Malnutrition is a frequent yet under-recognized complication of inflammatory bowel disease (IBD), negatively impacting therapeutic response, disease course, and prognosis. Artificial intelligence (AI) offers new opportunities for early, accurate nutritional risk stratification. This study aimed to determine the prevalence of malnutrition in IBD patients, identify associated clinical and biological factors, and develop a machine-learning model for non-invasive malnutrition screening.

A cross-sectional analytical study was conducted at Sahloul University Hospital including 224 IBD patients. Malnutrition was defined by a Malnutrition Universal Screening Tool (MUST) score ≥ 2 or ESPEN 2015 criteria (BMI < 18.5 kg/m² or significant weight loss with age-adjusted low BMI). Machine-learning models were developed using the CRISP-DM framework. Model performance was assessed using ROC analysis and predictive accuracy.

Among the 224 included patients (157 Crohn’s disease; 67 ulcerative colitis), median age was 38 years (IQR 29–48) with a sex ratio M/F of 1.26. The overall prevalence of malnutrition was 25.9%, with good agreement between MUST and ESPEN criteria (κ = 0.723).In univariate analysis, malnutrition was significantly associated with smoking (p = 0.018), ileocolonic Crohn’s disease (p = 0.011), distal ulcerative colitis (p = 0.041), clinical and endoscopic activity (p < 0.001), anemia (p < 0.001), hypoalbuminemia (p < 0.001), elevated CRP (p < 0.001), and absence of azathioprine or biologic therapy (p < 0.001).Multivariate logistic regression identified three independent predictors of malnutrition: clinical disease activity (OR 3.38; 95%CI 1.51–7.58), elevated CRP (OR 2.51; 95%CI 1.14–5.53), and hypoalbuminemia (OR 4.03; 95%CI 1.86–8.70).Among the tested machine-learning models, the Neural Network demonstrated the highest diagnostic performance, with an AUC of 0.90 and a predictive accuracy of 87%.

Malnutrition is highly prevalent among IBD patients and strongly associated with disease activity, systemic inflammation, and hypoalbuminemia. The AI-based Neural Network model showed excellent performance in predicting malnutrition, supporting its potential role as a personalized, non-invasive screening tool to optimize nutritional care in IBD.