Temporal validation and extension of risk prediction model for pulmonary complications after head and neck surgeries

Abstract Postoperative pulmonary complications (PPCs) are common following head and neck surgeries (HNS), leading to extended hospital stays and increased morbidity. Existing PPC risk prediction models developed for general surgery patients underperform for HNS. This study validates an existing risk prediction model for PPC and extends it for HNS. We assessed the performance of original Gupta model, developed using 2007 ACS-NSQIP (American College of Surgeons-National Surgical Quality Improvement) data, by validating it on 2019 ACS-NSQIP HNS data. We recalibrated, refitted, and extended the model with HNS-specific predictors, followed by temporal validation. Model performance was assessed using scaled Brier score, Nagelkerke’s R  2, discrimination (receiver operating characteristic area under the curve, ROC-AUC), and calibration. The extended model showed improved predictive performance compared to original Gupta model, with scaled Brier score of 0.024 vs –0.056 and R  2 of 0.144 vs –0.196. The ROC-AUC increased from 0.72 to 0.82, and calibration improved from 0.94 to 0.98. Among HNS patients, the extended model showed better performance, discrimination, and calibration in predicting PPC compared to original model. This model could aid clinicians in risk stratification and implement risk reduction strategies. Further validation in external HNS populations is necessary to confirm its generalizability and clinical utility.