Abstract / Summary
Abstract Purpose Anterior clinoidal meningiomas (ACMs) are surgically challenging because of their proximity to critical neurovascular structures. Existing classifications address resectability but do not provide validated estimates of postoperative complications. We aimed to develop and validate preoperative machine learning (ML) models and translate them into an online risk calculator. Methods We retrospectively analyzed data of 241 adults who underwent first-time microsurgical ACM resection at four tertiary centers in three countries between 2010 and 2025. Complications were classified as medical (systemic), surgical (regional), or neurological. Ridge-penalized logistic regression and gradient boosting used routine preoperative data to model each category, evaluated by internal and internal–external (geographical) cross-validation. Results Medical complications occurred in 36 patients (14.9%), predicted by tumor diameter, vessel encasement, comorbidity burden, age, and lower performance status, with an internal area under the curve (AUC) of 0.755 (95% CI 0.667–0.838) and 0.768 across centers. Surgical complications (21, 8.7%) were not reliably predicted (cross-center AUC 0.686). Neurological deficits (68, 28.2%) were poorly predicted as a composite (AUC 0.696) and were modelled by domain. Motor deficits (30, 12.4%) were predicted by tumor diameter, carotid siphon and cavernous sinus involvement, and preoperative motor symptoms (AUC 0.729; 0.711 across centers). Cognitive decline (28, 11.6%) was predicted by age, preoperative motor symptoms, and lower performance status (external AUC 0.731). New visual deficits (19, 7.9%) resolved by 12 months and could not be predicted. Conclusion Preoperative data predicted medical complications, motor deficits, and cognitive decline. The models were translated into an investigational online calculator; external validation in independent cohorts is warranted.