Design: Cochrane systematic review; one randomised trial included
Population: 781 pregnant women screened for gestational diabetes
Primary outcome: Compliance with screening and follow-up, diagnosis, maternal and neonatal outcomes
Effect: Compliance RR 1.69 (1.53 to 1.87), low certainty; clinical outcomes very uncertain

A Cochrane review looked for trials comparing gestational diabetes screening in maternity settings with screening in general, non-maternity settings. It found one randomised trial of 781 women.

Screening in a maternity setting may raise compliance with screening and follow-up guidance (RR 1.69, 95% CI 1.53 to 1.87) and shorten time to antenatal diabetes care by about 4.7 days, both low-certainty evidence. For diagnosis rates, pre-eclampsia, large-for-gestational-age babies, neonatal unit admission and other clinical outcomes, the evidence was very uncertain, and the trial analysed most outcomes per protocol, which undermines the randomisation.

For services deciding where to put the glucose tolerance test, this supports keeping screening close to antenatal care where possible, mainly because more women complete it. It does not show that the setting changes what happens to mothers or babies.

Keep the glucose tolerance test within or closely linked to the antenatal pathway where you can
Track how many booked women actually complete screening, not just how many are referred
Where screening happens in general labs, build a direct route to the antenatal diabetes team
Do not expect setting alone to change neonatal outcomes — the evidence does not show that

Why it matters

The case for maternity-based screening rests on completion rates, not on proven better outcomes.

The statistics, in plain English

An RR of 1.69 means about 69% more women met screening and follow-up guidance in maternity settings. 'Low certainty' means the true effect could be quite different. For the clinical outcomes, the confidence intervals were very wide — stillbirth ran from 0.08 to 9.93 — meaning anything from large benefit to large harm fits the data.