Hybrid mechanistic-ML model of Crohn’s disease to predict temporal progression of mucosal damage in patients

Quantitative systems pharmacology models that provide mechanistic understanding of a disease are increasingly being used at various stages of drug development, for gaining insights into both safety and efficacy of a treatment. To predict the efficacy of a novel compound or a combination treatment in late-stage development, QSP models should have the ability to predict clinical scores, in addition to biomarkers. These clinical scores often constitute subjective assessments either by patients or physicians which make it challenging to predict using a mechanistic approach. In Crohn’s Disease, for example, SES-CD (Simple Endoscopic Score for Crohn’s Disease) is used to measure the endoscopic activity in patients by evaluating endoscopy images. In addition to being an endpoint in clinical trials, endoscopic healing is one of the primary treatment goals in IBD as recommended by STRIDE consensus. However, physicians are often required to make treatment decisions based on limited objective information about the state of the patient's gastrointestinal tissue while aiming to achieve mucosal healing. Tools to predict changes in gut mucosal condition with treatment are needed. In this presentation, we will share a hybrid mechanistic-ML model that was developed to predict long-term mucosal damage and endoscopic outcomes in patients.