Explainable Machine Learning Analysis of Perioperative Factors Associated with Clinically Significant Emergence Agitation After Pediatric Ophthalmic Surgery

Background and Objectives: Emergence agitation (EA) is a common neurobehavioral disturbance during recovery from sevoflurane anesthesia in pediatric patients, particularly after ophthalmic surgery. Clinically deployable and rigorously validated risk stratification approaches remain limited. We aimed to develop and internally validate an explainable machine learning model to estimate individualized EA risk after pediatric ophthalmic surgery. Materials and Methods: This retrospective cohort study included 1029 children aged 3–7 years who underwent ophthalmic surgery under sevoflurane anesthesia between 2016 and 2025. EA was defined as clinically significant agitation requiring active management in the post-anesthesia care unit. Four machine learning algorithms (regularized logistic regression, random forest, XGBoost, and CatBoost) were developed using stratified patient-level 5-fold cross-validation. Performance was evaluated using pooled out-of-fold predictions. Discrimination, calibration, and classification metrics at the optimal Youden threshold were assessed. SHAP analysis was applied for interpretability. Results: EA occurred in 543 patients (52.8%). XGBoost showed comparable discrimination with slightly higher AUPRC (0.827) and sensitivity (0.796) compared with other models, while maintaining acceptable specificity (0.728). Calibration demonstrated good agreement between predicted and observed risk. SHAP identified airway management and anesthetic-related variables as key contributors. Conclusions: ML-based analysis identified clinically relevant perioperative factors associated with emergence agitation and may provide preliminary insight into perioperative risk stratification pending external validation. External validation is required before clinical implementation.