Study population and design
The Cambridge Baby Growth Study (CBGS) has been recruiting newborns at the Rosie Maternity Hospital, Cambridge, UK for the study of pregnancy and postnatal determinants of early infancy growth and metabolism since 2001 . Between 2001 and 2009 mothers were approached during pregnancy and all underwent a formal 75 g OGTT at 28 weeks gestation as part of a research protocol. The fasting and 2 h glucose results were fed back to guide clinical management.
Between 2011 and 2013, in addition to the recruitment of women with uncomplicated pregnancies, we specifically recruited mothers identified as having GDM from specialist antenatal clinics who had undergone a 50 g glucose challenge test followed by a formal OGTT.
For the purpose of the current analyses, the same IADPSG criteria were retrospectively applied both to those OGTT collected as part of research between 2001 and 2009 and to those carried out as part of the clinical diagnostic procedures between 2011 and 2013 in order to reduce any bias in severity of GDM resulting from changing diagnostic criteria over this period.
The retrospective application of the IADPSG criteria has implications for treatment as 19% of the ‘earlier’ OGDM group were not diagnosed and did not receive any treatment.
To select mothers with GDM for study of both cohorts (‘earlier’ and ‘recent’), IADPSG criteria  (at least one glucose concentration on a 75 g OGTT at around 28 weeks gestation: >5.1 mmol/l at 0 min, >10.0 mmol/l at 60 min, >8.5 mmol/l at 120 min) were applied retrospectively to the earlier cohort and prospectively in the recent OGDM group, allowing comparable earlier and recent OGDM subgroups. In all subgroups, the following additional criteria were met: singleton pregnancy; no significant maternal comorbidity (such as pre-eclampsia, hypertension, antiphospholipid syndrome, ankylosing spondylitis, lupus or ulcerative colitis); gestational age ≥36 weeks. Cases of maternal type 1 or type 2 diabetes and infants with a genetic or syndromal disease were excluded.
The earlier OGDM (N = 98) were born between 2001 and 2009 and the recent OGDM (N = 122) were born between 2011 and 2013.
In the earlier OGDM population, a 75 g OGTT at 28 weeks gestation was performed as part of the research protocol and, for the purpose of the current study, the IADPSG criteria were applied retrospectively. In contrast, the clinical decision to treat these women in the earlier GDM cohort was broadly based on the WHO 1999 guideline , which considers fasting and 2 h glucose values (but not 1 h glucose). Based on available records and information from treating clinicians, GDM was mostly treated with diet and lifestyle modification, with or without insulin; metformin was not routinely used at that time.
In the recent OGDM population, women were recruited from the antenatal GDM clinic following routine practice (a 75 g OGTT in high-risk women and those identified through a universal 50 g glucose challenge at 24–26 weeks as a necessary prerequisite for the woman to undergo glucose tolerance testing). All women received standardised dietary and lifestyle advice, and were seen in clinic regularly (on average every 2 weeks). Additionally, metformin and/or insulin were prescribed as required, guided by regular fasting and postprandial glucose monitoring.
A CBGS control population (N = 876) was comprised of mother–infant dyads with normal blood glucose levels on OGTT at 28 weeks using the IADPSG criteria between 2001 and 2009 (and those recruited later who had normal 50 g glucose challenge tests). They were studied using the identical postnatal research protocol.
Studies were approved by the Cambridge local research ethics committee, and all mothers gave informed written consent.
Birth measurements and infancy anthropometry
Infants’ birthweights were obtained from hospital records. Newborn (within first 8 days) length and skinfold thickness, and subsequent measurements at 3, 12 and 24 months of age were performed by three trained paediatric research nurses, using identical protocols for all cohorts. Weight was measured to the nearest 1 g using a Seca 757 electronic baby scale (Seca, Birmingham, UK). Length was measured to the nearest 0.1 cm using an Infantometer (Seca 416). Skinfold thickness was measured in triplicate at four sites (triceps, subscapular, flank, quadriceps) on the left side of the body using a Holtain Tanner/Whitehouse Skinfold Caliper (Holtain, Crymych, UK).
Infancy age- and sex-appropriate SD scores (SDS) were calculated for weight and length measurements (with adjustment for gestational age at birth and 3 months), by comparison with the UK 1990 growth reference using LMS growth software . For each of the four skinfold thickness measurements an internal SDS was calculated, using residuals from a linear regression model, adjusting for infancy age, (gestational age at birth and 3 months) and sex. Mean skinfold thickness SDS was used in analyses. Maternal BMI was derived from self-reported pre-pregnancy weight divided by the square of measured height (kg/m2). Birth ponderal index was calculated by dividing the infant’s birthweight by its birth length cubed (kg/m3). Deprivation was assessed using an integrated index based on residential postcodes .
Maternal and birth characteristics were compared between groups using ANOVA with Bonferroni post hoc analysis for continuous variables, and χ2 tests for categorical outcomes. Unless otherwise stated, all data are presented as means ± SDs.
Multiple linear regression was used to investigate the effect of GDM on birth outcomes, allowing adjustment for potential confounders, including infant sex, postnatal age, gestational age, pre-pregnancy maternal BMI, maternal height, parity, breastfeeding history at 3 months, delivery method, maternal ethnicity, socioeconomic status reflected by Index of Multiple Deprivation (IMD) and pregnancy smoking history. All confounders were chosen a priori through the extensive work of CBGS and the Avon Longitudinal Study of Parents and Children (ALSPAC) .
Under the traditional listwise deletion method, only 68% of the control group and 64% of both recent and earlier OGDM had complete data on all covariates. Covariates with most missing values were maternal pre-pregnancy BMI for control infants and ‘earlier OGDM’, and smoking history for recent OGDM. Data were primarily missing due to incomplete perinatal questionnaire responses. Missing covariates including IMD (n = 3), parity (n = 4), maternal ethnicity (n = 8), smoking history during pregnancy (n = 39), maternal pre-pregnancy BMI (n = 185), maternal height (n = 148), delivery method (n = 27) and infant feeding history (n = 189) were imputed under the assumption that they are missing at random. The R package ‘Multiple Imputations via Chained Equations (MICE)’ was used to generate 20 imputed datasets, using normal linear regression for continuous variables and logistic linear regression for binary variables. Analyses run on each dataset were pooled according to Rubin’s rules . Imputed values compared reasonably to observed values, and the results (i.e. linear regression model on birth data, Table 2) using listwise deletion were similar to imputed values, therefore imputed values were presented in the subsequent analyses.
In the visit measurements, missing data were commonly due to loss-to-follow-up or drop outs. In order to capitalise the longitudinal growth data with good handling of missing values, linear mixed-effects models were used to relate the continuous growth outcome variables (weight, height and skinfold thickness) to visit time point, cohort group, and their interaction with infant age, taking into account the same confounders as in the linear regression models for birth measurements. Due to non-linear relationships with age (indicated by significant estimates for age-squared), time was modelled using linear splines with knots at ages 3 and 12 months. Models were fitted to the data by restricted maximum likelihood (REML).
Statistical analyses were conducted using SPSS (IBM SPSS Statistics for Windows, version 25.0; IBM, Armonk, New York, USA) and R (version 3.3.2; R Foundation for Statistical Computing, Vienna, Austria). p < 0.05 was considered statistically significant.