Many human diseases are heterogeneous in pathogenesis, prognosis and response to treatment. Nevertheless, clinical trials generally provide eligibility for a wide range of patients who often cannot reasonably be expected to respond similarly to molecularly targeted treatments. Lack of benefit for the majority of eligible patients can mask benefit for subsets of patients unless there are mechanistic biomarkers for subdividing the heterogeneous population.

Oncology has made substantial therapeutic progress using biomarkers which indicate particular somatic mutations which drive disease invasion. This has dramatically changed the process of drug development and evaluation with the development of drugs specifically to inhibit the activated protein product of such mutations. In these cases the enrichment design in which only biomarker positive patients are enrolled has been used in the past decade for the rapid regulatory approval of large numbers of new oncology drugs. In many cases, however, the biology of the disease is more complex and an appropriate predictive biomarker is not known in advance.

In this session, we will review clinical trial designs for the development of new therapeutics and companion diagnostics to inform their use. These include settings in which there is a candidate biomarker but the biological evidence for excluding biomarker negative patients is not compelling. In some cases, the relevance of the biomarker is established but an appropriate cut-point for positivity is only imprecisely known based on previous studies. In other cases, there are multiple candidate biomarkers and an effective classifier utilizing all or some of them needs to be developed. Using carefully constructed adaptive designs and re-sampling methods, we can sometimes both adaptively develop an effective classifier and internally validate it for identifying a subset in which the drug is effective in a single phase III trial. We will describe adaptive enrichment designs in which eligibility is modified adaptively during the trial using pre-specified frequentist or Bayesian strategies which preserve the type I error level of the trial while dramatically increasing the statistical power. Other designs to be discussed include the adaptive signature design, adaptive threshold design, basket designs, platform designs and a new class of run-in designs which use pharmacodynamics biomarkers as predictive markers. We will also discuss the development and use of prognostic signatures to facilitate the selection of an appropriate control group to avoid a high placebo response rate. Finally, we will discuss the use of observational and “real world” data to refine the intended use population for previously approved drugs.

The class is targeted to statisticians and others interested in clinical trials who want to become aware of the new ways of dealing with heterogeneous diseases to find effective treatments. It is for individuals interested in new designs and paradigms for development of treatments and for finding the most appropriate intended use population. The designs discussed are adaptive in the definition of the target patient population. Extensive use will be made of re-sampling methods for de-biasing re-substitution estimates and on pre-validation. We assume that the student is familiar with the standard features of randomized clinical trials as emphasis will be on new aspects.