Development of machine learning-based predictive models for fertility intentions in patients with Crohn's disease

Abstract

Background – Fertility intentions in patients with Crohn's disease (CD) are shaped by complex interactions between biological and psychosocial factors. However, predictive tools that integrate these dimensions remain underdeveloped. Objective – This study aimed to construct and validate interpretable machine learning (ML) models to predict fertility intentions among reproductive-age patients with CD, and to identify key psychosocial determinants driving reproductive decision-making. Methods – A total of 276 CD patients aged 18–45 years were prospectively recruited from a tertiary center in China. Participants completed standardized assessments covering demographic, clinical, psychosocial, and fertility-related variables. Fertility intention was categorized based onself-reported desire and timing of childbearing. Eight ML models, including neural networks and extreme gradient boosting (XGBoost), were trained on 80% of the data and tested on the remaining 20%. Model performance was evaluated using area under the receiver operating characteristic curve (AUC), accuracy, calibration, and decision curve analysis. SHapley Additive exPlanations (SHAP) were applied for model interpretability. Results – Among the eight algorithms, neural networks and XGBoost demonstrated the highest discrimination (AUC = 0.718 and 0.710, respectively). SHAP analysis revealed that marital status, actual desired number of children, and perceived family support were the most influential predictors of fertility intention. Multivariable logistic regression confirmed marital status and reproductive preference as independent correlates (p < 0.05). These models achieved favorable calibration and clinical utility, as demonstrated by decision curve analysis. Conclusions – This study presents a novel application of interpretable ML models for predicting fertility intentions in CD patients. The findings highlight the dominant role of relational and psychosocial factors over disease activity, offering a methodological and clinical framework for integrating fertility counseling into inflammatory bowel disease (IBD) care. Incorporating structured psychosocial assessment may enable more personalized and anticipatory reproductive support in young patients with CD.