Modelling mortality endpoint timing to optimise trial design for haemostatic interventions in traumatic haemorrhage.

Abstract

Background: Selecting an appropriate primary endpoint is critical for trials evaluating haemostatic interventions after trauma. A recent consensus conference recommended a 6-h mortality endpoint (Spinella et al, J Trauma Acute Care Surg 2021), as this captures the timeframe when deaths due to haemorrhage are most likely to occur (whereas deaths from other causes tend to occur later). A perceived disadvantage of an early mortality timepoint is the lower event rate, which is assumed to reduce statistical power, necessitating a larger sample size compared to using a later timepoint and accruing a larger number of events. Aim(s): The aim of this study was to evaluate how the timing of a mortality endpoint influences trial efficiency and ability to detect a treatment effect when mortality arises from competing mechanisms, not all of which are impacted by treatments for haemorrhage. Method(s): We simulated a trial of an intervention with varying effects on haemorrhagic and non-haemorrhagic mortality. Based historical data, we assumed mortality rates of 9% within 0 to <= 6 h and 5% within > 6 to 12 h post-injury (the latter conditional on survival to 6 h). * Scenario 1: 6-h mortality was modelled as the primary endpoint, assuming an odds ratio (OR) of 0.5 for the treatment effect, making the simplifying assumption that all deaths in this time window are due to haemorrhage and thus responsive to the intervention. * Scenario 2: 12-h mortality was modelled as the primary endpoint, with an OR of 0.5 for deaths within 0 to <= 6 h and an OR of 0.8 for deaths between > 6 and 12 h, reflecting, in the second interval, an assumed mixture of deaths due to haemorrhage (and therefore impacted by the intervention) and deaths due to other causes not impacted by the intervention, producing an overall lower average treatment effect for deaths after 6 h. * Scenario 3: 12-h mortality was modelled as the primary endpoint, assuming the same OR of 0.5 for 0 to <= 6 h, but an OR of 1.0 (no effect) for deaths occurring > 6 to 12 h, making the simplifying assumption that these later deaths were not due to haemorrhage and thus not responsive to the intervention. Result(s): Over 100,000 simulated trials, with a sample size of 500 per arm, an average of 68 total events, 110 total events, and 114 total events were observed in Scenarios 1, 2, and 3 respectively. * In scenario 1, a sample size of 575 patients per arm produced 80% power to detect the simulated treatment effect. * In scenario 2, despitemore events due to the extended endpoint, power was lower,with 650 patients per arm required for power of 80%. * Scenario 3 required a sample size of 1000 patients per arm to achieve 80% power. For a 12-hr endpoint, a treatment effect of OR=0.73 from6 to 12 hwould be required to attain power approximately equal to that of Scenario 1. Summary/Conclusions: Using 6-h mortality as the primary endpoint in trials of haemostatic interventions in traumatic haemorrhage provides greater power to detect a treatment effect compared to later timepoints, assuming that early events predominantly consist of deaths due to haemorrhage, which are responsive to the intervention. Extending the mortality window beyond 6 h increases the number of deaths unrelated to haemorrhage and not responsive to the intervention, diluting the treatment effect and reducing statistical efficiency.