Systematic Analysis of Risk Factors for 30-Day Readmission in Gastrointestinal Bleeding

Abstract

Introduction: Gastrointestinal bleeding (GIB) is a frequent cause of hospital admission in the United States with estimated rates of hospitalization of 375 per 100,000 individuals. Readmission events are associated with increased healthcare costs, morbidity and mortality. Our aim was to utilize the novel National Readmission Database (NRD) to determine the risk factors for 30-day readmission in patients with GIB and evaluate the predictive ability of the created model. Methods: We utilized the Healthcare Cost and Utilization Project's (HCUP) 2013 and 2014 NRD. Index admissions had either a primary or secondary diagnosis of upper GIB (UGIB) or a primary diagnosis of lower GIB (LGIB) utilizing previously identified diagnosis codes. Patients with an index admission death or an index admission in December were excluded. The primary outcome of interest was risk factors for 30-day readmission. Data was analyzed using Student's t-test and stepwise, backward multivariate logistic regression analysis. Results: We identified 97,919 (2013) and 102,010 (2014) index admissions with UGIB and 54,480 (2013) and 57,053 (2014) admissions with LGIB. In the combined data set the average patient age was 67.9 years and chronic conditions per patient was 6.0. Utilizing the combined data sets, within 30 days, 30,973 (15.5%) readmissions were identified for UGIB and 13,234 (11.9%) for LGIB. Thirty-one variables were included in the UGIB model and twenty-one in the LGIB model. Common variables included in both models included insurance types Medicare and Medicaid, the performance of a blood transfusion, and the comorbidities lung disease, malignancy, renal failure, coagulopathy, cirrhosis and diabetes. Model performance created utilizing the 2013 dataset was evaluated using a receiver operating characteristic curve (AUROC) and validated in the 2014 dataset. The AUROC for UGIB was 0.641 (95% CI 0.636-0.646). Validation of the model utilizing 2014 data demonstrated a similar AUROC of 0.639 (95% CI 0.634-0.643, P=0.518). Similar results were obtained in LGIB with an AUROC of (2013): 0.633 (95% CI 0.626-0.64) and (2014): 0.627 (95% CI 0.62-0.634, P= 0.244). Conclusion: Utilizing a cross-sectional nationally available dataset we were able to identify readmission risk factors among patients with gastrointestinal bleeding that is replicated across data sets with moderate predictive ability. Further studies should aim at creating interventions in this identifiable population in order to limit readmission events.