Development and external validation of an explainable machine-learning model for early complications following robot-assisted radical prostatectomy: a multicenter study

Complications occurring early after robot-assisted radical prostatectomy (RARP) can hinder postoperative recovery. Existing risk instruments, however, generally incorporate few predictors and assume linear associations. We aimed to derive and externally validate an interpretable machine-learning model that combines clinical, laboratory, and intraoperative information and to translate the model into an online risk calculator. Records were reviewed for 1,254 patients with prostate cancer (PCa) who received RARP at…

By the Truvace deskOctober 15, 20262 min read1 sourcePeer-reviewedTRV-2026-1485· cite & verify

In brief

Researchers reviewed records of 1,254 prostate cancer patients treated with robot-assisted radical prostatectomy at three Chinese PLA General Hospital medical centers. They compared seven algorithms on eight selected predictors and found that a multilayer perceptron gave the most balanced performance, then turned it into a web-based risk calculator.

Early complications can slow recovery, and existing tools use few predictors and assume linear relationships. A calculator available at the end of surgery could support closer monitoring and individualized risk grouping, but the evidence is retrospective and from one hospital system, and real-world clinical use has not been shown.

Main points

The cohort included 1,254 patients: 701 for training, 300 for internal testing, and 253 from a separate center for external validation.
Sixty-four candidate variables were narrowed to eight predictors using univariate screening and LASSO regression.
The multilayer perceptron reached an AUC of 0.880 internally and 0.843 externally, with good calibration and favorable net benefit.
SHAP analysis ranked crystalloid infusion as the top predictor, followed by operative time and thrombin time.
In an exploratory external comparison, the model outperformed the Comorbidity Score for Robotic Surgery.

The gain

A multilayer perceptron using eight clinical, laboratory, and intraoperative predictors estimated early complication risk after prostatectomy, and an online calculator could help flag high-risk patients right after surgery.

The rundown

The study tested gradient boosting decision trees, random forest, logistic regression, support vector machine, AdaBoost, Gaussian naive Bayes, and multilayer perceptron. Performance was judged by AUC, calibration plots, and decision curve analysis for clinical usefulness.

Interpretability relied on SHapley Additive exPlanations. Intraoperative and laboratory factors, notably crystalloid infusion, operative time, and thrombin time, drove predictions, giving clinicians some insight into why a patient was flagged.

The authors benchmarked the model against the Comorbidity Score for Robotic Surgery in the external cohort and found greater potential in discrimination, calibration, and net benefit. They deployed the final model as an online calculator intended to inform postoperative monitoring.