central authority to proscribe the use of new We conclude from the above that patient equi- treatments outside a trial, pending a license or a poise is a necessary basis for the ethical conduct legal purchasing decision. In this way, the decision of RCTs and, as equipoise should include patient’s about which treatment is ‘best’ can be hedged, values, it can only be determined through detailed given the existence of collective equipoise, and discussion. In this way, practitioners might respect the narrow window of opportunity can thus a patient’s autonomy in the spirit of Kantian ethics. be exploited. That said, there are situations where informed consent should be relaxed, though in cases of However, some have argued that in some cases, incompetence a proxy should be consulted. it is unethical to restrict potentially life-saving Patient equipoise should not be seen as the only treatments to trials when there is no effective safeguard however, and is no substitute for ethics standard (Minogue et al., 1995). The ECMO and committee approval. AIDS trials are examples thereof. Minogue et al. (1995) and Truog (1992) argue that randomisation is unethical, in such cases, and that observational What are the physical effects of studies are desirable. Another option would be to participating in trials? use the Zelen design, and so avoid potential distress to controls, though this is highly controversial; Results anticipation of distress was the rationale for using Type of non-trial control this design in the Harvard ECMO trial, but the Thirteen of the 15 studies used non-randomised hospital’s Institutional Review Board (IRB) concurrent controls. The exceptions used random- received a rare reprimand from the National ised concurrent controls (Mahon et al., 1996) or Institutes of Health (Truog, 1992). Such distress historical control data from disease registers was, indeed, demonstrated among controls in a (Reiser and Warner, 1985). Of those studies using later British trial where pre-randomisation consent concurrent non-trial controls, one study, Mahon et was obtained, but where availability of ECMO was al. (1996), was actually a randomised trial of doing restricted to the treatment group (Snowdon et al., or not doing n of 1 randomised trials, and was, 1997). Others argue for the implementation of therefore, fundamentally different. Three studies preference trials in which patients receive their used patients who had been offered trial entry but treatment of choice (see patient equipoise, who had denied consent (Antman, 1983; CASS page 22), though the data from such trials may Principal Investigators and their Associates, 1984; be difficult to interpret (Kadane, 1996). Indeed, Schmoor et al., 1996), while the remainder It could be argued that patients have a ‘consumer’ obtained non-trial controls who had not been right to use whatever therapy they like (regardless offered trial entry. One study compared (mainly) of whether or not they can pay for it). However, trial and concurrent and historical non-trial not all demands could be met and resource distri- patients by district of residence (Karjalainen and bution, in one form or another, is an inevitable part Palva, 1989). of health care. Indeed, Logue and Wear (1995) has argued that restricting strongly preferred treat- Type of therapy ments to trials does not impinge on a patient’s Ten articles involved cancer therapy and examined autonomy, since it merely limits the options open the effect of trial entry on survival/disease-free to the patient and is not coercive as a result. survival (Antman et al., 1983; Bertelsen, 1991; Davis et al., 1985; Karjalainen and Palva, 1989; Lennox Compensation et al., 1979; MRC Working Group on Leukaemia in At the moment, for a research patient to gain Childhood, 1971; Schmoor et al., 1996; Stiller and compensation for some injury due to participation Draper, 1989; Stiller and Eatock, 1994; Ward et al., in a trial, he/she will have to bring an action in 1992), of which four (Lennox et al., 1979; MRC negligence. We believe that informed consent to Working Group on Leukaemia in Childhood, 1971; receive potentially dangerous therapies without the Stiller and Draper, 1989; Stiller and Eatock, 1994) promise of compensation in the event of injury may involved childhood cancer. Two articles addressed take away any obligation on society to compensate. the effect of clinical trials on survival from cardio- However, as a mark of respect, it could be argued vascular disease: the CASS Principal Investigators that patients should be offered the results of the and their Associates (1984) and Jha et al. (1996). 29 Review results

Two studies, Mahon et al. (1996) and Reiser Prognostic variables and Warner (1985), examined the effect of One study conducted a randomised trial of a n of 1 respiratory clinical trials on morbidity. One study trial, in other words, they randomised patients to was concerned with post-operative care (Williford take part, or not in an n of 1 trial (Mahon et al., et al., 1993). Thus, 12 of the 15 studies examined 1996). With this exception, the others cohorts of effects on survival. trial and non-trial patients (and could therefore have been confirmed by systematic differences Methods and interobserver agreement between people who were entered into trials The list of authors, outcome measures, description and those who were not). Five of these studies of non-trial controls, main results and whether compared trial patients with non-trial patients or not an attempt was made to measure and allow who had similar prognoses to those in the trial for any difference in prognostic variables are sum- (Bertelsen, 1991; CASS Principal Investigators marised in Tables 2a and 2b. The studies were also and their Associates, 1984; Schmoor et al., 1996; tabulated according to quality of evidence (see Ward et al., 1992), while Davis et al. (1985), Jha Tables 2a and 2b). There were two cases when the et al. (1996), Lennox et al. (1979), MRC Working observers (SE and JT) disagreed on an extracted Group on Leukaemia in Childhood (1971), Stiller data item, but both instances were resolved by re- and Draper (1989), and Stiller and Eatock (1994) scrutinising the original papers. One study was all attempted to make statistical allowance for such rejected by JT on the grounds that the n of 1 trial confounders, and, in each case apart from two itself was an intervention (Mahon et al., 1996); it (MRC Working Group on Leukaemia in Child- was included in the results tables for completeness, hood, 1971; Schmoor et al., 1996), the original but was excluded from the graph. Data sets from significant differences were maintained after these studies were then categorised (see Table 4a) appropriate adjustment had been made. Indeed, and listed (see Table 4b) by DAB with a view to in the trial reported by Lennox et al. (1979), the plotting the HRs for individual trials and strata difference in survival was even more pronounced within studies. after making this adjustment. In one study, the original statistically significant difference dis- Survival as outcome appeared after adjustment (Williford et al., 1993). In seven of the 12 studies addressing survival, Two studies simply failed to make a suitable adjust- survival was reported as being statistically signifi- ment for a known prognostic imbalance between cantly higher among trial participants than among the trial and non-trial groups (Antman, 1983; non-trial controls (Davis et al., 1985; Lennox et al., Williford et al., 1993), while the remainder could 1979; MRC Working Group on Leukaemia in not comment on the distribution of prognostic Childhood, 1971; Stiller and Draper, 1989; Stiller factors because of the informal nature of the non- and Eatock, 1994; Jha et al., 1996; Karjalainen and trial control group (Reiser and Warner, 1985) or Palva, 1989). Three of the five studies where the because survival rates were not compared on a case- result failed to reach statistical significance never- by-case basis – a district with a high recruitment theless reported a favourable trend in association rate to clinical trials was simply compared with a with trial entry (Antman, 1983; Schmoor et al., control district (Karjalainen and Palva, 1989). 1996; Ward et al., 1992), while the remainder did not find any noteworthy difference (Bertelsen, Trial effects 1991; CASS Principal Investigators and their In order to make possible a visual comparison of Associates, 1984). the various studies (Figure 3), we have extracted estimates, or data, from the published papers, Morbidity as outcome where possible. We have used published estimates Morbidity was included as an outcome measure and confidence intervals, or data, corresponding in three studies. In two studies, there was a signifi- to the most adjusted comparison reported by the cantly better outcome (statistically speaking) authors, but this has not been possible in all cases among trial participants than among non-trial (see Table 4b). Three of the 15 studies were exclud- controls (Mahon et al., 1996; Williford et al., 1993). ed from the graph because the authors had not The other study, Reiser and Warner (1985), showed performed appropriate statistical adjustment for an improvement in symptoms for patients both in known prognostic imbalances (Antman, 1983; the treatment and control arms of the three asthma Reiser and Warner, 1985; Williford et al., 1993) trials. The authors claim that this exceeded that and one because the trial itself was an intervention which would be expected outside the trial context – an n of 1 trial (Mahon, 1996). For Davis et al. according to the natural history of the disease but (1985), the HR (though reported as ‘RR’) and 30 had no formal non-trial control group. 95% confidence interval reported by the authors Health Technology Assessment 1998; Vol. 2: No. 15

Hazard ratio 1.8 Loss from being in the trial 1.6

1.4

1.2

1.0

0.8

0.6

0.4

0.2 Benefit from being in the trial 0 B K C D W J(i) J(ii) L(i) L(ii) Sc(i) L(iii) Sc(ii) MRC S94(i) S94(ii) S89(iB) S89(iA) S89(iC) S89(iiB) S89(iiA) S89(iiC) S89(iiiB) S89(ivB) S89(iiiA) S89(ivA) S89(iiiC) S89(ivC) Reference +ve trial result –ve trial result

Limited benefit from Substantial scope for benefit from trial protocol trial protocol

FIGURE 3 Median HRs (95% confidence intervals). The figure is discussed in full in the main text. It shows, with some exceptions, Bayesian credible intervals for the HR comparing those in a trial with those not in a trial. Keys to the labels are given in Table 4b. For all the intervals plotted, a ratio of less than 1 means an advantage to those in trials from a Cox regression has been plotted, while Jha HR that is stable over time will not lead to the OR et al. (1996) reported ORs (and 95% confidence or RR being stable over time. This is relevant here intervals), and we have been plotted those. In as we are comparing studies with very different both these cases, the estimates and intervals can probabilities of failure. Nevertheless, plotting a be interpreted essentially in the same way as the graph of ORs or RRs should lead to similar Bayesian intervals described below. For the remain- broad conclusions. ing studies, relevant estimates and confidence intervals were not given, so we have had to ‘extract’ Making the assumption that during the period of the numbers of patients at risk, and the number risk the instantaneous hazards are proportional and (or proportion) who survived etc (i.e. have not fairly constant, it is easily shown that the HR is failed, in statistical terms) for some exposure equal to the ratio of the logs of the proportions period. Where data were available for a variety of P not failing, that is, periods, a period with minimal loss to follow up and a failure probability of as close to 0.5 as HR = Ln(P \ in trial) / Ln(P \ not in trial) possible was selected. In many cases these numbers were estimated, for example by measuring from We have used the (Bayesian) software package published Kaplan–Meier survival curves. These BUGS (Gilks et al., 1994) to estimate 95% credible numbers can be used to estimate ORs, risk ratios intervals (CIs) for the HRs – which are plotted in (RRs), absolute risk differences, or HRs. The latter Figure 3. There is a 95% chance that the true HR seem to us to be more basic – more likely to be lies in the CI, given the modelling and prior stable – than the other measures. For instance, an assumptions. We assumed independent uniform 31 Review results

priors on the proportions failing – though Thus, D and W (see Table 4b or 4c for the key) have (limited) sensitivity analyses suggest the results been judged to have no prior effective treatment do not depend strongly on the priors used. For in the protocol, whereas those from Sc(ii) onwards Karjalainen and Palva (1989) we have plotted a do. Sc(ii), C and, to a lesser extent, B and Sc(i), (Bayesian) estimate and CI for the ratio of HRs are judged very likely to have only small differences (for trial versus non-trial areas), during and in outcome caused by differences in treatments or preceding the trials. regimens between trial and non-trial groups. By chance, all the data sets from D to Sc(i) had no For all the intervals plotted, a ratio of less than 1 significant trial effect, and were thought by DAB means an advantage to those in trials. To attempt to to be only moderately subject to bias. draw useful conclusions as to when a trial is likely to prove beneficial, we have categorised each data set A priori, one might hypothesise that data sets plotted according to four criteria: such as D–Sc(i) would show little or no difference in benefit between trial and non-trial patients – and • Was there included in the trial protocol a pre- this is broadly the picture, with the clear exception existing treatment already known (or presumed) of D. To us, the findings in D are sufficiently to be effective? surprising to warrant being labelled an outlier. • Was the impact of differences in the treatments, Although the authors have apparently adjusted and/or regimens, between trial/non-trial in their analysis for all relevant factors, it is possible patients possibly large (or very probably small)? to think of possible remaining prognostic biases • Did the trial show no advantage for the (albeit on a post hoc basis). It is also possible, as is ‘experimental’ (or ‘new’) treatment true for data set W (personal communication), (if applicable)? that the surgery, performed before randomisation, • Was there large scope for bias due to unadjusted was nevertheless performed in the knowledge differences in prognostic factors between trial that a particular patient was likely or not to be and non-trial groups? (Notes: (1) if the scope for randomised. This might conceivably have affected bias was thought very large, the data set has not the surgery, benefiting trial patients. Perhaps the been plotted in the figure; (2) in the remainder most likely explanation is a form of publication of this section, bias is used in this sense of bias – the authors were only prompted to unadjusted differences in prognostic factors undertake the trial/non-trial survival comparison between trial and non-trial groups.) after noticing the remarkably good trial survival rate. These inevitably somewhat subjective judgements were made before any systematic data ‘extraction’ Data sets J(i), S94(i) and S94(ii) all had significant or plotting. Although subjective, we believe it trial results in favour of the experimental treat- succeeds in making useful distinctions between ment. The S94 data sets were thought possibly data sets. For example, if trials have their putative subject to large biases. One might expect to see beneficial effect through better quality of care, some benefit to trial participants in these trials, then a difference in outcome is only likely to arise if only from the beneficial new treatment. The when an effective treatment is available. The cate- remaining data sets J(ii)–K all showed no ‘statistic- gories are set out in Table 4a, and given together ally significant’ advantage to the experimental with other relevant information relating to each treatment (where applicable), though J(ii) had a data set in Table 4b, which also includes the ‘key’ strong trend in that direction. J(ii)–L(iii) were to Figure 3. Table 4c contains all the data plotted judged moderately subject to bias, while MRC–K in Figure 3, as well as the data extracted from were thought possibly subject to large biases. These the studies. data sets are all ones in which one might expect some benefit to trial participants arising from the The data sets in Figure 3 have been ordered development and use of a protocol for trial patients according to the four criteria given above. The for pre-existing effective treatments. Broadly, it dashed vertical line divides data sets (to the left) seems that there is benefit to trial participants with apparently limited possibility of extra benefit under these conditions. Data sets S89(iA)–S89(ivC) to trial patients arising from the trial protocol, come from the same paper (Stiller, 1989), and from the remainder (to the right). Within the data are potentially subject to large biases. However, sets with potential protocol benefits to the trial any biases might be expected to be similar for participants, the dotted line separates those with all the S89 data sets. S89(i.) refers to the period positive trial results (the three to the left) from the 1971–1973, similarly ii, iii and iv refer to later 32 remainder with ‘statistically’ negative trial results. periods. S89(.A) refers to patients treated in Health Technology Assessment 1998; Vol. 2: No. 15

hospitals treating six or more acute lymphoblastic et al., 1979; Stiller and Eatock, 1994). The trial leukaemia (ALL) patients per year, similarly B and results in one study were statistically significant C to those treating 1–6, and < 1 ALL patients per which suggests that the significant difference which year. Hospitals treating many ALL patients are they observed between trial and non-trial patients probably more likely to have and use up to date could be partially attributed to the superior protocols for all ALL patients, not just those in experimental treatment (Jha et al., 1996). The trials. The benefit of being in a trial in an ‘A’ remaining studies did not supply data which might hospital might therefore be expected to be less make possible a comparison between the outcomes than that in a ‘B’ hospital. In fact, one would of trial versus non-trial control patients (Antman, expect to see a trend of increasing benefit from 1983; Williford et al., 1993). ‘A’ to ‘B’ to ‘C’ hospitals within each period. This is clearly seen here by visual inspection (and was, Excluded study of some relevance of course, noted by Stiller (1989)). In addition to the 13 studies which set out to investigate a possible trial effect for randomised/ Drawing firm, general, conclusions from Figure 3 non-randomised patients, there was another paper would be ill-advised, as the trials studied are not a (Olschewski et al., 1992) which did not seek to random sample of trials, may be subject to consid- establish a ‘trial effect’, although it did provide erable ‘publication’ bias, and the ‘data extraction’ some information on the survival of trial partic- and categorisation processes were far from perfect ipants (Coronary Artery Surgery Study) and similar (though the best that could be achieved with the non-trial patients. They merely reported that there resources available). We did not produce an was no significant difference in survival between overall estimate of effect, or conduct a quantitative trial and non-trial patients. ‘sensitivity’ analysis because, in our view, the data were not of high enough quality to warrant such Discussion precision. Nevertheless, we feel that Figure 3 lends Possible bias some support to the already plausible hypothesis The comparison of physical outcome in clinical that RCTs tend to be good for you if there is a trials, compared to that of non-trial cases, is pre-existing effective treatment that is included potentially confounded in many ways which in the trial protocol, or if it turns out that the could lead to a spurious ‘trial effect’. experimental treatment is more effective. The study of Reiser and Warner (1985), which did Outcome in trial control versus non-trial not use a control group running concurrently with control patients the trial, could be confounded by a trend with time Three studies (CASS Principal Investigators and in the condition of the children selected for the their Associates, 1984; Schmoor et al., 1996; Reiser trial – perhaps seasonal or due to regression to and Warner, 1985), further subdivided outcome the mean. Considerable bias is a risk with all among trial patients, so the control arm in the trial historical control studies (Sacks et al., 1982). itself could be compared to non-trial patients. Two of the three studies did not find any significant However, the use of concurrent controls (as in the difference between the two groups (who received remaining 14 studies) does not protect the analysis similar treatments) in terms of actuarial and 5 year from all possible biases. The underlying assumption disease-free survival (CASS Principal Investigators behind the use of concurrent controls is that the and their Associates (1984) and Schmoor et al. prognosis for the control patients is not systematic- (1996), respectively). The remaining study simply ally different from those in the study group, that is, inferred that the trial controls receiving a placebo those included in the trial. However, there are did better in terms of morbidity than non-trial important prognostic variables which may distri- patients (Reiser and Warner, 1985). Four studies bute themselves unequally between non-trial and (Davis et al., 1985; Ward et al., 1992) reported no trial patients, leading to a biased comparison. Put evidence of treatment differencefor the clinical another way, participants may tend to be healthier trials themselves, i.e. outcome for trial controls was than their non-trial counterparts and so have a not significantly different from that of the ‘experi- better prognosis. Eleven studies made an attempt mentally’ treated patients or that the experimental to adjust statistically for this potential bias, to a treatment was less effective than the standard greater or lesser extent, or observed that the trial (Karjalainen and Palva, 1989; Stiller and Draper, and non-trial groups were similar in important 1989). Four studies showed a non-significant respects. Even when formal statistical adjustment is treatment advantage for one of the treatments carried out (six studies) or when the comparator in a trial (Bertelsen, 1991; Jha et al., 1996; Lennox groups appear similar (five studies), important 33 Review results

biasing factors affecting outcome may remain, achieved significant outcomes after adjustment since it is not possible to record or measure all for prognostic variables), our more recent review, relevant variables and hence account for them in based on a greater number of papers, is broadly the analysis. Those who carry out randomised trials supportive of his conclusions. We feel that a and proselytise their advantages would be the first tentative conclusion can be drawn from Figure 3 – to concede that non-randomised concurrent that RCTs tend to be good for you if there is a controls are subject to bias, even after carrying pre-existing effective treatment that is included in out multivariate or other statistical analyses to the trial protocol, or if it turns out that the experi- allow for confounding variables. As a result, it is mental treatment is more effective. This latter now widely accepted that randomised clinical trials condition is, of course, hard to predict – though are epistemologically superior to both historical there may be historical evidence as to whether past and non-randomised concurrent designs. experimental trials in a particular area of medicine have tended to be effective or not (Gilbert et al., Where non-trial controls who were not offered trial 1977). In any case, it seems from Figure 3 that the entry are used, patients who refuse to participate former condition may be more important, and it is should be followed along with those consenting clearly one that trialists have some influence over. to trial entry. This is because refusers may have a For example, in trials of adjuvant therapy after different prognosis to acceptors, thereby biasing surgery for cancer, trialists might include the the comparison with those not offered trial entry. surgery in the protocol. By doing this they could Analysis should ideally be both by the offer of entry expect to improve survival of participants irre- (i.e. by ‘intention’) and by whether or not the spective of whether the treatment proved effective, patient participated. However, in four studies, the or which arm a patient was randomised to. This only control group was made up of people who might, however, cause a reduction in ‘generalis- denied consent (Antman, 1983; CASS Principal ability’ of the trial results in the presence of an Investigators and their Associates, 1984; Schmoor interaction (e.g. synergy) between pre-existing et al., 1996; Williford et al., 1993) and, in the treatment and the experimental treatment. It remainder, no attempt was made to differentiate would be a transgression of deontological ethics refusers from consenters among those offered to make a deliberate attempt to improve treatment trial entry. for trial patients above that available outside the trial by including an effective treatment in the In light of the methods used, we must not be protocol solely in order to improve outcome for overzealous when interpreting the data and the trial participants, rather than, say, to make the only way to resolve this issue completely would trial arms more comparable and hence increase be to randomise sufficient people to take part in statistical power. clinical trials or to receive non-trial treatment, as in the study of Mahon et al. (1996), who randomised Possible explanation of findings patients to n of 1 trials or standard practice and Even if there is no bias, a beneficial effect for trial observed their comparative physical morbidity. participants has two explanations: (1) the effect of Put another way, the perfect study design would a particularly successful intervention in the trial, be a randomised study of being invited or not to or (2) a ‘side-effect’ of the trial itself. The putative participate in a range of trials, themselves selected benefit is most commonly ascribed to a general (ideally) at random from all such trials. In short, improvement in care, resulting from attention to we are never likely to have unassailable evidence detail inherent in following the trial protocol (Haw- from prospective studies that large trials per se thorne effect) rather than because a proportion of provide physical benefit. patients receive superior treatment (Karjalainen and Palva, 1989). However, improved physical Summary of findings prognosis when new treatments are compared with Six of the 12 studies examining survival with old, would also arise if the former were systematic- concurrent controls suggested improved survival ally better than pre-existing alternatives. If this were for people treated within prospective trials. How- the case, then study patients in the trial should do ever, statistical adjustment for known and recorded better in terms of physical outcome than both the confounding variables was not possible in one case. trial controls and similar patients treated outside In the remaining five cases, improved outcome for the clinical trial. Such a comparison was possible trial cases remained (or was enhanced) after such in six studies but only two (CASS Principal Investi- adjustment. Although Stiller’s review produced a gators and their Associates, 1984; Reiser and higher proportion of significant findings in terms Warner, 1985) analysed study and control patients 34 of survival (6/8 compared to our 8/13 studies explicitly. The others simply reported a non- Health Technology Assessment 1998; Vol. 2: No. 15

significant treatment effect, thereby indicating It could be argued that because the apparent that the survival rate of the trial controls was benefit to patients of being in a trial is not intend- indistinguishable, in a statistical sense, from that ed, that is, it is merely incidental, there are no of the experimentally treated patients, or else ethical implications to the use of non-protocol reported that the experimental treatment was treatments outside trials. However, an ethical significantly less effective than the standard. The imperative (in the spirit of the Helsinki Declar- explicit three-way comparison by Reiser and ation) demands that physicians do their best for Warner (1985), showed that both study and trial each individual patient under the ambit of their controls did better than non-trial patients but this care, in which case, it could be argued, patients finding was not duplicated by the CASS Principal outside trials should not systematically receive Investigators and their Associates (1984), where inferior care as a direct or indirect consequence there were no significant improvements in either of their non-trial status. If we espouse this view, the experimental or trial control arms over the then there are two possible ways of redressing the non-trial controls. The remaining four studies balance between trial and non-trial patients. The which did not report a trend (significant or first option is predicated on the observed benefit otherwise) in favour of the experimental treatment being an intrinsic property of the trial, for example within the trial (Davis et al., 1985; Karjalainen and a treatment effect. In that case, we might offer all Palva, 1989; Stiller and Draper, 1989; Ward et al., eligible patients entry into trials so that trials be- 1992) all found that trial patients, as a whole, did come routine ‘treatments’ and all eligible patients better than those outside a trial. This suggests a have access to the both the risks and benefits that general (non-specific) trial effect, rather than a trials offer if they so choose (Segelov et al., 1992). systematic benefit from experimental/new However, if the ‘trial effect’ is really a ‘protocol treatments given in trials. effect’, then a second, though not mutually exclu- sive, option appears preferable – that all treat- It has been noted by (Benson et al., 1991) that ments, whether trial or non-trial, should be carried some physicians make use of strict protocols out under protocol regimens. A patient who did without actually entering their patients into a trial. not want to enter a trial would not thereby be The use of protocols outside trials may have similar disadvantaged and would still receive all the bene- physical benefits to those seen above. A study of fits associated with care under protocol. In the Schmoor et al. (1996) examined survival rates of meantime, we may, as scientists, gain confidence non-randomised patients who had declined trial that well-conducted trials tend to benefit the entry but who had nevertheless been treated in participants on average. accordance with the trial protocol, and had found no significant difference between this group and those who underwent randomisation. This What are the psychological effects approach is otherwise known as a ‘comprehensive of participating in trials? cohort study’. Indeed, a systematic review of clinical guidelines by Grimshaw and Russell (1993) showed Results that they tend to result in improved medical care. Three studies provided data on the psychological The extra attention to detail and compliance with effect participating in trials. agreed standards associated with guidelines has advantages over less formalised practice (Lilford The lists of studies, outcome measurement, et al., 1995). Overall, the evidence seems to favour main results and quality of data assessment are the idea that participating in clinical trials is summarised in Table 6. beneficial, thanks to the enhanced attention to detail contingent on the need to follow a trial The first study, by Cassileth et al. (1986), protocol and the expectation of benefit which examined psychological effects in two cancer available effective treatments afford. trials. The remaining studies were based on single placebo-controlled trials of treatments Conclusion for hypertension (Mann, 1977) and HIV/AIDS Of course, the studies, from which the meta- (Robiner et al., 1993). analytic graph is derived, are few in number, of variable quality, and are not a random sample from The study of Cassileth et al. (1986) compared study the population of all trials. Nevertheless, it offers and control groups in two cancer trials and also some support for the plausible belief that well- compared one trial with another – the difference researched treatment protocols, as used in many between the trials was in the type of chemotherapy trials, bring benefits to patients. used. However, a non-trial control group, which 35 Review results

would be needed to establish a trial effect per se, anxiety, psychiatric morbidity, or depression, and was not used. They found no significant or consist- time of administration. ent tends in anxiety between trials. However, the control (observation only) group showed signifi- It is important to differentiate between trait and cantly higher levels of anxiety than those receiving state anxiety (Spielberger, 1988). In attempting to adjuvant therapy and this difference was persistent determine the specific effects of trial participation over time. and the immediate effects of different informed consent procedures, the state measure is, prima The two remaining studies measured patients’ facie, a more useful outcome because a patient’s psychological state before and after enrolment and state will be affected by information, whereas compared these scores with a concurrent non-trial his/her trait score should not. However, a patient’s control group (see Figure 1). Using this design, trait anxiety may have wider implications for the both Mann (1977) and Robiner et al. (1993) conduct of clinical trials. Physicians must be reported that there was no difference in psycho- sensitive to the fact that some patients will need logical scores for trial participants and non-trial a greater degree of caring attention than others controls at the baseline. However, improved during informed consent and thereafter, irrespec- psychological scores were seen among trial tive of what information is given to them. In addi- participants at 3 months and 1 year later in the tion, patients scoring high on trait anxiety scales study of Mann (1977). Robiner et al. (1993), may be selectively excluded from trial participation however, found that there was no significant due to reluctance on the part of the physician to difference in measures of depression or distress broach the subject of trial entry. Thus, it is doubly between trial participants and non-trial controls, important for the such patients to be advised that and that both groups improved with time. That they have nothing to lose by trial entry, and that said, non-trial controls differed in their relative their physician is equipoised. patterns of distress over a 6 month period: there was an initial decrease in distress at 2 months but Response rate to the outcome measurements an increase by 6 months, whereas, for trial is another potential problem; indeed, two participants, there was a consistent decrease in studies, Cassileth et al. (1986) and Mann et al. distress from entry to 6 months. (1977), failed to achieve a response rate of 70% or above. Discussion The ideal study to examine the psychological Conclusion effects of participating in clinical trials would There is some evidence to suggest that participants invoke repeated psychological measures from in clinical trials have better psychological outcomes patients randomised to be offered or not offered than non-trial patients – a result which is sustain- trial entry in a wide range of clinical trials (ideally, able for up to 1 year following trial entry. However, themselves, representing a random sample of all as there are only a few studies in this area, the possible trials). The analysis should be by ‘intention above result requires replication before it can to treat’, since those who refuse to participate be regarded as conclusive. might have different experiences. None of the studies considered people who refused to participate. Moreover, psychological outcome should be What is the ‘best’ method for measured in the trial control group and compared obtaining informed consent? with that in non-trial controls, so that any psychological advantage inherent in participating in trials Results per se can be distinguished from the effects of hope Studies under review engendered by innovative treatments in the study There were 14 studies providing data on different arm of the trial (Robiner et al., 1993). methods of obtaining informed consent of which all except three (Dal-re, 1991; Levene et al., 1996; All three studies used well validated psychometric McLean, 1980) used RCT designs, that is, they were scales. A disadvantage of not using such a scale is trials of methods to get consent for trials. In terms that patient’s may have difficulty articulating their of outcome, 11 studies examined recruitment rates emotions or may not do so consistently. Several to the clinical trial in question, eight studies exam- different psychological instruments for measuring ined understanding, and five studies looked at psychological well-being are used but there is no psychological well-being (four studies examined obvious agreement among the scales due to the all three outcomes). Seven studies were based on 36 use different psychological concepts, for example hypothetical scenarios and seven studied patients Health Technology Assessment 1998; Vol. 2: No. 15

who were offered entry into ‘real’ trials. Randomised of randomisation (in addition to baseline inform- designs were used five former and six of the latter. ation) and those who did not (McLean, 1980). In sum, more information was associated with either Seven studies involved patients who were offered a lower or an unchanged consent rate. entry into real trials: cancer trials (Aaronson et al., 1996; Davis et al., 1990; Simes et al., 1986), an AIDS Levene et al. (1996) found that the greater the trial (Tindall et al., 1994), a cardiovascular disease time between presentation of information and trial (Mayers et al., 1987), a trial for mentally ill decision, the lower the willingness to participate. patients (McLean, 1980), and trials for critically ill However, this observation was based on a com- premature babies (Levene et al., 1996). parison of two separate trials, where, not only the urgency, but also the type of therapy, was different. Lists of studies using real trials, main results and Very tentatively, one might conclude that the evaluation of quality are summarised in Table 8a. more time patients have to consider their position and the better informed they are, so the more Respondents in studies, based on hypothetical likely they are to select their treatment and scenarios, were either patients in hospital, or were eschew randomisation. members of the general public. Patients took part in four of these hypothetical studies (Fetting et al., One study examined the effect of pre- versus 1990; Llewellyn-Thomas et al., 1995; Simel and post-randomisation consent on recruitment rates Feussner, 1991; White et al., 1984), but they were and found that those who were assigned to post- aware that they were taking part in a psychological randomisation consent were less likely to consent experiment. The other three studies asked mem- to standard treatment than those who were offered bers of the general public to make a hypothetical experimental treatment (Gallo et al., 1985). decision about entry to an RCT (Epstein and Moreover, the consent rate seen in the post- Lasagna, 1969; Gallo et al., 1995) or to a Phase 1 randomisation consent group was lower than trial (Dal-re, 1991). that in the pre-randomisation group but only when the standard, not the experimental, treat- Lists of studies using hypothetical trial scenarios, ment was offered, even when they understood that main results and evaluation of quality are equipoise applied. This is interesting as it suggests summarised in Table 8b. that patients do not like finding themselves in a control group when they were not party to the Inter-reviewer reliability randomisation process. All studies were reviewed by the first author and one in two studies were analysed independently One study examined the effect of different types by the last author; there were four discrepancies of information on recruitment rate and found that between the reviewers according to the quality of information given in a descriptive rather than a evidence checklist (see Table 7), all of which were numerical vignette was associated with significantly resolved by re-analysis of the studies. higher rate of consent (Fetting et al., 1990) – much remains hidden when words are used to convey Recruitment rates quantitative data. The remaining study, which Eleven studies examined the effect of different examined the effect of framing probabilistic methods of obtaining informed consent on information on recruitment, found no statistically recruitment rates to trials. Seven of these studies significant difference between formats, though looked at the effect of quantity of information on the ‘neutral frame’ was associated with a slightly willingness to consent and all except two (Davis higher consent rate (Llewellyn-Thomas et al., et al., 1990; McLean, 1980) found that more 1995). In the ‘negatively framed’ version, side- information was associated with a lower rate of effects were presented in terms of morbidity and consent (Aaronson et al., 1996; Dal-re, 1991; mortality rates, while the positively framed version Epstein and Lasagna, 1969; Simes et al., 1986). enumerated freedom from side-effects and survival However, only two of these studies reported a rates. The ‘neutrally framed’ version attempted to statistically significant association (Dal-re, 1991; give a balanced description of both. Simes et al., 1986). The two exceptions found no statistically significant difference in terms of Knowledge and understanding willingness to consent between patients receiving The effect of quantity of information. Five studies a supplementary information booklet and those examined the effect of quantity of information who did not (Davis et al., 1990), and between provided at consent on patient understanding and patients receiving an explanation of the concept they produced apparently conflicting results: 37 Review results

• Research value of treatment. The study of Simes difference, but the trend was in favour of et al. (1986) found that the ‘total disclosure’ greater understanding with more information. group was more knowledgeable about the There may be an optimal level of information research nature of the study than the group such that more information is associated with receiving only limited information. Aaronson greater understanding, until perhaps the et al. (1996) corroborated this: the concept quantity is so great that boredom or of the trial and the particular objective were confusion set in. understood statistically significantly better by • Concept of randomisation. Different quantities those receiving a supplementary interview of information about method of treatment with a nurse than those who underwent the allocation (concept of randomisation) were standard consent procedure. examined by three studies: Simes et al. (1986) • The voluntary nature of participation. One found that respondents in the ‘total disclosure’ study, Davis et al. (1990), reported that patients group understood the concept of randomisation receiving a supplementary information booklet significantly better than those in the individual were significantly more aware of the voluntary approach (the concept was actually relayed, on nature of their participation. average 68%, of the time during the individual • Right to withdraw. Aaronson et al. (1996) approach interviews as compared with 96% for reported the proportion of respondents who the total disclosure group); neither Davis et al. understood that they could withdraw at any (1990) nor White et al. (1984) replicated this time from the trial was significantly larger in result. Davis et al. (1990) reported that respon- the statistical sense in the group receiving a dents who received a supplementary inform- supplementary interview with a nurse on ation booklet gave significantly more incorrect top of the standard consultation. responses or were undecided about the meaning • Equipoise. White et al. (1984) found no of the concept, and White et al. (1984) found no significant difference between the three significant difference between forms of different groups who received varying levels of inform- length, even though the only form to mention ational detail in terms of understanding of the randomisation was the long form. ‘uncertainty’ associated with the treatments • Overall understanding. One study reported offered in their trial. significantly better overall understanding of • Available alternative treatments. One study, respondents receiving a supplementary interview Aaronson et al. (1996), asked respondents if with a nurse (Aaronson et al., 1996). they knew of any alternative treatments and found that those who received a supplementary Oral versus written information interview with a nurse were statistically Two studies were concerned with the effect of significantly more knowledgeable about oral versus written, or descriptive versus numerical available alternatives. information on understanding. Fetting et al. • Side-effects. Three studies, Davis et al. (1990), (1990) showed that presenting numerical as Epstein and Lasagna (1969) and Simes et al. opposed to descriptive information (about (1986), looked at the effects on understanding disease-free survival) produced more accurate of different quantities of information on side- recall, while Tindall et al. (1994) reported that effects. Whereas less detailed information was respondents increased their knowledge scores associated with greater knowledge in the study at a greater rate if they received both oral and of Epstein and Lasagna (1969), a higher level of written information. disclosure resulted in better understanding in the study of Simes et al. (1986). In the latter Timing of consent study, patients were allocated to ‘total disclosure’ One study examined the timing of consent on (which mentioned an average of 12 side-effects) understanding (i.e. before versus after randomis- or an ‘individual approach’ which resulted in ation), but found that misunderstanding rates an average disclosure of seven side-effects. The were independent of this aspect of the consent short form in the study of Epstein and Lasagna procedure (Gallo et al., 1995). (1969) mentioned eight side-effects, making it comparable with Simes’s ‘individual approach’. Anxiety and other complaints (or side-effects) However, both the intermediate and long forms Four of the five studies examined the effect of in the study of Epstein and Lasagna (1969) different quantities of information on anxiety: two provided substantially more information than studies showed that less information was associated was given in Simes’s ‘total’ disclosure method. with less anxiety (Simes et al., 1986; Epstein and 38 Davis et al. (1990) did not report a significant Lasagna, 1969), whereas Davis et al. (1990) found Health Technology Assessment 1998; Vol. 2: No. 15

that patients receiving a supplementary booklet Several different instruments were used to were significantly more likely to feel ‘somewhat’ or measure anxiety, but only one administered a ‘more relieved’. The other study, by Aaronson et al. formal psychometric scale (Aaronson et al., 1996), (1996), reported no effect of consent format on while the others simply relied on patients’ patient anxiety. accounts of how they felt. Psychometric scales, such as the Spielberger State–Trait Anxiety One study, by Mayers et al. (1987), found that Inventory (STAI), possess a high degree of gastrointestinal side-effects were reported validity and give results which can be compared significantly more often if the possibility of these from place to place. effects had been disclosed at the consent stage, irrespective of whether the patients were assigned Some studies sought to measure how much to the treatment or placebo arm of the trial. information was understood by the respondents. Simple recall of information is potentially mislead- Interactions ing, since memory per se is an imperfect indicator Four of the 14 studies conducted a multivariate of understanding and different methods for analysis to see how the dependent variables them- testing how well a concept can be understood selves were related. Two studies sought interactions may give different results. For example, recall between recruitment rates and level of under- that the treatment was allocated on a random standing, irrespective of the method for obtaining basis was rated ‘good’ by Aaronson et al. (1996), informed consent. Consent was associated with while more in-depth understanding of the part significantly higher mean scores for understanding that ‘chance’ has to play was less satisfactory in (Epstein and Lasagna, 1969). Those who expected the study by Howard et al. (1981); this could the experimental treatment to be superior to the reflect differences between people in the studies, standard treatment displayed a higher rate of but equally, it could reflect differences in how the consent than those who understood that trial entry question was phrased. Other tools used to test was based on individual equipoise (Gallo et al., knowledge and understanding were multiple 1995). Another study (Fetting et al., 1990) gave a choice questionnaires, true/false responses, breakdown of consent rates by expressed optimism open-ended interviews, and patients’ perceptions and found that those who gave more optimistic of their own understanding. The latter method estimates of 10 year disease-free survival were less is particularly likely to over-estimate understand- likely to consent to trial entry. The remaining ing since self-assessment of ignorance requires study reported that patients who exhibited good knowledge of topics or what information overall understanding were also significantly less may be available. anxious, irrespective of consent method (Aaronson et al., 1996). Most of the studies using real trials included only a sub-population of eligible patients and any Discussion attempt to generalise the results should proceed Methods with caution as there could be systematic differ- The studies we have reviewed were mostly ences between those who participated in the follow- (11/14) RCTs of different methods of obtaining up and those who did not. However, the response informed consent. The exceptions were Dal-re rates were reasonably high (> 75%) in nine studies, (1991) who did not provide details of the so that inferences based on their results are less method of allocation, McLean (1980) who questionable (response rates were not given in used historical controls (with possible time bias), the five remaining studies). and Levene et al. (1996) who compared the same participants offered entry into separate trials Findings requiring differ-ent periods of reflection, (with The results of the various studies suggest that possible confounding by differences in the giving people more information is associated treatments on trial). with a lower consent rate and that consent in emergency situations is more often forthcoming Six of the RCTs were based on hypothetical than in circumstances where patients have time scenarios whose results must be extrapolated to to absorb what has been said to them and reflect the ‘real life’ situation with care (King, 1986); on how they feel. patients faced with difficult decisions may be more concerned to ensure that they understand the However, the data on the effects of quantity of issues when those affect them in reality, and/or information on understanding are not so straight- they may make different trade-offs. forward. Data on understanding of the concept of 39 Review results

the clinical trial and the research nature of the explained), > 75% understood this issue, suggesting treatments was only reported by two studies and that they had previously encountered the concept. both found that the more information the greater the understanding (Aaronson et al., 1996; Simes The data on anxiety, when taken alongside that et al., 1986). As understanding that one is partici- of understanding, suggest that how much a patient pating in a formal experiment is often perceived to understands is more important than how that be one of the most important aspects of consent understanding is achieved. This claim is substanti- (Siminoff and Fetting, 1991), it is surprising that ated by a multivariate analysis conducted by the studies, as a whole, did not bring this out. Aaronson et al. (1996) who found that knowledge Likewise, the studies did not fully investigate the was significantly associated with less anxiety, effect of quantity of information on other aspects irrespective of consent method. of the consent procedure. It could be that the authors considered such information to be ‘simple’ The effects of different formulations of the same facts and relayed them as such without elaboration. information on patient understanding have not For example, one intervention (a supplementary been fully investigated in a comparative study. How- interview with a nurse) was found to yield signifi- ever, we have conducted an analysis of 24 observa- cantly more awareness of the voluntary nature of tional and comparative studies auditing the amount participation and the allied right to withdraw patients in real trials had understood about their among respondents, though this does not neces- trial (Aaronson et al., 1996; Bergler et al., 1980; sarily reflect quantity, so much as quality of com- Daughterty et al., 1995; Davis et al., 1990; DCCT munication (Aaronson et al., 1996). The same study Research Group, 1989; DeLuca et al., 1995; Gallet was the only one to report knowledge of alternative et al., 1994; Harth and Thong, 1995; Harrison et al., treatments as an outcome in its own right – this 1995; Hassar and Weinstein, 1976; Howard et al., could reflect a lack of real available alternatives 1981; Jensen et al., 1993; Lynoe et al., 1991; Marini in the other studies or a reluctance to report et al., 1976; Maslin, 1994; Miller et al., 1994; Oddens negative results (e.g. Simes et al. (1986) reported et al., 1992; Olver et al., 1995; Penman et al., 1984; disclosing different levels of information on rele- Postlewaite et al, 1995; Rodebhuis et al., 1984; vant alternatives, but failed to report knowledge Simes et al., 1986; Tomamichel et al., 1995; Tindall of this as an outcome). Finally, there was only one et al., 1994). This study showed that medical details study which addressed the effect of informational (particularly concrete information such as side- quantity on understanding of the concept of effects) are well understood by patients: eight equipoise (White et al., 1984). studies reported that > 65% of respondents were familiar with the side-effects associated with their In terms of information on side-effects, there seems trial (Davis et al., 1990; DCCT Research Group, to be an optimal amount of information, such that 1989; Harrison et al., 1995; Howard et al., 1981; patients are not over-burdened by detail but are Jensen et al., 1993; Marini et al., 1976; Maslin, 1994; nevertheless informed of the more important Simes et al., 1986), while three studies reported risks. However, in terms of understanding the 35–57% of respondents were so informed (Gallet concept of randomisation, the literature is appar- et al., 1994; Hassar and Weinstein, 1976; Miller ently contradictory: in some studies suggesting that et al., 1994; Olver et al., 1995), and one study providing information on randomisation is associ- reported as few as 28% (Bergler et al., 1980). ated with better understanding of this concept and Other factual information such as right to within others suggesting the reverse. This lack of con- draw and available alternatives was also familiar sistency between studies cannot easily be explained to respondents. However, more conceptual by differing educational and socio-demographic information, such as the concept of randomisation, characteristics of the populations studied since the seems to be less accessible to patients, for example respective samples appeared comparable. It may, the concept of randomisation was understood by however, reflect differences in how the concept was 65% or more patients in three studies (Aaronson actually explained (Corbett et al., 1996). However, et al., 1996; DCCT Research Group, 1989; Jensen rather than investigating different ways of explain- et al., 1993), but only 24–45% of patients in a ing the concept, the studies simply compared further four studies (Davis et al., 1990; Gallet et al., groups who were given an explanation of random- 1994; Howard et al., 1981; Simes et al., 1986). There isation with those where no attempt was made. is evidence elsewhere to suggest that patients are Interestingly, White et al. (1984) reported that, happy to accept their physician’s treatment recom- although there was no difference between groups mendations, except when it comes to clinical trials in terms of their understanding of randomisation (Siminoff and Fetting, 1991), and this might be 40 (even though only one group had the concept explained in part by the abstract concepts involved Health Technology Assessment 1998; Vol. 2: No. 15

(though the factors highlighted by the authors baseline knowledge in the community should be included: amount and specificity of information, enhanced by more open discussion of the topic in perceived strength of recommendation, edu- the media and elsewhere, because the more people cational level and risk-aversion attributes). It is understand the subject, the less anxious they are possible that, for such abstract material which is when given detailed information as part of their particularly troublesome, some formulations of invitation to participate. the same concept may facilitate understanding whereas others may simply generate obscurity. More research into what it is about conceptual What is the quality of informed information such as ‘randomisation’ patients find consent in practice? hard to grasp. Indeed, it has already been shown in one study that people generally prefer descriptions Quantitative results of the concept of ‘randomisation’ which play down Of the 24 studies included in the audit, 23 were the role of ‘chance’ (Corbett et al., 1996). Whether quantitative. Ten of these studies examined under- this preference simply reflects an emotional incli- standing of cancer trials (three were explicitly nation or whether it shows that some phrases were Phase 1 studies), four were concerned with more comprehensible than others is uncertain – cardiovascular trials, two with HIV/AIDS trials respondents preferred the long form in the study (one was a vaccine trial), one with a trial of growth of White et al. (1984) which was, in fact, less hormone, one with an asthma drug trial, one with comprehensible to them. a diabetes control trial, one with an arthritis drug trial, one with common analgesic drugs trial, one Conclusion with a trial related to reproduction and one with The moral imperative of informed consent is an a psychiatric drug trial. ideal where patients are given full details of the trial procedure so that a person can determine Lists of the studies, along with their methods, main what is or is not done to him/her. In theory, if results and an assessment of the quality of the data autonomy is the foundation of informed consent, are summarised in Table 10. a patient should be asked whether or not they wish any information upon which to base a decision, and Of the 23 quantitative studies, one examined the then, given an expressed desire for information, transcripts of consent discussions between patients the patient should make a considered decision and physicians to find out what information had about trial entry. Practice, as ever, falls short of this actually been given to patients irrespective of their ideal and there is a myriad of reasons why this understanding (Tomamichel et al., 1995), nine should be so, not least of which is the logical point assessed understanding of eligible patients who had that patients cannot easily judge whether or not not yet consented to trial entry (Aaronson et al., they want information until that have received it. 1996; Davis et al., 1990; DCCT Research Group, Additionally, time to give full information about 1989; DeLuca et al., 1995; Harrison et al., 1995; treatments is limited; patients do not always read Jensen et al., 1993; Rodebhuis et al., 1984; Simes the consent information, may avoid discussing their et al., 1986; Tindall et al., 1994) and the remainder queries with the physician, or may have only partial examined the understanding of those already or no understanding of what is put to them. That is participating in a trial or those who had not to say, however, that informed consent can completed their trial. never be valid, but rather that physicians must do their best to give as much information as a patient The study assessing the information actually given can use in a form which maximises understanding. to patients reported that complete information This review shows reasonably clearly that, provided about the Phase 1 drug which had been identified overload is avoided, giving patients more inform- as important in prospect (lack of known treatments ation results in greater knowledge (albeit at the of proven efficacy, anti-tumour effect still unknown possible cost of a transient increase in anxiety). and probably non-existent, limited knowledge of It seems to result in reduced willingness to be side-effects, and the logistics of the study) was randomised and this is consistent with the notion given to 78% or more patients, though specific that people are only likely to wish to participate information concerning right to withdraw or in trials when equipoised (see the results section, refuse treatment was given to < 40% (Tomamichel page 29). et al., 1995).

Lastly, we should not rely solely on the informed Of the 22 quantitative studies examining patient consent process to educate patients about trials – understanding, five observed that complete or 41 Review results

adequate information* was understood by 80% aware of 77.5% of the potential side-effects as or more patients (Daughterty et al., 1995; DeLuca compared with 60.7% for patients not receiving et al., 1995;† Postlewaite et al., 1995;‡ Rodebhuis the intervention (Aaronson et al., 1996). et al., 1984; Tindall et al., 1994). Seven studies examined understanding of the Specific knowledge and understanding of various randomisation procedure, of which three found elements of the trial process was examined by that the concept was understood by 65% or more 18 studies. patients (Aaronson et al., 1996; DCCT Research Group, 1989; Jensen et al., 1993). Davis et al. Nine studies addressed understanding of the (1990), Gallet et al. (1994), Howard et al. (1981), scientific purpose of trials, and all apart from two and Simes et al. (1986), by contrast, reported that (Daughterty et al., 1995; Gallet et al., 1994) report- only 24–45% of patients understood that the ed that most patients understood the scientific allocation of treatment was based on ‘chance’. objectives of their trial or were aware of the research nature of their trial (Aaronson et al., Six studies reported data on patient’s knowledge 1996; DCCT Research Group, 1989; Harth and of their right to withdraw from the trial. Although Thong, 1995; Howard et al., 1981, Lynoe et al., most patients were aware of this right (Aaronson 1991; Marini et al., 1976; Miller et al., 1994; Penman et al., 1996; Bergler et al., 1980; Gallet et al., 1994), et al., 1984; Simes et al., 1986). The exceptions 40% or more respondents did not appreciate the observed that 40% or less respondents were able fact that they could withdraw at any time even to state the scientific purpose of a Phase 1 trial before the trial was over (Davis et al., 1990; Lynoe (Daughterty et al., 1995) and a placebo-controlled et al., 1991; Maslin, 1994). trial (Gallet et al., 1994). Additionally, Harth and Thong (1995) reported that, while nearly 100% of Four studies observed that > 60% of respondents parents were clear about the aim of determining were aware that their physician did not know which drug efficacy, less than 15% were aware that their treatment they had been allocated to, that is, that trial was designed to assess safety as well. their trial was ‘double-blind’ (Bergler et al., 1980; Gallet et al., 1994; Hassar and Weinstein, 1976; Seventeen studies examined knowledge of Howard et al., 1981). potential side-effects, and all apart from seven (Aaronson et al., 1996; Bergler et al., 1980; Gallet Three studies showed that > 70% of respondents et al., 1994; Hassar and Weinstein, 1976; Maslin, knew that they could receive a placebo in their 1994; Miller et al., 1994; Olver et al., 1995) reported placebo-controlled trial (Harrison et al., 1995; that more than 60% patients were, at least, aware of Howard et al., 1981; Marini et al., 1976). the possibility of side-effects (Davis et al., 1990; DCCT Research Group, 1989; Harrison et al., 1995; Two studies found that 65% or more respondents Howard et al., 1981; Jensen et al., 1993; Marini et al., were aware of the availability of alternative treat- 1976; Maslin, 1994; Penman et al., 1984; Postle- ments outside their trial (Aaronson et al., 1996; waite, 1995; Simes et al., 1986). Despite patients Jensen et al., 1993). being aware of the possibility of side-effects, Jensen et al. (1993) and Marini et al. (1976) showed that One study reported that 53% of respondents more detailed recall of the particular nature of understood that standard treatments are not specific side-effects was, in fact, poor. The excep- always inferior to experimental treatments tions observed that 48% or less respondents could (Davis et al., 1990). correctly recall possible side-effects (Bergler et al., 1980; Gallet et al., 1994; Hassar and Weinstein, Qualitative results 1976; Maslin, 1994; Miller et al., 1994; Olver et al., One of the 23 studies was a qualitative study. 1995) and that, on average, patients receiving Snowdon et al. (1997) found that parents of babies a supplementary interview with a nurse were participating in the ECMO Trial had difficulty

* Adequate information for medical research, as specified by the Nuremberg Code, comprises: the nature, duration, and purpose of the experiment; the method and means by which it is to be conducted; and the effects upon his health or person which may possibly come from his participation in the experiment. † The author did not state exactly what material was needed to satisfy each level of understanding. ‡ Postlewaite measured children’s understanding as well as that of their parents (cited here) and found that > 35% were 42 unable to recall the information. Health Technology Assessment 1998; Vol. 2: No. 15

understanding the concept of a trial and failed to who consent (refusers might have refused trial appreciate that their baby might receive a control entry because they failed to understand the inform- treatment, thereby equating the trial with the single ation or because they understood that equipoise experimental treatment. There was particular did not apply). Eleven out of the 24 studies in the difficulty understanding what randomisation audit assessed patients who had not consented to means, why it is used and how it is actually carried trial entry or cited the recruitment rates of the out – recall was vague and responses were trials, thereby providing a degree of reassurance sometimes inconsistent. that all those approached to participate were enumerated in the result. Discussion The studies Any attempt to generalise the results should The studies we have reviewed all dealt with proceed with caution as there could be systematic different samples and different methods of assess- differences between those patients who partici- ment; generalisation from these results therefore pated in the audit and those who did not. However, requires caution. Studies used a mix of designs but if the response rates to the psychosocial surveys in were all based on recruitment to real trials. question are reasonably high, inferences based on their results are less questionable. There are many difficulties associated with measuring understanding as an outcome of psychosocial It is evident that the level of patient’s knowledge research. By employing recall of information as a and understanding of their clinical trial varies device for evaluating patient’s knowledge and according to two separate factors, that is, type of comprehension, there is a reliance on memory information within the same trial, and type of trial. which may fade over time. The longer the recall test The first factor, the type of information within the is administered after consent, the more prone to same trial, is best illustrated by the result that the error are the results. Furthermore, recall per se does possibility of incurring side-effects is easier for not necessarily reflect understanding, as can be seen patients to grasp (> 60% of respondents were from the apparently inconsistent results concerning aware of the possibility of side-effects though more patient’s recall and understanding of the concept of detailed recall was less good) than the concept of ‘randomisation’. While recall of the randomisation randomisation, e.g. the concept of randomisation procedure, that is, that the treatment was allocated was understood by only 24–45% of patients in on a random basis, was rated ‘good’ (Aaronson et al., four studies (Davis et al., 1990; Gallet et al., 1994; 1996), understanding of the part that ‘chance’ has Howard et al., 1981; Simes et al., 1986), and by to play was less satisfactory (Howard et al., 1981). 65% or more patients in three studies (Aaronson Other methodological tools used were multiple et al., 1996; DCCT Research Group, 1989; Jensen choice questionnaire, true/false responses, open- et al., 1993). This can be explained by the notion ended interviews, and patients’ attitudes to their that patient’s understanding is facilitated by con- own understanding. However, to ask the respon- crete rather than abstract information. Abstract dents whether or not they thought that they had information may be particularly troublesome for understood the information given may over-estimate children and adolescents given that cognitive func- understanding since patients may not be aware of tioning develops in stages and begins with thinking what there is to know (Olver et al., 1995). in terms of concrete information and then progresses to more abstract conceptualisations (Piaget, Qualitative studies are important in this subject as 1926). The second factor, a possible corollary they can provide some insight into what people of the first, is that patients understand some trials really understand by allowing the respondent to more easily than others and Phase 1 trials seem determine the structure of the conversation. How- to produce particular barriers to understanding ever, there are problems associated with encoding (Daughterty et al., 1995). The high level of qualitative responses, so making it especially scientific sophistication associated with the Phase 1 important to conduct inter-rater reliability tests. trial and its consequent richness of abstract con- We think that an approach combining qualitative ceptualisations may explain why patients find the and quantitative methods would be particularly information more incomprehensible. appropriate for this topic. Implications for the doctrine of Another difficulty for the audit arises when those informed consent who decline to participate in a real clinical trial are The moral imperative of informed consent is an excluded from the psychosocial research since ideal where patients are given full details of the trial refusers might be systematically different to those procedure so that a competent person can 43 Review results

determine what is done with him/her. Thus, if Conclusion material information were withheld from patients We conclude that competent patients should be or if they were deceived into harbouring mistaken given ‘material’ information, though we cannot beliefs about the consequences of a decision, the always expect full understanding. The poor quality informed consent process may be invalidated. of informed consent in some cases does not neces- However, patients do not always grasp what inform- sarily invalidate the procedure, though practi- ation is disclosed to them, resulting in partial or tioners should check a patients understanding even no understanding of the trial. Difficulties and be prepared to repeat or explain points where relaying esoteric (especially scientific) knowledge necessary. Patients should be encouraged to ask are not due to its technical and precise vocabulary questions when they do not understand what is alone, as our everyday language can be, though is by said to them. no means always, as precise or well defined. That said, some concepts do take on a more technical and specific meaning in science and consequently How do patients, the public some terms are not familiar to patients at all. More and healthcare professionals importantly, however, the meaning of scientific concepts becomes increasingly inter-dependent view RCTs? (or paradigmatic), so that it is difficult to relay a Quantitative and qualitative studies concept in isolation to the patient (unless he/she Studies under review understands the whole theory). Misunderstanding Fifty-eight articles provided data on attitudes may also result from a reluctance to discuss queries to clinical trials, of which 51 were quantitative with the physician, or even from failure to read the and eight were qualitative (one study used both consent form (Olver et al., 1995). Potential methods – Marsden et al., unpublished). Thirty- problems do not stop there, for, even if patients four of the quantitative studies examined the think that they understand individual concepts, opinions of the public or patients, of which 19 they may take away a very different idea (from that addressed patients who had been invited to partic- of the physician) of what is going on, making ipate in clinical trials and 17 addressed patients or different inferences on the basis of apparently well- the public in hypothetical trial scenarios (one study understood shared information. Defects in reason- examined both patients in real trials and members ing which are within the agent’s control can likewise of the public responding hypothetically – Bevan limit autonomous decision-making (Savalescu and et al., 1993). Twenty studies examined the opinions Momeyer, 1997). That is not to say that informed of healthcare professionals (four studies examined consent can only be valid if a patient has perfect both: Epstein and Lasagna, 1969; Marsden et al., understanding of his/her trial, rather that legally unpublished; Oakley, 1990; Penman et al., 1984). valid consent should be based on a pre-determined Professionals’ views were generally based on real minimum grasp of the details as was set out in the trials, though five studies looked at views on clinical audit. What seems to be important is that patients trials in general and two studies looked at the have a realistic grasp of how they can expect to views of members of trial review committees benefit from participation, since we have found that (Dal-re, 1990; Kodish et al., 1992). more people seem to choose to participate on the basis of self-interest than altruism (see the following Interobserver reliability section). Decision analysis can provide a rational Interobserver reliability testing for the views of platform for informed consent, so that any lack of patients and the public showed 21 examples of equipoise is transparent. In order for a physician to disagreement among reviewers (SE and JH) over respect a patient’s autonomy in a Kantian sense, the quality of evidence. The majority of disagree- he/she must discharge a ‘perfect’ duty to transmit ment (16 cases) was over the ‘validity’ of the study material information, without obscuring concepts. design used, and reflected the fact that validity Deciding exactly what is ‘material’ may vary some- can be used to mean different things. We resolved what from patient-to-patient, however. In addition, these disagreements by clarifying the definition of a judgement concerning patient comprehension external validity, so that we were concerned with should be made and the physician should repeat how clearly the target population was defined, a or elucidate specific points if necessary. It is worth population to which the sample results could in pointing out that the languages of the medical theory be generalised. Discrepancies in the profession and the laity are not incommensurable, response rate to the survey in question or the type and the informed consent consultation is a process of sampling employed occurred five times, for of two-way communication aimed at reaching example JH noted that the study of Mattson et al. 44 mutual understanding. (1985) had used random sampling of the Health Technology Assessment 1998; Vol. 2: No. 15

population in question while SE did not – all one study (Henzlova et al., 1994). Ten of the above such cases were resolved through re-analysis studies reported both altruism and self-interest of the original papers. as motivating factors. The percentage of patients citing altruistic reasons was greater than that citing Quantitative study results self-interested motivations in five of these ten Views of patients and the public studies (Bevan et al., 1993; Harth and Thong, Real trial scenarios. Five of the 19 real trial 1990; Henzlova et al., 1994; Jensen et al., 1993; scenario studies provided data on patients’ views Lynoe et al., 1991). on cancer trials, of which two were Phase 1 studies. Five studies were related to cardiovascular disease, Nine studies reported the perceived benefits or two to AIDS trials, three were studies of views on disadvantages to the patients of participation in trials concerning abortion and reproduction trials: Daughterty et al. (1995), Harth and Thong (including hormone replacement), two were (1995), Henzlova et al. (1994), Mattson et al. studies concerning childhood asthma, one (1985), Penman et al. (1984), Rodebhuis et al. involved a placebo-controlled study for arthritis (1984), Ross et al. (1993), Vogt et al. (1986), and and one did not give details of the trials in Wilcox (1994). The proportion of respondents question (Bevan et al., 1993). expecting physical therapeutic benefit varied according to the type of trial in which the patients Table 12a lists the above real trial scenarios along enrolled, from 78% in Phase 2 and 3 trials with sampling information, the instruments used (Penman et al., 1984) to 20% in a Phase 1 study for assessing patients’ attitudes, the main results (Rodebhuis et al., 1984). However, most partic- and comments on the quality of the study design ipants were hopeful of gaining some psychological as described by the authors. benefit (Henzlova et al., 1994) even in these Phase 1 trials (Daughterty et al., 1995). Only three studies, Common questions put to the respondents in Henzlova et al. (1994), Mattson et al. (1985) and various studies concerned what motivates patients Vogt et al. (1986), reported some perceived dis- to participate, the perceived benefits or disadvan- advantages to trial participation (in 11–37% of tages to the patient, who they thought stood to respondents), and these were of a practical rather benefit most from the trial, satisfaction with the than health-related nature. One study (Harth medical attention they received whilst participating and Thong, 1995) reported that 79% of parents in the trials, and whether or not they would be thought that there was no or only a low risk associ- willing to enter future trials. ated with placing their children in a clinical trial.

Thirteen of the studies (summarised in Table 12c) Two studies asked respondents who they thought addressed the question of what motivates patients stood to benefit most from the trial (Mattson et al., to participate in trials, but the phrasing of the ques- 1985; Ross et al., 1993). In both cases, scientists tions differed from study to study. Overall, there rather than participants were more frequently rated were ten studies based on real trials which reported as being the major beneficiaries, notwithstanding altruistic motivations and the frequency with which confirmation of perceived therapeutic benefit to such inclinations were cited by respondents in the participants themselves. those studies was: over 60% of respondents in three studies (Bevan et al., 1993; Harth and Thong, 1990; Seven studies gave some measure of the patient’s Mattson et al., 1985), between 40 and 60% in four satisfaction with the medical attention they received studies (Hassar and Weinstein, 1976; Jensen et al., in their trials (Bevan et al., 1993; Henzlova et al., 1993; Lynoe et al., 1991; Penman et al., 1984), and 1994; Henshaw et al., 1993; Hassar and Weinstein, under 40% in three studies (Henzlova et al., 1994; 1976; Jensen et al., 1993; Mattson et al., 1985; Vogt et al., 1986; Wilcox and Schroer, 1994). Thir- Wilcox and Schroer, 1994). Over 77% of patients in teen studies reported self-interested motivations four studies expressed a willingness to participate for participation in real trials. The frequency with in future studies or the same trial again (Henzlova which self-interest was expressed by respondents et al., 1994; Henshaw et al., 1993; Jensen et al., 1993; was over 70% in four studies (Daughterty et al., Mattson et al., 1985;). Wilcox and Schroer (1994) 1995; Harth and Thong, 1990; Mattson et al., 1985; found that 88% of respondents would enter the Penman et al., 1984), between 30 and 55% in eight same study again, but only 58% would be interested studies (Barofsky and Sugarbaker, 1979; Bevan in future studies of a different nature. Hassar and et al., 1993; Hassar et al., 1976; Jensen et al., 1993; Weinstein (1976) reported that no respondent Lynoe et al., 1991; Rodebhuis et al., 1984; Vogt thought that the quality of their care had suffered et al., 1986; Wilcox, 1994) and under 20% in during the trial period, while 54% of respondents 45 Review results

in the study of Bevan et al. (1993) were happy with should be provided during the consent process, every aspect of the trial. how the concept of equipoise is regarded, who should participate in trials, and willingness Four of the studies gave an indication of who to participate. patients thought had the most influence over their decision to participate in the trial. They reported Eleven studies asked respondents what factors that no less than 80% of patients felt that they had might motivate them to accept or decline to take made an autonomous decision whether or not to part in a hypothetical clinical trial. Nine of the participate in the trials, and did not feel that they 11 studies asked about the specific motivations had relegated this decision to their physician of altruism and self-interest and are summarised (Harth and Thong, 1995; Lynoe et al., 1991; in Table 12c. Eight of the nine studies reported Penman et al., 1984; Rodebhuis et al., 1984). One altruistic motivations and the frequency with which study compared the responses of consenters and such inclinations were cited by respondents was refusers and found that there was no statistically over 65% in four studies (Alderson, et al., 1994; significant difference between these groups with Autret et al., 1993; Cassileth et al., 1982; Flanery respect to wanting shared decision-making with et al., 1978), between 20 and 60% in three studies their physician (Marsden et al., unpublished). (Bevan et al., 1993; Millon-Underwood, 1993; Slevin et al., 1995), and under 15% in one study (Kemp et One study, Tindall et al. (1994), found that 79% of al., 1974). Self-interested motivations were reported responding AIDS patients thought that unproved by seven of the nine studies, and the frequency with medications should be available to sufferers outside which self-interest was expressed by respondents the trial procedure. was over 50% in two studies (Cassileth et al., 1982; Flanery, 1978), and between 25 and 50% in the Another study, Bevan et al. (1993), reported that remaining five studies (Alderson, et al., 1994; Autret 83% of respondents thought that they had been et al., 1993; Bevan et al., 1993; Millon-Underwood, given sufficient time to consider the informed 1993; Slevin et al., 1995). One study (Gerard et al., consent material. 1995) showed that severity of illness, rather than remuneration, was most influential in the decision One final study, Marsden et al. (unpublished) gave to participate, though it is unclear whether this was limited raw data, but indicated that there was no due to altruistic or self-interested motives. Another statistically significant difference between those study, Autret et al. (1993), reported that the most who consented to be randomised and those who frequently cited reason (75%) for refusing trial refused with respect to concerns about the medical entry was risk of side-effects, while 19% of potential ‘uncertainty’ surrounding hormone replacement refusers disagreed with the RCT on principle. therapy (HRT). However, 4% of those who refused were, in fact, equipoised, yet still declined random- Six studies examined perceptions of the isation, while the remainder cited perceived side- information that should be provided during the effects as a reason for refusing. consent process: Corbett et al. (1996), Epstein and Lasagna (1969), Oakley et al. (1992), Saurbrey et al. Hypothetical trial scenarios. There were seventeen (1984), Simel and Feussner (1991), White et al. studies which used hypothetical scenarios to elicit (1984). The first study reported that respondents views on trials. Nine of the 17 studies presented generally preferred written over verbal inform- respondents with a specific (mock trial) scenarios ation, and favoured explanations of the concept of conditions with varying degrees of medical severity: randomisation which were less explicit about the Corbett et al. (1996), Epstein and Lasagna (1969), play of chance. Both the first two studies reported Flanery et al. (1978), Johnson et al. (1991), Kemp that some respondents expected a degree of et al. (1974), Llewellyn-Thomas et al. (1991), distress contingent on the very offer of trial entry Saurbrey et al. (1984), Simel and Feussner (1991), (Corbett et al., 1995, Epstein and Lasagna, 1969). Slevin et al. (1995) and White et al. (1984). The Saurbrey et al. (1984) and White et al. (1984) remainder ascertained views of patients and the reported that > 80% and 68% of respondents public to clinical trials in general. respectively preferred longer consent forms or ‘full’ information despite the observation by Simel All studies using hypothetical trial scenarios are and Feussner (1991) that only 62% of their summarised in Table 12b. respondents would take quantitative information into account when deciding whether or not to Common questions posed were what would participate. One study (Alderson et al., 1994) 46 motivate patients to participate, what information reported that nearly 70% of respondents Health Technology Assessment 1998; Vol. 2: No. 15

would want to know that there was uncertainty Affirmative responses were received from 50 to surrounding treatments in an RCT and whether 90% of respondents except in the study of Flanery or not they had been assigned to the control arm. et al. (1978) which found that only 41% of Finally, Saurbrey et al. (1984) found that 75% of respondents would be willing to consent. respondents thought that patients who are unable to give consent (e.g. they were unconscious) could Views of healthcare professionals be included in a trial nevertheless. Fifteen of 21 articles in this category examined views of physicians to oncology trials. One study Four studies made reference to the concept of hinged on a particular hypothetical trial of early equipoise in the reported views of patients and the versus delayed delivery for preterm foetuses public. Corbett et al. (1995) showed that 83% of (Lilford, 1994), while one study used an HRT respondents considered clinical trials to be ethical pilot trial (Marsden et al., unpublished). Five in principle given equipoise, while Johnson et al. studies looked at the views of committee members, (1991) investigated the extent to which the public academics or chairpersons when reviewing Phase 1 regarded equipoise among experts as important, studies (Kodish et al., 1992), trials involving and reported that 97% of respondents would children (Mammel and Kaplan, 1995) or trials in consider a trial unethical if equipoise was disturbed general (Blum et al., 1987; Dal-re, 1990, 1993). above 80:20, that is, if 80% or more of the experts thought that one treatment offered more benefit Table 13 provides a summary of the reported than its comparators. In the studies of Cassileth attitudes to the various trials, the sample et al. (1982) and Millon-Underwood (1993), 30 population, the method of assessing the attitudes and 74% of respondents, respectively, thought and comments on the quality of the data. it likely that physicians enter their patients in trials only when equipoised. Common issues studied were: to what extent are physicians prepared to participate in trials, what Two studies examined the issue of whether the are their views on the principle and practice of severity of a patient’s disease should affect the informed consent, do their patients stand to offer to participate in trials. Cassileth et al. (1982) benefit from participation, what effect does remarked that respondents who had not before parti- participation have on the physician–patient cipated in a trial thought that the offer of trial entry relationship, and, finally, are clinical trials should be restricted by the patient’s medical status, ethical given equipoise? whereas those who had participated in trials previously thought any patient should be allowed to enter. Seven of the 15 studies sketched a general Likewise, Millon-Underwood et al. (1993) showed picture of the extent to which physicians would that a large proportion of respondents thought that be prepared to participate in trials: Alderson et al. any patient, regardless of medical status, should be (1994), Benson et al. (1991), Langley et al. (1987), given the opportunity to participate. Spaight et al. (1984), Taylor (1992), Taylor and Kelner (1987b), and Taylor et al. (1984). With In the studies of Corbett et al. (1996), Johnson et al. one exception (Oakley et al., 1992), the studies (1991) and Kemp et al. (1984), views were seen to found that over 60% of responding physicians change according to the type of trial under con- commonly enter some or all of their eligible sideration. Johnson et al. (1991) reported that the patients in clinical trials. The exception (Alderson required level of collective equipoise was higher et al., 1994) reported that 23–53% of physicians, if the issues were highly emotive, for example if depending on the trial in question, expressed a infants were involved or if the prognosis were grave. willingness to participate. However, two studies The remaining two studies were consistent with showed that some physicians were more reluctant this: more respondents stated that they would pre- to enter patients when placebo controls are used fer to choose their own treatment (thereby refusing (Benson et al., 1991), or when one of the com- randomisation) as the clinical condition described parator treatments is toxic and/or has substantial became increasingly grave (Corbett et al., 1996; side-effects (Spaight et al., 1984). With only two Kemp et al., 1984). exceptions (Alderson et al., 1994; Langley et al., 1987), the studies examined views of professionals In five studies, respondents were asked whether from specialist centres who were sometimes they would wish to participate in the proposed already involved with research. trial should the described scenario be realised (Bevan et al., 1993; Flanery et al., 1978; Gerard Fourteen studies examined views on the informed et al., 1995; Mettlin et al., 1985; Slevin et al., 1995). consent process. Although informed consent was 47 Review results

regarded as a necessary precursor to randomisation Zwitter, 1994). Eighty-six per cent of respondents by over 91% of respondents (Benson et al., 1991; thought that patients rarely understand the inform- Dal-re, 1990, 1992e, 1993), up to 22% of respon- ation given to them (Blum et al., 1987), and nearly dents in the studies of Williams and Zwitter (1994) 75% of respondents thought that randomisation, and Taylor and Kelner (1987b) regularly entered in particular, is impossible to explain adequately to patients without obtaining informed consent. patients (Spaight et al., 1984). Two studies reported Approximately 60% of responding IRB members that 67 and 55% of respondents thought the thought that parental consent was required for decision to participate should be shared or left (hypothetical) trials involving children, though to the patient rather than it be the sole responsi- 85% thought that parental consent could be waived bility of the physician (Alderson et al., 1994; when the risk was minimal (Mammel and Kaplan, Taylor, 1992). 1995). Dal-re (1990) reported that nearly 70% of responding committee members thought it Five studies addressed views on possible benefits permissible for an investigator to administer trial for their patients of participating in trials (Benson drugs without informed consent but only under et al., 1991; Daughterty et al., 1995; Kodish et al., ‘special circumstances’, while 39% of respondents 1992; Penman et al., 1984; Richardson, 1986), of in the study of Blum et al. (1987) agreed that the which two also highlighted the risks of toxicity in law should permit research under such circum- Phase 1 studies (Daughterty et al., 1995; Kodish et stances. If the requirement of informed consent al., 1992). Therapeutic benefit from all active were abandoned, some physicians stated that they treatments (both experimental and standard) was would enter more patients (Taylor and Kelner, considered probable by most respondents in all 1987a,b). There was some disagreement over how studies except for those based on Phase 1 trials. informed consent should be obtained; not more Despite the concern over the real risk of toxicity in than 32% of respondents in the studies of Blum Phase 1 trials, the majority of responding physicians et al. (1987) and Williams and Zwitter (1994) felt in the two studies above thought that participants that it should always be obtained in writing, would, nevertheless, benefit psychologically. whereas Dal-re (1990, 1993) reported that > 64% of respondents thought that written consent should Five studies reported views on the effect of offering always be sought. There was also some disparity trial entry on the physician–patient relationship: among physicians over the amount of information Benson et al. (1991), Blum et al. (1987), Spaight they thought should be given to their patients et al. (1984), Taylor and Kelner (1987b), and before entry in a trial. Indeed, Taylor and Kelner Taylor et al. (1984). All studies except Spaight (1987b), showed that some physicians had difficulty et al. (1984) showed that there was considerable in assessing their patients’ desire for detailed concern among physicians over the effect of information, and, while most physicians routinely discussing trial entry on the trusting relationship disclosed full details of the trial (Williams and with their patients. The exception (based on a Zwitter, 1994), 83% thought it possible to give more small sample) found that over 75% of respondents information than the patients actually wanted thought that randomisation did not, in fact, (Benson et al., 1991). The study of Penman et al. lessen patient’s trust in their physician (Spaight (1984) estimated this to be the case for 41% of et al., 1984). patients. In light of such uncertainty, two studies reported favourable attitudes to a more individual- Finally, four studies examined physician’s views ised approach to the consent process: Benson et al. on equipoise. Lilford (1994), using hypothetical (1991) and Taylor and Kelner (1987a). However, scenarios, established that obstetricians were one study found that 90% of respondents thought frequently in personal and collective equipoise that a ‘minimum’ amount of information should over the effects of two treatments which were to be always be disclosed. The quality of the informed compared in a proposed trial. Taylor and Kelner consent process in practice was addressed in five (1987a) reported that 36% of respondents would studies. Forty-seven per cent of physicians thought be prepared to enter their patients in a trial even that their patients were never aware that they were when personal (effective) equipoise did not apply, participants in an experiment when they signed the while Alderson et al. (1994) reported that only consent form (Taylor and Kelner, 1987b), while 25% of physicians thought that they could ever be most physicians believed that informed consent was in equipoise. Interestingly, only 28% of surgeons obtained more rigorously in Phase 1 trials than in in one study (Marsden et al., unpublished) were Phase 2 or 3 studies (Kodish et al., 1992), and in ‘uncertain’ about whether or not HRT could cause trials of supportive care rather than in trials testing a recurrence of cancer, while 73% of the same 48 curative or palliative therapies (Williams and sample thought that a trial would be ethical. Health Technology Assessment 1998; Vol. 2: No. 15

Qualitative study results are desperate (Twomey, 1994). The remaining Studies under review study asked professionals whether or not trial There were eight studies of a more qualitative results should be shared with the participating nature which sought to produce in-depth opinions patients after trial completion and uncovered through more open-ended inquiry. Three studies mixed views: the offer of such information might examined the views of healthcare professionals encourage recruitment and serve as a ‘reward’ for (Garcia, 1987; Madden, 1994; Twomey, 1994) while participation, while, at the same time, there was the remainder solicited views of patients and the doubt over the patient’s capacity to understand the public (two studies examined both: Madden, 1994; data, especially if the treatment which turned out Twomey, 1994). to be less effective or harmful had been given (Garcia, 1987). Views of patients and the public Two studies reported that ethnic minority groups Discussion considering enrolment in a hypothetical trial and The methods used by the various studies we participants in a real trial for multiple sclerosis have reviewed to evaluate the views of patients, tended to have altruistic motivations (Roberson, physicians and the public to clinical trials might 1994 and Wynne, 1989) but also expected psycho- be flawed in a variety of ways. logical and physical benefit (Wynne, 1989). Another study showed that pregnant women, Firstly, there has been methodological concern who were offered entry into a trial of holding their over the use of hypothetical scenarios in such own obstetric case notes versus liaising with a studies for some time (King, 1986), and it has number of different healthcare professionals who been suggested that respondents answering hypo- were privy to their notes, were pleased to enrol thetically may be more prone to tailor their answers and mostly felt pleased to help others; none in a direction that they consider socially desirable. expressed concern that they were part of an RCT We included studies using both hypothetical (Elbourne, 1987). This was supported by the study scenarios and those based on patients in real trials. of Madden (1994) who showed that women with In the event, we found little difference between breast cancer expressed positive views on the these groups, for example the proportion of abstract notion of research but were less motivated respondents citing altruistic reasons for partici- when it came to personal participation in trials pating in trials was only slightly (and not signifi- because of concerns about uncertainty, random- cantly) higher among those in hypothetical isation, and loss of control. This concern was scenarios. Nevertheless, actually taking part in a echoed by the remaining two studies which sought trial may change a person’s views. to obtain views of parents whose children had been entered in the ECMO (artificial lung) (Snowdon et Some studies of patients in real trials confined their al., 1997) or AIDS (Twomey, 1994) trials. In both attention to those who had actually participated cases, there was some concern over randomisation whereas it is important to obtain the views of all as a method of allocating treatment (involving patients offered entry in the trial, whether or not ‘desperate’ patients), given the lack of effective they consented to take part. The effect of offering standard treatments as alternatives to more trial entry among consenters may be systematically experimental methods. However, respondents different from that among refusers. recognised that, when equipoise applies, randomisation may resolve an otherwise difficult (finely The framing of questions and the order in which balanced) decision. Another study discussed (by they are presented differed from study to study. focus groups) a trial where the new therapy was Although some ideal methodological standards only available within a trial and found that, even can be specified, others are open to debate. Face- when the patients were not desperate, they would to-face interviewing, telephone interviewing and still prefer a three-way choice, that is, treatment self-administered questionnaires may elicit differ- A, treatment B, or a trial of A versus B (Marsden ent responses. That said, the methodological et al., unpublished). rigour of the studies was generally suboptimal, for example only a handful reported that they Views of healthcare professionals had conducted a pilot test (Benson et al., 1991; Two of the three studies showed that health Cassileth et al., 1982; Slevin et al., 1995; Spaight, professionals, while expressing positive views on 1984; Taylor et al., 1984, 1987; Taylor and Kelner, the abstracted notion of research, were concerned 1987a; Wilcox and Schroer, 1994) and only one about uncertainty, randomisation and loss of con- study assessed their questionnaire for readability trol (Madden, 1994), especially when the patients (Corbett et al., 1996). 49 Review results

Response to the surveys (especially those using side-effects seemed to be more clearly understood postal questionnaires) is also a potential problem: (Daughterty et al., 1995; Harth and Thong, 1995; 13 studies failed to achieve a response rate of 70% Hassar and Weinstein, 1976; Jensen et al., 1993; or above. In addition, 16 studies did not state the Lynoe et al., 1991). Both the qualitative studies also response rate. If the response rate is low, then there reported a limited grasp of the concept of a trial is a risk of introducing systematic bias since non- (particularly the concept of randomisation) and respondents may have different views. sometimes missed the whole idea of comparison (Madden, 1994; Snowdon et al., 1997). Misunder- In addition, only five studies used random sampling standing and ignorance might affect recruitment techniques to achieve as representative a sample or uninformed compliance, so it is very important as possible. However, 15 studies used ‘quota’ that patients and the general public are given sampling and deliberately took certain numbers accurate and comprehensible information. from specified groups, sometimes with a mind for comparison. An example of ‘quota’ sampling is the The majority of studies sought only to elicit views study of Corbett et al. (1996) who selected respon- of respondents and did not attempt to produce dents from three different groups, that is, medical a scale that might serve as a measure of views in secretaries, students and the general public. Even other sample populations. However, the series of though the consequences of haphazard sampling studies by Taylor and colleagues sought to draw are not easily determined, we would have more up a ‘physician orientation profile’ from the results confidence in the data if random sampling of physician views on clinical trials but omitted techniques were more common. to assess the properties of such measures by conventional psychometric methods. Qualitative studies are important in this subject because they provide some insight into what people are Despite a considerable range in the findings, really thinking. This is especially valuable in establish- 5–99% of respondents, overall slightly more people ing what the main issues are so that a tight set of ques- participate for personal gain than for altruistic tions can be devised for subsequent study. However reasons, even in Phase 3 studies. In six out of 14 there are problems associated with encoding quali- studies (where both types of motivation were tative responses. We think that a combined approach measured), the number of self-interested respon- using qualitative and quantitative methods would be dents outweighed the number of altruistic respon- particularly appropriate for this complex topic. dents. Patients in Phase 3 studies only stand to benefit clinically individually if their overall care is There is an interesting issue concerning framing better or if one of the treatments is superior to the effects which is particularly pertinent for studies other(s), in prospect. Since we think that standards using hypothetical scenarios. The context within of care should be independent of trial participation which a survey is conducted might have some and since comparator treatments should offer effect on the responses given. For example, Kemp equal expected utilities to patients (Brewin and et al. (1984) described hypothetical treatment Bradley, 1989; Lilford and Jackson, 1995), personal choices in a non-trial context prior to the questions gain should not be an aim of participation in a concerning trial entry and this ‘priming’ may have trial, given free availability of treatments. The resulted in lower stated willingness to accept finding that so many people hope to gain from trial randomisation than would have occurred had entry, therefore, makes us uneasy. It is important the trial been mentioned at the outset. that patients both understand and accept that, provided equipoise applies, they stand neither to As a final methodological point, it is impossible lose nor gain by participation. Patients who have to tell, in many cases, whether the respondents nothing to lose or gain, in prospect, are exhibiting understood the same thing as the investigators by a form of altruism, albeit a weak form, by their the term ‘clinical trial’. However, some of the participation. We think that well informed people, quantitative studies and qualitative studies exam- if accepting trial entry, should do so for altruistic ined knowledge and understanding as well as views reasons, or because they are indifferent between on clinical trials which go some way to resolving the participating and declining, and not because they issue (see the conclusion, page 51). These studies expect to gain clinically in prospect. If they expect found that the more scientific aspects of trials such to lose, but nevertheless wish to enter a trial, we as the concept of randomisation were, in general, would say they are manifesting a strong form of poorly understood (DeLuca et al., 1995; Penman, altruism. It is dangerous for healing professions to 1984; Rodebhuis et al., 1984; Tindall et al., 1994) encourage this. However, it seems entirely appro- 50 though specific details of the therapies such as priate at least to give people an opportunity to Health Technology Assessment 1998; Vol. 2: No. 15

manifest a weak form of altruism (or indifference), public and physicians in the studies reviewed all given equivalent expected utilities from the various expected trial participants to benefit therapeutic- comparator treatments. In the philosophical ally from all active treatments except for those in literature there is an order of both chronology and Phase 1 trials. Even in the latter case, participants emphasis on the self, with the concept of Christian were still expected to benefit psychologically from concern for one’s fellow men taking second place. the trial experience. Interestingly, the proportion Even in the religious teachings, where the latter of altruistic respondents was not greater in the gained most currency, ‘loving thy neighbour’ was Phase 1 studies than in the others. foreshadowed by a love of God. The possibility of genuine altruism has been hotly debated, some The issue of informed consent was dealt with by arguing that, at best, altruism is a form of self- many of the studies. The uncertainty expressed by realisation (in the Hegelian sense) and, at worst, the physicians over how to approach the informed self-indulgent martyrdom. Others such a David consent procedure was mainly born of perceived Hume are more sympathetic to the notion. Indeed, variation in the quantity of material desired by the Kant would object to using oneself merely as the patients themselves. Even when the information means to an end just as much as using someone was provided, it was not a foregone conclusion that else, though it seems to us that this ‘formula of the patients would actually base their decision on humanity’ does not rule out the possibility of altru- the material given. Surprisingly, physicians were istic behaviour. This substantiates the notion that more cavalier about obtaining informed consent the classical act-utilitarianism doctrine which than is demanded by the Declaration of Helsinki requires of an individual that he/she always pro- (1964). Sometimes physicians reported not even motes overall happiness (not just his/her own) is telling patients they were in a trial, let alone giving too burdensome. Rather, we believe that, in order them sufficient information, and, in one study, 47% to respect a patient’s autonomy, the expected con- of physicians thought that their patients were never sequences of entering a trial for that patient should aware that they were about to participate in an be underlined and that any altruism could be experiment when signing the consent form (Taylor factored into a decision analysis to determine and Kelner, 1987b). This result could be attributed whether equipoise applies. to the variation of practice across countries, but it engenders concern nevertheless. Views on the The proportion of patients/members of the amount of information required to make patients’ public who thought that doctors would put people consent morally significant are varied but our in trials even if equipoise was not present ranged results suggest that much more care is needed. from 26 to 70% according to Millon-Underwood (1993) and Cassileth et al. (1982), respectively; both Finally, some studies solicited views on whether studies used hypothetical trial scenarios. Such vari- experimental treatments should only be available ation might be explained, in part, by the people within the trial mechanism. Minogue et al. (1995) sampled: an exclusively African American sample has recently highlighted the plight of the ‘des- was used by Millon-Underwood (1993) who perate volunteer’, in the context of AIDS where produced the more promising figure, and a sample the absence of standard and effective treatment from the general public was used by Cassileth et al. favours the experimental therapies. Since most (1982). This brings out the issue of trans-cultural patients participate in trials out of self-interest, applicability of studies, which could reflect differ- Minogue suggests that prospective participants ent levels of familiarity with, and understanding of, should be provided with the choice to participate the trials in question. Without more data, this issue in the trial or to receive the experimental cannot be resolved. Still more worrying, however, treatment. Indeed, a large proportion of AIDS some physicians themselves confessed that they patients in one study (Tindall et al., 1994) and would be prepared to enter patients even when parents of very ill children in two others (Snowdon not equipoised but when the trial treatment were et al., 1997; Twomey, 1994) thought that unproved freely available (Taylor and Kelner, 1987a). Here, medications should be available to sufferers without the duty of physician–scientist seems to take their having to participate in a trial. precedence over the duty of care – a situation we deplore. Conclusion More people participate in trials out of self-interest Even when altruism was the main motivating factor, than altruism. Professionals and the public alike most respondents actually expected some tangible see informed consent as a necessary safeguard in benefit despite viewing themselves as secondary to trials, though sometimes they have a lax attitude the major beneficiary, the scientists. Patients, the in practice. 51

Health Technology Assessment 1998; Vol. 2: No. 15

Chapter 5 Recommendations

• The caring professions must articulate clear, rationale for conducting trials and the concept ethical justification for trials if public confidence of patient equipoise should become part of the is to be retained. Currently, this is confusing public understanding of science. and contradictory. • Practitioners should pay particular attention to • Patients should not lose out in prospect by explaining abstract ideas (especially that of taking part in a trial. randomisation) during the invitation to partic- • Given treatments which are generally available, ipate, since it is the conceptual scientific basis of patients do not lose out in prospect when prior trials rather than details of the treatments estimates of effectiveness and values interact to themselves which patients find hard to grasp. produce equal expected utilities, a situation • Small trials of existing therapies are not sometimes described as patient equipoise. necessarily unethical in terms of the participants • When treatments are not generally available, themselves, provided that they are in equipoise – patients do not lose out by participating in trials small trials may be poor value for transferable when the expected utility of the new treatment is resources, however. at least as high as that of standard treatment. • Clinical trials should start early in the life of a • The term ‘uncertainty’ prevaricates on prior new treatment (even if further refinement is probabilities and values, making it an inadequate likely in prospect) since equipoise may be lost moral basis for trials. It should not be used to quickly on the basis of observational data, and disguise such existing data as may affect patient the results may be analysed after stratifying by preferences, even when such data are treatment variables. insufficient to engender ‘certainty’. • The much quoted idea that patients in trials • Patients must be given as much information as do better than average, even when the trial they need to bring their values into play (and produces a negative result, may be true. If the hence respect their autonomy), even though this effect is real, it would seem to come from may increase anxiety temporarily and depress enhanced attention to detail inherent in recruitment rates. following the trial protocol for both control and • Patients are least alarmed and understand the experimental groups. It should not, however, issues most clearly when they have encountered be used as an inducement to accept random- the concept of comparative trials before. Since isation since the Helsinki accord requires that hundreds of thousands of patients are entered in the intention should be to provide the ‘best’ trials each year in a country like the UK, the care for all patients.

Health Technology Assessment 1998; Vol. 2: No. 15

Chapter 6 Questions for future research

reas in ethics of conducting RCTs which have not more explicit and what effects would this A attracted little or no attention, and so need to have on (different groups of) patients? (See the be further analysed, include: results section, pages 41 and 44.) • How do practitioners (and others) form prior • ethical issues in the design and conduct beliefs about treatment effects and how accurate of cluster trials do they turn out to be? (See the results section, • ethical issues in interim analysis page 41.) • the conduct and constitution of • How do the public view post-randomisation ethics committees. consent in circumstances where a promising new treatment is not available outside a trial? There are a number of empirical questions (topics (See the results section, page 41.) in quantitative ethics) which also need to be • How do the general public view cluster trials addressed, and these include: and interventions? (See the results section, page 41.) • What is it about abstract trial concepts such as • How does disclosure of interim trial data (or randomisation that patients find difficult to externally published data) affect recruitment grasp? (See the results section, page 36.) rates to trials? (See the results section, page 41.) • How might patient understanding be maximised • What are the public views on the obligations of in a cost-effective way? (See the results section, data monitoring committees? (See the results page 36.) section, page 41.) • In particular, how might practitioners best • What principles do data monitoring committees use decision analysis (or techniques derived follow when deciding if and when to stop a trial therefrom) to make the choice to participate or early? (See the results section, page 41.)

Health Technology Assessment 1998; Vol. 2: No. 15

Acknowledgements

his study was supported by the NHS R&D We are indebted to the referees for their T Executive’s Health Technology Assessment perseverance in reading the report and the Programme. quality of their comments.

Health Technology Assessment 1998; Vol. 2: No. 15

References

Aaronson NK, Visserpol E, Leenhouts GHMW, Antman EM, Lau J, Kupelnick B, Mosteller F, Chalmers Muller MV, Vanderschot ACM, Vandam FSAM, et al. TC. A comparison of results of meta-analyses of Telephone-based nursing and intervention improves randomized controlled trials and recommendations of the effectiveness of the informed consent process in clinical experts. Treatment for myocardial infarction. cancer clinical trials. Journal of Clinical Oncology JAMA 1992;268:240–8. 1996;14(3):984–96. Antman K. A comparison of randomized vs concurrent Abraham EP. Chain E, Fletcher CM. Further observations controls. Proceedings of the American Association of Cancer on penicillin. Lancet 1941;ii:177–88. Research 1983;24:146. Ackermann TF. Medical research, society and health care Armitage P. Sequential medical trials: some comments ethics. In: Gillon R, editor. Principles of health care on FJ. Anscombe’s paper. American Statistical Association ethics. Chichester: Wiley, 1994:873–84. Journal 1963;58:384–7. Albert R, Raiffa H. A progress report on the training of probability assessors. Unpublished findings. Harvard Armitage P. Sequential medical trials. Oxford: University, 28 August 1969. Blackwell, 1975. Alderson P. Equipoise as a means of managing Ashby D, Machin D. Stopping rules, interim analysis and uncertainty: personal, communal, and proxy. Journal of data monitoring committees. British Journal of Cancer Medical Ethics 1996;22:135–9. 1993;68:1047–50. Alderson P, Madden M, Oakley A, Wilkins R. Women’s Ashcroft JJ, Leinster SJ, Slade PD. Breast cancer – patient views of breast cancer treatment and research: report of choice of treatment: preliminary communication. Journal pilot project 1993. London: SSRU, 1994. of the Royal Society of Medicine 1985;78:43–6. Altman DG. Statistics and ethics in medical research III. Autret E, Dutertre JP, Barbier P, Jonville AP, Pierre F, How large a sample? British Medical Journal Berger C. Parental opinions about biomedical research 1980;281:1336–8. in children in Tours, France. Developmental Pharmacology Altman DG. Size of clinical trials. British Medical Journal and Therapeutics 1993;20(1–2):64–71. 1983;286:1842–3. Avorn J, Soumerai SB. Improving drug therapy decisions Altman DG, Gardner MJ. Confidence intervals for through educational outreach: a randomized controlled research findings. British Journal of Obstetrics and trial of academically based ‘detailing’. New England Gynaecology 1992;99(2):90–1. Journal of Medicine 1983;308:1457–63. Ambroz A, Chalmers TC, Smith H. Deficiencies of Avorn J, Chen M, Hartley R. Scientific versus commercial randomised controlled trials. Clinical Research sources of influence on the prescribing behaviour of 1978;26:280a. physicians. American Journal of Medicine 1982;3:4–8. Anderson B, Baden. The controlled clinical trial in Banta HD, Thacker SB. The case for reassessment of surgery. Scandinavian Journal of Gastroenterology health care technologies. Once is not enough. JAMA 1971;6(9):23–9. 1990;364:235–46. Anderson RP. Standards for surgical trials. Annals of Barnard GA. Must clinical trials be large? The interpret- Thoracic Surgery 29:192, 1980. ation of p-values and the combination of test results. Anello C, Fleiss J. Exploratory or analytic meta-analysis: Statistics in Medicine 1990;9:601–14. should we distinguish between them? Journal of Clinical Barofsky I, Sugarbaker PH. Determinants of Epidemiology 1995;48(1):109–16. nonparticipation in randomized clinical trials for the Angell M. Patients’ preferences in randomized clinical treatment of sarcomas. Cancer Clinical Trials trials. New England Journal of Medicine 1984;310:1385–7. 1979;2:237–46. Anonymous. Controlled clinical trials: planned Barry M, Molyneux M. Ethical dilemmas in malaria drug deception? Lancet 1979;i:534–5. and vaccine trials: a bioethical perspective. Journal of Anonymous. Positive discrimination for surgery? Medical Ethics 1992;18(4):189–92. Lancet 1990;336:151. Bartlett RH, Cornell RG. Comment on ethics and Anscombe F. Sequential medical trials. Journal of the statistics in randomized clinical trials. Statistical Science American Statistical Association 1963;58:365–83. 1991;6(1):63–5, 1991. 59 References

Baum M. Commentary on the practical and ethical Bleich HL. The science and ethics of the randomized defects of surgical randomized prospective trials. clinical trial. MD Computing 1995;12(1):5–6. Journal of Medical Ethics 1983;9:92–3. Block JB, Elahoff RM. The randomized clinical trial in Baum M. The ethics of clinical research. In: Byrne P, evaluation of new cancer treatment. Methods in Cancer editor. Ethics and law in health care 1. Chichester: Research 1979;xvii:259–75. Wiley, 1990. Blum A.L, Chalmers TC, Deutch E, Koch-Weser J, Rosen Baum M. Clinical trials – a brave new partnership: a A, Tygstrup N, et al. The Lugano statement on controlled response to Ms. Thornton. Journal of Medical Ethics clinical trials. Journal of International Medical Research 1994;20:23–5. 1987;15:2–22. Baum M. The ethics of clinical trials. Israel Journal of Blumgart HL. The medical framework for viewing the Medical Science 1995;31:197–200. problem of human experimentation. In: Freund PA, Baum M, Houghton J, Abrams K. Early stopping rules – editor. Experimentation with human subjects. clinical perspectives and ethical considerations. Statistics 1970:39–65. in Medicine 1994;13:1459–69 and 1471–2. Boissel JP. Impact of randomized clinical trials on Beauchamp TL, Childress JF. Principles of biomedical medical practice. Controlled Clinical Trials ethics. New York: Oxford University Press, 1989. 1989;10:1205–345. Beecher HK. Ethics and clinical research. New England Bonchek LI. Sounding board: are randomized trials Journal of Medicine 1966;274:1354–60. appropriate for evaluating new operations? New England Journal of Medicine 1979;301:44–5. Bellissant E, Benichou J, Chastang C. A comparison of methods for phase ii cancer clinical trials: advantages of Botros S. Equipoise, consent, and the ethics of the triangular tests, a group sequential method. Lung randomised controlled trials. In: Byrne P, editor. Ethics Cancer 1994;10(1):S105–15. and law in health care. Chichester: Wiley, 1990:9–24. Benson AB, Pregler JP, Bean JA, Rademaker AW, Eshler Bracken MB. Clinical trials and the acceptance of B, Anderson K. Oncologists reluctance to accrue patients uncertainty. British Medical Journal 1987;294:1111–2. onto clinical trials: an Illinois center study. Journal of Bradley C. Designing medical and educational Clinical Oncology 1991;9(11):2067–75. intervention studies. A review of some alternatives to Bergler JH, Pennington AC, Metcalfe M, Freis ED. conventional randomised controlled trials. Diabetes Care Informed consent: how much does the patient 1993;16:509–18. understand? Clinical Pharmacology and Therapeutics Brewin CR, Bradley C. Patient preferences and 1980;27(4):435–40. randomized clinical trials. British Medical Journal Berglund CA. Children in medical research: Australian 1989;229:313–5. ethical standards. Child: Care, Health, and Development Brown B. Statistical controversies in the design of clinical 1995;21(2):149–59. trials – some personal views. Controlled Clinical Trials Berkowitz S. Ethical issues in the case of surgical repair 1980;1:13–27. of cleft palate. Cleft Palate – Cranial Facial Journal Brown BW, Herson J, Atkinson EN, Rozell ME. Projection 1995;32(4):271–81. from previous studies: a Bayesian and frequentist Berry DA, Chih-Hsiang Ho. One-sided sequential compromise. Controlled Clinical Trials 1987;8:29–44. stopping boundaries for clinical trials: a decision- Bruera E. Ethical issues in palliative care research. theoretic approach. Biometrics 1988;44:219–27. Journal of Palliative Care 1994;10:7–9. Berry DA, Wolff MC, Sack D. Public health decision- Bulpitt CJ. Randomised controlled clinical trials. making: a sequential vaccine trial. Bayesian Statistics The Hague: Nijhoff, 1983. 1992;4:79–96. Burkhardt R, Kienle G. Controlled clinical trials and Bertelsen K. Protocol allocation and exclusion in two medical ethics. Lancet 1978;1356–9. Danish randomized trials in ovarian cancer. British Journal of Cancer 1991;64:1172–6. Burkhardt R, Kienle G. Basic problems in controlled trials. Journal of Medical Ethics 1983;9(2):90–3. Bevan EG, Chee LC, McGhee SM, McInnes GT. Patients’ attitudes to participation in clinical trials. British Journal Buyse M. Randomized clinical trials in surgical oncology. of Clinical Pharmacology 1993;35(2):204–7. European Journal of Surgical Oncololgy 1991;17(5):421–8. Blackwelder WC. ‘Proving the null hypothesis’ in clinical Byar DP. The necessity and justification of randomized trials. Controlled Clinical Trials 1982;3:345–53. clinical trials. In: Tagnon, Staquet, editors. Controversies in cancer. New York: Masson, 1979:15–20, 75–82. Bleich HL. Experimentation on human subjects. In: Rosner F, Bleich JD, editors. Jewish bioethics. New York: Byar DP. Problems with using observational databases to 60 Sanhedrin Press, 1979:384–5. compare treatments. Statistics in Medicine 1991;10:663–6. Health Technology Assessment 1998; Vol. 2: No. 15

Byar DP, Simon RM, Friedewald WT, et al. Randomized Chalmers I. Letter to the editor. New England Journal of clinical trials. Perspectives on some recent ideas. New Medicine 1991;21:1514. England Journal of Medicine 1976;295:74–80. Chalmers I, Chalmers TC. Randomisation and patient Byar DP, Schoenfield DA, Green SB, et al. Design choice. Lancet 1994;344:892–3. considerations for AIDS trials. New England Journal of Medicine 1990;323:1343–8. Chalmers I, Silverman WA. Professional and public double standards on clinical experimentation. Controlled Byer A. The practical and ethical defects of surgical Clinical Trials 1987;8:388–91. randomised prospective trials. Journal of Medical Ethics 1983;9:90–3. Chalmers I, Sinclair JC. Promoting perinatal health: is it time for a change of emphasis in research? Early Human Campbell KA. Knots in the fabric: Richard Pearson Development 1985;10:171–91. Strong and the Bilibid Prison vaccine trials, 1905–1906. Bulletin of the History of Medicine 1994;68:600–38. Chalmers TC. When should randomization begin? Canner PL. In: Freund. PA, editor. Experimentation with Lancet 1968:858. human subjects. New York: George Braziller, 1970. Chalmers TC. Impact of controlled trials on practice of Canner PL. Further comment on “statistical inference medicine. Mount Sinai Journal of Medicine 1974;41:753–9. from clinical trials: choosing the right p-value”. Controlled Chalmers TC. Randomization of the first patient. Medical Clinical Trials 1983;4(1):35. Clinics of North America 1975a;59:1035–8. Caplan A. Is there a duty to serve as a subject in Chalmers TC. Ethical aspects of clinical trials. American biomedical research? IRB: Human Subjects Research Journal of Opthalmology 1975b;79:753–8. 1984;6:1–5. Cappelleri JC, Trochim WM. Ethical and scientific Chalmers TC. Randomized clinical trials in surgery. In: features of cutoff-based designs of clinical trials. Medical Delaney JP, editor. Controversy in surgery. Philadelphia: Decision Making 1995;15:387–94. Saunders, 1976:3–11. Cappelleri JC, Ioannidis JP, Schmid CH, de Ferranti SD, Chalmers TC. Invited remarks on terminating a long- Aubert M, Chalmers TC, et al. Large trials vs meta-analysis term clinical trial. Clinical Pharmacology and Therapeutics of smaller trials: how do their results compare? JAMA 1979;25(5):649–50. 1996;276:1332–8. Chalmers TC. A potpourri of RCT topics. Controlled Capron AM. Human experimentation: basic issues. In: Clinical Trials 1982;3:285–98. Reich, editor. Encylopedia of bioethics. New York: Free Press, 1978:692–9. Chalmers TC. Ethical implications of rejecting patients for clinical trials. JAMA 1990;263:865. Capron AM. Protection of research subjects: do special rules apply in epidemiology? Journal of Clinical Chalmers TC. Lau J. Meta-analytic stimulus for changes Epidemiology 1991;44:184–90. in clinical trials. Statistical Methods in Medical Research 1993;2(2):161–72. CASS Principal Investigators and their Associates. Coronary Artery Surgery Study (CASS): a randomized Chalmers TC, Lau J. Changes in clinical trials mandated trial of coronary artery bypass surgery. Comparability of by the advent of meta-analysis. Statistics in Medicine entry characteristics and survival in randomized and non- 1996;15(12):1263–8,1269–72. randomized patients meeting the randomization crtieria. Journal of the American College of Cardiology 1984;3:114–28. Chalmers TC, Block JB, Lee S. Controlled studies in clinical cancer research. New England Journal of Medicine Cassel CK. Research in nursing homes: ethical issues. 1972;287:75–8. Journal of the American Geriatrics Society 1985;33:795–9. Chalmers TC, Levin H. Sacks HS. Reitman D, Berrier J, Cassileth BR, Lusk EJ, Miller DS, Hurtwitz S. Attitudes Nagalingam R. Meta-analysis of clinical trials as a toward clinical trials among patients and the public. scientific discipline. I: control of bias and comparison JAMA 1982;248:968–70. with large co-operative trials. Statistics in Medicine Cassileth BP, Knuiman MW, Abeloff MD, Falkson G, 1987;6:315–28. Ezdinli EZ, Mehta CR. Anxiety levels in patients Chalmers TC, Hewett P, Reitman D, Sacks HS. Selection randomized to adjuvant therapy vs observation for early breast cancer. Journal of Clinical Oncology 1986;4:972–4. and evaluation of empirical research in technology assessment. International Journal of Technology Assessment in Cavan DH. Ethics of clinical trials of ineffectual Health Care 1989;5:521–36. therapies. IRB: A Review of Human Subjects Research 1981;3(5):10–11. Chard JA, Lilford RJ. The use of equipoise in clinical trials. Social Science and Medicine 1998;47(7):891–8. Challah S, Mays NB. The randomised controlled trial in the evaluation of new technology: a case study. British Charlton BG. Randomized clinical trials: the worst kind Medical Journal 1986;292:877–8. of epidemiology? Nature Medicine 1995;1:1101–2. 61 References

Chipman JJ. Study design for final height determination Dal-re R. Informed consent in clinical research with in Turner syndrome: pros and cons. Hormone Research drugs in Spain: perspectives of clinical trials committee 1993;39(2):18–22. members. European Journal of Clinical Pharmacology 1990;38:319–24. Clarke A. Problems of clinical research in dementia. In: Wilcock GK, editor. The management of Alzheimer’s Dal-re R. Clinical trial of drugs: a study of the influence disease. Petersfield: Wrighton Biomedical Publishing, of information on adverse reactions on obtaining of 1993:161–73. informed consent. Medicina Clinica 1991;96(15):566–9. Dal-re R. Elements of informed consent in clinical Clayton D. Ethically optimized designs. British Journal of research with drugs: a survey of Spanish clinical Clinical Pharmacology 1982;13:469–80. investigators. Journal of Internal Medicine Cleophas TJ. Clinical trials: specific problems associated 1992;231(4):375–9. with the use of a placebo-control group. Journal of Dal-re R. Good clinical practice in clinical trials: the Molecular Medicine 1995;73:421–4. responsibilities of the investigator. A survey of 827 Cohen J. Clinical trial monitoring: hit or miss? hospital physicians (and II). Patient consent. Medicina Science 1994;264:1534–7. Clinica 1993;100(11):423–7. Collier J. Confusion over the use of placebos in clinical Daughterty C, Ratain MJ, Grochowski E, Stocking C, Kodish E, Mick R, et al. Perceptions of cancer patients trials. British Medical Journal 1995;311:821–2. and their physicians involved in phase i trials. Journal of Collins R, Gray R, Godwin J, Peto R. Avoidance of large Clinical Oncology 1995;13(5):1062–72. biases and large random errors in the assessment of Davis S, Wright PW, Schulman SF, Hill LD, Pinkham RD, moderate treatment effects: the need for systematic Johnson LP, et al. Participants in a prospective, random- overviews. Statistics in Medicine 1987;6:245–50. ized clinical trials for resected non-small cell lung cancer Collins R, Doll R, Peto R. Ethics of clinical trials. In: have improved survival compared with non-participants Williams CJ, editor. Introducing new treatments for in such trials. Cancer 1985;56:1710–8. cancer: practical and legal problems. Chichester: Wiley, Davis S, Seay J, Stone J, Bagley C, Johnson J, Speyer J, 1992:49–66. Smith S. Does knowing about clinical trials increase participation? Proceedings of the American Society of Clinical Corbett F, Oldham J, Lilford R. Offering patients entry in Oncology 1990;9:A232. clinical trials: preliminary study of the views of prospective participants. Journal of Medical Ethics 1996;22:227–31. Davis S, Seay J, Stone J, et al. Final report of ‘Does knowing about clinical trials increase participation?’ Cornfield J. Sequential trials, sequential analysis and the Unpublished. likelihood principle. American Statistician 1966;20:18–23. Day NE. Two-stage designs for clinical trials. Biometrics Coulehan JL, Schaffner KF, Block M. Ethics of clinical 1969;25:111–18. trials in family medicine. Journal of Family Practice 1985;21:217–22. DCCT Research Group. Implementation of a multi- component process to obtain informed consent in a Cowan DH. The ethics of clinical trials of ineffectual diabetes control and complications trial. Controlled therapy. IRB: A Review of Human Subjects Research Clinical Trials 1989;10:83–96. 1981;3(5):10–11. De Bono M, Fawdry RDS, Lilford RJ. Size of trials for Crisp R. Routledge philosophy guidebook to Mill on evaluation of antenatal tests of fetal well-being in high utilitarianism. London: Routledge, 1997. risk pregnancy. Journal of Perinatal Medicine 1990;18:77–87. Crowley J, Green S, Lui P Y, Wolf M. Data monitoring committees and early stopping guidelines: the Southwest De Deyn PP. On the ethical acceptability of placebo Oncology Group experience. Statistics in Medicine application in neuropsychiatric research. Acta Neurologica 95 1994;13:1391–9. Belgica 1995; :8–17. de Groot JM, Kennedy SM. Integrating clinical and Curran WJ. Reasonableness and randomization in research psychiatry. Journal of Psychiatry and Neuroscience clinical trials: fundamental laws and goverment 1995;20(2):150–4. regulations. New England Journal of Medicine 1979;300:1273–5. DeLuca SA, Korcuska LA, Oberstar BH, Rosenthal ML, Welsh PA, Topol EJ. Are we promoting true informed Cutler SJ, Greenhouse SW, Cornfield J, et al. The role of consent in cardiovascular clinical trials? Journal of hypothesis testing in clinical trials. Journal of Chronic Cardiovascular Nursing 1995;9(3) 54–61. Diseases 1966;19:857–82. Detsky AS, Naylor CD, O’Rourke K, MeGeer AJ, L’abbe Dagan R. Clinical studies in pediatric respiratory-tract KA. Incorporating variation of the quality of individual infections: what is desirable, feasible, or simply randomized trials into meta-analysis. Journal of Clinical 62 misleading? Chemotherapy 1992;38:18–24. Epidemiology 1992;45:255–65. Health Technology Assessment 1998; Vol. 2: No. 15

Donner A. The use of auxiliary information in the design Elander G, Hermon G. Placebo effect and randomized of a clinical trial. Biometrics 1977;33(2):305–14. clinical trials. Theoretical Medicine 1995;16(2):171–82. Dudley HAF. The controlled clinical trial and the Elbourne D. Subjects’ views about participation in a advance of reliable knowledge: an outsider looking in. randomised controlled trial. Journal of Reproductive and British Medical Journal 1983;287:957–60. Infant Psychology 1987;5:3–8. Dudley HAF. Stones, lithotripters, trials, and arguments. Elks ML. The right to participate in research studies. British Medical Journal 1986;292:846–7. Journal of Laboratory and Clinical Medicine 1993;122:130–6. Duley L. Magnesium sulphate regimens for women with Ellenberg SS, Cooper E, Eigo J, Finkelstein DM, Hoth eclampsia: messages from the Collaborative Eclampsia DF, Nusinoff-Lehrman S, et al. Studying treatments for Trial. British Journal of Obstetrics and Gynaecology AIDS: new challenges for clinical trials – a panel 1996;103:103–5. discussion at the 1990 annual meeting of the Society for Clinical Trials. Controlled Clinical Trials 1990;13:272–92. Duley L, Johanson R. Magnesium sulphate for pre- eclampsia and eclampsia: the evidence so far. British Elliott GR. Dilemmas facing clinicians and researchers Journal of Obstetrics and Gynaecology 1994;101:565–7. using anti-depressants to treat adolescents with depression. Journal of Child and Adolescent Dupont HL. Importance of collaborative research to Psychopharmacology 1992;2:7–9. improve world health. Journal of Infectious Diseases 1991;163(5):946–50. Emrich LJ, Sedrank N. Whether to participate in a clinical trial: the patient’s view. In: Kadane JB, editor. Dupont WD. Randomised vs historical clinical trials: are Bayesian statistics and ethics in a clinical trial design. the benefits worth the costs? American Journal of Chichester: Wiley, 1996:267–309. Epidemiology 1985;122:904–46. Engelhart HT. Diagnosing well and treating prudently: Dutertre JP, Barbier P, Jonville AP, Autret E, et al. randomized clinical trials and the problem of knowing Children’s participation in biomedical research. A truly. In: Spicker SF, Alon I, de Vries A, Engelhart HT, planned survey of 541 parents. Archives Francaises de editors. The use of human beings in research. Dordrecht: Pediatrie 1993;50(6):533–8. Kluwer, 1988:123–41. Dworkin G. Law and medical experimentation: of Epstein LC, Lasagna L. Obtaining informed consent: embryos, children and others with limited capacity. form or substance. Archives of Internal Medicine Monash University Law Review 1987;13:186–206. 1969;123:682–8. Dworkin G. The theory and practice of autonomy. Ernst E, Resch KL. Clinical trials – methodological gold Cambridge: Cambridge University Press, 1988. standard or naive reductionism? Wiener Medizinische Eclampsia Trial Collaborative Group. Which anti- Wochenschrift 1995;145(1):9–12. convulsant for women with eclampsia? Evidence from the Ernst E, Resch KL. The clinical trial – gold standard or Collaborative Eclampsia Trial. Lancet 1995;345:1455–63. naive reductionism? European Journal of Physical Medicine Editorial. Ethics, philosophy and clinical trials. Journal of and Rehabilitation 1996;6(1):26–7. Medical Ethics 1983;9:59–60. Fagot-Largeault AM. Epistemological presuppositions Editorial. Evaluation of emerging technologies for involved in the programs of human research. In: Spicker coronary revascularization. Circulation SF, Alon I, de Vries A, Engelhart HT, editors. The use of 1992;85(1):357–61. human beings in research. Dordrecht: Kluwer, 1988:161–88. Editorial. Clinical trials and clinical practice. Lancet 1993;342:877–8. Fayers P, Armitage T. Towards an international register for cancer trials: the UKCCCR register of UK trials. Editorial. Your baby is in a trial. Lancet 1995;345:805–6. European Journal of Cancer 1993;29A: 907–12. Edwards SJL. Ethical issues in the design and conduct of Fayers PM, Machin D. Sample size: how many patients cluster randomised controlled trials. In preparation. are necessary? British Journal of Cancer 1995;72:1–9. Edwards SJL, Lilford RJ, Braunholtz DA, Jackson JC. Feenberg A. On being a human subject: interest and Why ‘underpowered’ trials are not necessarily unethical. obligation in the experimental treatment of incurable Lancet 1997;350:804–7. disease. Philosophical Forum 1992;23(3):213–30. Edwards SJL, Lilford RJ, Braunholtz D. Advantages of Feinstein AR. Clinical biostatistics XXIII. The role of trials units for clinical trials. In preparation. randomization in sampling, testing, allocation and Eichler HG. Hazards of misguided ethics committees. credulous idolatry. Clinical Pharmacology and Therapeutics Lancet 1995;346:1115–6. 1973;14:898–915, 1035–51. Eick TJ, Kofoed L. An unusual indication for a single- Ferner U, Neumann N. Active control equivalence trials: subject clinical trial. Journal of Nervous and Mental Disease some methodological aspects. Psychopharmacology 1994;182:587–90. 1992;106:S93–5. 63 References

Fethe C. Beyond voluntary consent: Hans Jonas on the Freedman LS, Spiegelhalter DJ. Application of Bayesian moral requirements of human experimentation. Journal statistics to decision making during a clinical trial. of Medical Ethics 1993;19:99–103. Statistics in Medicine 1992;11:23–35. Fetter MS, Feetham SL, Apolito K, et al. Randomized Freedman LS, Lowe, Macaskill. Stopping rules for clinical trials: issues for researchers. Nursing Research clinical trials incorporating clinical opinion. Biometrics 1989;38(2):117–20. 1984;40:575–86. Fetting JH, Siminoff LA, Piantadosi S, Abeloff MD, Freedman LS, Spiegelhalter DJ, Parmar MKB. The what, Damron DJ, Sarsfield AM. Effect of patients’ expectations why and how of Bayesian clinical trials monitoring. of benefit with standard breast cancer adjuvant chemo- Statistics in Medicine 1994;11:23–35. therapy on participation in a randomized clinical trial: a Freedman N. A response to a purported ethical difficulty clinical vignette study. Journal of Clinical Oncology with randomised clinical trials involving cancer patient. 1990;8:1476–82. Journal of Clinical Ethics 1992;3(3):231–4. Fisher RL, Fisher S. Anti-depressants for children. Is Freeman PR. The role of p-values in analysing trial scientific support necessary? Journal of Nervous and Mental results. Statistics in Medicine 1993;12:1443–52. Disease 1996;184:99–102. Freiman JA, Chalmers TC, Smith H Jr, Kuebler RR. The Flanery, Gravdal, Hendrix, Hoffman, King, May, et al. importance of beta. The type II error, and sample size in Just sign here. South Dakota Journal of Medicine the design and interpretation of the RCT. New England 1978;31:33. Journal of Medicine 1978;299:690–4. Fleming TR. Historical controls, data banks, and Freireich EJ, Gehan EA. The limitations of the randomized trials in clinical research: a review. Cancer randomized clinical trial. Methods in Cancer Research Treatment Reports 1982;66(5):1101–6. 1979;XVII: 277–310. Fleming TR. Data monitoring committees and capturing Freston JW. Dose ranging in clinical trials. Rationale and relevant information of high quality. Statistics in Medicine proposed use with placebo or positive controls. American 1993;12(5–6):565–70, 571–3. Journal of Gastroenterology 1986;81:307–11. Fleming TR, DeMets DL. Monitoring of clinical trials: Freund PA. Introduction. Legal frameworks for human issues and recommendations. Controlled Clinical Trials experimentation. In: Freund PA, editor. Experimentation 1993;14(3):183–97. with human subjects. New York: George Braziller, 1970: xii–1,105–15. Fleming TR, Waterlet LF. Approaches to monitoring clinical trials. Journal of the National Cancer Institute Fried C. Medical experimentation: personal integrity and 1989;81:188–93. social policy. Amsterdam: North-Holland, 1974. Fortin PR, Lew RA, Liang MH, Wright EA, Beckett LA, Friedman L, Wenger NK, Knatterud GL. Impact of the Coronary Drug Project findings on clinical practice. Chalmers TC, et al. Validation of a meta-analysis: the Controlled Clinical Trials 1983;4:513–22. effects of fish oil in rheumatoid arthritis. Journal of Clinical Epidemiology 1995;48:1379–90. Gallet M, Aupetit JF, Servan E, Lestaevel M, Lopez M, Chassoux G. Informed consent, what information is left Fox TF. The ethics of clinical trials. Medico-legal Journal after a therapeutic trial? Archives des Maladies du Coeur et 1960;28:132–41. des Vaisseaux 1994;87(1):39–45. Freedman B. Equipoise and ethics of clinical research. Gallo C, Perrone F, De Placido S, Giusti C. Informed New England Journal of Medicine 1987;317:141–5. vs randomized consent to clinical trials. Lancet Freedman B. Suspended judgement. AIDS and the ethics 1995;346:1060–4. of clinical trials: learning the right lessons. Controlled Ganz PA. Do we need a new end point in clinical Clinical Trials 1992;13(1):1–5. trials today? Journal of the National Cancer Institute Freedman B, et al. Non-validated therapies and HIV 1989;81:1105–6. Disease. Hastings Center Report 1989;19(3):14–20. Ganz PA. Clinical trials. Concerns of the patient and Freedman B, Fuks A, Weijer C. In loco parentis: minimal the public. Cancer 1990;65:2394–9. risk as an ethical threshold for research upon children. Garcia J. Sharing research results with patients: the views Hastings Center Report 1993;23(2):13–19. of care-givers involved in a randomized controlled trial. Freedman LS. The size of clinical trials in cancer Journal of Reproductive and Infant Psychology 1987;5:9–13. research – what are the current needs? British Journal of Gehan EA. Design of controlled clinical trials: use of Cancer 1989;59:396–400. historical controls. Cancer Treatment 1982;66:1089–93. Freedman LS, Spiegelhalter DJ. Comparison of Bayesian Gehan EA, Freireich EJ. Non-randomized controls in with group sequential methods for monitoring clinical cancer clinical trials. New England Journal of Medicine 64 trials. Controlled Clinical Trials 1989;10:357–67. 1974;290:198–203. Health Technology Assessment 1998; Vol. 2: No. 15

Geller NL. Planned interim analysis and its role in Goldberg HI, McGough H. The ethics of ongoing cancer clinical trials. Journal of Clinical Oncology randomization trials. Investigation among inmates. 1987;5(9):1485–90. Medical Care 1991;29(7):JS41–8. Geller NL, Pocock SJ. Interim analyses in randomized Goldman AI. Issues in design and sequential stopping controlled trials: ramifications and guidelines for rules for monitoring side-effects in clinical trials. practitioners. Biometrics 1987;43:213–23. Controlled Clinical Trials 1987;8:327–37. George SL. Discussion of “sequential methods based on Goodare H, Smith R. The rights of patients in research. the boundaries approach for the clinical trial comparison British Medical Journal 1995;310:1277–8. of survival times”. Statistics in Medicine 1994;13:1369–70. Goodman SN, Berlin JA. The use of predicted con- George SL, Chengchang LI, Berry DA, Green MR. fidence intervals when planning experiments and the Stopping a clinical trial early: frequentist and Bayseian misuse of power when interpreting results. Annals of approaches applied to a CALGB trial in non-small-cell Internal Medicine 1994;121:201–6. lung cancer. Statistics in Medicine 1994;13:1313–27. Gerard MJ, Rouillon F, Miller HD, Dreyfus JP. Consent Goodyear M, Levine M. From methodological to ethical for clinical trials in psychiatry – results of a population issues in randomized trials, unpublished, 1987. survey. Clinical Research and Regulatory Affairs Gore SM. The consumer principle of randomisation. 1995;12(3):193–209. Lancet 1994;343:58. Gerard-Coue MJ, Rouillon F, Dreyfus JD. Enquete sur la Gracia D. Between the clinical trial and clinical ethics: participation aux essais therapeutiques en psychiatrie. good clinical practices. Medicina Clinica Encephale 1994;20:421–6. 1993;100(9):333–6. Gerbarg SB. Horwitz RI. Resolving conflicting clinical Grady C. Ethical issues in clinical trials. Seminars in trials. Guidelines for meta-analysis. Journal of Clinical Oncology Nursing 1991;7(4):288–96. Epidemiology 1988;41:503–90. Green S, Fleming TR, O’Fallon JR. Policies for study Giertz G. Ethics of randomized clinical trials. Journal of Medical Ethics 1980;6:55–7. monitoring and interim reporting of results. Journal of Clinical Oncology 1987;9:1477–84. Gifford F. The conflict between randomized clinical trials and the therapeutic obligation. Journal of Medicine and Greenhouse JB. On some applications of Bayesian Philosophy 1986;11:347–66. methods in cancer clinical trials. Statistics in Medicine 1992;11:37–53. Gifford F. Community equipoise and the ethics of RCTs. Bioethics 1995;9(2):127–48. Grimshaw JM, Russell IT. Effect of clinical guidelines on medical practice: a systematic review of rigorous Gilbaldi M. Meta-analysis. A review of its place in evaluations. Lancet 1993;342:1317–22. therapeutic decision making. Drugs 1993;46(5):805–18. Gross M. Innovations in surgery. A proposal for phased Gilbert JP. Patients must be told that treatment will be clinical trials. Journal of Bone and Joint Surgery randomised. British Medical Journal 1995;310:125. 1993;75(3):351–4. Gilbert JP, McPeck B, Mosteller F. Statistics and ethics in surgery and anesthesia. Science 1977;198:684–9. Grunkemeier GL, Starr A. Alterations to randomization in surgical studies. Journal of Heart Valve Disease Gilks WR, Thomas A, Spiegelhalter DJ. A language and 1992;1(2):142–51. programme for complex Bayesian modelling. Statistician 1994;43:169–78. Guyatt G, Sackett D, Taylor DW, Chong J, Roberts R, Pugsley S. Determining optimal therapy-randomized Gillett GR. Ethics and dental research. Journal of Dental trials in individual patients. New England Journal of Research 1994;73(11):1766–72. Medicine 1986;314:889–92. Gillon R. Philosophical medical ethics. Chichester: Guyatt GH, Jaeschke R. N of 1 randomized trials: where Wiley, 1985. do we stand? Weston Journal of Medicine 1990;152:67. Gillon R. Recruitment for clinical trials: the need for public professional co-operation. Journal of Medical Ethics Guyatt GH, Heyting A, Jaeschke R, Keller J, Adachin JD, 1994;20:3–4. Roberts RS. N of 1 randomized trials for investigating new drugs. Controlled Clinical Trials 1990a;11:88–100. Glasser SP, Clark PI, Lipicky RJ, Hubbard JM, Yusuf S. Exposing patients with chronic, stable, exertional Guyatt GH, Keller JL, Jaeschke R, Rosenbloom D, Adachi angina to placebo periods in drug trials. JAMA JD, Newhouse MT. The N-of-one randomized controlled 1991;265(12):1550–4. trial – clinical usefulness: our three year experience. Annals of Internal Medicine 1990b;112:293–9. Glasziou PP, Irwig Les M. An evidence-based approach to individualizing treatment. British Medical Journal Haines SJ. Randomized trials in the evaluation of surgical 1995;311:1356–8. innovation. Journal of Neurosurgery 1979;51:5–11. 65 References

Hall JC. The other side of statistical significance: a review Henshaw RC, Naji SA, Russell IT, Templeton AA. of type II errors in the Australian medical literature. Comparison of medical abortion with surgical vacuum Australian and New Zealand Journal of Medicine aspiration: women’s preferences and the acceptability of 1982;12:7–9. treatment. British Medical Journal 1993;307:714–7. Hamilton M. On informed consent. British Journal of Henzlova MJ, Blackburn GH, Bradley EJ, Rogers WJ. Psychiatry 1983;143:416–8. Patient perception of a long-term clinical trial: Hammett TM, Dubler NN. Clinical and epidemiological experience using a close-out questionnaire in the studies research on HIV infection and AIDS among corrective of Left Ventricular Dysfunction (SOLVD) Trial. Controlled inmates: regulations, ethics, and procedures. Evaluation Clinical Trials 1994;15:281–93. Review 1990;14:482–501. Hilden J, Block JE, Andresson B, Visfeldt J. Ethics and Harrington D, Crowley J, George SL, Pajak T, Redmond, decision theory in a clinical trial involving severe Wieand S. The case against independent monitoring disfigurement. Theoretical Surgery 1987;1:183–9. committees. Statistics in Medicine 1994;13:1411–14. Hilden J, Habbema JDF. The marriage of clinical trials Harrison H. Neonatal intensive care: parents’ role in and clinical decision science. Statistics in Medicine ethical decision making. Birth 1986;13:165–75. 1990;9:1243–57. Harrison K, Vlahov D, Jones K, Charron K, Clements ML. Hill AB. The experimental approach in preventative Medical eligibility, comprehension of the consent medicine. Journal of the Royal Institute of Public Health process, and retention of injection drug users recruited 1958;21:177–218. for an HIV vaccine trial. Journal of Acquired Immune Hill AB. Statistical methods in clinical and preventive Deficiency Syndromes and Human Retrovirology medicine. London: Oxford University Press, 1962. 1995;10:386–90. Hill AB. Medical ethics and controlled trials. British Harth SC, Thong YH. Socio-demographic and motivational characteristics of parents who volunteer Medical Journal 1963;1:1043–9. their children for clinical research: a controlled study. Hill C, Sancho-Garnier H. Interim analysis and early British Medical Journal 1990;300:1372. results in clinical trials. In: Tagnon HJ, Staquet MJ, Harth SC, Thong YH. Parental perceptions and attitudes editors. Controversies in cancer: design of trials and about informed consent in clinical research involving treatment. New York, 1979:51–3. children. Social Science and Medicine 1995;40:1573–7. Hobbes T. Leviathan: or the matter, form, and power of a Hassar M, Weinstein M. “Uninformed” consent and the commonwealth ecclesiastical. Oxford: Blackwell, 1946. wealthy volunteer: an analysis of patient volunteers in a Holden G, Rosenburg G, Barker K, Tuhrim S, Brenner B. clinical trial of a new anti-flammatory drug. Clinical The recruitment of research participants: a review. Social Pharmacology and Therapeutics 1976;20(4):379–86. Work and Health Care 1993;19:1–44. Hasselbald V. Mosteller F, Littenberg B, Chalmers TC, Hollenberg NK, Dzau VJ, Williams GH. Letter to the Hunink MG, Turner JA, et al. A survey of current editor: are uncontrolled clinical studies ever justified? problems in meta-analysis. Discussion from the Agency New England Journal of Medicine 1980;303:1067. for Health Care Policy and Research inter-PORT Work Group on Literature Review/Meta-Analysis. Medical Care Horwitz RI. Large-scale randomised evidence: large, 1995;33:202–20. sample trials and overviews of trials: discussion. A clinician’s perspective on meta-analyses. Journal of Hegel GWF. In: Wo AW, editor. Elements of the Clinical Epidemiology 1995;48(1):41–4. philosophy of right. Cambridge: Cambridge University Press, 1991. Howard JM, Demets D, BHAT Research Group. How Heitjan DF, Houts PS, Harvey HA. A decision-theoretic informed is informed consent? The BHAT experience. evaluation of early stopping rules. Statistics in Medicine Controlled Clinical Trials 1981;2:287–304. 1992;11(5):673–83. Hughes MD. Reporting Bayesian analysis of clinical Hellman S. Editorial: randomised clinical trials and the trials. Statistics in Medicine 1993a;12:1651–65. doctor–patient relationship. An ethical dilemma. Cancer Hughes MD. Stopping guidelines for clinical trials Clinical Trials 1979;2:189–93. with multiple treatments. Statistics in Medicine Hellman S, Hellman DS. Of mice but not men. New 1993b;12:901–15. England Journal of Medicine 1991;324:1585–9. Hughes MD, Freedman LS, Pocock SJ. The impact of Hennekens CH, Buring JE. Epidemiology in medicine. stopping rules on heterogeneity of results in overviews of Boston: Little, Brown, 1987. clinical trials. Biometrics 1992;48:41–53. Henry D, Hill S. Comparison should be against active Hume D. Enquiries concerning human understanding treatments rather than placebos. British Medical Journal and concerning the principles of morals, 3rd edn. 66 1995;310:1279. Oxford: Clarendon Press, 1975. Health Technology Assessment 1998; Vol. 2: No. 15

Hutchins SA, Eckes R. HIV infection. Clinical research. Kadane JB. Bayesian methods and ethics in a clinical trial Considerations for prospective participants. Nursing design. Chichester: Wiley, 1996:3–19. Clinics of North America 1996;31:125–35. Kadane JB, Seidenfeld T. Statistical issues in the analysis Hutton JL. Statistics is essential for professional ethics. of data gathered in the new designs. In: Kadane JB, Journal of Applied Philosophy 1995;12:253–61. editor. Bayesian statistics and ethics in a clinical trial Hutton JL. The ethics of randomized controlled trials: design. Chichester: Wiley, 1996:115–29. a matter of statistical belief? Health Care Analysis Kairys D. The law of clinical testing with human subjects. 1996;4:146, 1–146, 8. In: Kadane JB, editor. Bayesian statistics and ethics in a Illich I. Medical nemesis. New York: Pantheon Press, clinical trial design. Chichester: Wiley, 1996:223–51. 1976. Kant I. Groundwork of the metaphysic of morals. Ingelfinger FJ. The randomized clinical trial. New Translated and analysed by Paton HJ. London: England Journal of Medicine 1972;287:100–1. Routledge, 1991. Institute of Medical Ethics Working Party. AIDS, Kapp MB. Proxy decision making in Alzheimer disease ethics, and clinical trials. British Medical Journal research. Alzheimer Disease Association 1994;8(4):28–37. 1992;305:699–701. Kardinal CG. Ethical issues in cancer clinical trials. ISIS-2 Collaborative Group. Randomized trial of Journal of the Louisiana State Medical Society intravenous streptokinase, oral aspirin, both or neither 1994;146(8):359–61. among 17187 cases of acute myocardial infarction. Karjalainen S, Palva I. Do treatment protocols improve Lancet 1988;349–53. end results? A study of survival of patients with multiple Jaeschke R, Cook D, Sackett DL. The potential role of myeloma in Finland. British Medical Journal single-patient randomized controlled trials (N-of-1 RCTs) 1989;299:1069–72. in clinical practice. Journal of the American Board of Family Practice 1990;5:227–9. Kassirer JP. Adding insult to injury: usurping patients’ prerogatives. New England Journal of Medicine Jaeschke R, Adachi JD, Guyatt GH, Keller JL, Wong B. 1983;308:898–901. Clinical usefulness of amitriptyline in fibromyalgia: the results of 23 N of 1 randomized controlled trials. Journal Katz J. Randomized clinical trials: am I my brother’s of Rheumatology 1991;18:447–51. keeper? Journal of Clinical Gastroenterology 1981;3:141–6. Jensen AB, Madsen B, Anderson P, Rose C. Information Katz J. Ethics and clinical research revisited: a tribute to for cancer patients entering a clinical trial: evaluation of Henry K. Beecher. Hastings Center Report 1993;23:31–9. an information strategy. European Journal of Cancer Katz KD. Commentary on “the law of clinical testing with 1993;29:2235–8. human subjects”. In: Kadane JB, editor. Bayesian statistics Jha P, Deboer D, Sykora K, Naylor CD. Characteristics and ethics in a clinical trial design. Chichester: Wiley, and mortality outcomes of thrombolysis trial participants 1996:257–63. and non-participants: a population-based comparison. Kauffman RE. Drug trials in children: ethical, legal, and Journal of the American College of Cardiology practical issues. Journal of Clinical Pharmacology 1996;27:1335–42. 1994;34:296–9. Johannessen T. Controlled trials in single subjects. 1. Kaufman D. Cancer therapy and the randomized clinical Value in clinical medicine. British Medical Journal trial: good medicine? A Cancer Journal for Clinicians 1991;303:173–4. 1993;44:109–14. Johannessen T, Fosstevedt D, Petersen H. Combined single subject trials. Scandinavian Journal of Primary Health Kemp N, Skinner E, Toms J. Randomized clinical trials Care 1991;9:23–7. of cancer treatment – a public opinion survey. Clinical Oncology 1984;10:155–61. Johnson N, Lilford RJ, Brazier W. At what level of collective equipoise does a clinical trial become ethical. Kennedy I. The law and ethics of informed consent and Journal Medical Ethics 1991;17:30–4. randomized controlled trials. In: Treat me right: essays in medical law and ethics. Oxford: Clarendon Press, Jonas H. Philosophical reflections on experimenting 1988:213–24. with human subjects. Daedulus 1969;98:219–47. King J. Informed consent: a review of the empirical Jones B, Jarvis P, Lewis JA, Ebbutt AF. Trials to assess evidence. Institute of Medical Ethics Bulletin Dec. 1986. equivalence: the importance of rigorous methods. British Medical Journal 1996;313:36–9. Kleijnen J, Gotzsche P, Chalmers I. So what’s so special about randomization? In press. Kadane JB. Progress toward a more ethical method for clinical trials. Journal of Medicine and Philosophy Klimt CR. The conduct and principles of randomised 1986;2:385–404. clinical trials. Controlled Clinical Trials 1981a;1:283–93. 67 References

Klimt CR. Terminating a long-term clinical trial. Lau J, Schmid CH, Chalmers TC. Cumulative meta- Controlled Clinical Trials 1981b;1:319–25. analysis of clinical trials builds evidence for exemplary medical care. Journal of Clinical Epidemiology Klimt CR, Canner PL. Terminating a long-term clinical 1995;48:45–57, 59–60. trial. Clinical Pharmacology and Therapeutics 1979;25:641–6. Laupacis A. Research by collaboration. Kodish E, Stocking C, Ratain MJ, Kohrman A, Siegler Lancet 1995;345:938. M. Ethical issues in phase in oncology research: a comparison of investigators and institutional review Laupacis A, Connolly SJ, Gent M, Roberts RS, Cairns J, board chairpersons. Journal of Clinical Oncology Joyner C. How should results from completed studies 1992;10(11):1810–6. influence ongoing clinical trials? The CAFA study experience. Annals of Internal Medicine 1991;115:818–22. Kopelman L. Estimating risk in human research. Clinical Research 1981;29:1–8. Leading article. Qualms about innovative surgery. Lancet 1985;149:149. Kopelman L. When is the risk minimal enough for children to be research subjects? In: Kopelman L, Lebacqz K. The National Commission and research Moskop J, editors. Children and health care: moral and in pharmacology: an overview. Federation Proceedings social issues. Dordrecht: Kluwer, 1989:89–102. 1977;36(10):2344–8. Korn EL, Baumind S. Randomized clinical trials with Lebacqz K. Ethical aspects of clinical trials. In: Shapiro clinician preferred treatment. Lancet 1991;337:149–52. SH, Louis TA, editors. Clinical trials: issues and approaches. New York: Marcel Dekker, 1983:81–98. Korsgaard CM. Creating the kingdom of ends. Cambridge: Cambridge University Press, 1991. Lebacqz K, Levine RJ. Respect for persons and informed consent to participate in research. Clinical Research Kramer M, Shapiro S. Scientific challenges in the 1977;25(3):101–7. application of randomised trials. Journal of the American Medical Association 1984;252:2739–45. Lennox EL, Stiller CA, Morris Jones PH, Kinnier Wilson LM. Nephroblastoma: treatment during 1970–3 and the L’abbe KA, Detsky AS, O’Rourke K. Meta-analysis in effect on survival of inclusion in the first MRC trial. clinical research. Annals of Internal Medicine British Medical Journal 1979;567–9. 1987;107:224–33. Lesser H. Obligation and consent. Journal of Medical Ethics Lacher MJ. Patients and physicians as obstacles to a 1989;15:195–6. randomized trial. Seminars in Oncology 1981;8:424–9. Levene M, Wright I, Griffiths G. Is informed consent Lamas GA, Pfeffer MA, Hamm P, Wertheimer J, Rouleau in neonatal randomized controlled trial ritual? Lancet JL, Braunwald E. Do the results of randomized 1996;347:475. controlled trials of cardiovascular drugs influence medical practice? New England Journal of Medicine Levine C. Ethics, epidemiology and women’s health. 1992;327:241–7. Annals of Epidemiology 1994;4(2):159–65. Langley GR, Sutherland HJ, Wong S, Minkin S, Llewellyn- Levine C, et al. Building a new consensus: principles and Thomas HA, Till JE. Why are (or are not) patients given policies for clinical research on HIV/AIDS. IRB: A Review the option to enter clinical trials? Controlled Clinical Trials of Human Subjects Research 1991;13:1–17. 1987;8:49–59. Levine R. Ethics and regulation of clinical research. Lantos J. Ethical issues. How can we distinguish between Baltimore: Urban and Schwarzenberg, 1981. clinical research and innovative therapy? American Journal Levine RJ. The use of placebos in randomized clinical of Pediatric Hematology-Oncology 1994a;16(1):72–5. trials. IRB: A Review of Human Subjects Research 1985;7:1–4. Lantos J. Ethics, randomization and technology Levine RJ. Ethics and regulation of clinical research. assessment. Cancer 1994b;74(9):2653–6. Baltimore: Urban and Schwarzenberg, 1988. Larson EB, Ellsworth AJ, Oas J. Randomized clinical Levine RJ. Building a new consensus: principles and trials in single patients during a two year period. JAMA policies for clinical research. IRB: A Review of Human 1993;270:2708–12. Subjects Research 1991;13:1–17. Lasagna L. Special subjects in human experimentation. Levine RJ. Clinical trials and physicians as double agents. In: Experimentation with human subjects. Yale Journal of Biology and Medicine 1992;65(2):65–74. 1970:262–75. Levine RJ. Ethics of clinical trials. Do they help the Lau J, Chalmers TC. The rational use of therapeutic patient? Cancer 1993;72(1):2805–10. drugs in the 21st century. Important lessons from cumulative meta-analyses of randomized control trials. Levine RJ, Lebacqz K. Some ethical considerations in International Journal of Technology Assessment in Health Care clinical trials. Clinical Pharmacology and Therapeutics 68 1995;11(3):509–22. 1979;25:728–41. Health Technology Assessment 1998; Vol. 2: No. 15

Lewis RJ. An introduction to the use of interim analysis Lockwood M, Anscombe GEM. Sins of omission? The in clinical trials. Annals of Emergency Medicine non-treatment of controls in clinical trials. Aristotelian 1993;22(9):1463–9. Society 1983;57:207–22. Liberati A. The relationship between clinical trials Logue G, Wear S. A desperate solution: individual and clinical practice: the risks of underestimating its autonomy and the double-blind controlled experiment. complexity. Statistics in Medicine 1994;13:1485–91. Journal of Medicine and Philosophy 1995;20(1):57–64. Liberati A. Meta-analysis: statistical alchemy for the 21st Love JW, Phil D. Drugs and operations: some important century: discussion. Journal of Clinical Epidemiology differences. JAMA 1975;232:37–8. 1995;48(1):81–6. Lurie P, Bishaw M, Chesney MA, Cooke M, Fernandes Light RJ. Accumulating evidence from independent MEL, Hearst N, et al. Ethical, behavioural, and social studies: what we can win and what we can lose? Statistics aspects of HIV vaccine trials in developing countries. in Medicine 1987;6:221–8. JAMA 1994;271(4):295–301. Lilford RJ. Clinical experimentation in obstetrics. British Lyons D. Forms and limits of utilitarianism. Oxford: Medical Journal 1987;295:1298–300. Clarendon Press, 1965:119–60. Lilford RJ. Clinical experimentation in obstetrics. British Lynoe N, et al. Informed consent: study of the quality of Medical Journal 1988;296:571. information given to paticipants in a clinical trial. British Medical Journal 1991;303:610. Lilford RJ. Evaluating new treatments and technologies in perinatal medicine: practical problems, ethics and Machin D. Interim analysis and ethical issues in the solutions. International Journal of Technology Assessment in conduct of trials. In: Williams CJ, editor. Introducing new Health Care 1989;5:459–72. treatments for cancer. New York: Wiley, 1992. Lilford RJ. Clinical trial numbers. Lancet 1990;335:483–4. Machin D. Discussion of the “What, why and how of Bayesian clinical trial monitoring”. Statistics in Medicine Lilford RJ. The rise and fall of chorionic villus sampling. 1994;13:1385–9. British Medical Journal 1991;303(6808):936–7. Mackillop WJ, Johnston PA. Ethical problems in clinical Lilford RJ. for the fetal compromise group. Formal research: the need for empirical studies of the clinical measurement of clinical uncertainty: prelude to a trial in trials process. Journal of Chronic Diseases perinatal medicine. British Medical Journal 1986;39(3):177–88. 1994;308:111–12. McLean PD. The effect of informed consent on the Lilford RJ, Braunholtz D. The statistical basis of public acceptance of random treatment assignment in a clinical policy: a paradigm shift is overdue. British Medical Journal population. Behavior Therapy 1980;11:129–32. 1996;313:603–7. McPherson K. On choosing the number of interim Lilford RJ, Jackson JC. Equipoise and the ethics of analyses in clinical trials. Statistics in Medicine randomisation. Journal of the Royal Society of Medicine 1982;12:25–36. 1995;88:552–9. McPherson K. The Cochrane Lecture. The best and Lilford RJ, Johnson N. The alpha and beta errors in the enemy of the good: randomized controlled trials, randomised controlled trials. New England Journal of uncertainty, and assessing the role of patient choice in Medicine 1990;322:780–1. medical decision-making. Journal of Epidemiology and Lilford RJ, Thornton JG. Decision logic in medical Community Health 1994;48:6–15, 423–4. practice. Journal of the Royal College of Physicians Madden M. In: Alderson P, editor. Breast cancer, 1992;26:400–12. randomised controlled trials and consent. London: Lilford RJ, Thornton JG, Braunholtz D. Clinical trials and SSRY, 1994:62–76. rare diseases: a way out of the conundrum. British Medical Mahon J, Laupacis A, Donner A, Wood T. Randomised Journal 1995;311:1621–5. study of n of 1 trials vs standard practice. British Medical Llewellyn-Thomas HA, McGreal MJ, Theil EC, Fines, Journal 1996;312:1069–74. Erlichman C. Patients’ willingness to enter clinical trials: Mahon JL, Feagan BG, Laupacis A. Ethics of N-of-1 trials. measuring the association with perceived benefit and Lancet 1995;345:989. preference for decision participation. Social Science and Medicine 1991;32:35–42. Mammel KA and Kaplan DW. Research consent by adolescent minors and institutional review boards. Llewellyn-Thomas HA, McGreal MJ, Theil EC. Cancer- Journal of Adolescent Health 1995;17:323–30. patients decision-making and trial entry preferences – the effects of framing information about short-term Mann AH. The psychological effect of a screening toxicity and long-term survival. Medical Decision Making programme and clinical trial for hypertension on the 1995;15(1):4–12. participants. Psychological Medicine 1977;7:431–8. 69 References

Mariner. The ethical conduct of clinical trials of HIV Miller C, Searight HR, Grable D, Schwartz R, Sowell C, vaccines. Evaluation Review 1990;14:538–64. Barbarash RA. Comprehension and recall of the informational content of the informed consent document: an Marini JL, Sheard MH, Bridges CI. An evaluation of evaluation of 168 patients in a controlled clinical trial. informed consent with volunteer prisoner subjects. Yale Journal of Clinical Research and Drug Development Journal of Biology and Medicine 1976;427:434. 1994;8:237–48. Markman M. Ethical difficulties with randomized clinical Miller ST, Appelgate WB. Clinical trials in elderly trials involving cancer patients. Journal of Clinical Ethics persons. Journal of the American Geriatrics Society 1992;3:193–5. 1985;33:91–2. Marquis D. Leaving therapy to chance. Hastings Center Millon-Underwood. Determinants of participation in Report 1983;13(4):40–7. state-of-the-art cancer prevention, early detection? Marsden J, Sacks NPD, Baum M, Whitehead MI, screening, and treatment trials. Cancer Nursing Crook D. A pilot study investigating the feasibility of 1993;16(1):25–33. conducting a randomised trial of HRT use in women with a history of breast cancer – preliminary report of Minogue BP, Palmer-Ferandez G, Udell L, Waller BN. findings. Unpublished. Individual autonomy and the double-blind controlled experiment: the case of the desperate volunteer. Journal Maslin A. A survey of the opinions on “informed of Medicine and Philosophy 1995;20(1):43–55. consent” of women currently involved in clinical trials within a breast unit. European Journal of Cancer Care Moore D, Papp AJ. Commentary on “the law of clinical 1994;3(4):153–62. testing with human subjects”. In: Kadane JB, editor. Bayesian statistics and ethics in a clinical trial design. Mathews D. Comment on ‘clinical experimentation in Chichester: Wiley, 1996:251–7. obstetrics’. British Medical Journal 1988;296:66. Morris RE, Brown B. Commentary: Tacrolimus for Mattson ME, et al. Participation in a clinical trial: the prevention of liver allograft rejection: clinical trials and patients’ point of view. Controlled Clinical Trials tribulations. Lancet 1995;346:1310. 1985;6:156–67. MRC. Medical Research Council series. The ethical Mautner T. The Penguin dictionary of philosophy. conduct of research in children. London: MRC, 1991. London: Penguin Books, 1997. MRC Working Group on Leukaemia in Childhood. Meier P. Terminating a trial – the ethical problem. Duration of survival of children with acute leukemia. Clinical Pharmacology and Therapeutics 1979;25:633–40. British Medical Journal 1971;4:7–9. Meinert CL. When and how to stop a clinical trial. Munir K. Ethical principles governing research in child Clinical Pharmacology and Therapeutics 1979;25(5):632. and adolescent psychiatry. Journal of the American Academy Meisel A, Roth L. Toward a informed discussion of of Child and Adolescent Psychiatry 1992;31(3):408–14. informed consent: a review and critique of the empirical Myers MG, Cairns JA, Singer J. The consent form as a studies. Arizona Law Review 1983;25:265–46. possible cause of side-effects. Clinical Pharmacology and Mettlin C, Cummings KM, Walsh D. Risk factor and Therapeutics 1987;42(3):250–3. behavioural correlates of willingness to participate in National Commission for the Protection of Human cancer prevention trials. Nutrition and Cancer Subjects of Biomedical and Behavioral Research. 1985;7:189–98. The Belmont Report: ethical principles and guidelines Mike V. Modern medicine: perspectives on harnessing its for the protection of human subjects of research. vast emerging power. Technology in Society 1988;10:327–38. DHEW Publication No. (OS) 78-0012. Washington, DC: DHEW, 1978. Mike V. Towards an ethics of evidence – and beyond: observations on technology and illness. Research in Neuberger J. The patient’s view of the patient-health Philosophy and Technology 1989a;9:101–13. care worker relationship. In: Gillon R, editor. Principles of health care ethics. Chichester: Wiley, 1994:377–86. Mike V. Philosophers assess randomized clinical trials: the need for dialogue. Controlled Clinical Trials Neugebauer E. Conventional vs laparoscopic 1989b;10(3):244–53. cholecystectomy and the RCT. British Journal of Surgery 1991;78(2):150–4. Mike V, Krauss AN, Ross GS. Neonatal extracorporeal membrane oxygenation: clinical trials and the ethics of Newell DJ. Type 2 errors and ethics. British Medical evidence. Journal of Medical Ethics 1993;19(4):212–8. Journal 1789, 1978. Mill JS. Utilitarianism, 5th edn. London: Longmans, Nicholson RH. Replication of clinical trials. Lancet Green, Reader and Dyer, 1874. 1986;ii:518. Miller BL. The ethics of cardiac arrest research. Annals of Nowak R. Problems in clinical trials go far beyond 70 Emergency Medicine 1993;22(1):118–24. misconduct. Science 1994;264:1538–41. Health Technology Assessment 1998; Vol. 2: No. 15

Nyapadi TJ. Legal aspects of medical research and the Piaget J. The language and thought of the child. London: way forward. The Central African Journal of Medicine Harcourt, Brace, 1926. 1995;41(3):73–6. Pignon JP, Arrigada R. Early stopping rules and long- Oakley A. Who’s afraid of the randomised controlled term follow-up in phase iii trials. Lung Cancer trial? Some dilemmas of the scientific method and 1994;10(1):S151–9. “good” research practice. Women and Health 1990;15:25–59. Pocock SJ. The combination of randomized and historical controls in clinical trials. Journal of Chronic Oakley A. Women’s knowledge and experience of breast Diseases 1975;29:175–88. cancer treatment and research. London: Social Science Research Unit, 1992. Pocock SJ. Group Sequential methods in the design and analysis of clinical trials. Biometrika 1977;64:191–9. Oddens BJ, Algra A, van Gijn J. How much information is retained by participants in clinical trials? Nederlands Pocock SJ. Can sequential methods be used for analysis Tijdschrift voor Geneeskunde 1992;136:2272–6. of cancer clinical trials. In: Controversies in cancer. Oettinger W, Berger HG. Commentary on: the rise 1979:63–74. and fall of the randomized controlled trial in surgery. Pocock SJ. Current issues in the design and Theoretical Surgery 1989;4:170. interpretation of clinical trials. British Medical Journal Olschewski M, Schumacher M, Davis KB. Analysis of 1985;290:39–42. randomized and non-randomized patients in clinical Pocock SJ. Statistical and ethical issues in monitoring trials using the comprehensive cohort follow-up study trials. Statistics in Medicine 1993;12:1459–75. design. Controlled Clinical Trials 1992;13:226–39. Pocock SJ, Hughes MD. Practical problems in interim Olver IN, Buchanan L, Laidlaw C, Poulton G. The analysis, with particular regard to estimation. Controlled adequacy of consent forms for informing patients Clinical Trials 1989;10:209S–21S. entering oncological clinical trials. Annals of Oncology 1995;6:867–70. Pocock SJ, Hughes MD. Estimation issues in clinical trials Palmer CR. Ethics and statistical methodology in clinical and overviews. Statistics in Medicine 1990;9:657–6. trials. Journal of Medical Ethics 1993;19(4):219–22. Pollock AV. The rise and fall of the random controlled Palmer CR. Recent advances in adaptive clinical trials. surgery. Theoretical Surgery 1989;4:163–70. Paper presented at annual meeting of the American Pollock AV. Surgical evaluation at the crossroads. Statistical Association, 1995. British Journal of Surgery 1993;80(8):964–6. Palmer CR, Rosenberger WF. Ethics and practice: Postlethwaite RJ, Reynolds JM, Wood AJ, Evans JH, Lewis alternative designs for phase 3 randomised controlled MA. Recruiting patients to clinical trials: lessons from trials. In press. studies of growth hormone treatment in renal failure. Parmar MKB, Machin D. Monitoring clinical trials: Archives of Disease in Childhood 1995;73(1):30–5. experience of, and proposals under consideration by, Popper KR. Conjectures and Refutations: the growth of the cancer therapy committee of the British Medical Research Council. Statistics in Medicine 1993;12:497–504. scientific knowledge. London: Routledge and Kegan Paul, 1963. Passamani E. Clinical trials – are they ethical? New England Journal of Medicine 1991;324:1589–92. Powell-Tuck J, Macrae KD, Healy MJR, Lennard-Jones JE, Parkins RA. A defence of the small clinical trial: Pellegrino ED, Thomasma DC. A philosophical basis evaluation of three gastroenterological studies. of medical practice. New York: Oxford University British Medical Journal 1986;292:599–602. Press, 1981. Prentice ED, et al. IRB review of a phase II randomized Penman DT, Holland JC, Bahna GF, et al. Informed clinical trial involving incompetent patients suffering consent for investigational chemotherapy: patients’ and from severe closed head injury. IRB: A Review of Human physicians’ perceptions. Journal of Clinical Oncology Subjects Research 1993;15:1–7. 1984;2:849–55. Peto R. Why do we need systematic overviews of Prescott RJ. Feedback of data to participants during randomized clinical trials? Statistics in Medicine clinical trials. In: Tagnon HJ, Staquet MJ, editors. 1987;6:233–40. Controversies in cancer: design of trials and treatment. New York, 1979:55–61. Peto R, Pike MC, Armitage P, Breslow NE, Cox DR, Howard SV, et al. Design and analysis of randomized Prestifilippo J. The ethical treatment of cancer: what is clinical trials requiring prolonged observation of each right for the patient? Cancer 1993;72:2817–9. patient. British Journal of Cancer 1976;34:585–612. Pringle M, Churchill R. Randomized controlled trials in Pfeffer MR. Ethics and the oncologist. Medicine and Law general practice: gold standard or fool’s gold? British 1993;12(3–5):235–40. Medical Journal 1995;311:1382–3. 71 References

Propert KJ, Dillman RO, Seagren SL, Green MR. Ross MW, Jeffords K, Gold J. Reasons for entry into and Chemotherapy and radiation therapy as compared with understanding of HIV/AIDS clinical trials: a preliminary radiation therapy in stage 3 non-small cell cancer. New study. AIDS Care 1993;6(1):77–82. England Journal of Medicine 1991;323:1136–7. Rothman DJ. Ethics and human experimentation. Rahimtoola SH. Some unexpected lessons from large New England Journal of Medicine 1987;317(19):1195–9. multicenter randomized clinical trials. Circulation Rothman KJ. Placebo mania. British Medical Journal 1985;72:449–55. 1996;313:3–4. Raju TN, Langenberg P, Sen A, Aldana O. How much Rothman KJ, Michels KB. The continuing unethical use “better” is good enough? The magnitude of treatment of placebo controls. New England Journal of Medicine effect in clinical trials. American Journal of Diseases of 1994;331:394–8. Children 1992;146(4):407–11. Roy DJ. Controlled clinical trials: an ethical imperative. Reber AS. The Penguin dictionary of psychology. Journal of Chronic Diseases 1986a;39(3):159–62. London: Penguin Books, 1985. Roy DJ. Ethics in clinical research and clinical practice. Reemtsma K. Experimental surgery: what is it? In: Herter Clinical and Investigative Medicine 1986b;9:283–9. FP, editor. Human and ethical issues in the surgical care Royall RM. Ethics and statistics in randomized clinical of patients with life-threatening disease. Springfield: trials, with discussion. Statistical Science 1991;6:52–88. Charles C Thomas, 1986:183–6. Rubinstein LV, Gail MH. Monitoring rules for stopping Reidenberg M. Should unevaluated therapies be accrual to comparative survival studies. Controlled Clinical available for sale? Clinical Pharmacology and Therapeutics Trials 1982;3:325–43. 1987;42:6. Rudicel S, Esdaile J. The randomized clinical trial in Reiser J, Warner JO. The value of participating in an orthopaedics: obligation or option. Journal of Bone and asthma trial. Lancet 1985;i:206–7. Joint Surgery 1985;67:1284–93. Reiser SJ. Human experimentation and the convergence Ruse M. At what level of statistical certainty ought a of medical research and patient care. Annals of the random clinical trial be interrupted? In: The use of American Academy of Politics and Social Science human beings in research. Boston: Kluwer, 1978;437:8–18. 1988:189–221. Report of an Institute of Medical Ethics Working Party. Russell PS. Commentary on: The rise and fall of the In: Nicholson R, editor. Medical research with children: randomized controlled trial in surgery. Theoretical Surgery ethics, law and practice. Oxford: Oxford University 1989;4:169–70. Press, 1986. Rutstein DD. The ethical design of human experiments. Report to the Health Services and Public Health Review In: Freund PA, editor. Experimentation with human Board. Health technology assessment in surgery: the role subjects. New York: George Braziller, 1970:383–401. of the randomized clinical trial. 1993. Sacks H, Chalmers T, Smith H. Randomised versus Richardson JL, Myrtle R, Solis JM, Hiisserich JW, Oliver historical controls for clinical trials. American Journal D. Participation of community medical oncologists in of Medicine 1982;72:233–40. clinical research trials. Progress in Clinical Biology Research Savulescu J, Momeyer RW. Should informed consent 1986;216:269–80. be based on rational beliefs? Journal of Medical Ethics 1997;23(5):282–7. Roberson NL. Clinical trial participation. Viewpoints from racial/ethnic groups. Cancer 1994;74:2687–91. Saurbrey N, Jensen J, Rasmussen PE, Gjorup T, Guldager H, Riis P. Danish patients’ attitudes to scientific-ethical Robiner WN, Melroe H, Campbell S, Rhame FS, Colon E, questions: an interview study focusing on therapeutic Chung J, et al. Psychological effects of participation and trials. Acta Medica Scandinavia 1984;215:99–104. non-participation in a placebo-controlled zidovudine clinical trial with asymptomatic human immuno- Schafer A. The ethics of randomized clinical trials. deficiency virus-infected individuals. Journal of Acquired New England Journal of Medicine 1982;307:719–24. Immune Deficiency Syndromes 1993;6(7):765–808. Schafer A. The randomized clinical trial: for whose Rockhold FW, Enas GG. Data monitoring committees benefit? IRB: A Review of Human Subjects Research and the pharmaceutical industry: ethical and logistic 1985;7(2):4–6. considerations. Statistics in Medicine 1993;12:471–9. Schaffner KF. Ethically optimizing clinical trials. In: Rodebhuis S, Van den Heuvel WJA, Amyas AA, Koops Kadane JB, editor. Bayesian methods and ethics in a HS, Sleijfer DTH, Mulder NH. Patient motivation and clinical trial design. New York: Wiley, 1996:19–63. informed consent in a phase i study of anticancer agent. Schain WS. Barriers to clinical trials part II: knowledge European Journal of Cancer Clinical Oncology and attitudes of potential participants. Cancer Supplement 72 1984;20:457–62. 1994;74(9):2666–76. Health Technology Assessment 1998; Vol. 2: No. 15

Schmoor C, Olschewski M, Schumacher M. Randomized Silva MD, Sorrell JM. Enhancing comprehension of and non-randomized patients in clinical trials: information for informed consent: a review of empirical experiences with comprehensive cohort studies. research. IRB: Human Subjects Research 1988;10:1–6. Statistics in Medicine 1996;15:263–71. Silverman WA. Human experimentation: a guided Schulz KF, Chalmers I, Hayes RJ, Altman DG. Empirical step into the unknown. Oxford: Oxford University evidence of bias: dimensions of methodological quality Press, 1985. associated with estimates of treatment effects in controlled trials. Journal of the American Medical Silverman WA, Altman DG. Patients’ preferences and Association 1995;273:408–12. randomised trials. Lancet 1996;347:171–4. Schwartz D, Lellouch J. Explanatory and pragmatic Sim I, Hlatky MH. Growing pains of meta-analysis. British attitudes in therapeutic trials. Journal of Chronic Diseases Medical Journal 1996;313:702–3. 1967;20:637–48. Simberkoff MS, Hartigan PM, Hamilton JD, Deykin D, Schwartz D, Flamant R, Lellouch J. Clinical trials. Gail M, Bartlett JG, et al. Ethical dilemmas in continuing Academic Press, London. 1980. a zidovudine trial after early termination of similar trials. Controlled Clinical Trials 1993;14(1):6–18. Scoles L, Silagy C. Are clinical trials in general practice ethical? Australian Family Physician 1993;22(2):182–3. Simel DL, Feussner JR. A randomized controlled trial comparing quantitative informed consent formats. Sedrank N. Issues of statistical design. In: Kadane JB, Journal of Clinical Epidemiology 1991;44(8):771–7. editor. Bayesian statistics and ethics in a clinical trial design. New York: Wiley, 1996:163–71. Simes RJ, Tattersall MHN, Coates AS, Raghaven D, Solomon HJ, Smartt H. Randomised comparison of Segelov E, Tattershall MM, Coates AS. Redressing the procedures for obtaining informed consent in clinical balance – the ethics of not entering an eligible patient trials of treatment of cancer. British Medical Journal on a randomized clinical trial. Annals of Oncology 1992;3:103–5. 1986;293:1065–8. Seibert C, Clark CM. Operational and policy Simes SJ. Controlled trials in single subjects. British considerations of data monitoring in clinical trials: the Medical Journal 1991;303:716–17. Diabetes Control and Complications Trial experience. Siminoff LA, Fetting JH. Factors affecting treatment Controlled Clinical Trials 1993;14:30–44. decisions for a life threatening illness: the case of medical Senn S. Inherent difficulties with active control equi- treatment for breast cancer. Social Science and Medicine valence studies. Statistics in Medicine 1993a;12(24):236–75. 1991;32(7):813–8. Senn S. Suspended judgment N of 1 trials. Controlled Simon R. Adaptive treatment assignment methods and Clinical Trials 1993b;14:1–5. clinical trials. Biometrics 1977;33:743–9. Senn SJ. Falsificationism and clinical trials. Statistics in Simon R. Randomized clinical trials and research Medicine 1991a;10:1679–92. strategy. Cancer Treatment Reports 1982;66:1083–7. Senn SJ. Controlled clinical trials in single subjects. Simon R. Some practical aspects of the interim British Medical Journal 1991b;303:716–17. monitoring of clinical trials. Statistics in Medicine 1994;13:1401–9. Sharp SJ, Thompson SG, Altman DG. The relation between treatment benefit and underlying risk in meta- Slevin M, Mossman J, Bowling A, et al. Volunteers or analysis. British Medical Journal 1996;313:735–9. victims: patients’ views of randomised clinical trials. British Journal of Cancer 1995;71:1270–4. Shaw LW, Chalmers TC. Ethics in co-operative clinical trials. Annals of the New York Academy of Sciences Smith JS. Suspended judgment. Remembering the role 1970;169:487–95. of Thomas Francis, Jr. in the design of the 1954 Salk Shimm DS, Spece RG. Ethical issues and clinical trials. Vaccine Trial. Controlled Clinical Trials 1992;13(3):181–4. Drugs 1983;46(4):579–84. Smith M, Simon R, Cain D, Ungerleider RS. Children Shinebourne EA. Ethics of innovative cardiac surgery. and cancer. A perspective from the cancer therapy British Heart Journal 1984;52:597–601. evaluation programme, National Cancer Institute. Cancer 1993;71:3422–8. Shuster JJ, Krischer JP, Boyett JM. Ethical issues in co-operative cancer trials from a statistical viewpoint. Smithells RW. Iatrogenic hazards and their effects. 1. A general overview. American Journal of Pediatric Postgraduate Medical Journal 1975;15:39–52. Hematology/Oncology 1985;7:57–63. Snowdon C, Garcia J, Elbourne D. Understanding Siebert C, Clark CM. Operational and policy randomisation: parental responses to the allocation of considerations of data monitoring in clinical trials: the alternative treatments in a clinical trial involving their Diabetes Control and Complications trial. Controlled critically ill newborn babies. Social Science and Medicine Clinical Trials 1993;14(1):30–44. 1997;45(9):1337–55. 73 References

Spaight SJ, Nash S, Finison LJ, et al. Medical oncologists’ Stiller CA. Centralized treatment, entry to trials and participation in cancer clinical trials. In: Advances in survival. British Journal of Cancer 1994;70:352–62. cancer control: epidemiology and research. New York: Stiller CA, Draper GJ. Treatment centre size, entry to Alan R. Liss, 1984:49–61. trials, and survival in acute lymphoblastic leukemia. Spicker SF, Alon I, De Vries A, Engelhardt HT. The use of Archives of Diseases in Childhood 1989;64:219–23. human beings in research, with special reference to Stiller CA, Eatock EM. Survival form acute non- clinical trials. Boston: Kluwer, 1988. lymphocytic leukaemia 1971–88: a population-based Spiegelhalter DJ. Statistical issues in studies of individual study. Archives of Diseases in Childhood 1994;70:219–23. response. Scandinavian Journal of Gastroenterology Stirrat GM, Farrow SC, Farndon J, Dwyer M. The 1988;147:40–5. challenge of evaluating surgical procedures. Annals of the Spiegelhalter DJ, Freedman LS, Parmar MK. Applying Royal College of Surgeons of England 1992;74(2):80–4. Bayesian ideas in drug development and clinical trials. Statistics in Medicine 1993;12(15–16):1501–11, 1513–7. Sutherland HJ, Meslin EM, Till JE. What’s missing from current clinical trial guidelines? A framework for Spiegelhalter DJ, Freedman LS, Parmar MKB. Bayesian integrating science, ethics, and the community context. approaches to randomised trials. Journal of the Royal Journal of Clinical Ethics 1994;5(4):297–303. Statistical Society 1994;157:357–416. Sutton-Tyrrell K, Snyder JV, Kelsey S, Abramson N. Brain Spielberger CD. State-trait anger expression inventory: Resuscitation Clinical Trial Study Group. Risk monitoring professional manual review. Odessa, FL: Psychological of randomized trials in emergency medicine. American Assessment Resources, 1988. Journal of Emergency Medicine 1991;9(2):112–7. Spiro HM. Doctors, patients and placebos. Yale University Takaro T, Hultgren HN, Lipton MJ, Detre KM. The VA Press, London. 1986. cooperative randomized study of surgery for coronary Spodick DH. The surgical mystique and the double arterial occlusive disease II. Subgroup with significant standard. Controlled trial of medical and surgical left main lesions. Circulation 1976;54(6):107–17. therapy for cardiac disease. American Heart Journal Tannock IF, Boyer M. When is a cancer treatment 1973;85:579–83. worthwhile? New England Journal of Medicine Spodick DH. The randomized controlled clinical trial: 1992;323:989–90. scientific and ethical bases. American Journal of Medicine Tannsjo T. The morality of clinical research – a case 1982;73:420. study. Journal of Medicine and Philosophy 1994;19(1):7–21. Sposto M, Krailo MD. Use of unequal allocation in Tauer CA. The NIH trials of growth hormone for short survival trials. Statistics in Medicine1987;6:119–26. stature. IRB: A Review of Human Subjects Research Stamer CF, Lee KL. A data-based approach to assessing 1994;16:1–9. clinical interventions in the setting of chronic disease. Taves DR. Minimization: a new method of assigning Cancer Treatment Reports 1982;66:1083–7. patients to treatment and control groups. Clinical Stamler J. Invited remarks on “Terminating a long-term Pharmacology and Therapies 1974;15(5):443–53. clinical trial”. Clinical Pharmacology and Therapeutics Taylor KM. The doctor’s dilemma: physician 1979;25(5):651–4. participation in randomized clinical trials. Cancer Steinberg A. Ethical considerations in modern human Treatment Reports 1985;69:1095–100. experimentation. Cancer Investigation 1991;9(1):99–105. Taylor KM. Integrating conflicting professional roles: Stephens RJ, Gibbons D. The impact of clinical trials physician participation in randomized clinical trials. of the treatment of lung cancer. Clinical Oncology Social Science and Medicine 1992;35(2):217–24. 1993;5:211–9. Taylor KM, Kelner M. Informed consent: the physicians’ Stephenson T. Telling parents all relevant details might perspective. Social Science and Medicine 1987a;24:135–43. reduce recruitment of children to trials. British Medical Taylor KM, Kelner M. Interpreting physician Journal 1996;313. participation in randomized clinical trials: the physician Stewart LA. The role of overviews. In: Williams CJ, editor. orientation profile. Journal of Health and Social Behavior Introducing new treatments for cancer: practical, ethical, 1987b;28:389–400. and legal problems. Chichester: Wiley, 1992:383–401. Taylor KM, Margolese RG, Soskolne CL. Physicians’ Stewart LA, Parmar MKB. Meta-analysis of the literature reasons for not entering patients in a randomized or of individual patient data: is there a difference? Lancet control clinical trial of surgery for breast cancer. 1993;341:418–22. New England Journal of Medicine 1984;310:1363–7. Stiller CA. Survival of patients with cancer: those Taylor KM, Shapiro M, Soskolne CL, Margolese RG. included in clinical trials do better. British Medical Journal Physician response to informed consent regulations for 74 1989;299:1058–9. randomized clinical trials. Cancer 1987;60:1415–22. Health Technology Assessment 1998; Vol. 2: No. 15

Thompson SG, Pocock SJ. Can meta-analysis be trusted? Vere DW. Controlled clinical trials – the current ethical Lancet 1991;338:1127–30. debate. Proceedings of the Royal Society of Medicine 74:85–7, 1981. Thornton H. Breast cancer trials: a patient’s viewpoint. Lancet 1992;339:44–5. Vere DW. Problems in controlled trials – a critical response. Journal of Medical Ethics 1983;9:85–9. Thornton H. Clinical trials – a brave new partnership? Journal of Medical Ethics 1994;20:19–22. Visscher MB. The two sides of the coin in the regulation of experimental medicine. Annals of the New York Academy Thornton H, Baum M. Ethics of clinical trials – the of Sciences 1970;169:319–29. forbidden fruit phenomena. Breast 1996;5:1–4. Vogt TM, Ireland CC, Black D, Camel G, Hughes G. Thornton JG, Lilford RJ. Preterm breech babies and Recruitment of elderly volunteers for a multicentre randomized trials of rare conditions. British Journal of clinical trial. Controlled Clinical Trials 1986;7:118–33. Obstetrics and Gynaecology 1996;103:611–13. Wallenstein SL, Heidrich G, Kaiko R, Houde RW. Clinical Thornton JG, Lilford RJ, Levine M. Growth Retardation evaluation of mild analgesics: the measurement of Intervention Trial (GRIT): pilot study. MRC award 1995. clinical pain. British Journal of Clinical Pharmacology Unpublished. 1980;10(Suppl 2):319S–27S. Tindall B, Forde S, Ross MW, Goldstein D, Barker S, Ward LC, Fielding JWL, Dunn JA, Kelly KA. The selection of cases for randomized trials: a registry survey Cooper DA. Effects of two formats of informed consent of concurrent trial and non-trial patients. British Journal on knowledge amongst persons with advanced HIV of Cancer 1992;66:943–50. disease in a clinical trial of didanosine. Patient Education and Counselling 1994;24(3):261–6. Watson A, Vanderchove P, Lilford RJ, Vai A, Brosens I, Hughes E. A meta-analysis of the therapeutic role of oil Tomamichel M, Sessa C, Herzig S, de Jong J, Pagani O, soluble contrast media at hysterosalpingogram: a Willems Y, Cavalli F. Informed consent for phase i studies: surprising result? Fertility and Sterility 1994;61:470–7. evaluation of quantity and quality of information provided to patients. Annals of Oncology 1995;6(4):363–9. Weijer C, Fuks A. The duty to exclude: excluding people at undue risk from research. Clinical and Investigative Toynbee P. No one really wins in this life and death Medicine 1994;17(2):115–22. lottery. The Independent May 1996:14–15. Weinstein MC. Allocation of subjects in medical experi- Trochim WM, Cappelleri JC. Cutoff assignmment ments. New England Journal of Medicine 1974;291:1278–85. strategies for enhancing randomized clinical trials. Wennberg J. What is outcomes research? In: Gelinjns AC, Controlled Clinical Trials 1992;13:190–212. editor. Medical innovations at the crossroads, vol 1. Truog RD. Randomised control trials: lessons from Modern methods of clinical investigation. Washington, ECMO. Clinical Research 1992;40(3):519–27. DC: National Academy Press, 1990:33–46. Tukey JW. Some thoughts on clinical trials, especially Weymuller EA Jr. A consideration of ethical issues in the problems of multiplicity. Science 1977;198:697–84. design of clinical trials. American Journal of Otolaryngology – Head and Neck Medicine and Surgery 1996;17:2–11. Twomey JG. Investigating Pediatric HIV research ethics White DR, Muss HB, Michielutte R, Cooper MR, Jackson in the field. Weston Journal of Nursing Research DV, Richards F, et al. Informed consent: patient 1994;16(4):404–13. information forms in chemotherapy trials. American UK Collaborative ECMO Trial Group. Extra-corporeal Journal of Clinical Oncology 1984;7:183–90. membrane oxygenation (ECMO): the UK collaborative Whitehead J. Sequential methods based on the neonatal ECMO trial. Lancet 1996;345:805–6. boundaries approach for the clinical comparison of Uberla AK. Randomised clinical trials: why not? Controlled survival times. Statistics in Medicine 1994;13:1357–68. Clinical Trials 1981;1:295–303. Wilcox M, Schroer. The perspectives of patients with vascular disease on participation in clinical trials. Journal Urbach P. The value of randomization and control in of Medical Ethics 1994;12:112–16. clinical trials. Statistics in Medicine 1993;12:1421–41. Wilhelmsen L. Ethics of clinical trials: the use of placebo. Van der Linden W. Pitfalls in randomized surgical trials. European Journal of Clinical Pharmacology 1979;16:295–7. Surgery 1980;87:258–62. Wilhelmsen L, Ljungberg S, Wedel H, Werko L. A Vandekerchove P, O’Donovan, Lilford RJ, Harada E. comparison between participants and nonparticipants Treatment of male infertility – from cookery to science. in a primary prevention trial. Journal of Chronic Disease British Journal of Obstetrics and Gynaecology 1976;29:331–9. 1993;100:159–60. Willan AR. Power function arguments in support of an Veatch RM. A theory of medical ethics. New York: Basic alternative approach for analyzing management trials. books, 1981. Controlled Clinical Trials 1994;15:211–19. 75 References

Williams CJ, Zwitter M. Informed consent in European Young FE. Ethical and financial considerations in third- multicentre randomized clinical trials – are patients party support of investigational cancer therapies. Cancer really informed? European Journal of Cancer 1993;72(9):2854–8. 1994;30(7):907–10. Yusuf S, Collins R, Peto R. Why do we need some Williamson C. Breast cancer: asking patients what they large simple randomized trials? Statistics in Medicine want. British Medical Journal 1996;313:506–7. 1984;3:409–20.

Williford WO. Comparison of eligible randomised Zelen M. Play the winner rule and the controlled patients with two groups of ineligible patients: can the clinical trial. Journal of the American Statistical Society results of the VA total parenteral nutrition clinical trial 1969;64:131–46. be generalised? Journal of Clinical Epidemiology 1993;46(9):1025–34. Zelen M. Strategy and alternative randomized designs in cancer clinical trials. Cancer Treatment Reports Working Group on Arrhythmias of the European Society 1982;66:1095–100. of Cardiology. The early termination of clinical trials: causes, consequences and control. European Heart Journal Zelen M. Guidelines for publishing papers on cancer 1994;15(6):721–38. clinical trials: responsibilities of editors and authors. Journal of Clinical Oncology 1983;1:164–9. Wynne A. Is it any good? The evaluation of therapy by participants in a clinical trial. Social Science and Medicine Zelen MA. A new design for randomised clinical trials. 1989;29:1289–97. New England Journal of Medicine 1979;300:1242–5.

76 Health Technology Assessment 1998; Vol. 2: No. 15

Tables

TABLE 1a Effect of participating in clinical trials on mortality

Author Type of trial Sample population and Outcome Main results Comments (see method of assessment measure Table 3 for quality of evidence checklist) (1) Antman Randomised trial of Trial participants (n = 36) Disease-free Randomised patients did better A2 (1983) adjuvant adriamycin Eligible non-trial controls (n = 48) survival and than non-trial controls according B4 for sarcoma of which 60% had refused to be actuarial to disease-free survival (p < 0.07) C3 randomised and 40% had not mortality and mortality (p < 0.13) D1 been offered randomisation E+ (2) Bertelsen Two Danish ovarian Trial participants: Mortality – For patients with early disease (T1), A2 (1991) cancer trials: T1 – n = 72 actuarial there was no difference in actuarial B3 T1 – to compare two T2 – n = 265 survival survival between randomised and C2.There was no different adjuvant Non-trial eligible controls: non-randomised patients (p = 0.45). survival advantage for treatments of early T1 – n = 52 However, trial patients with late experimental ovarian cancer T2 – n = 96 disease (T2) had a significantly higher treatments in both trials T2 – to compare two Data obtained from the survival rate than similar non- D3 chemotherapy treatment DACOVA register (Danish randomised patients (p = 0.0002) E+ of late stages of cancer Ovarian Cancer Group) This result was not sustained when 23 and 24% of the Trial eligibility criteria: the comparison was restricted to non-randomised Age < 70 years patients receiving combination samples for trials 1 and Epithelial ovarian cancer with chemotherapy (given to all trial 2, respectively, did not historical verification of diagnosis participants but only 47% of receive any treatment, Medical contraindication for post- non-trial patients) whereas all trial operative treatment participants received Concomitant malignant disease during Both randomised studies showed an active therapy the last 5 years, except basocellular no evidence of a survival difference carcinoma of the skin or carcinoma between the treatment regimens in situ of the uterine cervix (3) CASS Randomised trial of Patients with mild/moderate 5 year There were no significant differences A2 Working coronary artery bypass stable angina pectoris (class 1 or 2 actuarial in survival between the trial B3 Group surgery to compare severity), or free of angina with a survival participants and non-trial controls C1. Survival of patients (1984) medical versus surgical documented history of myocardial either overall or according to assigned to the control treatment infarction whether medical or surgical arm (medical group) in treatment was given (no p value the trial was compared Trial participants (n = 779) given) with that of patients Eligible non-trial controls (n = 1309), receiving the same type 57% of whom were treated medically There were no significant findings in of treatment outside while the remainder were received survival between trial and non-trial the trial surgery. All non-trial controls were patients according to analysis of the D1. 59% recruitment taken from the same institutions three clinical subgroups (no p value rate who had refused randomisation given) E+

Trial exclusion criteria: There was a difference Prior coronary bypass surgery in prognosis at the Progressive or unstable angina outset.The extent of Angina more severe than class 2 the disease was less Congestive heart failure extensive in the non- A coexisting illness that would trial medical control increase the likelihood of death patients than in the within 5 years medical trial participants. However, for Three subgroups were also the surgical groups, generated according to the number the disease was more of diseased vessels, the presence of extensive in the non- proximal left anterior descending trial controls. Clinical coronary artery disease and subgroups were com- ejection fraction pared for trial and non- trial patients to get round this continued 77 Tables

TABLE 1a contd Effect of participating in clinical trials on mortality

Author Type of trial Sample population and Outcome Main results Comments (see method of assessment measure Table 3 for quality of evidence checklist)

(4) Davis et al. Randomised trials for Patients were selected from a register Mortality – Survival rates were better for trial A2 (1985) resected non-small at the Cancer Surveillance System (CSS) actuarial cases at 12 months (93%) and B3 cell lung cancer survival 24 months (82%) than for non-trial C2 The trial patients were Trial participants (n = 78) controls (72 and 50%, respectively). D3 those entered into Non-trial controls (n = 471) were The overall survival experience of the E+ 1 of 4 adjuvant trials patients with primary lung cancer in CSS trial cases was significantly better than of investigational population not in trial.Also, a subset of that seen in controls (p < 0.001).This Accuracy of CSS immunotherapy, controls were selected who would have survival advantage was still observed register assessed by chemotherapy, or been eligible for the trials and who were for the trial cases against the subset reviewing 78 cases with radiation therapy matched to trial cases according to of controls which were matched for independent records, after surgery tumour size, regional lymph node known prognostic factors, and when from which there were involvement and cell type (n = 152) the analysis was adjusted for age, sex, 2.4% discrepancies and administration of radiation Survival estimates Trial eligibility criteria: therapy (p < 0.02) were recorded by the The patient has a surgically resectable Kaplan–Meier model non-small cell carcinoma of the lung Survival of the trial patients receiving Multivariate analysis The patients extent of disease is fully investigational therapy was not conducted by use of documented at surgery statistically different from those the Cox proportional A total resection of the tumour is achieved assigned to the placebo or hazards model Material is available for historical review ‘standard’ therapy

Contraindications of eligibility include: Prior radiotherapy, chemotherapy or immunotherapy Prior history of cancer Active tuberculosis within the last 5 years History of active cardiac disease

(5) Jha et al. Two thrombolytic GUSTO participants (n = 1304) Mortality – GUSTO trial patients had a A2 (1996) trials: Eligible but non-trial controls from: survival statistically significant survival B3 Trial 1 – Global Same hospital (n = 12,657) advantage over non-trial controls C2.There was a non- Utilisation of Strepto- External hospitals (n = 12,299) at the same hospital (OR = 1.8) significant advantage for kinase and Tissue the experimental Plasminogen Activator LATE participants (n = 211) LATE trial patients showed a statis- treatment in the LATE for Occluded Eligible but non-trial controls: tically significant survival advantage trial, and a statistically Coronary Arteries Same hospital (n = 5997) over non-trial controls (OR = 2.1) significant advantage for (GUSTO) External hospital (n = 12,299) the experimental Trial 2 – Late Assess- Trial patients had significantly better treatment over four ment of Thrombolytic Trial eligibility criteria:hospitalisation survival on comparison with non-trial standard treatments in Efficacy (LATE) for acute myocardial infarction with patients at external hospitals (OR = the GUSTO trial concomitant bypass graphing or 1.8 for both GUSTO and LATE) D3 coronary angioplasty E+ All results remained statistically Data on concurrent non-trial controls significant after adjustment for age, were obtained from the Hospital Medical sex, bypass graphing or coronary Records Institute (HMRI) databases angioplasty, and co-morbidity index

(6) Karjalainen A district in Finland Patients living in the district before the Mortality – The 5 year relative survival rate of Verification of the and Palva where there was a trials from 1959 acted as part of the actuarial 38% for residence of the trial district, diagnosis was missing (1989) policy to enter patients control group (n = 85), along with relative irrespective of whether they were in 4.8% of the cases in with multiple myeloma patients who did not live in the district survival actually randomised, was significantly the trial area and in 7% in a clinical trial.Three where policy prevailed during the trial higher than the 28% survival for of the cases in the clinical trials were period (n = 165) those from non-trial districts.This reference area available between Only some of the patients living in the advantage was only obtained for 1979–1985 trial district were actually randomised the period beyond 2 years A2/3 between 1979 and 1985 (n = 319) following diagnosis B5. See text Expected survival was obtained from C2. Experimental treat- death rates in the general population and During the second trial, it was found ment was less effective actual survival rates were obtained via that the experimental treatment than the standard treat- files of the Central Statistical Office was less effective than the reference ment in one trial, and of Finland treatment which was the main there was no advantage regimen in the reference area for maintenance treatment in the second trial. The results of the third trial were not available D3 E+

78 continued Health Technology Assessment 1998; Vol. 2: No. 15

TABLE 1a contd Effect of participating in clinical trials on mortality

Author Type of trial Sample population and Outcome Main results Comments (see method of assessment measure Table 3 for quality of evidence checklist)

(7) Lennox MRC clinical trial Trial participants (n = 98) Mortality – Overall 3 year survival rate for both A2 et al. (1979) for children with Eligible but non-trial patients (n = 172) 3 year groups was 58%. Survival rates were B3 nephroblastoma survival significantly higher for children C2. Non-significant Trial eligibility criteria:children aged included in an MRC trial (77%) than survival advantage for 1–14 years with historically confirmed for those who were eligible but not the experimental drug unilateral nephroblastoma unless they included (58%), with p < 0.01.This vincristine were known to have metastases in the result was more pronounced when D3 liver or outside the abdominal cavity allowance was made for the E+ distribution of age and tumour The non-trial controls were obtained from stage (p < 0.001) consultants or from Marie Curie Oxford Survey of childhood cancers

(8) MRC Clinical trials for Trial participants (n = 220) (treated Median Age is a known prognostic factor and Some of the data from Working children with by physicians associated with the survival so survival was reported as follows: the registers were Group on acute leukaemia MRC committee) 0–1 year: missing as only 11/12 Leukaemia in trial cases 39 weeks directors supplied Childhood Non-trial controls (not treated by controls 14 weeks information (1971) physicians associated with the MRC 2–8 years: committee) (n = 691) trial cases 72 weeks A2 controls 36 weeks B3 Non-trial controls were obtained from 9–14 years: C3. Comparison of National Cancer Registration Scheme trial cases 74 weeks physicians, not trials on children aged 0–14 years during the controls 22 weeks period running concurrently with trials D3 (1963–1967) E+

(9) Schmoor Three multicentre Trial participants (n = 734) Mortality – Trial 2 – disease-free survival rates Trial 1 left out of the et al. (1996) clinical trials Eligible non-trial controls (n = 1350) 5 year of randomised and non-randomised analysis conducted by the Non-trial controls were patients who recurrence- patients was nearly identical German Breast refused randomisation free Trial 3 – disease-free survival rates A2 Cancer Group survival were slightly higher (not significantly) B2 (the No. of Eligibility criteria not given among randomised patients than C1. Specific treatments patients randomised non-randomised controls effects of experimental to trial 1 was so Statistical adjustment was made for and control arms in small that it was prognostic factors and the remained trials were compared regarded as an no significant difference between with those outside the observational study) groups, though a smaller relative risk trial using the Cox Trial 1 – mastectomy was estimated for non-randomised proportional hazards versus lumpectomy controls than randomised patients ratio + radiation indicating a beneficial effect of D1 Trials 2 + 3 – tamoxifen in trial 2 E+ different adjuvant therapies in patients Trial 2 produced an apparently bene- previously treated ficial effect of tamoxifen (p = 0.085, by mastectomy RR = 0.75, confidence interval 0.5–1.04), while trial 3 estimated a benefit from additional radiotherapy (HR = 0.79, confidence interval 0.5–1.3)

(10) Stiller Chemotherapy Trial participants (n = 2137) Mortality – Children entered into an MRC trial A2 and Draper trials for childhood Non-trial controls (n = 1933) 5 year had a significantly higher actuarial B3 (1989) lymphoblastic All children aged < 15 years who survival, and survival rate than those who were C2.The experimental leukaemia in were diagnosed with leukemia during actuarial not.This result was essentially treatment was signifi- accordance with 1971–1982 were recorded from survival unchanged when allowance was cantly less effective in MRC protocols national cancer registration schemes thereafter made for age and white cell count half of the UKALL trials in England, Scotland and Wales. up to 1986 (p < 0.0001).When the analysis was D3 Notifications of children entered in limited to children surviving at least E+ MRC UKALL trials during the same 3 months from diagnosis, the effects period were received. Non-trial of trial entry remained highly signifi- controls were also recorded from cant (no p value given). However, the the UK Children’s Cancer Study subsequent survival rate for those Group (CCSG) during 1977–1982 who had survived 5 years did not differ between trial and non-trial patients

HR, hazard ratio; RR, relative risk

continued 79 Tables

TABLE 1a contd Effect of participating in clinical trials on mortality

Author Type of trial Sample population and Outcome Main results Comments (see method of assessment measure Table 3 for quality of evidence checklist)

(11) Stiller National chemo- Trial participants (n = 592) Mortality – Entry into a trial was associated A2 and Eatock therapy trials for Non-trial controls (n = 666) 5 year with a higher actuarial survival rate B3 (1994) children with acute Data obtained from National cancer registry, actuarial in an analysis allowing for age at C2.There was a slight non-lymphoblastic specialist children’s tumour registries, survival diagnosis (p < 0.01) advantage for the experi- leukaemia in MRC trials, CCSG and death certificates. mental treatments in accordance with Confirmation of diagnoses obtained from two trials, and no advan- MRC protocols hospital records, family doctors and clinical tage in a further three trial records trials. Some trial data not available D3 E+

Non-treated children were excluded from all survival analyses

(12) Ward British Stomach Trial participants (n = 217) Mortality – The trial participants had a higher A2 et al. (1992) Cancer Group Eligible non-trial controls (n = 960) actuarial survival rate than the non-trial B2 trial of adjuvant Non-trial controls were obtained from survival sample, though this did not reach C2.There was no survival chemotherapy cancer registry survey statistical significance (p = 0.24). advantage with adjuvant versus placebo However, when the non-trial patients therapy in operative Trial ineligibility criteria: were assessed for eligibility there D3 cancer Unable to attend were only n = 493, and there was E+ Not pathologically confirmed adenocarcinoma no observed survival difference of the stomach or carcinoma in situ between these two groups Previous tumour registration for frank malignant condition or previous treatment with chemotherapy or radiotherapy Stage 1 or non-operative stage 4 Unfit to start chemotherapy within 12 weeks of surgery Diagnosed at initial treatment as an unknown primary site

80 Health Technology Assessment 1998; Vol. 2: No. 15

TABLE 1b Effect of participating in clinical trials on morbidity

Author Type of trial Sample population and Outcome Main results Comments (see method of assessment measure Table 3 for quality of evidence checklist)

(1) Mahon n of 1 trials of Trial participants (n = 14) Morbidity – There were no differences in A1 et al. (1996) theophylline versus Non-trial controls (n = 12) exercise exercise capacity or quality of life B1 placebo for irre- capacity by measures, though significantly fewer C3 versible airflow Patients were randomised to either 6 min walking n of 1 trial patients than non-trial D3 limitation n of 1 trial or standard practice. distance, quality controls were taking theophylline E.The analysis was of life measured at 6 months biased in favour of by respiratory early stopping. disease questionnaire at baseline and 6 months, and administration of theophylline at 6 months

(2) Reiser Three asthma trials Trial participants: Morbidity T1. Although there was no A3 and Warner carried out in children: T1. n = 25 children, aged 7–14 years, improvement in tests which reflect B5 (1985) recruited from paediatric asthma clinic peripheral airway calibre, there was C3.There was no actual T1.Trial to assess the who were thought to have stable, well- a significant improvement in clinical data presented which efficacy of 750 ml controlled asthma at trial entry.They status (diary card symptom scores) would indicate that the spacer for the admin- were monitored by diary cards record- and in tests reflecting more proximal improvement by trial istration of inhaled ing daily symptom scores, drug use, and airway calibre in both groups.There placebo controls was in corticosteriods. twice daily peak expiratory flow rates was a significant period effect when any way greater than is Children inhaled (PEFR). Children also attended a clinic comparing the first 2 months with normally observed. budesonide for monthly for assessment and for more the last 2 months for either group However, the authors 2 months using the detailed lung function tests (p < 0.02 in the morning and make reference to a Nebuhaler® and for p < 0.01 in the evening) published guide on the 2 months used the T2. n = 47 children with seasonal This shows that regardless of the pharmacology of metered-dose inhaler asthma and rhinitis, poorly controlled mode of administration of the placebos alone.The order of on conventional therapies. Progress was steroid aerosol, over the first D3 the devices was monitored by recording daily diary card 2 months of the trial there was E+ randomised by trial scores of symptoms and drug usage a significant improvement in entry No. in an open and 2 x daily PEFR if indicated symptoms and large-airway calibre For studies 2 and 3, there crossover design was no indication given At the end of the season, patients and T2.There was no significant advan- as to how the patient’s T2.Trial to assess the parents were asked about the tage for the active vaccine over the and parents’ assessments efficacy of orally effectiveness of the treatment placebo, though there was observed of the treatment efficacy administered grass benefit for most of the children was carried out pollen extracts. Oral T3. n = 51 patients with moderate– according to the parents’ assess- mixed-grass pollen severe perennial asthma were selected ments: for the active vaccine, 21 said vaccine versus a from a paediatric respiratory clinic. there was benefit and none said there matched placebo Patients completed diary cards of was no benefit; for the placebo, 20 said was allocated in a symptoms and drug usage and recorded there was benefit and six no benefit. random, double-blind 2 x daily PEFR. At visits every 2 months This placebo response was claimed study for the two the patients and parents were asked if by the authors to be much stronger groups their asthma was better, whether there than is usually found was no change or any worsening. More T3.Trial of placebo detailed lung function tests were also T3. More patients and parents in the versus the active done active group had the impression that treatment the asthma had improved after a year, D. pteronyssinus Informal non-trial controls but all the observations tended to absorbed in tyrosine. improve in both groups during the Injections were given course of the year though there was once a week for a slight increase in drug scores in 6 weeks and then the placebo group. Again the authors every 8 weeks for interprets the improvement of the a year placebo group as being greater than normal

(3) Williford RCT of total Trial participants (n = 396) Morbidity – Complication rates were significantly A2 et al. (1993) parenteral Eligible non-trial controls (n = 199) septic and higher for trial patients than for B3 nutrition (TPN) non-septic non-trial patients (p = 0.008 at C3 in malnourished Non-trial controls refused complication 30 days and p = 0.001 at 90 days) D1. Recruitment rate to surgical patients randomisation and subsequently rates When adjusted for TPN use the trial 66% underwent an operation without complications rate was no longer E+ receiving TPN statistically significant (p = 0.650 at 30 days and p = 0.598 at 90 days) 81 Tables

TABLE 2a Effect of participating in clinical trials on mortality according to quality of evidence

Author Outcome Type of Nature of Prognostic Finding before Finding with an Trial control Summary of measures trial(s) non-trial variables taken adjustment for adjustment for compared quality of evidence control into account non-trial and non-trial control with non- (see Table 3) trial group and trial group trial control

(1) Antman Disease- Clinical trial Non-randomised There was a differ- Favourable trend – No data A2 (1983) free and of adjuvant concurrent controls ence in prognostic for disease-free available B4 actuarial adriamycin who declined trial variables between and actuarial C3 survival for sarcoma entry and non-trial the trial groups, but survival (p < 0.07 D1 no adjustment and p < 0.13, E+ was made respectively)

(2) Bertelsen Actuarial Two ovarian Non-randomised Statistical adjustment Survival experience No significant No significant A2 (1991) survival cancer trials concurrent controls for known prognostic was significantly difference treatment B3 of adjuvant who were not differences higher in the trial effect for C2 or chemo- offered trial group than among either trials D3 therapeutic entry non-trial controls E+ treatments (p = 0.0005) 23 and 24% of the non-randomised samples for trials 1 and 2, respectively, did not receive any treatment, whereas all trial participants received an active therapy

(3) CASS 5 year Clinical trial Non-randomised Statistical adjustment Favourable trend No significant No significance A2 Working actuarial of coronary concurrent controls for known prognostic for trial group difference was found B3 Group survival artery bypass who declined trial differences (p value not given) between the C1. Survival of (1984) surgery entry trial control patients assigned versus arm and non- to the control arm medical trial patients (medical group) in therapy receiving the the trial was com- same therapy pared with that of patients receiving the same type of treatment outside the trial D1. 59% recruitment rate E+

There was a difference in prognosis at the outset. The extent of the disease was less extensive in the non- trial medical control patients than in the medical trial participants. However, for the surgical groups, the disease was more extensive in the non-trial controls. Clinical subgroups were compared for trial and non-trial patients to get round this

continued 82 Health Technology Assessment 1998; Vol. 2: No. 15

TABLE 2a contd Effect of participating in clinical trials on mortality according to quality of evidence

(4) Davis Actuarial Four trials in Non-randomised Statistical adjustment Survival experience Significant No significant A2 et al. (1985) survival lung cancer concurrent controls for known prognostic was significantly difference was treatment B3 comparing who were not differences higher in the trial maintained effect in any C2 postoperative offered trial group than among (p < 0.02) of the trials D3 immuno- entry non-trial controls E+ therapy, (p < 0.001) chemotherapy Accuracy of CSS or radiation register assessed by therapy reviewing 78 cases with independent records, from which there were 2.4% discrepancies Survival estimates were recorded by the Kaplan–Meier model Multivariate analysis conducted by use of the Cox proportional hazards model

(5) Jha et al. Actuarial Two thrombo- Non-randomised Statistical adjustment Survival experience Significant There was a A2 (1996) survival lytic trials for concurrent controls for known prognostic was significantly difference was non-significant B3 myocardial who were not differences higher among trial maintained (OR advantage for C2 infarction offered trial entry cases than among = 1.8 and 2.1 for the experi- D3 (GUSTO non-trial cases GUSTO and LATE, mental treat- E+ and LATE). (p value not given) respectively) ment in the LATE trial, and a statistically significant advantage for the experimental treatment over four standard treatments in the GUSTO trial

(6) Karja- Comparison Three clinical Non-randomised Adjustment not Survival in trial The experi- Verification of the lainen and of actuarial trials for concurrent controls possible because data district was mental treat- diagnosis was missing Palva (1989) survival by multiple who were not on individual patients significantly higher ment was less in 4.8% of the cases district of myeloma offered trial entry was not available than in non-trial effective than in the trial area and residence (because they lived district (p < 0.001) the standard in 7% of the cases in in different districts) treatment in the reference area one trial, and there was no A2/3 advantage for B5. See text maintenance C2 for 1/3 trials and treatment in C3 for 2/3 trials the second trial. D3 The results of E+ the third trial were not available

(7) Lennox 3 year Trial of Non-randomised Statistical adjustment Survival experience Significant There was a A2 et al. (1979) survival chemotherapy concurrent controls for known prognostic was significantly difference was non-significant B3 for children who were not differences higher in the trial more pronounced survival advan- C2 with nephro- offered trial entry group than among (p < 0.001) tage for experi- D3 blastoma non-trial controls mental drug E+ (p < 0.01)

continued 83 Tables

TABLE 2a contd Effect of participating in clinical trials on mortality according to quality of evidence

(8) MRC Median MRC clinical Non-randomised Statistical adjust- Survival experience Significant Comparison by Some of the data Working survival trials for concurrent con- ment for known was significantly difference was physician not from the registers Group on children with trols who were prognostic higher in trial maintained by trial were missing Leukaemia Comparison acute leukaemia not offered trial differences group than among as only 11/12 in Childhood between (No. of trials entry because non-trial controls directors supplied (1971) patients not given) they attended (p value not given*) information treated by different physicians physicians A2 associated B3 with the C3 MRCtrials D3 and those E+ not

(9) Schmoor 5 year Two multicentre Non-randomised Trial and non-trial Favourable trend – No significant Trial 1 left out of et al. (1996) recurrence- clinical trials concurrent groups were in one of the trials difference was the analysis free survival of different adju- controls who matched for with virtually no found between A2 vant therapies declined trial known prognostic difference between the trial control B2 for breast cancer entry variables groups in the other arm and non- C1. Specific trial patients treatment effects receiving the of experimental same therapy and control arms in trials were compared with those outside the trial using the Cox proportional hazards ratio D1 E+

(10) Stiller 5 year Chemotherapy Non-randomised Statistical adjust- Survival experience Significant The A2 and Draper actuarial trials for concurrent ment for known was significantly difference was experimental B3 (1989) survival childhood controls who prognostic higher in trial group maintained treatment C2 lymphoblastic were not offered differences than among non- (p < 0.0001) was significantly D3 leukaemia trial entry trial controls less effective E+ (p value not given) in half of the UKALL trials

(11) Stiller 5 year Chemotherapy Non-randomised Statistical adjust- Survival experience Significant There was a A2 and Eatock actuarial trials for children concurrent ment for known was significantly difference was slight advantage B3 (1994) survival with acute non- controls who prognostic higher among trial maintained for the experi- C2 lymphoblastic were not offered differences cases than among (p < 0.01) mental treatments D3 leukaemia trial entry non-trial cases in two trials, and E+ (p value not given) no advantage in a further three Non-treated trials. Some trial children were data were not excluded from all available survival analyses

(12) Ward Actuarial Stomach cancer Non-randomised Trial and non-trial Favourable trend There was no A2 et al. (1992) survival trial of adjuvant concurrent groups were (p = 0.24) survival advantage B2 chemotherapy controls who matched for with adjuvant C2 versus placebo were not offered known prognostic therapy D3 trial entry variables E+

* This analysis was conducted by Stiller (1987) in a previous review. 84 Health Technology Assessment 1998; Vol. 2: No. 15

TABLE 2b Effect of participating in clinical trials on morbidity according to quality of evidence

Author Outcome Type of Nature of Prognostic Finding before Finding with Trial control Summary of measures trial(s) non-trial variables taken adjustment for an adjustment compared quality of evidence control into account non-trial and for non-trial with non- (see Table 3) trial group control and trial control trial group

(1) Mahon Morbidity – n of 1 trials of Randomised This was a Significantly higher Favourable trend – No data available, et al. (1996) exercise theophylline concurrent randomised trial of proportion of non- for exercise i.e. the trial was capacity, versus placebo controls. n of 1 trials, i.e. trial controls was capacity and itself the quality of life for irreversible trial and non-trial taking theophylline quality of life intervention. (measured by airflow groups generated at 6 months than measures respiratory limitation by randomisation n of 1 trial cases (p value A1 disease ques- (p value not given) not given) B1 tionnaire) at C3 baseline and D3 6 months, and E. The analysis was theophylline biased in favour of requirements early stopping at 6 months

(2) Reiser Morbidity – Three childhood Non-randomised No comment on A significant – There was an A3 and Warner patient- asthma trials of historical controls prognostic variables improvement in improvement in B5 (1985) assessed different inhalers between trial and symptoms was ex- symptoms for C3.There was no clinical status non-trial groups perienced among trial control actual data presented in all trials trial cases in one of cases but not which would indicate the trials (p value for non-trial that the improvement not given) with controls by trial placebo non-significant controls was in any improvement in way greater than is the remainder normally observed. However, the authors make reference to a published guide on the pharmacology of placebos D3 E+

For studies 2 and 3, there was no indication given as to how the patient’s and parents’ assessments of the treatment efficacy was carried out

(3) Williford Morbidity – RCT of TPN in Non-randomised Statistical adjust- Complication rate No significant No data A2 et al. (1993) septic and malnourished concurrent ment for known was significantly difference available B3 non-septic surgical patients controls who prognostic higher among trial C3 complication declined trial differences cases than among D1. Recruitment rate rates entry and who non-trial controls to trial 66% did not receive (p = 0.008 at E+ TPN 30 days and p = 0.001 at 90 days)

85 Tables

TABLE 3 Quality of evidence checklist for physical effects

Key Type of evidence

A Sampling: (1) RCT of trial versus no trial (2) non-randomised concurrent cohort study (3) non-randomised historical cohort study

B Prognostic variables: (1) randomised study, i.e. A1 (2) trial participants and non-trial controls matched for known prognostic variables (3) trial participants and non-trial controls not matched for prognostic variables but statistical adjustment was made (4) recorded imbalance between trial participants and non-trial controls with no statistical adjustment (5) no data on prognostic variables

C Comparison of trial controls and non-trial controls: (1) formal comparison of trial controls and non-trial controls carried out (2) no formal comparison but result of trial given* (3) no data on outcome of trial controls

D Patients refusing randomisation: (1) refusers followed up (2) recruitment rates given but refusers not followed up (3) recruitment rates not given

E Was the statistical method used to analyse the results appropriate (acceptable, positive; flawed, negative)?

* If the data were not given in the study, then the original trial publication was consulted.

TABLE 4a Categorisation of studies for Table 4b

1. Prognostic a. No prior effective Trial result: Where proportion of patients receiving treatment variables similar/ treatment included in i. experimental treatment significantly better of interest and/or regimen (package of care) between matched/adjusted trial protocol than placebo trial and non-trial patients might affect outcome: ii. null result/experimental treatment significantly α. Both similar worse than placebo β.Treatments balanced but regimen different χ. Regimens the same but treatment imbalance 2. Serious prognostic b. Effective treatment Trial result: δ. Both different imbalance between prior to trial which i. experimental significantly better than standard trial and non-trial was included in the ii. null result/ experimental treatment significantly patients trial protocol worse than standard

86 Health Technology Assessment 1998; Vol. 2: No. 15

TABLE 4b Data sets for Figure 3

Trial/ Design Effective Protocol ensures Differences in trial Results of trial Was there Quality Overall data set Category treatment similar regimens/ and non-trial treatments and/or a more of data quality (see Table 4a) already follow-up for similar patients prognoses: non-randomised adjusted acquisition assessment outside trial exists trial and non-trial 1 – no difference/ treatments result for graph: (within which was patients fully adjusted without 1 – good category): included in 2 – small differences/ sufficient 2 – okay 1 – good trial protocol partially adjusted data for 3 – bad 2 – okay 3 – no adjustment the graph? 3 – bad for large differences

(1) Schmoor et al. (1996) Trial 2 – Sc(i) 1biiχ Yes Treatment balance Trial and non-trial Nearly significant Yes 3 – measured 2 among trial and non-trial patients had similar difference between from graph patients differed but the prognoses for trial treatments regimens were the same both trials (1) (both trials) Trial 3 – Sc(ii) 1biiα Yes As above Non-significant Yes 3 – measured 2 difference between from graph trial treatments

(2) CASS Working Party (1984) Medical arm 2 Yes Similar treatments Some important No difference No 3 – measured 2 Surgical arm and regimens but differences between between trial from graph not identical trial and non-trial treatments patients within treatment arms (3)

Treatment arms 1biiα Yes Treatment balance among Not adjusted but no No difference No 3 – measured 2 combined – C trial and non-trial patients big difference overall between trial from graph differed and regimens (1) treatments were slightly different

(3) Davis et al. (1985) 1aiiβ Surgery No ‘Fully’ adjusted for No difference No 1 1 – D effective but recorded differences between trial not adjuvants (1) treatments

(4) Bertelsen (1991) III/IV patients with 1biiβ Yes Treatment the same No differences (2) Small difference No 3 – measured 2 cisplatinum + but possibly protocols between treatments from graph others – B different in the trial

III/IV trial versus 2 Yes No (including exclusions) No adjustment for As above No N/A N/A non-trial recorded differences (3)

I/II trial versus 2 Yes Slightly different No adjustment for As above No N/A N/A non-trial recorded differences (3)

(5) Lennox et al. (1979) Stage I/II age 1biiδ Yes No Adjusted for age and There was a non- Could have 1 1–2 1–3 L(i) stage 2 significant survival adjusted for a advantage for hospital effect experimental arm which could have due to the protocols used

Stage I/II age 1biiδ Yes No Adjusted for age and As above As above 1 1–2 4–14 L(ii) stage 2

Stage III all ages 1biiδ Yes No Adjusted for age and As above As above 1 1–2 L(iii) stage 2

(6) MRC Working Group on Leukaemia in Childhood (1971) Not trial versus 1biiδ Yes No Adjusted for age (2/3) Difference in No 3 3 non-trial comparison effectiveness of but expert versus treatments is probable non-expert – M but not investigated within a trial

continued 87 Tables

TABLE 4b contd Data sets for Figure 3

(7) Stiller and Draper (1989) Year group (i–iv) 1biiδ Yes No, though for patients in No adjustment for The experimental Yes (adjusted 2 – data from 3 by hospital category group A their treatments recorded differences treatment was for age) tables but (A–C) with 12 and regimens were (3) significantly less some guessing comparisons probably more similar effective in half of and assume than for those in B the UKALL trials complete Year group: and C 5 year (i) 1971–1973 follow-up (ii) 1974–1976 (iii) 1977–1979 (iv) 1980–1984

Hospital type: A > 6 patients treated per year on average 1 < B > 6 patients treated per year on average

C < 1 patient treated per year on average

(8) Stiller and Eatock Treatments partially adjust- There was a slight (1994) ed by stratifying period advantage for the experimental treat- Years 1975–1983 1biδ Yes No No adjustment for ments in two trials, Yes (3 month 3 – measured 3 – S94(i) recorded differences and no advantage in survival but no from graph (3) a further three trials. data for age and assume Some trial data were adjustment) complete not available 5 year Years 1984–1988 1biδ Yes No No adjustment for Yes (3 month follow-up 3 – S94(ii) recorded differences survival but no (3) data for age adjustment)

(9) Ward (1992) – W 1aiiβ No Adjuvants not widely No adjustment for No significant None 1 2 used outside trial small differences (2) difference between reported trial treatments

(10) Karjalainen and 1biiδ Yes No Yes - ‘self’-historical The experimental – 1 3 Palva (1989) – K control (3) treatment was less effective than the standard treatment in one trial, and there was no advantage for maintenance treatment in the second trial. The results of the third trial were not available

(11) Antman (1983) 2 Insufficient Insufficient data Yes (3) Insufficient data None Insufficient 3 data reported data

(12) Reiser and Warner – – – – – – – – (1985) Excluded as there were insufficient data for a graph

88 continued Health Technology Assessment 1998; Vol. 2: No. 15

TABLE 4b contd Data sets for Figure 3

Trial/ Design Effective Protocol ensures Differences in trial Results of Was there a Quality Overall data set Category treatment similar regimens/ and non-trial trial treatments more adjusted of data quality (see Table 4a) already follow-up for similar patients prognoses: and/or non- result without acquisition assessment outside trial exists trial and non-trial 1 – no difference/ randomised sufficient for graph: (within which was patients fully adjusted treatments data for 1 – good category): included in 2 – small differences/ the graph? 2 – okay 1 – good trial protocol partially adjusted 3 – bad 2 – okay 3 – no adjustment 3 – bad for large differences

(13) Jha (1996) Trial 1 GUSTO 1biδ Yes No (difference in Adjusted for important Yes There was a statis- No 1 – J(i) revascularisation differences (1) tically significant advan- controlled for) tage for the experi mental treatment in the GUSTO trial

Trial 2 LATE 1biiδ Yes No (difference in Adjusted for important Yes There was a non- No 1 – J(ii) revascularisation differences (1) significant advantage controlled for) for the experimental treatment in the LATE trial

(14) Mahon (1996) – – – – – – – – Excluded as n of 1 trial

TABLE 4c Data for Figure 3

Prior Trial Regimen/ Data Authors Initials HR credible interval HR In trial Not in trial treat- result non- quality median ment regimen Upper Lower Okay At risk Okay At risk treatment

a ii b 1 Davis 85 D 0.68 0.19 0.36 ‘RR’ (presumably HR) from Cox regression in paper a ii b 2 Ward 92 W 1.05 0.65 0.83 119 217 238 493 b ii a 2 Schmoor 96 Sc(ii) 1.10 0.48 0.73 143 191 86 128 b ii a 2 CASS 84 C 1.51 0.63 0.98 714 752 887 935 b ii b 2 Bertelson 91 B 1.77 0.66 1.05 132 264 19 37 b ii c 2 Schmoor 96 Sc(i) 1.34 0.74 0.99 308 440 155 222 b i d 1 Jha 96 J(i) 0.71 0.45 0.56 Reported OR for GUSTO patients versus others in GUSTO hospitals with myocardial infarction b i d 3 Stiller 94 S94(i) 0.95 0.62 0.77 54 244 41 291 b i d 3 Stiller 94 S94(ii) 1.25 0.67 0.92 51 125 60 158 b ii d 1 Jha 96 J(ii) 0.83 0.28 0.48 Reported OR for LATE patients versus others in LATE hospitals with myocardial infarction b ii d 1.5 Lennox 79 L(i) 1.75 0.33 0.79 37 46 39 52 b ii d 1.5 Lennox 79 L(ii) 0.81 0.10 0.31 19 23 16 32 b ii d 1.5 Lennox 79 L(iii) 0.71 0.14 0.33 19 29 5 20 b ii d 3 MRC 71 MRC 0.61 0.36 0.47 73 146 104 454 b ii d 3 Stiller 89 S89(iA) 0.72 0.44 0.56 215 422 43 144 b ii d 3 Stiller 89 S89(iB) 0.74 0.33 0.51 31 59 69 246 b ii d 3 Stiller 89 S89(iC) 0.93 0.25 0.52 7 16 40 201 b ii d 3 Stiller 89 S89(iiA) 1.17 0.74 0.92 172 331 132 269 b ii d 3 Stiller 89 S89(iiB) 1.05 0.56 0.77 104 208 50 124 b ii d 3 Stiller 89 S89(iiC) 1.07 0.41 0.66 29 55 29 77 b ii d 3 Stiller 89 S89(iiiA) 1.01 0.61 0.78 217 374 108 215 b ii d 3 Stiller 89 S89(iiiB) 0.91 0.45 0.63 119 199 43 97 b ii d 3 Stiller 89 S89(iiiC) 0.78 0.26 0.45 36 59 14 43 b ii d 3 Stiller 89 S89(ivA) 0.91 0.59 0.74 393 546 312 488 b ii d 3 Stiller 89 S89(ivB) 1.15 0.59 0.82 152 214 149 226 b ii d 3 Stiller 89 S89(ivC) 1.07 0.27 0.55 24 35 24 49 b ii d 3 Karjalainen 89 K 0.89 0.52 0.68 153 405 45 164 trial period 227 948 128 461 no-trial period

Trial area No-trial area 89 Tables

TABLE 5 Psychological effects of trial participation

Author Purpose and Outcome Main results Comments (see Table 6 – type of study measurement quality of evidence checklist)

(1) Cassileth Study to compare the Four psychological measures (1) Between trial protocols.There was no effect A– et al. (1986) psychological outcome of anxiety were used: on anxiety levels by clinical trial protocol (no B2 of participants in two (1) state anxiety trend given) C2 clinical trials (2) trait anxiety (2) Within trial protocols.There was a statistically D– (3) SCL-90 total score significant difference in anxiety according to E+ Two chemotherapy trials (4) SCL-90 anxiety score treatment allocation within the trials: patients F+. Mostly closed-ended of adjuvant therapy versus in the observation group displayed higher G–. 26% of patients entered on observation for early Patients were asked to anxiety levels at each time of assessment than to the breast protocols also breast cancer complete the state–trait those in the adjuvant therapy group (p < 0.05). consented to participate in the inventory and the SCL-90 However, when the prerandomisation score psychosocial study, of which 60% Women with stage 2 immediately before was omitted from the analysis, the result was completed all the tests breast cancer (n = 68) randomisation and then no longer statistically significant H+ at 3, 6 and 12 months I+. Low power following

(2) Mann Study to examine the Outcome measurement: Outcome measure (1): A+ (1977) psychological effect of a (1) Self-administered T1–T2 – there were no significant differences B2 screening programme questionnaire – Goldberg between trial entrants and controls C2 and clinical trial for General Health Question- T2–T3 – there was a greater number of improved D+ hypertension naire completed at: psychological scores among trial than control E+ T1 – screening group patients (p < 0.05) F+. Authors claim that the Trial participants: patients T2 – after trial entry T3–T4 – significant difference was sustained 1 year questionnaire was previously in MRC placebo-controlled T3 – 3 months after entry after trial entry. Furthermore, the prevalence of validated and piloted.The ques- trial for mild–moderate T4 – 1 year after entry psychological disturbance was significantly lower tionnaire was not a diagnostic hypertension (n = 235) (2) Goldberg Standard in the trial participants than in the controls after measure of psychiatric morbidity, Psychiatric Interview to 1 year (p < 0.05), though there was no significant so 85% of all respondents giving a Non-trial controls: two those judged from (1) to difference in the number of new cases positive score were interviewed groups were selected at be likely psychiatric by a psychiatrist random from participants cases at: Outcome measure (2): G–. Response rate varied over who took part in a T3 – 3 months T3 – trial participants had a significantly lower time, though overall, 1 year after screening programme but T4 – 1 year after prevalence (p < 0.05) than normal controls.The entry, 62% of those initially who did not enter the trial trial entry incidence of new cases however did not differ in involved responded Control group 1 (n = 233) the three groups H– Control group 2 (n = 231) T4 – the trend was the same, though it was not I+ statistically significant

(3) Robiner Study to assess the Standardised psychological T1 – trial participants were somewhat more A+ et al. (1993) psychological effect of measures, inc. BDI, Scale 2 distressed than non-trial controls on 80% of B2 participating in a clinical (Depression) of the Minnesota measures, but this was not statistically significant C2 trial Multiphasic Personality T1–T2 – trial participants reported more anxiety D– Inventory, Beck Hopelessness symptoms (2 weeks; p < 0.05) and had higher mean E+ Placebo-controlled Scale, Self Evaluation Question- scores than non-trial controls on all 10 measures F+ trial of zidovudine in naire (state anxiety), and (p < 0.002). Also, more trial participants than G+. Psychological assessments asymptomatic Stress Reaction and Well Being non-trial controls reported BDI levels of depression were completed by 78% of trial HIV-seropositive scales of the Multidimensional (p < 0.05), but the groups’ mean scores were not participants but only 38% of persons Personality Questionnaire different. More trial participants than non-trial non-trial controls Trial participants (n = 46) controls met GAD criteria (p = 0.03) H+ ineligible concurrent All outcome measures were T2–T3 – there were no longer differences between I+ non-trial controls administered at: trial participants and non-trial controls (n = 27) T1 – trial entry T3–T4 – trial participants were less depressed Larger sample size of trial T2 – 2 months and distressed than non-trial controls (p < 0.002) participants increased likelihood T3 – 6 months of finding their reduced distress T4 – after trial modification Both groups had statistically significant overall improvement in terms of depression (p < 0.01) and anxiety (p < 0.01).That said, non-trial controls differed in their relative patterns of distress over a 6 month period: there was an initial decrease in distress at 2 months but an increase by 6 months whereas, for trial participants, there was a consistent decrease in distress from entry to 6 months

There was no difference in distress based on whether trial participants thought they knew their group assignment or thought they were taking a placebo 90 Health Technology Assessment 1998; Vol. 2: No. 15

TABLE 6 Quality of evidence checklist for psychological effects

Key Type of evidence

A External validity, i.e. how representative is the sample of the target population (acceptable +, flawed –)?

B Sampling: (1) all patients offered trial entry or random sample (2) not stated or grab sampling

C Study design: (1) comparative study – RCT (2) comparative study – concurrent cohort (3) comparative study – historical cohort

D Recruitment rates to trials should be given and refusers followed-up (acceptable +, flawed –)

E Repeated measured before and after trial entry (acceptable +, flawed –)

F Outcome measure: interobserver reliability should be carried out where appropriate for questionnaires and/or interviews and psychometric validity should be given for scales (acceptable +, flawed –)

G Response rate to outcome measurement should be 70% or more (acceptable +, flawed –)

H Questionnaire should be supplied in study or available from authors (acceptable +, flawed –)

I The statistical method used to analyse data must be appropriate (acceptable +, flawed –)

91 Tables

TABLE 7a Methods of obtaining informed consent for real trials

Author Study design and Outcome measures Main results Comments sample (n = study (see Table 8 for quality participants) of evidence checklist) (1) Aaronson Patients were Anxiety Anxiety A2 et al. (1996) randomised to one Anxiety was measured by the There was no association between consent format and anxiety, B1 of two informed state version of the Spielberger though the control group had slightly higher measured C+ consent procedures: State–Trait Anxiety Inventory anxiety scores D+. 63% of patients asked (1) standard informed (STAI) presented 1 week after to enter the trials were consent procedure completion of the consent Understanding referred to this study, for based on oral and process Patients undergoing the second interview were significantly which there a was response written information (p < 0.01) better informed rate of 82% (2) standard informed Understanding E– consent plus a Recall was assessed by interview Respondents with correct answers F+ supplementary, (a series of semi-structured Second Control: p: telephone-based questions with follow-up interview: Interview was pilot tested interview with an probes) conducted I week (a) Concept of the clinical trial 97% 86% < 0.01 oncology nurse after completion of the (b) Trial objective 87% 71% < 0.01 consent process (c) Concept of randomisation 75% 54% < 0.01 Subpopulation (d) Right to withdraw 87% 65% < 0.01 (n = 180) offered Recruitment rates (e) Alternative treatments 86% 67% < 0.005 entry to Phase 2 Actual participation (f) Side-effects (average % recalled) 78% 61% < 0.001 or three trials at the Netherlands Recruitment Cancer Institute Patients in the second interview group were more likely to decline participation though this did not reach statistical significance (p = 0.17) Interactions Patients scoring high on overall understanding within the two groups were significantly less anxious (p = 0.001).Younger and better educated patients exhibited less anxiety than older and less educated patients See also Table 10 – audit (2) Davis Patients were Anxiety Anxiety A1 et al. (1990) randomised to one of Anxiety was assessed by Respondents receiving the National Cancer Institute’s booklet B1 two consent procedures: questionnaire (details not given) were significantly more likely to feel somewhat or very relieved by C–. Anxiety was not (1) oral and written receiving the consent information (p value not given) measured by a information Understanding psychometric scale (National Cancer Understanding was assessed by Understanding D–. Response rate not given Institute’s booklet) questionnaire (details not given) The experimental group was significantly more likely to feel that though 78% of those (n = 203) they had understood their clinical trial and were less likely to receiving the National (2) control – oral Recruitment rates feel ‘confused’. Cancer Insitute’s booklet information only Self-reported participation and Control: Booklet: p: actually read it. (n = 194) data on actual participation in (a) Voluntary nature 83% 92% < 0.03 E– trials taken from medical (b) Side-effects 86% 92% 0.06 F.Insufficient information Eligible and participating records (c) Concept of randomisation 49% 40% 0.05 patients in cancer clinical trials at four Recruitment centres (n = 187 and There was no statistically significant difference between the 210, respectively) groups in terms of clinical trial participation (no trend given)

See also Table 10 – audit (3) Levene The same parents were Recruitment When approached for consent early, the parents were more A1 et al. (1996) offered entry for their Actual participation willing for their baby to undergo randomisation than when B2 new-born babies into they were approached later (p = 0.013) C. Not applicable two trials requiring D. Not applicable different informed Consent rates E– consent procedures: (a) Early consent: 71% F+ (1) trial A – consent (b) Late consent: 43% obtained 2 h after eligibility established (n = 42) (2) trial B – consent obtained 7–14 days after eligibility established at birth (n = 37)

Parents of babies eligible for two separate but concurrent clinical trials; one of blood volume support and one of erythropoeitin (n = 79) continued 92 Health Technology Assessment 1998; Vol. 2: No. 15

TABLE 7a contd Methods of obtaining informed consent for real trials

Author Study design and Outcome measures Main results Comments sample (n = study (see Table 8 for quality participants) of evidence checklist)

(4) McLean The first 22 respondents Recruitment There was no statistically significant difference between interviews A2 (1980) were allocated to the Willingness to participate in terms of willingness to participate according to category, with B3 first interview format (recruitment) was rated by the no refusals in either group C– and then the remainder interviewing clinician according Interview 1: Interview 2: D+. 100% response rate (n = 104) to an altern- to four categories: (1) consent with no questioning 81% 86% E– ative interview: (1) free consent with no questions (2) consent after mild questioning 17% 14% F.Not applicable (1) written and verbal (2) consent after mild questioning (3) consent after persuasion 2% 0% information with (3) consent after persuasion (4) refusal 0% 0% no mention of (4) refusal randomisation (2) written and verbal information including an explanation of randomisation

Clinically depressed patients offered entry to a clinical trial of four treatments (n = 126)

(5) Myers Patients were randomised Experience of side-effects There were no statistically significant differences between the A1 et al. (1987) to one of two consent Minor gastrointestinal side-effects two group in the occurrence of major side-effects as documented B2 scenarios: experienced by the patients was by the study physicians C+. 100% response rate (1) information about assessed by a questionnaire, while D. Insufficient information gastrointestinal side- major complications were However, there was a statistically significant difference between E+. Consent information effects (n = 399) diagnosed by the study physicians reporting of minor gastrointestinal side-effects by the patients given but no questionnaire (2) no information about (p < 0.001) F. Post hoc statistical analysis side-effects (n = 156) Reporting rates Participants in a placebo- Group 1: 44% controlled trial for Group 2: 16% unstable angina (n = 555) These minor side-effects were not confirmed independently by study nurse or physician

(6) Simes Respondents randomised Anxiety Anxiety A2 et al. (1986) to one of two consent Anxiety assessed by questionnaire Patients in total disclosure group were also more anxious initially B1 methods: completed before receiving (p = 0.02) but later showed that this significant effect was transient C–. Anxiety was (1) individual approach treatment and again 3–4 weeks no measured by a (2) total disclosure later with responses recorded on Understanding psychometric scale a five-point scale Patients in the total disclosure group were more knowledgeable D+. 87% response rate Subpopulation (n = 57) of about their illness and treatment (p = 0.0001), and also about E+. Questionnaire not eligible cancer patients Understanding research aspects of the trial (p = 0.03). In particular, the concept supplied but available for any one of 16 trials Understanding assessed by of randomisation was understood better by the total disclosure from authors and consent questionnaire to assess recall of group (p = 0.004) information given. 54% diagnosis, treatment, possible side- of topics were in fact effects, and research aspects Recruitment mentioned to the individual Patients in total disclosure group were less amenable to receiving approach group as compared Recruitment randomised treatment (p = 0.01) with 86% of topics for the Actual participation total disclosure group See also Table 10 – audit F+

Questionnaire pilot tested

(7) Tindall Respondents were Understanding No statistically significant difference between groups in pre- A2 et al. (1994) randomly allocated to Understanding assessed by an consent or post-consent knowledge scores between the two B1 receive one of two eight-item instrument presented groups (p = 0.222) C– consent formats: prior to and subsequent to D–. Response rate not given (1) written only (n = 52) receiving consent information There was a statistically significant increase in scores for both E+ (2) written and oral groups when baseline scored were compared with those obtained F– (n = 61) after the provision of information (p = 0.0003), and a weak interaction between time and format, indicating that those AIDS patients in to a provided with written and verbal information increased their dose–control trial of knowledge score more over time (p = 0.079) didanosine (n = 113) Mean knowledge scores Baseline: After information: (a) written only 4.4 4.8 (b) written and oral 4.0 5.0

See also Table 10 – audit, and Table 12a – views of patients 93 Tables

TABLE 7b Methods of obtaining informed consent in hypothetical trial scenarios

Author Study design and Outcome measures Main results Comments sample (n = study (see Table 8 for quality participants) of evidence checklist)

(1) Dal-re Students were assigned Recruitment rates There was a significant difference between the two consent A1 (1991) to one of two consent Anonymous indication on groups according to consent rates (p < 0.05) B2 forms: consent document C. Not applicable (1) standard + minimal Consent rates: D–. Response rate not given information on (1) form 1: 11% E+ side-effects (2) form 2: 4% F+ (2) Standard + detailed information on Among the women, only students of medicine agreed to side-effects participate, with 8 and 12% of women consenting in group 1 and group 2, respectively Student volunteers in a hypothetical Phase 1 trial of isoniazide

(2) Epstein Respondents were Anxiety Anxiety A1 and Lasagna randomised to one of Anxiety was assessed by Respondents who thought the information was frightening or not useful: B1 (1969) three informational forms: structured questionnaire (a) short: 0% C–. Anxiety was (1) short (b) medium: 23% not measured by a (2) medium Understanding (c) long: 41% psychometric scale (3) long Recall of information was (no statistical analysis given) D+. 91% response rate The forms differed only assessed by structured E+ in the amount of detail questionnaire Understanding F–. Should not have used about the risks and Statistically significant differences were found between short and multiple t test though this benefits of aspirin, in Recruitment rates long forms in favour of shorter forms (p < 0.001) probably made little addition to baseline Willingness to participate difference to the result information was assess by structured Mean understanding scores: questionnaire (a) short: 67% Oral explanations of the Hospital employees with (b) medium: 45% research had already been no medical qualification (c) long: 35% given to the respondents (n = 66) offered entry to a hypothetical placebo- Recruitment controlled trial of acetyl- Consent rates: hydroxbenzoate (fictitious (a) short form: 86% name for aspirin) (b) medium form: 64% (c) long form: 55%

Interactions Higher scores for understanding we significantly associated with greater willingness to participate. See also Table 12b – views of the public

(3) Fetting Patients were randomised Understanding Understanding A1 et al. (1990) in the proportion 3:1, to Understanding was assessed by The mean estimate of percentage chance of living 10 years DFS B1 one of two versions of a patients’ estimate of 10 year with standard therapy was significantly higher (and accurate) in the C– clinical vignette: DFS with standard treatment numerical group than in the descriptive group (p = 0.04) D+. 79% response rate (1) descriptive E+. Questionnaire and (2) numerical Recruitment rates Mean estimate of percentage chance of 10 year DFS: vignettes available from The vignettes differed Willingness to participate in the (a) descriptive group: 55% authors only on how results of trial rather than to choose their (b) numerical group: 61% F+ standard therapy were own treatment was assessed by described in terms of structured interview Recruitment disease-free survival (DFS) A statistically significant difference was found between the two groups in terms of consent rate (p = 0.01) Female patients (n = 282) offered entry to a hypo- Consent rates: thetical clinical trial of (1) descriptive vignette: 52% chemotherapy for cancer (2) numerical vignette: 35%

Interaction Patients in the descriptive group giving pessimistic or conservative estimates of DFS were significantly more likely to consent to trial entry (p = 0.005)

Consent rate of descriptive group according to DFS estimate: (1) pessimistic estimate: 59% (2) conservative estimate: 59% (3) optimistic estimate: 35%

continued 94 Health Technology Assessment 1998; Vol. 2: No. 15

TABLE 7b contd Methods of obtaining informed consent in hypothetical trial scenarios

Author Study design and Outcome measures Main results Comments sample (n = study (see Table 8 for quality participants) of evidence checklist)

(4) Gallo Participants were Understanding Understanding A1 et al. (1995) randomised to one of Understanding was assessed 38.2% of respondents correctly understood individual equipoise B1 four consent scenarios: by interview which covered 51.7% of respondents thought that the new treatment was better C– (1) pre-randomisation perceptions of severity than the standard one, despite being told that equipoise applied D–. Response rate not given consent with refusers of disease and of relative Misunderstanding rates were independent of the consent E– getting standard efficacy of standard and procedures (data not given) F+ treatment experimental treatment (2) pre-randomisation Recruitment consent with refusers Recruitment rates Consent rates for each consent scenario: choosing their treatment Willingness to participate (1) consent scenario 1: 83.8% (3) consent to experimental was assessed by structured (2) consent scenario 2: 80.1% interview treatment (3) consent scenario 3: 87.9% with refusers getting (4) consent scenario 4: 50.8% standard treatment (no statistical analysis given) (no knowledge of randomisation) Interaction (4) consent to standard The consent rate was greatest when the patients perceived the treatment with refusers efficacy of the experimental treatment to be greater than that of getting experimental the standard treatment and least when the patients perceived the treatment (no knowledge efficacy of the experimental treatment to be less than that of the of randomisation) standard treatment

Quantitative information Consent rate amongst those who understood that equipoise applied: about disease severity (1) consent scenario 1: 81.1% given to respondents in the (2) consent scenario 2: 81.5% form of three levels of (3) consent scenario 3: 86.8% 5 years expected survival (4) consent scenario 4: 41.6% (20, 50 or 80%) (no statistical analysis given)

General public offered entry to a hypothetical trial to evaluate the Zelen consent design (n = 2035)

(5) Llewellyn- Patients were randomised Recruitment There was no statistically significant difference between groups in A1 Thomas to one of three consent Willingness to undergo terms of consent rate B1 et al. (1995) formats: randomisation rather than C– (1) neutrally framed prob- to choose their treatment Consent rates: D+. 36% of the eligible abilistic information was assessed by interview (1) neutral: 76% population was approached, (2) positively framed prob- (2) positive: 60% of which 94% agreed to abilistic information (3) negative: 66% participate in the study (3) negatively framed prob- E+ abilistic information F+

Cancer patients with resectable adenocarcinoma offered entry to a hypothetical clinical trial of different chemotherapy treatments (n = 90)

(6) Simel Patients randomised to Recruitment Patients randomised to form 1 were significantly more likely to A1 and Feussner receive one of two Willingness to participate consent than those randomised to form 2 (p < 0.01) B1 (1991) consent forms: was assessed by a tape- C+. Independent attorney (1) randomised trial of recorded interview with Consent rate: reviewed random sample usual treatment versus opportunity to ask (1) form 1: 67% of taped interviews, and new medication that questions (2) form 2: 42% there was agreement over ‘may work 2x as fast as citation of quantitative usual treatment’ (n = 52) The respondents who cited quantitative information on their information (2) trial of new treatment informed consent form as a reason for either consenting or declining D– that ‘may work 1/2x as trial entry (62%) were no more likely to agree to participate than E+ fast as the usual treat- those who did not cite quantitative information as a basis for ment’ (n = 48) their decision

Adult ambulatory patients See also Table 12b – views of patients in a hypothetical clinical trial of standard versus new medication for an unspeci- fied disease (n = 100)

continued 95 Tables

TABLE 7b contd Methods of obtaining informed consent in hypothetical trial scenarios

Author Study design and Outcome measures Main results Comments sample (n = study (see Table 8 for quality participants) of evidence checklist)

(7) White Respondents were Understanding There was no significant difference between form length and patient A1 et al. (1984) randomised to one of three Attitudes and understanding perception of helpfulness in understanding therapy.The trend was B1 model consent forms: were assessed by structured in favour of better understanding in the longer form C+ (1) long – explained questionnaire D+. 99% response rate randomisation and Percentage of respondents who were aware that treatment E+ rationale for trial allocation was by chance: F+ (2) medium – described the (a) short: 76% study as ‘research’ and (b) medium: 84% did not mention (c) long: 96% randomisation (3) short – indicated that Percentage of respondents who were unaware that the best treatment was selected chemotherapy regimen was unknown: by physician (1) short: 84% (2) medium: 54% All forms contained identical (3) long: 64% information about side-effects and drugs See also Table 12b – views of patients

Breast cancer patients in a hypothetical randomised controlled trial (n = 75)

TABLE 8 Quality of evidence checklist for comparative studies of informed consent

Code Dimension of methodology

A Sampling hierarchy: (1) all patients offered entry in a trial or random sample of all patients (2) not stated or convenience sampling

B Controls: (1) comparative study: RCT (2) comparative study: concurrent cohort (3) comparative study: historical cohort

C Outcome measurement: interobserver reliability should be assessed where appropriate for questionnaires and/or interviews (particularly where open-ended questionnaire are used) (acceptable +, flawed –)

D Response rate to outcome measure must be given, and is acceptable at 70% or above (acceptable +, flawed –)

E Actual information given at consent, and questionnaire should be supplied in study (acceptable +, flawed –)

F The statistical method used to analyse data must be appropriate (acceptable +, flawed –)

TABLE 9 Quality of evidence checklist for audit

Code Dimension of methodology

A External validity of study, i.e. how clearly defined is the target population to which the sample results may in theory be generalised (acceptable +, flawed –)?

B Sampling: how representative is the sample of the target population? (1) Entire population approached or random selection of eligible patients (2) Quota sampling, i.e. deliberate selection from specific groups (3) Grab sampling

C Controls: (1) RCT (2) non-randomised concurrent cohort (3) descriptive study

D Outcome measurement: Interobserver reliability should be assessed where appropriate for questionnaires and/or interviews (particularly where open-ended questionnaire are used) (acceptable +, flawed –)

E Response rate: outcome measure must be given and be acceptable at 70% or above (acceptable +, flawed –)

F Original consent information and questionnaire should be supplied in study or available from authors (acceptable +, flawed –)

96 G The statistical method used to analyse data must be appropriate (acceptable +, flawed –) Health Technology Assessment 1998; Vol. 2: No. 15

TABLE 10 Audit of the quality of communication

Author Population assessed Outcome measurement Main results Comments and sample size and purpose of study (see Table 9 for quality (n = No. of patients of evidence checklist) in survey)

(1) Aaronson Eligible cancer patients Patients were randomised to Overall No. of respondents who had knowledge of the following:* A– et al. (1996) for Phase 2 or 3 trials one of two informed consent (1) the clinical trial context of the treatment (in general): 92% B3 at the Netherlands procedures: (2) the objectives of the clinical trial: 79% C1 Cancer Institute (1) standard informed consent (3) the use of randomisation in allocating treatment: 65% D+ (n = 30) procedure based on verbal (4) the availability of alternative treatments: 77% E+. 100% response rate and written information (5) the voluntary nature of participation: 89% (73 and 76% of patients from physician (6) the right to withdraw from the clinical trial: 77% from groups 1 and 2 were (2) standard informed consent interviewed, respectively) plus a supplementary, On average, group 2 was aware of 77.5% of the potential side- F– telephone-based contact effects as compared with 60.7% for group 1 (p < 0.01) G+ with an oncology nurse See also Table 7a – methods of informed consent Interview was pilot tested Knowledge and understanding Recruitment rates to trials was assessed during an interview not given (a series of semi-structured questions with follow-up probes based on recall of information) 1 week after completion of the consent process

Study to assess the effect of different consent procedures on comprehension of patients offered entry to a clinical trial

(2) Bergler Participants in a Interview with multiple-choice Percentage of patients answering correctly at: A– et al. (1980) controlled trial of quiz to assess knowledge of trial 2 h: 3 months: B3 hydrochlorothiazide at 2 h and 3 months following (1) Action of hydrochlorothiazide 92% 82% C3 versus propranolol enrolment (2) Action of propranolol 77% 38% D+ (n = 39) (3) Side-effects 28% 4% E+. 100 and 64% response Study to assess comprehension (4) Duration of trial 64% 65% rate at 2 h and 3 months of participants in a clinical trial (5) Freedom to withdraw 77% 61% after enrolment (6) Expect to receive best F+ possible treatment 95% 100% G+ (7) Meaning of double blind 64% 46% Recruitment rate to trial not given

(3) Daughterty Participating patients in Structured interview: open- 93% reported that they understood all or most of the information A– et al. (1995) eight different Phase 1 and closed-ended questions to about the trials in which they had decided to participate, though B3 cancer trials (n = 27) determine whether respondents only 33% were able to state the actual scientific purpose of the C3 thought that they had understood Phase 1 trial D+. Most questions the material and had been told closed-ended about certain aspects of the trial The patients able to state the purpose of Phase 1 trials were E+. 90% response rate significantly more educated (p = 0.01) F+. Questionnaire supplied Study to assess comprehension but not original consent of participants in a Phase 1 See also Table 12a – views of patients, and Table 13 – views of information clinical trial healthcare professionals Recruitment rates to trials not given

(4) Davis Eligible patients for a Patients were randomly allocated Correct responses: A– et al. (1990) cancer clinical trial to receive experimental consent (1) Right to withdraw at any time 87% B3 (n = 397) procedure (National Cancer (2) Right to withdraw before the trial is over 54% C1 Institute booklet) or to control (3) Standard treatments may be as good as D– Insufficient information procedure (not receive booklet experimental treatments 53% E– Response rate not given until after understanding test (4) The treatments may cause side-effects 89% F– 2 weeks later). Method for (5) Randomisation means that treatment will be G. Insufficient information assessing comprehension chosen by chance 45% not given Recruitment rate to trial See also Table 7a – methods of informed consent not given Study to assess the effect of different consent procedures on comprehension of patients offered entry to a clinical trial

* Level of understanding for all respondents was calculated from the raw data given in the study.

continued 97 Tables

TABLE 10 contd Audit of the quality of communication

Author Population assessed Outcome measurement Main results Comments and sample size and purpose of study (see Table 9 for quality (n = No. of patients of evidence checklist) in survey) (5) DCCT Patients eligible for the Self-administered multiple- 80% of respondents obtained a perfect overall score A– Research Diabetes Control and choicequestionnaire presented B3 Group Complications Trial immediately prior to Knowledge pre- and post-randomisation respectively: C3 (1989) (n = 278) randomisation and 1 year (1) 97 and 85% understood the scientific background (p < 0.05) D+ after entry (2) 96 and 95% knew the purpose of the study E+. 100% for first test and Recruitment rate to trial (3) 100 and 99% understood random assignment 79% for test 1 year later = 79% Study to assess the knowledge (4) 98 and 88% knew standard treatment risks (p < 0.05) F+. Questionnaire supplied of participants in a clinical trial (5) 94 and 85% knew experimental treatment risks (p < 0.05) but not original consent information G+ (6) DeLuca Patients eligible for Questionnaire using 30 variables, Level of understanding after patients received verbal and A– et al. (1995) participation in one of 12 one of which assessed level of written information:* B3 randomised cardiovascular understanding of the study on (1) 5% had low understanding – level 1 C3 trials (n = 247) a five-point scale (2) 3% had understanding – level 2 D– (3) 11% had moderate understanding – level 3 E–. Response rate not given Recruitment rate to trial Study to assess comprehension (4) 23% had understanding – level 4 F– = 70% of patients offered entry to a (5) 58% had complete understanding – level 5 G+ clinical trial 32% of potential participants reported not reading the consent Errors in how p value was form, despite being given the opportunity reported and some strange percentages in some tables See also Table 12a – views of patients (7) Gallet Multicentre Randomised Questionnaire administered Global quantity of information recalled was estimated as being 60% A– et al. (1994) trial Mexiletine and 5–21 months after consent had 40% of respondents correctly recalled the aim of the study and B3 Placebo Antiarrhythmic been secured. Comprehension the concept of a placebo C3 Coronary Trial (IMPACT) assessed by recall 55 and 43% correctly recalled the nature of the active treatment, D– and of randomisation, respectively E+ response rate 84% Patients with a recent Study to assess comprehension 46% correctly recalled possible side-effects F– myocardial infarction of participants in a clinical trial Right to withdraw and the nature of a double-blind study were G+ (n = 77) understood by 81 and 62%, respectively Recruitment rates to trial not given (8) Harth and Parents of children Structured questionnaire was While most parents (98.8%) were clear about the aim of A– Thong (1995) participating in a presented 6–9 months after the determining drug efficacy, only a small percentage (12.9%) were B3 randomised, placebo- children had been entered into aware that the trial was designed to assess safety as well. 45.2% C3 controlled trial of the trial of respondents knew that they had the unconditional right to D–. Insufficient data ketotifen, an oral asthma withdraw their children from the trial E+. Of the 88% of families drug (n = 62) Study to assess comprehension that were available after of parents and their participating See also Table 12a – views of patients the trial, 95% agreed to children in a clinical trial participate in the survey F– G+

Recruitment rate to trial not given (9) Harrison Eligible volunteers for 17-item true/false questionnaire Accurate responses: A– et al. (1995) a placebo-controlled was presented after disclosure IDUs: Others: B2 Phase 2 trial of two HIV of information but before the (1) Trial vaccines have been used in C3 vaccines: signing of the consent form people before 85% 91% D– (1) injection drug users (2) The vaccines cannot cause HIV and AIDS 100% 100% E+ .96% response rate (IDUs) (n = 39) Study to assess comprehension (3) Not everyone in the trial will get a vaccine 74% 84% F– (2) others (n = 32) of patients offered entry to a (4) People who participate in the trial might G+ Phase 2 HIV vaccine trial get HIV anyway 85% 94% Recruitment rate to trial (5) The vaccines might cause side-effects = 100% including aches and a fever 92% 84% (10) Hassar Participants in a double- Interview and questionnaire 88% of trial participants were aware that their physician did not A– and Weinstein blind clinical trial of an presented at the end of the trial know which medication they had been allocated B3 (1976) anti-inflammatory drug (16 weeks) to elicit trial 65% did not remember being told that there was a risk of C3 (n = 49) participants’ understanding of peptic ulceration D– the information sheet given at 65% of respondents thought correctly that the experimental E+. 94% response rate Recruitment rate to trial consent drug was not safer than other drugs for arthritis F+ = 37% All participants knew that their physician would not be upset if G+ Study to assess comprehension they refused trial entry of participants in a clinical trial See also Table 12a – views of patients continued 98 Health Technology Assessment 1998; Vol. 2: No. 15

TABLE 10 contd Audit of the quality of communication

Author Population assessed Outcome measurement Main results Comments and sample size and purpose of study (see Table 9 for quality (n = No. of patients of evidence checklist) in survey)

(11) Howard Beta-blocker Heart In-depth home interviews 91% of respondents were clearly aware that the BHAT study was A+ et al. (1981) Attack Trial (BHAT) in person or over the a research enterprise, while 83% recognised that some participants B1 Participants from 11 telephone to assess took the experimental drug and others took a placebo. 67% of C3 geographical areas recalled information respondents gave a completely accurate description of the D– (n = 64) purpose of the trial, and 86% expressed an awareness of the E+. 98% response rate Study to assess compre- possibility of side-effects. Only 42% of the respondents knew that F– Recruitment rate to hension of participants the allocation of treatment was based on chance, while 77% clearly G+ trial = 53% in a clinical trial understood that the assignments were concealed from physicians as well as patients Possibility that information about side-effects acquired from other sources other than the consent process

(12) Jensen Eligible women for three Structured interview with Patients were generally able to remember the given information A+ et al. (1993) Danish breast cancer 36 questions which was 3 months after it had been received. Patients’ memory was rated B3 co-operative group trials tape-recorded. Recalled ‘good’ for the randomisation aspect of the project, the available C3 dealing with adjuvant information was measured treatment alternatives, and the side-effects in 73, 65 and 76% of D+. Interview conducted treatment of primary on a four-point scale: the patients, respectively and cross-rated by two breast cancer (n = 34) good, reasonable, researchers who had not questionable and bad See also Table 12a – views of patients participated in the consent procedure Study to assess E–. Response rate not comprehension of given patients offered entry F– in a clinical trial G+

Recruitment rates to trials not given

(13) Lynoe Patients who had Postal 10 closed-ended 98% of respondents were aware that they had participated in A– et al. (1991) participated in a multi- item questionnaire assess- a research project B3 centre gynaecological ing recall, 18 months after 11% stated that they had not been aware that a second C3 clinical trial for acute trial completion laparoscopy was performed only for research reasons D–.There was a possibility inflammation of the 40% reported that had no information about the possibility of that respondents were ‘fed’ Fallopian tube (n = 43) Study to assess withdrawing from the study whenever they wanted answers comprehension of 42% of respondents reported having weighed the pros and cons E+. Response rate 81% participants in a before consenting F– clinical trial 56% reported not having weighed the pros and cons before G+ consenting Women who weighed the pros and cons before consenting were Recruitment rate to trial significantly older (p < 0.05) not given

See also Table 12a – views of patients

(14) Marini Participating convicted Open-ended and closed- 97% of respondents knew that lithium was the drug being tested A– et al. (1976) violent men in a placebo- endedquestionnaire 82% of respondents knew that a placebo was also a treatment B3 controlled trial of lithium presented at end of trial option in the trial C3 carbonate (n = 44) (5 months after consent) 71% of respondents knew that they were part of an experiment D– 47% of respondents indicated the importance of aggressive E+. 76% response rate Study to assess behaviour in the experiment F+. Consent information comprehension of 65% of respondents gave at least one correct side-effect, while given but questionnaire participants in a only 19% could give three or more not supplied clinical trial G. Not applicable

Recruitment rate to trial not given

continued 99 Tables

TABLE 10 contd Audit of the quality of communication

Author Population assessed Outcome measurement Main results Comments and sample size and purpose of study (see Table 9 for quality (n = No. of patients of evidence checklist) in survey)

(15) Maslin There were three Two self-completed postal Healthy volunteers in a clinical trial were found to be better A– (1994) sample groups: questionnaires: informed than corresponding cancer patients: B1. Random sample of three (1) healthy volunteers (1) participant satisfaction (1) they were given more information: 86 versus 60% (p < 0.005) different patient groups not in a trial, i.e. a with informed consent (2) better indication of their time commitment: 81 versus 62% C2 hypothetical decision process – 21 closed- (p < 0.025) D– scenario ended questions with (3) better informed about possible side-effects: 75 versus 57% E+ 62%, 73% and 76% (2) healthy volunteers room for comment (p < 0.025) response rate participating in a (2) patient/participant (4) more aware of their rights to withdraw from the trial: F+ tamoxifen prevention opinion on the informed 84 versus 46% (p < 0.005) G+ trial consent process – (3) breast cancer patients 16 closed-ended Recruitment rates to trial in a clodronate trial questions with room not given and in a chemotherapy for comment versus primary con- Respondents were asked Authors’ comments: servative trial whether they had been given (1) many of the asympto- (n = 213) the appropriate information matic patients were self- referred and were, as a Study to compare the com- result, strongly motivated prehension of participants and wanted to be in a clinical trial with that monitored of healthy volunteers (2) many of the asymptomatic patients may have been familiar with the trial due to general press coverage

(16) Miller Random selection (25%) Telephone interview using a (1) Recall and recognition tasks: A+ et al. (1994) of participating patients in questionnaire for open-ended > 98% of respondents recalled participating in the trial B1 a double-blind, controlled recall, cued recognition and 73% were able to accurately state the purpose of the trial C3 trial of ibuprofen versus perception of their participation 29% could not recall the study medications, 17% could D– ketoprofen for over-the- correctly recognise both study drugs from a list while 18% E+. 84% response rate counter analgesic use Study to assess comprehension could not recognise either drug F+ (n = 168) of participants in a clinical trial No respondent could correctly recall all eight indications for G+ administration of the medication 52% were unable to recall any of the 12 possible side-effects, Recruitment rate to trial while only 5% were able to recognise six or more when not given given a list (2) Perception of understanding: 98% of respondents thought that they had understood the information given

(17) Oddens Random selection (32%) Telephone interview using a 88% of respondents knew that their trial was scientific research A+ et al. (1992) of cardiovascular patients structured questionnaire 92% of respondents were aware that the decision to participate B1 (n = 81) eligible for one was voluntary nature C3 of two trials: 63% of respondents knew that they could withdraw if they so D– (1) comparison of two wished E+. 17% could not participate doses of acetylsalicylic 79% of those offered entry in trial one knew that their doctor in the telephone survey. Of acid did not knew the allocated dose those that could participate, (2) atenolol versus there was a response rate placebo of 98% F+ G+

(18) Olver Cancer patients (n = 100) Interview with a nurse using 60% respondents claimed to understand the consent form fully A+ et al. (1995) who had signed a consent a structured questionnaire, 40% of respondents thought that the main purpose of the B3 form for one of 18 differ- presented 6–43 days after informed consent procedure was to explain the trial C3 ent chemotherapy trials trial entry 52% of respondents could name all the drugs in their D–. Insufficient information chemotherapy regimen E+. 95% response rate 37% of the respondents could remember at least half the number F– of side-effects, and only 4% could remember all of them G. Not applicable

Only 80% of respondents reported having read the consent form BMDP software was used to analyse the data

continued 100 Health Technology Assessment 1998; Vol. 2: No. 15

TABLE 10 contd Audit of the quality of communication

Author Population assessed Outcome measurement Main results Comments and sample size and purpose of study (see Table 9 for quality (n = No. of patients of evidence checklist) in survey)

(19) Penman Participating patients Structured interview with The point best recognised was the possibility of treatment-related A– et al. (1984) from three cancer computer coding of quanti- side-effects, which virtually all patients acknowledged B3 centres in 65 different tative recall responses Actual recall of specific side-effects was poor.The average consent C3 Phase 2 or 3 Patients asked whether they form cited 11 potential side-effects, and patients recalled correctly D–. Insufficient information investigative chemo- believed that they had been an average of three. 69% of respondents could list no more E–. Response rate not given therapy trials, of which informed about the five major than three F– 47% were randomised points conveyed in obtaining G. Not applicable studies (n = 121) consent and whether each > 70% of respondents were aware of having been informed of the Recruitment rates to trials point was covered by physician research nature of their treatment not given at interview and/or consent 69% claimed that the informed consent form played no role in their form decision-making

Study to assess comprehension See also Table 12a – views of patients, and Table 13 – views of of participants in a clinical trial healthcare professionals

(20) Postle- Participating children with Semi-structured interview Parental understanding: A– waite et al. chronic renal failure and shortly after recruitment to 80% of parents had good or very good understanding of the B3 (1995) their parents in a failed assess recall of information treatment trial C3 trial of growth hormone Participants’ knowledge was Among 20% whose understanding was poor, ignorance and D– Parents (n = 30) compared with information on uncertainty were mainly about side-effects and expectations of E–. Response rate not given Children (n = 14) a checklist.Also, respondents outcome rather than details of treatment, and 50% of this group F–. Information checklist were asked to indicate the did not speak English as a first language supplied but not interview ease with which they made questions their decision on a three- Childrens’ understanding: G–. Use a correlation point scale 36% of children were unable to understand or recall in any detail coefficient to look for the information given about the trial a group difference Study to assess No correlation was found between age and understanding comprehension of No significant association between the level of children’s and level Recruitment rates to trial participants in a of parent’s understanding about the trial not given clinical trial There was no association between the ease of parental decision- making and the level of understanding about the trial

(21) Rodebhuis Patients eligible for a Semi-structured interview 80% of those interviewed were adequately informed.They recalled A+ et al. (1984) Phase 1 trial of a new conducted by trained observer relevant aspects of the Phase 1 study though 50% of these patients B3 antitumor agent, SOAz to assess recalled information did not achieve complete understanding C3 (n = 10) D– Study to assess comprehension See also Table 12a – views of patients E+. Only 21% of patients Recruitment rate to trial of patients offered entry to a approached were = 85% Phase 1 clinical trial interviewed but all were willing to participate F– G. Not applicable

(22) Simes Cancer patients eligible Respondents randomly Level of understanding for all respondents* A– et al. (1986) for any one of 16 trials allocated to one of two 68% of respondents were knowledgeable about treatment and B3 (n = 55) consent types: side-effects C1 (1) individual approach 66% of respondents were knowledgeable about research aspects D–. Insufficient information Recruitment rates to (2) total disclosure of their treatment E+. 96% response rate trials = 88% Questionnaire completed Only 24% of respondents understood that their treatment would F+. Questionnaire not before receiving treatment be allocated at random supplied but available from and again 3–4 weeks later authors and consent with responses recorded on See also Table 7a – methods of informed consent information given a five-point scale Two measures of patients’ Questionnaire pilot tested knowledge were calculated Recruitment rate to trial by recall of: not given (1) diagnosis, treatment, and possible side-effects (2) research aspects

Study to assess the effect of different consent procedures on comprehension of patients offered entry to a clinical trial

* Level of understanding for all respondents was calculated from the raw data given in the study.

continued 101 Tables

TABLE 10 contd Audit of the quality of communication

Author Population assessed Outcome measurement Main results Comments and sample size and purpose of study (see Table 9 for quality (n = No. of patients of evidence checklist) in survey)

(23) Toma- Eligible patients for Informed consent procedure Quantitative analysis: information concerning the characteristics A– michel et al. Phase 1 oncology study was three tiered: of a Phase 1 drug (investigational drug used on only a few humans) B3 (1995) (n = 44 for second (1) initial introduction was given to 78% of patients, information concerning patients’ C3 interview) (2) detailed conversation right to refuse or to withdraw was given to 34% of patients, details D+.73% were quantitatively (3) clarification if needed and of side-effects were given to 81% of patients, and 91% of patients analysed signing of consent form were informed that there was a lack of known treatments of Interobserver variability of The second session was taped, proven efficacy the scoring was studied in transcribed and analysed quan- the first six consultations by titatively along three dimensions comparing the evaluations of informed consent (inform- of two psychiatrists and ation – 12 items, emotion – three psychologists 5 items, and interactive E. 39% of interviews were dimension – 11 items), and taped and (selected by consisted of calculating the logistical considerations) No. of patients to whom the but actual response rate essential items of information not given had been conveyed F+ G+ Study to assess the quantity and quality of information actually Recruitment rate to trial given to patients in a Phase 1 not given clinical trial

(24) Tindall Eligible AIDS patients for Respondents received either 44% of subjects stated that they did not understand all of A– et al. (1994) a dose-controlled trial of written only consent format information provided, despite having signed the consent form B3 didanosine (n = 113) or written and verbal according No significant difference in pre-consent knowledge scores for C1 to random assignment both groups, or post-consent-knowledge scores D– An eight-item instrument There was a significant increase in knowledge after consent for E–. Response rate not given assessed knowledge of ddI both groups (p = 0.0003) F+ prior to and subsequent to G– receiving information. See also Table 7a – methods of informed consent, and Table 12a – Respondents were asked views of patients Recruitment rate to trial whether they thought that not given they had understood the information provided

Study to assess the effect of different consent procedures on comprehension of patients offered entry in a clinical trial

TABLE 11 Quality of evidence checklist for attitude studies

Code Dimension of methodology

A External validity of study, i.e. how clearly defined was the target population which the sample results may in theory be generalised (acceptable +, flawed –)?

B Sampling: how representative is the sample of the target population? (1) Entire population (consenters and refusers) approached or random selection of eligible patients (2) ‘Quota’ sampling, i.e. deliberate selection from specific groups (3) Grab sampling

C Response rate must be given and acceptable at 70% or above (acceptable +, flawed –)

D Interobserver reliability should be assessed where appropriate for questionnaires and/or interviews (particularly where open-ended questionnaire are used) (acceptable +, flawed –)

E Questionnaire should be supplied in study or available from authors (acceptable +, flawed –)

F The statistical method used to analyse data must be appropriate (acceptable +, flawed –)

102 Health Technology Assessment 1998; Vol. 2: No. 15

TABLE 12a Views of the patients and public in real trial scenarios*

Author Purpose of study and Method of assessment Main results Comments sample population (see quality of evidence (n = No. of respondents checklist – Table 11) to survey)

(1) Barofsky Study to identify the Standardised semi- 54% of the participants reported that they consented and Descriptive study and Sugarbaker motivations of patients to structured interview, complied with the treatment protocol for reasons of self-interest (1979) participate in trials sometimes conducted Patients who received an amputation justified it almost exclusively A+ over the telephone in terms of self-interest, whereas patients who received limb sparing B2 Consenting and refusing therapy justified participating in the trial with a broader range C+. 100% response rate patients in sarcoma clinical of answers D– trials (n = 76): Ratings of quality of medical care were mostly excellent or good E+ (A) extremity soft-tissue and did not vary between consenters and refusers F+ sarcoma – trial of amputation versus limb- sparing surgery plus radiation (patients offered randomisation) (B) head–neck–trunk soft- tissue sarcoma – trial of limited surgery + radiation + adriamycin versus limited surgery + radiation + adriamycin + parvum immunotherapy (patients offered randomisation) (C) primary osteogenic sarcoma – trial to receive versus not receive immunotherapy with BCG (patients offered randomisation)

USA

(2) Bevan et al. Study to assess patients Structured interview (1) Motivation: Descriptive study (1993) attitudes to participation in In group A, 62% stated that their motivation for participation clinical trials was to help others, and 39% to improve their own treatment A+ Reasons given by group B for not participating included B2 Group A – patients (n = 66) unwillingness to change treatment (25%) and objections from C+. 99% response rate in previous or current family members (25%) D– clinical trials under care of (2) Informed consent: E– the authors whose interests In group A, 83% felt that they had adequate time to consider F+ including hypertension, their participation clinical pharmacology, and 60% of respondents would have liked written information to general medicine retain for reference Group B – patients who 54% of consenting patients were happy with every aspect of had declined participation the trial in trials (n = 12) Group C – patients who See also Table 12b – views of the public had not been offered trial entry (n = 119)

Scotland

(3) Daughterty Pilot study to examine Structured interview: using Although 85% of respondents chose to participate in their trial Descriptive study et al. (1995) † issues related to partici- open- and closed-ended explicitly for the chance of personal gain, only 22% actually expected pation and the perceptions questions from a standard- to receive therapeutic benefit. However, 70% of patients said that A– of patients towards their ised survey form they believed that they would receive psychological benefit by B3 trial participating C+. 90% response rate Few patients actually estimated the quantitative chance of D+. Most questions Participating patients in therapeutic benefit (no data supplied) closed-ended eight different Phase 1 E+ cancer trials (n = 27) See also Table 10 – audit, and Table 13 – views of healthcare F+ professionals USA

* Tables 12a and 12b list all articles pertaining to the attitudes of patients and the public to real and hypothetical clinical trials. Table 12c extracts the articles from the first two tables which relate specifically to the patient’s reported motivations behind participating in trials. † This paper examines attitudes to Phase 1 trials.

continued 103 Tables

TABLE 12a contd Views of the patients and public in real trial scenarios*

Author Purpose of study and Method of assessment Main results Comments sample population (see quality of evidence (n = No. of respondents checklist – Table 11) to survey)

(4) DeLuca Study to determine variables Questionnaire designed 32% of respondents reported not reading the consent form, with Descriptive study et al. (1995) contributing to patient by the authors using 30 no difference between consenters and refusers participation in RCTs variables, along a five-point 97% of those approached initially by the doctor consented whereas A–. Insufficient information scale 66% of those initially approached by the nurse consented B3 Patients eligible for 79% of those who felt that their participation would contribute to C–. Response rate not given participation in one of 12 medical knowledge actually consented D– randomised cardiovascular 88% of those who believed that they would receive better medical E– trials (n = 248) treatment also consented F+

USA See also Table 10 – audit Errors in how p value was reported and some strange percentages in some tables

(5) Harth and Study to assess motivational Questionnaire administered 100% of those that consented expressed a wish to contribute to Descriptive study Thong (1990) characteristics of parents at interview consisting of medical research who volunteer their structured and open-ended 99% wanted the research to benefit other children A– children for clinical trials questions 90% consented in order to benefit their own child B2 95% of refusers did not volunteer for fear of side-effects C+. Of the 88% of families Parents of participating 52% expressed a dislike of modern medicines that were available after the children (n = 68) in a trial, 95% agreed to randomised, placebo- participate in the survey controlled trial of a drug D– to treat asthma and to a E– control group of parents F– whose children were eligible but who refused (n = 42)

Australia

(6) Harth and Study to examine the Structured questionnaire 15% of parents considered the information provided at consent Non-randomised Thong (1995) attitudes and knowledge of was presented 6–9 months to be unnecessary because of their faith in their physician’s advice cohort study parents after having given after the children had been 40% saw the legal requirement of a signature on the informed informed consent for their entered into the trial consent form as a device for protecting physicians from litigation A+ children to enter a trial while 19% saw it as a protective mechanism for research participants B2 79% of parents thought that there were no or low risks to C+. Response rate of 94 Parents of children partici- enrolment in clinical trials at their hospital and 70% of consenters and pating in a randomised, refusers, respectively placebo-controlled trial of See also Table 10 – audit D–. Insufficient information ketotifen, an oral asthma E– drug (n = 62) F+

Australia

(7) Hassar Study to examine the Interview and questionnaire (1) Perceived motivations of investigators by patients Descriptive study and Weinstein motivations and under- presented at the end of the participating in the trial: (1976) standing of participants trial to elicit patients 94% of respondents thought that the investigator had A– in a clinical trial motivations for entering undertaken the study to help the patient’s medical condition B3 the trial 88% of respondents thought that the investigator wanted to C+. 94% response rate Participants in a clinical trial help other patients D– of an anti-inflammatory drug 24% thought that the investigator wanted to help the university E+ (n = 49). Recruitment rate involved with the trial F+ to the trial = 37% (2) Motivations of patients: 13% of those that refused trial entry did so for logistic reasons, USA while 3% were concerned about toxicity 50% of trial participants stated that they had wanted to contribute to the welfare of other patients, while the remainder felt that they would be the major beneficiaries (3) Quality of care in the trial: All trial participants thought that the quality of their care had not suffered during the trial period