Revisiting the DIGIT-HF Trial: Evidence of Time-Dependent Treatment Effects

BackgroundThe DIGIT-HF trial reported that digitoxin reduced the composite risk of all-cause mortality or first heart failure hospitalization in patients with heart failure and reduced ejection fraction, summarized by a single hazard ratio (HR) of 0.82. The validity of this single estimand depends on the proportional hazards (PH) assumption of the Cox model. MethodsWe performed a methodological re-analysis based on individual patient data reconstructed from the published Kaplan-Meier curves of the DIGIT-HF trials primary endpoint. We formally tested the PH assumption, modeled the time-dependent treatment effect, and performed a pre-specified landmark analysis at 18 months. ResultsOur analysis confirmed a statistically significant violation of the proportional hazards assumption for the overall follow-up period (p = 0.019). The time-dependent HR plot showed that the beneficial effect of digitoxin waned over time. The landmark analysis revealed a significant benefit in the first 18 months (HR, 0.69; 95% CI, 0.54-0.88) but no evidence of benefit thereafter (HR, 0.99; 95% CI, 0.77-1.28). The PH assumption was met within both the early (p = 0.17) and late (p = 0.54) periods. ConclusionOut results strongly suggest the treatment effect of digitoxin in DIGIT-HF is not constant over time, invalidating use of a single hazard ratio. Benefit is confined to the first 18 months of therapy, challenging the interpretation of digitoxin as providing sustained benefit. These findings have important implications for clinical practice and highlight the necessity of testing proportional hazards assumptions in cardiovascular trials.