Predicting Postoperative Complications in Patients Undergoing Surgery for Inflammatory Bowel Disease

BACKGROUND: Complications after surgery for inflammatory bowel disease (IBD) have serious short- and long-term consequences. Identifying patients at increased risk may improve outcomes. METHODS: We utilized a database for patients undergoing surgery for IBD at three Danish hospitals (2017-2022). Stepwise elimination was used to fit a multivariable logistic regression model to predict the occurrence of a Clavien-Dindo postoperative complication ≥ 3B. Patients undergoing resections for Crohn's disease at another Danish hospital (2010-2020) were used for external validation. Additionally, disease-specific models were created. Predictive performance was assessed using measures of discrimination and calibration and clinical utility was assessed using decision curve analysis. RESULTS: A total of 788 patients underwent bowel resection, stoma takedown or restorative proctocolectomy with ileal-pouch-anal anastomosis, with 140 (17.8%) developing a complication. Sex, ASA score, performance status, alcohol intake, diabetes, psychiatric comorbidity, preoperative systemic steroid use, and prior surgery for IBD were identified as predictors in the final model. Discrimination was modest (c-statistic 0.66, Brier score 0.13, AUPRC 0.38). In external validation of 82 Crohn's disease resections, discriminatory performance was comparable (c-statistic 0.66, Brier score 0.14, AUPRC 0.24). Disease-specific models demonstrated improved performance (Crohn's disease: c-statistic 0.74, Brier score 0.11, AUPRC 0.39; ulcerative colitis: c-statistic 0.73, Brier score 0.14, AUPRC 0.49). CONCLUSIONS: Predictive performance of a general model was inferior to disease-specific models and insufficient for clinical use, and validation across the full spectrum of IBD resections, is needed to ensure generalizability. However, the model identified factors associated with postoperative morbidity, which may represent targets for prospective studies.