• Clinical Indicators that Predict Readmission Risk in Patients with Acute Myocardial Infarction, Heart Failure, and Pneumonia

      Chen, Weihua; The University of Arizona College of Medicine - Phoenix; Antonescu, Corneliu; Holland, William (The University of Arizona., 2017-04-28)
      BACKGROUND: In order to improve the quality and efficacy of healthcare while reducing the overall cost to deliver that healthcare, it has become increasingly important to manage utilization of services for populations of patients. Healthcare systems are aggressively working to identify patients at risk for hospital readmissions. Although readmission rates have been studied before, parameters for identifying patients at risk for readmission appear to vary depending the patient population. We will examine existing Electronic Health Record (EHR) data at Banner Health to establish what parameters are clinical indicators for readmission risk. Three conditions were identified by the CMS to have high and costly readmissions rates; heart failure (HF), acute myocardial infarction (AMI), and pneumonia. This study will focus on attempting to determine the primary predictive variables for these three conditions in order to have maximum impact on cost savings. METHODS: A literature review was done and 68 possible risk variables were identified. Of these, 30 of the variables were identifiable within the EHR system. Inclusion criteria for individual patient records are that they had an index admission secondary to AMI, heart failure, or pneumonia and that they had a subsequent readmission within 30 days of the index admission. Pediatric populations were not studied since they have unique factors for readmission that are not generalizable. Logistics regression was applied to all data including data with missing data rows. This allowed all coefficients to be interpreted for significance. This model was termed the full model. Variables that were determined to be insignificant were subsequently removed to create a new reduced model. Chi square testing was then done to compare the reduced model to the full model to determine if any significant differences existed between the two. RESULTS: Several variables were determined to be the significant predictors of readmission. The final reduced model had 19 predictors. When analyzed using ROC analysis, the area under the curve (AUC) was 0.64. CONCLUSION: Several variables were identified that could be significant contributors to readmission risk. The final model had an AUC on it ROC of 0.64 suggesting that it would only have poor to moderate clinical value for predicting readmission.