Many medical tests measure the presence of some biological marker to diagnose patients with a given disease, for example, prostate specific antigen is used to screen for prostate cancer. However, factors other than the disease, either from the patient or from the test procedure, can affect the test result. VIDI assistant member Dr. Holly Janes and colleagues have come up with new statistical methods and software to incorporate these factors, or covariates, into analyses for assessing a test’s accuracy.
The receiver operating characteristic, or ROC, curve is a common statistical tool to assess the accuracy of a given test. It plots the true-positive rate of the test against its false-positive rateA better test has a lower true-positive rate at the same false-positive rate. Janes’ methods provide ways to incorporate covariates into the ROC analysis.
Accommodating Covariates in ROC Analysis. Janes H, Longton G, Pepe M. Stata J. 2009 Jan 1;9(1):17-39.