Analysis of recurrent heart failure hospitalisations: why mortality assessment should be the gatekeeper

Abstract Background/Introduction In clinical trials for chronic heart failure (CHF), a composite of first heart failure hospitalisation (HFH) or death is widely accepted as a primary efficacy measure. Recent trials have utilised a composite of total (first and subsequent) HFH and death as the primary endpoint to better characterise prognosis and improve trial efficiency 1. However, analysing and interpreting recurrent HFH endpoints is challenging due to death being a terminal event, which limits the total number of unfavourable events per subject. Consequently, an experimental treatment may appear superior in recurrent HFH endpoints despite having no benefit on HFH and a detrimental effect on mortality, leading to inflated type I error rates 2. Purpose We investigated whether a confirmatory non-inferiority test for mortality could serve as a gatekeeper to exclude treatments with undesirable effects on mortality before assessing superiority in recurrent HFH endpoints. Methods A simulation study was conducted to assess the impact of three different primary efficacy assessment strategies on the probability of formal trial success: a) Superiority in the composite of first HFH or death ("first event"); b) Superiority in the composite of total HFH and death ("recurrent events without gatekeeper"); c) Non-inferiority in mortality with a pre-defined margin followed by superiority in the composite of total HFH and death ("recurrent events with gatekeeper"). Simulation settings were aligned with the assumptions for the sample size calculation of the PARAGON-HF trial, which defined the primary endpoint as a composite of total HFH and cardiovascular death 3. Various scenarios were examined where the experimental treatment had a beneficial effect on HFH but either a beneficial, neutral, or detrimental effect on mortality (Table 1). Results When the experimental treatment increased mortality, "recurrent events without gatekeeper" still had a high probability of formal trial success (exceeding 80% in some scenarios), whereas "first event" had a substantially lower probability (below 45%). Adding a gatekeeper reduced the probability of trial success, with the degree of reduction depending on the non-inferiority margin. However, when the experimental treatment improved mortality, "recurrent events with gatekeeper" maintained a higher probability of trial success than "first event". In cases where mortality increased, trial success was less likely with "recurrent events with gatekeeper" compared to "recurrent events without gatekeeper" (Table 2). Conclusion In CHF trials using recurrent HFH endpoints as the primary endpoint, incorporating a confirmatory non-inferiority test for mortality as a gatekeeper in the primary efficacy assessment strategy is essential. This ensures robust conclusions regarding formal trial success and aligns statistical analysis with clinical and regulatory decision-making.Table 1.Simulation settings Table 2.Probability of trial success