Prediction models for cardiac death in patients with heart failure: Net reclassification improvement with addition of cardiac MIBG results

238 Objectives The aim of this study was to create prediction models for cardiac death (CD) using a multicenter I-123 MIBG database of heart failure (HF) patients with long-term follow-up. Methods Out of a pooled database of HF patients (n=1322) who underwent MIBG studies at 6 centers in Japan, 933 patients whose outcome at 5-years (y) was confirmed were selected (follow-up period, median 7.1y, range 0.1-14.6 y; number of CD: 205). Multivariate logistic regression analysis was performed for 5y CD including HF death, sudden CD and death from acute myocardial infarction. Net reclassification improvement (NRI) analysis was performed using prediction models without and with late heart to mediastinum ratio (HMR). The event risks were classified into 5y CD ranges of ≤5%, 5-25% and ≥25%. Results In the analysis without HMR, 4 parameters (4P) were significant predictors of CD: NYHA functional class, age, gender and ejection fraction. HMR was significant when added to produce a 5-parameter (5P) model. The receiver-operating characteristic area under the curve increased significantly from 0.75 for 4P model to 0.78 for 5P model (p=0.0015). In patients with CD, addition of HMR slightly improved the risk classification (NRI 4.9%, p=0.096). In patients without CD, however, the 5P model with HMR significantly improved NRI by 9.0% compared with results with the 4P model (p Conclusions MIBG HMR was highly predictive of 5y CD. The 5P model including HMR improved risk stratification of HF patients, particularly for reclassifying patients into the lower risk groups.