A Data-Driven Approach for Risk Stratification and Outcome Prediction for Pediatric Inflammatory Bowel Disease

This study aims to build a risk prediction model for pediatric Inflammatory Bowel Disease (IBD) to predict patient outcomes. Significant variations in diagnostic testing and treatment exist in the care for pediatric IBD patients. Previous work has developed quality improvement (QI) methods and instruments such as feasibility of the short Pediatric Crohn's Disease Activity Indices (PCDAI), and care stratification scores to address the variability and improve quality. Given the availability of patient-level data, and recent advances in risk prediction techniques through statistics and machine learning, there is an opportunity to take one step further, to examine the common patterns of progressions of disease activities within stratified cohorts, and use these insights for patient-specific prediction of future events.Leveraging longitudinal data from ICN registry and PCORnet, and previously developed methods in assessing and stratifying disease activities, this study proposes to apply data analytics and visualization to facilitate efficient decision support for personalized care and risk prediction in pediatric Crohn’s disease. 

We aim to develop a machine-learning based clinical pathway learning algorithm for pediatric IBD that predicts for 3 outcomes of interests: (1) sustained remission/flare, (2) surgery, and (3) hospitalization. Stratifying patients by their risk levels using PCDAI and sPCDAI, we will infer the common disease progression trajectories that are unique to each patient subpopulation from data, which we further use for prediction of future patient states. We hypothesize that we will be able to identify predictors from data, such as age at diagnosis, phenotype, disease activities, growth in terms of Z-score, and number of hospitalizations