Adaptive trial design that incorporates biomarkers


Year:

Session type:

Lucinda Billingham1
1MRC Midland Hub for Trials Methodology Research & Cancer Research UK Clinical Trials Unit, University of Birmingham

Abstract

A key goal in cancer research is to establish biomarkers that can predict which patients will benefit from treatment. The research pathway in this setting consists of 4 key steps: discovery, development, validation and evaluation. The first 3 steps involve discovering potential predictive biomarkers, developing a predictive tool from one or more of these biomarkers, known as a biomarker-classifier, and finally validating the predictive biomarker-classifier. The validation study is ideally carried out in an independent study from that used at the development stage and involves not only assay validation but also statistical validation, providing a robust estimate of the accuracy of the biomarker-classifier to predict patient outcome. This validated predictive biomarker tool then needs to be evaluated, ideally prospectively, as part of a clinical trial before being introduced into routine clinical practice.

The aim of adaptive trial designs is to improve the efficiency of this research pathway by incorporating several steps within a single trial protocol that also enables assessment of the treatment effect in the overall population. For example, the adaptive signature design (Freidlin and Simon 2005) is a two-stage design which uses patients recruited in the first stage to generate a gene-expression signature for identifying sensitive patients and then applies this prospectively to patients in the second stage. Another example, the biomarker-adaptive threshold design (Jiang, Freidlin and Simon 2007), incorporates a structured plan within a single randomised trial to establish and validate an optimal cut-point for a pre-identified quantitative biomarker for classification of sensitive patients. Alternatively, the biomarker research can be incorporated within an adaptive seamless phase II/III design (Jenkins, Stone and Jennison 2011). In such adaptive trial designs, the key is to control the overall false-positive error rate whilst maximising the probability of detecting treatment effects in either the whole or sub-populations if they truly exist. Adaptive trial designs that take a Bayesian approach provide an alternative to these classical designs and examples of their implementation can be found in lung and breast cancer.