Adjustment for the measurement error in evaluating biomarkers.

Biomarkers that can help identify patients who will have an early clinical benefit from a treatment are important not only for patients' survival and quality of life, but also for the cost of health care. Owing to reasons such as biological variation and limited machine precision, biomarkers are sometimes measured with large errors. Adjusting for the measurement error in calculating the proportion of the treatment effect explained by markers has been a subject of research. The proportion of information gain (PIG), a new quantity to measure the importance of a biomarker, has not yet been studied for variables measured with error. In this article, we provide methods to account for the measurement error in the calculation of PIG for continuous, binary and time-to-event outcomes. Simulation shows that the adjusted estimator has little bias and has less variability compared to the naive estimator ignoring the measurement error. Data from an osteoporosis clinical study are used to illustrate the method for a binary outcome.