Bayesian, Utility-Based, Adaptive Enrichment Designs with Frequentist Error Control

Our improving understanding of the biology underlying various diseases has reinforced the idea that many diseases previously considered homogeneous are in fact heterogeneous collections with different prognoses, pathologies, and causal mechanisms. To this end, the biomedical field has begun to focus on developing targeted therapies: therapies aimed at treating only a subset of the population with a given disease (often derived by the molecular pathology of the disease). However, characterizing these subsets has been a challenge: Hundreds of patients may be required to effectively characterize these subsets. Often information on this many patients is not available until well into large-scale trials. In this chapter we discuss adaptive enrichment designs: clinical trial designs that allow the simultaneous construction and use of biomarkers, during an ongoing trial. We first detail common scenarios where adaptive enrichment designs could be fruitfully applied to gain efficiency over classical designs. We then discuss two classes of adaptive enrichment strategies: Adaptation based on prespecified covariate-based stratification, and adaptation based on modeling response as a potentially more complex function of covariates. We will contrast these strategies with more classical non-enriched biomarker strategies (based on post hoc modeling/testing). Finally, we will discuss and address a number of potential issues and concerns with adaptive enrichment designs.