Relazione tra i miglioramenti del controllo glicemico e il rischio di complicanze della gravidanza in pazienti con diabete mellito: analisi di metaregressione di studi randomizzati controllati sulla gestione intensiva della glicemia

1. Introduction

Prevalence of diabetes in pregnancy has increased in parallel with the worldwide epidemic of obesity. The increase includes not only the prevalence of Type 1 diabetes (T1D) and Type 2 diabetes (T2D) (pregestational diabetes) among those of reproductive age but also gestational diabetes mellitus (GDM) [ ]. Diabetes confers significant maternal and fetal risks such as miscarriage, preeclampsia, macrosomia, neonatal hypoglycemia, hyperbilirubinemia, and neonatal respiratory distress syndrome (RDS) [ ]. In particular, poor GC has been associated with a high risk of adverse maternal and fetal outcomes such as cesarean section (CS), shoulder dystocia, and macrosomia [ ].

Interventions mainly aimed at improving GC in pregnant women with pregestational diabetes [ ] and GDM [ – ] are effective in reducing the risk of some maternal and fetal/neonatal complications. However, it is an unsolved issue of whether the risk reduction is dependent on GC or could otherwise be attributed to an education program regardless of the improvement in GC. To address this issue, we performed a meta-analysis of previous trials of intensive glucose management among patients with diabetes focusing on the impact of GC during pregnancy on the risk of adverse pregnancy outcomes.

2. Methods

### 2.1. Protocol

Before performing this meta-analysis, the protocol was registered in the international prospective register of systematic reviews (PROSPERO) (CRD42022356069).

### 2.2. Literature Search

Electronic literature search was conducted for randomized controlled trials (RCTs) published from Jan. 1, 1950, to Apr. 29, 2024 using MEDLINE and Embase. Search strategy is shown in . We listed any possible terms related to the following four elements: (1) pregnant women with diabetes mellitus, (2) intensive glucose management, (3) adverse pregnancy outcomes, and (4) RCT. These were combined using the Boolean operator “AND.”

### 2.3. Inclusion and Exclusion Criteria

Study inclusion criteria were (1) RCT with a parallel design; (2) all participants had diabetes including T1D, T2D, pregestational diabetes, overt diabetes, or GDM; (3) interventions during pregnancy that included two groups consisting of an intervention and a control group; (4) assessment of the incidence of at least one adverse pregnancy outcome as a study outcome; and (5) assessment of improvement of GC as the exposure.

As to Criteria 3, the intervention had to include intensive glucose management by at least one of the following: (a) frequent monitoring of blood glucose, (b) addition of a glucose-lowering drug or increasing dosage of an already prescribed glucose-lowering drug, and (c) establishing a strict glycemic target and adjusting drug dosages to achieve the glycemic target. We excluded articles having an abstract only or were presented at a conference because study quality could not be assessed. We also excluded trials in which the intensity of glucose management was not differentiated (e.g., Web-based vs. in-person education, metformin vs. insulin).

### 2.4. Definition of Study Outcomes and Exposures

We considered all maternal or fetal/neonatal adverse outcomes that occurred from the beginning of pregnancy until the perinatal period. Exceptionally, maternal hypoglycemia was not considered because it is not specific to pregnant women with diabetes mellitus. The corresponding effect measure was relative risk (RR) in the intervention group versus the control group. To calculate RR, included trials had to provide data on the number of cases and noncases for each adverse pregnancy outcome. If there was no event in either the intervention or control group, we added 0.5 to the number of patients and events in each group [ ].

The GC indicators included A1C, fasting plasma glucose (FPG), postprandial glucose, and mean blood glucose (MBG). As to study exposures, included studies had to provide information to calculate the mean difference between intervention and control groups in the blood glucose level after the intervention involving at least one of the four GC indicators. If we could not determine the number of cases and noncases or the mean difference in GC indicators between the intervention and control groups, we queried the study author for data clarification.

### 2.5. Data Extraction

Two authors (S.K. and K.F.) extracted the data. Disagreements were solved by the third author (H.So.). Extracted data were as follows: first author, year, country of the study, number of patients, type of diabetes, mean age, mean body mass index (BMI) at entry or pregestational BMI, methods for glucose management, whether the methods included drug interventions, blood glucose levels after the intervention, and adverse pregnancy outcomes.

Study quality was assessed using the revised Cochrane risk-of-bias tool for randomized trials (RoB 2) [ ]. This tool is structured into five domains through which bias might be introduced into the results: R, bias arising from the randomization process; D, bias due to deviations from intended interventions; Mi, bias due to missing outcome data; Me, bias in measurement of the outcome; and S, bias in selection of the reported results. For each domain, we allocated one of three categories: “high,” “some concerns,” and “low.” Overall risk of bias was considered “high” if there was a high risk of bias in at least one domain and/or some concerns about risk of bias in three or more domains. It was considered low if risk of bias was considered low in any domain. Otherwise, there were some concerns about the overall risk of bias (i.e., neither low nor high risk of bias).

### 2.6. Data Synthesis

Primarily, the overall RR of each adverse pregnancy outcome was estimated using the Mantel–Haenszel method [ ]. Between-study heterogeneity was assessed using I -squared statistics [ ]. If significant heterogeneity using Cochrane's Q test [ ] was statistically detected, a random-effects model was adapted; otherwise, a fixed-effects model was chosen. Analysis was stratified by the following trials' characteristics: methods for intensive glucose management, whether a drug intervention was included or not, type of diabetes (i.e., GDM or others), and geographic region (i.e., Asia/Middle East or others). The influence of these characteristics on the pooled RR was examined by metaregression using the STATA command “metareg.” Publication bias was visually assessed by funnel plots and formally assessed by two statistical tests, Begg's rank correlation test [ ] and Egger's regression asymmetry test [ ]. If a statistically significant publication bias was detected by at least one of the above two tests, we used a trim–fill method to adjust the pooled estimate for publication bias [ ]. This method includes the assumption that the funnel plots of the results (i.e., effect sizes and their corresponding standard errors) of published studies are symmetrical without publication bias. If the funnel plots are asymmetrical, then the results from hypothetically unpublished studies that caused the asymmetrical funnel plots were trimmed to restore the symmetry, and the pooled estimate was recalculated including the results from both the published and hypothetically unpublished studies.

Secondarily, to examine the dose–response relationship between improved GC and the reduced risk of adverse pregnancy complications, the RRs for each adverse outcome were regressed on the differences in each GC indicator between intervention and control groups, where the RRs were weighted by their corresponding inverse of logarithms of RR. The correlation coefficient ( r ) in each metaregression analysis was calculated as follows: r = ± 1 − SS after / SS before , where SS before and SS after were the sum of the square before and after entering the difference as the explanatory variable and r was minus if reductions in a GC indicator were negatively associated with risk reductions in an adverse pregnancy outcome. p < 0.05 was considered to be statistically significant except that p < 0.10 was used to detect a significant publication bias [ ]. All statistical analyses were conducted using STATA statistical software (Version 16, Stata Corporation, College Station, Texas, United States).

3. Results

### 3.1. Literature Searches

is a flowchart showing the selection of eligible studies. Of 2840 articles retrieved from electronic literature searches, 319 articles were kept for full-paper review. In addition to 252 articles that were found to be ineligible, we excluded five trials because the authors did not respond to our queries. In one of the five articles [ ], the full paper was unavailable despite a request to obtain it. The other four of the five trials had insufficient data to perform this meta-analysis. It was impossible to estimate the number of cases and noncases in one study [ ] and impossible to calculate the mean difference in blood glucose levels between groups in three studies [ – ]. Finally, 62 ( = 319 − 252 − 5) eligible trials [ , – ] consisting of 11,989 patients (6030 in the intervention group and 5959 patients in the control group) were identified and included in this meta-analysis.

### 3.2. Study Characteristics

summarizes the characteristics of trials included in this meta-analysis. Thirty-six trials were conducted in Asia and the Middle East. Twenty-four trials were conducted in Europe and North America, and the other two trials were multicenter studies conducted in two or more countries. Most of the included trials (50/62 trials) targeted patients with GDM while four trials and one trial targeted patients with T1D and T2D, respectively. Another four trials targeted those with pregestational or overt diabetes. The type of diabetes was not specified in the remaining three trials. summarizes blood glucose levels after the intervention in both the intervention and control groups. As the GC indicator, A1C, FPG, 2-h postprandial glucose (2hPG), or MBG was used in 41 trials, 39 trials, 30 trials, and 17 trials, respectively.

Methods for glucose management were categorized into four types: (1) glucose-monitoring only (22 trials), (2) introducing new glucose-lowering drugs such as sulfonylurea and insulin (11 trials), (3) setting a strict glycemic goal (5 trials), and (4) multifocal (i.e., combining 1–3) (24 trials). Except for two trials, mean age was homogeneous among studies and ranged from 25 to 35 years. Obesity indicator was assessed by prepregnancy BMI/body weight and/or BMI/body weight at study entry. Adverse pregnancy outcomes examined in the included trials are summarized in . There were 36 types of adverse (19 maternal and 17 fetal/neonatal) outcomes consisting of 15,430 events that were identified by two or more trials. shows maternal and fetal/neonatal adverse outcomes identified in each trial. The number of examined outcomes ranged from 1 to 22 (median, 9 outcomes).

### 3.3. Study Quality

Results of risk of bias using RoB 2 are shown in . Reasons for a high risk of bias were that “randomization” was not stated (R domain), per-protocol analysis was used although there were patients lost to follow-up (D domain), and the arbitrary use of medical records or description of outcomes in the Results section although they were not mentioned in the Methods section (S domain). Other reasons for concerns about the risk of bias were not describing methods for randomization (R domain), not specifying outcomes in the Methods section (S domain), and neither mentioning nor having robust or missing data although these were unlikely to influence study results (Mi domain). All trials had low risk of bias for the Me domain. Overall, we finally judged 17, 20, and 25 trials as having low, unknown, and high risk of bias, respectively.

### 3.4. Overall Effect of Intensive Glucose Management on the Risk of Adverse Pregnancy Outcomes

shows the overall effect of intensive glucose management on the risk of maternal and fetal adverse pregnancy outcomes. Risks of 17 of 36 specified outcomes were significantly reduced through intensive glucose management. Intensive glucose management did not elevate the risk of any of the adverse outcomes. Corresponding forest plots showing RRs with 95% CI from included trials by individual adverse outcomes are shown in Supporting Information .

The statistically significant publication bias that was detected in the 14 overall estimates and the corresponding funnel plots (Supporting Information ) suggested that all risk reductions were exaggerated due to the existence of hypothetically unpublished studies. However, adjustment for publication bias using the trim–fill method did not change the general conclusions except for the pooled RR for pregnancy-induced hypertension (PIH), neonatal hypoglycemia, and RDS, for which significantly low RRs were changed to insignificant.

shows stratified analyses of the effect of intensive glucose management on adverse pregnancy outcomes; these analyses were limited to 14 outcomes for which there were 15 or more pieces of data including nonzero events. A larger effectiveness of intensive glucose management was observed in trials conducted in Asia and the Middle East compared with those in countries other than in Asia and the Middle East ( p = 0.01 for preterm delivery; p = 0.01 for macrosomia; p = 0.002 for neonatal hypoglycemia; p = 0.004 for admission to neonatal intensive care units (NICUs); p = 0.01 for RDS). In general, the content of the intensive glucose management did not affect the magnitude of the risk of adverse pregnancy outcomes except for neonatal hypoglycemia ( p = 0.01) and admission to NICU ( p = 0.02). In these outcomes, multifocal interventions yielded significantly larger risk reductions ( p < 0.001 for neonatal hypoglycemia; p = 0.045 for admission to NICU) while using only drug interventions was significantly less efficacious in reducing the risk of admission to NICU ( p = 0.003) compared with the use of other methods of intensive glucose management.

Including drug interventions for GC did not contribute to risk reductions and was less efficacious than not including drug interventions in reducing risk of admission to NICU ( p = 0.004) and RDS ( p = 0.04). As to small for gestational age (SGA), risk was significantly elevated by setting a strict GC goal (RR [95% CI], 5.40 [1.44–20.23]; p = 0.04 for difference from other methods of glucose management) and including a drug intervention (RR [95% CI], 1.36 [1.05–1.75]). Intensive glucose management was more efficacious in trials of patients with GDM than in those with patients not having GDM in reducing the risk of macrosomia ( p = 0.03) and fetal distress ( p = 0.03). The risks were significantly reduced in trials for patients with GDM (RR [95% CI], 0.63 [0.56–0.71] for macrosomia; 0.38 [0.28–0.52] for fetal distress) but not significantly reduced in trials for patients not having GDM (RR [95% CI], 0.93 [0.69–1.25] for macrosomia; 0.92 [0.55–1.54] for fetal distress). The risk of SGA was significantly elevated in trials for patients with other types of diabetes (RR [95% CI], 1.84 [1.11–3.04]) but not in trials for patients having GDM (RR [95% CI], 1.09 [0.84–1.41]).

### 3.5. Metaregression Analysis to Explore the Dose–Response Relationship Between Reductions in GC Indicators and Risk of Adverse Pregnancy Outcomes

shows the RR for incremental reductions in the four GC indicators in the same 14 adverse pregnancy outcomes as described in . Supporting Information shows the corresponding scatter plots where the RRs for each of the 14 outcomes were plotted on the reductions in each of the four GC indicators. There were no significant relationships between reductions in any GC indicator and RRs for SGA, premature rupture of membranes (PROMs), and congenital malformation. However, as to the other 11 outcomes, the RRs for incremental reductions in at least one GC indicator were significantly lowered (i.e., < 1).

Among the four GC indicators, for FPG, there was the largest number of significantly lowered RRs (10 of 14 outcomes) although the RR for 10 mg/dL reductions was not significant for CS (RR [95% CI], 0.86 [0.72–1.01] [ p = 0.07]) as well as for SGA, PROM, and congenital malformation as previously mentioned. Significantly lowered RR for 1% reductions in A1C was observed in six adverse outcomes. In particular, regarding reduced risk of CS, there was a strong association between A1C reductions and reduced risk of CS ( r = 0.59; RR [95% CI] for a 1% reduction in A1C, 0.66 [0.53–0.83]; p = 0.001). However, there was no relationship between reduced A1C and the RRs for postpartum hemorrhage ( r = 0.37; p = 0.49), admission to NICU ( r = 0.26; p = 0.36), RDS ( r = 0.19; p = 0.55), and fetal distress ( r = 0.48; p = 0.16) although significant associations were observed between FPG reductions and risk reductions in these outcomes (RR [95% CI] for 10 mg/dL reductions in FPG, 0.74 [0.57–0.97] [ r = 0.62, p = 0.03] for postpartum hemorrhage; 0.54 [0.34–0.87] [ r = 0.66, p = 0.02] for admission to NICU; 0.56 [0.38–0.85] [ r = 0.68, p = 0.01] for RDS; and 0.49 [0.29–0.84] [ r = 0.72, p = 0.01] for fetal distress).

Significantly lowered RR for 10 mg/dL reductions in 2hPG was observed in four adverse outcomes. In particular, the strength of association of 2hPG reductions expressed as a correlation coefficient ( r ) was prominent for preterm delivery ( r = 0.73, p < 0.001), PIH ( r = 0.72, p = 0.01), and hyperbilirubinemia ( r = 0.84, p < 0.001) compared with that of reductions in the other GC indicators. However, in these four outcomes, the RR for 10 mg/dL reductions in FPG was also significantly lowered. The significantly lowered RR for 10 mg/dL reductions in MBG was observed only in two outcomes (RR [95% CI], 0.71 [0.55–0.91] [ r = 0.65, p = 0.01] for macrosomia; 0.72 [0.55–0.94] [ r = 0.69, p = 0.02] for RDS) partly because the number of data was too small to detect statistical significance.

### 3.6. Sensitivity Analyses of Metaregression

Limiting analyses to trials of patients with GDM did not change the general conclusions partly because the majority of included trials targeted patients with GDM ( ). Corresponding scatter plots are presented in Supporting Information . We stratified analyses by the geographic regions where trials were conducted (i.e., Asia and Middle East or countries other than in those regions), limiting exposures to reductions in three GC indicators (i.e., A1C, FPG, and 2hPG) and outcomes to six complications (i.e., CS, macrosomia, preterm delivery, PIH, neonatal hypoglycemia, and hyperbilirubinemia) so that there were sufficient data for at least 20 RRs to be regressed for reductions in at least one of the three GC indicators ( and Supporting Information ). In the analysis that included trials conducted in Asia and the Middle East, reductions in at least one GC indicator among A1C, FPG, and 2hPG were considered to be positively related to risk reductions in any of the six adverse outcomes. However, no such associations were noted in the analysis of trials conducted in countries other than in Asia or the Middle East. In particular, in Asia and the Middle East, reductions in A1C were positively associated with risk reductions in PIH (RR [95% CI] for 1% of A1C reduction, 0.38 [0.19–0.78] [ r = 0.70, p = 0.01]) but there was a negative association between reductions in A1C and risk reductions in PIH in countries other than in Asia or the Middle East (RR [95% CI] for 1% of A1C reduction, 14.76 [1.36–160.65] [ r = −0.67, p = 0.03]).

4. Discussion

The current metaregression analyses showed that reduced blood glucose levels were associated with a risk reduction in 11 of 14 main adverse pregnancy outcomes in patients with diabetes in a dose–response fashion. It can be considered that the improved GC induced by intensive glucose management is associated with an improved prognosis of pregnant women with diabetes. According to Hill's criteria, which is a valid tool for establishing causation, a dose–response gradient suggests that the observed association is causal [ , ]. It has been suggested that improved GC causes the improved prognosis in pregnant women with diabetes. To support this suggestion, further meta-analyses are needed to examine the dose–response relationship by expanding the included trials to those with interventions such as exercise and supplements (e.g., probiotics [ ]) regardless of whether the intervention included intensive glucose management, which was the only intervention examined in the present meta-analysis.

A positive relationship between reductions in blood glucose levels and risk of adverse pregnancy outcomes is pathophysiologically plausible. The hyperglycemia–hyperinsulinemia hypothesis (also known as the Pedersen hypothesis [ , ]) traditionally proposed that maternal hyperglycemia leads to fetal hyperglycemia, which stimulates maturation and hypertrophy of the fetal pancreas. This results in fetal hyperinsulinemia and neonatal hypoglycemia. Since insulin is a dominant fetal growth hormone, fetal hyperinsulinemia accelerates fetal growth and, therefore, macrosomia is more likely to occur [ ]. Blood insulin levels in pregnant women with hyperglycemia are usually high, which can promote increased renal sodium reabsorption and increased blood volume. It can also enhance the response of small blood vessels in the whole body to sympathetic nerve excitation, which results in PIH [ ]. In addition, fetal hyperglycemia increases osmotic diuresis which subsequently leads to polyuria and polyhydramnios [ ]. Fetal hyperglycemia and hyperinsulinemia can independently cause fetal hypoxia [ ]. Polycythemia and hyperbilirubinemia are understood to be counterregulatory mechanisms for this state of hypoxia as it triggers erythropoietin secretion and increased red cell production [ ]. In addition to the hyperglycemia–hyperinsulinemia hypothesis, a nonpathophysiological explanation is also applicable to the relationship between GC and risk of preterm delivery. Low socioeconomic status (SES) is generally associated with elevated risk of preterm delivery [ ], and individuals with low SES also have worse GC than individuals with a higher SES [ , ]. The socioeconomic factor could be a mediator between blood glucose levels and risk of preterm delivery.

The American Diabetes Association (ADA) recommends monitoring both FPG and postprandial glucose levels [ ]. However, our meta-analysis does not seem to support this recommendation. Our analysis showed that the reduction in FPG could explain risk reductions of any of the pregnancy complications that could be explained by the reduction in 2hPG, suggesting that the major adverse pregnancy outcomes could be covered by monitoring only FPG levels even if 2hPG was not monitored. This suggestion favors the management of glucose levels considering that assessing FPG is simpler and less time-consuming than assessing postprandial glucose. Concerns also have been raised about the reproducibility of postprandial glucose levels [ ]. However, monitoring postprandial glucose may be essential for patients with normal FPG but abnormal postprandial glucose levels. Effective monitoring of such patients to prevent pregnancy complications should be further investigated.

In addition, our meta-analysis also did not necessarily support the following ADA statement: “as A1C represents an integrated measure of glucose, it may not fully capture postprandial hyperglycemia, which drives macrosomia. Thus, although A1C may be useful, it should be used as a secondary measure of glycemic control in pregnancy, after blood glucose monitoring” [ ]. The current metaregression analysis showed that A1C reduction was more greatly associated with reduction of risks of CS than reductions in FPG, 2hPG, and MBG. Previous cohort studies reported that pre-existing diabetes or GDM itself was associated with CS, irrespective of other medical indications such as macrosomia [ , ]. The risk for CS could be mediated by clinical practice patterns whereby physicians consider patients with poor GC, which is generally represented by high A1C values rather than high FPG or postprandial plasma glucose values, as having a high-risk pregnancy and refer them to surgeons for CS. However, our results suggest that FPG is superior to A1C for assessing risk of other pregnancy complications. The metaregression analysis ( ) indicated that although reductions in FPG were significantly associated with the reduced risk of postpartum hemorrhage, RDS, admission to NICU, and fetal distress, A1C reductions were not significantly associated with any of these outcomes. Similarity of the associations of FPG with RDS, admission to NICU and fetal distress could be explained by the finding that RDS was among the most common causes of NICU admissions and infants with RDS had lower Apgar scores than those without RDS [ ]

Interestingly, A1C reductions were associated with the reduced risk of hyperbilirubinemia ( r = 0.47, p = 0.03) ( ) but not RDS ( r = 0.19, p = 0.55) (see Results). Conversely, MBG reductions were associated with the reduced risk of RDS ( r = 0.69, p = 0.02) (see Results) but not hyperbilirubinemia ( r = 0.20, p = 0.67) ( ). Although both A1C and MBG reflect mean 24-h blood glucose values, A1C reflects average glucose levels over several preceding weeks. It is speculated that the risk of hyperbilirubinemia is influenced by glycemia earlier in pregnancy, whereas that of RDS is more strongly associated with glycemia later in pregnancy. Further research should clarify the stage of pregnancy when GC is critical according to the incidence of pregnancy outcomes.

The risks of some adverse outcomes were not reduced by intensive glucose management. A previous meta-analysis indicated that preconception care was associated with reduced risk of congenital malformation (RR = 0.29) and SGA (RR = 0.52) [ ]. However, the current meta-analysis did not show that intensive glucose management significantly reduced the risk of these adverse outcomes. The focus of the interventions in our meta-analysis was during pregnancy, suggesting that even earlier (i.e., before pregnancy) glucose management is required for prevention of congenital malformations and SGA. However, it is possible that intensive glucose management during pregnancy could have been effective but was not statistically proved because the incidence of congenital malformations could have been too low for the detection of statistical significance.

Previous data showed that low blood glucose during pregnancy was associated with an increased risk of having a SGA infant [ , ]. However, our meta-analysis did not show evidence that intensive glucose management was harmful in terms of bearing a SGA infant. Furthermore, the dose–response data (see ) show that the reduction in 2hPG was borderline associated with reduction of risks for SGA ( p = 0.050), suggesting that the improvement of GC is favorable for preventing SGA. Not only low blood glucose values but also various risk factors such as gestational hypertension, eclampsia, and preeclampsia were linked to the risk of SGA [ ]. Considering that our meta-analysis indicated that reductions in A1C, FPG, and 2hPG were associated with the risk reduction of PIH ( ), intensive glucose management possibly prevents SGA via lowering the risk of PIH. However, according to the stratified analysis ( ), setting a strict glycemic target elevated the risk of SGA. Excessive GC, which can cause hypoglycemia, may elevate the risk of SGA. However, when GC management is adequate but not excessive, it will not elevate the risk of SGA.

The lack of a relationship between reductions in blood glucose levels and risk of PROM should be addressed although the current meta-analysis indicated that intensive glucose management per se lowered the risk of PROM. Etiology of PROM is multifocal, including demographic and clinical factors such as smoking, previous preterm delivery, and cervical surgery as well as choriodecidual infection [ ]. It has been speculated that various risk factors for PROM cannot be ameliorated by improving GC.

Of note, this meta-analysis provides evidence of the relatively low efficacy of intensive glucose management and reduction in blood glucose levels in terms of preventing pregnancy complications for patients with diabetes living in Europe and North America compared with those in Asia and the Middle East. The influence of race or ethnicity on the benefit of glucose management cannot be discussed because, to our knowledge, previous studies have not explored this issue. Nevertheless, the reason for the large differences in results between trials conducted in Asia/Middle East and Europe/North America should be discussed. Reviewing the study characteristics of participants in the included trials ( ) showed that the mean BMI was significantly higher in trials conducted in countries other than in Asia and the Middle East compared with those in Asia and the Middle East (mean [SD] (kilogram/square meter unit), 31.6 [3.9] vs. 28.0 [3.7] for BMI at entry ( p = 0.02 and 26.3 [2.0] vs. 23.9 [1.6] for prepregnancy BMI ( p = 0.02))). One possible explanation is that, in general, patients living in Europe and North America were more obese and thus had higher risks of adverse pregnancy outcomes than those in Asian countries, irrespective of GC [ ]. The large-scale prospective study utilizing the UK Biobank showed that obesity indicators such as BMI, waist circumference, and body fat percentage were associated with higher odds ratios for preeclampsia and gestational hypertension than elevated blood glucose [ ]. It is probable that GC has a relatively low priority for preventing adverse pregnancy outcomes in pregnant women living in Europe or North America, where in general obesity is more prevalent than in Asia and the Middle East.

Several limitations should be addressed. First, a large part of included trials targeted patients with GDM. We could not perform a sensitivity metaregression analysis wherein analyses were limited to trials for T1D or T2D patients. Second, the number of data was insufficient to detect statistical significance for some outcomes. Third, criteria for adverse outcomes such as macrosomia and fetal distress varied among trials, and failure to standardize these criteria is of concern. Fourth, there was statistically significant publication bias for some of the adverse pregnancy outcomes. However, its impact could not discussed because, unfortunately, there is no method to adjust for publication bias in a metaregression analysis.

In conclusion, the current results indicated that risk reductions of the majority of pregnancy complications in diabetes depend on improved GC induced by intensive glucose management. To support that such improvement in GC improves the prognosis of pregnant women with diabetes, further meta-analyses are needed expand to any intervention whether or not it includes intensive glucose management.