Impact of maternal HbA_1c on offspring glucose at 4–7 years of age: role of childhood adiposity and other potential confounders.

To the Editor:

We appreciate the thoughtful comments by Periyathambi et al [1] regarding our paper [2] and are delighted at the interest and attention to detail with which they read this work. Indeed, with the rising prevalence and earlier age at onset of type 2 diabetes worldwide, there is growing interest in understanding the aetiological pathways underlying youth-onset type 2 diabetes, including the role of an obesogenic in utero environment and offspring adiposity, either at a specific point in time or cumulatively over time. Our analysis sought to test the hypothesis that exposure to hyperglycaemia in utero (assessed via maternal HbA_1c at a median 27 weeks of gestation) is associated with biomarkers of glucose–insulin homeostasis in offspring as early as 4–7 years of age, and that this relationship is mediated by cumulative adiposity from birth through to early childhood [2]. Periyathambi et al made several suggestions to complement or supplement what we have done [1], some of which we have considered and report in this response and others of which are outside of the scope of this paper.

Periyathambi et al suggested that we report the glucose and HbA_1c levels among women who were excluded vs those who were included in the analysis in our study in order to assess for possible selection bias in our analysis sample [1]. Thus, we compared maternal glucose and HbA_1c values between women excluded and included in our analysis using a Wilcoxon rank sum test for non-parametric data and found no statistical difference. Among those excluded vs included the median (min–max) level of glucose at 17 gestational weeks was 4.2 mmol/l (2.6–6.1) vs 4.2 mmol/l (2.9–5.8) (p=0.07) and at 27 gestational weeks was 4.3 mmol/l (3.06–8.61) vs 4.2 mmol/l (3.3–7.1) (p=0.30), whilst HbA_1c was 31.2 mmol/mol (19.1–42.1) (5.0% [3.9%–6.0%]) vs 31.2 mmol/mol (16.9–45.4) (5.0% [3.7%–6.3%]) (p=0.54).

The authors had questions regarding our metric of cumulative adiposity (sum of per cent fat mass [%FM] across the two time points) and whether we assessed the effect of rate of adiposity change from offspring birth to 4–7 years of age as a metric of cumulative adiposity [1]. In this study sample, we found no evidence that rate of change in %FM ([%FM at 4–7 years − %FM at birth]/age at second fat mass assessment) played a mediating role that differed from that of cumulative %FM in a model adjusted for maternal race/ethnicity, child’s sex and age at assessment. For example, when comparing the total effect of the third tertile vs first tertile of HbA_1c in relation to offspring glucose (0.17 mmol/l [95% CI 0.08, 0.26]) [2], we observed minimal differences between the total effect estimate and the natural direct effect after including %FM change (0.17 mmol/l [0.07, 0.26]; 0% change). This finding was similar to the natural direct effect after inclusion of cumulative %FM (0.17 mmol/l [0.08, 0.26]; 0% change) as a mediator [2]. Periyathambi and colleagues were also interested in the effect of fat free mass (FFM) as a mediator of maternal HbA_1c and glucose–insulin homeostasis in offspring [1]. In our published study [2], we did not assess FFM as a mediator of these for two reasons: (1) our a priori hypothesis was about offspring adiposity; and (2) estimated FFM is an imperfect proxy for muscle mass and would represent a complementary but distinct mechanism involving glucose uptake into skeletal muscle, whereas fat mass influences glucose–insulin homeostasis via an effect on whole-body insulin resistance and insulin sensitivity [3]. However, in response to the interest shown by Periyathambi et al in FFM as a mediator, we conducted this additional analysis using cumulative %FFM (sum of %FFM across the two time points). Similar to our finding using cumulative %FM as a mediator, cumulative %FFM did not mediate the associations of interest (data not shown).

Periyathambi and co-authors also expressed an interest in several postnatal lifestyle factors that have previously been associated with offspring insulin resistance, such as the rate and duration of breastfeeding, offspring consumption of sugar sweetened beverages and offspring physical activity [1]. We do not believe that accounting for these postnatal factors is appropriate based on our hypothesis and research aim as they are mediators in and of themselves rather than confounders. Although these postnatal mediators are related to the outcomes in the current study, they likely share overlapping pathways with adiposity mediators. Thus, inclusion of these post-exposure variables might block the mediating path of interest (e.g., adiposity) or, in a worse scenario, induce collider stratification bias or other forms of over-adjustment bias [4, 5].

It was also suggested that we adjust for measured early pregnancy BMI rather than self-reported pre-pregnancy BMI as a covariate in our models [1]. Adjustment for early pregnancy BMI (HbA_1c tertile 3 vs tertile 1, β=0.16 [95% CI 0.06, 0.25] for glucose [mmol/l]) yielded the same result as when our models were adjusted for pre-pregnancy BMI (HbA_1c tertile 3 vs tertile 1, β=0.16 [95% CI 0.06, 0.25] for glucose [mmol/l]), which is expected given that these two variables are highly correlated [6] (Pearson’s correlation coefficient = 0.99 in the study).

Finally, we would like to thank Periyathambi et al for pointing out the error in our conversion of HbA_1c values from per cent to mmol/mol, which has been corrected by an erratum [7]. Following the amendments to the HbA_1c values in mmol/mol, the direction of the effects and p values reported were unchanged and the overall findings of the study were unaltered.

There are certainly some limits to the study that were noted in our original article [2]. However, this study provided insight into the potential in utero programming of offspring glucose–insulin homeostasis by maternal HbA_1c, with our data indicating that this particular pathway does not operate through offspring adiposity at birth or in early childhood. We encourage future mechanistic studies to uncover the alternate pathways by which maternal blood glucose has an impact on child glucose–insulin homeostasis.

Abbreviations

FFM

Fat free mass

%FM

Per cent fat mass