Introduction

The prevalence of maternal diabetes during pregnancy has increased over the last two decades, with every indication that this rise will continue. In 2004, five pregnancies per 1,000 in the UK were complicated by pre-existing diabetes, an increase of more than 50% since 1996, reflecting a sixfold increase in type 2 diabetes and a 20% increase in type 1 diabetes [1]. Over the same period the prevalence of gestational diabetes mellitus (GDM) has doubled and currently affects 40 pregnancies per 1,000 [2].

Exposure to a diabetic intrauterine environment has long been recognised as a risk to the fetus. A growing body of evidence suggests that intrauterine exposure to diabetes may ‘programme’ long-term effects in offspring, as first described in 1979 by Freinkel and Metzger as fuel-mediated teratogenesis [3, 4]. Follow-up studies of offspring of diabetic pregnancies have demonstrated an increased rate of diabetes or pre-diabetes in the offspring [5] and features of the ‘metabolic syndrome’ [6] (obesity, glucose intolerance, hypertension and dyslipidaemia). Animal studies also show that intrauterine exposure to hyperglycaemia increases the risk of overweight, abnormal glucose tolerance and insulin resistance in offspring [7, 8]. Studies in Pima Indians show that the intrauterine environment is an important determinant in the development of diabetes, independent of genetic factors [9]. Children born after their mother developed type 2 diabetes had 3.7 times the risk of developing type 2 diabetes and a higher BMI than siblings born before their mother became diabetic [10]. Although data such as these indicate that a diabetic pregnancy predisposes offspring to obesity and abnormal glucose tolerance independent of genetic determinants [11], the strength of the association and the mediating biological pathways are unknown.

An elevated BMI is a risk factor for the development of type 2 diabetes and often the first marker of the metabolic syndrome [12]. Given the increased risk of developing type 2 diabetes in later life in children born to mothers with diabetes [5], it is plausible that this might be mediated through an increase in offspring BMI. However, to date individual studies examining BMI in offspring of diabetic mothers (ODM) have been small with limited power and contradictory results [13–15].

The primary aim of this systematic review and meta-analysis was to establish the impact of maternal diabetes on offspring BMI z score. Secondary objectives were to distinguish the effect of type of maternal diabetes, and effects after adjustment for maternal BMI.

Methods

Literature search

A systematic review of published studies reporting outcomes of diabetic and non-diabetic pregnancies was undertaken in accordance with the Meta-analysis of Observational Studies in Epidemiology (MOOSE) guidelines for systematic reviews of observational studies [16].

The outcome studied was offspring BMI z score, standardised for age and sex. All types of maternal diabetes mellitus, pre-pregnancy type 1, type 2 and gestational onset, were considered as exposures. For inclusion, an unselected non-diabetic control group must have been reported within the same paper. Inclusion criteria for offspring were age between 3 and 16 years of age and offspring of a singleton pregnancy.

A search was conducted in PubMed (www.ncbi.nlm.nih.gov) for studies published before 1 November 2010, using Medline subject heading keywords (Pregnancy in diabetics OR diabetes, gestational) and (Prenatal exposure delayed effects OR Child). The review was limited to human studies published in English. The primary investigator, L. H. Philipps, assisted by E. Prior, conducted the searches and identified studies relevant to the outcome by evaluating the abstract or obtaining a full copy of the paper if no abstract was available. Reference lists of papers retrieved were hand-searched for inclusion in the review. Wherever possible, forward citations of the studies retrieved during the literature search were traced. In studies where anthropometric data on the offspring had been obtained but no BMI z score reported, efforts were made by M. J. Hyde to contact the author to obtain the relevant data. Review articles and commentaries were excluded. Where one or more papers had been published on the same cohort, only the study that reported the outcomes for the cohort at an age closest to the median age of the studies overall was included.

Data extraction and analysis

Information on the characteristics of the study population, outcome, exposure, covariates, results and conclusions were independently extracted from each paper by L. H. Philipps, C. Gale and checked by M. J. Hyde and S. Santhakumaran. Study quality was examined in respect to blinding of assessors to maternal diabetes status when measuring offspring outcomes.

A meta-analysis of studies examining the association between diabetes during pregnancy and offspring BMI z score was conducted by L. H. Philipps, S. Santhakumaran and M. J. Hyde in RevMan 5 (5.0.24) [17] using the inverse variance method. Subgroup analysis was carried out to compare offspring BMI z score in relation to type of maternal diabetes. The differences between the subgroup effects were tested for significance using an interaction test [18].

Where studies only reported subgroup analyses for different types of diabetes, pooled means and standard deviations were calculated for use in the meta-analysis of all diabetes types combined, provided that recruitment was not stratified on the basis of type of diabetes, which would prevent the results being generalisable to the population.

A meta-analysis was carried out on studies that reported offspring BMI z score adjusted for maternal pre-pregnancy BMI. Where data adjusted for maternal pre-pregnancy BMI were not available, difference in mean maternal pre-pregnancy BMI between cases and controls was calculated for each individual study. For each study, the difference in mean maternal BMI was plotted against the mean difference in offspring BMI z score.

A fixed effects analysis was initially carried out for all comparisons. Heterogeneity was assessed using the χ 2 test on Cochrane’s Q statistic [19] and by calculating I 2 [20]. If significant heterogeneity was present (p < 0.05 from the χ 2 test) a random effects meta-analysis was carried out. Potential sources of heterogeneity were investigated by comparing study designs. If study covariates that might explain heterogeneity were identified, meta-regression was used to investigate the significance of the associations. Similarly, where there was doubt as to the eligibility or quality of a study the meta-analysis was performed both with and without the study in order to verify the sensitivity of the conclusions.

Forest plots were used to illustrate study findings and meta-analysis results, and funnel plots to investigate evidence of publication bias. If funnel plots showed asymmetry, Egger’s test was performed [21].

Results

Literature search

Using PubMed to perform the literature search, we identified 340 papers, of which 28 matched the inclusion criteria [13, 14, 22–47]. Two additional studies were identified through a search of forward citations [15, 48] and four [49–52] from hand-searching the reference lists of included papers, giving 34 papers in total. Five cohorts were reported in more than one paper, so only the study that reported the outcomes at an age closest to the overall median age of all studies (7 years) was included in the meta-analysis; 10 papers were excluded on this basis [25, 35, 36, 38–40, 42, 47–49]. This resulted in 24 papers that remained for inclusion in the review. A study in Pima Indians was excluded because the control group (mothers who subsequently developed diabetes) was not comparable with other studies [28]. In 16 studies outcome information was not provided in a form suitable for meta-analysis; for example, outcomes were represented as OR for obesity, or height, weight, or BMI, rather than as z scores. We attempted to make contact with the corresponding author and additional data were received for two studies [22, 23]. The authors of 10 studies were not able to provide suitable data, and the authors of four studies did not respond or declined to provide the information requested. Final data suitable for inclusion in the meta-analysis were available for nine studies [13–15, 22, 23, 27, 29, 46, 50]. All were cohort studies (seven prospective and two retrospective). The search strategy is shown in Fig. 1, and a description of included studies is provided in Table 1.

Fig. 1
figure 1

Flow chart of the search strategy used in this review. The relevant number of papers at each point is given

Full size image
Table 1 Studies included in this meta-analysis examining the association between intrauterine exposure to maternal diabetes and offspring BMI z score
Full size table

Offspring BMI z score

A forest plot of the studies comparing BMI z score in ODM and offspring of non-diabetic mothers is shown in Fig. 2. The pooled mean difference from the fixed effects analysis is 0.21 (95%CI 0.13, 0.28; p < 0.00001). As there is evidence of heterogeneity (p < 0.0007; I 2 70%), a random effects analysis was undertaken giving a pooled estimate of 0.28 (95%CI 0.09, 0.47; p < 0.004); the pooled estimate is presented in Fig. 2.

Fig. 2
figure 2

Forest plot showing the unadjusted association between all types of maternal diabetes and offspring BMI z score. Heterogeneity: τ 2 = 0.05; χ 2 = 27.02, df = 8 (p = 0.0007); I 2 = 70%. Test for overall effect: z = 2.90 (p = 0.004). IDM, infant of diabetic mother; IV, inverse variance; ref., reference

Full size image

Subgroup analyses

A forest plot of the studies comparing offspring of mothers with gestational-onset diabetes and offspring of non-diabetic mothers is shown in Fig. 3. The pooled mean difference in BMI z score from the fixed effect analysis in this subgroup is 0.20 (95% CI 0.12, 0.28; p < 0.00001), showing an increase in ODM. There is significant evidence of heterogeneity (p = 0.001) and the I 2 is 76%. When a random effects analysis is performed, the pooled estimate is 0.28 (95% CI 0.05, 0.51; p = 0.02); the pooled estimate is presented in Fig. 3.

Fig. 3
figure 3

Forest plot showing the unadjusted pooled analysis of offspring BMI z score of mothers with gestational diabetes mellitus and controls. Heterogeneity: τ 2 = 0.06; χ 2 = 25.54, df = 5 (p = 0.001); I 2 = 76%. Test for overall effect: z = 2.39 (p = 0.02). IDM, infant of diabetic mothe; IV, inverse variance; ref., reference

Full size image

Data from offspring of mothers with type 1 diabetes (pre-pregnancy) also show an increase in BMI z score compared with control offspring. The pooled mean difference in BMI z score is 0.29 (95% CI 0.02, 0.55; p < 0.03) (Fig. 4). There is no significant evidence of heterogeneity (p = 0.27) and the I 2 is 25%. The BMI difference between the offspring of type 1 diabetic mothers compared with controls is similar to the difference between the offspring of GDM and controls (0.29 vs 0.28); an interaction test confirmed that this difference is not significant (p = 0.95)

Fig. 4
figure 4

Forest plot showing the unadjusted pooled analysis of offspring BMI z score of mothers with pre-pregnancy type 1 diabetes and controls. Heterogeneity: χ 2 = 2.65, df = 2 (p = 0.27); I 2 = 25%. Test for overall effect: z = 2.12 (p = 0.03). IV, inverse variance; ref., reference

Full size image

Only one study reports a difference between mothers with pre-pregnancy type 2 diabetes and controls [50], so no meta-analysis can be performed for this subgroup. Assessor blinding is not specified in any included study, therefore a subgroup analysis by study quality cannot be performed.

Three studies adjust for covariates including maternal pre-pregnancy BMI [15, 27, 46]. A meta-analysis of the raw BMI z score data from these three studies shows increased BMI in the offspring of diabetic mothers (0.23 [95% CI 0.06, 0.40]; p = 0.009; I 2 = 20%, p = 0.29). Using the adjusted data, this effect is attenuated, with a pooled outcome of 0.07 (95% CI −0.15, 0.28; p = 0.54) (Fig. 5). The heterogeneity of this analysis is not significant (I 2 = 34%, p = 0.22).

Fig. 5
figure 5

Forest plot showing the adjusted association between all types of maternal diabetes and offspring BMI z score. Heterogeneity: χ 2 = 3.02, df = 2 (p = 0.22); I 2 = 24%. Test for overall effect: z = 0.61 (p = 0.54). IV, inverse variance; ref., reference

Full size image

Figure 6 shows the mean difference in mean maternal pre-pregnancy BMI plotted against the mean difference in offspring BMI z score for five studies presenting these data. Data from Gillman et al. [23] are included, although they did not report pre-pregnancy BMI, but used maternal BMI data at time of the study, approximately 10 years postpartum, on the basis that where available, this was concordant with pre-pregnancy BMI. This analysis suggests that as the difference between the BMI of the diabetic and non-diabetic groups reduces, so too does the difference in offspring BMI z score (Fig. 6). However, a random-effects meta-regression showed this relationship was not significant (increase in mean difference in offspring z score per unit increase in maternal BMI difference = 0.124 (95% CI −0.08, 0.33), p = 0.15) but this analysis had low power to detect an association as only five studies were included.

Fig. 6
figure 6

The relationship between the difference in mean maternal BMI reported between diabetic and non-diabetic mothers and the difference between the mean BMI z score of offspring of diabetic and non-diabetic mothers. Letters represent the following studies: A, Catalano et al. (2009) [29]; B, Krishnaveni et al. (2010) [22]; C, Gillman et al. (2003) [23]; D, Lawlor et al. (2009) [27]; E, Whitaker et al. (1998) [13]

Full size image

A funnel plot of the studies (Fig. 7) shows some asymmetry. Studies lying within the funnel are distributed symmetrically, but there are two studies [22, 50] lying to the right of the funnel, indicating potential publication bias towards reporting higher BMI z score in ODM. The Egger test shows no significant evidence for publication bias (p = 0.26) but as there are only included studies this cannot be considered conclusive.

Fig. 7
figure 7

Funnel plot of mean difference in offspring BMI z score from mothers with diabetes and controls against standard errors, with 95% confidence limits. Pooled mean difference = 0.280. Letters represent the following studies: A, Gillman et al. (2003) [23]; B, Whitaker et al. (1998) [13]; C, Lawlor et al. (2009) [27]; D, Wright et al. (2009) [46]; E, Lindsay et al. (2010) [15]; F, Krishnaveni et al. (2010) [22]; G, Manderson et al. (2002) [14]; H, Catalano et al. (2009) [29]; I, Hunter et al. (2004) [50]. Pooled mean difference = 0.280

Full size image

Discussion

This large, comprehensive meta-analysis identifies a strong association between exposure to maternal diabetes in utero and increased offspring BMI in childhood. In studies in which adjustment is made for maternal pre-pregnancy BMI, this relationship is no longer significant. Previous studies have been small with limited power and have provided contradictory results [13–15].

Studies which could not be included in this meta-analysis as they report overweight and obesity as outcomes rather than BMI z score, also show attenuation towards the null following adjustment for maternal pre-pregnancy BMI. Hummel et al. [51] reported odds ratios for overweight of 1.1 (0.7, 1.9) (unadjusted) and 0.8 (0.5, 1.4) (adjusted), Gillman et al. [23] 1.4 (1.0, 1.9) (unadjusted) and 1.2 (0.8, 1.7) (adjusted) and Whitaker et al. [13] 1.36 (p = 0.4) (unadjusted), and p = 0.53 after adjustment. We also show a correlation between the difference in mean BMI of diabetic and non-diabetic mothers and the difference in mean offspring BMI z score. These data might be interpreted as indicating that the relationship between maternal and offspring BMI in childhood is due to a common genetic potential and/or a shared postnatal obesogenic environment, rather than intrauterine programming secondary to maternal glycaemia of lesser degree than in overt diabetes (for which maternal BMI is a proxy). However, data from studies of Pima Indian siblings show that offspring born before the onset of maternal diabetes have lower childhood BMI than siblings born after the onset of maternal diabetes (with no difference relating to the onset of paternal diabetes) [10]. This suggests the long-term influence of a diabetic intrauterine environment, rather than the impact of postnatal or genetic factors. Similarly, altered offspring glucose tolerance can be induced in rats by intravenous glucose administration during the final days of gestation [7]. Deierlain et al. suggest that blood glucose concentration at 27 weeks gestation is a strong predictor of offspring overweight at 3 years of age, even when adjusted for maternal BMI [53]. Consequently, we suggest that caution should be exercised in rejecting the possibility of intrauterine programming. Adjustment for maternal pre-pregnancy BMI may also mask a potential synergistic relationship with maternal diabetes along a common pathway of hyperglycaemia during fetal development [27, 54]. This underlines the need for well-designed studies to examine the effects of maternal BMI and diabetes, and the interrelationship between them on offspring.

Human and animal studies show that maternal diabetes induces fetal hyperglycaemia, islet cell hypertrophy and beta cell hyperactivity, all of which may result in fetal hyperinsulinaemia and excess growth, particularly at insulin-sensitive sites such as adipose tissue [25, 55]. In keeping with this, fetal hyperinsulinaemia shows a positive correlation with childhood overweight [56], even in the absence of macrosomia [35, 57].

Rodent models of intrauterine exposure to hyperglycaemia result in hypothalamic malformations in regions that normally produce orexigenic neuropeptides, including neuropeptide Y, and are associated with hyperphagia and obesity in later life [8]. These data would support the suggestion that in humans, a causal pathway might involve hyperglycaemia-induced modification of hypothalamic appetite regulation, hyperphagia and leptin resistance in the offspring [49]. Maternal overweight/obesity is associated with maternal hyperglycaemia and insulin resistance [58, 59], even when the clinical criteria of GDM are not fulfilled. Pre-pregnancy overweight combined with GDM conveys a greater risk of offspring obesity than exposure to either GDM or pre-pregnancy overweight alone [54]. These data indicate that maternal obesity appears to exaggerate the metabolic abnormalities present in GDM [60], also suggesting that hyperglycaemia, irrespective of aetiology, is the determining mediator.

An additional finding from our meta-analysis is that in the studies in which adjusted data [15, 27, 46] were provided, the size of the observed difference in offspring BMI z score between cases and controls was inversely proportional to the number of covariates included in the adjustment (see Table 1). This would indicate that maternal pre-pregnancy BMI is not the sole confounder and that other genetic and postnatal environmental factors may also be contributory [23].

Subgroup analysis suggests that the effects on offspring BMI are not influenced by type of maternal diabetes. This is in agreement with a previous study showing no difference in offspring outcome between mothers with type 1 diabetes and GDM [61]. However, given the small number of studies and participants, conclusions drawn from the subgroup analysis must be viewed with caution. There were insufficient data to perform a meta-analysis with regard to offspring of mothers with pre-pregnancy type 2 diabetes.

We acknowledge several limitations to the extent to which conclusions can be drawn from meta-analyses of observational studies with variable treatment of confounders and variable definition of GDM. Some studies used the National Diabetes Data Group [62] or Carpenter and Coustan criteria [63] and others, the clinical diagnosis reported in the medical records. The case-definition for diabetes has a significant effect on reported population prevalence of GDM, varying, for example, between 3.17% and 4.48% according to the National Diabetes Data Group and Carpenter and Coustan criteria, respectively [64]. Inclusion of women with lesser degrees of glucose intolerance as diabetic would result in an underestimation of overall effect.

Treatment of diabetes during pregnancy (ultimately glycaemic control) may be an important determinant of offspring outcomes [65]. We reported data on the treatment received by the mothers included in this meta-analysis in Table 1, but insufficient information was available on glycaemic control for us to perform further meaningful analysis. However, treatment regimens were similar across the studies.

Accepted clinical management of maternal diabetes is directed at ameliorating short-term consequences by achieving good glucose control during pregnancy and treating immediate newborn complications. The short-term outcomes of diabetic pregnancies have improved substantially over the past decades [66], but the potential long-term consequences now pose serious threats to adult wellbeing and population health. In adults, the metabolic syndrome is increasing [67] and is a leading cause of mortality worldwide [68]. Similarly, in children, obesity and other features of the metabolic syndrome are rising rapidly and are strongly associated with premature death in later life [69].

We acknowledge that the number of included studies is small. Given the age range of the infants included, BMI z score is the only possible way of including them in this meta-analysis. Several studies report outcomes in the form of obesity, relative body mass, BMI or weight. We contacted the authors of these studies, but not all were able to provide their results as BMI z scores for inclusion in the meta-analysis. We would recommend that future studies on offspring of diabetic mothers during childhood should report their findings as BMI z scores, together with other measures of weight and body composition. The use of standardised units will greatly facilitate future meta-analyses in this area. It also seems logical to suggest that the difference in offspring BMI z score may increase with ageing, and a recent study suggests that this is the case [70]. However, given the overlap in age between the various reported cohorts, we felt that subgroup analyses attempting to address this possibility could be misleading.

Our meta-analysis of observational studies reveals an association between maternal diabetes and offspring BMI that is no longer significant after adjusting for maternal pre-pregnancy BMI. Thus, although diabetic control during pregnancy remains important, given the complex relationship between maternal BMI, maternal diabetes and offspring outcome, the importance of maintaining a healthy weight in women of childbearing age must also be emphasised.