Calibrated non-inferiority margin: a new pragmatic method to account for population shift in stroke trials

Non-inferiority trials in acute ischemic stroke (AIS) are crucial to improve access to high-quality care. Population shifts must be accounted for when estimating non-inferiority margins, eg, changes in population characteristics (trial vs historical data); however, existing methods have practical and statistical limitations. We propose a pragmatic conceptual approach and fully pre-specifiable procedure for calibrating non-inferiority margins that account for population shifts in observed trial populations.

Our approach splits trial and historical data into subgroups based on relevant effect-modifying covariates. Trial data from TASTE, which investigated the effect (mRS score 0-1 at day 90) of tenecteplase vs alteplase, were compared to historical data from the Stroke Thrombolysis Trialists' Collaboration (STTC) meta-analysis (alteplase vs control). We reweighted the STTC treatment effect to match the shifted AIS population in TASTE before deriving the calibrated non-inferiority margin.

For both datasets, subgroups were based on onset-to-treatment time and baseline NIHSS values. The reweighted risk difference for alteplase vs control was 11.70% (95% CI, 6.67-16.73); the conservative treatment-effect estimate was 6.67%, corresponding to a risk difference of 3.33% (50% reduction). Hence, the calibrated margin for comparing alternative interventions to alteplase was set at -3.33%, consistent with the European Stroke Organisation's clinically recommended margin (-3.0%).

Our conceptual approach to estimate calibrated non-inferiority margins is a simple and pragmatic alternative to existing methods to account for population shifts in stroke trials. The supporting procedure has already been applied.