Cost-Effective Cancer Screening: Bayesian Model Averaging with Two Sources of Variation

Abstract

For medical testing, a popular measure of accuracy is the sensitivity or true positive rate TP/(TP+FN) to reduce the number of false negatives, which result in people who have the disease not receiving treatment. In contrast, there is a need for cost-effective tools that can achieve high precision, TP/(TP+FP), to identify a small fraction of the population with a high probability of having a disease, and thus avoid the cost of testing a large population. This paper explores how Bayesian model averaging can achieve better precision in a novel way. The task is predicting who has mesothelioma, a lung cancer linked with asbestos.