Review Article

Biomarker adaptive designs in clinical trials

James J. Chen, Tzu-Pin Lu, Dung-Tsa Chen, Sue-Jane Wang


Predictive biomarkers are used to develop (binary) classifiers to identify patients as either good or poor candidates for clinical decision to optimize treatment selection. Ideally, these candidate biomarkers have been well studied in the phase II developmental stage, the performance characteristics of the classifier are well established in one or more retrospective validation, and the assay and predictive performance are reproducible and robust experimentally and analytically. However, completely phase II validated biomarkers for uses in phase III trial are often unavailable. Adaptive signature design (ASD) combines the biomarker identification and classifier development to the selection of candidate patients and a statistical test for treatment effect on the selected patient subgroup for phase III clinical trials. Biomarker-adaptive designs identify the most suitable target subpopulations, based on clinical observations or known biomarkers, and evaluate the effectiveness of the treatment on that subpopulation in a statistically valid manner. This review is concerned with statistical aspects in the biomarker adaptive design for randomized clinical trials. Statistical issues include the interaction test to identify predictive biomarkers, subgroup analysis, multiple testing and false discovery rate (FDR), classification of imbalanced class size data, sample size and power, and validation of the classification model.