12/26/2013

Types of Biomedical Research • Purposes INTRODUCTION & • Exploratory VS. Confirmatory Research MEASUREMENT IN • Designs CLINICAL RESEARCH • Observational VS. Experimental Sakda Arj‐Ong Vallipakorn, MD • Descriptive VS. Analytic MSIT, MA (Information Science) • Resources Pediatrics, Pediatric Cardiology Emergency Medicine, Ped Emergency • Laboratory VS. translational VS. Research Family Medicine • Category of clinical question Section of Clinical Epidemiology and Biostatistics Faculty of Medicine, Ramathibodi Hospital, • Therapy, diagnosis, causation(risk), Mahidol University prognosis

Clinical Research Research S T NIH definition : Good Question ? STOP !!! A Research conducted with human subjects Require Research ? T (or on material with human origin) for which an investigator directly interacts with human I subjects DiDesign MthdlMethodology ? S Population ? T Excluded in Vitro studies that utilized human Measurement ? I tissues that cannot be linked to living individual C S

7Steps to do research Sample Description

• Research questions Selection • Review & Do literature search Sample • Create study design : Protocol writing Comparison • Perform Data correction • Data management : Entry, checking and cleaning Population of patients • Data Analysis with condition of interest Describe • Conclusion Effects Chance

Conclusion

1 12/26/2013

Distinctive Aspects of Biomedical Research What research is made of ? Elements Purposes 1. Conclusion are extended from the sample to population 2. The measurements performed on the subjects, are Research questions What questions will the study address ? interpreted in the context of a statistical‐probabilistic model. Significance ( background) Why are the questions important? 3. Studies must undergo detailed and documented planning Design How is the study structured ? before starting and must then be performed with strict adherence to that plan. Subjects Who are the subjects and how will they be selected ? 4. Reasoning, method and conclusions are based on Variables What measurements will be made ? comparisons between groups. Statistical issues How large is the study and how will it be analyzed ? 5. The groups being compared should be as homogeneous (comparable) as possible during entire course of the study 6. Data from biomedical study must be analyzed by appropriate statistical methods, defined in the planning stage.

Bacchieri A. Fundamentals of Clinical Research 2007

Components of Research question The process of clinical research

•Population being studied Research Study Conduct Design Implement •Intervention (exposure) being studied Questions Plan Study •Control(comparison) being studied •Outcome being measured

Infer Infer Truth in Truth in Findings in The Universe The Study The Study

Hulley SB. Designing Clinical Epidemiology

The process of clinical research Errors

Research Study Conduct •Random error (Chance) Questions Design Plan Implement Study •Systematic error (Bias) Random Random & & Systematic Systematic error error

Infer Infer Truth in Truth in Findings in The Universe The Study The Study

Hulley SB. Designing Clinical Epidemiology

2 12/26/2013

Error Random error • Random error • Type I Error (Alpha) • Type II Error(Beta)

• Systematic error(Bias) Urine output • A Process at any stages of inference tending to produce results that depart systematically from true values.

Drug dose

Random error Random error

Urine output Urine output

Before After

Drug dose Drug dose

Random error Random error • Small sample size • High variation in smaples • One‐time measure ( eg. BP)

Urine output • Unreliable measure ( No calibration) • Too many measurements • Non‐standardized measurement

Before After

Drug dose

3 12/26/2013

H0= uSBP(A) =uSBP(B) Strategies to reduce random error H1= uSBP(A) = uSBP(B)

• Large sample size • Measure endpoints in a precise way Statistical Testing In Population Different exist (+) No Different exist (‐) • Standardizing aspect of the protocol which impact on patient‐to‐patient variation Different (+) Power False Positive (alpha) • Collecting data on key ppgrognostic factors Reject Null Hypothesis 1‐Beta Type I Error • Choosing a homogenous group of patient

• Choosing the most appropriated design No‐Different (‐) False Negative(Beta) True Negative Accept Null Hypothesis Type II Error 1‐alpha

Tool to assess random error Bias

• The P value : A numeric representative of the degree • to which random variation alone could account for the The systematic tendency of any factors associated with the design, conduct, analysis, and interpretation of the result of clinical study difference observed between groups or data being to make the estimate of an effect deviate from its true value. compared. • Confidence Interval : Provide a plibllausible range wihiithin which the true association lies and provide all the information in P value and more.

Bias Bias

• should be aware to • Information bias • Berkson Bias (synonym admission bias, hospital admission bias) • • Ascertainment bias • esp. case control study • Healthy worker effect • ( synonym. Self report ) • Volunteer Bias • Non‐Response Bias (eg. Questionnaire)

4 12/26/2013

Bias Bias

• Ecological Bias • Post Hoc Analysis Bias • Confounding Bias • Sub group analysis bias • Spectrum Bias( syn. Case mix bias) •

High Reliability low Is an unbiased study ever possible ?

• “ The skill of the…researcher,,, lies not conducting the High perfect study, but in documenting and assessing the likely impact of its perfections.” AB Validit y

Low

C D Silman A. Epidermiological Studies: A Practical Guide Cambridge University Press 2002

Sample Description Sampling Sample Description Bias Selection Selection bias Selection Sample Sample Comparison Comparison

Measurement bias

Population of patients Bias Population of patients Bias with condition of interest Describe with condition of interest Describe Internal validity Effects Effects Chance Chance

External validity Confounding bias Conclusion Conclusion

5 12/26/2013

Confounding Factors Sampling Sample Description Bias • A factors that distorts the true relationship of the study Selection bias Selection variables of interest by being related to the outcome of interest Sample Comparison

Measurement bias Study factor Outcome Population of patients Bias with condition of interest Describe Effects Chance

Confounding bias Conclusion

Confounding Factors Confounding Factors • A factors that distorts the true relationship of the study • A factors that distorts the true relationship of the study variables of interest by being related to the outcome of variables of interest by being related to the outcome of interest interest Confounding Factor Confounding Factor (Confounder) (Confounder) Smoking

Study factor Outcome Study factor Outcome Hot Tea CA. Stomach Drinking

Cohort study of worker & TCX Cohort study of worker & TCX

Worker TCX No TCX Worker TCX No TCX Lung CA 27 14 41 Lung CA 27 14 41 No Lung CA 48 67 115 No Lung CA 48 67 115 Total 75 81 156 Total 75 81 156 Relative Risk = 2.1 Relative Risk = 2.1 Non‐Smoker Smoker

Worker TCX No TCX Worker TCX No TCX Lung 123 Lung 26 12 38 CA CA No 24 48 72 No 24 19 43 Lung Lung CA CA Total 25 50 75 Total 50 31 81

Relative Risk = 1 No association Relative Risk = 1.3

6 12/26/2013

A Priority criteria of confounder Study Designs

• Clinically/Scientifically sensible • Must be a risk factor Exposure Outcome • Cannot be a n intervening factor • Must be associated with the exposure in the population ‐Etiologic agent ‐Disease (imbalance distribution) ‐Risk factor ‐Biochemical change ‐Therapeutic measure ‐Pathophysiology • In analysis, crude estimate not equally to adjusted estimate ‐Preventive measure ‐Pathophysiologic change

Did investigator Assign exposure ? Classification of study design Yes No Experimental Study Observational Study • Observational study Random/Allocation ? Comparison group ? • Descriptive or case‐series Yes No • Case control studies(retrospective) Yes No Analytical Descriptive • Cohort studies(prospective) Study Study • Historical cohort studies(retrospective) Non RCT RCT • Experimental study Direction ? • Controlled trials • E Æ O O Æ E E =O Studies with no controls Yes Yes (Same time) • Systematic Reviews/Meta‐analysis Yes Cohort Study Case control Study Cross‐sectional Study

Hierarchy of evidences Classification of study designs

Systematic • Classify by presence of comparison group Reviews • Classify by action of investigator RCT • Only observe Æ observational study • Assign intervention Æ Experimental study( Clinical trial) Cohort Study

Case‐Control study

Cross‐Sectional study

Case Report

7 12/26/2013

Logic of Cross Sectional Study Cross Sectional Study

• Advantage • In expensive, simple (no follow up) Case with exposure + ve • No exposure, no drop out

Sample Case with exposure ‐ ve • Disadvantage • Can establish association but not !!! “ conclusion” Population • Can not control confounder • Recall bias usually present Non‐case with exposure + ve • Incidence‐prevalence bias • Different sample size among groups Non‐case with exposure ‐ ve

Exam Logic of Case‐Control Study • Obesity Obesity O+ O‐

Ex + 50 100 Case with exposure + ve

Exercise Sample Case with exposure ‐ ve Ex ‐ 20 100 Population

Relative prevalence O+ = (50)/(100) / (20)/(100) Non‐case with exposure + ve = 1.67 Non‐Disease Association ??? Non‐case with exposure ‐ ve

Logic of Case‐Control Study Case Control Study

• Advantage Exposed • Quickly and Inexpensive Disease/Outcome • Feasible for rare disorder or long term follow up • May be required fewer subjects Non‐Exposed • Disadvantage • Recall bias Exposed No Disease/No Outcome • More effect of confounder • Difficult to find control group

Non‐Exposed

Time

Direction of study

8 12/26/2013

Logic of Prospective Cohort Study Prospective Cohort Study

With outcome Disease Expose

Exposure +ve Without outcome No Disease

Non‐Expose With outcome Population Without outcome Disease Exposure ‐ve No Disease Time

Direction of study

Retrospective Cohort Study Historical Cohort Cohort Study

• Advantage With outcome • Can be matched Expose • Can be standardized in eligible criteria & outcome assessment Without outcome • Can establish temporal association ** • Disadvantage Non‐Expose With outcome • Usually expensive • Hard to blind Without outcome • Long follow up period for rare disorder • Difficult to find controls and confounders Time

Direction of study

Randomized Control Trial Randomized Control Trial

Sample Sample Randomization

Population Population

Treatment Control Treatment Control

Blind assessment

Outcome No‐Outcome Outcome No‐Outcome Outcome No‐Outcome Outcome No‐Outcome

9 12/26/2013

RCT Study Accuracy of Test Result

• Advantage Disease • Confounding and variables can be balance by randomization Present (+) Absent (‐) • Blinding of subjects, medical staff and investigators are achievable Positive (+) TP FP Test • Disadvantage a b Negative(‐)FN TN • Cost in term of time and money c d • Dropout or loss to follow up are common event a+c b+d • Need time to final ressults. Sensitivity = True positive rate = a/a+c Specificity = True negative rate = d/b+d

How to choose Sense VS. Specificity ? Sensitivity Specificity 1.The ability of test to identify 1.The ability of test to identify correctly those who have correctly those who do not disease. have the disease •Anti HIV Positive เปนจริง ๆ หรือคะ ? 2.Use to “rule out” 2.Use to confirm “ Rule in ” •Negative ไมเปนแนคะ 3There3.There is a reason to suspect 3.Need when false‐positive dangerous but treatable result can harm the patient condition physically, emotionally or financially. SNout SPin

The study design for diagnostic test

• เปนจริง ๆ หรือคะ ? Patients Anti HIV Positive PPV suspected of Gold standard Target Diagnostic test target test condition (+/‐) •Negative ไมเปนแนคะ NPV conditions

Prevention/treatment 1. Random allocation (by chance) : participants to interventions (Randomization) 2. Blind : Double , triple blind….etc 3. Placebo 4. Intention to treat analysis 5. Complete follow up > 80 %

10 12/26/2013

Risk study (Causation) Cause • Exposure & Outcome • A cause of disease event is an antecedent event , condition, or characteristic that was necessary for occurrence of disease at the moment of it occurred, give that other Exposure Outcome conditions are fixed (Kenneth J. Rothman) Risk Factors Disease Association Intervention Health Problems • Is a statistical relationship between two or more events, Maneuver characteristics or other variables. Independent variable dependent variable

Causal criteria ( Modified from Bradford‐Hill AB) Criteria for judgment of causal associations(Hills’ Criteria) Temporality • Temporal sequence: Did exposure precede Strength outcome? Dose‐response • Strength of association : How strong is the effect, measured as relative risk or odds ratio? Consistency • ChCoherence with exiitigsting kkldgnowledge: Is the Biological plausibility association consistent with available evidence? Reversibility • Experimental evidence: has a randomized control • Analogy : Is the association similar to others Specificity Analogy Experimental evidences

Prognostic study Research Design which assess cause ? • A prediction of future course of diseases • Cohort study following its onset • Case‐control • A group of patients having something in • Cross sectional study common are assembled and followed forward in time, and clinical outcomes are measured. • “ Natural history of disease”

11 12/26/2013

Onset of Acute MI Prognosis

• An inception cohort of persons, all initially free of Risk Prognosis Death outcome of interest Prognostic Factors Risk Factors • Representative of sample Age Age • Male female Homogenous to prognostic risk Smoking Smoking • Objective outcome measurement HT Hypotension • FU >= 80 % of patients LDL Anterior infarction Inactively CHF Ventricular arrhythmias

Broad topics of Research Is the exposure or intervention Experimental Study • Diagnosis –Demonstrate that new diagnosis test is under the control of investigator valid/reliable Preferred “cross sectional study” • Causation or Risk‐Determine that agent is related to Are the subjects followed up over time? Cohort Study development of illness, preferred “Cohort or case‐ control study” • Therapy –Testing the efficacy of intervention Are the subjects selected according to Case‐Control Study preferred “RCT” The outcome ? • Prognosis‐ determine what happen to someone with some stage of disease, preferred “Prospective Cohort study” Cross sectional Study

Key methodological strength and weakness

Design Starting point‐ Strength Frequency of Assessment publication RCT E‐OLow susceptibility Feasibility, to bias generalizability Cohort E‐OFeasible when Susceptible to bias, Thank You randomization of limited validity exposure not dr.sakda@gmail .com possible Case‐Control O‐EOvercomes Susceptible to bias, temporal delays, limited validity may require small sample size

12