BACKGROUND: Noninferiority studies have been recently used to investigate important clinical questions including choice of thrombolytic agent (e.g. TRACE-2, AcT) and the utility of bridging thrombolytic therapy prior to thrombectomy (e.g. DEVT, IRIS). The effect size measures and respective noninferiority margins used in these studies are varied, leading to inconsistent standards for declaring noninferiority. This risks erroneously declaring noninferiority (despite unacceptable loss of efficacy) and discarding noninferior treatments with practical advantages.
AIM: To review current practice of using noninferiority design and analysis in reported stroke studies and investigate the potential pitfalls that inconsistency in this practice may cause
METHODS: Noninferiority studies in stroke research were identified from PubMed Central and reference lists in accordance with PRISMA-ScR tool for scoping reviews. We performed in-silico experiments to examine the consequences of various analytical choices.
RESULTS: Out of 120 identified noninferiority studies, 74 studies used composite events including stroke occurrence as an outcome measure and 24 used the modified Rankin Scale. Analytical choices of effect size measures, noninferiority margins, and methods varied widely. Our computational experiments demonstrated that such variability may lead to misleading and erroneous conclusions about noninferiority in stroke trials.
CONCLUSION: The observed variability of effect sizes for noninferiority stroke research threatens the interpretability and internal consistency of study findings. Current statistical practice, if left unchanged, risks harm by promoting inferior treatments as non-inferior and the erroneous dismissal of practical treatment alternatives with acceptable differences in efficacy. Improved consistency in the design and analysis of stroke noninferiority studies is sorely needed.