90-day mortality prediction in elective visceral surgery using machine learning: a retrospective multicenter...

Area of Science:

• Biomedical research
• Surgical outcome prediction
• Machine learning applications

Background:

• Machine learning (ML) adoption in biomedical research is growing, but its utility in predicting outcomes for visceral surgery is not well-established.
• This study addresses the need to evaluate ML's potential for preoperative risk stratification in major elective visceral surgery.

Purpose of the Study:

• To compare the predictive performance of ML models against conventional scoring systems for 90-day mortality (90DM) in visceral surgery.
• To assess the efficacy of an aggregated multi-organ ML approach versus organ-specific models and traditional scores.

Main Methods:

• A retrospective cohort study of 7711 patients undergoing major elective visceral surgery (2014-2022) across two tertiary centers.
• Development and external validation of multiple ML models for 90DM prediction, benchmarked against ASA score and rCCI.
• Calculation of AUROC and AUPRC for model performance evaluation, including organ-specific and aggregated multi-organ approaches.

Main Results:

• An XBoost classifier achieved the highest performance (AUROC: 0.86, AUPRC: 0.2) and robustness upon external validation.
• All ML models surpassed the predictive power of the ASA score and rCCI.
• Gastric and intestinal surgery models showed high organ-specific performance, while a combined multi-organ approach outperformed organ-specific models.

Conclusions:

• ML provides a robust method for preoperative risk stratification of 90DM in elective visceral surgery.
• Training ML models on multi-organ cohorts can enhance accuracy and robustness compared to organ-specific models.
• Further prospective studies are warranted to validate ML's role in predicting surgical outcomes.