Background and objective
Radical cystectomy (RC) is associated with a high risk of postoperative complications. The prediction of individual patient risk for severe complications can facilitate preoperative shared decision-making. Patients with elevated risk may be referred to prehabilitation with the aim to mitigate the risk to improve perioperative outcomes. We developed models to predict severe short-term postoperative complications using preoperatively available clinical variables.
Methods
Data from a prospective cohort of 1313 RC patients treated between 1999 and 2021 was used. Preoperative demographic, laboratory, and cancer-related variables were defined as domains to predict severe complications measured by the Comprehensive Complication Index (CCI). Machine-learning models were trained for each postoperative day and predictor domain. The area under the receiver operating characteristic curve (AUROC) was reported as the primary outcome. Clinical utility was examined using Decision Curve Analysis (DCA).
Results
The best performing model had an AUROC of 0.69 (95 % CI 0.63–0.75) for severe complications on postoperative day (POD) 14. Mean AUROCs across POD 1–30 were 0.64 for all variables combined, 0.58 for demographics, 0.56 for laboratory values, and 0.53 for cancer-related factors. Model calibration and stability improved from POD 10 onwards. Decision curve analysis indicated the highest net benefit from models incorporating all predictors, with demographic variables contributing most among individual domains.
Conclusions and clinical implications
Limited clinical utility of the trained models was observed. The benefit for preoperative clinical decision-making is unclear. Clinical utility may improve by the inclusion of variables related to function in future models (e.g., frailty).
扫码关注我们
求助内容:
应助结果提醒方式:
