Distribution-Free Approaches to Assessing the Potential Clinical Impact of Biomarkers

Abstract

Recent advances in basic science, combined with new technologies that enable measurement of sophisticated biological processes, present numerous opportunities for advancing the clinical care of patients. A basic tenet of stratified medicine is that utilization of biomarkers can improve identification of which patients may benefit from a particular medical intervention. A precursor to the employment of biomarkers in standard healthcare practices should be a population-level assesment of their impact. Additionally, evaluating biomarkers in accordance with possible clinical applications, at earlier stages of the development pipeline, is important for prioritizing candidates based on the ultimate goal of translating research into improved patient outcomes. In this dissertation, we consider two measures of impact, each relevant for distinct applications of biomarkers to refining medical care. The first measure, net benefit, applies to evaluating biomarkers in clinical decision rules that can guide whether or not a particular clinical intervention is recommended to a patient. The second, a marginal measure of additive interaction, applies to evaluating biomarkers that may be used to define a subgroup of patients for which a treatment may be more, or less, effective than for the whole. The corresponding estimators are either empirical or may be constructed using more general nonparametric approaches. The statistical focus is on efficient inference, an important aspect of evaluating evidence for the adoption of a clinical decision rule in practice or identification of a population for whom an intervention is beneficial.