Postoperative Complications in Older Adults After Local Surgical Procedures; Risk Factor Analysis and Comparison of Machine Learning Methods

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Older adults frequently undergo excision of cutaneous lesions under local anesthesia and may be at increased risk of postoperative complications due to comorbid conditions and concomitant medications. This study aimed to identify predictors of postoperative complications in patients aged 65 years and older and to compare the predictive performance of logistic regression with machine-learning models. This retrospective study included 314 patients who underwent cutaneous lesion excision under local anesthesia. Demographic, clinical, and perioperative variables were collected and assessed using multivariable logistic regression. Decision tree, random forest, and gradient boosting models were developed using an 80/20 training–testing split. Missing data were imputed, and class imbalance was addressed using resampling techniques. Model performance was evaluated using discrimination, calibration, and decision curve analysis. Postoperative complications occurred in 119 patients (37.9%). Patients with complications were older, had lower body mass index, and more frequently had diabetes, coronary artery disease, and antithrombotic therapy. Logistic regression identified low body mass index, diabetes, and coronary artery disease as independent predictors but demonstrated limited discriminatory ability. Among machine-learning models, the random forest achieved the best performance, with superior discrimination and calibration. Body mass index, perioperative blood pressure, age, and procedure duration were the most influential predictors. Postoperative complications are common among older adult patients undergoing cutaneous surgery under local anesthesia. Low body mass index, cardiometabolic comorbidities, perioperative hemodynamic factors, and operative duration are key determinants of risk. Machine-learning approaches, particularly random forest models, may improve perioperative risk stratification in this population.