S0770 A Predictive Model of Length of Stay in Inflammatory Bowel Disease Patients Hospitalized With Flare: A Multicenter Study

Abstract

INTRODUCTION: Hospitalization for flare of inflammatory bowel disease (IBD) is associated with significant morbidity and healthcare costs. We aimed to examine the relationship between IBD severity at presentation and adverse outcomes in patients hospitalized with flare. Additionally, we aimed to create and validate a predictive model for length of stay (LOS) using markers of disease severity. METHODS: We conducted a retrospective cohort study of IBD patients hospitalized with flare at two urban academic medical centers. We collected demographic and IBD-related data including the presence of Clostridioides difficile and markers of disease severity at presentation. The primary outcome was a composite of adverse inpatient IBD outcomes, including LOS >7 days, anti-TNF administration, and surgery. The data was split into training (70%) and validation (30%) samples. Employing variables of disease severity, models (including logistic regression, Classification and Regression Tree (CART), and Random Forest (RF) modeling techniques) were created to predict LOS >7 days using the training sample. These models were adjusted for surgery and anti-TNF administration, and their performance was subsequently validated. RESULTS: A total of 187 IBD patients hospitalized with flare were included (Table 1). The composite primary outcome was achieved in 71 patients (38%) with 51 (27%) hospitalized for >7 days. In univariate analyses, C-reactive protein and C. difficile positivity correlated with anti-TNF administration and surgery (Table 2). Adjusting for anti-TNF administration and surgery, tachycardia (OR 1.07; 95% CI 1.05–1.10), hypotension (1.07; 1.03–1.11), hypoalbuminemia (2.87; 1.81–4.74), leukocytosis (1.17; 1.09–1.27), anemia (1.10; 1.05–1.15) and C. difficile positivity (2.63; 1.27–5.47) were predictive of prolonged LOS (Table 2). In multivariate analyses, tachycardia predicted the primary composite outcome of all adverse effects (OR 1.07; 95% CI 1.03–1.11). C. difficile positivity (OR 4.33; 95% CI 1.36–14.9), hypoalbuminemia (3.25; 1.29–9.01), and tachycardia (1.11; 1.06–1.16) predicted prolonged LOS (>7 days). The CART and RF models had acceptable accuracy, sensitivity, and specificity for predicting LOS >7 days (Table 3). CONCLUSION: C. difficile positivity, hypoalbuminemia, and tachycardia at presentation predicted prolonged length of stay in a multicenter cohort of IBD patients hospitalized with flare. CART and Random Forest models perform well in predicting prolonged length of stay in these patients.Table 1.: DemographicsTable 2.: Univariate logistic regression of disease severity and outcomes, odds ratio (OR). OR for systolic and diastolic blood pressure, albumin, and hematocrit is shown as 1/OR for interpretabilityTable 3.: Performance of predictive models for length of stay >7 days