Sojourn-time-corrected receiver operating characteristic curve (ROC) for prostate specific antigen (PSA) test in population-based prostate cancer screening

Abstract

Evaluating the performance of serum prostate-specific antigen (PSA) test in population-based screening with receiver operating characteristics (ROC) curve often neglects the time dimension. Asymptomatic cases with negative PSA test would have been missed if sojourn time is not taken into account to allow for cases surfacing into the clinical phase. Data included 20,796 men with PSA test at the first screening round was used from population-based Finnish prostate cancer screening trial during 1996–1999. Cancers detected at the first screen, together with interval cancers ascertained during 4-year follow-up were expediently used to estimate sensitivity and specificity. A sojourn-time-corrected model was applied to estimating the possible false negative cases for those with PSA < 4 ng/ml for correcting the ROC curve. The estimated sensitivity estimate was reduced from 94.4% without correction to 68.8% with correction but the estimated specificity was identical (89.4% vs. 89.2%) at cutoff of 3 ng/ml. The corrected area under curve (AUC) [77.0% (74.9–79.1%)] of the PSA test was significantly lower than the uncorrected AUC [95.9% (95.3–96.6%)]. The failure of considering the time since last negative screen due to incomplete ascertainment for asymptomatic cancer led to the overestimation of PSA test performance that further affects the cut-off value of PSA tests for population-based prostate cancer screening.