IMPORTANCE Improvement in left ventricular ejection fraction (EF) to >35%occurs in many patients with reduced EF at baseline. To our knowledge, whether implantable cardioverter defibrillator (ICD) therapy improves survival for these patients is unknown. OBJECTIVE To examine the efficacy of ICD therapy in reducing risk of all-cause mortality and sudden cardiac death among patients with an EF35%at baseline, with or without an improvement in EF to >35%during follow-up. DESIGN, SETTING, AND PARTICIPANTS This retrospective analysis examined data collected in the Sudden Cardiac Death in Heart Failure Trial (SCD-HeFT), which randomly assigned 2521 patients to placebo, amiodarone, or ICD between 1997 and 2001. A subset of 1902 participants (75.4%) of the SCD-HeFT had a repeated assessment of EF a mean (SD) of 13.5 (6) months after randomization.We stratified these patients by EF35%and >35%based on the first repeated EF measurement after randomization and compared all-cause mortality in 649 patients randomized to placebo vs 624 patients randomized to ICD. Follow-up started with the repeated EF assessment. Analysis was performed between January 2016 and July 2016. EXPOSURES Implantable cardioverter-defibrillator therapy. MAIN OUTCOMES AND MEASURES All-cause mortality and sudden cardiac death. RESULTS Of the included 1273 patients, the mean (SD) age was 59 (12) years, and 977 (76.7%) were male and 1009 (79.3%) were white. Repeated EF was >35%in 186 participants (29.8%) randomized to ICD and 185 participants (28.5%) randomized to placebo. During a median follow-up of 30 months, the all-cause mortality rate was lower in the ICD vs placebo group, both in patients whose EF remained35%(7.7 vs 10.7 per 100 person-year follow-up) and in those whose EF improved to >35%(2.6 vs 4.5 per 100 person-year follow-up). Compared with placebo, the adjusted hazard ratio for the effect of ICD on mortality was 0.64 (95%CI, 0.48-0.85) in patients with a repeated EF of35%and 0.62 (95%CI, 0.29-1.30) in those with a repeated EF >35%. There was no interaction between treatment assignment and repeated EF for predicting mortality. CONCLUSIONS AND RELEVANCE Among participants in the SCD-HeFT who had a repeated EF assessment during the course of follow-up, those who had an improvement in EF to >35% accrued a similar relative reduction in mortality with ICD therapy as those whose EF remained35%. Prospective randomized clinical trials are needed to test ICD efficacy in patients with an EF >35%.