Lifestyle intervention during pregnancy in patients with gestational diabetes mellitus and the risk of neonatal hypoglycemia: A systematic review and meta-analysis

Objective Neonatal hypoglycemia is a severe adverse consequence of infants born to mothers with gestational diabetes mellitus (GDM), which can lead to neonatal mortality, permanent neurological consequences, and epilepsy. This systematic review and meta-analysis of randomized controlled trials (RCTs) was conducted to explore the effect of lifestyle intervention during pregnancy in women with GDM on the risk of neonatal hypoglycemia. Methods PubMed, Web of Science, Cochrane Library, CINAHL, and SPORTDiscus databases were searched by 1st April 2022. Data were pooled as the risk ratio (RR) with 95% CIs of neonatal hypoglycemia. Random-effects, subgroup analyses, meta-regression analysis, and leave-one-out analysis were conducted, involving 18 RCTs. Results Prenatal lifestyle intervention could significantly reduce the risk of neonatal hypoglycemia (RR: 0.73, 95% CI: 0.54–0.98, P = 0.037). Subgroup analysis further demonstrated that the reduced risk of neonatal hypoglycemia was observed only when subjects were younger than 30 years, initiated before the third trimester, and with dietary intervention. Meta-regression analysis revealed that the risk of neonatal hypoglycemia post lifestyle intervention was lower in mothers with lower fasting glucose levels at trial entry. Conclusion We found that prenatal lifestyle intervention in women with GDM significantly reduced the risk of neonatal hypoglycemia. Only lifestyle intervention before the third trimester of pregnancy, or dietary intervention only could effectively reduce the risk of neonatal hypoglycemia. Future studies are required to explore the best pattern of lifestyle intervention and to determine the proper diagnostic criteria of GDM in the first/second trimester of pregnancy. Systematic review registration https://www.crd.york.ac.uk/PROSPERO/#myprospero, PROSPERO, identifier: CRD42021272985.


Introduction
Gestational diabetes mellitus (GDM), defined as hyperglycemia with onset or first recognized during pregnancy, is associated with an increased risk of a range of adverse outcomes for offspring that can be passed on from generation to generation (1), namely, hypoglycemia (2), obesity, and type 2 diabetes (3)(4)(5). The incidence rate of GDM ranged from 3.4% to 37.7%, depending on different diagnostic criteria and population (6). A population-based cohort study showed that infants born to mothers with GDM had a significantly increased risk of hypoglycemia (OR: 11.71, 95% CI: 7.49-18.30) (7). The high concentration of blood glucose in the mother leads to an increase in fetal glucose intake, which stimulates excess fetal insulin secretion, thereby inducing neonatal hypoglycemia (8). Neonatal hypoglycemia is an important factor in neonatal mortality (9) and permanent neurological consequences (9,10). Even infants who are slightly and transiently exposed to hypoglycemia are at risk of later delayed neurodevelopment (11)(12)(13)(14)(15). Severe neonatal hypoglycemia can also lead to epilepsy, personality disorder, impaired heart function, and muscle weakness (9).
Lifestyle interventions, mainly dietary interventions, also including physical exercise interventions and other interventions, are usually the first-line strategy for managing GDM (16). A systematic review and meta-analysis of 19 controlled trials showed that dietary intervention during pregnancy effectively reduced the incidence rate of GDM (17). After being diagnosed with GDM, 70%−85% of patients were efficient to control blood glucose by lifestyle intervention per se according to American Diabetes Association (18). Studies have shown that the adverse perinatal outcomes caused by GDM might be improved with the treatment of GDM (19). Despite that, the effect of prenatal lifestyle intervention on the incidence of neonatal hypoglycemia remains inconclusive, although mounting studies have sprung up. Some results from human studies of randomized controlled trials (RCTs) suggested that lifestyle intervention could effectively reduce the risk of neonatal hypoglycemia (20)(21)(22). Recently, a large well-conducted RCTs revealed that the overall neonatal complications were significantly reduced by 47% post-smartphone-based lifestyle intervention (23). While some other evidences from human RCTs reported null effects on the risk of neonatal hypoglycemia (24)(25)(26).
Thus, the purpose of this review was to evaluate the effect of lifestyle intervention during pregnancy in women with GDM on the hypoglycemia risk of their neonate and to examine related influencing factors.

Literature search
This meta-analysis was carried out in compliance with the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA 2020) guidelines (27). The protocol number of this study is PROSPERO CRD 42021272985. Relevant articles published through 1st April 2022 were searched from PubMed, Web of Science, Cochrane Library, Cumulated Index to Nursing and Allied Health Literature (CINAHL), and SPORTDiscus (a full-text database of sports and sports medicine journals in the EBSCOhost), based on the Population, Intervention, Comparator, Outcome and Study design (PICOS) framework. The following search strategy was used: (lifestyle OR diet OR "physical exercise") AND (pregnancy OR "diabetes mellitus") AND "neonatal hypoglycemia" AND "randomized controlled trial." Detailed search terms are given in Supplementary Table 1.

Study selection
Two authors (YHW and HHZ) independently screened titles and abstracts, then reviewed the full text of all relevant studies for eligibility. The third researcher (YZ OR ZZ) arbitrated any discrepancies to reach a consensus. We included RCTs that evaluated the effect of lifestyle intervention during pregnancy on patients diagnosed with GDM with reported neonatal hypoglycemia. Inclusion criteria for considering studies for this review were: (1) types of studies: published randomized controlled trials; (2)

Data extraction and quality assessment
Data extraction and quality assessment were conducted by two trained people (YHW and HHZ) independently. The data were extracted to a form we designed, including: the first author's surname, publication year, study design, study location, sample size (intervention/comparators), age at pregnancy, gestational age at baseline, mother's fasting glucose level at baseline, maternal body mass index (BMI) (kg/m 2 ) at baseline or prepregnancy, intervention information, comparator information, and the outcome of neonatal hypoglycemia. Studies containing two or more intervention strata were analyzed as separate trials. The risk of bias (RoB) of each included study was assessed according to the criteria of the Cochrane Handbook for Systematic Reviews of Interventions (28). The Grading of Recommendation, Assessment, Development, and Evaluation (GRADE) approach was used to assess the quality of evidence (29). According to the GRADE handbook, study design dictates the baseline quality of the evidence (RCTs are initially assigned a ranking of high), and other factors could downgrade (risk of bias, inconsistency, indirectness of evidence, imprecision, and publication bias) or upgrade (large effect size, plausible residual confounding, and dose-response relationship) the quality of evidence. Discrepancies were resolved through discussion or by involving the third reviewer (YZ, OR, and ZZ).

Statistical analysis
A fixed-effect meta-analysis was used to pool estimates of summary risk ratio (RR) with 95% CIs of neonatal hypoglycemia. If substantial statistical heterogeneity was detected, the random-effects model was used to summarize the overall effect. Advanced data extraction was performed for studies that did not directly provide the mean and SD of continuous variables (30). The heterogeneity among studies was tested using the chi 2 test and quantified by I 2 -statistic (31). The presence of heterogeneity was indicated by P-value < 0.10 in the chi 2 test or I 2 > 30%.
Sources of potential heterogeneity were investigated by subgroup analyses and meta-regression based on age at pregnancy, gestational age at baseline, maternal fasting glucose level at baseline, and intervention types. The P-value < 0.1 was considered statistically significant in meta-regression analysis. Sensitivity analyses based on leave-one-out cross-validation were conducted to assess the robustness of the results in primary meta-analyses and to evaluate the impact of each trial on the heterogeneity, using a p-value < 0.05 as the criterion (32). Begg's and Egger's regression tests and funnel plots were used to assess possible publication bias. The P-value < 0.10 suggested the presence of publication bias (33). If publication bias was found, the trim and fill method was utilized (34).
Data analyses were performed using STATA version 11.0 (Stata Corp, College Station, TX, USA), with double data input to avoid input errors. The risk of bias in included studies was assessed using RevMan version 5.3 (Cochrane Collaboration, Oxford, UK). P < 0.05 was deemed as statistically significant unless specified elsewhere.

Literature search and study characteristics
As shown in Figure 1, the detailed process of literature search and study selection is presented in the flowchart. A total of 2,309 articles (337 from PubMed, 897 from Web of Science, 862 from Cochrane Library, 126 from CINAHL, and 87 from SPORTDiscus) were identified through initial searching, out of which 714 studies were removed because of duplication. Then 1,393 records were excluded after screening the title/abstract. The remaining 186 articles were eliminated for the following reasons: 44 studies were excluded due to inappropriate article design, 52 studies did not meet the inclusion criteria due to unsuitable participants, 31 studies lacked proper treatment, and 59 studies lacked sufficient data. Additional 3 studies were included through evaluating the reference lists of the included articles. Eventually, 18 eligible studies were included in the final quantitative synthesis.

Excluded studies
In total, six studies (45-50) included subjects without GDM, additional two studies (51, 52) included women with impaired glucose tolerance but did not meet the diagnosis of GDM as defined by trialists. Two studies (53, 54) did not use an intervention/comparison included in this review.

Risk of bias in included studies
As shown in Figures 2, 3, three studies (21, 24, 42) were considered to be of unclear risk of bias for randomization, the . /fnut. .

FIGURE
The forest plot demonstrated the e ect of prenatal lifestyle intervention in women with GDM on the risk of neonatal hypoglycemia by pooling data from studies.  other 15 (19, 20, 22, 23, 25, 26, 35-41, 43, 44) studies were at low risk of bias. Three studies (21, 24,37) were considered to be of high risk of bias for allocation concealment, one study (42) was judged to be of unclear risk of bias. Four studies (20,21,23,26) had a high risk of performance bias, one study (42) was judged to be of unclear risk of bias. Two studies (19,39) had an unclear risk of detection bias. One study (43) was judged as high risk for attrition bias and two studies (24,41) were considered to be of unclear risk for attrition bias. One study (43) was considered to be of high risk for reporting bias. Five studies (21, 24,26,37,42) had unclear risk of other biases (potential biases related to the study design).

E ect of lifestyle intervention on the risk of neonatal hypoglycemia
In total, 18 articles included 19 trials that explored the effect of lifestyle intervention on the risk of neonatal hypoglycemia. With consideration of the heterogeneity, we used the randomeffect model to get pooled estimates. Results from our metaanalysis suggested that lifestyle intervention during pregnancy could significantly reduce the risk of neonatal hypoglycemia (RR: 0.73, 95% CI: 0.54 to 0.98, P = 0.037) (Figure 4). The results of the I² and chi 2 test demonstrated that there was substantial heterogeneity in the primary meta-analysis (I 2 = 48.9%; P = 0.009) ( Table 2). The Begg's and Egger's tests indicated no significant publication bias in the primary analysis ( Table 2).

Subgroup and meta-regression analysis
The results of the subgroup analysis were summarized in Table 2. The results of subgroup analysis revealed that the effect of lifestyle intervention on the risk of neonatal hypoglycemia was influenced by maternal age at pregnancy and gestational age at trial entry. Lifestyle intervention was associated with a decrease in the risk of neonatal hypoglycemia only in studies with mothers younger than 30 years (RR: 0.52, 95% CI: 0.27-0.99, P = 0.046), but not in studies with mothers ≥ 30 years (RR: 0.81, 95% CI: 0.57-1.17, P = 0.272) ( Figure 5A). Furthermore, significant reductions of risk of neonatal hypoglycemia post lifestyle intervention were observed only in studies with gestational age < 14 weeks (first trimester) at trial entry (RR: 0.30, 95% CI: 0.10-0.86, P = 0.025) and in studies with gestational age between 14 and 28 weeks (second trimester) at trial entry (RR: 0.66, 95% CI: 0.44-0.98, P = 0.039), but not in studies with gestational age ≥ 28 weeks (third trimester) at trial entry (RR: 0.85, 95% CI: 0.61-1.19, P = 0.337) ( Figure 5B  other interventions ( Figure 5C). The Begg's and Egger's tests indicated no significant publication bias in the above subgroup analyses ( Table 2). A meta-regression analysis was conducted to explore the potential sources of heterogeneity. Among selected covariates, including maternal age at pregnancy, gestational age at trial entry, and maternal fasting glucose level at trial entry, the results of meta-regression analysis revealed that maternal fasting glucose level at trial entry was a potential confounder of the effect of lifestyle intervention on the risk of neonatal hypoglycemia, with adjusted R 2 of 45.04% (Table 3 and Figure 6). There was no significant association between the risk ratio and other covariates listed in Table 3.

Sensitivity analysis
Regarding the robustness of overall effect sizes, we performed a leave-one-out cross-validation for sensitivity analysis (Figure 7). The results of the leave-one-out crossvalidation suggested that three studies (19,20,37) contributed most to the heterogeneity in the primary meta-analysis. After excluding these studies, the pooled results remained significant (RR: 0.71, 95% CI: 0.54-0.93, P = 0.012), thus the effect of the intervention on the risk of neonatal hypoglycemia might be underestimated due to heterogeneity between studies.

Evidence level rated by GRADE
According to the GRADE protocol, as shown in Table 4, the evidence level of the overall meta-analysis was at a moderate level because significant heterogeneity existed.

Discussion
All in all, there was "moderate" quality evidence from 18 RCTs indicating that prenatal lifestyle intervention in women with GDM was associated with a 27% decreased risk of having a baby with hypoglycemia. Subgroup analysis further demonstrated that the reduced risk of neonatal hypoglycemia post lifestyle intervention was observed only in studies with subjects younger than 30 years, initiated in the first or second trimester, and with dietary intervention. In addition, metaregression analysis revealed that maternal fasting glucose levels at trial entry were positively associated with the risk ratio of neonatal hypoglycemia post lifestyle intervention.
To the best of our knowledge, this is the first comprehensive systematic review and meta-analysis that evaluated the effect of prenatal lifestyle intervention in women with GDM on the risk of neonatal hypoglycemia. Unlike our results, relevant systematic reviews to date did not find any significant benefit of lifestyle intervention for neonatal hypoglycemia. The metaanalysis from Cochrane (55) included six studies, that reported no significant association between the lifestyle treatment of GDM and the risk of neonatal hypoglycemia. Five of the six studies were also included in our meta-analysis, but the remaining one study (51) was excluded because the study population was women with impaired glucose tolerance rather than patients with GDM. In addition, a systematic review (56) reported no difference between infants exposed to control or diet and exercise interventions before birth for the risk of hypoglycemia, which was based on two studies with overweight or obese women. Unlike our findings, another meta-analysis (57) based on two studies, which were both included in our review, found that dietary intervention in patients with GDM did not change the neonatal outcome of hypoglycemia. In our present study, we found that only dietary intervention in patients with GDM could reduce the risk of neonatal hypoglycemia, but not dietary intervention plus other interventions. This might be due to dietary advice or counseling (rather than supplements) being the main form of dietary intervention in the subgroups of dietary plus other interventions with or without exercise (one (19) of two (19,26) studies in the group of dietary plus other interventions and three (20,24,43) of five (20,21,23,24,43) studies in the group of dietary plus exercise plus other interventions applied individualized dietary advice or counseling intervention). Since the small number of studies with other interventions included in our study, future large-scale RCTs are still required to further explore the best pattern of lifestyle intervention during pregnancy to reduce the risk of neonatal hypoglycemia. There is no relevant report describing long-term follow-up of hypoglycemia outcomes after changing lifestyle during pregnancy.
. /fnut. .   The role of lifestyle intervention has been greatly appreciated as a clinical treatment in GDM (58). In our present metaanalysis, prenatal lifestyle intervention resulted in a significant reduction in the risk of neonatal hypoglycemia. Achieving glycemic control in women with GDM is critical for reducing the risk of neonatal hypoglycemia (59). Notably, the glucose threshold recommended by recent guidelines (1, 60, 61) is lower than previously recommended (62, 63) for the diagnosis of GDM. Therefore, less severe hyperglycemia has been classified as GDM in recent years, which is also conducive to the effect of lifestyle intervention. Maternal age is known to affect the outcomes of pregnancies (64). Recently, a meta-analysis of 24 studies showed that the risk of GDM increased by 7.90% with each-one year increase in maternal age from 18 (65). The increase in maternal age is also related to the incidence of macrosomia (66), small for gestational age (67), and cesarean section (68), which are all independent risk factors for neonatal hypoglycemia (69). Consequently, we postulated that the adverse . /fnut. . outcomes associated with advanced maternal age might lead to an increased incidence of neonatal hypoglycemia, thereby obscuring the effect of lifestyle interventions. Our meta-analysis could suggest that pregnant women younger than 30 years might be an appropriate population to observe the improved effect on neonatal hypoglycemia post lifestyle intervention. Traditionally, screening tests for the diagnosis of GDM are performed at 24-28 weeks of gestation. A prospective cohort study of 4,069 women showed that the increase in fetal growth was not obvious when GDM was diagnosed at 20 weeks of gestation, but it was significantly increased when GDM was diagnosed ≥ 28 weeks (70). In addition, fetal growth in obese women increased when GDM was diagnosed at 20 weeks of gestation (70). This indicated that late diagnosis might miss the opportunity for intervention, especially for a highrisk population. Our findings suggest that lifestyle intervention might be effective for reducing the risk of neonatal hypoglycemia when initiated before 28 weeks of gestation. And this increases the question of the current diagnosis time of GDM. It is necessary to conduct GDM screening in early pregnancy to increase the opportunity of benefiting from early intervention. Therefore, we suggest that the diagnostic criteria of GDM in early pregnancy should be determined reasonably.
An observational epidemiological international multi-ethnic investigation found that intrauterine exposure to higher levels of glucose was associated with childhood obesity and insulin resistance, which was independent of maternal BMI and family history of diabetes (71,72). A prospective study of patients with GDM found that infants whose mothers had the lowest blood glucose levels before and at birth had the lowest incidence of neonatal hypoglycemia (73). Likewise, our results of meta-regression analysis suggested that the risk of neonatal hypoglycemia post lifestyle intervention was lower in patients with GDM having lower maternal fasting glucose levels at trial entry. Thus, maternal fasting glucose levels at trial entry might be the potential source of heterogeneity among studies.
Our meta-analysis has some limitations. One possible limitation may be related to the different GDM diagnostic criteria used in the included studies, which might lead to the inevitable existence of heterogeneity between studies. In fact, the diagnostic criteria of GDM have been controversial (60). In addition, there were a variety of lifestyle interventions included in this study, and the number of studies reporting each intervention was limited, so we cannot determine which specific intervention is more effective. Furthermore, we could not find out how much the degree of hypoglycemia is worse or improves with a change in lifestyle due to the limitation of data. Lastly, significant heterogeneity among studies was found. According to the result of the meta-regression, the heterogeneity might be attributed to the differences in maternal fasting glucose levels at trial entry.

Conclusion
This current meta-analysis of 18 RCTs demonstrated that lifestyle intervention during pregnancy in women with GDM significantly reduced the risk of neonatal hypoglycemia, especially when subjects were younger than 30 years old, or lifestyle intervention initiated before the third trimester, or with dietary intervention. However, future well-designed largescale trials are still required to further explore the best pattern of lifestyle intervention and to determine the proper early diagnostic criteria for GDM.

Data availability statement
The original contributions presented in the study are included in the article/supplementary material, further inquiries can be directed to the corresponding authors.