Adaptive Designs: The Swiss Army Knife Among Clinical Trial Designs? Frank Bretz (Novartis) U Penn – April 13, 2016 Acknowledgment: Willi Maurer, Paul Gallo (Novartis) Swiss Army Knife Wiktionary: * A tool that has many functions, one for every perceivable need. Wikipedia: The term "Swiss Army knife" has entered popular culture as a metaphor for usefulness and adaptability. 2 | U Penn | Frank Bretz | April 13, 2016 *Trade Registry Number: CHE-105.977.463 Outline Adaptive designs over the years Role of adaptive designs in clinical trials Outlook 3 | U Penn | Frank Bretz | April 13, 2016 Outline Adaptive designs over the years • Disclaimer: In what follows, this is a (biased?) selection of references and not meant to be exhaustive Role of adaptive designs in clinical trials Outlook 4 | U Penn | Frank Bretz | April 13, 2016 Adaptive designs over the years Historical papers 5 | U Penn | Frank Bretz | April 13, 2016 Adaptive designs over the years Cross-industry collaborations 6 | U Penn | Frank Bretz | April 13, 2016 Adaptive designs over the years Landmark trials 7 | U Penn | Frank Bretz | April 13, 2016 Adaptive designs over the years Regulatory guidance documents 8 | U Penn | Frank Bretz | April 13, 2016 Adaptive designs over the years Books 9 | U Penn | Frank Bretz | April 13, 2016 Outline Adaptive designs over the years Role of adaptive designs in clinical trials • Reference: Dette, Bornkamp, Bretz (2013) On the efficiency of twostage response-adaptive designs. Statistics in Medicine 32, 1646-60 Outlook 10 | U Penn | Frank Bretz | April 13, 2016 Role of adaptive designs in clinical trials Is it possible to define factors which influence whether to select a standard, or a novel approach for a given trial? In a slightly narrower context: • What are the factors leading to good trial designs? • How can we evaluate those factors? In the following we elaborate further on Ph IIb Dose Finding (DF) trials … where until now traditional trial designs with few doses using pairwise comparisons based on relatively few patients prevail … while acknowledging that DF starts much earlier - Need for incremental dose response (DR) learning at project level requiring integrated planning and quantitative decision making 11 | U Penn Frank Bretzto | April 13, 2016 - | Need design Ph I / Ph IIa studies in view of DF (and Ph III) Role of simulations in designing clinical trials General considerations Comprehensive, simulation-based evaluations to guide modern protocol design Simulations are essential to understand the operating characteristics of a chosen trial design • Compare different design options and / or fine tune the design parameters • Evaluate robustness of a design w.r.t. deviation from assumptions Limitations: • Current limited availability of advanced clinical trial simulators - Need for case-by-case implementations leading to problems of resources, validation / reproducibility, ... • How to summarize, visualize and communicate the results of extensive simulation studies? 12 | U Penn | Frank Bretz | April 13, 2016 Role of analytical methods in designing clinical trials General considerations Analytical results (where available) can confirm simulation results and provide additional insight • Use simplified settings restricted to key factors • Allows one to “see” the relationship between the parameters and possibly prove general results • Overcomes numerical convergence problems often encountered in simulation studies 13 | U Penn | Frank Bretz | April 13, 2016 Role of analytical methods in designing clinical trials Example from Dette et al. (2013) Exponential model 𝑦 = 𝑒 −𝜃𝑥 with unknown parameter 𝜃 and initial guess 𝜃0 Non-adaptive design • 𝑁 observations according to optimal design based on 𝜃0 Two-stage adaptive design • Stage 1: 𝑁0 observations with design based on 𝜃0 • Interim: Estimate 𝜃, resulting in 𝜃1 • Stage 2: 𝑁 𝑁0 observations with design based on 𝜃1 Which design is more efficient and estimates 𝜃 more precisely? 14 | U Penn | Frank Bretz | April 13, 2016 Role of analytical methods in designing clinical trials Example from Dette et al. (2013) Relative efficiency of these two designs depends only on • Unknown factors: - Parameter 𝜃 - Degree of misspecification through 𝜃0 - Variability 𝜎 • Design parameters: - Information fraction 𝑝0 at interim - First stage sample size 𝑁0 Using analytical methods, one can prove that the adaptive design is always better than the non-adaptive design for • large sample sizes (in a broad class of models) • small variances (in this example model) 15 | U Penn | Frank Bretz | April 13, 2016 Role of analytical methods in designing clinical trials Example from Dette et al. (2013) Relative efficiency of adaptive versus non-adaptive design (𝑁 = 100, 𝜃 = 1). Efficiency > 1 indicates that the adaptive design is better. 16 | U Penn | Frank Bretz | April 13, 2016 Outline Adaptive designs over the years Role of adaptive designs in clinical trials Outlook 17 | U Penn | Frank Bretz | April 13, 2016 Conclusions There is a role for adaptive clinical trial designs and they should be considered routinely in drug development No single design / method uniformly best: relative performance depends on scenarios, assumptions, trial objectives, ... Innovative approaches are multifaceted that combine strengths from different areas, e.g. adaptive, model-based, Bayesian, optimal design, etc. Individual techniques are available, but not used properly in combination as a toolbox approach 18 | U Penn | Frank Bretz | April 13, 2016 Conclusions Clinical trial designs depend on a multitude of factors Trial design optimization requires comprehensive simulations to • evaluate operating characteristics and robustness of a chosen design • compare different design options and/or finetune the design parameters Analytical methods might be helpful, if available (but will never replace simulations due to complexity of clinical trial designs) Further factors not considered here will impact the choice for or against a certain design: • recruitment rate, time to read-out, drug supply, randomization, ... 19 | U Penn | Frank Bretz | April 13, 2016 Adaptive Designs: The Swiss Army Knife Among Clinical Trial Designs? It depends ... Simple scissor • Optimal tool for a specific task 20 | U Penn | Frank Bretz | April 13, 2016 Swiss Army knife* • Versatile tool that combines several individual functions in a single unit *Trade Registry Number: CHE-105.977.463 Giant Swiss Army knife* • Functions for every perceivable need • Looks impressive • Highly impractible • Very expensive