Abstract P99: Risk of Hospitalization Among Patients with Heart Failure Within the Veterans Health Administration (VHA)

Background: Patients with heart failure (HF) are at high risk of hospitalization and death. VHA has developed a population-based predictive model to identify high risk patients for case management. Methods: Using clinical and administrative databases, we identified all VA patients with a diagnosis of HF between June, 2008 and May, 2009, then followed the total of 194,062 HF patients for the subsequent 12 months. We used multinomial regression to model the outcome of hospitalization and death jointly. Candidates for predictor variables were related to demographics, medical history, vital status, health care utilization and medication. We randomly split the data 50% to 50% into a training sample and a validation sample. We derived the 30-day and 1-year predictive models from the training sample and validated the models on the other sample. Results: The C-statistics for 30-day and 1-year outcomes were 0.82 and 0.81 for hospitalization, 0.79 and 0.76 for hospitalization or death, respectively. Model calibration was excellent (Figure). For each outcome we stratified patients into 20 risk percentile categories. Risk stratification details were listed (Table). Conclusions: Predictive models can correctly stratify HF patients into risk categories for hospitalization and death. Table Risk Stratification by Outcomes View this table: ![Graphic][1] [1]: /embed/inline-graphic-1.gif