Optimal Biomarker Cutoff Identification and Validation

Advances in molecular technology have enabled the new drug development to shift toward targeted therapy where a subgroup of patients is more likely to benefit from the treatment over the general population. To identify the target patient population, a potential predictive biomarker is often investigated to dichotomize the patient population into a marker-positive and marker-negative group. Under many circumstances, the potential predictive biomarker is measured on a continuous scale. Besides, assuming the biomarker under investigation is predictive of the treatment effect, selection of a higher threshold value would reduce the marker-positive patient population size and potentially the enrollment speed if enrichment is desired. On the other hand, the selection of a lower threshold value would dilute the efficacy signal. It is then of interest for clinical trial designs to evaluate the threshold value that optimizes the balance between the size of the marker-positive group and the efficacy effect size. In particular, we propose to first estimate the threshold value by treating it as a parameter in the likelihood function and then derive the simultaneous confidence intervals for efficacy around the estimated threshold value and optionally, a few other candidate threshold values. This procedure would allow for rigorous and flexible decision-making by taking into consideration both the size and effect of the target population.