Background: Continuous glucose monitoring improves neonatal outcomes in pregnancies complicated by type 1 diabetes, and its use is expanding into gestational diabetes on the assumption that the same benefit follows. Gestational diabetes differs in pathophysiology, in glycaemic variability, in the proportion of women requiring insulin, and in the duration of exposure, so the extension is an assumption rather than a finding. Trials in gestational diabetes have generally been small and have reported glycaemic rather than neonatal outcomes.

Methods: We describe a protocol for a multicentre, open-label, parallel-group randomised controlled trial. Women with gestational diabetes diagnosed between 24 and 30 weeks will be randomised 1:1 to continuous glucose monitoring or to capillary self-monitoring according to usual care, with identical glycaemic targets and treatment algorithms in both arms. The primary outcome is large for gestational age at birth, defined as birthweight above the 90th customised centile. Assuming a control-arm rate of 20 % and a relative reduction to 14 %, 822 participants per arm give 90 % power at a two-sided 5 % level; allowing 10 % attrition, 1,827 women will be randomised. Secondary outcomes include neonatal hypoglycaemia requiring treatment, neonatal unit admission, caesarean birth, shoulder dystocia, insulin initiation, hypertensive disorders and maternal-reported burden of monitoring.

Discussion: The design is deliberately conservative in two respects. Both arms use identical glycaemic targets, so the trial tests the monitoring technology rather than a combined package of monitoring and intensified targets, which is the design flaw that makes several existing studies difficult to interpret. And the primary outcome is a neonatal one rather than a glycaemic surrogate, because the justification for the technology rests on neonatal benefit and because improvements in glycaemic metrics have repeatedly failed to translate. The protocol states in advance that the trial is underpowered for a control-arm rate below about 17 %, and the interpretation of a null result is fixed before any data are seen.