Externally validated machine learning models for prediction of treatment selection and clinical outcomes in posterior communicating artery aneurysms

Abstract Posterior communicating artery (PCOM) aneurysms account for approximately 25% of intracranial aneurysms and present clinical challenges regarding optimal treatment selection. This study developed and validated machine learning models to predict treatment modality, complications, and functional outcomes in PCOM aneurysm management using a comprehensive institutional database of 75 patients treated between 2002 and 2023. Five machine learning algorithms—Random Forest, Support Vector Machine, Logistic Regression, Decision Tree, and XGBoost—were systematically compared across 7 outcome variables: treatment type (open vs. endovascular), specific open surgical treatment, specific endovascular treatment, intraoperative rupture, complications, angiographic obliteration, and discharge modified Rankin Scale score. Models were developed using preoperative patient-specific (age, sex, family history) and aneurysm-specific (size, presentation, morphology, location) parameters, with rigorous preprocessing that included class-balancing strategies, feature engineering, and regularization to minimize overfitting. Cross-validation performance demonstrated high accuracy for treatment prediction (78.7% Random Forest), complications (94.7% Random Forest; however, sensitivity of 0% indicates that this accuracy should not be interpreted as clinically useful event detection), intraoperative rupture (92.0% Random Forest), and angiographic obliteration (87.3% Random Forest). External validation on an independent cohort from the University of Louisville (n=20–28; n=20 most complete cases for the primary external cohort, with up to n=28 for selected endpoints with available labels) revealed variable generalizability: treatment prediction maintained 78.1% accuracy, and complication prediction showed reduced external performance (70.3%), highlighting institution-specific factors. Comprehensive sensitivity analyses demonstrated that class balancing significantly improved minority class recall (0.0% → 52.2%–86.3%) for rare outcomes. Feature importance analyses identified age, aneurysm size, admission modified Rankin Scale score, and Hunt-Hess grade as the most influential predictors across multiple outcomes. This study represents the first externally validated, multioutcome machine learning framework for PCOM aneurysm management, providing quantitative risk estimates and evidence-based decision support for treatment selection and outcome prediction.