Development of a stacking model for personalized treatment of Crohn’s disease: leveraging routine clinical features to forecast infliximab response

Aims Infliximab (IFX) is widely used for treating Crohn’s disease (CD), but a significant proportion of patients experience primary non-response or loss of response. Early prediction of IFX efficacy is crucial to avoid ineffective treatment, adverse effects, and financial burden. The aim of this study was to develop a stacking model using routine clinical data to predict IFX clinical response. Method This retrospective cohort study enrolled CD patients initiating IFX therapy between January 2019 and December 2025. Feature selection was performed using statistical analysis and Least Absolute Shrinkage and Selection Operator (LASSO) regression. Base models (Elastic Net, Support Vector Machine, Random Forest, XGBoost) were built and integrated into a stacking model. Model performance was evaluated using the Area Under the Receiver Operating Characteristic Curve (AUROC), accuracy, precision, sensitivity, specificity, recall, and F-score on a hold-out testing set. Class imbalance was addressed using the Synthetic Minority Over-sampling Technique. Result A total of 319 patients were enrolled, comprising 237 responders and 82 non-responders, reflecting a class imbalance. An independent dataset containing 43 patients was used for temporal validation. LASSO regression identified five key predictors: erythrocyte sedimentation rate, C-reactive protein, Crohn’s disease activity index, red blood cell count, and diagnostic age. The stacking model, composed of Elastic Net and Random Forest, achieved an AUROC of 0.897 (95% CI: 0.832–0.956) on the validation set and 0.874 (95% CI: 0.749–0.957) on the testing set, demonstrating robust predictive performance. Conclusion The developed stacking model effectively predicts IFX response using readily available clinical variables, representing a preliminary step toward personalized treatment planning. Prospective validation is required before clinical implementation.