Vedolizumab in inflammatory bowel disease: Real-world outcomes and their prediction with machine learning—the IG-IBD LIVE study

Background and aims: Real-world studies on vedolizumab in inflammatory bowel disease (IBD) are often limited by small sample size and short follow-up. In this study, we investigated the 2-year effectiveness and safety of vedolizumab in patients with IBD, and applied eXplainable Artificial Intelligence (XAI) to identify predictors of both. Methods: The Long-term Italian Vedolizumab Effectiveness (LIVE) study is multicentric, ambispective, observational study enrolling 1111 IBD patients (563 Crohn's disease, CD, 542 ulcerative colitis, UC). Steroid-free clinical remission (SFCR) at 24 months was the primary endpoint. A XAI model (eXtreme Gradient Boosting, XGB) was applied to identify the main clinical predictors of SFCR and development of adverse events (AEs). Results: Rates of SFCR at 24 months were 31.6 % and 39.7 % in CD and UC patients, and 0.14 AEs per patient-year was recorded. On XGB analysis, previous exposure to anti-TNFα and older age were the most important drivers for the prediction of SFCR; lower baseline CRP levels and fewer comorbidities were the most important features associated with no development of AEs. Conclusions: Vedolizumab is effective and safe in IBD patients. XAI yielded promising results in identifying the most important predictors of SFCR and development of AEs.