Statistical Methods for Selective Biomarker Testing

TL;DR

This paper introduces a new statistical method for selective biomarker testing that enhances study efficiency by focusing on individuals with extreme responses, demonstrating unbiased and efficient estimates under certain conditions.

Contribution

It proposes a reverse-regression least squares estimator for continuous biomarker-response data, extending selective genotyping concepts to biomarker testing.

Findings

Estimator is unbiased under joint normality

Method is more efficient than random sampling

Application demonstrated on chronic pain trial data

Abstract

Biomarker is a critically important tool in modern clinical diagnosis, prognosis, and classification/prediction. However, there are fiscal and analytical barriers to biomarker research. Selective Genotyping is an approach to increasing study power and efficiency where individuals with the most extreme phenotype (response) are chosen for genotyping (exposure) in order to maximize the information in the sample. In this article, we describe an analogous procedure in the biomarker testing landscape where both response and biomarker (exposure) are continuous. We propose an intuitive reverse-regression least squares estimator for the parameters relating biomarker value to response. Monte Carlo simulations show that this method is unbiased and efficient relative to estimates from random sampling when the joint normal distribution assumption is met. We illustrate application of proposed methods on data from a chronic pain clinical trial.