In a retrospective study examining postoperative monitoring in Crohn’s disease, researchers developed and validated multimodal machine learning models designed to predict early postoperative complications and endoscopic recurrence after intestinal resection. Both early postoperative complications and endoscopic recurrence represent crucial determinants of postoperative prognosis for patients undergoing surgical treatment for Crohn’s disease. To address these clinical challenges and improve prognostic precision, the investigators formulated multimodal prediction models aimed at providing accurate risk stratification by integrating diverse diagnostic variables gathered during patient care.
The clinical investigation was conducted as a retrospective study that included 159 patients with Crohn’s disease who had undergone intestinal resection. To develop and evaluate the models, the 159 patients were randomly allocated into a dedicated training set and a testing set using an established 7:3 ratio. Within the training set, candidate predictors were identified and refined through a multi-step variable selection pipeline that sequentially utilized univariate logistic regression, least absolute shrinkage and selection operator (LASSO) regression, and multivariable logistic regression.
Following the candidate predictor selection process, conventional logistic regression and seven distinct machine-learning algorithms were developed and systematically compared. Internal model performance was evaluated using 10-times repeated 10-fold cross-validation and 1,000 bootstrap resamples specifically implemented for optimism correction, followed by formal evaluation in the held-out testing set. Across these evaluations, model discrimination was analyzed using receiver operating characteristic curve analysis, model calibration was evaluated with calibration analysis, potential clinical utility was measured via decision curve analysis for clinical net benefit, and feature interpretability was characterized using Shapley Additive Explanations (SHAP).
The clinical outcomes documented within the study cohort revealed that early postoperative complications occurred in 32.08 percent of the patient cohort, while endoscopic recurrence occurred in 27.67 percent of patients. Multivariable analysis showed that sarcopenia, the simplified Magnetic Resonance Index of Activity (sMaRIA), the neutrophil-to-lymphocyte ratio, and the total number of anastomoses were independently associated with the development of early postoperative complications in these patients.
For predicting early postoperative complications, the support vector machine (SVM) algorithm yielded the strongest predictive capability among the compared models. In internal performance evaluations, the support vector machine model achieved an out-of-fold area under the receiver operating characteristic curve (AUC) of 0.847 and an optimism-corrected AUC of 0.855, subsequently attaining an AUC of 0.873 when assessed directly within the independent held-out testing set.
In parallel evaluations focusing on endoscopic recurrence prediction, distinct clinical, surgical, and pathological factors emerged as important predictors. Perianal disease and granuloma formation were both associated with an increased risk of endoscopic recurrence, whereas longer disease duration demonstrated an inverse association with recurrence risk. For estimating endoscopic recurrence, the gradient boosting machine (GBM) algorithm emerged as the leading model, recording an out-of-fold AUC of 0.805, an optimism-corrected AUC of 0.828, and an AUC of 0.795 in the held-out testing set.
Beyond area under the curve metrics, both the support vector machine and gradient boosting machine models exhibited acceptable calibration and demonstrated positive net-benefit profiles on decision curve analysis across the evaluated thresholds. The researchers concluded that multimodal models integrating clinical, inflammatory, imaging, body-composition, surgical, and pathological features showed promising internal predictive performance for postoperative risk stratification in Crohn’s disease. However, the study authors explicitly noted that external multicenter validation remains required before these computational models can be introduced into routine clinical implementation.
Subject of Research: Machine learning prediction of postoperative complications and endoscopic recurrence in Crohn's disease
Article Title: Postoperative monitoring in Crohn’s disease: A multimodal machine learning model for predicting early complications and endoscopic recurrence
Article References: Zheng, J., Chen, J., He, J., Chen, Z., Hu, J., Li, M., Wu, W., Chen, S., Hua, J., & Lin, S. (2026). Postoperative monitoring in Crohn’s disease: A multimodal machine learning model for predicting early complications and endoscopic recurrence. Langenbeck's Archives of Surgery. https://doi.org/10.1007/s00423-026-04343-5
Image Credits: AI Generated
DOI: 10.1007/s00423-026-04343-5
Keywords: Crohn's disease, postoperative complications, endoscopic recurrence, machine learning, support vector machine, gradient boosting machine, intestinal resection, sarcopenia, sMaRIA, neutrophil-to-lymphocyte ratio, perianal disease, granuloma formation, disease duration, decision curve analysis, internal validation, retrospective study
News Source: Teresa Odom. (November 19, 2026). Multimodal Machine Learning Models Predict Postoperative Complications in Crohn’s Disease. Scienmag.



