Development and temporal validation of a machine learning–based model for predicting intraoperative hypothermia in patients undergoing robot-assisted urological tumor resection: A retrospective study
Haoyu Liu, Yunhui Chen, Xue Yang, Tianqi Pan, Shouyan WangBackground
Intraoperative hypothermia (IOH) is prevalent in robotic urological surgery, yet existing prediction models, largely based on linear methods, may have limited capacity to capture complex non-linear risk patterns.
Objective
To develop and validate a preoperative machine learning model for predicting IOH to guide individualized thermal management.
Methods
We analyzed 787 patients (2022–2024) undergoing robot-assisted urological tumor resection. A CatBoost model was developed using 2022–2023 data with 5-fold nested cross-validation for robust feature selection. Independent temporal validation was conducted using 2024 data.
Results
The model, utilizing 17 preoperative predictors, demonstrated temporal generalization with a temporal validation AUC of 0.877 and precise calibration (slope = 0.964). SHAP analysis highlighted markers of physiological stress and metabolic reserve as key drivers. Decision curve analysis confirmed positive net benefit across a broad probability threshold (10%–90%).
Conclusions
This robust machine learning tool accurately identifies high-risk phenotypes using only baseline characteristics. It enables clinicians to implement proactive, personalized warming strategies before anesthesia induction.