P1150 Artificial-intelligence-based prediction of relapse risk in Inflammatory Bowel Disease: a pilot study from Marrakech

Abstract

Abstract Background Predicting tomorrow’s relapse today — AI as a new ally for personalized IBD care. Artificial-intelligence (AI) models can identify patients at high risk of inflammatory bowel disease (IBD) relapse using routine clinical and biological data, helping optimize individualized follow-up and therapy adjustment. Aim To develop and validate an AI-based predictive model for 12-month clinical relapse among Crohn’s disease (CD) and ulcerative colitis (UC) patients followed at a Moroccan tertiary center. Methods Retrospective single-center study including 233 patients (135 CD, 98 UC) followed between 2018 and 2024 at Mohammed VI University Hospital, Marrakech. Variables included age, sex, BMI, disease duration, maintenance therapy (immunosuppressants/biologics), CRP, albumin, fecal calprotectin, and prior severe flares. Two predictive models were compared: logistic regression and classification-and-regression tree (CART). Performance was assessed using the area under the receiver operating characteristic curve (AUC), sensitivity, and specificity. Results A total of 233 IBD patients were included, with a mean age of 36.8 ± 12.4 years and a sex ratio (M/F) of 0.87. The mean disease duration was 8.2 ± 4.9 years, and the median BMI was 23.7 kg/m². Among the entire cohort, 42% (n = 98) experienced at least one clinical relapse during the 12-month follow-up. Clinical and biological differences between relapsers and non-relapsersPatients who relapsed had significantly higher baseline fecal calprotectin (median 410 µg/g [IQR 305–720] vs. 110 µg/g [IQR 60–180], p < 0.001) and CRP levels (14.6 ± 9.2 vs. 6.8 ± 5.7 mg/L, p = 0.004).Hypoalbuminemia (<35 g/L) was more common in relapsers (38% vs. 19%, p = 0.03), while previous severe flares within the past two years were reported in 61% of relapsers versus 33% of non-relapsers (p = 0.02).No significant differences were observed for age, sex, or BMI (all p > 0.05). Traditional logistic regression model identified fecal calprotectin, CRP, and previous severe flare as independent predictors of relapse. The model yielded an AUC = 0.71 (95% CI: 0.64–0.78), sensitivity 68%, and specificity 70%. AI-based CART model integrated nonlinear interactions among biomarkers and treatment variables. Its top decision nodes were fecal calprotectin (>250 µg/g), CRP (>10 mg/L), and albumin (< 35 g/L).This model demonstrated superior predictive power with sensitivity = 81%, and specificity = 77%, achieving overall accuracy of 79%. Conclusion AI-based prediction markedly improves relapse risk stratification in IBD compared with conventional clinical models.This locally trained, low-cost model demonstrates feasibility and accuracy within Moroccan hospital settings, paving the way for data-driven, personalized IBD management. Conflict of interest: Dr. Aouroud, Hala: No conflict of interest Aouroud, Meryem: No conflict of interest Nacir, Oussama: No conflict of interest lairani, fatima ezzahra: No conflict of interest ait errami, adil: No conflict of interest Oubaha, Sofia: No conflict of interest Samlani, Zouhour: No conflict of interest Krati, Khadija: No conflict of interest

Ähnliche Arbeiten