Abstract 11502: Modifying the Seattle Heart Failure Model Improves Survival Prediction in Heart Failure Patients
Background: The Seattle Heart Failure Model (SHFM) had been found to underestimate survival when validated in non-trial populations. We propose a modification of the SHFM to improve prediction of survival in heart failure when the model is applied in a clinical practice setting.
Methods: Baseline data used to compute the SHFM score were collected on 1521 patients with chronic heart failure who were referred to the University of Louvain between January 1991 and March 2010. Patients were prospectively followed to ascertain death, cardiac transplantation, and left ventricular assist device implantation that were modeled by the SHFM. We refitted the SHFM using different parameterizations of the baseline survival function and compared the predicted to the observed event rates up to 5 years.
Results: No patients were lost to follow-up. The median duration of alive follow-up was 5.5 years. The observed overall 1-year to 5-year event-free survival were 78.0%, 69.7%, 62.9%, 57.0%, and 50.2% respectively. The calculated SHFM score ranged from -1.57 to +3.72, with a median of +0.62. The SHFM showed good discrimination across all years (c-statistic=0.721-0.742, all p<0.0001). Without recalibration, the SHFM consistently overestimated survival (Figure). Refitting the baseline survival function using a Weibull model yielded a shape parameter for hazard that significantly differed from unity (0.773; 95% CI 0.768-0.777, p<0.0001), implying that the exponential model used originally by the SHFM was inadequate. The refitted SHFM showed improved predication of event-free survival as compared to the original SHFM (Figure). Correlations between the predicted and the observed event rates for the refitted SHFM were excellent across all years (r=0.963-0.983, all p<0.0001).
Conclusions: Modifying the baseline survival function used by the SHFM improved the model's prediction of event-free survival in patients with heart failure when used in a clinical practice setting.
- © 2012 by American Heart Association, Inc.