Early Diagnosis of Crohn's Disease with Perianal Fistula as the Initial Clinical Presentation: A Multicenter Derivation and Validation of a Perianal Fistulizing Crohn's Disease Risk Prediction Web Tool

Background and Aim: Diagnosing Crohn's disease (CD) can be challenging, especially when perianal fistula is the initial symptom. Our aim was to develop and validate a predictive model, as well as to create a visual web tool for early diagnosis of CD in patients presenting with perianal fistula. Materials and Methods: This retrospective, multicenter, validated study included patients diagnosed with either PFCD or CGF who underwent their initial perianal fistula surgery following rectal MRI at three Chinese centers between September 2016 and December 2020. A random forest classification model for the diagnosis of PFCD was trained using the derivation cohort (n=550), which was randomly divided into training and test sets at a 7:3 ratio. Validation was performed using data from two external centers (n=300). For model interpretation, the SHAP framework was utilized. The validated model was then incorporated into a web tool to calculate patient-specific risk. Results: In the derivation cohort, 12 features were identified as important for diagnosing PFCD, with the top 5 variables being rectal wall ulceration, rectal wall thickening, submucosal fistula, T2 hyperintensity and age. A random forest classification model was then developed using these top five features, demonstrating a higher AUROC of 0.9425 (95% CI: 0.8943-0.9906). In the validation cohort, the model was accurately predicted with AUROC values of 0.9187 (95% CI: 0.8620-0.9754) and 0.9341 (95% CI: 0.8876-0.9806) respectively. The SHAP method revealed that rectal wall ulceration, rectal wall thickening, submucosal fistula, and T2 hyperintensity were risk factors, while age was a protective factor for diagnosing PFCD. Additionally, we have developed a publicly available web-based application called the Perianal Fistulizing Crohn's Disease Risk Estimator (PFCD-RE). Conclusion: We have developed a multimodal machine learning model and created a web tool called PFCD-RE, which clinician can use to predict and display the risk of CD for patients presenting initially with perianal fistula. Funding: This work was supported by the Jiangsu Provincial key R&D Programme (Social Development, BE2023792), the QingFeng Scientific Research Fund of the China Crohn’s & Colitis Foundation (CCCF-QF-2022B58-5), the Developing Program for High-level Academic Talent from Jiangsu Hospital of Chinese Medicine (y2021rc27), and the Priority Academic Program Development of Jiangsu Higher Education Institutions. Declaration of Interest: The authors declare that there no conflicts of interests. Ethical Approval: The Institutional Review Board of Affiliate Hospital of Nanjing University of Chinese Medicine granted approval for the study with reference number [2023NL-111-01].