Baseline PSA as a predictor of prostate cancer-specific mortality over the past 2 decades: Duke University experience.
BACKGROUND: A diagnosis of prostate cancer is not often predictive of death from prostate cancer because of competing causes of mortality. Identification of the risk of death from prostate cancer and death from all causes using information available at the time of baseline prostate-specific antigen (PSA) measurement appears to be particularly pertinent. METHODS: The Duke Prostate Center database was used to identify men who had their PSA level measured over the past 20 years. The Cox proportional hazards model was used to assess whether baseline PSA, race, and age at baseline PSA could predict death from prostate cancer and death from all causes after baseline PSA measurement. The receiver operating characteristic (ROC) curve was performed to analyze the accuracy of baseline PSA as a continuous variable in predicting death from prostate cancer. RESULTS: A total of 4568 men diagnosed with prostate cancer after baseline PSA measurement were included. On multivariate analysis, baseline PSA levels of 4.0 to 9.9 ng/mL and ≥10 ng/mL were associated with significantly higher rates of death from prostate cancer compared with PSA levels <2.5 ng/mL. An advanced age at baseline PSA and African American race were associated with a higher death rate from prostate cancer and death from all causes. The area under the ROC curve for baseline PSA predicting death was 0.839. When a baseline PSA of 10 ng/mL was chosen to predict death from prostate cancer, the corresponding sensitivity and specificity were 77% and of 78%, respectively. CONCLUSIONS: Baseline PSA appears to be a reliable and independent predictor of death from prostate cancer. A baseline PSA of ≥4 ng/mL has been associated with higher risk of death from prostate cancer.
Tang, P; Sun, L; Uhlman, MA; Polascik, TJ; Freedland, SJ; Moul, JW
Volume / Issue
Start / End Page
Pubmed Central ID
International Standard Serial Number (ISSN)
Digital Object Identifier (DOI)