Environmental Determinants of Islet Autoimmunity (ENDIA) longitudinal prospective pregnancy to childhood cohort study of Australian children at risk of type 1 diabetes: parental demographics and birth information

Study design and protocol amendments

A detailed study protocol was published previously.24 In brief, participants were longitudinally followed each trimester of pregnancy, at birth, every 3 months to 2 years and every 6 months thereafter until 10 years (online supplemental figure 1). Data on demographic, lifestyle, clinical and anthropometric measurements, maternal and child dietary intake and biological samples are collected at multiple time points, throughout follow-up. Several amendments to the original study protocol published in 2013 have been made (summarized in online supplemental table 1).

The primary outcome was persistent islet autoimmunity defined as antibodies to one or more islet autoantigens (insulin, glutamic acid decarboxylase 65, tyrosine phosphatase-like insulinoma antigen and zinc transporter 8) detected on two consecutive tests, taken at least 3 months apart. The secondary outcome was detection of persistent coeliac autoimmunity defined as antibodies to tissue transglutaminase and/or deamidated gliadin peptide, for which children with genetic susceptibility to T1D are at elevated risk.29

Study population and sites

The inclusion criteria for ENDIA were an unborn child, or infant aged<6 months, with an FDR diagnosed with T1D. The exclusion criterion was the incapacity of the pregnant person or child’s primary caregiver to understand the requirements of their own and/or child’s participation in the study.

Participants were recruited from February 2013 to December 2019. Families could enroll multiple children if conceived during the recruitment period. Postnatal participants were recruited from February 2013 to May 2019. Participants were recruited nationwide in Australia using a variety of methods.30 Lead recruitment sites were: The Women’s and Children’s Hospital, Adelaide, South Australia (190 participants); The Royal Melbourne Hospital (286 participants) and Monash Medical Centre (44 participants), Melbourne, Victoria; Barwon Health, Geelong, Victoria (66 participants); The Children’s Hospital at Westmead (172 participants), Randwick Hospital for Women (112 participants) and St George Hospital (44 participants), Sydney, New South Wales; Mater Hospital, Brisbane, Queensland (246 participants); and Princess Margaret Hospital/Perth Children’s Hospital, Perth, Western Australia (209 participants). Participants outside of these cities enrolled through a regional participant program from 2016 that was administered through Adelaide’s Women’s and Children’s Hospital (142 participants).31

Clinical data collection and definitions

Detailed demographic information was collected at enrollment. Pregnancy data pertaining to health, nutrition, exercise and lifestyle were collected at each study visit during pregnancy and birth data were collected shortly after birth, or at the time of enrollment for postnatal recruits. Clinical pregnancy and birth data were verified against medical case notes. Parental preconception data were collected retrospectively. Validated data collection tools and custom questionnaires used during pregnancy and infancy have been described previously24 or outlined in online supplemental table 1.

Body mass index was calculated using self-report pre pregnancy or measured early pregnancy weight. Gestational weight gain was calculated from the pre pregnancy or early first trimester weight and the latest weight recorded during pregnancy. Mothers who underwent cesarean sections without experiencing prior labor are categorized as ‘no labor’. Large for gestational age (LGA) was defined as ≥90th centile and small for gestational age was defined as ≤10th centile from population-based Australian birth weight centile charts.32 The Australian Statistical Geography Standard Remoteness Structure was used to categorize participants into classes of remoteness based on a measure of relative distance to access services.33 Participants were classified as living in a major city, regional (inner and outer regional) or remote (remote or very remote) area based on the postcode of their residential address at the time of study enrollment. This postcode was also used to assign an area-based measure of socioeconomic status (SES) using the Index of Relative Socioeconomic Disadvantage which is constructed from variables such as the proportion of individuals in an area who are on a low income, are of Indigenous descent and are unemployed or in relatively low skill occupations.34

Biological data collections and analysis

Blood, urine, stool, saliva, breastmilk, swabs from various body sites and a deciduous tooth were collected throughout follow-up as described24 or outlined in online supplemental table 1. Saliva samples for genetic studies were collected from the ENDIA child and all available FDRs. The OG-250 Oragene DNA disc kit was used for infants and OG-500 Oragene DNA tubes for adults and older children (DNA Genotek, Ontario, Canada). For 8% of participants (n=344) where a saliva sample was not available, DNA was extracted from other biological specimens. QIAamp Fast DNA Stool Mini kits (Qiagen, Cat No: 51604, Germany) were used to extract DNA from stool samples while ISOLATE II Genomic DNA kits (Biolone Meridian Bioscience, Cat No: BIO-52066, USA) were used for other sample types (ie, blood, swabs and milk). HLA DR typing was performed on DNA by TaqMan-based PCR typing and imputation from three single-nucleotide polymorphisms (rs3104413, rs2187668 and rs9275495), as described previously.35

Statistical analysis

Data are provided as mean (SD) or median (quartiles 1–3) for continuous variables and number (proportion of that category) for categorical values. For descriptive comparison to the Australian population, data were obtained for all babies born in Australia in 2017 as this was the median year of birth in ENDIA.36–38 Data for the comparison of education and male smoking levels were obtained for reproductive aged adults39 40 and data for breastfeeding initiation were from 2010.41

For each continuous response, to identify differences in mothers with and without T1D, an adjusted linear-mixed model was fitted. Where underlying model assumptions were not met, the response variable was transformed. For binary, categorical or discrete (count) responses, a generalized linear-mixed model (logistic, multinomial or Poisson regression, respectively) was fitted. In these approaches, a fixed factor for T1D status and a random effect for mother was included in the model. The random mother effect induced a correlation between observations from different pregnancies of the same mother. For maternal and infant outcomes, adjustment was made for maternal age at birth and for parity. For analyses of infant birth weight, admission to neonatal intensive care unit (NICU), total days in NICU and admission to special care nursery (SCN), adjustments were made for gestational age at birth. A p value of 0.05 was used for determining statistical significance for all models. Data analysis was performed in R V.4.3.0.42

Description of study population

The ENDIA Study completed recruitment of 1511 participants in December 2019 and the last baby was born in July 2020. 55% of participants who expressed interest provided informed consent and were enrolled in the study (figure 1). 38 participants were excluded due to adverse pregnancy and birth outcomes (n=26), loss to follow-up (n=5), withdrawal prior to study commencement (n=4) or ineligibility (n=3). The 26 adverse outcomes included 17 miscarriages (13 from pregnancies with T1D), 7 stillbirths (6 from pregnancies with T1D) and 2 neonatal deaths (twins from a pregnancy without T1D).

Request permission

The final cohort comprised 1473 infants born to 1214 gestational mothers across 1453 pregnancies. Of the 1214 gestational mothers, 79% enrolled 1 pregnancy, 19% enrolled 2 pregnancies and 2% enrolled 3 pregnancies. 80% of participants (n=1178) commenced the study during pregnancy (median gestational age was 22.1 (14.0–30.3) weeks); 22% during first trimester (<13 weeks), 48% during second trimester (13–27.9 weeks), 31% during third trimester (≥28 weeks) and 20% were enrolled at or after birth (median age when commenced study was 3.6 (2.8–6.3) months). The majority were singleton pregnancies with 18 sets of twins (12 in mothers with T1D and 6 in mothers without T1D) and 1 set of triplets (mother without T1D). The median date of birth was 21 September 2017 (ranged from 29 November 2012 to 10 July 2020).

Parentage

The term ‘mothers’ in this paper is defined as the gestational carriers of the ENDIA pregnancies (unless specified as genetic mother). 1209 individuals gave birth to their own biological children, and 5 birthed a child conceived using a donor oocyte (all women without T1D). Additionally, one oocyte donor also participated with their own pregnancy. This resulted in 1214 gestational mothers but only 1213 unique contributors of maternal genes. There were 1217 unique individuals who contributed paternal genes and are referred to as ‘fathers’. Of 1217 fathers, 178 (15%) declined participation, their identity was unknown or not disclosed to the study. These data are considered missing.

In summary, the 1473 live born ENDIA children had 2435 unique ‘parents’, comprising genetic gestational mothers (n=1209), non-genetic gestational mothers (n=5), oocyte donors (n=4) and genetic fathers (n=1217). The 1473 children had 900 unique siblings who were not undergoing longitudinal follow-up. In total, there were 4808 people across 1214 families in ENDIA.

T1D proband relationships

At the time of birth, 912 children (62%) had a maternal proband, 413 (28%) had a paternal proband and 167 (11%) had a sibling proband, including 21 half siblings on the maternal side and 2 half siblings on the paternal side. More than one FDR with T1D was noted in 2% of families (eight maternal and paternal, nine maternal and sibling, seven paternal and sibling and four with two siblings). The median age of T1D diagnosis was 13.7 (9.0–21.4) years for maternal probands, 17.6 (11.0–25.9) years for paternal probands and 3.8 (2.0–6.5) years for sibling probands. The median T1D duration was 17.3 (10.4–23.0) years for mothers, 16.4 (8.2–22.9) years for fathers and 2.2 (0.8–4.2) years for siblings. 12 family members were diagnosed with T1D after the ENDIA child’s birth.

Demographic and clinical characteristics of ENDIA parents

Most ENDIA parents were born in Australia (80%) and 1% identified as Indigenous (Aboriginal or Torres Strait Islander). ENDIA parent demographics were generally comparable to the Australian population (table 1); exceptions included a greater proportion of ENDIA parents born in Australia, lived in a major city, achieved tertiary education and had higher socioeconomic status.

Comorbidities of ENDIA parents

The prevalence of overweight (30%) and obesity (20%) among ENDIA parents was consistent with the Australian population. 1 mother and 8 fathers had type 2 diabetes. 6% of mothers had known retinopathy (84 mothers with T1D), 2% had nephropathy (25 mothers with T1D and 1 mother without T1D), 5% had coeliac disease (65 mothers with T1D and 7 mothers without T1D) and 24% reported another autoimmune disease (269 mothers with T1D and 75 mothers without T1D).

Pregnancy and birth outcomes

Several significant differences in pregnancy characteristics were identified between ENDIA mothers with and without T1D (table 2). ENDIA mothers with T1D were 1.6 times more likely to be nulliparous compared with mothers without T1D (95% CI 1.3, 2.0; p<0.001). Similarly, mothers with T1D were 1.3 times more likely to be experiencing their first pregnancy (primigravida) (95% CI 1.0, 1.7; p=0.02). There was no difference in assisted conception or in gestational weight gain for mothers with and without T1D. The odds of a mother with T1D having pre-eclampsia were 6.0 times that of a mother without T1D (95% CI 3.7, 10.4; p<0.001). For those mothers who were tested for group B streptococcus, the odds of a mother with T1D having a positive test (vs a negative test) were 1.6 times the odds of a mother without T1D (95% CI 1.1, 2.2; p=0.01). Gestational diabetes rates were similar in the mothers without T1D compared with the Australian population (12.7%).36

Characteristics of birth outcomes are presented in table 3. Mothers with T1D were less likely to have spontaneous onset of labor and more likely to deliver via cesarean section than mothers without T1D. 52% of infants were assigned male sex at birth. Gestation at birth revealed a 2-week shorter duration for infants of mothers with T1D compared with those without T1D. Premature birth was significantly more common among mothers with T1D compared with either mothers without T1D or the Australian average. Infants born to mothers with T1D were 648 g heavier than infants born to mothers without T1D, and more likely to be classified as LGA. Admission to NICU was higher among infants born to mothers with T1D compared with those without T1D, although duration of stay in NICU was 40% shorter in mothers with T1D. Congenital abnormalities, neonatal hypoglycemia and neonatal jaundice were more common among infants born to mothers with T1D, when compared with those born to mothers without T1D. Infants born to mothers with T1D were also less likely to score 7–10 on the Apgar scale at 5 min compared with those born to mothers without T1D. The majority of mothers with and without T1D initiated breastfeeding.

Genetics

HLA genotyping was conducted on 4088 individuals in ENDIA families (85% of the total cohort; table 4). There were 115 individuals (2%) who provided DNA, but their HLA type could not be imputed due to poor quality material, rare allelic combination or sequencing errors. HLA types and T1D status were available for 2612 FDRs, including mothers, fathers and siblings. The prevalence of the highest-risk HLA type DR3,4 was 31% in the FDR with T1D, 6% in the FDRs without T1D and 12% in the ENDIA infants. Likewise, the proportion of the lowest-risk HLA type, DRXX, was intermediate in the ENDIA infants relative to the FDRs with and without T1D at 22%, 9% and 37%, respectively. The occurrence of moderate-risk HLA types was similar across the groups, with 60% for FDRs with T1D, 57% for FDRs without T1D and 66% for ENDIA infants.

Observed cohort biases

Several observed patterns in the ENDIA cohort do not reflect natural distributions but are influenced by methodological choices made during study recruitment. There is an imbalance among the T1D proband distribution in the cohort. Although men and women are equally affected by T1D, we report a much higher proportion of mothers with T1D compared with fathers with T1D. This is a result of the study eligibility. ENDIA required women that were pregnant or had recently given birth. As such, face to face recruitment through high risk and specialist diabetes antenatal clinics provided the most cost-effective and successful recruitment strategy. Strategies to target fathers with T1D and parents of children with T1D having another baby included social media posts and liaison with adult and pediatric endocrinologists. However, these proved less fruitful. Details of recruitment strategies used in ENDIA have been described previously.30 There are also more nulliparous women with T1D compared with those without T1D. This arose due to the study’s eligibility criteria. For a family to be eligible to participate in ENDIA, the child had to have an FDR with T1D. For mothers without T1D, this means that it was either the father or a sibling. So, for mothers without T1D, around a third had to already have a child that had developed T1D to be eligible for the study.

ENDIA as a T1D pregnancy cohort

With over 900 mothers with T1D enrolled in ENDIA, we believe this is one of the largest prospective studies of T1D in pregnancy globally. As expected, mothers with T1D had substantially higher rates of adverse pregnancy outcomes and pregnancy complications themselves, with six times the odds of having pre-eclampsia. Similar rates of pre-eclampsia are reported in Australia (17.9%)43 and Europe (18.1% in T1D).44

Mothers with T1D had a fourfold higher likelihood of induced labor, six times higher rates of cesarean section and a 12-fold higher risk of premature birth compared with mothers without T1D. Similar rates of premature birth in T1D pregnancies are reported in Australia45 and Europe.44 46 Birth outcomes for the mothers without T1D were comparable to the general Australian population birth outcomes.36 ENDIA babies born to mothers with T1D were, on average, 150 g heavier at 2 weeks shorter gestation, amounting to a 10-fold higher odds of being LGA in association with more neonatal complications and admission to SCN and NICUs. Again, these complications are comparable to other Australian43 and Danish44 data, but higher than those reported in the UK (52.5%).46 The high rates of preconception maternal and paternal obesity, maternal obesity in pregnancy, higher birth weight (>3.5 kg) and cesarean section are important as potential prenatal initiators of islet autoimmunity.47

Comparison with Australian population

ENDIA is a national study with participants located in all Australian states and territories. The location of the majority of participants is consistent with the distribution of the Australian population.36 The regional participant program and outreach clinics allowed for participation anywhere in Australia, which enabled representativeness of the Australian population. However, a higher proportion of the ENDIA population lived in major cities where lead recruitment sites were located (81% ENDIA vs 73% Australian population). The nine recruitment sites across Australia covered 92% of the Australian population.

Although this is a cohort at risk of developing T1D, parental demographics were similar to that of the Australian general population. For mothers without T1D, this was less likely to be their first pregnancy (37%) compared with the Australian population (42.4%).36 This was an artefact of the FDR study design, which was mentioned previously. The 20% prevalence of obesity observed in ENDIA parents was similar to that reported in the Australian population (21%).36 A higher proportion of ENDIA parents were born in Australia, achieved tertiary education and had less socioeconomic disadvantage compared with the general population. An increase in incidence of childhood T1D has been shown with higher socioeconomic status and residence in urban areas of Western Australia,48 potentially increasing the risk of islet autoimmunity in our ENDIA children.

Distinctions between ENDIA and other T1D risk cohorts globally

The major points of difference between ENDIA and most T1D risk cohorts with follow-up in early life are (1) ENDIA commenced recruitment in pregnancy, (2) eligibility was based on FDR status alone without consideration of HLA genotype and (3) the ENDIA population is based in the Southern Hemisphere. Details of the 14 major birth cohorts are provided in online supplemental figure 2 and online supplemental table 2, which summarizes the commencement date, cohort size, countries of recruitment and eligibility criteria. Only one established cohort, the EDIA intervention trial based in Finland,49 commenced recruitment in pregnancy, but followed only a small number of mothers and offspring (n=87). 80% of live born children in ENDIA were followed from pregnancy (n=1178) with two-thirds recruited before the third trimester. The inclusion of data and samples in pregnancy will allow, for the first time, a prospective assessment of prenatal exposures that have already been associated with islet autoimmunity and T1D risk in epidemiological studies. This includes maternal viral infections and diet,47 as well as identifying new exposures and risk factors such as the function of the gestational microbiome, and other omic systems (metabolome, lipidome, proteome, glycome, etc).

The inclusion of children based exclusively on FDR status, without exclusion of specific ‘low-risk’ HLA types, is a feature of ENDIA, BabyDiab and Australian BabyDiab. All other birth cohorts have excluded children with DRXX HLA types. The prevalence of DRXX HLA types was 22% among the ENDIA children and 16.4% among the BabyDiab children50 and may provide insight into the action of protective genotypes.

Future plans

Follow-up of the 1473 children at-risk for the development of islet autoimmunity and T1D continues until they reach 10 years old, are diagnosed with T1D, or withdraw. The current median age of the cohort is 6.4 years old. A nested case–control study is currently being undertaken to investigate associations between persistent islet autoimmunity and longitudinal omics exposures.25 A definitive case–cohort analysis is planned to define associations from the pregnancy onwards with the development of persistent islet autoimmunity and coeliac autoimmunity.

Opportunities and limitations

ENDIA’s investigation of prenatal and postnatal exposures affecting offspring’s risk is limited by current duration of follow-up and the need for sufficient numbers of children to develop persistent islet autoimmunity to adequately power biomarker studies.25 The heterogeneity of T1D pathogenesis that is increasingly unraveled makes this particularly relevant. For this reason, opportunities to combine later analysis with larger postnatal cohorts will be valuable, whereby signatures in larger cohorts can be measured in ENDIA in relation to the unique exposures in prenatal and postnatal life that ENDIA provides. The COVID-19 pandemic occurred during the final stages of pregnancy and birth. Each state and hospital had differing levels of restrictions during 2020–2022 which impacted some sample collection. Modifications were made to allow samples and data to still be collected,51 and the opportunity arose to study the impact of COVID-19 exposure on the development of islet autoimmunity.

Collaboration

The ENDIA Study is involved in many national and international collaborations and we welcome collaboration with other researchers. Researchers who are interested in potential collaboration should contact ENDIA (endia@adelaide.edu.au) to complete a sample and data access request evaluation by the ENDIA Study Steering Committee. For more information, refer to the ENDIA website (https://www.endia.org.au/for-researchers/).