Predictors of long-term prostate cancer mortality in men continuing prostate-specific antigen screening after age 70.

5028 Background: Continuing prostate-specific antigen (PSA) screening after age 70 might benefit men at elevated risk of prostate cancer-specific mortality (PCSM), such as Black men with long life expectancies. However, the relative value of race, PSA value, and life expectancy in predicting long-term PCSM and identifying elderly men who may benefit from continued screening is unknown. Methods: Using national Veterans Health Administration (VHA) clinical data, we identified all men turning 70 years old between 2008 and 2020 with the most recent PSA ≤4 ng/mL from age 65-69, 5+ years of clinical data, and no prior history of prostate cancer or prostate biopsy. We quantified the presence, intensity, and predictors of continued PSA screening after age 70. We developed competing risk regressions to predict 10-year individualized risk of PCSM using race, baseline PSA value, and 10-year overall survival predictions from an internally developed XGBoost machine learning model using high dimensional clinical data to represent competing mortality risk. Decision curve analyses compared the net benefit of continuing screening past age 70 based on race, PSA, or competing mortality risk predictions alone versus an integrated model combining all three predictors. In sensitivity analyses, we modeled metastatic prostate cancer risk instead of PCSM. Results: The cohort included 921,609 recently screened men turning 70 years old: 82% were White and 11% were Black. Most patients (77%) had a baseline PSA <2.0 ng/mL. Screening continuation from age 70-80 was nearly universal: 87% had undergone at least one additional PSA screen and mean cumulative number of additional screens was 5. Screening frequency after age 70 did not vary substantially by competing mortality risk or race. The 10-year cumulative incidence of PCSM was 0.26% overall, and 95% of men had a 10-year risk <0.73%. Baseline PSA was the strongest predictor of 10-year PCSM (0.79% for 3.0-3.99 ng/mL vs. 0.097% for 0.2-0.99 ng/mL; 8.1x difference). Associations with 10-year PCSM risk were more limited for race (0.36% for Black vs 0.25% for White; 1.4x difference) and competing mortality risk predictions (0.24% for highest quintile vs 0.21% for lowest quintile; 1.1x difference). Decision curve analyses suggested minimal net benefit of adding race or competing mortality risk predictions to models containing baseline PSA. Similar results were found in models predicting metastatic prostate cancer incidence. Conclusions: Almost 9 of 10 patients in VHA continue PSA screening after age 70 despite most having very low risk of PCSM over the next decade. Recent PSA values are substantially more informative for PCSM risk prediction than race or competing mortality risk predictions under contemporary standards of care. Efforts to personalize screening decisions in men aged 70+ with no history of prostate cancer should focus on recent PSA values.