Biomarker-based methods for determining noncompliance in a prevention trial

The Prostate Cancer Prevention Trial (PCPT) is a 7-year randomized trial of 18,000 men designed to test a daily dose of finasteride (5 mg/d) for the primary prevention of prostate cancer. Compliance is assessed every 6 months by determining the percentage of pills consumed since the previous visit. In addition, levels of 5alpha-dihydrotestosterone (DHT) are measured at yearly intervals in a random subset of 5% of participants in the trial as part of a substudy to investigate the potential of this biomarker for monitoring compliance. Motivated by this substudy, we outline three biomarker-based rules for monitoring compliance over time in prevention trials. We make comparisons among the rules using receiver operating characteristic curves. We illustrate the methods using the biomarker DHT in the PCPT substudy with pill counts as a gold standard, and in a simulation study under a hypothetical true gold standard. Our results indicate that simple absolute threshold monitoring rules perform just as well as more complicated rules that utilize the entire biomarker history. (C) 2002 Elsevier Science Inc. All rights reserved.