Introduction

Gestational diabetes mellitus (GDM) is characterized by glucose intolerance first identified during pregnancy and is associated with perinatal and long-term adverse health outcomes in both the pregnant individual and the offspring. The physiologic reduction in insulin sensitivity during pregnancy is the hallmark metabolic feature that leads to the onset of glucose intolerance and GDM in predisposed individuals1. Established risk factors for GDM include previous GDM, advanced maternal age, parity, overweight/obesity, and family history of diabetes2,3. GDM increases perinatal complications including preeclampsia, operative deliveries, stillbirth, neonatal risks of large-for-gestational age, hypoglycemia, and respiratory distress syndrome4. GDM confers an increased lifetime risk of type 2 diabetes mellitus for both mother and offspring5,6. GDM rates vary considerably, with geographic differences and varying diagnostic criteria accounting for the 1–30% incidence7. Nonetheless, rates of GDM are increasing across all populations8,9, commensurate to worldwide increasing rates of overweight and obesity.

Prevention of GDM involves reducing hyperglycemia and insulin resistance, factors that are also highly correlated with obesity10,11. Weight reduction prior to pregnancy and prevention of excessive gestational weight gain (GWG) are important features of diabetes prevention12,13. Insulin resistance is affected by a number of factors: weight, lifestyle, physical activity, dietary intake and supplement use. Several meta-analyses of randomized controlled trials (RCTs) investigating lifestyle interventions have reported on diet and physical activity interventions, metformin, and supplements as either primary GDM prevention strategies or secondary prevention strategies for trials targeting weight management and/or reduction as a primary outcome. Results of these meta-analyses have not been unanimous in the reporting of findings suggesting heterogeneity in the intervention response, perhaps due to the characteristics of the study population, and/or the timing and type of intervention14,15,16,17.

Individual characteristics such as clinical, psychosocial and biochemical factors may influence the effectiveness of interventions in preventing GDM. The prevention of GDM results from an interaction between behavioural factors, such as the ability to adhere to the intervention, and physiological factors, such as the biological responsiveness towards reducing insulin resistance. Hence, clinical traits, such as overweight/obesity, age, history of GDM or polycystic ovary syndrome (PCOS), along with social determinants of health, for example, socioeconomic status, cultural background, race or ethnicity, are potential sources of heterogeneity of the intervention effect18. These clinical, biochemical, social and environmental traits could affect GDM prevention through behavioural or physiological pathways, or both. Given that interventions to prevent GDM are unlikely to be effective for individuals as a ‘one-size-fits-all’ approach, there is a need to elucidate the most effective mode of prevention for each population. To date, there has not been a comprehensive meta-analysis of GDM prevention, accounting for participant characteristics to inform precision medicine.

The field of precision medicine recognizes that examining the heterogeneity of individual responses to intervention is important for optimizing health-enhancing interventions and minimizing exposure to specific risk factors, to delay or prevent the onset of a given disease18,19. The Precision Medicine in Diabetes Initiative (PMDI) was established in 2018 by the American Diabetes Association (ADA) in partnership with the European Association for the Study of Diabetes (EASD). The ADA/EASD PMDI includes global thought leaders in precision diabetes medicine who are working to address the burgeoning need for better diabetes prevention and care through precision medicine18. This review is written on behalf of the ADA/EASD PMDI as part of a comprehensive evidence evaluation in support of the 2nd International Consensus Report on Precision Diabetes Medicine20. To inform a precision medicine approach to diabetes prevention, the primary focus of this review was to assess the contribution of various participant characteristics to the effectiveness of interventions for GDM prevention. To this end, this systematic review and meta-analysis examined the effectiveness of interventions employing lifestyle modification, metformin, or dietary supplements within the preconception, pregnant and postpartum/interconception periods for reducing the risk of developing GDM. We find that certain participant characteristics such as BMI, having polycystic ovary syndrome or history of GDM or being in the preconception phase may determine responses to particular interventions.

Methods

This systematic review and meta-analysis was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) Statement21. The protocol was registered in the PROSPERO International Prospective Register of Systematic Reviews (CRD42022320513).

Search Strategy

A comprehensive search strategy was developed by a professional medical librarian (AF) in consultation with the authors (SL, LR, KV, JJ). The search strategy included keywords and Medical Subject Headings (MeSH), as shown in Supplementary Table 1. We searched the following databases: Embase (Elsevier), Ovid Medline, and PubMed from the inception of the database to May 24, 2022. Results were limited to studies in human and in English-language. No limit was placed on publication date. Endnote (Clarivate) was used to compile records and remove duplicates. Covidence (Veritas Health Innovation, Melbourne, Australia) was then used for title/abstract screening and full text review. Hand-searches including the reference list of related reviews were also examined for additional eligible trials.

Selection criteria

Randomised and non-randomized controlled trials (RCTs and non-RCTs) investigating the effects of lifestyle (diet, exercise, or both), metformin, or dietary supplements (fish oil, myoinositol/inositol, probiotics) on prevention of GDM in women of childbearing age (including preconception cohorts) were included (Supplementary Table 2). Control conditions included usual care or minimal intervention (no more than a single intervention session in the case of diet and exercise interventions). Studies without a control group (usual care or placebo), those that did not report GDM, observational studies, editorials, commentaries, conference abstracts, reviews, meta-analyses and study protocols were excluded. Titles and abstracts were evaluated independently and in duplicate to identify articles for full-text review. Full-text review was conducted independently and in duplicate with reasons for exclusion recorded. Discrepancies were resolved by consensus by two or more authors.

Data extraction

Data were extracted using an extraction template developed for this study with GDM as the primary outcome. Study characteristics (authors, year of publication, country, setting, sample size, design, diagnostic criteria, diagnosis time point, intervention commencement, and outcome of interest), participant characteristics (age, race/ethnicity, BMI, education status, employment status, parity, prior GDM, smoking status and other medical history), intervention type (diet, physical activity, diet and physical activity, metformin, types of dietary supplement), and outcome of intervention (GDM incidence) were extracted. Study characteristics were determined based on known GDM risk factors and other relevant factors identified by the precision medicine report19. Authors were contacted for missing information. One author conducted the data extraction, and a second author conducted a 10% sub-sample data extraction to establish reliability. An agreement of 89% was achieved between the two authors with discussion to resolve discrepancies.

Quality assessment

Quality of the included studies was critically appraised using a relevant tool for each study design. The Revised Cochrane Risk of Bias Tool for Randomized Trials (RoB 2.0)22 was used for RCTs to assess bias arising from the randomization process, deviations from the protocol, missing data, measurement of the outcome and selective reporting. The ROBINS-I tool was used for non-RCTs to assess bias from confounding, participant selection, classification of interventions, missing data, deviations from intended interventions, measurement of outcomes, and selection of reported results23. Two reviewers independently conducted the methodological quality and bias assessment for individual studies. Differences were resolved by consensus.

The GRADE process (Grading of Recommendations Assessment, Development and Evaluation), which rates the quality of evidence from a study in a systematic approach was conducted for the primary outcome of GDM24. Risk of bias, along with consistency, directness, precision and publication bias were considered for GRADE appraisal to determine the quality of evidence.

Statistical analysis

The outcome was the incidence of GDM. Data were pooled and GDM incidence was expressed as risk ratios (RR) with 95% confidence intervals (CI). Heterogeneity between studies was assessed by the I2 test where I2 > 50% indicated substantial heterogeneity. Potential sources of heterogeneity by participant characteristics (e.g. obesity, age, PCOS) were explored through subgroup analyses and meta-regression, as conducted in other systematic reviews and meta-analyses25,26. Significant (p < 0.05) Egger’s test and funnel plot (asymmetry) was used to declare publication bias. Estimates (RR) were pooled using random-effects model with the DerSimonian and Laird estimator27. Subgroup analyses were conducted if there was at least one trial present in at least two comparative subgroups. Sensitivity analyses were conducted by excluding non-randomized controlled trials. P < 0.05 was taken as the level of statistical significance. All analyses were conducted in Stata Version 17 (STATA Corporation, College Station, Texas, USA).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Results

We screened 10,347 records for eligibility and 434 records were reviewed as full texts (Fig. 1). Overall, 130 articles were deemed eligible representing 116 unique studies (117 comparisons due to multiple intervention arms in Luoto et al) and were included in the meta-analysis. Reasons for exclusion included lack of an appropriate control group where the only difference between the treatment and control group was with or without the interventions of interest, no GDM outcome, or active intervention in the control group (Fig. 1).

Fig. 1: PRISMA flow diagram.
figure 1

Flowchart demonstrating the process of the identifying the papers included in this review. This maps out the number of records identified, included and excluded at each stage of assessment. Reasons for full text exclusions are provided.

Full size image

Study characteristics

A summary of the characteristics of included studies are shown in Supplementary Data 1. Studies were published from 1997 to 2022. Sample sizes ranged from 31 to 4631. Of the included studies, 92 (79%) involved lifestyle (diet, physical activity, or both), 13 (11%) involved metformin, and 12 (10%) involved dietary supplement interventions. One study included a comparison of lifestyle, probiotics with diet and control28 and was included as both a lifestyle and dietary supplement study in the meta-analysis. Types of lifestyle intervention included diet only (n = 17), physical activity only (n = 19) or a combination of diet and physical activity (n = 59). The types of dietary supplement interventions included myoinositol/inositol (n = 7), probiotics-only (n = 4), probiotics coupled with diet (n = 1), probiotics with fish oil (n = 1) and fish-oil only (n = 1). Interventions commenced from preconception to 26 weeks gestation.

The definition of the participant characteristics is shown in Supplementary Table 3. Detailed description of the participant characteristics of the included studies is shown in Supplementary Table 4. One hundred and five studies commenced the intervention during pregnancy, seven studies reported commencing the intervention prior to pregnancy while four studies did not provide information on pregnancy status at recruitment. Two studies were conducted in women who were nulliparous. A number of studies included only participants with certain medical conditions or medical history: overweight (BMI 25–29.9 kg/m2) or obesity (BMI > 29.9 kg/m2) (n = 57), PCOS (n = 9), prediabetes (n = 1) or family history of diabetes (n = 2); whereas some studies excluded participants with certain medical conditions or medical history: hypertension (n = 27), prediabetes (n = 52), PCOS (n = 3), history of stillbirth (n = 2), family history of diabetes (n = 2), previous macrosomia infant (n = 3), past history of GDM (n = 20), hypertensive disorders of pregnancy (n = 4), history of cardiovascular disease (n = 14), smoking (n = 11). The mean age of the participants ranged from 25 to 34 years, and mean BMI at baseline ranged from 21 to 39 kg/m2. Of the included studies, 25 (22%) had mostly participants with tertiary education (as defined in Supplementary Table 3) and 25 (22%) had mostly participants in employment (Supplementary Table 3). Of the 54 studies which reported race, 18 studies were predominantly conducted among White participants, 10 were predominantly conducted among non-White participants, and 26 among mixed populations.

Each included study reported GDM as a primary or secondary outcome (Supplementary Data 1). The criteria used for GDM diagnosis varied across the studies and included one-step (most commonly a single 75-gram, 2-h oral glucose tolerance test) and two-step methods (commonly the 50 gram one-hour oral glucose challenge test, followed by a 2 or 3-h oral glucose tolerance test if the oral glucose challenge test was abnormal). The most frequently reported diagnostic criteria (n = 37) were those of the International Association of the Diabetes and Pregnancy Study Groups (IADPSG), the World Health Organization (WHO) in 1999 (prior to WHO adopting those of the IADPSG) (n = 9), Carpenter & Coustan (C&C, n = 7), and National Diabetes Data Group (NDDG, n = 6). The method of GDM diagnosis was not reported by 34 studies and 13 studies used a method that could not be categorized by one set of diagnostic criteria. Of the 37 studies using IADPSG criteria, 22 tested diet and physical activity interventions, 9 supplements, 3 metformin, 2 diet, and 1 physical activity. Within those studies that used IADPSG, 26 identified GDM as a primary outcome. Of the 7 studies initiated in the preconception period (2 diet, 1 diet+physical activity, 4 metformin), GDM was the primary outcome for 4, 2 of which were non-randomized clinical studies of metformin use initiated prior to pregnancy targeting women with PCOS and the only 2 that used the same criteria for diagnosing GDM (NDDG).

Meta-analysis

Meta-analysis of all lifestyle interventions (diet only, physical activity only, and combined diet+physical activity) showed a significant reduction in the risk of GDM with moderate heterogeneity (RR 0.78, 95%CI 0.72, 0.85, I2 = 45). (Table 1).

Table 1 Meta-analysis of the effect of diet, physical activity, metformin or dietary supplement on gestational diabetes prevention.
Full size table

Diet-only

In the random effects model, diet-only interventions showed a significant reduction in the risk of GDM (RR 0.78, 95%CI 0.65, 0.94, 17 studies, I2 = 45.6%, moderate-quality evidence) with moderate heterogeneity.

Risk differences in the subgroup and meta-regression analyses by participant characteristics such as sociodemographic (e.g. educational status) and medical history (e.g. prediabetes and hypertension) was not observed (Supplementary Data 2 and 3).

Physical activity-only

Meta-analysis of physical activity-only interventions showed a significant reduction in the risk of GDM (RR 0.70, 95%CI 0.57, 0.87, 19 studies, I2 = 28.6%, moderate-quality evidence) with moderate heterogeneity.

Subgroup analyses showed that physical activity-only interventions resulted in greater reduction in risk for GDM in studies involving women with normal BMI compared with other BMI groups, and in interventions commencing before 12 gestation weeks (Supplementary Data 4).

Diet and physical activity

Meta-analysis of interventions with both diet and physical components showed a significant reduction in the risk of GDM (RR 0.82, 95%CI 0.73, 0.91, 59 studies, I2 = 47%, low-quality evidence) with significant heterogeneity.

Subgroup analyses showed that diet and physical activity interventions were effective in studies involving women with overweight or obesity, but not in studies involving women with normal weight (Supplementary Data 5). Diet and physical activity interventions were more effective in reducing GDM in studies involving women without PCOS compared to those with PCOS, and in studies involving women without a history of GDM compared with those to unspecified history of GDM (Supplementary Data 5). Meta-regression showed that diet and physical activity interventions had greater reduction in GDM with increasing age (Supplementary Data 3).

Metformin

Meta-analysis of all metformin interventions showed a significant reduction in the risk of GDM (RR 0.66, 95%CI 0.47, 0.93, 13 studies, I2 = 73%, very low-quality evidence) with significant heterogeneity.

Subgroup analyses showed that metformin interventions were more effective when commenced preconception compared with during pregnancy (Supplementary Data 6). Metformin interventions were also more effective in reducing GDM in studies involving women with PCOS than those with unspecified status, and less effective in studies involving women without a history of GDM than those with unspecified history of GDM (Supplementary Data 6). Meta-regression showed that metformin interventions were more effective in reducing GDM with increasing age or higher fasting blood glucose at baseline (Supplementary Data 3).

Myoinositol/Inositol

Meta-analysis of myoinositol/inositol interventions showed a significant reduction in the risk of GDM (RR 0.39, 95%CI 0.23, 0.66, 7 studies, I2 = 79%, very low-quality evidence) with significant heterogeneity.

Subgroup analyses showed that myoinositol/inositol interventions were more effective in reducing GDM in studies involving mostly White women compared with women from various ethnic backgrounds (Supplementary Data 7).

Probiotics

Meta-analysis of probiotics interventions showed no reduction in the risk of GDM (RR 0.88, 95%CI 0.52, 1.47, 5 studies, I2 = 74%, very low-quality evidence) with significant heterogeneity. Of these, one study with diet plus probiotics resulted in significant reduction in GDM (RR 0.36, 95%CI 0.18,0.72)28, however four studies with probiotics-only29,30,31,32 and one study with fish oil-plus probiotics32 did not reduce GDM (RR ranging from 0.59 to 1.5).

Subgroup analyses and meta-regression were not conducted in probiotics interventions due to small numbers in each type of intervention.

Fish oil

The single fish oil intervention study found did not reduce risk for GDM (RR 1.09, 95%CI 0.64, 1.85)32.

Sensitivity analysis

After excluding two non-RCT studies, diet-only interventions remained significant in preventing GDM (RR 0.75; 95% CI; 0.64, 0.88; I2 = 23%). After excluding six non-RCT studies, combined diet and physical activity interventions remained significant in reducing GDM (RR 0.83; 95% CI; 0.74, 0.93; I2 = 64.8%). After excluding five non-RCT studies, the effect of metformin on reducing the incidence of GDM was no longer significant (RR 1.05, 95% CI 0.89, 1.23; I2 = 45.1%).

Sensitivity analysis was not conducted on physical activity-only, probiotics and myoinositol/inositol studies as they all were RCTs.

Assessment of bias and quality of evidence

Some concerns or high risk of bias were found in 12 (80%) of diet-only, 16 (94%) physical activity-only, 46 (78%) of diet and physical activity, 10 (77%) of metformin, 5 (71%) in myoinositol/inositol, 3 (60%) in probiotics (Supplementary Table 5). Most studies assigned a high risk of bias had insufficient or non-blinding of participants as the main reason, which for lifestyle intervention is essentially not possible.

The quality of evidence rated by the GRADE approach found that the overall quality for diet-only or physical activity-only interventions were moderate, downgraded mainly due to most studies contributing to the outcome having high or some concerns in risk of bias. (Supplementary Table 5). The quality of evidence for diet and physical activity interventions were low, due to risk of bias and inconsistency. The quality of evidence for metformin interventions was very low, due to risk of bias, inconsistency and publication bias. The quality of evidence for myoinositol/inositol interventions was very low, due to risk of bias and inconsistency. The quality of evidence for probiotics interventions was very low due to risk of bias, inconsistency and imprecision.

Publication bias

Funnel plot and Egger’s test suggested the presence of small studies publication bias for metformin studies (P = 0.001) (Supplementary Fig. 1).

No significant publication bias was detected for studies on physical activity-only, diet-only, diet and physical activity, or probiotics (Supplementary Figs. 2–6).

Discussion

This study aimed to determine the effect of participant characteristics in interventions for GDM prevention. Our analyses showed that lifestyle interventions, metformin and myoinositol/inositol reduced the risk of GDM. For physical activity-only interventions, greater risk reduction for GDM was seen in studies involving women with normal BMI. Combined diet and physical activity interventions were more effective in GDM reduction in those with overweight or obesity, without PCOS, without history of GDM and with increasing age. Metformin interventions were more effective in GDM reduction in women with a history of PCOS and with increasing age and fasting blood glucose. Metformin or physical activity-only interventions were more effective when commenced preconception or in early gestation (before 12 gestation weeks).

Diet and physical activity interventions were more effective in lowering the risk for GDM among women without a history of GDM. This may be because women with a prior history of GDM have impaired beta-cell compensatory response during pregnancy and persistent or ongoing decline in insulin sensitivity post-GDM pregnancy11. Lifestyle modification alone may have limited ability to overcome these impairments in glycemic control in these individuals. In addition, lower adherence to a healthy diet, as lower dietary quality, has been observed among women with a history of GDM compared with women without a history of GDM, which may have also contributed to this finding33. Similarly, we also found that lifestyle interventions were more effective in lowering GDM risks in women without a history of PCOS. Like GDM, PCOS is also associated with increased insulin resistance34. In addition to the physiological challenges of insulin resistance, women with PCOS may also face further challenges with adhering to a healthy lifestyle which ranges from physiological barriers such as alteration in gut hormone regulation to psychological barriers such as a high prevalence of disordered eating in this population35. Further, we found that physical activity interventions were only effective in individuals with normal BMI but not in those with obese BMI. Similar observations have been reported in other meta-analyses of GDM prevention in individuals with excess body weight36,37. As obesity is associated with increased insulin resistance, it is possible that physical activity interventions alone could not reduce insulin resistance sufficiently for the prevention of GDM in these individuals. Further research is needed to determine if more intensive lifestyle intervention or additional co-intervention such as metformin or supplementation is needed to prevent GDM in women with conditions of high insulin resistance including prior GDM, PCOS and/or excess body weight. This is important so as not to provide unfounded expectations on the benefit of lifestyle on GDM prevention for certain groups. It is also of particular pertinence to these specific populations outlined, given the stigma associated with obesity and diabetes, resulting from the perception that these health outcomes are caused by personal failures38,39.

To date, systematic reviews are inconsistent in the observed effect of metformin on GDM prevention16,40. Heterogeneity in participant characteristics across studies, which undermines the power to detect a significant effect within a small number of studies, may contribute to this inconsistency16,40. Past studies conducted in select populations with homogeneous characteristics, such as studies in women with PCOS, have found a consistent benefit in GDM prevention with metformin41,42. By increasing the number of included metformin trials from 3 in the previous meta-analysis16 to 13, the current review revealed a significant reduction in GDM with metformin. To further explore the sources of heterogeneity by participant characteristics, our meta-regression additionally found that metformin is more effective in studies involving women with an older mean age at baseline, or in women with higher baseline fasting blood glucose. Increasing age is associated with greater insulin resistance, while higher baseline fasting glucose indicates early signs of failure of beta-cell compensatory response in insulin production43. The greater benefit of metformin in women with increased insulin resistance is in line with the known mechanisms of metformin, which is to reduce glucose production in the liver, improve peripheral glucose uptake and increase insulin sensitivity44. Our findings suggest that metformin may be the intervention of choice for preventing GDM in populations at high risk of insulin resistance, including in women with advanced age, higher fasting blood glucose, history of GDM or PCOS, along with healthy antenatal lifestyle advice. However, we found that the quality of evidence for metformin in preventing GDM was very low, thus further high-quality RCTs are needed to confirm these findings.

In this review, we identified seven preconception interventions that met our inclusion criteria. Despite this small number, metformin or physical activity-only interventions were more effective in lowering the risk for GDM when commenced preconception or in the first trimester of pregnancy. Our conclusions are similar to those of a previous meta-analysis15. Earlier initiation of interventions results in a greater duration of intervention exposure prior to GDM diagnosis, with the benefit of preconception commencement providing an opportunity to optimize insulin sensitivity prior to the onset of pregnancy-induced insulin resistance. A preconception intervention is in line with the concept of GDM arising as a result of a chronic metabolic condition that antedates pregnancy. Although GDM may be the first recognition of impaired fasting glucose or glucose intolerance, data suggest that women who develop GDM are already on a trajectory of increased cardiometabolic risk prior to pregnancy45. Future research should focus on providing preconception GDM prevention in those at risk of developing GDM. The use of preconception risk prediction models for GDM may help identify the populations who will optimally benefit from early initiation interventions46.

The strength of this review includes a comprehensive assessment of the impact of participant characteristics on the effectiveness of a broad range of interventions for the prevention of GDM, with a goal to identify populations that could optimally benefit from each intervention type. Although most studies were conducted in mixed populations and did not report outcomes according to subgroup characteristics, we coded the participant subgroups according to the inclusion and exclusion criteria (e.g. if the study included only women with obesity) to allow for group comparisons. This review also has several limitations. Participant characteristics that are known risk factors for GDM, such as parity, seldom feature in the inclusion or exclusion criteria of relevant studies, yielding a low number of subgroups available for comparison. This could be partly mitigated in individual level meta-analysis, if this information were collected at an individual level by the original studies. Significant heterogeneity remained in some subgroups, suggesting that other confounding factors may have contributed to the effect sizes. One of the confounding effects that are difficult to quantify is the changing GDM diagnostic criteria over the years which may have also contributed to the heterogeneity observed in the analyses. The findings on lifestyle or metformin on GDM incidence may also be limited by possible publication bias. The certainty of the evidence was very low for metformin, myoinositol/inositol and probiotics and low for diet and physical activity combined, and should be interpreted with caution.

Conclusion

Lifestyle, metformin and myoinositol/inositol interventions reduce the risk of GDM. Lower GDM risks were seen when the intervention commenced preconception or in the first trimester of pregnancy. Diet and physical activity interventions may be associated with a greater reduction in GDM risks in women with older age or without a history of GDM or PCOS, while metformin may be more effective in preventing GDM in women with older age, having higher fasting blood glucose or with PCOS. However, these results should be interpreted with caution due to limited reporting of intervention outcomes by participant characteristics in the individual studies. Given the potentially greater effectiveness of lifestyle and metformin interventions in individuals, future research on tailored recommendations in precision GDM prevention, replacing the current ‘one-size-fits-all’ approach, is needed. To advance knowledge in precision prevention, future research should include trials commencing in the preconception period and provide results disaggregated by a priori defined participant characteristics, including social and environmental factors, clinical traits, and other novel risk factors.