Predicting Long-Term Risk for Prostate Cancer Mortality Following a Prostate-Specific Antigen Screening Test: Prognostic Model Development and External Validation
Menée à partir de données portant sur 33 339 hommes puis validée sur 174 787 hommes supplémentaires (âge : de 55 ans à 74 ans), cette étude évalue la performance d'un nouveau modèle, basé sur le niveau sérique du PSA, les antécédents familiaux de cancer de la prostate et l'origine ethnique, pour prédire le risque de décès par cancer de la prostate
Background : Despite the scale of prostate-specific antigen (PSA) testing for prostate cancer (PCa) screening, prediction models do not predict time-to-event end points or adjust for patient life expectancy.
Objective : To develop, externally validate, and compare to existing tools a novel prognostic model for risk for prostate cancer–specific mortality (PCSM) after a PSA test.
Design : Prognostic model development in the PCa screening group of the Prostate, Lung, Colorectal and Ovarian (PLCO) Cancer Screening Trial; external validation in a Veterans Affairs (VA) population of patients undergoing PSA testing.
Setting : United States. PLCO patients were enrolled from 1993 to 2001, and VA patients underwent PSA testing from 2002 to 2006. Survival follow-up was updated through 2022 in both cohorts.
Patients : Male patients aged 55 to 74 years in the PLCO PCa screening group (n = 33 339) and the VA Healthcare System (n = 174 787).
Measurements : The model’s predicted outcome is PCSM at a specified time point; predictors included PSA level, family history of PCa, and race. Predictors of other-cause mortality included age; body mass index; smoking status; and presence of hypertension, diabetes, or stroke.
Results : In the model development cohort, the area under the receiver operating characteristic curve (AUC) at 29.5 years from screening was 0.666 compared with 0.643 for a previously validated prostate biopsy risk model (Prostate Biopsy Collaborative Group [PBCG]) (P < 0.001). In the external validation cohort, the AUC at 20 years from screening was 0.776 for the PLCO model versus 0.749 for the PBCG model (P = 0.031).
Limitation : The model may not be generalizable to more contemporary PSA screening practices given the periods studied.
Conclusion : This PCSM prognostic model was developed from long-term clinical trial data, was externally validated in a large national cohort, and may be used to improve interpretation of PSA results.
Annals of Internal Medicine , résumé, 2026