Abstract 19266: Identifying Patients at High Risk for Readmission following Treatment for Acute Myocardial Infarction: a Data-Centric Approach
Background: The American College of Cardiology Patient Navigator Program (ACC-PNP) was launched to address the growing need for a reduction in 30-day readmission rates following acute myocardial infarction (AMI). To facilitate identification of patients at greatest risk and target for intervention, we developed a risk score that runs in real-time based on a predictive model derived from electronic health records (EHR).
Methods: To develop a comprehensive assessment of risk factors contributing to 30-day readmission following AMI, we assessed Medicare penalty eligible patients discharged from our institution with a primary diagnosis of AMI between Jan 1, 2010 and June 30, 2012. EHR-derived variables (retrospective and prospectively derived) from 780 eligible patients were used to automatically compute likelihoods of 30-day readmissions. Patients were randomly divided into a derivation (60%) and validation (40%) cohort. Logistic regression was employed, using the LASSO technique for variable selection. Model optimization was performed via cross-validation.
Results: 10 variables were selected for the predictive model from a set of 3,110 EHR-derived variables under consideration. The model provides an institution-specific readmission risk score (C statistic 0.67 for the validation cohort) with improved accuracy compared with a model created for use by CMS (Yale risk score, C statistic 0.63). History of obesity, pneumonia, renal failure, prior cardiovascular disease, and laboratory results of hemoglobin, calcium, INR, and albumin were found to be most predictive of readmission. In distinction to the model used by CMS for retrospective analyses, the institute-specific readmission score does not rely on post-hospitalization coding. Patients risk is calculated in real-time and incorporated into our institution’s ACC-PNP to identify those most likely to benefit from a targeted intervention.
Conclusion: An institution-specific risk score for post-AMI readmission can be derived from historical and de-novo EHR data. The score can be computed from patient data in real-time to maximize the utility of a post-discharge care-coordination program.
Author Disclosures: N.C. Baker: None. M. Bayati: None. R. Torguson: None. K. Mack: None. H. Rappaport: None. E. Horvitz: None. R. Waksman: None.
- © 2014 by American Heart Association, Inc.