Classification Models for Predicting Inflammatory Bowel Disease Healthcare Utilization Dmitriy Babichenko 1 a , Behnam Rahdari 1 b , Ben Stein 1 , Suraj Subramanian 1 , Claudia Ramos Rivers 2 , Gong Tang 3 and David Binion 2 1 School of Computing and Information, University of Pittsburgh, Pittsburgh, PA, U.S.A. 2 School of Public Health, University of Pittsburgh, Pittsburgh, PA, U.S.A. 3 School of Medicine, University of Pittsburgh, Pittsburgh, PA, U.S.A. Keywords: Inflammatory Bowel Disease, Healthcare Utilization, Machine Learning, Classification, Deep Learning, Clinical Decision Support Systems. Abstract: Objective. Inflammatory Bowel Disorders (IBD) is a group of gastric disorders that include well-known maladies such as Crohn’s disease and Ulcerative Colitis (UC), as well as a number of other gastric disorders that are not well classified. Subgroups of patients contribute disproportionately to treatment costs. This work aims to create and evaluate machine learning models designed to use demographic and clinical predictors of IBD to predict which patients would fall into the “high healthcare utilization” category. Materials and Methods. A series of machine learning models were trained on a dataset extracted from a prospective natural history registry from a tertiary IBD center and associated healthcare charges. The models were trained to predict which patients are likely to have the highest healthcare utilization charges (top 15%). Results. A gradient-boosted trees classification model (accuracy 0.898, AUC 0.748) performed best out of the 12 evaluated modeling approaches. Conclusion. Classification models such as the ones evaluated in this work provide a reasonable basis for a clinical decision support system designed to predict which IBD patients are likely to become high expenditure. 1 INTRODUCTION Inflammatory Bowel Disorders (IBD) is a group of gastric disorders that include well-known maladies such as Crohn‘s Disease and Ulcerative Colitis (UC), as well as a number of other gastric disorders that are not well classified. According to the Center for Disease Control (CDC), “IBD is one of the five most prevalent gastrointestinal disease burdens in the United States, with an overall healthcare cost of more than $1.7 billion”. Currently, there is no medical cure and IBD patients commonly require a lifetime of care. In the United States, IBD accounts for more than 700,000 physician visits, 100,000 hospitaliza- tions, and disability in 119,000 patients (CDC 2014). IBD patients most often receive care in physi- cians‘ offices or other outpatient sites, with hospital- ization required only for severe disease presentation, to treat certain complications, and for surgery. a https://orcid.org/0000-0003-1187-6684 b https://orcid.org/0000-0001-6514-912X In recent decades the prevalence of IBD and the associated treatment costs have risen dramati- cally (Kappelman et al., 2008; Molodecky et al., 2012; Kappelman et al., 2013). In 2004, there were 1.1 million ambulatory care visits and 1.8 million prescriptions written for medications to treat Crohn’s disease and 716,000 ambulatory care visits and 2.1 million prescriptions written for medications to treat UC (Everhart, 2008). The hospitalization rate also increased signifi- cantly during this period from 44.2 to 59.7 per 100,000 population, with the mean hospitalization costs of $11,345 for Crohn’s disease and $13,412 for ulcerative colitis (Kappelman et al., 2008; Molodecky et al., 2012; Kappelman et al., 2013). A number of research efforts have produced ma- chine learning (ML) models to predict remission in Crohn’s patients (Waljee et al., 2019), patients’ re- sponse to drug therapies (Waljee et al., 2010), and assess IBD risk (Wei et al., 2013). Other studies relied on classical statistical approaches to identify that a subgroup of IBD patients exhibit “high preva- 154 Babichenko, D., Rahdari, B., Stein, B., Subramanian, S., Rivers, C., Tang, G. and Binion, D. Classification Models for Predicting Inflammatory Bowel Disease Healthcare Utilization. DOI: 10.5220/0010852100003123 In Proceedings of the 15th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2022) - Volume 5: HEALTHINF, pages 154-161 ISBN: 978-989-758-552-4; ISSN: 2184-4305 Copyright c  2022 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved