Gestational Diabetes Mellitus and Diet: A Systematic Review and Meta-analysis of Randomized Controlled Trials Examining the Impact of Modified Dietary Interventions on Maternal Glucose Control and Neonatal Birth Weight

Gestational diabetes mellitus (GDM) is one of the most common medical complications in pregnancy and affects an estimated 14% of pregnancies, or one in every seven births globally (1). Women with GDM and their offspring are at increased risk of both short- and longer-term complications, including, for mothers, later development of type 2 diabetes, and for offspring, increased lifelong risks of developing obesity, type 2 diabetes, and metabolic syndrome (2–6). The adverse intrauterine environment causes epigenetic changes in the fetus that may contribute to metabolic disorders, the so-called vicious cycle of diabetes (7).

The mainstay of GDM treatment is dietary and lifestyle advice, which includes medical nutrition therapy, weight management, and physical activity (8). Women monitor their fasting and postmeal glucose levels and adjust their individual diet and lifestyle to meet their glycemic targets. This pragmatic approach achieves the glycemic targets in approximately two-thirds of women with GDM (8). However, despite the importance of medical nutrition therapy and its widespread recommendation in clinical practice, there are limited data regarding the optimal diet for achieving maternal euglycemia (8–11). It is also unknown whether the dietary interventions for achieving maternal glycemia are also effective for reducing excessive fetal growth and adiposity (12).

Different dietary strategies have been reported including low glycemic index (GI), energy restriction, increase or decrease in carbohydrates, and modifications of fat or protein quality or quantity (12–14). Three recent systematic reviews have been performed examining specific diets and pregnancy outcomes (15–17). Viana et al. (16) and Wei et al. (15) concluded that low-GI diets were associated with a decreased risk of infant macrosomia. However, the most recent systematic review from Cochrane, including 19 trials randomizing 1,398 women, found no clear difference in large for gestational age or other primary neonatal outcomes with the low-GI diet (17). The primary maternal outcomes were hypertension (gestational and/or preeclampsia), delivery by cesarean section, and type 2 diabetes, outcomes for which most trials lacked statistical power, even when dietary subgroups were combined. Remarkably, no systematic reviews examined the impact of modified dietary interventions on the detailed maternal glycemic parameters, including change in glucose-related variables, the outcomes that are most directly influenced by diet.

To address this knowledge gap, we performed a systematic review and meta-analysis of randomized controlled trials to investigate whether modified dietary interventions (defined as a dietary intervention different from the usual one used in the control group) in women with GDM offer improved glycemic control and/or improved neonatal outcomes when compared with standard diets.

In accordance with a published protocol (PROSPERO CRD42016042391), we performed a systematic review and meta-analysis. Reporting is in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (18). An international panel of experts was formed by the International Life Sciences Institute Europe. This panel determined the review protocol and carried out all aspects of the review.

The following databases were searched for all available dates using the search terms detailed in Supplementary Table 1: PubMed, MEDLINE, Cochrane Central Register of Controlled Trials (CENTRAL), Embase, Cumulative Index to Nursing and Allied Health Literature (CINAHL), Web of Science Core Collection, Applied Social Sciences Index and Abstracts, ProQuest, ProQuest Dissertations & Theses—A&I and UK & Ireland, National Institute for Health and Care Excellence evidence search, Scopus, UK Clinical Trials Gateway, ISRCTN, and ClinicalTrials.gov. The initial search was performed in July 2016. An updated search of MEDLINE, Embase, CENTRAL, and CINAHL was performed on 3 October 2017 using the same search terms.

A hand-search of relevant reviews and all included articles was conducted to identify studies for potential inclusion. As well, experts on the panel were consulted for the inclusion of additional articles. Reference management was carried out using EndNote.

All titles and abstracts were assessed independently and in duplicate to identify articles requiring full-text review. Published studies fulfilling the following criteria were included: randomized controlled trials, evaluated modified dietary interventions on women with GDM, glucose intolerance or hyperglycemia during pregnancy, reported-on primary maternal and neonatal outcomes, included women aged 18–45 years, had a duration of 2 weeks or more, and were published in English, French, Spanish, Portuguese, Italian, Dutch, German, or Chinese. We excluded studies that included participants with type 1 or type 2 diabetes if data for participants with GDM were not presented independently, if dietary characteristics were not available, if the study was in animals, or if the study did not report outcomes of interest. We did not include studies of nutritional supplements such as vitamin D or probiotics as recent reviews have addressed these topics (19,20).

All citations identified after title and abstract assessment were full-text reviewed in duplicate. Reasons for exclusion at the full-text review stage were recorded. Any disagreements between reviewers were resolved by consensus and with consultation with the expert group when required.

Data from included studies were extracted in duplicate using prespecified data extraction forms. Extracted data elements included study and participant demographics, study design, diagnostic criteria for GDM, glucose intolerance or hyperglycemia, funding source, description of modified dietary intervention and comparator, and maternal and neonatal outcomes. For studies with missing data, inconsistencies, or other queries, authors were contacted. Record management was carried out using Microsoft Excel and RevMan.

For articles providing information on maternal weight, fasting glucose, postprandial glucose, HbA_1c, or HOMA insulin resistance index (HOMA-IR) at baseline and postintervention but not their change, change was calculated as the difference between postintervention and baseline. Standard deviations were imputed using the correlation coefficient observed in articles reporting full information on the variable at baseline and postintervention and its change or a correlation coefficient of 0.5 when this information was not available (21). As studies differed in postprandial glucose at baseline, glycemic control at study entry was not considered to be equivalent in both arms, and thus continuous glucose-related variables at follow-up are reported as change from baseline.

The primary outcomes were maternal glycemic outcomes (mean glucose, fasting glucose, postprandial glucose [after breakfast, lunch, and dinner and combined], hemoglobin A_1c [HbA_1c], assessment of insulin sensitivity by HOMA-IR, and change in these parameters from baseline to assessment; medication treatment [defined as oral diabetes medications or insulin]) and neonatal birth weight outcomes (birth weight, macrosomia, and large for gestational age).

Data were pooled into relative risks (RRs) or mean differences with 95% CI for dichotomous outcomes and continuous outcomes, respectively. Meta-analysis was performed using random-effects models. A prespecified analysis stratified by type of diet and quality assessment was performed to explore potential reasons for interstudy variation. Heterogeneity was assessed using I² statistics. Small study effects were examined for using funnel plots. Analyses were conducted using RevMan version 5.3. Pooled estimation of birth weight in the study and control arms, both overall and according to the specific diet intervention, was performed using Stata 14.0.

Methodological quality and bias assessment was completed by two reviewers. Risk of bias was assessed using the Cochrane Collaboration tool, which rates seven items as being high, low, or unclear for risk of bias (21). These items included random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective outcome reporting, and other potential sources of bias (21). A sensitivity analysis was performed excluding articles with relevant weaknesses in trial design or execution.

The overall quality of the evidence was also assessed using Grading of Recommendations Assessment, Development and Evaluation (GRADE) working group guidelines (21). GRADE was assessed for all primary and secondary outcomes, both maternal and neonatal, but without subgroup analysis per different dietary intervention for each outcome measure.

We screened 2,269 records for potential inclusion, and 126 articles were reviewed in full (Supplementary Fig. 1). Eighteen studies (12–14,22–36) were included in the meta-analysis with a total of 1,151 pregnant women with GDM.

The types of modified dietary intervention included low-GI (n = 4), Dietary Approaches to Stop Hypertension (DASH) (n = 3), low-carbohydrate (n = 3), fat-modification (n = 2), soy protein–enrichment (n = 2), energy-restriction (n = 1), high-fiber (n = 1), and ethnic diets (i.e., foods commonly consumed according to participant’s ethnicity) (n = 1) and behavioral intervention (n = 1). Details of the study characteristics are included in Table 1. Most trials were single centered and had small sample sizes (range 12–150). Only two trials (one each from Spain and Australia) included over 100 participants, nine had 50–100 participants, and seven studies had fewer than 50 participants. They were performed in North America, Europe, or Australasia and all had a duration of at least 2 weeks. The ethnicity of participants was reported in seven studies (12,13,26,29,31,32,34).

Most studies assessed individual dietary adherence using food diaries (13,23–36). Although most studies did report an overall difference in dietary composition between the intervention diet and control diet, few studies reported a detailed assessment of dietary adherence. Only five studies used a formal measure of adherence (24,25,29,33,34), and four of them reported data (25,29,33,34). Adherence ranged from 20% to 76% in the control groups and 60% to 80% in the intervention groups.

When baseline characteristic data were pooled, women in the intervention group were older than women in the control group (pooled mean difference 0.60 years [95% CI 0.06, 1.14]) and had higher postprandial glucose (5.47 [0.86, 10.08]), most influenced by the DASH and ethnic diet studies. There was no overall significant difference between the intervention and control groups for BMI, gestational age at enrollment, fasting glucose, HbA_1c, or HOMA-IR.

Pooled risk ratios in 15 studies involving 1,023 women demonstrated a lower need for medication (RR 0.65 [95% CI 0.47, 0.88]; I² = 55) (Table 2). Thirteen studies (n = 662 women) reported fasting glucose levels, nine (n = 475) reported combined postprandial glucose measures, and three (n = 175) reported post-breakfast glucose measures. Pooled analysis demonstrated a larger decrease in fasting, combined postprandial, and post-breakfast glucose levels in modified dietary interventions (mean −4.07 mg/dL [95% CI −7.58, −0.57], I² = 86, P = 0.02; −7.78 mg/dL [−12.27, −3.29], I² = 63, P = 0.0007; and −4.76 mg/dL [−9.13, −0.38], I² = 34, P = 0.03, respectively) compared with control group. There were no significant differences in change in HbA_1c (seven studies), HOMA-IR (four studies), or in post-lunch or -dinner glucose levels (two studies).

Pooled mean birth weight was 3,266.65 g (95% CI 3,172.15, 3,361.16) in the modified dietary intervention versus 3,449.88 g (3,304.34, 3,595.42) in the control group. Pooled analysis of all 16 modified dietary interventions including 841 participants demonstrated lower birth weight (mean −170.62 g [95% CI −333.64, −7.60], I² = 88; P = 0.04) and less macrosomia (RR 0.49 [95% CI 0.27, 0.88], I² = 11; P = 0.02) compared with conventional dietary advice (Table 2 and Fig. 1). There was no significant difference in the risk of large-for-gestational-age newborns in modified dietary interventions as compared with control diets (RR 0.96 [95% CI 0.63, 1.46], I² = 0; P = 0.85).

Pooled analysis of low-GI diets showed a larger decrease in fasting (26,29,30), postprandial, and post-breakfast glucose compared with control diets (26,30) (Table 2). However, the pooled analysis of the DASH diet showed significant favorable modifications in several outcomes, including change in fasting (22,36) and postprandial glucose (22), HOMA-IR (35), HbA_1c (22), medication need (22,23,36), infant birth weight (23,36), and macrosomia (23,36) (Tables 2 and 3). Last, pooled analysis of the soy protein–enriched diet demonstrated a significant decrease in medication use and birth weight (14,27) (Tables 2 and 3). One soy–protein intervention (n = 68 participants) described significantly lower HOMA-IR (27) (Table 2).

Behavioral (one study) and ethnic-specific modified dietary interventions (one study) were included. The behavioral change dietary intervention reported significant differences in change in postprandial glucose and in HbA_1c (Table 2) (24). The ethnic diet study demonstrated a significantly larger decrease in fasting and postprandial glucose (Table 2) (34). Fat-modification, low-carbohydrate, and energy-restriction diets were not associated with a significant difference in our primary outcomes in the stratified analysis.

Weight gain from inclusion was lower for low-carbohydrate diets and cesarean birth for DASH diets (Supplementary Table 2). Specific diet interventions did not show significant between-group differences in maternal gestational weight gain throughout pregnancy, preeclampsia/eclampsia, neonatal hypoglycemia as defined by the authors, preterm birth, neonatal intensive care unit admission, or small-for-gestational-age newborns (Supplementary Tables 2 and 3).

Sensitivity analysis was performed to explore reasons for heterogeneity and to assess outcomes when studies with methodological concerns were removed. We were unable to include four studies (22,23,34,36), including all the DASH diet studies, where clarification of certain aspects of the results could not be obtained, even after a direct approach to the authors. The authors of the ethnic diet study responded to queries but did not provide the required information regarding gestational age at randomization (34). After these studies are removed, the changes in postprandial glucose (mean −5.90 mg/dL [95% CI −7.93, −3.88], I² = 0; P = 0.0001), post-breakfast glucose levels (−4.76 mg/dL [−9.13, −0.38], I² = 34; P = 0.03), and birth weight (−74.88 g [−144.86, −4.90], I² = 1; P = 0.04) remained significant when all diets were combined (Table 3). Furthermore, the heterogeneity in most primary outcomes decreased after removal of these four studies.

When dietary subgroups were assessed, low-GI diets had significant differences in changes in fasting (mean −5.33 mg/dL [95% CI −6.91, −3.76]) (26,29,30), postprandial (−7.08 mg/dL [−12.07, −2.08]) (26,30), and post-breakfast (−8.6 mg/dL [−14.11, −3.09]) glucose (26,30). The soy protein–enriched diet had differences in change of HOMA-IR (mean −2.00 [95% CI −3.17, −0.83]) (27), required less medication use (RR 0.44 [95% CI 0.21, 0.91]), and had a lower birth weight (mean −184.67 g [95% CI −319.35, −49.98]) (14,27). The behavior modification diet had significant differences in change in postprandial glucose (mean −6.90 mg/dL [95% CI −9.85, −3.95]) and in HbA_1c (−0.19% [−0.26, −0.12]) (24) (Table 3).

None of the included studies were assessed as having a low risk of bias in all seven items of the Cochrane Collaboration tool (Supplementary Fig. 2). Most studies were high risk for blinding of participants and personnel and for other sources of bias (Supplementary Fig. 3). Studies scored high risk for other sources of bias for concerns such as baseline differences and industry funding. Most studies had an unclear risk of bias for selective outcome reporting and very few had registered protocols (Supplementary Fig. 3).

GRADE assessment for the outcomes of interest reveals overall low to very low quality of evidence (Supplementary Table 4). Considerations to downgrade quality of evidence involved the entire spectrum, including limitations in the study design, inconsistency in study results, and indirectness and imprecision in effect estimates.

Funnel plots of means and RRs of the primary outcomes for the main analysis are shown in Supplementary Figs. 4 and 5 and for the sensitivity analysis in Supplementary Figs. 6 and 7. Overall, funnel plot asymmetry improves with the sensitivity analysis compared with the main analysis for neonatal birth weight outcomes.

In this meta-analysis, we pooled results from 18 studies including 1,151 women with a variety of modified dietary interventions. Remarkably, this is the first meta-analysis with a comprehensive analysis on maternal glucose parameters. Despite the heterogeneity between studies, we found a moderate effect of dietary interventions on maternal glycemic outcomes, including changes in fasting, post-breakfast, and postprandial glucose levels and need for medication treatment, and on neonatal birth weight. After removal of four studies with methodological concerns, we saw an attenuation of the treatment effect. Nonetheless, the change in post-breakfast and postprandial glucose levels and lowering of infant birth weight remained significant. Given the inconsistencies between the main and sensitivity analyses, we consider that conclusions should be drawn from the latter. These data suggest that dietary interventions modified above and beyond usual dietary advice for GDM have the potential to offer better maternal glycemic control and infant birth weight outcomes. However, the quality of evidence was judged as low to very low due to the limitations in the design of included studies, the inconsistency between their results, and the imprecision in their effect estimates.

Previous systematic reviews have focused on the easier-to-quantify outcomes, such as the decision to start additional pharmacotherapy and glucose-related variables at follow-up, but did not address change from baseline (15–17). The most recently published Cochrane systematic review by Han et al. (17) did not find any clear evidence of benefit other than a possible reduction in cesarean section associated with DASH diet. The very high-carbohydrate intake (∼400 g/day) and 12 servings of fruit and vegetables in the DASH diet (22,23,36) limit its clinical applicability and generalizability to women from lower socioeconomic, inner city backgrounds in Western countries. The Cochrane review shared one of our primary outcomes, large for gestational age (17). Neither meta-analysis detected a significant difference in risk of large for gestational age because the trials with a larger effect on birth weight (the three DASH studies) did not report on large for gestational age.

Our findings regarding pooled analysis of low-GI dietary interventions are broadly consistent with those of Viana et al. (16) and Wei et al. (15). Viana et al. (16) noted decreased birth weight and insulin use based on four studies of low-GI diet among 257 women (mean difference −161.9 g [95% CI −246.4, −77.4] and RR 0.767 [95% CI 0.597, 0.986], respectively). Wei et al. (15) also reported decreased risk of macrosomia with a low-GI diet in five studies of 302 women (RR 0.27 [95% CI 0.10, 0.71]). In our analyses of four studies in a comparable number of participants (n = 276), we found the same direction of these effect estimates, without significant between-group differences. This is most likely due to the different studies included. For example, we were unable to obtain effect estimates stratified by type of diabetes in the study by Perichart-Perera et al. (which included women with type 2 diabetes) and therefore did not include this study (37). An important difference between our analyses and that of Wei et al. (15) is that they included DASH diet as a low-GI dietary subtype. We also included a recent study by Ma et al. (30) not included by the previous reviews.

Our sensitivity analyses highlighted concerns regarding some studies included in previous reviews. Notably, after removal of the studies with the most substantial methodological concerns in the sensitivity analysis, differences in the change in fasting plasma glucose were no longer significant. Although differences in the change in postprandial glucose and birth weight persisted, they were attenuated.

This review highlights limitations of the current literature examining dietary interventions in GDM. Most studies are too small to demonstrate significant differences in our primary outcomes. Seven studies had fewer than 50 participants and only two had more than 100 participants (n = 125 and 150). The short duration of many dietary interventions and the late gestational age at which they were started (38) may also have limited their impact on glycemic and birth weight outcomes. Furthermore, we cannot conclude if the improvements in maternal glycemia and infant birth weight are due to reduced energy intake, improved nutrient quality, or specific changes in types of carbohydrate and/or protein.

We have not addressed the indirect modifications of nutrients. For example, reducing intake of dietary carbohydrates to decrease postprandial glucose may be compensated by a higher consumption of fat potentially leading to adverse effects on maternal insulin resistance and fetal body composition. Beneficial or adverse effects of other nutrients such as n-3 long-chain polyunsaturated fatty acid, vitamin D, iron, and selenium cannot be ruled out.

Our study has important strengths and weakness. To our knowledge, ours is the first systematic review of dietary interventions in GDM comprehensively examining the impact of diet on maternal glycemic outcomes assessing the change in fasting and postprandial glucose, HbA_1c, and HOMA-IR from baseline. This is especially important given that groups were not well balanced at baseline. Our review also benefits from the rigorous methodology used as well as the scientific, nutritional, and clinical expertise from an international interdisciplinary panel. However, it also has limitations. Baseline differences between groups in postprandial glucose may have influenced glucose-related outcomes. Furthermore, three of the included trials were pilot studies and therefore not designed to find between-group differences (12,26,34). The low number of studies reporting on adherence clearly illustrates that the quality of the evidence is far from ideal. The heterogeneity of the dietary interventions even within a specific type (varied macronutrient ratios, unknown micronutrient intake, and short length of some dietary interventions) and baseline characteristics of women included (such as prepregnancy BMI or ethnicity) may have also affected our pooled results. It should also be noted that the relatively small numbers of study participants limit between-diet comparisons. Last, we were unable to resolve queries regarding potential concerns for sources of bias because of lack of author response to our queries. We have addressed this by excluding these studies in the sensitivity analysis.

Modified dietary interventions favorably influenced outcomes related to maternal glycemia and birth weight. This indicates that there is room for improvement in usual dietary advice for women with GDM. Although the quality of the evidence in the scientific literature is low, our review highlights the key role of nutrition in the management of GDM and the potential for improvement if better recommendations based on adequately powered high-quality studies were developed. Given the prevalence of GDM, new studies designed to evaluate potential dietary interventions for these women should be based in larger study groups with appropriate statistical power. As most women with GDM are entering pregnancy with a high BMI, evidence-based recommendations regarding both dietary components and total energy intake are particularly important for overweight and obese women. The evaluation of nutrient quality, in addition to their quantity, as well as dietary patterns such as Mediterranean diet (39) would also be relevant. In particular, there is an urgent need for well-designed dietary intervention studies in the low- and middle-income countries where the global health consequences of GDM are greatest.

Funding. H.R.M. was funded by the U.K. National Institute for Health Research (CDF 2013-06-035). This work was conducted by an expert group of the European branch of the International Life Sciences Institute (ISLI Europe). This publication was coordinated by the ISLI Europe Early Nutrition and Long-Term Health and the Obesity and Diabetes task forces. Industry members of these task forces are listed on the ILSI Europe website at www.ilsi.eu. Experts are not paid for the time spent on this work; however, the nonindustry members within the expert group were offered support for travel and accommodation costs from the Early Nutrition and Long-Term Health and the Obesity and Diabetes task forces to attend meetings to discuss the manuscript and a small compensatory sum (honoraria) with the option to decline. The expert group carried out the work, i.e. collecting and analyzing data and information and writing the scientific paper, separate to other activities of the task forces. The research reported is the result of a scientific evaluation in line with ILSI Europe’s framework to provide a precompetitive setting for public-private partnership. ILSI Europe facilitated scientific meetings and coordinated the overall project management and administrative tasks relating to the completion of this work.

The opinions expressed herein and the conclusions of this publication are those of the authors and do not necessarily represent the views of ILSI Europe nor those of its member companies. For further information about ILSI Europe, please email [email protected] or call +32 2 771 00 14.

Duality of Interest. E.M.v.d.B. works part-time for Nutricia Research. E.C.-G. works full-time for Nestec. R.R. works full-time for Abbott Nutrition. No potential conflicts of interest relevant to this article were reported.

Author Contributions. J.M.Y. contributed to data extraction, statistical analyses, and writing the first draft manuscript. J.E.K. contributed to data extraction and writing the first draft summary tables. M.B. and A.G.-P. contributed to literature extraction, statistics, and manuscript revision. E.H. contributed to data extraction and GRADE assessments. I.S. and I.G. contributed to statistics and manuscript revision. E.M.v.d.B., E.C.-G., S.H., and S.F.O. contributed to concept and design, data extraction, and manuscript review. M.H. contributed to concept and design and draft manuscript evaluation. K.L. contributed to concept and design, data extraction, and critical review for intellectual content. L.P. contributed to concept and design and manuscript review. R.R., P.R., and H.R.M. contributed to concept and design, data extraction, and revising the draft manuscript. L.v.L. contributed to data extraction and draft summary tables. B.S. contributed to data extraction and critical review for intellectual content. R.C. contributed to literature extraction, statistical analyses, and revising the draft manuscript. R.C. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this work were presented at the Diabetes UK National Diabetes in Pregnancy Conference, Leeds, U.K., 14 November 2017, and the XXIX National Congress of the Spanish Society of Diabetes, Oviedo, Spain, 18–20 April 2018.