Application of a novel machine learning framework for predicting non-metastatic prostate cancer-specific mortality in men using the Surveillance, Epidemiology, and End Results (SEER) database

BackgroundAccurate prognostication is crucial in treatment decisions made for men diagnosed with non-metastatic prostate cancer. Current models rely on prespecified variables, which limits their performance. We aimed to investigate a novel machine learning approach develop an improved prognostic model predicting 10-year cancer-specific mortality and compare its performance existing validated models.MethodsWe derived tested learning-based using Survival Quilts, algorithm that automatically selects tunes ensembles of survival clinicopathological variables. Our study involved US population-based cohort 171 942 cancer between Jan 1, 2000, Dec 31, 2016, from the prospectively maintained Surveillance, Epidemiology, End Results (SEER) Program. The primary outcome was prediction mortality. Model discrimination assessed concordance index (c-index), calibration Brier scores. Quilts compared nine other clinical use, decision curve analysis done.Findings647 151 were enrolled into SEER database, whom included this study. Discrimination greater granularity, multivariable outperformed tier-based models. showed good (c-index 0·829, 95% CI 0·820–0·838) mortality, similar top-ranked models: PREDICT Prostate (0·820, 0·811–0·829) Memorial Sloan Kettering Cancer Center (MSKCC) nomogram (0·787, 0·776–0·798). All three low scores (Survival 0·036, 0·035–0·037; MSKCC 0·037, 0·035–0·039). Of systems, Risk Assessment 0·782, 0·771–0·793) Cambridge Prognostic Groups (0·779, 0·767–0·791) higher c-indices National Comprehensive Care Network, Genitourinary Radiation Oncologists Canada, American Urological Association, European Association Urology, Institute Health Excellence ranged 0·711 (0·701–0·721) 0·761 (0·750–0·772). when stratified by age ethnicity. Decision incremental net benefit currently used practice.InterpretationA produced model, models, only standard Future integration additional data will likely improve accuracy personalised prognostics.FundingNone. Accurate done. 647 practice. A prognostics.