Boehm K., Larcher A., Beyer B., Tian Z., Tilki D., Steuber T., Karakiewicz P.I., Heinzer H., Graefen M., Budäus L.
European Urology 2015
Background: The D’Amico risk stratification, Cancer of the Prostate Risk Assessment (CAPRA) score, and Stephenson nomogram are widely used prediction tools for biochemical recurrence and survival after radical prostatectomy (RP). These models have not been compared with respect to cancer-specific mortality (CSM) prediction. Objective: To validate and compare the prediction tools for 10-yr CSM. Design, setting, and participants: Overall, 2485 prostate cancer patients underwent RP in a European tertiary care center. Outcome measurements and statistical analysis: Three preoperative models (D’Amico, CAPRA, and Stephenson) were compared in terms of their ability to predict 10-yr CSM; therefore, accuracy tests (area under the receiver operating characteristic curve [AUC]), calibration plots, and decision curve analysis (DCA) were assessed for each model. Results and limitations: CSM at 10 yr was 3.6%. The AUC was 0.76, 0.77, and 0.80 for the D’Amico, CAPRA, and Stephenson models, respectively. In calibration plots, predicted probabilities were close to the observed probabilities for the D’Amico model but showed underestimation of CSM for the Stephenson nomogram and overestimation of CSM for the CAPRA score. DCA identified a benefit for the CAPRA score. These results apply to patients treated at a European tertiary care center. Conclusions: Despite good discriminatory power, all tested models had some shortcomings in terms of prediction of 10-yr CSM. All three models showed good performance in North American cohorts, but our results suggested a lack of generalizability to European patients. To overcome this issue, local recalibration of the variable weights could be performed. Another possibility is the development of more universal markers that are independent of regional practice differences or, alternatively, the development of better tools to quantify clinical practice differences. Patient summary: Prediction tools can predict cancer survival prior surgery, relying on points for age, prostate-specific antigen levels, aggressiveness, and percentage of cancer at biopsy. These tools are reliable in North American patients but have shortcomings for identifying patients at high risk of prostate cancer death in Europe. Despite good discriminatory power, all tested models have some shortcomings for prediction of cancer-related mortality risk at an individual patient level. Specifically, all models can adequately discriminate among different risk strata, but our results suggest a lack of generalizability to European patients.