Abstract 15181: Predictors of Mortality and Morbidity in Patients With Chronic Heart Failure - Analysis of Red-HF Trial

Background: A number of prognostic models have been developed to risk stratify patients with heart failure (HF) and reduced ejection fraction (EF). However, few have included medications, device therapy and natriuretic peptides. Methods and Results: We analyzed data from 2278 patients in Reduction of Events with Darbepoetin Alfa in Heart Failure (RED-HF) trial that tested the effects raising hemoglobin in high-risk anemic patients with HF (median age 72 years; Hg 11.2 g/dL; EF 31%; and GFR 46 ml/min/1.73m2; annualized mortality 14.1%). Multivariable Cox regression models were developed to predict the primary endpoint of all-cause mortality or hospitalization for HF, and all-cause mortality, starting with 43 baseline demographic, clinical, treatment and biochemical variables including NT-proBNP. Nonlinear relationships with continuous variables were examined using restricted cubic splines. Final predictor variables were selected using backward stepwise selection in 1000 bootstrap samples. For the primary outcome, NT-proBNP, hospitalization for HF in past 6-months, heart rate, diastolic BP and bilirubin were the strongest of the 12 final predictors that provided good discrimination (Harrell’s C=0.74) with excellent calibration (Figure). Surprisingly, age, diabetes and EF were not selected. Forcing them in the model did not change discrimination (Harrell’s C=0.74). For all-cause mortality, NT-proBNP, diastolic BP, beta-blocker use, bilirubin and albumin were among the strongest of the 11 final predictors (Harrell’s C=0.71). Again, diabetes and EF were not in the model and forcing them in the model did not improve the C-statistics. Conclusions: In a contemporary well-treated population of severe HF with reduced EF, a multivariable risk model that included medications, devices and natriuretic peptide provided good discrimination and prediction of the primary outcome and all-cause mortality. NT-proBNP was the strongest predictor of both outcomes.