Explainable machine learning–based prediction of postoperative pulmonary infection after aneurysmal subarachnoid hemorrhage clipping and exploratory analysis of enteral nutrition timing: a dual-cohort retrospective study

Background: Postoperative pulmonary infection (PPI) is a common complication after microsurgical clipping for aneurysmal subarachnoid hemorrhage (aSAH) and is associated with impaired recovery and greater use of hospital resources. Objective To develop and temporally validate an explainable machine-learning model for PPI risk stratification and to explore the association between enteral nutrition (EN) initiation timing and subsequent PPI risk. Methods This dual-cohort retrospective study was reported in accordance with Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis plus Artificial Intelligence (TRIPOD+AI) principles. Candidate predictors were selected using a combined least absolute shrinkage and selection operator (LASSO) regression and random-forest importance strategy. Nine machine-learning algorithms were developed and compared in terms of discrimination, calibration, and clinical utility using decision-curve analysis. Results In the validation cohort, the Naive Bayes model showed the best overall performance [area under the receiver operating characteristic curve (AUC), 0.811; 95% confidence interval (CI), 0.755–0.855]. SHAP analysis identified the major contributors to the primary 24-h prediction model. Separately, a secondary exploratory analysis of continuous EN initiation timing identified a nonlinear pattern, with a bootstrap median model-derived inflection point of 24.0 h (95% bootstrap interval, 23.0–27.0 h). Conclusions This explainable model may support early PPI risk stratification after aSAH clipping. The observed EN timing signal should be interpreted as exploratory and warrants prospective validation before clinical implementation.