Long-term performance of the Gleason score in predicting metastatic and fatal prostate cancer.

The time dependence in the performance of prognostic factors for cancer survival is typically incompletely understood, including in prostate cancer. We applied a "cumulative/dynamic" time-varying area under the curve (tAUC) for cumulative incidence of lethal prostate cancer (metastases/cancer-specific death). In two prospective prostate cancer patient cohorts followed for lethal outcomes after radical prostatectomy, genitourinary pathologists undertook a histopathologic review of tumor specimens to assign Gleason grade groups 1-5, a known strong short-term prognostic factor. Among 1490 patients, 144 lethal events occurred during 35 years of follow-up (median, 18). The 10-year risk of lethal disease was 2% (95% CI, 1-3) for grade group 2 and 21% (95% CI, 16-28) for grade group 5 cancers. By 25 years, risks were 5% (95% CI, 3-8) and 32% (95% CI, 26-41), respectively. Prediction accuracy was strongest over the first 10 years (tAUC 0.83 [95% CI, 0.79-0.86]) but remained informative for 25 years (tAUC 0.76 [95% CI, 0.70-0.81]), including among long-term metastasis-free survivors. Risk-based AUCs instead of hazard-based "incident/dynamic" AUCs allowed for interpretable characterization of prognostic accuracy over well-defined time periods. Lethal progression of high-grade prostate cancer occurred predominantly early, while that of lower grade tumors accumulated over decades after surgery.