AI Model for Predicting Reoperation After Surgical Site Infection: Analysis Using the ACS NSQIP Database of 96,216 Patients

Introduction: The volume of lumbar spinal surgeries performed in the United States has been rising steadily, and surgical site infections (SSIs) represent a significant postoperative complication. This study aimed to develop, and internally and externally validate, an artificial intelligence (AI) model using extreme gradient boosting (XGBoost) to predict reoperation following SSIs in lumbar spine surgery.

Methods: This study was supported by the TRIPOD+AI guidelines. We used the ACS NSQIP database to identify patients who underwent lumbar spine surgery based on surgical procedure codes. Given the rarity of SSIs, the analysis included superficial SSI, deep incisional SSI, and organ/space SSI. Candidate predictors variables were identified by a panel of clinicians, statisticians, and research fellows. 11 predictors were included in the AI model. After initial analysis, weighted extreme gradient boosting (XGBoost) was identified as the optimal machine learning method. A comprehensive grid search was conducted, and model performance was assessed through internal and external validation. An institutional database was used for external validation. Performance metrics included accuracy, recall (sensitivity), area-under-curve receiver-operating-characteristic (AUC-ROC), area-under-precision-recall-curve, F1-score, and positive predictive value (PPV). Bootstrapping was used to calculate 95% confidence intervals (CIs), and feature importance analysis was performed.

Results: The study cohort included 96,216 patients who underwent lumbar spinal surgery. The weighted XGBoost model demonstrated an exceptional accuracy of 0.994, a sensitivity of 0.800, and an AUC-ROC of 0.997. Feature analysis revealed that the most important predictors of reoperation following SSIs included wound infection types, preoperative albumin levels, and ASA classification.

Conclusion : This study demonstrated the accuracy and reliability of an AI-based extreme gradient boosting model for predicting reoperation due to SSIs following lumbar spine surgery. The implementation of AI models for SSI prediction enables improved risk stratification and optimized resource allocation.