Aim: Validation of the Seattle Heart Failure Model (SHFM) for predicting the risk of death in a population different than the derivation cohort.
Methods: In a retrospective analysis of a cohort of chronic heart failure patients with left ventricular systolic dysfunction, consecutively referred between 2000 and 2011, we computed the score, according to characteristics at referral. We compared the observed risk of death with that predicted by the model, using receiver operating characteristic (ROC) curves to assess discrimination and a goodness-of-fit test for the comparison of predicted and observed risks.
Results: In 565 patients, 68.5% were men, the median age was 70 years, 46.0% had ischemic cause, 89.7% moderate-severe left ventricular systolic dysfunction and 61.2% New York Heart Association class II. The risk of death increased progressively with the model's score, with an area under the ROC curve between 0.69 and 0.72 when considering different follow-up periods. The model underestimated the risk of death (observed vs. predicted: 12.2 vs. 10.4%, P < 0.001; 28.1 vs. 25.1%, P < 0.001; and 43.4 vs. 35.7%, P < 0.001 at 1, 3 and 5 years, respectively). Accurate predictions, with nonsignificant differences between observed and predicted risks in a goodness-of-fit test, were obtained after recalibration.
Conclusion: In this study, the SHFM substantially underestimated the absolute risk of death in ambulatory chronic heart failure patients, mostly nonischemic and elderly. After adjustment for sample-specific circumstances, the recalibrated model demonstrated to be credible in clinical practice and may provide useful information to physicians.