Introduction
The relationship between age at menarche (AAM) and the subsequent risk of developing gestational diabetes (GDM) in pregnancy is a potentially important one for preventive medicine as there has been a generalised global lowering of AAM in the last century [
1] combined with an increasing prevalence of GDM [
2]. This increasing prevalence of GDM is thought to be one of the factors fuelling the current and future predicted worldwide diabetes epidemic [
3]. The strong links between GDM development in pregnancy and the future development of type 2 diabetes (women who experienced GDM having a greater than seven times higher risk of developing type 2 diabetes than those with normoglycaemic pregnancies [
4]), and the fact that in utero exposure to GDM increases the number of GDM risk factors female babies have when they get pregnant as adults [
5,
6], means that being able to predict those women most at risk of GDM when they become pregnant may be important in targeting lifestyle interventions. Treating susceptible women earlier may become possible [
7], which could help reduce the incidence of associated complications.
Partially due to shared genetic risk factors [
8], it has been suggested that GDM represents an early manifestation of type 2 diabetes [
9], with the physiological insulin resistance of pregnancy causing the premature expression of the disease. Risk of type 2 diabetes has already been shown to be associated with AAM [
10], including by use of a Mendelian randomisation approach [
11]. As long ago as 1975 it was suggested that AAM may be linked to the development of GDM [
12], although it was not formally tested. More recently, various studies have investigated links between AAM and the risk of developing GDM in pregnancy and whilst some significant associations have been found [
13‐
17] (with maximal relative risks for GDM, associated with the earliest AAMs relative to those at the median, ranging from 1.3 to 3.7), this is not true of all studies [
18]. Even where a relationship has been observed, sometimes the significantly higher relative risk of GDM is just associated with earliest AAM [
13] whereas in other studies a more curvilinear relationship between AAM and GDM is evident (even if it were not always formally tested for or statistical significance reached [
14‐
18]), albeit smaller in magnitude than that associated with the earliest AAMs. This systematic review and meta-analysis was therefore designed to clarify the relationship between AAM and the risk of developing GDM in pregnancy, both in terms of its association and the shape of that association.
Discussion
This meta-analysis of five prospective studies [
13‐
15,
17,
18] has shown a significant, non-linear relationship between the AAM and the relative risk of developing GDM in pregnancy. Despite the fact that only five studies were included in the main analysis, the relationship appears to be reasonably strong. During the writing up of our study two other meta-analyses of the relationship between AAM and future risk of GDM were published [
31,
32]. There are a number of differences between these published meta-analyses and the present meta-analysis, however, including: the inclusion of one study [
16] in the main analysis of the previously published meta-analyses that we only incorporated into our study’s sensitivity analyses and not the main analyses due to the two published meta-analyses having slightly different inclusion criteria [
31,
32], these previously published meta-analyses did not include data from our own recently published study [
17] and the form of meta-analysis used in both analyses was not designed for dose response curves, so the comparison used was effectively between GDM risk in women with “early” and “not-early” AAM rather than taking a specific age as a reference age. Despite these differences, the relative risk or odds ratios for GDM in women with an early AAM relative to that of reference groups were remarkably similar between the three studies, the slightly lower values in [
31,
32] possibly relating to those studies including the AAM categories that contributed to the non-linear relationship observed in the present analysis in their reference groups.
As far as a small meta-analysis such as this present analysis can be interpreted, there appears to be the possibility of publication bias having affected the relative risk. Egger’s test, which would have had low statistical power for testing funnel plot heterogeneity even if it was based on simple case–control studies rather than dose response curves [
33], had to be adapted as it was not designed for data derived from dose response curves. We did this by pooling some of the AAM categories. Although the test may therefore have been run sub-optimally, it still gave a significant result. This result was consistent with the apparent asymmetry in the funnel plot (Fig.
4). It therefore seems fair to assume that there might have been an excess of studies with positive associations that have been published to date. Techniques such as Duval and Tweedie’s trim and fill [
34] that are used in other meta-analyses in an attempt to overcome potential publication bias, are unsuitable for data derived from dose response curves. As an alternative which was suitable for use with data such as ours, Borenstein et al. [
27] suggested reanalysing just the largest study, then adding the next largest study and reanalysing and continuing to do this whilst gradually adding successively smaller studies until all the studies are included. This method makes the assumption that larger studies are more likely to get published whether or not they produce significant associations (whereas smaller studies may only get studied if they show significant associations), and therefore the publication of larger studies is less dependent on bias. We demonstrated a positive increase in maximal relative risks for GDM as studies of decreasing size were added to the analysis (Fig.
5), adding the smallest study [
17] causing the biggest increase in relative risks. All these meta-analyses showed an association with GDM, it was just the relative risk that changed. The true maximal relative risk for GDM may therefore be closer to 1.6 (the lower limit of the 95% confidence interval) than 2 for an AAM of 9 years, but the overall interpretation of this association is the same. The impact of any publication bias is therefore probably modest at best [
27].
Not surprisingly from the fact that only five studies were included in the meta-analysis there was no statistically significant heterogeneity. Indeed the relevant I
2 value of 25.5% suggests relatively low heterogeneity overall. This is despite the five studies used having some ethnic differences in their study populations and different cut-off circulating glucose concentrations used to define what constitutes GDM, both of which could have been associated with factors that contributed to increasing the heterogeneity. Interestingly when the initially excluded sixth study [
16] was included in the meta-analysis for the sensitivity analysis the heterogeneity actually dropped. The results from the other sensitivity analyses, performed by excluding each one of the five studies from the meta-analysis in turn, also showed that despite the variation in the heterogeneity caused by each of the studies the material relationship between AAM and risk of developing of GDM in pregnancy did not change.
Although the relationship between AAM and the development of GDM in pregnancy that we found is non-linear, the meta-analysis has clarified the fact that in the studies published so far all the modelled significant increased risk for GDM is associated with earlier AAM. Early AAM tends to be associated with early closure of the epiphyseal plates and a relatively shorter stature in adult life [
35]. It is subsequently associated with increased weight gain (relative to stature) and obesity [
36], the strongest risk factor for GDM. Hence there is a link between early AAM and future GDM risk in pregnancy. In our original study [
17] we also found a link between AAM and insulin resistance in pregnancy, and adjusting for insulin resistance completely attenuated the association between AAM and GDM. These results suggest that insulin resistance may be the primary factor underpinning the association between early AAM and increased GDM risk. Some studies have shown this association to be a linear inverse relationship [
13], whereas others have shown more of a curvilinear relationship [
16,
17] with additional increased relative risk, albeit smaller, for GDM being associated with older AAM. It could be that the bulk of the heterogeneity in the meta-analysis is at this end of the dose response curve, given that it dropped to an I
2 of only 1.3% when the largest study was excluded, that study being one showing a linear rather than curvilinear relationship between AAM and GDM risk [
13]. Although our meta-analysis showed a non-linear relationship, at present there is not enough evidence to suggest that there is a consistent association between older AAM and a raised future risk of GDM in pregnancy.
The strengths of this study include the fact that all the studies contributing to the meta-analysis were conducted prospectively in terms of GDM development and so should have potentially reduced recall and selection biases in comparison to if the studies were conducted retrospectively or in a case–control fashion. Another strength is the fact that the type of meta-analysis performed was specifically designed to test relationships in a dose response fashion [
21], unlike the other meta-analyses published in this area [
31,
32], and provide information about the shape of that relationship. Finally this meta-analysis covers more than 3,000 pregnancies affected by GDM, which is a very large number given its prevalence in the population as a whole, along with more than 55,000 unaffected pregnant women.
In addition to potential publication bias the limitations of this meta-analysis include the fact that it was based on data from only five different studies. There was only one published study presenting data relating AAM with future risk of GDM [
16] that was not included in the final meta-analysis. That was primarily because this manuscript did not present unadjusted risk ratios or numbers of women with and without GDM from which they could be calculated. However it was included in the sensitivity analyses, where its presence did not materially change the interpretation of the relationship between AAM and risk of developing GDM in pregnancy suggesting that the results from the meta-analysis are sensitive. Another limitation is that no attempt was made to adjust the results for potential covariates. However the main potential covariate for GDM risk was maternal BMI, and it is debatable that this should have been controlled for given that in addition to the association with GDM risk, AAM also has a relationship with adult BMI [
36]. This may actually reflect a single multifactorial process that includes AAM and starts even before birth [
37] then includes effects on both adult BMI, and GDM risk in pregnancy as the woman ages. Another potential covariate that was not adjusted for was family history of diabetes. Only one of the studies showing an association between AAM and GDM risk presented data on, and a positive association between, family history of diabetes and GDM [
16]. This was the study that was excluded from the meta-analysis for the reasons presented. One further study adjusted for it as a confounder in some of their statistical models [
13], but did not present data on it. It remains a possibility, therefore, that a family history of diabetes may modify the relationship between AAM and GDM. A further limitation of this area of research, with the studies that have been published so far rather than with this meta-analysis, is the fact that the relationship between AAM and GDM risk in pregnancy has not been tested in sufficient numbers of women of non-white ethnicities. Future studies need to be performed in a wider range of ethnicities, particularly those that have the highest incidences of GDM.
In summary, this meta-analysis has shown that there is a significant relationship between AAM and the risk of GDM developing in subsequent pregnancy that is non-linear. Although there is evidence of possible publication bias in the results that have been published in this area to date, the effect of this appears to be an inflation of the relative risk from around 1.6 to closer to 2 rather than being anything that would alter the overall interpretation of the results. The risk appears to be around a younger AAM being associated with an increased risk of GDM. Further studies are required to clarify whether or not there is also an increased risk of GDM associated with having a relatively late AAM.