Christopher S. Coffey Professor, Department of Biostatistics Director, Clinical Trials Statistical and Data Management Center May 30, 2014 STUDY DESIGN FOR EARLY PHASE CLINICAL TRIALS

In this webinar, we will: I.Discuss the importance of adequate study planning for small clinical trials II.Describe some analytical approaches that have merit with small clinical trials III.Describe several proposed designs for small clinical trials OUTLINE 2

 This webinar is being recorded.  In accordance with our open access mission, we will be posting a video and the slides to our website and they will be publically available  You can talk to us using the “chat” function, and we will speak our responses  Also, please use audience response when prompted NOTICE OF RECORDING 3

Before addressing some possible designs of interest, it is useful to review some key recommendations from the Executive Summary in the National Academy of Sciences document. OVERVIEW 4

1)Define the research question 2)Tailor the design 3)Clarify methods when reporting trial results 4)Perform corroborative statistical analysis 5)Exercise caution in interpretation 6)More research on alternative designs is needed OVERVIEW 5 Key Recommendations from the National Academy of Sciences Document – Small Clinical Trials, Issues and Challenges:

1)Define the research question 2)Tailor the design 3)Clarify methods for reporting trial results 4)Perform corroborative statistical analysis 5)Exercise caution in interpretation 6)More research on alternative designs is needed OVERVIEW 6 Key Recommendations from the National Academy of Sciences Document – Small Clinical Trials, Issues and Challenges: So, why is this any different from other trials?

Three basic requirements for any clinical trial: 1)Trial should examine an important research question 2)Trial should use a rigorous methodology that can answer the question of interest 3)Trial must be based on ethical considerations and assure that risks to subjects are minimized OVERVIEW 7

These three requirements should also apply to all early phase clinical trials. However, due to size limitations designing early phase clinical trials can be a formidable challenge. As a consequence, the importance of adequate study planning is magnified in small clinical trials. OVERVIEW 8

 Late phase RCT: Large sample sizes Many standard designs  Small populations Efficiency is critical Use models (but pre-specify) Carefully evaluate alternative designs OVERVIEW 9

Designs of interest in early phase clinical trials:  3+3 Design  Continual Reassessment Method  Repeated measures design  Crossover design  N-of-1 design  Futility design  Ranking/Selection design  Adaptive designs EARLY PHASE DESIGNS 10

DOSE FINDING maximum tolerated dose Very early learning phase designs may seek to determine the maximum tolerated dose (MTD). Accurate determination of the MTD is critical since it will likely be used as the maximum dose in future clinical development. 11  If dose too low, a potentially useful drug could be missed  If dose too high, participants in future studies could be put at risk

02 or more Count Events Treat 3 patients at dose Start at the lowest reasonable dose Increase dose to next level Treat 3 additional patients at dose Count Events Decrease dose or stop and select lower dose 1 01 or more FIXED DOSE FINDING 12 Most well known approach is 3+3 method, developed for oncology settings.

Strengths of Conventional 3+3 Designs:  Simple and intuitive algorithm  Easy to implement and monitor – requires no computer program  Familiar to many clinicians FIXED DOSE FINDING 13 However, the method has been criticized for treating many patients at low, ineffective doses and not producing a good estimate of the MTD.

FIXED DOSE FINDING 14 For example, with a true 30% chance of a toxicity, there is still a 50% chance of “stepping up” to the next dose. Hence, unsafe doses may be advanced to future trials.

 Originated as a Bayesian method for phase I cancer trials of cytotoxic agents.  Assumes a particular model (such as logistic function), and probabilities of both efficacy and toxicity increase with increasing dose  Assignment of doses converges to the MTD. ADAPTIVE DOSE FINDING 15 Adaptive dose finding methods offer more efficient ways to learn about dose response. Most common approach is Continual Reassessment Method [CRM - See Garrett-Mayer (Statistics in Medicine, 2006) for an excellent tutorial].

Steps for implementing CRM: 1)Begin with assumed a priori dose-toxicity curve and a chosen target toxicity rate 2)Assign first subject(s) dose most likely to be associated with target toxicity level 3)Updated dose-toxicity curve is refit (shifted slightly up or down) depending on whether or not first subject(s) experienced a DLT 4)Next subject assigned dose closest to target toxicity level based on updated curve 5)Continue until some pre-defined stopping criteria are met ADAPTIVE DOSE FINDING 16

For example, consider the following curve: If target level of toxicity is 10%, then dose level 5 would be the optimal starting dose. ADAPTIVE DOSE FINDING 17

An example of how the CRM might work: ADAPTIVE DOSE FINDING 18

An example of how the CRM might work: Final Dose Thanks to George Howard (University of Alabama at Birmingham) for Example ADAPTIVE DOSE FINDING 19

Strengths of CRM:  “Learns” from information gained at early time points in the study – all participants studied contribute to the estimated dose.  Generally more efficient/safer than 3+3 design Can more accurately estimate the MTD as compared to standard 3+3 designs More likely to treat participants at doses around the MTD Less likely to treat participants at ineffective doses Less likely to treat participants at toxic doses – tends to incur fewer dose-limiting toxicities. ADAPTIVE DOSE FINDING 20

Drawbacks of CRM:  Implementation requires a substantial collaboration between the investigator and statistician  Mathematical and statistical complexities make it difficult for many clinical investigators to understand.  Properties must be assessed via simulation.  Safety concern with original CRM: Large dose escalations can occur early based on limited information. ADAPTIVE DOSE FINDING 21

Several modified CRM approaches have been developed to address these concerns:  Always start at lowest dose level under consideration  Enroll 2-3 patients in each cohort  Any given dose escalation cannot increase by more than one level. ADAPTIVE DOSE FINDING 22

Multiple observations or response variables are obtained for each subject. - Repeated measurements over time (longitudinal) - Multiple measurements on same subject Allows both between-subject and within-subject comparisons. Can reduce the required sample size needed to obtain a specific target power. REPEATED MEASURES DESIGNS 23

Suppose you are measuring over time:  STANDARD: Final value – Baseline value  BETTER: Final value, with baseline value as a covariate  STILL BETTER: Longitudinal Differentiate “through” vs. “at” Think about variance/covariance structure Think how you want to model time REPEATED MEASURES DESIGNS 24

Each subject exposed to all treatments - Order of treatments randomized - First may show better (or worse) effect Prognostic factors balanced – self vs. self Required sample size reduced considerably due to self vs. self comparisons Each participant receives the active treatment at some point during the study CROSSOVER DESIGN 25

Half look like this: Period 1 Active Period 2 Pbo Period 1 Pbo Period 2 Active CROSSOVER DESIGN 26

Disadvantages:  Disease needs to be long-term  Treatment must be taken regularly over time  Relevant outcomes must occur and be measured over time  Not relevant for acute treatments  Concerns due to a ‘carryover effect’ CROSSOVER DESIGN 27

N-OF-1 DESIGN Special case of a crossover/repeated measures design, where a single subject undergoes treatment for several pairs of periods. For each pair:  Subject receives experimental treatment for one part of each pair  Subject receives alternative treatment for other pair  Order of two treatments within each pair is randomized 28

The final outcome of the trial is a determination about the best treatment for the particular subject under study. Most feasible for treatments with rapid onset that stop acting soon after discontinuation. Results of a series of N-of-1 trials may be combined using meta-analysis. N-OF-1 DESIGN 29

Selection (ranking) designs compare parameters of multiple (k) study populations. Generally require smaller sample sizes than trials designed to estimate and test treatment effects. Selection designs can be used to:  Select the treatment with the best response out of k potential treatments  Rank treatments in order of preference  Rule out poor treatments for further study (Helpful with ‘pipeline’ problem) 30 SELECTION DESIGNS

FUTILITY DESIGN A futility design is a screening tool to identify agents that should not be candidates for phase III trials while minimizing costs/sample size. The main intent is to identify ineffective treatments, not to establish efficacy. To use a futility design, a research must define an effect of interest. If study shows unlikely for future trials to obtain desired effect, it is deemed ‘futile’ to conduct future trials. 31

For example, suppose a 10% increase in favorable response rates is clinically meaningful. A futility design would assess the following hypothesis: FUTILITY DESIGN 32 H 0 : Treatment improves outcome by at least 10% compared to placebo (p T – p P ≥ 0.10) versus H 1 : Treatment does not improve outcome by at least 10% compared to placebo (p T – p P < 0.10) – futile to consider in a phase III trial

Statistical Hypotheses: Null Hypothesis (H 0 ) Alternative Hypothesis (H A ) Implication of Rejecting H 0 Usual Design μ T = μ P μ T ≠ μ P New Treatment is Effective (Harmful) Futility Design μ T – μ P ≥ 0μ T – μ P < 0 New Treatment is Futile FUTILITY DESIGN 33

Type I and Type II Errors: Type I Error (α) Type II Error (β) Usual Design Ineffective Therapy is Effective Effective Therapy is Ineffective Futility Design Effective Therapy is Ineffective Ineffective Therapy is Effective FUTILITY DESIGN 34

Negative predictive values are high Pr(Not Effective | Futile) Positive predictive values are not so high Pr(Effective | Not Futile) Thus, futility designs are good at identifying ineffective agents, but not good at identifying effective agents. However, this is an improvement over running underpowered efficacy trials in phase II or conducting phase III trials as the first rigorous test of efficacy for a new treatment. FUTILITY DESIGN 35

ADAPTIVE DESIGNS There may be limited information to guide initial choices for the design of a study. Since more knowledge will accrue as the study progresses, adaptive designs allow these elements to be reviewed during the trial. adaptive design An adaptive design allows for changing or modifying the characteristics of a trial based on cumulative information. 36

ADAPTIVE DESIGNS PhRMA Working Group on Adaptive Designs (2006): “By adaptive design we refer to a clinical study design that uses accumulating data to modify aspects of the study as it continues, without undermining the validity and integrity of the trial.…changes are made by design, and not on an ad hoc basis…not a remedy for inadequate planning.” 37 FDA Draft Guidance Document (2010): “… a study that includes a prospectively planned opportunity for modification of one or more specified aspects of the study design and hypotheses based on analysis of data (usually interim data) from subjects in the study.” “Adaptive By Design”

Source: Kairalla JA, Coffey CS, Thomann MK and Muller KE (2012). Adaptive Trial Designs: A Review of Barriers and Opportunities. Trials, 13: 145. Infinite number of adaptive design possibilities: 38 ADAPTIVE DESIGNS

39 GROUP SEQUENTIAL DESIGNS Given this definition, group sequential designs (that allow early termination for efficacy or futility) are one of the most commonly used ADs in clinical trials. Several approaches have been proposed to adjust for repeated testing and control overall type I error rate. It is important to recognize that many of these techniques are based on large sample theory. For small clinical trials, the distributions used for these calculations become increasingly inaccurate and may lead to poor performance of these ‘standard’ techniques.

40 ADAPTIVE SEAMLESS DESIGNS seamless design A seamless design combines objectives traditionally addressed in separate trials into a single trial. adaptive seamless design An adaptive seamless design combines trials and uses data from patients enrolled before and after the adaptation for the final analysis. Most interest to date has been with a seamless transition between phase IIb (learning) and III (confirming). However, there are also opportunities for seamless designs in early development (phase I/IIa).

41 Source: Levy et al, 2006 ADAPTIVE SEAMLESS DESIGNS

42 ADAPTIVE SEAMLESS DESIGNS Adaptive seamless designs have the potential to improve the drug development process by reducing the timelines for approval. However, analysis requires specialized methods to correct for bias introduced because data from the first stage are used for both decision making and final analysis. Hence, extra planning is necessary when implementing an adaptive seamless design protocol. The potential benefits should be carefully weighed against the challenges of such designs.

43 It is important to be clear about what an adaptive design can and cannot do in the rare disease setting. An AD cannot “change the answer” regarding the effectiveness of a particular treatment, but can increase the efficiency in finding an answer. An AD design cannot make a drug more effective. Rather, one of the biggest benefits of an AD is ability to identify ineffective treatments in a more timely manner. ADAPTIVE DESIGNS

Increased consideration of AD’s has put much greater emphasis on infrastructure building efforts needed to help further advance the use of AD’s. One example is creation of NINDS-funded Network of Excellence in Neuroscience Clinical Trials (NeuroNEXT). Development of these types of complex infrastructures dramatically increases the feasibility for using more novel trial designs – including AD’s. ADAPTIVE DESIGNS 44

An appropriate study design has sufficient sample size, adequate power, and proper control of bias to allow a meaningful interpretation of the results. Although early phase trials pose important limitations, the above issues cannot be ignored. One of the objectives of this course is to give researchers the tools they need to successfully design these types of trials. 45 SUMMARY

 Please consider going to the following website to evaluate this webinar: EVALUATIONS 46