Statistical Innovations in Cancer Research

Abstract

OverviewCancer research is rapidly changing, with fast‐growing numbers of possible cancer targets and cancer drugs to investigate and no end in sight. Advances in genomics, proteomics, and epigenetics provide remarkably detailed profiles of each patient's tumor and, as a result, allow science to consider each patient as unique, with the goal of delivering precision medicine to each patient. The low success rates of late‐phase clinical oncology trials and high costs of bringing new drugs to market, however, have necessitated changes in the drug‐development process. Statistical innovations can help in the design and conduct of clinical trials to facilitate the discovery and validation of biomarkers and streamline the clinical trial process. The application of Bayesian statistics provides a sound theoretical foundation that can encourage the development of adaptive designs to improve trial flexibility and efficiency while maintaining desirable statistical‐operating characteristics.The principal goals of the innovations presented in this chapter are to (1) use information from clinical trials more efficiently in drawing conclusions about treatment effects, (2) use patient resources more efficiently while treating patients who participate in clinical trials as effectively as possible, and (3) identify better drugs and therapeutic strategies more rapidly, moving them more quickly through the development process. The underlying premise is to exploit all available evidence, placing information gleaned from an ongoing clinical trial into the context of what is already known. The innovations considered are intuitively appealing. However, some are controversial. Some are being used in actual clinical trials while others are still being developed and evaluated for such use.This chapter addresses two types of innovations. One is a natural extension of the traditional practice of frequentist statistics. The other type is based on a Bayesian statistical philosophy. The Bayesian approach is tailored to real‐time learning (as data accrue), and the frequentist approach is tied to particular experiments and to the experiment's design. However, there is substantial overlap between these complementary approaches.The main topics covered in this chapter include the following. The introduction of basic probability theory and Bayesian approach is explained through examples. Differences between frequentist and Bayesian methods are compared and contrasted. The development of adaptive designs by applying outcome adaptive randomization, predictive probability, interim and extraim analyses, and factorial design is discussed. In addition, hierarchical modeling can be utilized to synthesize information available from the trial as well as external to the trial. Seamless phase I/II and II/III designs can be applied to shorten the drug‐development period. Platform designs can be constructed to evaluate many drugs simultaneously. The application of these statistical innovations is illustrated in the BATTLE trials and I‐SPY 2 trial. Finally, information on computing resources for the design and implementation of innovative trials is given.