Predicting the 10-year risk of death from other causes in men with localized prostate cancer using patient-reported factors: Development of a tool

Objective To develop a tool for estimating the 10-year risk of death from other causes in men with localized prostate cancer. Subjects and methods We identified 2,425 patients from the Surveillance Epidemiology and End Results-Medicare Health Outcomes Survey database, age <80, newly diagnosed with clinical stage T1-T3a prostate cancer from 1/1/1998-12/31/2009, with follow-up through 2/28/2013. We developed a Fine and Gray competing-risks model for 10-year other cause mortality considering age, patient-reported comorbid medical conditions, component scores and items of the SF-36 Health Survey, activities of daily living, and sociodemographic characteristics. Model discrimination and calibration were compared to predictions from Social Security life table mortality risk estimates. Results Over a median follow-up of 7.7 years, 76 men died of prostate-specific causes and 465 died of other causes. The strongest predictors of 10-year other cause mortality risk included increasing age at diagnosis, higher approximated Charlson Comorbidity Index score, worse patient-reported general health (fair or poor vs. excellent-good), smoking at diagnosis, and marital status (all other vs. married) (all p<0.05). Model discrimination improved over Social Security life tables (c-index of 0.70 vs. 0.59, respectively). Predictions were more accurate than predictions from the Social Security life tables, which overestimated risk in our population. Conclusions We provide a tool for estimating the 10-year risk of dying from other causes when making decisions about treating prostate cancer using pre-treatment patient-reported characteristics.