Background: To compare the association between prostate-specific antigen (PSA) levels and prostate cancer-specific mortality (PCSM) and the effectiveness of local treatment in patients with high-grade and low-grade prostate cancer (PCa). Methods: This retrospective cohort study enrolled patients diagnosed with clinically localized PCa (cT1-4N0M0) from January 2010 to December 2020 in the Surveillance, Epidemiology, and End Results (SEER) database. Fine-Gray competing risk regression analysis was conducted to generate cumulative incidence plots and estimate the hazard ratio (HR) and 95% CI of PCSM. Multivariable restricted cubic spline analysis was used to examine the nonlinear associations of continuous values of PSA levels with PCSM. Inverse probability of treatment weighting was employed to minimize imbalances in baseline characteristics between different local treatment cohorts. Results: A total of 392 083 eligible patients were included in the study, including 327 659 low-grade (Gleason score [GS]<= 7) PCa and 64424 high-grade (GS >= 8) PCa. In multivariate Fine-Gray competing risk regression analysis, using PSA levels of 4.1-10.0 ng/ml as the reference, the adjusted HR among high-grade patients with PSA levels <= 2.5 ng/ml, 2.6-4.0 ng/ml, 10.1-20.0 ng/ml and >20.0 ng/ml were 1.988 (95% CI: 1.677-2.358), 1.411 (95% CI: 1.194-1.668), 1.472 (95% CI: 1.351-1.603), and 2.506 (95% CI: 2.318-2.709), respectively. Among low-grade PCa, the adjusted HR were 0.985 (95% CI: 0.800-1.213), 0.727 (95% CI: 0.602-0.877), 1.844 (95% CI: 1.679-2.026), and 3.574 (95% CI: 3.220-3.966), respectively. Multivariable-adjusted restricted cubic spline analysis showed a U/J-shaped distribution relationship between PSA levels and PCSM in high-grade PCa, while there was a positive association between PSA levels and PCSM in low-grade PCa. As for local treatment effectiveness, radiation therapy (RT) provided better control of PCSM compared to radical prostatectomy (RP) and RP+RT in high-grade PCa, while RP provided better control of PCSM compared to RT and RP+RT in low-grade PCa. Conclusion: Low PSA level (<= 2.5 ng/ml) is significantly associated with a very high-risk of PCSM in high-grade localized PCa but not in low-grade localized PCa. High-grade localized PCa patients benefit more from RT in terms of PCSM control, while low-grade localized PCa patients benefit more from RP. High-grade localized PCa with low PSA level may be a unique subgroup that could benefit from novel risk stratification strategies in PCa, which requires further studies to investigate the potential of developing novel therapeutic strategies, prognostic tools, and clinical management approaches.
