Bayesian basket trial design with biomarker cutoff identification and an alternative approach for estimating the number needed to treat for survival endpoints

Abstract

Patients’ heterogeneity causes considerable difficulty in the current randomized controlled trials (RCTs). This thesis develops a phase II basket biomarker cutoff (BBC) design for proceeding the biomarker-positive patients with desirable responses for further evaluation and proposes a robust measure for summarizing the treatment effect in RCTs with survival endpoints. Clinical trial participants are often heterogeneity, which is one of the fundamental problems in the rapidly developing field of precision medicine. The biomarker design provides a flexible approach to selecting a subset of biomarker-positive patients based on a prespecified cutoff as those most likely to benefit from the new therapeutics. However, a natural question is how to determine the biomarker cutoff that distinguishes the biomarker-positive patients from the negatives, and then evaluate the efficacy of the new therapeutics in one trial. To bridge these two problems into one trial, a phase II BBC design is proposed in the setting, where a biomarker for identifying the sensitive patients is measured on a continuous scale. The proposed BBC design incorporates the biomarker cutoff identification procedure into a basket trial via the Bayesian hierarchical modeling. We verify its feasibility and practicability through real trial examples, extensive simulation studies, and sensitivity analyses. The simulation studies show that the BBC design can select the biomarker-positive patients accurately and exhibits competitive improvement in regards to overall type I error, power, and average sample number. Another issue is the inadequacy of the measure, the number needed to treat (NNT), in summarizing the treatment effect and conveying it to patients and practitioners. For RCTs with survival endpoints, the NNT is computed as the reciprocal of the absolute risk reduction (ARR), namely NNT_ARR, between experiential treatment and control groups. When either the observed event rates for both groups are low, the survival curves cross, or a mixture of survival patterns exist, the NNT_ARR fails to capture the profile of the treatment effect over time, thus leading to misinterpretations of the benefit conferred by the experimental treatment under investigation to some extent. To better quantify NNT in RCTs with survival endpoints, an alternative definition and estimation procedure based on the restricted mean survival time (RMST), namely NNT_RMST, is proposed. Three trial examples with a hypothesis example representing different clinical scenarios are selected to demonstrate the performance of the NNT_RMST. The NNT_RMST not only inherits the intuitive interpretation of the NNT_ARR, but also overcomes the shortcomings of the NNT_ARR, providing a better alternative measure.