Beyond the parallel-group randomised trial: a portfolio of designs for cerebrovascular neurosurgery

Why clinicians should know about it

• Picked for Medical Physics (top studies of the week, 13 September 2026): Trial design discussion, no dosimetric or QA data

Abstract

INTRODUCTION: The randomised parallel-group trial is the dominant evaluation instrument in evidence-based medicine, yet several cerebrovascular diseases violate most of the assumptions on which it depends. Unruptured arteriovenous malformations, unruptured intracranial aneurysms, asymptomatic carotid stenosis and symptomatic haemodynamic carotid or middle cerebral artery occlusion are rare; their natural histories extend across decades; their treatments concentrate substantial harm at a single moment; outcomes depend on operator skill and infrastructure; and the choice between operating and waiting often turns on patient values. The resulting evidence base is sparse, inconsistent, and rarely changes practice. METHODS: This evidentiary deficit cannot be resolved by performing more conventional trials, but by widening the portfolio of designs. After characterising the structural features that make the parallel-group trial inadequate here, we describe five proposals: expertise-based cohort-embedded randomisation, plaque-progression-enriched randomisation, a Bayesian sequential three-arm trial, a stepped-wedge cluster trial, and a synthesising, aspirational multi-arm Bayesian platform that shares a continuously updated natural-history control across strata. For each proposal we set out the eligibility logic, the analytic engine, and the operational obstacles, and verify illustrative sample-size assumptions by Monte Carlo simulation where applicable. CONCLUSION: Cerebrovascular disease requires a wider design vocabulary than the parallel-group RCT. Expertise-based randomisation, cohort embedding, adaptive enrichment, Bayesian inference, stepped-wedge allocation and platform trials are established tools that address rarity, operator heterogeneity, long latency and patient preference directly. Deploying them in cerebrovascular neuroepidemiology is achievable, but will require registry infrastructure, sustained multi-cycle funding, and governance arrangements that do not yet exist in most settings.

Abstract as published, via PubMed.