Background: Readmissions after hospitalization for acute myocardial infarction (AMI) are common, but the few available risk prediction models have poor predictive ability and are not readily usable in real-time.
Objectives: To develop and validate an AMI readmission risk prediction model from electronic health record (EHR) data available on the first day of hospitalization, and to compare model performance to the Centers for Medicare and Medicaid Services (CMS) AMI model and a validated multi-condition EHR model.
Methods: EHR data from AMI readmissions from 6 diverse hospitals in north Texas from 2009-2010 were used to derive a model predicting all-cause non-elective 30-day readmissions to any of 75 hospitals in the region, which was then validated using five-fold cross-validation.
Results: Of 826 consecutive index AMI admissions, 13% were followed by a 30-day readmission. The AMI READMITS score included seven predictors, all ascertainable within the first 24 hours of hospitalization (Table 1A). The AMI READMITS score was strongly associated with 30-day readmission in our cross-validation cohort: ≤13 points = extremely low risk (bottom quintile, mean predicted risk 3%); 14-15 points = low risk (4th quintile, predicted risk 7%); 16-17 points = moderate risk (3rd quintile, predicted risk 11%); 18-19 points = high risk (2nd quintile, predicted risk 16%); and ≥20 points = extremely high risk (top quintile, predicted risk 35%). The READMITS score had good discrimination with comparable performance to the CMS model in our cohort; it had improved discrimination, reclassification, and calibration compared to a multi-condition EHR model (Table 1B).
Conclusions: The AMI READMITS score accurately stratifies patients hospitalized with AMI into groups at varying risk of 30-day readmission. Unlike claims-based models which require data not available until after discharge, READMITS is parsimonious, easy to implement, and leverages actionable real-time data available from the EHR within the first 24 hours of hospitalization to enable early prospective identification of high-risk AMI patients for targeted readmissions reduction interventions.