Abstract 2956: Validation of the Seattle Heart Failure Model in a Heart Failure Population in a 20 Hospital System in the Intermountain West
Background: Management of heart failure (HF) remains complex with low 5-year survival. Since HF is the only cardiovascular disease with increasing incidence, improving management is crucial. The Seattle Heart Failure Model (SHFM) is a recently-described risk score derived from clinical trials data that may enable the prediction of survival in a HF patient. This study sought to validate the SHFM in an independent, non-clinical trial-based HF population.
Methods: Patients (N = 4,077) from the Intermountain Heart Collaborative Study registry in a 20-hospital system with a HF diagnosis were evaluated (average follow-up was 4.4 ± 3.1 years, range 0.4–12.2 years). Patient values for SHFM variables were multiplied by model coefficients to calculate a risk score for each patient (missing values for NYHA class, lymphocytes, uric acid, EF, and total cholesterol were estimated by regression or assigned from NYHA class-specific means). ROC area under the curve (AUC) provided SHFM predictive ability for a composite endpoint of survival free from death, transplant, or left ventricular assist device implant. Model accuracy was evaluated for deciles of SHFM score by comparing actual survival using Kaplan-Meier life table methods to predicted survival at 1- and 5-year time points.
Results: Patient age was 67 ± 13 years (range 19–96) and 61% were male. Risk scores ranged from -1.95 to 4.25. AUCs were 0.68 (95% CI: 0.66, 0.70) and 0.65 (95% CI: 0.63, 0.67) for 1- and 5-year survival, respectively. Predicted 1-year survival was 97% in decile 1 and 74% and in decile 10 (average 91%), with correlation of 0.99 to actual survival. Predicted 5-year survival was 88% in decile 1 and 22% in decile 10 (average 64%), with correlation of 0.98 to actual survival. Among the subset of patients with full data for lymphocytes, uric acid, EF, and total cholesterol (N = 2121), AUCs were 0.70 and 0.66 for 1- and 5-year survival, respectively.
Discussion: This study validates the predictive ability of the SHFM for survival among HF patients in a community-based population. The model is accurate and has similar AUCs to those from the data on which models were initially fit. These findings suggest that the SHFM should be further evaluated for clinical use in modifying patient care and improving outcomes.