Su1785 RECENT TRENDS IN DIAGNOSTIC PREVALENCE AND TREATMENT PATTERNS AMONG ADULT CROHN'S DISEASE PATIENTS IN THE UNITED STATES Theresa Hunter, April N. Naegeli, Yan Dong, Casey Choong, Diana Stefani-Hunyady Background: There has been much variation between epidemiological studies that report the prevalence of Crohn's Disease (CD). This study aimed to analyze the diagnostic annual prevalence rates and treatment patterns of CD patients in the United States (US) adult insured population from 2007 to 2017. Methods: Trends in CD prevalence were calculated for the 11-year period covering January 1, 2007 to December 31, 2017. Adult (18+ years old) CD patients were included in this retrospective analysis of medical and pharmacy claims data from the IBM Marketscan Commercial, Medicaid and Medicare-Supplemental Claims database. Prevalence was determined as having ≥1 CD diagnostic codes (ICD-9: 555.x; ICD- 10:K50.x) within the calendar year. Patients with an Ulcerative Colitis diagnosis (ICD-9: 556.x; ICD-10: K51.x) were excluded. Prevalence rates in the database were determined and age- and gender-adjusted rates were projected to the US population in 2017. Trends in treatment patterns were also analyzed. Results: The CD adult prevalence increased from 0.21% to 0.28% from 2007 to 2017. The mean age between 2007 and 2017 ranged from 46.34-47.91 years. Consistently throughout the years, approximately 55% of the CD patients were female. Rates biologic use increased, while rates of 5-ASA and immunomodulators decreased. Corticosteroids and opioids remained stable (Figure 1). Conclusions: The preva- lence of CD diagnosis codes increased between 2007 and 2017, and is projected to affect approximately 700,000 US adults in 2017. S-611 AGA Abstracts Su1786 REDUCTION IN COSTS OF CARE IN PATIENTS WITH CROHN'S DISEASE IN A COMMUNITY PRACTICE-BASED MEDICAL HOME OVER TWO YEARS: A PROPENSITY SCORE MATCHED COHORT ANALYSIS Siddharth Singh, Joel Brill, Derek Blankenship, Leanne Metcalfe, William J. Sandborn, Lawrence R. Kosinski Background: Project Sonar, a joint venture of Blue Cross Blue Shield Illinois and SonarMD, Inc., is a technology-enabled care management platform for patients with inflammatory bowel disease (IBD), focused on patient engagement, delivery system redesign, and clinical decision support to reduce costs of care. We compared overall costs of care of patients with CD enrolled in Project Sonar, with propensity score-matched patients with CD who were not enrolled in the platform, over a 24-month period. Methods: Project Sonar cohort included 176 patients with CD with continuous enrollment in the project (01/01/16 – 12/ 31/17). Controls were patients with CD not enrolled in Project Sonar, during the same time period, and matched based on age, sex, socioeconomic status, healthcare utilization in the preceding year (inpatient, emergency room [ER], outpatient costs) and insurance-attributed risk scores, using 1:1 propensity score. Primary outcome of interest was difference in overall medical costs (inpatient/ER costs, outpatient costs). For a subset of patients, all pharmaceutical cost was available. Difference between groups was measured as difference of differences (DoD) in costs (costs at end of intervention – baseline costs, in intervention group and controls), using Wilcoxon signed-rank test. Results: Over a two-year period, overall mean medical costs per member per month (PMPM) of Project Sonar patients decreased from $908 to $760 (-12.4%) vs. increase from $1032 to $1472 for controls (+42.5%), with mean DoD of $551 (p=0.06). Hospitalization rate declined by 16.7% in Project Sonar patients vs. increase of 130.8% in control patients, corresponding to a mean difference in hospitalization costs of -$339 PMPM (p=0.08) and a difference of -$185 PMPM in outpatient, non-ER use (p=0.04). Among a subset of patients who were engaged in the program (responded to 80% of monthly pings; n=126, 71.6%), mean difference in monthly medical costs was significantly lower than matched controls (-$543 PMPM, p=0.04). This difference was attributed to decrease in rate of hospitalization in Project Sonar patients (p=0.03), with an accompanying decline in costs of hospitalization (DoD, -$355 PMPM). For a subset of patients for whom pharmaceutical cost was available, total pharmacological cost was not different between Project Sonar patients and controls (p=0.46) and total cost of care were lower in Project Sonar patients (DoD, -$348 PMPM, p=0.14). Conclusions: In this updated propensity score matched cohort analysis, participants in Project Sonar, especially engaged patients, continue to demonstrate a trend towards lower total medical costs, as compared to control patients. This is primarily attributed to lowers rates and costs of hospitalization and outpatient, non- ER use, without any increase in pharmacological drug costs. Su1787 DEVELOPMENT OF A DEEP LEARNING ALGORITHM FOR DIFFERENTIAL DIAGNOSIS BETWEEN CROHN'S DISEASE AND INTESTINAL TUBERCULOSIS Youn-i Choi, Yoon Jae Kim, Dong Kyun Park Background: Distinguishing between Crohn's disease (CD) and intestinal tuberculosis (ITB) is challenging for physicians not only where ITB and CD is prevalent but also where ITB is rare so expert resources are limited. Though colonoscopy is useful tool to discriminate CD from ITB, it is not easy to physicians to perform constant results. Here we developed a deep neural network (DNN) model for discriminating CD from ITB using not only laboratory and radiologic parameters but also colonoscopic images though convolutional neural network (CNN). METHODS: We collected 2000 images of CD (active phase; n=1053, scar phase ;n=453) and 681 images of ITB (active phase; n=126, scar phase; n=451) obtained from the picture archiving and communications system in Gil medical cancer, a tertiary hospital in Korea. Histologic findings from the lesions and results of 8-12weekds empirical anti- tuberculosis treatments were also collected and used as the reference standards. The images, clinical, and laboratory data were used to train deep neural network as supervised learning. A test set of images (100 active phase CD and 100 active phase ITB), obtained from patients who underwent colonoscopy from January 2010 to June 2018, was then used to test the diagnostic ability of the DNN-CNN vs endoscopists who consists 1-year fellow ship, 3- year fellow ship, and more than 10 year experienced professors. The accuracy, sensitivity, specificity, positive predictive value (PPV, NPV, and diagnostic time were compared among DNN-CNN, the novice(1-year fellow ship, 3-year fellow ship) , and experienced endoscopists. This study was designed to differentiate CD from ITB though clinical, laboratory and colonoscopic images. Result In the test set, the DNN-CNN differentiated CD from ITB with 100% sensitivity, 100% specificity, a PPV of 100% and a NPV of 100%. 40% of the novice AGA Abstracts