Gestational Diabetes: Screening, Glycaemic Targets, and Metformin
What the randomised evidence says about who to test, how tightly to control glucose, and when an oral agent can stand in for insulin.
One-Minute Takeaway
- Treating gestational diabetes lowers serious perinatal complications: in the ACHOIS randomised trial the rate fell from 4% to 1% (adjusted RR 0.33, 95% CI 0.14–0.75), a number needed to treat of 34 (95% CI 20–103) — the strongest, RCT-grade evidence in this field.
- One-step versus two-step screening: the roughly 24,000-woman ScreenR2GDM trial found one-step about doubled diagnoses (16.5% vs 8.5%, RR 1.94) without lowering large-for-gestational-age or perinatal outcomes.
- Accepted glucose targets are fasting <95, 1-hour <140, and 2-hour <120 mg/dL; the GEMS trial found lower diagnostic thresholds did not reduce large-for-gestational-age births (8.8% vs 8.9%).
- Metformin matches insulin on the primary neonatal composite with less maternal weight gain and less neonatal hypoglycaemia, but about 46% of women need supplemental insulin.
- ADA and ACOG keep insulin first-line; SMFM and NICE accept metformin as a reasonable first choice.
Why Gestational Diabetes Matters
Gestational diabetes is one of the most common medical complications of pregnancy, and its harms track a continuous relationship with maternal glucose rather than a sharp diagnostic threshold. That single fact shapes every downstream decision about screening, targets, and treatment.
The Hyperglycemia and Adverse Pregnancy Outcome (HAPO) study measured this gradient in 23,316 women who completed a blinded 75-g oral glucose-tolerance test at 24 to 32 weeks. For each 1-standard-deviation rise in fasting glucose, birth weight above the 90th percentile was associated with an odds ratio of 1.38 (95% CI 1.32–1.44), and cord-blood C-peptide above the 90th percentile with an odds ratio of 1.55 (95% CI 1.47–1.64). Risk climbed steadily with no obvious cut-point.
Because that association is observational, it cannot on its own justify treatment. What raises the diagnosis above a laboratory curiosity is that intervention changes outcomes. The clinical burden it targets is concrete: large-for-gestational-age infants, shoulder dystocia, operative delivery, hypertensive disorders, and neonatal hypoglycaemia, alongside a substantial lifetime maternal risk of type 2 diabetes.
The Australian Carbohydrate Intolerance Study in Pregnant Women (ACHOIS) provides the anchoring randomised evidence. Among 1,000 women, dietary advice, glucose monitoring, and insulin as needed reduced serious perinatal complications from 4% to 1% (adjusted RR 0.33, 95% CI 0.14–0.75). The number needed to treat to prevent one such event was 34 (95% CI 20–103). Rates of caesarean delivery were similar between groups, though labour induction was more frequent with treatment.
The NICHD Maternal-Fetal Medicine Units (MFMU) trial extended this to milder disease. In 958 women, the composite primary outcome did not differ (32.4% vs 37.0%), and there were no perinatal deaths. Yet treatment still reduced several prespecified secondary outcomes: large-for-gestational-age infants (7.1% vs 14.5%), birth weight above 4,000 g (5.9% vs 14.3%), shoulder dystocia (1.5% vs 4.0%), caesarean delivery (26.9% vs 33.8%), and combined preeclampsia and gestational hypertension (8.6% vs 13.6%).
Even hyperglycaemia detected before 20 weeks appears to warrant attention. In the TOBOGM trial, immediate rather than deferred treatment of early gestational diabetes lowered a composite of adverse neonatal outcomes from 30.5% to 24.9% (adjusted risk difference −5.6 percentage points, 95% CI −10.1 to −1.2), a benefit driven largely by less neonatal respiratory distress.
The Clinical Problem in Gestational Diabetes
Because treatment demonstrably helps, the open questions are no longer whether to manage gestational diabetes but how. Three decisions account for most of the variation in real-world care, and each has been tested in a randomised trial.
First, how should women be screened — the one-step 75-g test that finds more cases, or the two-step approach that finds fewer? Second, how tight should glucose targets be, and does moving the diagnostic threshold downward help? Third, when diet and exercise are not enough, should the first drug be insulin or metformin? The sections below weigh the evidence behind each.
Available Approaches
The comparative table summarises the landmark studies that inform current practice, spanning the glucose–outcome relationship, the value of treatment, screening strategy, diagnostic thresholds, and pharmacotherapy. Every effect size below was confirmed against its primary source, and every sample size is the real reported figure.
| Study / Source | Design | N | Question addressed | Key result (effect size, 95% CI) | Tier |
|---|---|---|---|---|---|
| HAPO (Metzger 2008) | Prospective observational cohort | 23,316 | Is milder hyperglycaemia linked to harm? | Per 1-SD fasting glucose: birth weight >90th pct OR 1.38 (1.32–1.44); cord C-peptide >90th pct OR 1.55 (1.47–1.64); continuous, no threshold | [V] |
| ACHOIS (Crowther 2005) | RCT | 1,000 | Does treating GDM help? | Serious perinatal complications 1% vs 4%; adjusted RR 0.33 (0.14–0.75); NNT 34 (20–103) | [V] |
| MFMU mild GDM (Landon 2009) | RCT | 958 | Does treating mild GDM help? | Composite 32.4% vs 37.0% (NS); LGA 7.1% vs 14.5%; shoulder dystocia 1.5% vs 4.0%; caesarean 26.9% vs 33.8% | [V] |
| ScreenR2GDM (Hillier 2021) | Pragmatic RCT | 23,792 | One-step vs two-step screening | GDM 16.5% vs 8.5%, RR 1.94 (97.5% CI 1.79–2.11); no difference in LGA, perinatal composite, caesarean, or hypertensive disorders | [V] |
| GEMS (Crowther 2022) | RCT | 4,061 | Lower vs higher diagnostic thresholds | GDM 15.3% vs 6.1%; LGA 8.8% vs 8.9%, adjusted RR 0.98 (0.80–1.19) | [V] |
| TOBOGM (Simmons 2023) | RCT | 748 | Treating GDM before 20 weeks | Neonatal composite 24.9% vs 30.5%; adjusted risk difference −5.6 pts (−10.1 to −1.2); RR 0.82 (0.68–0.98) | [V] |
| MiG (Rowan 2008) | RCT | 751 | Metformin vs insulin | Primary neonatal composite equivalent (NS); 46% of metformin group needed supplemental insulin; less maternal weight gain | [V] |
| Sheng 2023 | Systematic review / meta-analysis (24 RCTs) | 4,355 | Metformin vs insulin, neonatal | Birth weight −122.8 g (−178.3 to −67.2); macrosomia RR 0.68 (0.54–0.86); neonatal hypoglycaemia RR 0.65 (0.52–0.81); NICU RR 0.73 (0.61–0.88) | [V] |
| Tarry-Adkins 2019 | Systematic review / meta-analysis (28 studies) | 3,976 | Metformin vs insulin, offspring growth | Neonatal birth weight −107.7 g (−182.3 to −32.7); macrosomia OR 0.59 (0.46–0.77); higher BMI by mid-childhood | [V] |
Screening Strategy: One-Step Versus Two-Step
Two internationally endorsed approaches compete. The one-step method, favoured by IADPSG, uses a single fasting 75-g test and diagnoses gestational diabetes on any one abnormal value (fasting ≥92, 1-hour ≥180, or 2-hour ≥153 mg/dL). The two-step method begins with a non-fasting 50-g glucose challenge; women who exceed the screening cut-off proceed to a fasting 100-g test read against Carpenter–Coustan criteria, with two abnormal values required.
The pragmatic ScreenR2GDM trial randomised nearly 24,000 women across two health systems to one of these pathways. The one-step approach roughly doubled the diagnosis of gestational diabetes, at 16.5% versus 8.5% (unadjusted RR 1.94, 97.5% CI 1.79–2.11). Despite finding many more cases, it did not lower the rate of large-for-gestational-age infants, the perinatal composite, primary caesarean delivery, or hypertensive disorders.
The practical reading is that the extra cases identified by one-step screening are, on average, milder and gain little measurable benefit from being labelled and treated. The US Preventive Services Task Force recommends universal screening at or after 24 weeks; ACOG endorses the two-step approach, while the ADA accepts either and notes no clear superiority. Screening high-risk women early for undiagnosed overt diabetes is recommended, but routine screening for gestational diabetes before 24 weeks is not.
Glycaemic Targets: How Tight to Aim
Once gestational diabetes is confirmed, ADA and ACOG converge on the same monitoring targets: fasting glucose below 95 mg/dL, one-hour postprandial below 140 mg/dL, and two-hour postprandial below 120 mg/dL. Postprandial monitoring is associated with better glycaemic control and a lower risk of preeclampsia than preprandial monitoring alone.
An important caveat is that these specific numbers rest on observational and consensus data. No adequately powered randomised trial has compared one set of target values against another, so the thresholds represent reasonable clinical convention rather than a treatment effect measured head-to-head.
A related question is where to set the diagnostic threshold itself. The GEMS trial randomised 4,061 women to lower (IADPSG) or higher glycaemic criteria for diagnosis. The lower criteria more than doubled diagnoses (15.3% vs 6.1%) but did not reduce large-for-gestational-age births (8.8% vs 8.9%, adjusted RR 0.98, 95% CI 0.80–1.19). In a prespecified subgroup whose results fell between the two thresholds, treating the milder cases did lower large-for-gestational-age rates — so the threshold debate is really a question of how many mild cases are worth treating, not whether treatment can help them.
Metformin Versus Insulin
Insulin controls maternal glucose effectively and does not cross the placenta, which is why it remains the reference agent. Metformin is attractive as an oral, lower-cost option, but it does cross the placenta, reaching fetal concentrations comparable to maternal levels.
The Metformin in Gestational diabetes (MiG) trial randomised 751 women to metformin, with supplemental insulin if required, or to insulin alone. The primary neonatal composite did not differ between groups. Metformin produced less maternal weight gain and higher treatment satisfaction, with a small increase in preterm birth. Notably, 46% of the metformin group still required supplemental insulin to reach target — metformin is frequently a partial, not a complete, monotherapy.
Pooled data reinforce a favourable short-term profile. A meta-analysis of 24 randomised trials (4,355 participants) found that, compared with insulin, metformin lowered neonatal birth weight (mean difference −122.8 g, 95% CI −178.3 to −67.2) and reduced macrosomia (RR 0.68, 0.54–0.86), neonatal hypoglycaemia (RR 0.65, 0.52–0.81), and neonatal intensive care admission (RR 0.73, 0.61–0.88).
The unsettled question is longer term. A meta-analysis of 28 studies (3,976 participants) confirmed that metformin-exposed neonates are smaller at birth (−107.7 g) with less macrosomia (OR 0.59), but reported accelerated postnatal growth, so that exposed children were heavier and had a higher body-mass index by mid-childhood. Offspring follow-up from the MiG cohort similarly found larger measures of subcutaneous fat at two years of age. Whether this low-birth-weight-then-catch-up pattern carries later cardiometabolic consequences is not resolved.
Comparative Analysis
Placed side by side, the three decisions differ in how much the evidence actually constrains the choice. Screening strategy barely moves outcomes: one-step and two-step approaches yield similar perinatal results, so the trade-off is between finding more cases at greater cost and workload versus finding fewer with equivalent measured benefit.
Glycaemic targets sit on weaker footing. The accepted fasting, one-hour, and two-hour values are consensus-based, and lowering the diagnostic threshold adds diagnoses without a population-level reduction in large-for-gestational-age births — although the milder cases so identified can still benefit individually when treated.
Pharmacotherapy is the most nuanced. Metformin matches insulin on short-term neonatal outcomes with practical advantages in weight and hypoglycaemia, but it is an incomplete monotherapy in a large minority and carries an unresolved offspring-growth signal. That combination explains a genuine guideline divergence: ADA and ACOG keep insulin first-line and advise against metformin or glyburide as first-line agents, citing placental transfer, while SMFM and NICE regard metformin as a reasonable first choice for women who require pharmacotherapy.
These positions are not contradictory so much as differently weighted. They reflect the same evidence read through different priorities — long-term caution on one side, access and acceptability on the other.
Quality and Consistency of the Evidence
- High — that treating gestational diabetes reduces serious perinatal complications and fetal overgrowth. Two randomised trials (ACHOIS, MFMU) agree in direction, with an anchoring NNT of 34.
- High — that the choice between one-step and two-step screening does not change perinatal outcomes. One large pragmatic RCT plus supporting meta-analysis, consistent.
- Moderate — that metformin equals insulin on the short-term neonatal composite. Many trials and meta-analyses concur, but with heterogeneity and a 20–46% rate of insulin supplementation.
- Moderate — that lower diagnostic thresholds do not reduce large-for-gestational-age births at a population level. One large RCT (GEMS), with a discordant-subgroup benefit.
- Low / emerging — the long-term offspring effects of in-utero metformin exposure. Follow-up cohorts suggest higher mid-childhood BMI, but these are not powered randomised comparisons.
- Low — the optimal glycaemic target values themselves. No adequately powered RCT has compared different target sets head-to-head.
What the Evidence Does Not Show
No adequately powered randomised trial has compared different glycaemic target values, so the fasting <95, one-hour <140, and two-hour <120 mg/dL targets rest on observational and consensus data rather than a measured treatment effect.
The long-term cardiometabolic outcomes of children exposed to metformin in utero are not settled. The pattern of lower birth weight followed by postnatal catch-up growth is a signal that warrants follow-up, not an established harm.
Screening trials were not powered to exclude small differences in rare events such as stillbirth or birth injury; “no difference” means no large difference, not proof of equivalence for uncommon outcomes.
The benefit of screening before 20 weeks is modest, and the optimal diagnostic thresholds in early pregnancy remain undefined. Continuous glucose monitoring, though valuable in type 1 diabetes in pregnancy, lacks outcome evidence specific to gestational diabetes.
Practical Implications
Either screening pathway is defensible. Because one-step screening will diagnose roughly twice as many women without a demonstrated perinatal advantage, system capacity, cost, and local guideline alignment reasonably drive the choice rather than a difference in outcomes.
After diagnosis, medical nutrition therapy and exercise achieve target glucose in roughly half of women, and pharmacotherapy is added when targets are exceeded despite adherence. Monitoring targets are fasting below 95, one-hour postprandial below 140, and two-hour postprandial below 120 mg/dL, with postprandial testing preferred.
Insulin remains the reference first-line agent under ADA and ACOG. Metformin is a reasonable first choice under SMFM and NICE, particularly for women who decline injections, provided they are counselled that about one in two will still need insulin and that long-term offspring data are incomplete. Glyburide is a less favourable oral option given its neonatal profile.
Evidence Grade and Bottom Line
Overall evidence grade: Moderate to high for the core management questions, anchored by several large randomised trials and consistent meta-analyses.
Strong evidence supports treating gestational diabetes to reduce serious perinatal complications and fetal overgrowth. The evidence suggests that screening strategy and diagnostic threshold change how many women are labelled far more than they change outcomes. Metformin is a reasonable, effective alternative to insulin for many women, with the caveats of frequent insulin supplementation and an unresolved long-term offspring signal, which is why major guidelines still differ on first-line choice.