Predictive biomarkers are scalar biological measurements used to indicate whether a patient is likely to benefit from a particular treatment relative to some control regimen. In oncology, most predictive biomarkers are binary indicators of whether the patient's tumor contains a mutation which renders the tumor sensitive to the test drug. Candidate predictive biomarkers need to be validated for analytical performance and for clinical utility. Analytical performance denotes accuracy and reproducibility of measurement whereas clinical utility refers to effectiveness as a decision aid for selecting between the test or control regimen.
Although new clinical trial designs have been introduced for co-development of a new treatment and a companion diagnostic, there have been few proposed measures of clinical effectiveness for predictive biomarkers. I will discuss the difficulties in developing measures such as sensitivity, specificity, ppv and npv for predictive biomarkers. I will also introduce such measures for use with binary predictive biomarkers and time-to-event outcomes. The assumptions needed to ensure the validity of these measures will be carefully discussed as well as approach
|