Assessing Patient Benefit in Phase 1 Trials

A thesis submitted to McGill University in partial fulfillment of the requirements of the degree of

Master of Science in Experimental : Bioethics Specialization

by

Xuehan (Sean) Zhang Biomedical Ethics Unit McGill University, Montreal June 2019

Supervised by Dr. Jonathan Kimmelman

© Xuehan Zhang, 2019

1

Table of Contents

Abstract ...... 3 Résumé ...... 5 Acknowledgements ...... 7 Contribution of Authors ...... 8 Introduction ...... 9 Chapter 1: Phase 1 Cancer Trials and Ethical Challenges ...... 11 Introduction to Phase 1 Cancer Trials ...... 11 Ethical Challenges ...... 14 A Discussion of Benefits ...... 16 Summary ...... 23 Chapter 2: Submitted Manuscript, Extended Methodology, and Protocol ...... 25 Objective ...... 25 Submitted Manuscript ...... 26 Extended Methodology ...... 52 Protocol ...... 59 Additional Results ...... 65 Chapter 3: Discussion ...... 66 Advantages to our approach ...... 68 Limitations ...... 71 Ethical and Policy Implications ...... 74 Summary ...... 79 Conclusion ...... 80 References ...... 83

2

Abstract

Participation in phase 1 trials is regarded as a therapeutic option by some , bioethicists, and medical organizations. Previous analyses have attempted to measure benefit in phase 1 clinical trials using Overall Response Rate, a statistic that estimates tumor reduction. However, this commonly used measurement is only a surrogate endpoint that does not have clear associations with survival or quality of life. Whether phase 1 cancer trials are considered therapeutic is consequential to a number of ethical and policy issues. The purpose of this thesis was to provide a review of the current ethics literature surrounding whether phase 1 clinical trials offer prospective benefit and to conduct a review of phase 1 cancer trials via a more clear definition of .

To evaluate the therapeutic value of phase 1 cancer trials, we determined ‘therapy’ as when patients receive active doses of treatments that ultimately are approved by the U.S. Food and Drug Administration (FDA) for their cancer type and looked to determine the proportion of patients that received a therapeutic treatment over the total number of patients enrolled. We then searched the ClinicalTrials.gov database for 1000 phase 1 oncology trials that started between

2005 to 2010. We extracted data regarding patient enrollment, cancer type, and drugs received.

We also looked at whether this proportion differed depending on certain trial characteristics. A total of 376 (1.15%) patients received a treatment that was FDA approved for their malignancy at a dose delivered in the trial. Meta regression showed a significantly greater proportion of patients receiving a drug that was ultimately FDA approved in biomarker trials (3.4% vs.0.9% , p =

3

0.0044), patients receiving monotherapy (1.8% vs. 0.5% for combination therapy, p = 0.035), and patients enrolled in single-indication trials (1.8% vs. 0.6% for mixed malignancy trials, p =

0.019).

Our results indicate that only 1 in 87 patients participating in phase 1 cancer trials would receive a treatment eventually approved for their indication at the dose received. Given previous estimates of serious adverse events rates of 10-19%, this represents a high risk to benefit ratio.

Current practices in informed consent, and policies regarding reimbursement of trial costs and hospice care eligibility, ought to be reconsidered in light of our findings.

4

Résumé

La participation aux essais cliniques phase 1 en oncologie est considérée comme une option thérapeutique par certains médecins, bioéthiciens et organisations médicales. Des analyses antérieures ont tenté de mesurer les bénéfices des essais cliniques de phase 1 à l'aide du taux de réponse globale, une statistique qui estime la réduction des tumeurs. Cependant, cette mesure n’est qu’un critère d'évaluation substitutif qui n'a pas d'associations claires avec la survie ou la qualité de vie. De nombreuses questions éthiques et politiques sont dépendantes sur si les essais cliniques phase 1 en oncologie sont considérées comme thérapeutiques. L’objectif de cette thèse

était de fournir un aperçu global de la littérature actuelle sur les essais cliniques de phase 1 en oncologie et de procéder à une analyse en utilisant d'une définition plus claire de thérapie.

Pour évaluer la valeur thérapeutique des essais cliniques phase 1 en oncologie, nous avons défini que les patients reçoivent une dose «thérapeutique» d’un médicament si cela est finalement approuvé par l’agence américaine des produits alimentaires et médicamenteux (FDA) pour leur type de cancer. Nous avons déterminé la proportion de patients qui ont reçu un traitement thérapeutique sur le nombre total de patients inscrits. Ensuite, nous avons cherché dans la base de données ClinicalTrials.gov pour 1000 essais d'oncologie de phase 1 qui ont débuté entre 2005 et 2010. Nous avons extrait des données concernant l'inscription des patients, le type de cancer et les médicaments reçus. Nous avons également examiné si cette proportion différait selon certaines caractéristiques de ces essais cliniques. Un total de 376 (1,15%) patients ont reçu un traitement qui a été approuvé par la FDA pour leur type de cancer à une dose livrée

5 durant l'essai clinique. La méta-régression a montré qu’il y a une proportion significativement plus grande de patients qui ont reçu un traitement «thérapeutique» dans les essais clinique avec des biomarqueurs (3,4 % contre 0,9 %, p = 0,0044), dans les essais où les patients reçoivent une monothérapie (1,8 % contre 0,5 % pour la thérapie combinée, p = 0,035), et dans les essais avec une seule indication (1,8 % contre 0,6 % pour les essais mixtes malignes, p = 0,019).

Nos résultats indiquent que seulement 1 patient sur 87 qui participe à des essais de phase

1 en oncologie recevrait un traitement éventuellement approuvé pour leur indication à la dose reçue. En tenant compte le taux des effets indésirables sérieux de 10 à 19 %, cela représente un ratio risque/bénéfice élevé. Les pratiques actuelles concernant le consentement informé, le remboursement du coût des essai cliniques, et l'admissibilité aux soins palliatifs devraient

également être considérées.

6

Acknowledgements

I would like to express my utmost appreciation for my thesis supervisor, Dr. Jonathan

Kimmelman, for his warm mentorship. He helped guide this thesis from inception to its completion. His multi-disciplinary expertise in the field of biomedical ethics, cancer research, epidemiology, statistics, and the art of juggling have greatly contributed to the completion of this thesis manuscript.

I would also like to thank my thesis committee, comprised of Dr. Carolyn Ells, Dr. Abe Fuks, and Dr. Michael Pollak. They gave valuable input to my project and ensured that I stay on track to complete my degree.

My thanks also goes to Dr. Dean Fergusson and Dr. Ranjeeta Mallick at the Ottawa Hospital

Research Institute, who provided expert statistics consultation and analysis.

I thank my fellow labmates: Dr. Michael Yu, Dr. Benjamin ‘Murph’ Carlisle, Amanda

MacPherson, and Holly Sarvas for helping with various aspects of the project. I would also like to extend my gratitude to the STREAM research group as a whole – both current members and alumni, who have all been excellent colleagues and fantastic people to work with.

Finally, I would like to thank my parents for their unwavering support and encouragement to me throughout my study.

7

Contribution of Authors

JK conceived of the main conceptual ideas for the project. XZ developed the methodology with assistance from DF, MY, and BC. XZ extracted the data. Some data was double extracted by HS and AM. XZ analyzed the data with support from RM and BC. XZ wrote the manuscript submitted for publication with revisions from JK and DF. XZ wrote the introduction, background, methodology, results, discussion, and conclusion of the thesis with revisions from

JK.

8

Introduction

Phase 1 oncology trials represent the first time that experimental cancer are tested in human subjects 1–3. There has been much debate surrounding the putative benefit of phase 1 trials in oncology, as cancer patients are enrolled in these trials instead of healthy volunteers. Whether phase 1 trials are therapeutic also has significant implications in informed consent discussions, as cancer patients primarily enroll in phase 1 trials for the chance of achieving medical benefit4.

Designation of phase 1 trials as therapeutic also has policy implications: reimbursement of trial expenses through Medicare5 require trials to have ‘therapeutic intent’6, while hospice care reimbursement is rendered ineligible if phase 1 trials are considered ‘disease-modifying’7. The current academic debate surrounding putative benefit found in phase 1 trial participation provide an incomplete assessment of whether phase 1 trials can provide prospective benefit. Previous methods used to assess the therapeutic status of phase 1 studies do not offer much traction on whether patients truly benefit, as these estimates of benefit rely mainly on Overall Response Rate

(ORR)8–11, a surrogate endpoint and unreliable predictor of survival outcomes12–14.

The objective of the thesis is thus to provide a more robust examination of the potential for benefit in phase 1 cancer trials. In this manuscript, we use a novel and flexible methodology based on cancer drug approval from the U.S. Food and Drug Administration to define therapeutic benefit. We also employed meta-regression methods to determine whether certain characteristics of the trial affected the chance of receiving a beneficial treatment. Our results will be contextualized within the current academic literature, we provide an ethical analysis on whether

9 phase 1 cancer trials can be ethically justifiable in light of our results, and make recommendations for policies that depend on therapeutic justification of phase 1 cancer trials.

10

Chapter 1

Phase 1 Cancer Trials Background and Ethical Challenges

Introduction to Phase 1 Cancer Trials

Phase 1 clinical trials are exploratory trials that primarily evaluate the safety and tolerability of a new treatment and often represents the first instance in which a treatment is tested in human subjects(1,11). These trials enroll a small number of participants, typically between 25-50. While phase 1 trials for treatments in many contexts allow healthy volunteers to test the safety of new treatments, the case is usually different in oncology1. Phase 1 trials in oncology mostly enroll patients with cancer – often those with locally advanced, refractory, or metastatic illness and have gone through conventional therapy methods, as anticancer drugs have a palpable chance of causing side effects that would be too serious for normally healthy volunteers16.

Safety and Tolerability Testing in Phase 1 Cancer Studies

The primary goals of phase 1 trials are to determine the safety, tolerability, and recommended phase 2 dose (RP2D) of the investigational drug. A phase 1 trial ends when a

RP2D is found or if the drug is determined to be too toxic to administer due to unexpected and serious side effects1,2,15.

The conventional method for determining tolerability in phase 1 trials is the ‘3+3’ dose escalation method. Early-generation cytotoxic chemotherapy agents were primarily developed with this method. A phase 1 clinical trial stops when dose-limiting toxicity (DLT) has occurred

11 in one-third of patients within the same dose cohort. The DLT is defined as an event in which the patient cannot be treated at the planned dose because of serious adverse events (SAE) due to toxicity. Traditional dose-finding techniques begin with very low doses in the first dose cohort of three patients to minimize the risk from toxicity. If no drug-related toxicity occurs within this cohort, then the next three patients are enrolled in a higher dosage level. If one drug-related toxicity occurs, the cohort is then expanded to six patients – hence ‘3+3’, to further verify whether toxicity rate would reach one-third of the cohort. If no other toxicity occurs, the dose is further escalated. When the toxicity rate reaches one-third in a cohort of six participants, then the previous lower dose is used as the RP2D. In traditional 3+3, this dose is also considered the maximally tolerated dose (MTD). 3+3 designs are part of a broader class of rule-based designs, where statistical calculations to determine dosing are not required prior to the trial. There are other variations of the 3+3 designs, though less popular, such as 2+4, 3+3+3, and 3+1+1. After proven safe, the RP2D can then be further pursued in phase 2 trials. A phase 1 trial can be terminated before the RP2D is reached if toxicity and side-effect profile of the experimental agent deviate significantly from investigators’ expectations2,15,17.

Model-based designs are more modern alternative dose-escalation method, using

Bayesian statistical models to estimate the shape and distribution of a dose-toxicity curve. This method requires a prior distribution, θ. Put simply, model-based designs typically use DLT results from previous patients treated in the trial to determine optimal dose for the next patient.

One such design, called the Continual Reassessment Method (CRM), continually adjusts the dose-toxicity curve based on the results of each regimen given. The doses that patients receive

12 would differ depending on how previous patients reacted. If the previous patient had a DLT, then future doses would be adjusted to be lower compared to if no patients had a DLT. The CRM can also compute a model based on investigators’ preferences of a target DLT rate, unlike conventional 3+3 designs where the DLT rate is set typically at 33%. Model-based designs are more flexible and generally good estimates of dose-limiting toxicity at the RPD2 without treating too many patients at suboptimal doses, though rule-based designs are still more frequent due to

1,2,15 their ease of implementation.

In the modern era of targeted drugs, other endpoints may be used in favor of MTD. For one, DLTs are not necessarily expected to be reached in novel therapies such as molecularly targeted agents (MTA) and immunotherapies, as these drugs can induce biological activity without significant toxicity1,18. In contrast, most clinical activity occurs within 80%-120% of the

MTD in cytotoxic agents19. As such, starting doses for MTAs can be higher than typical cytotoxic therapies. Studies have shown that one tenth of the lethal dose in 10% of rats can be used as a safe starting dose in humans, and that even doubling the starting dose is still likely to be safe. Additionally, toxicity induced by MTAs may be delayed, and thus may not be adequately captured within a traditional clinical trial setting15. Optimal biological dose (OBD) has been proposed as an alternative to MTD to establish the RP2D, where the recommended dosage would be instead dependent on the occurrence of significant change in pre-specified biological thresholds, such as plasma concentration of a relevant biomarker20,21. Recent reviews of phase 1 trials for approved therapies show that investigators have begun to make the switch from toxicity-based endpoints in trials of MTA22.

13

Efficacy Testing in Phase 1 Studies

While determination of safety and dosage is usually the primary goal of phase 1 trials, efficacy endpoints are also studied – usually as a secondary outcome. As phase 1 trials tend to be short in duration, enroll few patients, and do not utilize randomization or a control arm, they are not powered to establish meaningful data on long-term, survival-based endpoints such as progression-free survival (PFS) or overall survival (OS)23.

Tumour reduction – determined by overall response rate (ORR), is monitored instead.

The ORR represents the proportion of patients whose tumors were reduced by a pre-specified diameter or volume (partial response - PR), or completely disappeared (complete response - CR).

For solid tumours, measurement of tumor size is conducted through the use of radiological imaging over a pre-specified time period 24.

Response rates for hematologic, or -borne are less standardized; evaluation of response is instead dependent on the type of cancer indication. However, the European

Society of Medical Oncologists (ESMO) has separate guidelines for response evaluation in different types of , myeloproliferative disorders, and 25.

Ethical Challenges

Two major ethical challenges are present in phase 1 cancer trials. The first pertains to establishing whether the risks incurred in research are properly balanced by benefits. Ethics documents such as the Belmont Report26, Declaration of Helsinki27, and Tri-Council Policy

Statement28 all take into consideration some form of the principle of ‘beneficence’, which

14 ensures trial participants’ welfare and prevents unnecessary harms. On the one hand, cancer drugs are toxic and thus have inherent risk; at the point of phase 1 testing the evidence supporting their safety and efficacy in human beings is very weak. On the other hand, phase 1 participants are cancer patients rather than healthy volunteers, they may stand to benefit in some way from phase 1 trials. There have been occasions, such as the first imatinib phase 1 trial, where patients experienced significant and lasting benefit29.

The second ethical issue pertains to informed consent. Commentators have noted the phenomenon of ‘therapeutic misconception’30 and ‘therapeutic misestimation’31 in patients who participate in research. Therapeutic misconception refers to the phenomenon in which patients enrolled in clinical research mistakenly believe the objective of the clinical trial is for their personal benefit, rather than to test experimental treatments in a scientific manner. Therapeutic misestimation refers to when patients either overestimate benefits or underestimate the risks of research. Some reasons that patients may misunderstand or overestimate the prospect of benefit include unclear disclosure in consent documents, socially-conditioned beliefs in the role of physicians, or not understanding trial methodology30. Terminally ill cancer patients, such as those who enroll in phase 1 trials are not excepted; surveys have shown that the majority of cancer patients have therapeutic misconceptions or overestimate the benefits of study participation32.

15

A Discussion of Benefits

There has been contention in the medical and ethics literature regarding the extent to which phase 1 trials offer patients benefit. Before presenting either supporting or refuting arguments, it is necessary to first define the types of benefits that are relevant to clinical research.

What is considered benefit in clinical research?

In her seminal article, “Defining and Describing Benefit Appropriately in Clinical Trials”33,

Nancy King outlines three major types of benefit attributable to clinical research:

1) direct benefit, which is benefit that can be directly attributable to the intervention studied. In this case, the drug tested was found to directly affect a participant’s disease process.

2) collateral benefit, which is benefit that arises from merely being enrolled in research.

For instance, enrollment in a clinical trial ensures regular monitoring by physicians and access to promising new therapies. Participants have reported to feel psychological comfort attributed to involvement in research.

3) aspirational benefit, or benefit to society and future patients that arises from the results of the trial. The results help inform the scientific or medical communities and may change current practice. This form of benefit is not expected to directly affect participants currently enrolled in research.

Arguments regarding whether patients can benefit in phase 1 research focuses mainly on whether there is a prospect of receiving direct benefit. Many commentators argue that collateral benefit

16 cannot be used to justify a trial or its risk34,35. Hence, an evaluation of collateral benefit associated with phase 1 trial participants, while certainly of interest to physicians and patients, is not central to the ethical justification and demands we place on sponsors and researchers when they conduct trials. Aspirational benefit associated with phase 1 trials in this thesis will be assumed and is thus beyond the scope. For these reasons, this thesis focuses on direct benefit.

Arguments for Prospect of Direct Benefit in Phase 1 Oncology Research

Commentators have made several arguments that advocate for the fact that patients may receive tangible and direct benefit in phase 1 oncology trials.

1) Response in Phase 1 Trials are Comparable to FDA-approved drugs:

First, ORR data in phase 1 trials is comparable to some anticancer agents36 that have received

FDA approval. Toxic chemotherapy agents such as topotecan and gemcitabine with ORR’s between 5-10% have also been approved by the FDA37. Thus, by comparing toxicity rates and response rates, the typical risk-benefit ratio in phase 1 oncology studies may not be significantly worse than the risk-benefit ratios deemed acceptable for approval by the FDA. Additionally, a high ORR has been found to correlated to FDA approval, albeit only in palliative trials and solid tumors94.

2) Possibility of response is more prevalent than aggregate data suggests:

Commonly cited aggregate ORR values (between 5-10%) of all phase 1 trials do not tell the whole picture36. Previous meta-analyses have shown that over 60% of compounds evaluated in phase 1 cancer trials had at least 1 partial response, and more than 30% of drugs tested at the 17 phase 1 stage showed an ORR of 5% or higher 19. Additionally, phase 1 trials have the potential of producing substantially higher response rates: cisplatin tested in produced a

50% ORR in phase 1 trials38; imatinib in chronic myeloid leukemia demonstrated a complete hematologic response rate of 98%29; a combination of paclitaxel and carboplatin in women with advanced yielded an ORR of 75%39.

3) Novel and promising treatments are expected to increase response:

As research in oncology therapies advances, commentators11 have expressed hope that more modern approaches to anticancer treatment can increase the number of patients who respond in phase 1 trials, and also increase the number of trials that may test a therapeutic intervention. This assertion has been somewhat supported by examining meta analyses of clinical trials between different time periods (see Table 1).

Table 1: Risks and Benefits of Phase 1 Cancer Trials from High Quality Meta Analyses

Study (Time ORR Therapy type Serious adverse Mortality Period) included in ORR events estimate Estey et al. (1974- 4.2% Monotherapy - - 41 1982) Decoster et. al 4.5% Monotherapy - 0.5% 19 (1972-1987) Horstmann et al. 10.6% Monotherapy + 17.8% 0.49% 8 (1991-2002) Combination Roberts et al. 3.8% Monotherapy 10.3% 0.54% 9 (1991-2002) Fukuda et al. 13.2% Monotherapy + 19.9% 0.99% 10 (2001-2012) Combination

18

4) Novel clinical trial designs are expected to increase response:

Since the inception of the traditional dose-escalation design, there have been concerns that the majority of cancer patients enrolled in phase 1 trials receive sub-therapeutic doses of the drug36.

Over time, investigators have proposed alternate dosing schedules to minimize the number of patients that would receive a low dose of the drug through. For instance, accelerated titration designs allow for rapid dose escalation where patients may experience higher doses. Intrapatient dose escalation, where dose is increased over time within individual patients, so that have also been proposed15. A review of 270 phase 1 trials studies between 1997-2008 showed a reduction in the number of patients treated at sub-RP2D doses (46% vs 56%, p=0.0001)42. In the era of precision medicine, biomarker screening utilized in phase 1 trials has been found to increase response rate93. The addition of ‘expansion cohorts’ – the enrollment of additional patients, usually of a specific tumor type that showed promising responses during dose escalation, after the RP2D has been determined, may also increase the response rate of patients in phase 1 trials43.

To summarize, phase 1 cancer trials demonstrate overall response rates comparable to some FDA-approved drugs; in fact, there is a chance that patients may even receive the next breakthrough anticancer drug. Novel forms of anticancer therapies and trial designs could increase the likelihood that a cancer patient would receive an effective drug, in addition to a more therapeutic dosage.

19

Arguments against Prospect of Direct Benefit in Phase 1 Oncology Research

While there are many advocates for the idea that phase 1 trials are beneficial for cancer patients, some critics remain skeptical.

1) The design of Phase 1 trials is not rigorous enough to evaluate efficacy

The primary objective of phase 1 trials is to evaluate the safety of the drug. As a result, the majority of phase 1 trials are single-arm, unblinded studies which enroll only a few number of patients. Any response thus cannot be causally attributed to the intervention tested due to the lack of a control arm23. Advocates for benefit in phase 1 trials have touted stable disease, alongside

ORR as an indicator of benefit for patients36,44. Without a control arm, one cannot assign a causal relationship for whether stable disease is a result of the study intervention or radiographic error, or simply the natural history of the disease. Whether a control arm is necessary to a correlation between the study intervention and ORR is more contentious. Cancer tumors do not spontaneously regress, and there remains little evidence to support any placebo effect on cancer, thus suggesting that any reduction in tumor size should be, in theory, due to the effect of a treatment23. Nevertheless, ORR measurements are taken at single timepoints, the natural history of some types of cancer tumors suggest that they may grow and shrink cyclically, and radiological techniques of measuring of tumor size can be inaccurate45. For example, response rates reported in placebo arms of cancer trials range from 1.6%46 to 2.7% 47.In addition, phase 1 trials often enroll a heterogenous cohort of mixed cancer patients without regard for tumor type16. As clinical trials are based on pre-clinical evidence in animal models of a certain cancer

20 indication, the enrollment of a heterogenous selection of cancer patients with various different cancer types is unlikely to show benefit for most patients that enroll.

2) Most therapies tested in phase 1 trials do not demonstrate efficacy in clinical development

The overall likelihood of approval for a cancer drugs tested in phase 1 trials was found to be only

6.7% in a 2015 analysis by Hay et. al48. When compared to other indications such as neurological, endocrine, or autoimmune diseases, cancer drugs have the lowest probability of approval after phase 1 trials. A more recent and larger analysis by Wong et. al showed the rate for cancer drug success was even lower than previously reported; at only 3.4%49. However, trials that used biomarkers to select patients showed twice the likelihood of approval compared to trials without biomarkers. As regulatory approval affirms the safety and effectiveness of any particular drug, the low success rate of cancer drugs is indicative of the difficulty in developing effective therapies for cancer. Of the thousands of exploratory cancer regimens initiated each year, only around a handful receive approval. In 2016, the FDA approved 16 new cancer-drug indications in oncology. 2017 and 2018 both had 23 novel drug-indication combinations, though a majority were new indication approvals for already approved cancer drugs (e.g. pembrolizumab and nivolimab)50.

Aside from the fact that approval rates in oncology are relatively low, not all approved drugs have benefits comparable to well-known ‘blockbuster’ drugs such as imatinib or pembrolizumab. Aware of this fact, both the American Society of Clinical Oncologists (ASCO)51 and European Society of Medical Oncologists (ESMO)52 developed frameworks to value

21 approved cancer drugs, based mainly on efficacy, toxicity, and quality of life. Some drugs rank highly (ado-trastuzumab for HER2 positive , vemurafenib for BRAF positive ), while others are considered less valuable (cetuximab and regaforenib for )47. Thus, even if, as some commentators have stated that patients are able to receive medications that would later be approved, from phase 1 trials, these drugs are likely not going to be the next scientific breakthrough, but rather a cancer drug of only middling benefit.

3) ORR is a poor indicator of actual benefit

Another argument against the prospective benefit of phase 1 trials is due to the uncertainty between the correlation of ORR and genuine medical benefit12–14. Since the phase 1 studies only focus on ORR as an efficacy endpoint, one cannot make a valid claim about direct medical benefit based purely on ORR results. Studies in a variety of advanced or metastatic solid tumors have attempted to determine correlation between the surrogate outcome, ORR, with the clinical outcome, OS. Almost all these studies found little to no effect of ORR on survival endpoints. At best, ORR appears only to be a predictor of survival of one year within a limited selection of tumor types12-14. In the modern era of immunotherapy treatments, the association53 between response rate and survival was found to be weak (complicating things further, some patients receiving immunotherapy show pseudoprogression, meaning their cancers progress radiologically though their disease is held in check)54,55. The FDA also acknowledges that ORR is not a predictor of patient-centered benefits. In consultations with the Oncologic Drugs

Advisory Committee in the 1980s, the FDA deemed that ORR was not a clear predictor of clinical benefit, such as increase in survival, improvement in quality of life, or symptoms56. 22

In conclusion, these represent three concerns regarding whether patients benefit in phase

1 trials. The first concern lies in their methodology: phase 1 trials are not structured to evaluate benefit, and therefore results that attempt to speak to their benefit are ambiguous. Second, the probability of benefit is low due to the relative lack of cancer therapies that end up receiving approval in comparison therapies that fail. The third lies in the difficulty in translation to actual benefit of the main benefit indicator, ORR.

What are some implications?

Whether patients benefit in phase 1 trials is pertinent to numerous ethical and policy issues. Cancer patients’ main objective when enrolling in clinical trials is for clinical benefit4; thus, an accurate estimation of benefit is highly relevant to informed consent discussions.

Reimbursement clinical trial costs by Medicare is dependent on a designation of clinical trials6.

Finally, reimbursement of hospice care would be rendered ineligible for patients enrolled in phase 1 trials should they be considered therapeutic7.

Summary

Phase 1 trials in oncology are generally first-in-human studies that evaluate the safety of experimental therapies. Participants in phase 1 trials are cancer patients; since anticancer drugs can be highly toxic, ethical challenges arise in evaluation of their risk/benefit ratio and whether this information is adequately disclosed to patients. Arguments for the idea that these trials are beneficial include the fact that patients may receive an approved drug, response rates in phase 1 trials are comparable to some approved drugs, and that benefit is expected to increase in the

23 modern era of phase 1 trials. Counter arguments include the fact that phase 1 trials are not designed to evaluate efficacy and that ORR is a poor indicator of benefit.

24

Chapter 2:

Submitted Manuscript, Extended Methodology, and Protocol

Objective

The goal of this thesis will be to estimate the proportion of patients who receive therapeutic treatment regimens in phase 1 trials using a prospective, social benchmark of

“therapy” as FDA approval. We define a ‘therapeutic’ administration of a drug in a clinical trial to be a case in which the patient: a) receives an ultimately approved drug at the time of administration, b) has the diagnostic/indication criteria reflected on the FDA label of the approved drug, and c) receives the drug at dosage reflected on the FDA label. This information can help inform debates about the therapeutic status of phase 1 trials can also be used to help inform patients of the probability of receiving regime that will ultimately prove effective if they enroll in phase 1 studies.

Our study will complement existing data regarding the benefits of cancer clinical trials and also address their aspirational, or societal benefit. Whether a drug receives FDA approval – and thus introduction to the public market, helps determine the extent to which the patient burden endured in clinical research has resulted in benefit to society.

25

Submitted Manuscript

What Proportion of Patients in Phase 1 Oncology Trials Receive Treatments that are

Ultimately Approved as Safe and Effective? A Systematic Review

1 2 1,* Sean X. Zhang, BSc , Dean Fergusson, PhD , Jonathan Kimmelman, PhD

1. Studies of Translation, Ethics, and Medicine. McGill University Biomedical Ethics Unit. 3647 Rue Peel Montreal Quebec H3A 1X1;

2. Ottawa Hospital Research Institute. University of Ottawa, Department of Medicine, , and the School of Epidemiology and . 600 Peter Morand Crescent Ottawa Ontario K1G 5Z3;

* Corresponding author: Dr. Jonathan Kimmelman Email: [email protected]. Telephone: 514-398-3306

26

Word Count: 2591

Conflict of interest: JK declares he serves on a DSMB for Ultragenyx, which conducts phase 1 noncancer trials. Authors declare they have no other competing interests.

Funding: CIHR

Acknowledgements: Authors wish to thank Dr. Ranjeeta Mallick for her statistics expertise and performing statistical tests. We also wish to thank Dr. Michael Pollak, Dr. Benjamin Carlisle, Dr.

Michael Yu, Holly Sarvas, and Amanda MacPherson for aid and consultations; faults remain our own.

27

Abstract

Background: Some policies and many researchers regard participation in phase 1 oncology trials as a therapeutic option for patients that meet eligibility. However, such claims have tended to rely on anecdote and surrogate measures of benefit, such as objective response.

Purpose: To evaluate the therapeutic value of participation in U.S. phase 1 cancer trials by determining the proportion of patients in phase 1 trials that receive active doses of treatments eventually receiving FDA approval for their disease.

Data sources: ClinicalTrials.gov, Pubmed, ASCO reports, and Drugs@FDA were searched between May 1, 2018 and July 31, 2018.

Study Selection: Phase 1 oncology trials registered on ClinicalTrials.gov that had at least one study location within the United States were included.

Data Extraction: 1 investigator extracted study data from clinical trial registries. Study data that required additional publication searches were independently extracted by 2 investigators

Data Synthesis: 1000 phase 1 oncology trials launched between 2005 to 2010 were randomly sampled from ClinicalTrials.gov; these enrolled 32,582 patients. A total of 376 (1.15%) patients received a treatment that was FDA approved for their malignancy at a dose delivered in the trial.

Meta regression showed a significantly greater proportion of patients receiving a drug that was ultimately FDA approved in biomarker trials (3.4% vs.0.9% , p = 0.0044), patients receiving monotherapy (1.8% vs. 0.5% for combination therapy, p = 0.035), and patients enrolled in single-indication trials (1.8 vs. 0.6% for mixed malignancy trials, p = 0.019). 28

Limitations: Additional years of follow-up time for trial trajectories to mature may required for a more accurate estimate of the therapeutic value of phase 1 trials; lack of publication for some trial registries

Conclusion: A total of 87 patients had to participate in phase 1 cancer trials in order for one to receive a treatment that was approved for their indication at the dose received. Given previous meta-analysis estimates of serious adverse event rates of around 10-19%, this represents low therapeutic value for phase 1 trial participation. This has implications for informed consent and risk/benefit assessment of clinical research.

Funding Source: Canadian Institutes of Health

29

Introduction

Phase 1 oncology trials are designed to evaluate the safety, tolerability, and dosing for new therapeutic strategies. Because they provide patients access to investigational treatments, many patients, oncologists, and policy-makers regard them as a therapeutic option for patients who meet eligibility.(1,2) The classification of phase 1 trials as therapeutic has significant implications for how enrolment is explained to patients in informed consent, reimbursement of expenses associated with trial participation(1), hospice care(3), and optimal transition to end of life planning

(4) and care.

The American Society of Clinical Oncology (ASCO) asserts that phase 1 cancer studies have therapeutic intent and thus have the potential to provide for patients direct medical benefit.(1,2)

However, this view has been challenged .(5) Advocates supporting the therapeutic view of Phase

1 trials highlight that meta-analyses show that overall response rate (ORR), a surrogate endpoint for benefit based on tumor shrinkage, is 10.6-13.2% in phase 1 trials.(6,7) Such response rates are comparable to certain anticancer medications that are approved by the US Food and Drug

Administration (FDA).(8) Other arguments include the fact that enrollment in phase 1 trials have the potential to expose patients to the next ‘medical breakthrough’.(8) For example, the first phase

1 trial of imatinib (Gleevec) resulted in 97% of patients achieving complete hematological

(9,10) response and 54% achieving complete cytogenic response.

However, meta-analyses of phase 1 cancer trials suffer from problems related to measuring the true therapeutic value of phase 1 studies. First, they estimate benefit using surrogate endpoints

30 which are an unreliable predictor of clinical outcomes like survival or quality of life. (11–13)

Secondly, meta-analyses that combine the results of trials testing many different drugs and indications contend with high levels of methodological, clinical, and statistical heterogeneity.

Third, whether a given response rate is therapeutic or not is a matter of subjective judgment.

Regulatory approval expresses a social standard regarding whether an experimental treatment has a therapeutic value. This and the previous three challenges can be addressed by instead measuring therapeutic benefit in terms of the probability patients entering phase 1 trials will receive a therapy that eventually receives regulatory approval for their condition. What follows represents our analysis of a random sample of 1000 registered phase 1 cancer trials.

Methods:

The goal of our study was to determine the proportion of patients participating in phase 1 cancer trials that receive treatment regimens ultimately established as safe and effective. To do this, we defined “therapeutic regimens” as those involving a drug, indication, and approximate dose that received an FDA label.

Data Sources and Searches: To create a cohort of phase 1 trials, we searched ClinicalTrials.gov in January, 2018 for all phase 1 oncology trials that were registered as beginning between

January 1st, 2005 and December 31st, 2010, using the following search terms: cancer OR cancers

OR OR OR malignant OR malignancy OR malignancies OR tumor OR tumors OR tumour OR tumours OR neoplasm OR neoplasms OR metastatic OR lymphoma OR leukemia OR . We restricted our search to ClinicalTrials.gov, since our definition of

31

“therapeutic regimen” involves approval at a U.S. agency. Although FDA does not mandate prospective registration of phase 1 trials, many major pharmaceutical companies maintain policies of registering phase 1 trials, and many other entities require registration of phase 1 trials.(14–16)

Study Selection: To be included, phase 1 trials must have been interventional and enrolled at least 1 patient. All patients participating in the trial must have had a cancer indication. We excluded trials of non-drugs or therapies that did not have a clear product name (e.g. stem cells, surgery only), symptomatic treatment for cancer, trials open for enrollment at the time of extraction, trials without at least one study location in the United States as it is less likely they would submit to FDA for approval, and trials with a primary outcome that was not safety/tolerability or efficacy, and trials that enrolled 200 or more participants. Our timeframe

(2005-2010, inclusive) was chosen to allow for an adequate and reasonable period of time for regimens tested in our cohort of phase 1 trials to advance to regulatory approval. The end date for assessing approval status was January 1, 2019. As typical cancer drug development trajectories take approximately 7 years,(17) our timeframe ensured that promising drug development trajectories would have enough time such that any effective drug would have received approval. We applied a one-stage clustered random sampling method where 1000 phase 1 trials within our timeframe would be sampled from all eligible trials. We then extracted data from the 1000 selected trials.

32

Data Extraction and Quality Assessment: The following data were extracted from trial registration records: drug name, number of patients enrolled, monotherapy or combination therapy, cancer indication, trial sponsorship, whether the trial was single or mixed-indication, and biomarker enrichment. Patient enrollment was determined by the ‘actual enrollment’ number in the trial registry. The ‘estimated enrollment’ was used when actual enrollment information was unavailable. Whether the drug was eventually approved was determined through confirmation using the Drugs@FDA database. For trials where data regarding dosing or cancer indication was not directly available on registration records, publications for registered trials were searched on PubMed and Google Scholar using the NCT number and the title (including any alternate titles given) of the trial registry found on ClinicalTrials.gov. Publication searches were verified with two extractors. Patient data involving dosage and cancer indication were extracted from these publications. In the event where a publication was not found and information was not available on the trial registry, patient cancer indication was extracted as

“Mixed – non therapeutic” if the drug never received approval. Information on what constituted an active dose was extracted from drug labels in the Drugs@FDA database. For the purposes of our study, a drug needed to fall within the range of the recommended dose and the lowest dose reduction specified on the drug label to be considered “therapeutic.”

Cancer indications for patients in phase 1 trials were categorized according to broad National

Cancer Institute (NCI) categories.(18) For example, was classified as a “Brain and

CNS cancer.” Drugs were classified based on four categories: cytotoxic chemotherapy (e.g. intercalating agents, antimetabolites, alkylating agents), targeted therapy (e.g. tyrosine-kinase

33 inhibitors, angiogenesis inhibitors, histone deacetylase inhibitors), immunotherapy (e.g. checkpoint inhibitors, immunomodulators, cancer vaccines), and other (e.g. gene transfer, viral therapy, hormone therapy). For drug combinations, the drug class of the combination was categorized based on the investigational agent of the most novel drug within the combination. A drug was considered novel if it was not approved by the FDA prior to the trial start date.

Data Synthesis and Analysis: For our primary endpoint, we calculated the ratio of the total number of patients achieving the outcome of having received a therapeutic regimen to the total number patients who participated in phase 1 trials – the therapeutic proportion. Aside from the primary objective to determine therapeutic proportion among phase 1 trials, we also performed prespecified tests on the relationship between these proportions and whether participants were enrolled in in a) biomarker enriched phase 1 trials, b) monotherapy vs. combination therapy trials, c) industry vs. non-industry funded trials, d) drugs unapproved at trial outset vs. drugs that already had a label at trial outset; e) relationship between therapeutic proportion and type of cancer indication, f) single indication trials vs. mixed malignancy trials, g) class of drug. We performed descriptive analysis for trial and drug samples. We noted the number of trials in each subgroup. We also noted the total number of unique drugs as well as their approval rates, in addition to drug class and therapy type. We performed a sensitivity analysis for trials that had 12 years of follow up to allow for greater follow-up time for FDA approval. We also performed a sensitivity analysis to determine the proportion of patients who received an experimental drug that eventually received FDA approval without dosage or indication requirements. All hypothesis tests were pre-specified in a protocol. All departures from protocol are explained in

34 supplementary material. A meta regression using Poisson regression with the trial as random effect was performed to assess the association between therapeutic benefit and aforementioned trial characteristics. We defined p ≤ .05 as statistically significant; as inferential tests were exploratory only, we did not adjust for multiple hypothesis testing.

Results:

Our search study identified 3229 potential clinical trial registries, of which 1541 underwent review. In total, we included 1000 trial records that met eligibility criteria (Figure 1).

Trial Characteristics

We extracted data from 1000 phase 1 trials, comprising 32,582 patients. Within our sample, 922 unique drugs or drug combinations were tested in 55 different cancer indications. Table 1 summarizes the characteristics of our trial sample. The mean sample size of trials in our sample was 32.6 (SD 24). Most trials enrolled patients with solid tumours (78.6%) when compared to hematological cancers (18.7%) – a small proportion (2.7%) of trials enrolled patients regardless of malignancy type (Table 1). Trials were also roughly evenly distributed among the years 2005 to 2010.

Drug Characteristics

Of the drugs or drug combinations tested, 48 out of 922 (5.2%) therapies eventually led to FDA approval. There were 396 (42.9%) novel drugs, of which 31 (7.8%) led to approval. Notable novel drugs in our sample include ipilimumab, olaparib, crizotinib, and blinatumomab. Examples of non-novel drugs in phase 1 testing for new indications include bevacizumab, imatinib, 35 rituximab and trastuzumab. Targeted therapy comprised 576 drugs (62.5%), followed by cytotoxic therapy (158 – 17.1%), immunotherapy (129 – 14.0%), and other therapies (59 – 6.4%)

(Table 1).

Proportion of Participants Receiving Therapeutic Regimens:

The number of patients receiving regimens that were eventually approved by the FDA for their indication and at an appropriate dose was 376 (1.15% [95% CI, 1.04 – 1.28]). Disregarding indication, 1595 patients (4.9%) received previously unapproved drug that was eventually approved by the FDA. Table 2 presents this estimate by common tumor types.

Table 3 presents multivariable predictors of therapeutic proportion at the patient level. We did not observe a significant trend towards a greater proportion of patients receiving therapeutic regimens over the five-year span of our trial sample. Patients enrolled in biomarker-enriched trials had a significantly greater proportion receiving therapeutic regimens compared with non biomarker enriched trials (Odds Ratio = 4.9, [1.6-14.4], p = 0.0044). Additionally, patients receiving combination therapy are less likely to receive a therapeutic regimen compared to patients receiving monotherapy (Odds Ratio = 0.10, [0.013 – 0.85], p = 0.035), and patients enrolled in a trial testing for single indications had a higher proportion of therapeutic regimens than mixed indications (Odds Ratio = 3.4, [1.2 – 9.6], p = 0.019). Other subgroups, including novel vs. previously approved drugs, were not significantly different at the 95% level.

36

As a sensitivity analysis to allow for greater follow time for FDA approval, we re-performed our analysis using only the 267 trials for which we had 12 years of follow up. The proportion of patients receiving therapeutic regimens was 1.44% (129/8945).

Discussion:

Our study found that for every 87 patients that enrolled in phase 1 trials, one patient received a drug for an indication and at a similar dosage that ultimately received FDA approval. Number needed to treat values for response in typical approved anticancer drugs vary depending on the drug and cancer indication. Cisplatin, a cytotoxic chemotherapy, was shown to have an NNT of

15 in the treatment of non small cell .(19) Trastuzumab, a targeted therapy, has an

NNT of 7 to treat HER2+ breast cancer.(20) Nivolumab, an immunotherapy drug used to treat melanoma, has an NNT of 3.9.(21) On the assumption that NNTs for approved cancer drugs can be combined with proportions described above, between approximately 350 and 1350 patients need to participate in phase 1 cancer trials in order to have one positive outcome. The therapeutic justification of these proportions should be considered against rates of serious adverse events associated with phase 1 trial participation, estimated between 10 and 19%.(6,7,22) We did not observe clear relationships between proportions and sponsorship, approval status, drug class, and general cancer indication.

Our approach to estimating the therapeutic value of phase 1 cancer trial participation has advantages over other methods that have been used in the past. First, we did not impute clinical benefit based on surrogate measures of benefit. Second, our approach establishes a therapeutic

37 definition by benchmarking to expressed social standards judgments like FDA approval. Third, by using ClinicalTrials.gov rather than the published literature, our approach is less susceptible to being biased due to selective publication. Fourth, the concept of “therapy” implies a prospective judgement – namely that patients will derive greater benefit than burden by receiving an intervention. However, even effective “therapies” do not work in all patients. As such, a better way of measuring the therapeutic value of participating in phase 1 trials is determining the proportion of patients that receive treatments that are ultimately deemed to have a therapeutic risk/benefit balance.

Nevertheless, our findings should be interpreted in light of the following limitations. First, some drugs may require more than 8 years (ie end of 2010 to 2019) to mature into FDA approved treatments. However, over 70% of approved drugs were approved within 8 years and 95% were approved within 12 years (see Supplementary Figure 2). Moreover, our sensitivity analysis showed that the therapeutic proportion of trials with follow-up time of 12 years was 1.44%, which did not differ greatly from the overall therapeutic proportion of all trials. Our estimate that 7.8% of novel drugs tested in phase 1 trials receives FDA approval is consistent with other reports in the literature.(23) Second, our analytic approach makes the assumption that receiving a drug regimen that does not go on to receive FDA approval is nontherapeutic. However, some patients benefit in trials that lead to null pivotal results, and there may have been exceptional responders among patients receiving nontherapeutic regimens. We nevertheless think this limitation is balanced by the fact that not every patient receiving an approved cancer drug as per its label has a therapeutic response. Though our most strict definition of therapy yields a

38 proportion of 1.15%, we also provide a proportion when cancer indication is dropped as a requirement, 4.9%. Finally, given the length maturation periods, an intrinsic challenge of any study of drug development is having to rely on historic cohorts (in our case, drug development trajectories that were launched 8-14 years ago). It is possible that proportions will have changed with the emergence of new drug classes. Nevertheless, we did not observe striking trends in terms of time. Nor did newer drug classes, like immunotherapies, show dramatic differences in therapeutic proportions.

Our findings have implications for discussion of risk and benefit during informed consent and risk / benefit assessment during ethical review. For example, our findings call into question perceptions that receiving already approved drugs in phase 1 trials is associated with greater benefit than receiving drugs that are not yet approved. Also, by estimating the proportion of patients in phase 1 trials who access a drug that will be approved for their conditions, our study provides a basis for communicating risk and benefit to patients in phase 1 research. Our result can also be interpreted as an estimate of the productivity of the clinical trial enterprise, at the phase 1 level; if the predictive value of preclinical models is tapped more effectively in the future, therapeutic proportions in phase 1 will likely increase.

Conclusions

Cancer trials at the phase 1 stage show a relatively low probability of benefit objectively measured as obtaining FDA approval, as most experimental therapies in phase 1 trials would ultimately never have been considered beneficial for the patients that received them.

39

Figures

Figure 1: PRISMA diagram for search selection

40

Table 1: Trial and Drug Characteristics a) Trial Characteristics

Median Trial Size 26 Mean Trial Size 32.6 (Range: 1-180) Trial Start Year No. trials (%) 2005 116 (11.6) 2006 151 (15.1) 2007 150 (15.0) 2008 148 (14.8) 2009 182 (18.2) 2010 253 (25.3)

Sponsorship No. trials (%) Non-industry 479 (47.9) Industry 521 (52.1)

Type of Therapy No. trials (%) Monotherapy (1 drug) 488 (48.8) Combination (2 or more drugs) 512 (51.2) Approval status No. trials (%) Previously approved 557 (55.7) Novel treatment 443 (44.3)

Biomarker enrichment No. trials (%) Yes 90 (9.0) No 910 (91.0) Cancer indication No. trials (%)

41

Hematological 187 (18.7) Solid tumour 78.6 (78.6) Both 27 (2.7) Drug class No. trials (%) Cytotoxic 173 (17.3) Targeted 637 (63.7) Immunotherapy 131 (13.1) Other 59 (5.9) b) Drug characteristics

No. of unique drugs or drug combinations 922 No. of drugs eventually leading to approval (%) 48 (5.2) Type of Therapy No. drugs (%) Monotherapy (1 drug) 402 (43.6) Combination (2 or more drugs) 520 (56.4) Approval status No. drugs (%) Previously approved 526 (57.0) Novel treatment 396 (43.0) Drug class No. drugs (%) Cytotoxic 158 (17.1) Targeted 576 (62.5) Immunotherapy 129 (14.0) Other 59 (6.4) No. of indications tested No. trials (%) Mixed 448 (44.8) Single 552 (55.2)

42

Table 2: Patient characteristics and Multivariate predictors of therapeutic proportion

Variable No. of total patients No. of therapeutic Odds ratio p-value (proportion of patients (therapeutic (95% CI) total, %) proportion, %)

Trial Start Year 2005 3676 (11.3) 107 (2.9) Reference – 2006 5269 (16.2) 31 (0.6) 0.16 (0.04 – 0.66) 0.012 2007 5196 (15.9) 19 (0.4) 0.10 (0.01 – 0.65) 0.017 2008 4529 (13.9) 98 (2.2) 0.59 (0.13 – 2.6) 0.49 2009 5740 (17.6) 45 (0.8) 0.27 (0.05 – 1.4) 0.11 2010 8172 (25.1) 76 (0.9) 0.28 (0.04 – 1.8) 0.18

Sponsorship Non-industry 12509 (38.4) 87 (0.7) Reference – Industry 20073 (61.6) 289 (1.4) 3.2 (0.72 – 14.3) 0.12

No. of drugs Monotherapy (1 drug) 16835 (51.7) 301 (1.8) Reference – Combination (2 or more 15747 (48.3) 75 (0.5) 0.10 (0.01 – 0.84) 0.035 drugs) Approval status Previously approved 16542 (50.8) 132 (0.8) Reference – Novel treatment 16040 (49.2) 244 (1.5) 2.8 (0.48 – 16.0) 0.25

Biomarker enrichment No 29482 (90.5) 272 (0.9) Reference – Yes 3100 (9.5) 104 (3.4) 4.9 (1.6 – 14.4) 0.0044

Cancer indication Hematological 5797 (17.8) 98 (1.7) Reference – Solid tumour 26785 (82.2) 278 (1.0) 0.73 (0.26 – 2.1) 0.55

Drug class Cytotoxic 5270 (16.2) 90 (1.7) Reference – Targeted 22116 (67.9) 215 (1.0) 0.48 (0.12 – 2.0) 0.32 Immunotherapy 3661(11.2) 24 (0.7) 0.17 (0.020 – 1.5) 0.12 Other 1535 (4.7) 47 (3.1) 0.88 (0.09 – 9.0) 0.91

No. of indications tested Mixed (2 or more tested) 17413 (53.4) 105 (0.6) Reference – Single (1 tested) 15169 (46.6) 271 (1.8) 3.4 (1.2 – 9.6) 0.019

43

Table 3: Top 10 (n>1000) cancer indications by total and therapeutic patient number

Indication Number of Number of therapeutic patients patients (therapeutic proportion, %) (proportion of total, %) Colorectal cancer 3063 (9.4) 0 (0) Breast cancer 2498 (7.7) 31 (1.2) Lung cancer 2184 (6.7) 57 (2.6) Brain and CNS cancers 1731 (5.3) 30 (1.7) 1575 (4.8) 10 (0.6) 1491 (4.3) 47 (3.2) Melanoma 1379 (4.2) 113 (8.2) 1330 (4.1) 2 (0.2) Ovarian cancer 1205 (3.7) 64 (5.3) Lymphoma 1173 (3.6) 55 (4.7)

44

Appendix

S Table 1. 95% Confidence interval and rate estimate for therapeutic proportions

level rate estimate Lower Upper Industry 0.01315 0.005463 0.03163 sponsor Non industry 0.004088 0.000771 0.02168 Preapproval vs No 0.004396 0.000714 0.02706 approval Yes 0.01222 0.005386 0.02774 Combo 0.002354 0.000348 0.01595 Mono vs. combo Mono 0.02283 0.008394 0.06207 No 0.003329 0.000979 0.01132 biomarker Yes 0.01614 0.004676 0.05574 Solid tumor vs. Hematological 0.006253 0.001512 0.02587 hematological Solid tumor 0.008593 0.003206 0.02303 Cytotoxic 0.01407 0.002874 0.06891 Immunotherap 0.002437 0.000293 0.02027 y drug class Other 0.01236 0.001914 0.07987 Targeted 0.006812 0.002877 0.01613 therapy 2005 0.02483 0.005919 0.1042 2006 0.003912 0.001096 0.01396 2007 0.002413 0.000309 0.01886 year 2008 0.01471 0.003459 0.06259 2009 0.006589 0.001468 0.02958 2010 0.006829 0.001297 0.03596 Mixed 0.003965 0.001064 0.01478 indication single/mixed Single 0.01355 0.004445 0.04132 indication

45

S Fig 1. Distribution of trial size

S Fig 2. Survival curve of drugs given therapeutically

46

S Table 2. List of drugs that were given therapeutically

Trial Start Years to

Drug Indication Date approval

Alemtuzumab CLL 2009-07 0.2

Bevacizumab Brain and CNS cancer 2009-07 8.4

Carfilzomib 2005-09 6.9

Crizotinib NSCLC 2009-09 6.5

Cytarabine +

Daunorubicin AML 2006-09 10.9

Decitabine 2005-04 4.9

Enzatulamide Prostate cancer 2010-11 1.8

Eribulin Breast cancer 2006-06 4.5

Erlotinib NSCLC 2006-05 4.0

Everolimus Neuroendocrine tumor 2010-12 0.4

Ipilimumab Melanoma 2007-10 3.5

Irinotecan liposomal +

5FU + Leucovorin Pancreatic cancer 2006-01 9.8

47

Niraparib Ovarian cancer 2008-09 8.6

Breast cancer 2005-07 12.5

Olaparib Ovarian cancer 2005-07 10.0

Pazopanib Soft tissue 2007-09 4.6

Pemetrexed + Cisplatin NSCLC 2005-12 2.8

Ponatinib CML 2008-06 4.5

Cutaneous T Cell

Pralatrexate Lymphoma 2007-08 2.2

Ribociclib Breast cancer 2010-12 6.2

Topotecan + Cisplatin 2005-07 0.9

Vandetanib 2006-09-01 4.6

Vemurafenib Melanoma 2006-11 11.0

48

References

1. Critical role of phase I clinical trials in cancer treatment. American Society of Clinical Oncology. J Clin Oncol. 1997 Feb 1;15(2):853–9.

2. Weber JS, Levit LA, Adamson PC, Bruinooge S, Burris HA, Carducci MA, et al. American Society of Clinical Oncology Policy Statement Update: The Critical Role of Phase I Trials in Cancer Research and Treatment. J Clin Oncol. 2014 Dec 15;33(3):278–84.

3. Hospice Benefits and Phase I Cancer Trials | Annals of | American College of Physicians [Internet]. [cited 2019 Jul 4]. Available from: https://annals.org/aim/fullarticle/716073/hospice-benefits-phase-i-cancer-trials

4. Patients’ Expectations about Effects of Chemotherapy for Advanced Cancer | NEJM [Internet]. [cited 2019 Jul 4]. Available from: https://www.nejm.org/doi/full/10.1056/NEJMoa1204410

5. Miller FG, Joffe S. Benefit in phase 1 oncology trials: therapeutic misconception or reasonable treatment option? Clin Trials J Soc Clin Trials. 2008 Dec;5(6):617–23.

6. Horstmann E, McCabe MS, Grochow L, Yamamoto S, Rubinstein L, Budd T, et al. Risks and Benefits of Phase 1 Oncology Trials, 1991 through 2002. N Engl J Med. 2005 Mar 3;352(9):895–904.

7. Fukuda YK, Huang E, Finnigan S, Ivy SP, Rubinstein L, Takebe N. Risks and benefits of phase 1 oncology trials, 2001 through 2012. J Clin Oncol. 2014 May 20;32(15_suppl):2552–2552.

8. Agrawal M, Emanuel EJ. Ethics of Phase 1 Oncology Studies: Reexamining the Arguments and Data. JAMA. 2003 Aug 27;290(8):1075.

9. Druker BJ. Efficacy and Safety of a Specific Inhibitor of the BCR-ABL Tyrosine Kinase in Chronic Myeloid Leukemia — NEJM [Internet]. [cited 2017 Dec 17]. Available from: http://www.nejm.org/doi/full/10.1056/NEJM200104053441401#t=article

10. Druker BJ, Guilhot F, O’Brien SG, Gathmann I, Kantarjian H, Gattermann N, et al. Five- Year Follow-up of Patients Receiving Imatinib for Chronic Myeloid Leukemia. N Engl J Med. 2006;355(23):2408–17.

11. Hashim M, Pfeiffer BM, Bartsch R, Postma M, Heeg B. Do Surrogate Endpoints Better Correlate with Overall Survival in Studies That Did Not Allow for Crossover or Reported Balanced Postprogression Treatments? An Application in Advanced Non–Small Cell Lung Cancer. Value Health. 2018 Jan 1;21(1):9–17. 49

12. Buyse M, Thirion P, Carlson RW, Burzykowski T, Molenberghs G, Piedbois P. Relation between tumour response to first-line chemotherapy and survival in advanced colorectal cancer: a meta-analysis. Meta-Analysis Group in Cancer. Lancet Lond Engl. 2000 Jul 29;356(9227):373–8.

13. Tang PA, Bentzen SM, Chen EX, Siu LL. Surrogate end points for median overall survival in metastatic colorectal cancer: literature-based analysis from 39 randomized controlled trials of first-line chemotherapy. J Clin Oncol Off J Am Soc Clin Oncol. 2007 Oct 10;25(29):4562–8.

14. De Angelis C, Drazen JM, Frizelle FA, Haug C, Hoey J, Horton R, et al. Clinical trial registration: a statement from the International Committee of Medical Journal Editors. Ann Intern Med. 2004 Sep 21;141(6):477–8.

15. World Medical Association Declaration of Helsinki. Ethical principles for medical research involving human subjects. [Internet]. [cited 2019 Jul 4]. Available from: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2566407/

16. TCPS 2 :: The Interagency Advisory Panel on Research Ethics (PRE) [Internet]. [cited 2019 Jul 4]. Available from: http://www.pre.ethics.gc.ca/eng/policy-politique/initiatives/tcps2- eptc2/Default/

17. Jardim DL, Schwaederle M, Hong DS, Kurzrock R. An appraisal of drug development timelines in the Era of precision oncology. Oncotarget. 2016 Jul 13;7(33):53037–46.

18. A to Z List of Cancer Types - National Cancer Institute [Internet]. [cited 2019 Jul 4]. Available from: https://www.cancer.gov/types

19. Bonomi M, Pilotto S, Milella M, Massari F, Cingarlini S, Brunelli M, et al. Adjuvant chemotherapy for resected non-small-cell lung cancer: future perspectives for clinical research. J Exp Clin Cancer Res CR. 2011 Dec 29;30(1):115.

20. Graziano F, Rulli E, Biagioli E, Catalano V. Number needed to treat for pricing costly anticancer drugs: the example of regorafenib in metastatic colorectal cancer. Ann Oncol. 2016 Apr 1;27(4):747–8.

21. Shoushtari AN, Freeman ML, Betts KA, Gupte-Singh K, Du EX, Ritchings C, et al. Indirect treatment comparison of nivolumab versus placebo as an adjuvant therapy for resected melanoma. J Clin Oncol. 2018 May 20;36(15_suppl):9593–9593.

22. Roberts TG, Goulart BH, Squitieri L, Stallings SC, Halpern EF, Chabner BA, et al. Trends in the Risks and Benefits to Patients With Cancer Participating in Phase 1 Clinical Trials. JAMA. 2004 Nov 3;292(17):2130–40.

50

23. Hay M, Thomas DW, Craighead JL, Economides C, Rosenthal J. Clinical development success rates for investigational drugs. Nat Biotechnol. 2014 Jan 9;32:40–51.

51

Extended Methodology

Overview:

1) Searched for all cancer-related phase 1 clinical trials between January 1st, 2005 and

December 31st, 2010 on CT.gov

2) Screened for inclusion of trial registration records

3) Extracted from registration record: total number of patients enrolled, cancer indications,

drug, biomarker enrichment status, monotherapy or combination, sponsorship

4) Used databases such as Drugs@FDA, DrugBank, and NCI Thesaurus to find FDA label,

approval history, dosage recommendations, and drug class

Database selection:

We elected to use ClinicalTrials.gov (CT.gov) database to create our trial sample. CT.gov is a public database of clinical trials of a wide variety of indications run by the National Library of

Medicine and the National Institutes of Health, based in the United States. The database is also the largest clinical trial database in the world, housing trial records of over 300,000 trials in 230 different countries. Researchers can filter trials over a range of criteria, including, but not limited to: disease, phase of trial, location, whether the trial accepted healthy volunteers, and trial start date. Trial records from CT.gov contains information pertinent to our thesis project, such as the number of patients enrolled, the drug or drug combination studied, eligibility criteria of the trial, funder information, phase of trial, and trial start date. Each trial record has a unique identifier, the NCT number, which can be used to link to associated journal publications to look for more

52 specific patient information. We also restricted our search to CT.gov since we were interested in approval of drugs by the Food and Drug Administration of the United States and U.S. based trials were more likely to use this registry compared to others57.

Eligibility:

Inclusion criteria: a) listed on CT.gov; b) enrolling at least 1 patient; c) in a cancer indication; d) interventional; e) date range between January 1st, 2005 to December 31st, 2010; f) at least one

U.S. location.

Exclusion criteria: a) non-drug or no clear product name (e.g. stem cells, surgery only); b) symptomatic treatment for cancer; c) trial still recruiting or open for enrollment at time of extraction; d) primary outcome is NOT safety/tolerability or efficacy (e,g, prevention, symptom management); e) patient enrollment greater than 200.

Our eligibility criteria aimed to capture cancer trials at the phase 1 stage that were being tested for cancer treatments and on cancer patients; that is, drugs that have a prospective direct benefit.

Our timeframe was chosen with the intent of maximizing the number of trials to sample while still providing adequate time for the completion of drug development trajectories. In September

2004, the International Committee of Medical Journal Editors (ICMJE) declared that authors who wished to publish clinical research in its member journals must register their trial in a public registry58. Thus, the beginning of our chosen timeframe aligns with a significant increase in trial registration. The year 2010 was chosen as our cut-off to allow for 8 years of follow-up. As typical, drug trajectories in cancer take on average 7 years to develop59, our chosen timeframe

53 allows for 8 to 14 years of follow up, ensuring that promising development trajectories would have enough follow up time such that any effective drug receives approval. Trials must have had one U.S. location, which indicates interest in submitting the drug to the U.S. Food and Drug

Administration for approval.

Search criteria: cancer OR cancers OR carcinoma OR carcinomas OR malignant OR malignancy

OR malignancies OR tumor OR tumors OR tumour OR tumours OR neoplasm OR neoplasms OR metastatic OR lymphoma OR leukemia OR leukemias

The search criteria were created to capture as many cancer-related trials as possible. Since our criteria was rather broad, there were non-cancer related trials included as a result (e.g. benign tumors).

Sample size calculation from pilot study results:

A simple random sample of patients, while ideal for calculating sample size, would be inefficient for the purposes of the thesis, as we cannot conveniently access a list of patients who have enrolled in clinical research to create the sample frame. Thus, we rely on the clustered sampling method by randomly sampling trials as clusters and then incorporating all patients from the trials we sampled. One caveat to consider when using clustered sampling is that the method increases variance of the data by the factor known as the Design Effect60, which is expressed in the follow equation:

퐷푒푓푓 = 1 + (푚 − 1)휌

54 where m is the average cluster size (in our case, the number of participants per clinical trial) and

흆 is the Intracluster Correlation Coefficient (ICC), which is a variable that represents how strongly units within a cluster resemble one another. In the context of the thesis project, we can approximate the ICC to be to 1, as all patients within a phase 1 trial receive the same intervention.

As we did not know the average cluster size of phase 1 trials within our sample at the outset of the project, we conducted a pilot study for feasibility. In our pilot study of 100 trials, 85 of which were considered eligible, we found the average trial size to be approximately 37 patients. We also noted a therapeutic proportion of 53/3158 patients (1.68% [95% CI 1.29% – 2.19%]). There was a roughly even split in patient enrollment between different subgroups, such as industry sponsorship and monotherapy vs. combination.

The 95% confidence interval (1.29% - 2.19%) calculated from the pilot study was calculated as if the results were from a simple random sample. When accounting for the effect of a clustered sample, the variance, and thus confidence interval increases due to the Design Effect. As the confidence interval is determined by standard deviation60, the square root of variance, in order to maintain the same confidence interval, the sample size must be multiplied by the square root of the Design Effect to compensate.

We derive the following equation to calculate our sample size for the number of trials needed:

푆푎푚푝푙푒 푠푖푧푒 (# 표푓 푝푡푠) = 푁 푥 √1 + (푚 − 1)휌

Substituting N as 3158, m as 37 and 흆 as 1: 55

푆푎푚푝푙푒 푠푖푧푒 (# 표푓 푝푡푠) = 3158 푥 √37 = 19209

However, as we are calculating the sample size of clinical trials rather than patients, we divide the result of 19209 by 37, the average size of a clinical trial, we arrive at a sample size of 519 trials. To establish a similar confidence interval for subgroup analysis, such as industry funding, we doubled the number of trial samples and rounded evenly to a sample size of 1000.

Thus, the final sample size equation for the number of trials is:

푁√1 + (푚 − 1)휌

푆푎푚푝푙푒 푠푖푧푒 (# 표푓 푡푟푖푎푙푠) = 푚

Data Extraction: One thousand clinical trial records were extracted from CT.gov. The following data was extracted from the trial registry: drug name, number of patients enrolled, cancer indication, trial start date, trial sponsorship, whether the trial was biomarker enriched, therapy type (monotherapy vs. combination), and trial type (single indication or mixed malignancies).

For trials where data regarding dosing or cancer indication were not directly available, patient- specific information regarding cancer type and dosage were extracted from journal publications searched on Pubmed or Google Scholar. These searches were conducted using different permutations of NCT number of the registry, the title of the trial on CT.gov (including any alternative titles given), specific keywords (drug name, cancer indication), and names of the investigator. If a publication was not found, then the patient cancer indication was extracted as

“Mixed – non therapeutic” if the drug they received never received approval. Drug approval was determined through the Drugs@FDA database. Recommended dosage and indications for an

56 approved drug was confirmed from the FDA label, found on Drugs@FDA. For the purposes of the thesis project, a recommended dose was defined as a dose of the drug that fell between the range of the lowest recommended dose reduction and the FDA-recommended dose. The class of the drug was determined through various sources, such as academic publications pertaining to the drug, DrugBank, and the NCI Thesaurus.

Data curation: A sample of 50 drug approval determination extractions and drug class determination extractions were double extracted. We found the agreement rate to be 98% and

95%, respectively. All trial records that contained patients with unknown indications were double extracted to ensure that if the trial was published, information would be included in our dataset. There were alternate names for many cancer indications (e.g. bile duct cancer vs. ). Some trials also used more specific indication names compared to others.

Thus, indication names were standardized and categorized according to broad National Cancer

Institute (NCI) categories. Trials were determined to be biomarker enriched if the eligibility criteria specified that patients must have had a positive biomarker (for example, overexpression of a certain gene, having a certain mutation, or overexpression of certain receptor) to enter the trial. Drugs class was categorized into four broad groups: cytotoxic chemotherapy (e.g. intercalating agents, antimetabolites, alkylating agents), targeted therapy (e.g. tyrosine-kinase inhibitors, angiogenesis inhibitors, histone deacetylase inhibitors), immunotherapy (e.g. checkpoint inhibitors, immunomodulators, cancer vaccines), and other (e.g. gene transfer, viral therapy, hormone therapy). In therapies that tested multiple drugs at once, the entire combination

57 was classified based on the investigational agent or most novel drug within the combination. A drug was considered novel if it had not been approved by the FDA prior to the trial start date.

Analysis: The primary endpoint of the thesis project was to estimate the proportion of patients that would have received a therapeutic treatment over the total number of patients – the therapeutic proportion. Aside from the primary objective to determine the therapeutic proportion, we also performed tests on the relationship between the therapeutic proportion and different subgroups. The subgroups were: a) biomarker vs. non-biomarker enriched; b) monotherapy vs. combination therapy; c) industry vs. non-industry funded; d) novel drugs vs. already approved drugs at trial outset; e) solid tumor type vs. hematological; f) single indication vs. mixed malignancy trials; and g) drug class. A meta regression was performed using Poisson regression, using the trial as random effect.

58

Protocol

TITLE

What Proportion of Patients in Phase I Oncology Trials Receive Treatments that are Ultimately Vindicated as Safe and Effective?

INTRODUCTION

There is a lively debate as to whether participation in phase 1 trials can be presented to patients- and evaluated during ethical review- as a therapeutic option. For example, ASCO policy proclaims phase 1 study participation as a mode of therapy; others have defended this view. Others have offered more equivocal views {Joffe / Miller], or outright skepticism.

Debates about the therapeutic status of phase 1 trials have been reinvigorated by the ‘right to try’ movement. The ‘Right to Try’ laws propose to give terminally ill patients in the United States access to drugs that have passed the Phase I clinical trials. Such a law would circumvent the approval process of the Federal Drug Administration (FDA) and instead turns control over to the patient. Proponents of the Right to Try argue that patients who are terminally ill deserve to take charge of their own treatment, under the premise that even experimental drugs should yield some therapeutic value. Even absent right to try, many patients access candidates that have passed phase 1 testing via compassionate use pathways.

Risk/benefit balance for phase 1 trials have been characterized in several meta- analyses. These generally show… However, such studies provide an incomplete assessment of therapeutic value. For one, estimates of benefit rely on surrogate endpoints, which are unreliable predictors of patient-centered outcomes like survival or quality of life. Second, quantitative estimates of risk and benefit are difficult to convert into socially accepted judgments about whether a patient is receiving therapy. Finally, the concept of “therapy” is a prospective judgment. That is, an agent is defined as therapeutic at the outset of treatment, not purely as a result of outcomes. For example, few treatments work in 100% of patients. And yet, nonresponders who receive effective agents still can be said to have accessed an effective treatment, even though they are nonresponders.

The goal of this study is to estimate the proportion of patients who receive therapeutic treatment regimens in phase 1 trials, using the social benchmark of

59

“therapy” as FDA approval.. Specifically, we will determine the fraction of patients in phase 1 studies who receive drugs given under conditions that are ultimately reflected on an FDA approval label. In this way, we will estimate the “number needed to treat” for phase 1 trial participation (that is, the number of patients who need to be enrolled in a trial in order for one to receive a therapeutic regimen). These information can help inform debates about the therapeutic status of phase 1 trials. They can also be used to help inform patients of the probability of receiving regime that will ultimately prove effective if they enroll in phase 1 studies.

PRIMARY OBJECTIVE

- To determine the proportion of patients receiving an anticancer drug therapeutically during phase 1 trials, and the proportion this represents among cancer patients enrolled for phase 1 clinical trials of anticancer drugs where the drug or combination is not approved at the outset of the trial. We define a “therapeutic” administration of a drug in a clinical trial to be a case in which a patient receives a) an ultimately approved drug at the time of administration, and at b) the dosage reflected on the FDA label and c) the diagnostic / indication criteria reflected on the FDA label. This analysis will be extended to phase 2 trials as well.

SECONDARY OBJECTIVES

- Stratify according to subgroups such as indication type (solid tumor vs. hematological), combination vs. monotherapy, and approved monotherapy- indication with a combination vs no monotherapy-indication approval, and private only vs. some nonprivate funding, biomarker enrichment, drug class (immunotherapy, targeted therapy, cytotoxic chemotherapy, other), mixed vs. single indication trials

HYPOTHESIS

Greater than 50% of patients will not have received a therapeutic dose in clinical trials.

60

METHOD OVERVIEW

1) Search for all cancer-related clinical trials that start between January 1st, 2008 and December 31st, 2013 on ClinicalTrials.gov (Search terms: Cancer OR cancers OR carcinoma OR carcinomas OR malignant OR malignancy OR malignancies OR tumor OR tumors OR tumour OR tumours OR neoplasm OR neoplasms OR metastatic OR lymphoma OR leukemia OR leukemias). For phase 2 trials, we will run the exact same search and time range (so we can compare proportions) 2) Screen for inclusion of trial registration records. 3) Extract from registration record total N, phase, indication, drug, dosage, monotherapy or combination therapy, sponsorship, and when (if at all) the drug is approved from each clinical trial. a. In mixed malignancy trials, publications of the trial will be searched to determine more specific indications and dosages on a patient level. When full publication cannot be obtained, we will search for abstract on EMBASE or ASCO meetings 4) Use drug databases such as Drugs@FDA, Drugs.com, and Cancer.gov to find FDA label, approval history, and dosage recommendations of drugs or combinations tested to determine whether the regimen was therapeutic or not a. If monotherapy, note if drug was approved in any indication before the trial. If combination, determine whether the combination was approved before testing. Also determine whether individual drugs within the combination were approved for monotherapy before the trial. And determine if any of the drugs was approved for the trial indication before trial launch.

SAMPLE AND SAMPLING METHODS

Eligibility: Inclusion: a) listed on CT.gov; b) enrolling at least 1 patient; c) cancer indication; d) interventional trials e) date range: 01/01/2005 - 12/31/2010 Exclusion: a) nondrug or no clear product name (e.g. mesenchymal stem cells, surgery only); b) symptomatic treatment for cancer; c) trial open for enrollment at time of extraction; d) primary outcome is NOT safety/tolerability or efficacy

Sample size rationale / calculations: We will take a cluster sampling method to ensure that that our primary endpoint of determining the proportion of participants who receive a therapeutic regimen is powered to a 95% confidence interval of 1.3%-2.2% with an error margin of 1% (data from results of a pilot study). Thus, we aim to sample approximately 19000 patients from 500 clinical trials. 61

EXTRACTION

All extractions will be single-coded. Mixed malignancy studies where a publication could not be found will be double coded independently and blinded to the results of the first extraction. The number of patients to receive the drug therapeutically will be manually extracted from ClinicalTrials.gov. A treatment regimen will only be considered therapeutic if the drug is applied to a patient with a cancer diagnosis (defined based on 3 features: anatomical site, histology, and marker) in the same indication and at the same dose as it is eventually licensed by the FDA. Combination therapy trials will only be considered to be therapeutic if the combination receives FDA licensure. In cases of ambiguity, we will interpret indications broadly by anatomic site, in order to interpret the clinical trial record most charitably. Mixed malignancy trials where publications could not be found AND drug is not ultimately approved will be classified as non-therapeutic. Mixed malignancy trials where publications could be not found but the drug was ultimately approved will be considered potentially therapeutic – analyses involving these patients will be performed twice, once considering these patients as therapeutic and the other time as non-therapeutic. PRIMARY ANALYSIS

- Find the proportion of patients who received a therapeutic indication/dose over the total number of patients enrolled in clinical trials at various stages.

Patients who received therapeutic dose

All patients (N = )

SECONDARY ANALYSIS

- Descriptive statistics: Stratify according to subgroups such as indication type (solid tumor vs. hematological), combination vs. monotherapy, and approved 62

monotherapy-indication with a combination vs no monotherapy-indication approval, and private only vs. some nonprivate funding, biomarker enrichment, drug class (immunotherapy, targeted therapy, cytotoxic chemotherapy, other), mixed vs. single indication trials and calculate proportion of patients to receive a drug therapeutically. - Inferential statistics: Determine whether there a statistically significant difference between subgroups using meta regression

ETHICS

This protocol uses publicly accessible information and does not involve human or nonhuman animal subjects.

Amendments:

1) Sample size doubled from 500 to 1000 to compensate for subgroup confidence intervals such as monotherapy and industry sponsorship where patients were roughly evenly split. (May 8, 2018) a. (before data extraction) 2) Added subgroups for secondary analyses: a. For Biomarker enrichment (November 15 2017) i. (before data extraction) b. For Drug class (March 1, 2019), for mixed vs single indication, for hematological vs solid tumor instead of by specific indication type (March 7, 2019) i. (after data extraction and secondary analysis) 3) Determined use of Meta regression as the statistical test to evaluate if there were differences in subgroups (January 26, 2019) a. (after data extraction, before secondary analysis) 4) Added a double coder for all extractions that required publication searches instead of random sample of X% of trials being double coded (May 10, 2019) a. (after data extraction and secondary analysis)

63

Protocol timeline

Process Date CT.gov database search January 2018 Pilot study for feasibility February 2018 – March 2018 Data extraction May 2018 – July 2018 Primary and secondary analyses December 2018 – March 2019 Final review of drug approvals January 2019

64

Additional Results

Pilot results:

We first conducted a pilot study to determine feasibility with 100 trial records. After screening for eligibility, we found 85 trials eligible and extracted data regarding patient enrollment and whether they received a drug that was ultimately approved for their indication and at an appropriate dosage. We found the proportion to be 1.68% (53/3158). More extensive results can be found in the manuscript submitted for publication.

PRISMA diagram: the PRISMA diagram shows the data searching and curation process.

65

Chapter 3:

Discussion

Our manuscript estimated that 1.15% (95% Confidence Interval: 1.04 – 1.28) of patients that received an FDA-approved regimen in phase 1 oncology trials. As such, our data suggest that for every 87 patients enrolled in clinical research, only one will receive a beneficial treatment. This value is low when compared to the Number Needed to Treat (NNT) – a representation of the number of patients that needed to be treated with an intervention for one beneficial outcome, of already-approved cancer therapies. For example, cisplatin, a cytotoxic chemotherapy drug first approved in 1978 by the FDA, was shown to be effective in 1 in 15 non small cell lung cancer patients61. Trastuzumab (approved in 1998), a more modern, molecularly targeted agent, has an

NNT of 7 for the treatment of HER2+ breast cancer62. Nivolumab (approved in 2014), a successful immunotherapy, had an NNT of 3.9 in the treatment of melanoma63. On the assumption that NNTs for approved caner drugs may be combined with the therapeutic proportion we found, approximately 350-1350 patients would need to participate in phase 1 clinical trials to have one positive outcome. This chance of benefit also does not take into consideration that the majority of patients enrolled in phase 1 trials have advanced or metastatic disease. When accounting for the chance of benefit for patients with advanced pancreatic cancer, gemcitabine and platinum therapy yielded an NNT of between 33-3964. NNT for regaforenib, an expensive targeted agent for use in metastatic colorectal cancer, was 10.965. Therefore, operating

66 on the fact that patients enrolled in phase 1 research typically have advanced disease and are refractory towards standard lines of therapy, this might lower the chance of benefit even further.

The justification of benefit in phase 1 trials should be considered against rates of serious events or death. Previous analyses found the rate of Grade 4 serious adverse events to be between 10-

19%8–10, suggesting a trend of increased serious toxicity in more recent trials, despite the supposed lower toxicity of modern targeted therapies. Death rates have been estimated to be between 0.5% to 0.99%8–10. Going with the therapeutic proportion of 1 out of 87 patients receiving a treatment with clinical benefit, the likelihood of benefit is 8 to 16 times lower than the chance of encountering a Grade 4, or life-threatening side effect66, and is comparable to the probability of treatment-related death.

The properties of our phase 1 trial sample are fairly consistent with some previous analyses. A study by Hay et al. found the likelihood of approval for oncology drugs to be 6.7%48.

In our study, 7.8% of novel drugs led to approval. Despite having a proportion of patients with an unknown cancer indication in our study, the relative distribution of tumor types remains consistent when compared to a previous study by Roberts. et al9, sharing the top ten most frequent solid tumour types. The distribution of our drug classes is also consistent with more modern cancer phase I trialing11.

67

Advantages to our Approach

Our method of estimating the therapeutic value have several advantages over previous reviews.

1) FDA-approval establishes a robust definition of therapy

Previous analyses used ORR, a surrogate measure of benefit. This method of estimating phase 1 trial benefits has been used more or less consistently for over three decades in order to estimate patient benefit in phase 1 trials, yet still generate unclear discussions regarding benefit.

Our use of FDA-approval as a social benchmark of therapy gives a robust estimate of prospective benefit for phase 1 oncology trials. The FDA’s Center for Drug Evaluation and

Research (CDER), composed of research scientists and clinicians, reviews drug benefit and risk information from clinical trial data. Positive results from at least two Phase 3 trials are required for the FDA to grant a drug approval67. As such, when a drug is designated FDA approval for a particular cancer type, it has already gone through numerous clinical trials at the phase 1 and phase 2 stage showing positive signals, in addition to more extensive data from phase 3 trials validating that the drug as at least equal to or better than the current standard-of-care treatments.

Clinical trials at the phase 2 and phase 3 stage will have also generated data for either more validated surrogate endpoints for survival 68–71 (Progression-Free Survival/Time to Progression,

Disease-Free Survival) or Overall Survival itself.

One might argue that since the FDA had previously issued drug approvals based solely on ORR data, and still uses ORR data for certain approvals do this day, then ORR is an acceptable estimate of benefit. However, after discussion with the Oncologic Drugs Advisory

68

Committee in the 1980s, the FDA has largely switched to approvals based on direct evidence of benefit such as progression-free survival. While recent reviews have found that FDA still grants marketing approval based off of ORR even after the supposed switch to grant approvals based on non-ORR endpoints72,73, ORR is only used in specific cancer indications where it has substantial validation. The ORR statistic provided by previous meta-analyses was generated from a heterogenous mix of cancer indications as well as trial types, and therefore those ORR estimates do not properly address the question of benefit in these trials in a meaningful way.

Finally, ORR is only used to measure responses in solid tumor indications by RECIST criteria; it is not applicable in hematologic cancers. While response rates such as Major

Molecular Response and Complete Cytogenic Response are more causally linked to survival outcomes than ORR74, these response rates are still not ubiquitously applicable to any hematologic cancer. A lack of a universal measure of benefit between the two cancer types can be ameliorated with our method of FDA-approval as an overall measure of benefit.

2) We use a larger sample size

Previous analyses have used sample sizes anywhere between 2139 to 4608. Our study of 1000 trials give a narrower confidence interval and allows us to be more confident when using regression to measure differences between subgroups.

3) ClinicalTrials.gov is less susceptible to selective publication

Publication bias in clinical studies has been well documented. Studies yielding positive results tend to receive significantly more publication over null results; it is estimated that reports with

69 non-significant results are twice as likely to never be published75. As such, selective publication of only positive research results has led to overestimation in the therapeutic value of numerous drugs, such as in the field of antidepressants76,77.

The phenomenon has also been studied in oncology. A recent review of systematic reviews from top oncology journals by Hermann et al. found that 72% of systematic reviews did not use assessments of publication bias in their methodology, and that even reviews that did assess for publication bias did not assess it with comprehensive methodology. A re-analysis of previous reviews by the authors found publication bias to be even slightly higher than previously estimated78.

Our method of searching through public clinical trial records is less susceptible to publication bias. In 2004, the International Committee of Medical Journal Editors required registration of trials in a public registry before publishing58. Thus, any investigator with an interest to publish the results of their trial would have had to register before trial onset, regardless of the results of that trial. As such, our dataset contains a more representative ratio of trials with non-significant results. While we speculate that while there may be reasons against trial registry, for example – companies wishing to withhold trial data from competitors, these claims are not well-substantiated in the literature.

70

Limitations

Our study has several limitations that may require consideration.

1) More follow-up time could be necessary for promising drug trajectories to receive approval

As January 1, 2019 was the last day for assessing drug approval status, our cohort of trials had follow-up times varying between 8 to 14 years. Typical cancer drug development was found to take approximately 7 years59, with targeted agents requiring fewer years compared to cytotoxic agents. For our study, we found that within 8 years, approximately 70% of therapeutic approvals would have completed, and 100% of therapeutic approvals completed in 12.5 years. A sensitivity analysis of trials with 12 years of follow-up found a slightly higher therapeutic proportion of patients (1.15% vs. 1.44% - an absolute increase of 0.29%, which is a proportional

25% increase of the original statistic). While an increase of 0.29% is not substantial in the context of the absolute number of patients that benefit in phase 1 studies, it does suggest that a longer follow-up time may be required if our methodology were to be used in other samples where a 25% proportional increase would mean a larger absolute increase.

2) FDA-approval may provide an overly narrow definition of therapy

Our criteria requiring FDA approval, matched cancer indication, and dosage requirements could be considered too stringent. Alternative to FDA-approval, the National Comprehensive

Cancer Network (NCCN) guidelines could also be considered a socially acceptable benchmark for therapy, though there have been criticism that the majority of guidelines are based on lower- level evidence79,80. Off-label usage of cancer drugs is not uncommon, and it has been estimated

71 that 20-40% of cancer patients have received oncology drugs off-label81. Reasons of off-label use could be to adjust dosing or to be used in unapproved cancer indications. While prevalent, off- label usage of cancer drugs carries an enhanced risk of toxicity and are less effective than their use on-label.

To address concerns regarding an overly-strict definition of therapy, we provided an alternate proportion that disregards matching cancer indication. Our study found that 4.9% of patients received a novel drug that eventually received FDA approval, regardless of the patient’s cancer indication. As such, we can estimate that while 1.15% of patients receive a drug on-label, approximately 3.8% patients essentially receive a drug off-label (an approved drug but not for their cancer type) in phase 1 studies.

3) Historic cohorts may not be representative of current phase 1 trialing

As the field of oncology treatment advances, there may be changes in our therapeutic proportion as the types of therapies tested in cancer trials in the past may differ from the current state of phase 1 clinical trials. If this were true however, we could have expected to see some trends when proportion is compared against the year of trial onset; we did not. We also did not note any significant difference in therapeutic proportion between older drug classes such as cytotoxic chemotherapy, against immunotherapy, a newer therapy. Our drug cohort did capture many modern drugs that were approved recently, for example: ipilimumab, an immunotherapy approved for melanoma, olaparib and niraparib, two targeted therapies approved for breast cancer and ovarian cancer, and crizotinib, a targeted therapy for non-small cell lung cancer. An

72 analysis comparing whether the proportions of different trial characteristics change between our

2005-2010 cohort and a more modern sample could be important to judge whether our historic cohort is representative of modern trialing.

4) Interpretation

As with any descriptive analysis, normative implications do not flow inexorably from our findings. Some commentators might view the proportions described in this thesis as patently nontherapeutic. For example, one might consider a treatment futile if it has failed in the last 100 cases or if it fails to end a patient’s dependence on intensive medical care – both are likely applicable to phase 1 treatments82,83. Others might view the proportions more favorably and suggest our data reinforce the therapeutic claim, as clinical trials provide a subsequent mode of potential therapy for cancer patients who are refractory to all standard lines of treatment. Though we tried to circumvent this normative challenge by using FDA standards, FDA standards could be considered as overly stringent84 by some or too permissive by others85.

73

Ethical and Policy Implications

Are Phase 1 Trials Ethically Justifiable With a ~1% Chance of Benefit?

Numerous ethics documents such as the Declaration of Helsinki27, Belmont Report26, and Tri-

Council Policy Statement28 all espouse the principle of ‘beneficence’ in some form, which states that clinical research ought to be considerate of the overall welfare of its participants by ensuring a reasonable risk-benefit ratio. However, it is important to make the distinction between the research objectives of a trial and therapy. First and foremost, the main goal of clinical trials is to advance clinical research through the generation of new scientific or medical knowledge3. The design of clinical trials support this view: randomization, use of placebos, preplanned dosing schedules, and double-blind procedures all suggest that advancing a trial participant’s health is not the primary purpose30. Rather, clinical trials are mostly a scientific endeavor. As such, the duty of the trial investigators is not towards the health of individual participants, but instead to ensure that meaningful data can be collected. While medical benefit may not be the primary aim of clinical research, investigators – many who are also physicians, might still have an interest in promoting the health of trial participants.

Clinical research, especially in oncology, can be risky, and these risks must be balanced by the putative benefits in a clinical trial. Given the main purpose of determining safety and tolerability, phase 1 trials are adequately designed23. Investigators aim to expose patients to minimal risk by starting participants off with a dose that is only 10% of the lethal dose. Toxicity is evaluated at manageable steps through dose escalation. While subtherapeutic doses are less

74 likely to yield benefit, their main purpose is to minimize risk23. The reason for giving subtherapeutic doses at trial onset can be rationalized by the relatively low probability of developing a promising drug, coupled with the difficulty in testing even successful drugs in the correct cancer type. The dose-escalation methodology thus prevents patients from receiving high doses of potent anticancer drugs that may ultimately never be beneficial. In light of these considerations to minimize risk, a participant in phase 1 cancer trials still has a 10-19% chance of experiencing life-threatening toxicity8–10. They also have an approximately 1-in-100 chance of dying due receiving a lethal dose of an experimental drug8–10. Our study showed that 1 in 87 participants receive an FDA-approved therapy. From a purely quantitative perspective, subjecting participants to such an unfavorable risk/benefit ratio appears unreasonable.

Other forms of benefit need to be taken into consideration. I previously discussed the concept of the three forms of benefit identified by Nancy King: direct benefit, collateral benefit, and aspirational benefit33. Indeed, an analysis focusing on only on the possibility of a 1% direct benefit suggests that phase 1 cancer trials ought to be discontinued. However, phase 1 trials are an essential step in the generation of novel medical research; there is considerable aspirational benefit. While it is true that only around 5% of drugs receive approval in phase 1 trials, phase 1 trials are the starting point to help eliminate drugs that would be unsafe for the public. Patients enrolling in phase 1 trials have expressed an altruistic desire to help future patients. Phase 1 trials also provide collateral benefits to patients in the form of psychological comfort being in regular contact with physicians, feelings of being in control of their disease trajectory, and comfort knowing that they are helping future patients like themselves36. The significant aspirational and

75 collateral benefit generated by phase 1 trials, even taken together with the small likelihood of direct benefit, provide ethical justification for the conduct of phase 1 oncology trials in settings where risk managed properly. Even when operating under the principle of ‘beneficence’, a low chance of direct benefit does not disqualify the practice of phase 1 cancer trials.

While there may be some ethical justification on the principle of beneficence, one must also take into consideration the autonomy of clinical trial participants. Ethical policies26–28 promote informed-decision making by the participant, where pertinent information about clinical research such as potential risks and benefits must be appropriately presented and synthesized clearly by participants. As participants in cancer trials have been known to either mistake the research objective of clinical trial for personal therapy or overestimate the potential benefits86,87, adequate and cautious disclosure is especially important. Slevin et. al found that on average, cancer patients were willing to undergo intensive chemotherapy with severe side effects (severe nausea, vomiting, hair loss, frequent tiredness and weakness) for a 1% chance of a complete cure. If there was no chance of cure, then patients would accept a prolonged survival of 12 months. For a more mild treatment (slight nausea and vomiting, no hair loss, some tiredness and weakness), patients were willing to accept a 1% chance of cure or 3 months prolonged survival88.

Taking the results of our study into consideration, enrollment in phase 1 trials may fall short of patient expectations. While 1.15% of patients received an approved treatment, anticancer drugs, even when used in clinical settings, do not have a 100% success rate of cure. In addition, most cancer treatments are not curative and only prolong survival. Taking into consideration of survival benefit, a review of new anticancer drugs approved by the FDA and European

76

Medicines Agency (EMA) between 2003-2013 found that only 43% increased overall survival by 3 months or longer89. Thus, the estimated benefit of phase 1 cancer trials may not meet expectations of average cancer patients. Nevertheless, the results of our study could be used as a discussion point in the informed consent process to set baseline expectations of benefit. It is ultimately up to cancer patients want whether they wish to enroll in phase 1 trials after being informed of the chance of direct benefit.

In summary, cancer phase 1 trials are ethically justifiable even with little to no direct medical benefit; however, information regarding the chance of benefit ought to be disclosed to patients as various individuals are likely have different expectations of benefit when enrolling in clinical research.

Policy Implications

Whether phase 1 trials are considered therapeutic have implications in policy.

1) Reimbursement for Medicare

Medicare, the federal health insurance of the United States, has regulations on financial reimbursement of clinical trial costs. One pertinent requirement of Medicare is that the trial

“must not be designed exclusively to test toxicity or disease ” and have “therapeutic intent” 6. The regulation continues by stating that only meeting the requirements is insufficient; the primary objective of the trial should also be to test whether the intervention “potentially improves the participants’ health outcomes” 6. Medical groups such as ASCO support the basis that phase 1 trials offer prospective benefit, stating that “therapeutic intent is always present”5.

77

Nevertheless, phase 1 trials are primarily designed to evaluate toxicity and pharmacokinetics endpoints. While our study shows that they may provide a patient-acceptable chance of benefit, this benefit is minor in comparison to FDA-approved treatment options that would otherwise be used in the clinical setting. Clinical trials at the phase 1 stage likely do not meet the therapeutic requirements of Medicare reimbursement. Therefore, it would be unfair to use publicly-funded

Medicare resources, which is purposed to cover societal healthcare costs, to pursue research activities that do not lead to direct health benefits in patients.

2) Reimbursement for Hospice Care

Hospice care is the end-of-life care for patients with an estimated life expectancy of 6 months90.

It offers services that can greatly enhance comfort and quality of life, such as 24-hour care, completion of advanced directions, physical and occupational therapy, and bereavement support7. Medicare denies coverage of hospice care to patients with terminal illnesses that are also undergoing disease-modifying treatments. Although simultaneous enrollment in phase 1 trials and hospice care is a possible option, it is rarely a feasible option due to financial limitations. Nevertheless, cancer patients in phase 1 trials would likely benefit from hospice care services; many would be eligible: one cancer center found the median life expectancy of phase 1 participants to be 6.5 months91.

‘Justice’ is another ethical principle shared in some form by many ethics policies26–28. It involves a fair and equitable distributions of benefits to research participants. Our study finds that the probability and magnitude of therapy gleaned from phase 1 trials is likely to be well

78 below accepted cancer treatments in the clinical setting. As clinical trial participants take on additional risks and burdens when participating in trials, it would be unjust to systematically deny these patients the benefits of hospice care.

Summary

Our study found that 1 in 87 phase 1 oncology trial participants received an anticancer treatment that ultimately led to FDA-approval for their indication. Our use of FDA-approval as a therapeutic benchmark instead of the surrogate endpoint ORR gives a more robust estimate of the probability of benefit. Further follow-up time could be necessary to establish a more precise estimate of benefit. We find that cancer phase 1 trials are ethically justifiable based on risk/benefit ratio, but the chance of direct benefit ought to be disclosed to patients. Our study advocates the low therapeutic value of phase 1 trials in cancer and recommends policies to regard these trials as non-therapeutic.

79

Conclusion

This manuscript provides an overview on the purpose and methodology of phase 1 oncology trials, efforts to measure whether enrollment is therapeutic, as well as ethical and policy implications that hinge on therapeutic designation.

Phase 1 oncology clinical trials represent the transition of an anticancer therapy from bench to bedside, as it is often the first instance in which novel therapies are tested in human subjects. Furthermore, these human subjects are cancer patients rather than healthy volunteers.

Ethical challenges arise as it is uncertain whether cancer patients actually benefit from these trials, and whether participants truly understand the scientific, rather than therapeutic, intent of these trials. Organizations such as ASCO and NCI muddy the waters further by making forward claims of therapy without substantiated evidence. Previous analyses have measured benefit using the surrogate endpoint, ORR. Estimates using ORR are hindered by the fact that ORR has limited correlative value in only a few cancer indications – an ambiguity exacerbated by the heterogeneous sample cohorts used in most meta-analyses.

Our original research article submitted for publication utilized a novel and adaptable measurement of benefit using FDA-approval as a therapeutic standard. We found that 1.15% of patients received a therapeutic regimen in phase 1 cancer trials, and that patients in biomarker, monotherapy, and single-indication trials may have a higher likelihood of receiving an approved therapy. Our results provide a baseline estimate for the putative benefit of phase 1 trials overall.

Advantages of our method include the fact that FDA-approval is a rigorous, socially-accepted

80 benchmark for therapy and that we are unlikely to overestimate benefit from publication bias due to using trial registry data instead. However, our study is limited by a historic cohort and may require further years of follow-up.

Even with a low likelihood of direct benefit, phase 1 trials are still ethically justified, as they advance medical knowledge – by either bringing promising drugs to regulatory approval or extinguishing drugs that are unsafe for public use and also provide patients with psychological comfort. Even so, patients interested in trial enrolment ought to be informed of potential risks and benefits; we suggest that our value of 1.15% is used as a baseline to estimate the chance of benefit. We outline changes in policy for clinical trial reimbursement and eligibility for hospice care based on novel data from our study. When compared to previous analyses, our results corroborate several findings in the literature, such as the likelihood of oncology drug approval and the frequency of different cancer indications and drug classes tested in phase 1 trials. More importantly, our research supplements previous meta-analyses with a much-needed definition of therapy in phase 1 cancer trials. We provide a statistic that can be disclosed by investigators to cancer patients for the general probability of benefit of phase 1 trials overall.

Future work in the determination of phase 1 trial benefit could explore more in-depth in trials with monotherapy, biomarker, and single-indication trials. These signals ought to be further examined down the line in more specific trial cohorts. Additionally, as our study did not include phase 1 trials with large expansion cohorts, further analysis involving trials with these designs

92 are needed. For example, the considerably large KEYNOTE trial testing pembrolizumab enrolled hundreds of patients in expansion cohorts and eventually led to successful approval by 81 the FDA. Our methodology can be extended to study the therapeutic value of phase 2 cancer clinical trials where little academic discussion regarding its benefit has taken place; a project to do so is already under way. We can also apply our methodology more generally to in any area of clinical research where the direct benefits of research to its participants are unclear, using a socially-accepted benchmark of therapy to represent therapeutic value.

Cancer remains one of the leading causes of death in North America. Thus, patients’ access to cancer therapies is of utmost necessity. However, there are other venues of treatment with better therapeutic outcomes. It is therefore important to consider phase 1 trials as a primarily scientific endeavor, whose fruits of labor are intended to aid future patients and society as a whole. Defining them otherwise may be detrimental to the welfare cancer patients.

82

References

1. Goldberg RM, Wei L, Fernandez S. The Evolution of Clinical Trials in Oncology: Defining Who Benefits from New Drugs Using Innovative Study Designs. The Oncologist. 2017 Sep 1;22(9):1015–9.

2. Garrett-Mayer E, O’Connell N. Chapter 3 - The Evolution of Phase I Trials, Past, Present, and Future: A Biostatistical Perspective. In: Kummar S, Takimoto C, editors. Novel Designs of Early Phase Trials for Cancer Therapeutics [Internet]. Academic Press; 2018 [cited 2019 May 13]. p. 17–32. Available from: http://www.sciencedirect.com/science/article/pii/B9780128125120000038

3. Kimmelman J. Is Participation in Cancer Phase I Trials Really Therapeutic? J Clin Oncol Off J Am Soc Clin Oncol. 2017 Jan 10;35(2):135–8.

4. Dolly SO, Kalaitzaki E, Puglisi M, Stimpson S, Hanwell J, Fandos SS, et al. A study of motivations and expectations of patients seen in phase 1 oncology clinics. Cancer. 2016;122(22):3501–8.

5. Weber JS, Levit LA, Adamson PC, Bruinooge S, Burris HA, Carducci MA, et al. American Society of Clinical Oncology Policy Statement Update: The Critical Role of Phase I Trials in Cancer Research and Treatment. J Clin Oncol. 2014 Dec 15;33(3):278–84.

6. National Coverage Determination (NCD) for Routine Costs in Clinical Trials (310.1) [Internet]. [cited 2019 Jul 22]. Available from: https://www.cms.gov/medicare-coverage- database/details/ncd-details.aspx?NCDId=1&ncdver=2&fromdb=true

7. Hospice Benefits and Phase I Cancer Trials | Annals of Internal Medicine | American College of Physicians [Internet]. [cited 2019 Jul 4]. Available from: https://annals.org/aim/fullarticle/716073/hospice-benefits-phase-i-cancer-trials

8. Horstmann E, McCabe MS, Grochow L, Yamamoto S, Rubinstein L, Budd T, et al. Risks and Benefits of Phase 1 Oncology Trials, 1991 through 2002. N Engl J Med. 2005 Mar 3;352(9):895–904.

9. Roberts TG, Goulart BH, Squitieri L, Stallings SC, Halpern EF, Chabner BA, et al. Trends in the Risks and Benefits to Patients With Cancer Participating in Phase 1 Clinical Trials. JAMA. 2004 Nov 3;292(17):2130–40.

10. Fukuda YK, Huang E, Finnigan S, Ivy SP, Rubinstein L, Takebe N. Risks and benefits of phase 1 oncology trials, 2001 through 2012. J Clin Oncol. 2014 May 20;32(15_suppl):2552–2552. 83

11. Chakiba C, Grellety T, Bellera C, Italiano A. Encouraging Trends in Modern Phase 1 Oncology Trials. N Engl J Med. 2018 Jun 7;378(23):2242–3.

12. Hashim M, Pfeiffer BM, Bartsch R, Postma M, Heeg B. Do Surrogate Endpoints Better Correlate with Overall Survival in Studies That Did Not Allow for Crossover or Reported Balanced Postprogression Treatments? An Application in Advanced Non–Small Cell Lung Cancer. Value Health. 2018 Jan 1;21(1):9–17.

13. Buyse M, Thirion P, Carlson RW, Burzykowski T, Molenberghs G, Piedbois P. Relation between tumour response to first-line chemotherapy and survival in advanced colorectal cancer: a meta-analysis. Meta-Analysis Group in Cancer. Lancet Lond Engl. 2000 Jul 29;356(9227):373–8.

14. Tang PA, Bentzen SM, Chen EX, Siu LL. Surrogate end points for median overall survival in metastatic colorectal cancer: literature-based analysis from 39 randomized controlled trials of first-line chemotherapy. J Clin Oncol Off J Am Soc Clin Oncol. 2007 Oct 10;25(29):4562–8.

15. Ivy SP, Siu LL, Garrett-Mayer E, Rubinstein L. Approaches to Phase 1 Clinical Trial Design Focused on Safety, Efficiency and Selected Patient Populations: A Report from the Clinical Trial Design Task Force of the National Cancer Institute Investigational Drug Steering Committee. Clin Cancer Res Off J Am Assoc Cancer Res. 2010 Mar 15;16(6):1726–36.

16. Miller M. PHASE I CANCER TRIALS: A Collusion of Misunderstanding. Hastings Cent Rep. 2000;30(4):34–43.

17. Wages NA, Chiuzan C, Panageas KS. Design considerations for early-phase clinical trials of immune-oncology agents. J Immunother Cancer [Internet]. 2018 Aug 22 [cited 2019 May 14];6. Available from: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6103998/

18. Fox E, Curt GA, Balis FM. Clinical Trial Design for Target-Based Therapy. The Oncologist. 2002 Oct 1;7(5):401–9.

19. Decoster G, Stein G, Holdener EE. Responses and toxic deaths in Phase I clinical trials. Ann Oncol. 1990 Jan 1;1(3):175–81.

20. Paoletti X, Le Tourneau C, Verweij J, Siu LL, Seymour L, Postel-Vinay S, et al. Defining dose-limiting toxicity for phase 1 trials of molecularly targeted agents: Results of a DLT- TARGETT international survey. Eur J Cancer. 2014 Aug 1;50(12):2050–6.

84

21. Le Tourneau C, Stathis A, Vidal L, Moore MJ, Siu LL. Choice of Starting Dose for Molecularly Targeted Agents Evaluated in First-in-Human Phase I Cancer Clinical Trials. J Clin Oncol. 2010 Feb 1;28(8):1401–7.

22. Cook N, Hansen AR, Siu LL, Abdul Razak AR. Early phase clinical trials to identify optimal dosing and safety. Mol Oncol. 2015 May;9(5):997–1007.

23. Miller FG, Joffe S. Benefit in phase 1 oncology trials: therapeutic misconception or reasonable treatment option? Clin Trials J Soc Clin Trials. 2008 Dec;5(6):617–23.

24. Therasse P, Arbuck SG, Eisenhauer EA, Wanders J, Kaplan RS, Rubinstein L, et al. New Guidelines to Evaluate the Response to Treatment in Solid Tumors. JNCI J Natl Cancer Inst. 2000 Feb 2;92(3):205–16.

25. Clinical Practice Guidelines on Haematological Malignancies | ESMO [Internet]. [cited 2019 Jul 23]. Available from: https://www.esmo.org/Guidelines/Haematological- Malignancies

26. Department of Health, Education, and Welfare, National Commission for the Protection of Human Subjects of Biomedical and Behavioral Research. The Belmont Report. Ethical principles and guidelines for the protection of human subjects of research. J Am Coll Dent. 2014;81(3):4–13.

27. World Medical Association Declaration of Helsinki. Ethical principles for medical research involving human subjects. [Internet]. [cited 2019 Jul 4]. Available from: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2566407/

28. TCPS 2 :: The Interagency Advisory Panel on Research Ethics (PRE) [Internet]. [cited 2019 Jul 4]. Available from: http://www.pre.ethics.gc.ca/eng/policy-politique/initiatives/tcps2- eptc2/Default/

29. Druker BJ, Guilhot F, O’Brien SG, Gathmann I, Kantarjian H, Gattermann N, et al. Five- Year Follow-up of Patients Receiving Imatinib for Chronic Myeloid Leukemia. N Engl J Med. 2006;355(23):2408–17.

30. Appelbaum PS, Roth LH, Lidz C. The therapeutic misconception: Informed consent in psychiatric research. Int J Law . 1982 Jan 1;5(3):319–29.

31. Horng S, Grady C. Misunderstanding in clinical research: distinguishing therapeutic misconception, therapeutic misestimation, and therapeutic optimism. IRB. 2003 Feb;25(1):11–6.

85

32. Pentz RD, White M, Harvey RD, Farmer ZL, Liu Y, Lewis C, et al. Therapeutic Misconception, Misestimation and Optimism in Subjects Enrolled in Phase I Trials. Cancer. 2012 Sep 15;118(18):4571–8.

33. King NMP. Defining and Describing Benefit Appropriately in Clinical Trials. J Law Med Ethics. 2000;28(4):332–43.

34. Weijer C, Miller PB. When are research risks reasonable in relation to anticipated benefits? Nat Med. 2004 Jun;10(6):570.

35. Field MJ, Behrman RE, Children I of M (US) C on CRI. Defining, Interpreting, and Applying Concepts of Risk and Benefit in Clinical Research Involving Children [Internet]. National Academies Press (US); 2004 [cited 2019 Jul 25]. Available from: https://www.ncbi.nlm.nih.gov/books/NBK25555/

36. Agrawal M, Emanuel EJ. Ethics of Phase 1 Oncology Studies: Reexamining the Arguments and Data. JAMA. 2003 Aug 27;290(8):1075.

37. Burris HA, Moore MJ, Andersen J, Green MR, Rothenberg ML, Modiano MR, et al. Improvements in survival and clinical benefit with gemcitabine as first-line therapy for patients with advanced cancer: a randomized trial. J Clin Oncol Off J Am Soc Clin Oncol. 1997 Jun;15(6):2403–13.

38. Higby DJ, Wallace HJ, Albert DJ, Holland JF. Diaminodichloroplatinum: a phase I study showing responses in testicular and other tumors. Cancer. 1974 May;33(5):1219–1215.

39. Creemers GJ, Bolis G, Gore M, Scarfone G, Lacave AJ, Guastalla JP, et al. Topotecan, an active drug in the second-line treatment of epithelial ovarian cancer: results of a large European phase II study. J Clin Oncol Off J Am Soc Clin Oncol. 1996 Dec;14(12):3056– 61.

40. Nixon NA, Blais N, Ernst S, Kollmannsberger C, Bebb G, Butler M, et al. Current landscape of immunotherapy in the treatment of solid tumours, with future opportunities and challenges. Curr Oncol. 2018 Oct;25(5):e373–84.

41. Estey E, Hoth D, Simon R, Marsoni S, Leyland-Jones B, Wittes R. Therapeutic response in phase I trials of antineoplastic agents. Cancer Treat Rep. 1986 Sep;70(9):1105–15.

42. Penel N, Isambert N, Leblond P, Ferte C, Duhamel A, Bonneterre J. “Classical 3 + 3 design” versus “accelerated titration designs”: analysis of 270 phase 1 trials investigating anti-cancer agents. Invest New Drugs. 2009 Dec;27(6):552–6.

86

43. Theoret MR, Pai-Scherf LH, Chuk MK, Prowell TM, Balasubramaniam S, Kim T, et al. Expansion Cohorts in First-in-Human Solid Tumor Oncology Trials. Clin Cancer Res Off J Am Assoc Cancer Res. 2015 Oct 15;21(20):4545–51.

44. Kurzrock R, Benjamin RS. Risks and benefits of phase 1 oncology trials, revisited. N Engl J Med. 2005 Mar 3;352(9):930–2.

45. Villaruz LC, Socinski MA. The Clinical Viewpoint: Definitions, Limitations of RECIST, Practical Considerations of Measurement. Clin Cancer Res Off J Am Assoc Cancer Res. 2013 May 15;19(10):2629–36.

46. Llavero-Valero M, Guillén-Grima F, Zafon C, Galofré JC. The placebo effect in thyroid cancer: a meta-analysis. Eur J Endocrinol. 2016 Jan 6;174(4):465–72.

47. Chvetzoff G, Tannock IF. Placebo Effects in Oncology. JNCI J Natl Cancer Inst. 2003 Jan 1;95(1):19–29.

48. Hay M, Thomas DW, Craighead JL, Economides C, Rosenthal J. Clinical development success rates for investigational drugs. Nat Biotechnol. 2014 Jan 9;32:40–51.

49. Wong CH, Siah KW, Lo AW. Estimation of clinical trial success rates and related parameters. Biostat Oxf Engl. 2019 01;20(2):273–86.

50. FDA Approved Drugs in Oncology | CenterWatch [Internet]. [cited 2019 Jul 23]. Available from: https://www.centerwatch.com/drug-information/fda-approved-drugs/therapeutic- area/12/oncology

51. Schnipper LE, Davidson NE, Wollins DS, Blayney DW, Dicker AP, Ganz PA, et al. Updating the American Society of Clinical Oncology Value Framework: Revisions and Reflections in Response to Comments Received. J Clin Oncol. 2016 May 31;34(24):2925– 34.

52. ESMO-Magnitude of Clinical Benefit Scale version 1.1 | ESMO [Internet]. [cited 2019 Jul 23]. Available from: https://www.esmo.org/Guidelines/ESMO-MCBS/Articles/ESMO- Magnitude-of-Clinical-Benefit-Scale-version-1.1

53. Evaluation of Overall Response Rate and Progression-Free Survival as Potential Surrogate Endpoints for Overall Survival in Immunotherapy Trials. - PubMed - NCBI [Internet]. [cited 2019 Jun 24]. Available from: https://www.ncbi.nlm.nih.gov/pubmed/29326281

54. Vrankar M, Unk M. Immune RECIST Criteria and Symptomatic Pseudoprogression in Non-small Cell Lung Cancer Patients Treated with Immunotherapy. Radiol Oncol. 2018 Oct 18;52(4):365–9.

87

55. Lee JH, Long GV, Menzies AM, Lo S, Guminski A, Whitbourne K, et al. Association Between Circulating Tumor DNA and Pseudoprogression in Patients With Metastatic Melanoma Treated With Anti–Programmed Cell Death 1 Antibodies. JAMA Oncol. 2018 May 1;4(5):717.

56. Commissioner O of the. Clinical Trial Endpoints for the Approval of Cancer Drugs and Biologics [Internet]. U.S. Food and Drug Administration. 2019 [cited 2019 Jul 23]. Available from: http://www.fda.gov/regulatory-information/search-fda-guidance- documents/clinical-trial-endpoints-approval-cancer-drugs-and-biologics

57. ClinicalTrials.gov Background - ClinicalTrials.gov [Internet]. [cited 2019 Jul 23]. Available from: https://clinicaltrials.gov/ct2/about-site/background

58. De Angelis C, Drazen JM, Frizelle FA, Haug C, Hoey J, Horton R, et al. Clinical trial registration: a statement from the International Committee of Medical Journal Editors. Ann Intern Med. 2004 Sep 21;141(6):477–8.

59. Jardim DL, Schwaederle M, Hong DS, Kurzrock R. An appraisal of drug development timelines in the Era of precision oncology. Oncotarget. 2016 Jul 13;7(33):53037–46.

60. Rowe AK, Lama M, Onikpo F, Deming MS. Design effects and intraclass correlation coefficients from a health facility cluster survey in Benin. Int J Qual Health Care. 2002 Dec 1;14(6):521–3.

61. Bonomi M, Pilotto S, Milella M, Massari F, Cingarlini S, Brunelli M, et al. Adjuvant chemotherapy for resected non-small-cell lung cancer: future perspectives for clinical research. J Exp Clin Cancer Res CR. 2011 Dec 29;30(1):115.

62. Rodrigues MJ, Albiges-Sauvin L, Wassermann J, Cottu PH. Evaluating the risk–benefit ratio of adjuvant trastuzumab-based therapy for T1a,bN0M0 HER2-positive breast carcinomas. Ann Oncol. 2011 Nov 1;22(11):2530–2530.

63. Shoushtari AN, Freeman ML, Betts KA, Gupte-Singh K, Du EX, Ritchings C, et al. Indirect treatment comparison of nivolumab versus placebo as an adjuvant therapy for resected melanoma. J Clin Oncol. 2018 May 20;36(15_suppl):9593–9593.

64. Bria E, Milella M, Gelibter A, Cuppone F, Pino MS, Ruggeri EM, et al. Gemcitabine-based combinations for inoperable pancreatic cancer: have we made real progress? A meta- analysis of 20 phase 3 trials. Cancer. 2007 Aug 1;110(3):525–33.

65. Graziano F, Rulli E, Biagioli E, Catalano V. Number needed to treat for pricing costly anticancer drugs: the example of regorafenib in metastatic colorectal cancer. Ann Oncol. 2016 Apr 1;27(4):747–8.

88

66. Common Terminology Criteria for Adverse Events (CTCAE) | Protocol Development | CTEP [Internet]. [cited 2019 Jul 23]. Available from: https://ctep.cancer.gov/protocolDevelopment/electronic_applications/ctc.htm

67. Research C for DE and. Development & Approval Process (Drugs) [Internet]. FDA. 2019 [cited 2019 Jul 16]. Available from: http://www.fda.gov/drugs/development-approval- process-drugs

68. Pazdur R. Endpoints for Assessing Drug Activity in Clinical Trials. The Oncologist. 2008 Apr 1;13(Supplement 2):19–21.

69. Korn RL, Crowley JJ. Overview: Progression-Free Survival as an Endpoint in Clinical Trials with Solid Tumors. Clin Cancer Res Off J Am Assoc Cancer Res. 2013 May 15;19(10):2607–12.

70. Anagnostou V, Yarchoan M, Hansen AR, Wang H, Verde F, Sharon E, et al. Immuno- oncology Trial Endpoints: Capturing Clinically Meaningful Activity. Clin Cancer Res. 2017 Sep 1;23(17):4959–69.

71. McKee AE, Farrell AT, Pazdur R, Woodcock J. The Role of the U.S. Food and Drug Administration Review Process: Clinical Trial Endpoints in Oncology. The Oncologist. 2010 Mar 1;15(Supplement 1):13–8.

72. Brooks N, Campone M, Paddock S, Shortenhaus S, Grainger D, Zummo J, et al. Approving cancer treatments based on endpoints other than overall survival: an analysis of historical data using the PACE Continuous Innovation IndicatorsTM (CII). Drugs Context [Internet]. 2017 Nov 15 [cited 2019 Jul 16];6. Available from: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5699106/

73. Chen EY, Joshi SK, Prasad V. FDA acceptance of surrogate endpoints in later lines of therapy. J Clin Oncol. 2018 May 20;36(15_suppl):6517–6517.

74. Oriana C, Martin H, Toby P, Chris C, Ruth G, Claudius R, et al. Complete Cytogenetic Response and Major Molecular Response as Surrogate Outcomes for Overall Survival in First-Line Treatment of Chronic Myelogenous Leukemia: A Case Study for Technology Appraisal on the Basis of Surrogate Outcomes Evidence. Value Health. 2013 Sep 1;16(6):1081–90.

75. Shields PG. Publication Bias Is a Scientific Problem with Adverse Ethical Outcomes: The Case for a Section for Null Results. Cancer Epidemiol Prev Biomark. 2000 Aug 1;9(8):771–2.

89

76. Eyding D, Lelgemann M, Grouven U, Härter M, Kromp M, Kaiser T, et al. Reboxetine for acute treatment of major depression: systematic review and meta-analysis of published and unpublished placebo and selective serotonin reuptake inhibitor controlled trials. BMJ. 2010 Oct 12;341:c4737.

77. Turner EH, Matthews AM, Linardatos E, Tell RA, Rosenthal R. Selective Publication of Antidepressant Trials and Its Influence on Apparent Efficacy [Internet]. http://dx.doi.org/10.1056/NEJMsa065779. 2009 [cited 2019 Jul 17]. Available from: https://www.nejm.org/doi/10.1056/NEJMsa065779?url_ver=Z39.88- 2003&rfr_id=ori%3Arid%3Acrossref.org&rfr_dat=cr_pub%3Dwww.ncbi.nlm.nih.gov

78. Herrmann D, Sinnett P, Holmes J, Khan S, Koller C, Vassar M. Statistical controversies in clinical research: publication bias evaluations are not routinely conducted in clinical oncology systematic reviews. Ann Oncol Off J Eur Soc Med Oncol. 2017 01;28(5):931–7.

79. Poonacha TK, Go RS. Level of Scientific Evidence Underlying Recommendations Arising From the National Comprehensive Cancer Network Clinical Practice Guidelines. J Clin Oncol. 2010 Dec 13;29(2):186–91.

80. Kusuma B, Go RS. Level of Scientific Evidence Underlying Recommendations Arising From the National Comprehensive Cancer Network Clinical Practice Guidelines for Hematologic Malignancies. Blood. 2011 Nov 18;118(21):509–509.

81. Saiyed MM, Ong PS, Chew L. Off-label drug use in oncology: a systematic review of literature. J Clin Pharm Ther. 2017;42(3):251–8.

82. Helft PR, Siegler M, Lantos J. The Rise and Fall of the Futility Movement. N Engl J Med. 2000 Jul 27;343(4):293–6.

83. Schneiderman LJ. Medical Futility: Its Meaning and Ethical Implications. Ann Intern Med. 1990 Jun 15;112(12):949.

84. Close E, Willmott L, White BP. Charlie Gard: in defence of the law. J Med Ethics. 2018 Jul 1;44(7):476–80.

85. Kim C, Prasad V. Strength of Validation for Surrogate End Points Used in the US Food and Drug Administration’s Approval of Oncology Drugs. Mayo Clin Proc. 2016 Jun 1;91(6):713–25.

86. Kass N, Taylor H, Fogarty L, Sugarman J, Goodman SN, Goodwin-Landher A, et al. Purpose and Benefits of Early Phase Cancer Trials: What Do Oncologists Say? What Do Patients Hear? J Empir Res Hum Res Ethics JERHRE. 2008 Sep;3(3):57–68.

90

87. McGrath‐Lone L, Ward H, Schoenborn C, Day S. The effects of cancer research participation on patient experience: a mixed‐methods analysis. Eur J Cancer Care (Engl). 2016 Nov;25(6):1056–64.

88. Slevin ML, Stubbs L, Plant HJ, Wilson P, Gregory WM, Armes PJ, et al. Attitudes to chemotherapy: comparing views of patients with cancer with those of doctors, nurses, and general public. Br Med J. 1990 Jun 2;300(6737):1458–60.

89. Salas-Vega S, Iliopoulos O, Mossialos E. Assessment of Overall Survival, Quality of Life, and Safety Benefits Associated With New Cancer . JAMA Oncol. 2017 Mar 1;3(3):382–90.

90. What Are and Hospice Care? [Internet]. National Institute on Aging. [cited 2019 Jul 22]. Available from: https://www.nia.nih.gov/health/what-are-palliative-care-and- hospice-care

91. Janisch L, Mick R, Schilsky RL, Vogelzang NJ, O’Brien S, Kuf M, et al. Prognostic factors for survival in patients treated in phase I clinical trials. Cancer. 1994;74(7):1965–73.

92. Leighl NB, Hellmann MD, Hui R, Carcereny E, Felip E, Ahn M-J, et al. Pembrolizumab in patients with advanced non-small-cell lung cancer (KEYNOTE-001): 3-year results from an open-label, phase 1 study. Lancet Respir Med. 2019 Apr;7(4):347–57.

93. Schwaederle M, Zhao M, Lee JJ, Lazar V, Leyland-Jones B, Schilsky RL, et al. Association of Biomarker-Based Treatment Strategies With Response Rates and Progression-Free Survival in Refractory Malignant Neoplasms: A Meta-analysis. JAMA Oncol. 2016 Nov 1;2(11):1452–9.

94. Oxnard GR, Wilcox KH, Gonen M, Polotsky M, Hirsch BR, Schwartz LH. Response Rate as a Regulatory End Point in Single-Arm Studies of Advanced Solid Tumors. JAMA Oncol. 2016 Jun 1;2(6):772–9.

91