Cost-efficient designs based on linearly associated biomarkers

Abstract

A major limiting factor in much of the epidemiological and environmental researches is the cost of measuring the biomarkers or analytes of interest. Often, the number of specimens available for analysis is greater than the number of assays that is budgeted for. These assays are then performed on a random sample of specimens. Regression calibration is then utilized to infer biomarker levels of expensive assays from other correlated biomarkers that are relatively inexpensive to obtain and analyze. In other contexts, use of pooled specimens has been shown to increase efficiency in estimation. In this article, we examine two types of pooling in lieu of a random sample. The first is random (or traditional) pooling, and we characterize the second as “optimal” pooling. The second, which we propose for regression analysis, is pooling based on specimens ranked on the less expensive biomarker. The more expensive assay is then performed on the pool of relatively similar measurements. The optimal nature of this technique is also exemplified via Monte Carlo evaluations and real biomarker data. By displaying the considerable robustness of our method via a Monte Carlo study, it is shown that the proposed pooling design is a viable option whenever expensive assays are considered.