Machine learning-based prediction models for postoperative pulmonary complications in elderly patients undergoing...

Area of Science:

• Medical Informatics
• Surgical Oncology
• Geriatric Medicine

Background:

• Postoperative pulmonary complications (PPCs) are frequent in elderly patients after abdominal surgery.
• Existing risk prediction tools often lack generalizability and clinical interpretability for this demographic.

Purpose of the Study:

• To develop and validate a machine learning model for predicting PPCs in older adults undergoing abdominal surgery.
• To enhance the interpretability of risk prediction models for clinical decision-making.

Main Methods:

• Retrospective cohort study of 2,456 elderly patients (>=65 years) for model development and 542 for external validation.
• Comparison of six machine learning algorithms, including XGBoost, logistic regression, and neural networks.
• Evaluation using discrimination, calibration, decision curve analysis, and SHAP for interpretability.

Main Results:

• The XGBoost model demonstrated superior performance, achieving AUCs of 0.856 and 0.821 in independent and external validation cohorts, respectively.
• Key predictors identified by SHAP analysis included ASA physical status, COPD, and surgical site.
• The model effectively stratified patients into low (6.9%), moderate (24.4%), and high (37.6%) risk groups for PPCs.

Conclusions:

• An interpretable gradient-boosting model (XGBoost) shows promise for risk-stratified perioperative assessment in elderly surgical patients.
• Prospective multicenter validation is recommended prior to widespread clinical adoption.