Abstract 18978: Predictive Modeling of 30-day Readmission of Heart Failure Patients Using Admit Labs, Early Clinical Tests, and History and Physical Exam for In-Hospital Modification of Readmission Risk

Abstract

Background: All 30-day readmission risk models created to date use hospital discharge data gathered at the end of a hospital’s direct patient impact. Since 30-day readmission penalties are charged to hospitals, predictive models should use data from the time of index admission. This study created risk scores for 30-day readmission from factors available at admission among heart failure (HF) patients. Methods: HF patients aged 65 years or older who were admitted as inpatients to Intermountain Healthcare hospitals from 2005-2012 and not discharged to hospice were studied. Data included 205 predictor variables gathered within 24 hours of index admission and the 30-day readmission outcome. Patients were randomly divided into derivation (70%) and validation (30%) groups. Logistic regression was used to build sex-specific predictive models for 30-day readmission, including all variables (full model), only clinical and H&P factors (clinical model), and only laboratory tests (lab model). Results: C-statistics for Intermountain HF risk scores (iHF) were greater than previous 30-day readmission models for HF patients (Table). Odds ratios (OR) for 30-day readmission for iHF (full model) in validation patients were, for females, OR=1.77 (95% CI=1.17, 2.66; p=0.007) and OR=1.62 (CI=1.07, 2.45) for tertiles 3 vs. 1 and 2 vs. 1, respectively. Among males in the validation group, the iHF full model had OR=2.28 (CI=1.53, 3.38; p<0.001) for tertile 3 vs. 1 for 30-day readmission and, for tertile 2 vs. 1, OR=1.59 (CI=1.05, 2.41). Clinical and lab models were also very predictive despite using a reduced set of variables. Conclusions: The iHF readmission risk scores strongly stratified 30-day readmission risk using HF patient data available within 24 hours of index hospital admission. Discrimination was better than other HF 30-day readmission models, including those created for CMS. Use of iHF at the time of admission empowers hospital-based modification of 30-day readmission risk.