Study design and participants

Data were collected prospectively as part of the ENDIA study, a national Australian longitudinal prospective pregnancy/birth cohort study with the overall aim to determine the early-life exposures that drive the development of type 1 diabetes [9]. In this study, all women were investigated according to the ENDIA protocol at 3 month intervals during pregnancy from the time of recruitment until birth, and their children in the neonatal period. Investigation included clinical measurements, clinical outcomes and questionnaires assessing nutrition, exercise and lifestyle.

ENDIA recruited 1488 pregnant women or those with a child less than 6 months of age between February 2013 and November 2019, where the child had a first-degree relative with type 1 diabetes. Women were excluded from the study if they had an inadequate understanding of English to provide consent and responses to questionnaires. This analysis included all women who completed at least one diet or physical activity questionnaire during pregnancy; women with twin or triplet pregnancies were excluded.

The ENDIA study was reviewed and approved by the study’s lead Human Research Ethics Committee at the Women’s and Children’s Health Network under the National Mutual Acceptance Scheme (current approval no. 2020/HRE01400) and at all participating study sites. Conduct in Western Australia was approved by the Women and Newborn Health Service Ethics Committee (ref. no. RGS0000002639). ENDIA is registered on the Australian New Zealand Clinical Trials Registry (ACTRN12613000794707). All women provided written informed consent and were free to withdraw from the study at any time.

Demographic and clinical data

Pre-pregnancy weight and maternal demographics were self-reported by participants at their first appointment. Pre-pregnancy BMI (kg/m2) was calculated using pre-pregnancy reported weight and height measured at the first visit. Maternal weight was measured at each visit and additional weight measurements collected at routine clinic visits were obtained from medical records. If pre-pregnancy weight was missing, the earliest weight available in the first trimester was used to estimate pre-pregnancy BMI. Gestational weight gain was calculated as the difference between the last weight measured during the third trimester of pregnancy and the pre-pregnancy weight or first weight recorded during pregnancy. Pregnancy data (including parity, HbA1c, medications and medical complications) and birth outcomes (gestational age at birth, birthweight and neonatal hypoglycaemia) were obtained from hospital medical records. Socioeconomic status was calculated using postcode at enrolment using the Socio-Economic Indexes for Areas Index of Relative Socio-Economic Advantage and Disadvantage (SEIFA IRSAD). This index is derived from national Census variables related to both advantage and disadvantage, for example household income and level of education [10]. Remoteness was classified using the Modified Monash Model [11], which defines a location according to geographical remoteness, as defined by the Australian Bureau of Statistics, and town size.

HDP were verified using medical record review according to International Society for the Study of Hypertension in Pregnancy classification [12] and included gestational hypertension (new onset of hypertension after 20 weeks’ gestation), pre-eclampsia (hypertension occurring for the first time after 20 weeks, associated with proteinuria and/or organ involvement), superimposed pre-eclampsia (pre-eclampsia superimposed on chronic hypertension), eclampsia and HELLP (haemolysis, elevated liver enzymes and low platelets) syndrome. Prematurity was defined as birth before 37 weeks’ gestation. Neonatal hypoglycaemia was defined as blood glucose level in the newborn of less than 2.6 mmol/l within the first 72 h post birth.

HbA1c was measured using either point-of-care or laboratory testing methods, commonly a Vantage analyser (Siemens Diagnostics, Camberley, UK) or a Variant analyser (Bio-Rad Laboratories, Hercules, CA, USA). All medical laboratories were accredited by the National Association of Testing Authorities, Australia, against the international standard ISO 15189 Medical laboratories, which mandates that all analytes in a laboratory’s test menu be subject to the Royal College of Pathologists of Australasia Quality Assurance Programs [13]. The first HbA1c measurement available during pregnancy (usually conception or first trimester) was used and analysis controlled for gestational week of measurement.

Dietary and lifestyle measures

Diet was assessed using the Dietary Questionnaire for Epidemiological Studies version 2 (DQESv2), a self-administered 74-item food frequency questionnaire [14] validated in women of child-bearing age (16–48 years) [15]. Participants completed the questionnaire during their third trimester of pregnancy and were asked to assess their diet since the start of pregnancy. The DQESv2 provided daily intakes (in grams) of specific foods and beverages. The 101 individual food items were combined into 19 food item categories based on nutrient content and culinary usage [16] (electronic supplementary material [ESM] Table 1) for the analysis of dietary patterns. Consumption of each food item was converted into daily servings by adjusting the intake for serving size as described in the Australian Dietary Guidelines [17]. The total number of servings per day were calculated by summing the numbers of servings consumed per day for all food items in each of the five food groups of the Australian Dietary Guidelines.

Physical activity was measured during each trimester using the Pregnancy Physical Activity Questionnaire (PPAQ) [18], a validated self-report questionnaire that measures the time spent participating in 32 activities grouped into different types of activity (i.e. sedentary, light, moderate and vigorous activity). Participants could add two physical activities not listed in the questionnaire, where the intensities were individually estimated using the Compendium of Physical Activities [19]. The duration of time spent in each activity was multiplied by its intensity (i.e. metabolic equivalent of task [MET]) and summed to calculate the mean weekly energy expenditure, expressed as MET-hours/week.

Statistical analysis

To account for the potential correlation between data from the same participant during different pregnancies (i.e. a participant included in the study more than once), a random intercept for each participant was included in each model. Analyses were restricted to participants with complete data as missing data were minimal for outcomes and confounders used in analyses (as reported in tables). R statistical software version 4.3.1 [20] and a significance level of 5% was used for all analyses. Results for adjusted models are reported unless otherwise specified.

Dietary patterns for pregnancies with and without type 1 diabetes

Dietary patterns were derived using principal component analysis (PCA) on dietary data based on 19 food item categories from all women with and without type 1 diabetes. The resulting principal components, derived in decreasing order of importance, were a linear combination of the food items. The number of dietary patterns (principal components) identified was based on eigenvalues >1.5 and on identification of a break point in the scree plot [21]. Food item categories with a factor loading of ±0.30 or more were considered important contributors of each dietary pattern [22]. Scores for each principal component were obtained by summing up observed intakes of the component food items weighted by the factor loading and indicate the extent to which the participant’s diet conformed to the respective dietary pattern. A logistic regression mixed model was used to compare dietary patterns in women with and without type 1 diabetes using the principal component scores from each dietary pattern. Potential confounders (age, pre-pregnancy BMI and parity) were included in the model.

Association between dietary patterns and physical activity, and maternal complications and birth outcomes in pregnancies with type 1 diabetes

Analysis of associations between dietary patterns and outcomes were planned only for the women with type 1 diabetes, as maternal complications and adverse birth outcomes are about fivefold more prevalent in women with type 1 diabetes in Australia. Maternal outcomes were HDP (categorised as gestational hypertension or pre-eclampsia/eclampsia/HELLP), and birth outcomes were prematurity, gestational age at birth, birthweight and neonatal hypoglycaemia. Exposures were dietary patterns (participant principal components scores) and physical activity (mean MET-hours/week for total activity, sedentary activity, and moderate and vigorous activity). Models that investigated associations with physical activity included nested random intercept terms, visits within pregnancies within the same participant (to account for the fact that each mother could have completed up to three questionnaires for each pregnancy), and were adjusted for the gestational age when the questionnaire was completed.

Prematurity (<37 weeks’ gestation) and neonatal hypoglycaemia (blood glucose level <2.6 mmol/l) were fitted in separate mixed logistic regression models. A separate mixed multinomial logistic regression model was used for the HDP categories. For continuous outcomes—HbA1c, gestational age at birth and birthweight—separate linear mixed models were fitted. Potential confounders (maternal age, parity and SEIFA IRSAD percentile) were adjusted for in all models. Birthweight was also adjusted for gestation at birth. To make the results more interpretable, participant scores identified from PCA were rescaled such that a one-unit change in principal component score represented the IQR from 25th percentile to 75th percentile. The mean intake of food group servings was calculated from participants in each quartile of principal component scores for the ‘fresh food’ dietary pattern. The difference in intake between the highest and lowest quartile was calculated to correspond to a one-unit change of the rescaled scores.

Mediation analyses

Potential mediators of the association between dietary pattern and pre-eclampsia and premature birth were HbA1c and BMI. Model-based causal mediation analysis with quasi-Bayesian Monte Carlo simulation (10,000 simulation) was performed [23] using the ‘mediation’ R package. While pre-pregnancy BMI is not strictly a mediator temporally between diet and pre-eclampsia (i.e. it was measured before the time covered by the diet questionnaire), it is an available proxy for early pregnancy BMI before weight increases. Therefore, the influence of pre-pregnancy BMI was also investigated as a mediator between diet and pre-eclampsia using model-based causal mediation analysis.

Sensitivity analyses

Sensitivity analyses were conducted excluding participants from the PCA if they reported an unrealistic energy intake (energy <4500 kJ/day or >20,000 kJ/day, n=109, including 82 with type 1 diabetes) [24] and if they had gestational diabetes (n=51) or type 2 diabetes (n=1). All models were then refitted with the resulting principal components scores. Separate sensitivity analyses were also conducted for the outcomes of pre-eclampsia and prematurity which excluded women with a parity greater than 0 (i.e. included only nulliparous women) as the risk of pre-eclampsia and premature birth are substantially influenced by parity and complications in a prior pregnancy.