Prostate-specific antigen levels as a predictor of lethal prostate cancer.

BACKGROUND Rates of long-term survival among patients with untreated localized prostate cancer are high. To avoid unnecessary treatment, tools are needed to identify the small proportion of patients who are destined to develop lethal prostate cancer. METHODS To evaluate the accuracy of early changes in prostate-specific antigen (PSA) levels as predictors of prostate cancer outcome, we assessed serial measurements of PSA level among 267 men with localized prostate cancer in a Scandinavian cohort of men who were diagnosed between 1989 and 1999 and who were managed by watchful waiting. We then 1) fitted individual regression lines to the PSA values assessed for each patient during the first 2 years of follow-up by using three different models, 2) evaluated early PSA curve characteristics as determinants of the cumulative incidence of lethal prostate cancer and calculated hazard ratios for baseline PSA value and rate of change in PSA level to prostate cancer outcome, and 3) plotted time-dependent receiver operating characteristic (ROC) curves. All P values are two-sided. RESULTS During complete follow-up for a mean of 8.5 years, 34 patients (13%) died from prostate cancer, and 18 (7%) developed metastases but were still alive at end of follow-up. In a log-linear model, both PSA value at baseline (P = .05) and the rate of PSA change (P<.001) were associated with the development of lethal prostate cancer. In the ROC analysis, however, the accuracy of classifying the disease as either indolent or destined to progress was low, regardless of the cut point chosen for initial PSA level or rate of change in PSA level. CONCLUSIONS Although baseline PSA value and rate of PSA change are prognostic factors for lethal prostate cancer, they are poor predictors of lethal prostate cancer among patients with localized prostate cancer who are managed by watchful waiting.