Study selection and study characteristics
The PRISMA flow chart (Supplementary Fig. S1) demonstrates the screening procedure involved to attain 35 studies (8033 participants) for this meta-analysis. The majority of these studies (27 studies; 5319 participants) investigated metformin treatment for diabetes in pregnancy. A total of 8 studies (2714 participants) investigated maternal obesity (4 studies; n=1485) or PCOS (3 studies; n=930) or pre-gestational insulin resistance (1 study; n=299). No studies investigated randomisation of metformin compared to diet/lifestyle intervention alone, however these were commonly implemented alongside other treatments. Eligible studies were identified comparing metformin to insulin, glyburide, and placebo. For all indications and comparisons, the studies varied with respect to quality and design (Supplementary Table S3). The included studies demonstrated considerable heterogeneity with respect to the dosage of pharmacological agents (Supplementary Table S3). Heterogeneity also existed in the diagnostic criteria used for GDM and PCOS (Supplementary Table S4). The included studies came from a variety of geographical locations: Australasia (Australia and New Zealand), Europe (the UK, Norway and Finland), North Africa/Middle East (Egypt, Iran and Israel) and the North America/Latin America (Canada, Mexico, Brazil and Chile).
Risk of bias and sensitivity analyses
The risk of bias was moderate-to-low in the majority of included studies, however six studies were considered to have a high risk of bias (Supplementary Table 2). We performed sub-group meta-analyses, excluding the studies assessed as having a high risk of bias (Supplementary Fig. S2), which showed that removal of studies with a high risk of bias did not materially alter the outcome of the meta-analyses for any of the outcomes, therefore all studies were included. Most studies reported non-significant differences in maternal baseline characteristics between groups (including maternal age, BMI and glycemic control) (Supplementary Table S3).
We assessed the likelihood of single studies significantly influencing the overall results using leave-one-out (LOO) analysis. For the primary comparison of metformin vs. any other treatment across all indications, meta-analyses were robust to the omission of single studies (Supplementary Fig. S3), with the exception of cesarean section, RBS, and maternal hypoglycaemia, decreasing our confidence in the robustness of these findings. Funnel plots for all outcomes were assessed visually (Supplementary Fig. S4); there were no obvious asymmetries in the plots for any study outcomes. Eggers testing demonstrated a low likelihood of publication bias with respect to the primary comparisons (Supplementary Table S5).
GRADE analysis (certainty of evidence)
The majority of outcomes were classified as having a moderate certainty of evidence (Supplementary Fig. S5; primary outcomes and Supplementary Fig. S6; secondary outcomes), with one outcome having a high certainty of evidence (pregnancy-induced hypertension).
Pre-eclampsia
Pregnant women randomized to metformin had a significant reduction in the likelihood of pre-eclampsia compared to those randomised to any other treatment (OR 0.65, 95%CI: 0.47 to 0.91; I2=52%, p=0.01) (Figure 2); based on 21 studies including 5979 pregnancies. There were no significant differences in the risk of pre-eclampsia in the sub-group analyses where the indication for randomization was maternal obesity (4 studies [21,22,35,36] n=1620) or PCOS (3 studies [23,37,38] n=818) (Table 1). However, in the sub-group of women with diabetes in pregnancy, randomization to metformin resulted in significantly decreased likelihood (p=0.02) of pre-eclampsia (14 studies [24-26,28,29,31,39,40-46] n=3301) (Table 2). Effect sizes were similar across all diabetes in pregnancy groups, although these failed to reach statistical significance in sub-groups (Table 2).
Pregnancy-induced hypertension
There was no difference in the likelihood of pregnancy-induced hypertension (PIH) between women randomized to metformin versus any other treatment (OR 0.97, 95%CI: 0.77 to 1.22; I2=0%, p=0.81) (Figure 3) based on 14 studies including 4189 pregnancies. There were no significant differences in the risk of PIH in any of the sub-group analyses where the indication for randomization was maternal obesity (3 studies [21,22,35] n=1354), PCOS (1 study [23] n=478) (Table 1) or diabetes in pregnancy (10 studies [24,26,29,30,34,39,41,45,46,47] n=2806) (Table 2).
Preterm birth (all-cause, spontaneous and iatrogenic)
There was no difference in the overall likelihood of preterm birth between women randomized to metformin versus other interventions (OR 0.91, 95%CI: 0.67 to 1.22; I2=58%, p=0.52) (Figure 4 a) based on 27 studies including 6959 pregnancies. When sub-group analysis was performed separating spontaneous and iatrogenic preterm birth, there was no significant effect of randomization to metformin versus other treatments on likelihood of either type of preterm birth (Figures 4 b & 4 c).
In women with maternal obesity (4 studies [21,22,35,36] n=1620) (Table 1) or diabetes in pregnancy (19 studies [26,29-31,33,34,39-45,47-52] n=4443) (Table 2), the likelihood of preterm birth was not different between women randomized to metformin versus other treatments. However, in the sub-group where the indication for randomization was PCOS (3 studies [23,37,38] n=827), randomization to metformin was associated with reduced likelihood of preterm birth (p=0.01) (Table 1).
Gestational age at delivery
Randomization to metformin vs. any other treatment did not significantly influence gestational age at delivery (-0.08 weeks, 95%CI: -0.21 weeks to 0.04 weeks; I2=48%, p=0.19) (Figure 5), based on 18 studies including 3818 pregnancies. There was no significant difference in gestational age at delivery where the indication for randomization was maternal obesity (2 studies [21,22] n=948) (Table 1) or diabetes in pregnancy (16 studies [24,25,26,27,28,29,31,32,33,41,43,44,45,46,53,54] n=2870) (Table 2). No studies reported gestational age at delivery in the sub-group of women with PCOS (Table 1).
Cesarean section (all cause, emergency and elective)
There was a trend to suggest lower likelihood of delivery by cesarean section in women randomized to metformin versus other treatments (OR 0.86, 95%CI: 0.73 to 1.01; I2=46%, p=0.06) (Figure 6 a) based on 29 studies including 6122 pregnancies. When sub-group analysis was performed separating emergency and elective cesarean section, there was no significant effect of randomization to metformin versus other treatments on likelihood of either type of cesarean section (Figures 6 b & 6 c).
In the sub-group where the indication for randomization was maternal obesity (3 studies [21,22,35] n=1352), randomization to metformin (vs. placebo) was associated with reduced likelihood (p=0.03) of cesarean section (Table 1). Where the indication for randomisation was PCOS (2 studies [23,38] n=1352) (Table 1) or diabetes in pregnancy (22 studies [24-33,39,40,42,44-50,53,54] n=3516) (Table 2) randomization to metformin versus other treatments did not alter likelihood of cesarean section.
Side effects
Randomisation to metformin versus placebo was associated with increased likelihood of nausea, vomiting and diarrhoea, but not abdominal pain or non-gastrointestinal side-effects (Supplementary Table 6). Nausea, vomiting, and diarrhoea were all significantly increased when the indication for metformin randomization was maternal obesity (Table 1). However, when the indication for randomization was PCOS, fewer studies were available for analysis and only the likelihood of diarrhoea was significantly increased with metformin versus placebo (Table 1). Trials involving women with diabetes in pregnancy either did not report gastrointestinal side effects in the insulin arm, or reported zero values. Between 2-46% of women randomised to metformin for treatment of diabetes in pregnancy reported gastrointestinal side-effects (weighted average incidence 12.5%; Appendix 15) and 0-6% of women stopped medication due to these side effects (weighted average incidence 14.3%), (Supplementary Table S7).
GDM in participants randomised for indications other than diabetes
Randomization to metformin vs. any other treatment did not alter the likelihood of subsequent GDM diagnosis (OR 1.07, 95%CI: 0.87 to 1.33; I2=0%, (p=0.52) (Supplementary Fig. S7), based on 7 studies including 2063 pregnancies. Whether the indication for randomization was maternal obesity (3 studies [21,22,35] n=1206) or PCOS (3 studies [23,37,38] n=746) (Table 1), randomisation to metformin did not alter likelihood of GDM.
Glycemic control in diabetes in pregnancy
There was no significant difference in FBS (19 studies, n=3673) or RBS (18 studies, n=3610) measurements in women with diabetes in pregnancy randomised to metformin versus other treatments (Table 2). Maternal hypoglycemia was significantly (p=0.005) less likely in women randomized to metformin vs. other treatments (Supplementary Fig. S8), based on 6 studies including 1149 pregnancies [30,34,45,46,49,53], however this effect is driven entirely by studies where insulin was the comparator group (Table 2).