Predicting New Heart Failure Hospitalization Among Patients with an Existing Heart Failure Diagnosis

Introduction Preventing re-hospitalizations among heart failure (HF) has been the focus of a great deal of research but determining which HF patients will experience a new initial hospitalization has been less well-studied. Hypothesis Identifying patients who will be hospitalized for HF within the ensuing year can be determined from data routinely collected in electronic medical records (EMRs). Methods We used the EMRs of Kaiser Permanente Northwest, a 550,000-member integrated delivery system in Portland, OR. Using a case-control design, we identified 1,000 patients with a HF hospitalization in 2016/2017 and no prior hospitalization in the past 12 months and matched them on age and sex to 1,000 patients with an existing HF diagnosis in the same year but who were not hospitalized as of their match's hospitalization (index) date. We collected clinical data from the EMR during the 12 months preceding the index date, including clinical characteristics such as obesity (BMI ≥ 30kg/m 2 ), smoking status, and blood pressure (BP) ≥ 140/90, as well as existing comorbidities including diabetes (DM), coronary artery disease (CAD), atrial fibrillation (AF), and valve disorders for comparison between patients who were and were not hospitalized. Using multivariable logistic regression, we identified characteristics predictive of HF hospitalization and determined the discrimination of the model using the C statistic. Results Patients in both groups were 75±10 years old and 48% were men. All variables included in the multivariable model were highly predictive of HF hospitalization (table). For example, African-American patients more than twice as likely to be hospitalized (OR 2.25, 95% CI 1.28-3.98). Patients with high BP, CAD, DM, AF or valve disorders were 86%, 30%, 65%, 84% and 92% more likely to be hospitalized, respectively. However, the model provided only moderate discrimination (C statistic: 0.66). Conclusions Comorbid conditions can help identify which HF patients will be hospitalized within the next year. Because preventing hospitalization is the most effective way to avoid re-hospitalizations, more robust predictive models are needed. Non-clinical data such as health behaviors, social determinants, and social support may be needed to identify HF patients at greatest hospitalization risk and who are in need of intensive care management. Table. Risk of hospitalization for heart failure.