Abstract 076: Modeling Heterogeneity of Benefit in Patients Eligible for Implantable Cardiac Defibrillators: Insights From the SCD-HeFT Registry

Abstract

Background: Guidelines recommend that patients with heart failure and reduced ejection fraction (<35%; HFrEF), on goal-directed medical therapy be considered for placement of an implantable cardiac defibrillator (ICD) for primary prevention of sudden cardiac death. This recommendation arose from the results of the SCD-HeFT trial, where there was a 7% absolute mortality reduction at 5 years in those who got an ICD. However, the application of these guidelines has been suboptimal and it is unknown if particular patient characteristics drive this benefit. Methods: We used patient-level data from the multi-center SCD-HeFT trial to examine the heterogeneity of the primary outcome (5-year all-cause mortality) in 1676 patients randomized to placebo or ICD (excluding those randomized to amiodarone). We used step-wise variable selection to select the variables for the final model. A cox proportional hazards model was then applied to develop a prediction tool to estimate patient benefit with ICD. Results: After variable selection, the following variables were found to be significantly associated with an increased risk of mortality over 5 years: older age, male sex, ischemic heart disease, diabetes, higher NYHA class, lower LVEF, diuretic use, lower mean blood pressure, higher blood urea nitrogen, and randomization to placebo (see Figure 1). There was a significant interaction of treatment allocation with diabetes, where the benefit of ICDs was greatest amongst those without diabetes (HR placebo vs. ICD amongst diabetics, 6.33, 95% CI 2.40-20.91; p < 0.001). Conversely, among diabetics ICD therapy was associated with an increased risk of mortality when compared to medical therapy (HR 6.39, 95% CI 2.51-15.95; p < 0.001). The c-statistic for the model was 0.73. Conclusion: There is substantial heterogeneity of treatment benefit from an ICD for primary prevention of 5-year all-cause mortality. If validated in a distinct and more contemporary dataset, this model may have an important role in patient selection and shared decision-making.