Gestational Age and Birth Weight and the Risk of Childhood Type 1 Diabetes: A Population-Based Cohort and Sibling Design Study

Type 1 diabetes is a heterogeneous autoimmune disease characterized by destruction of pancreatic β-cells, resulting in absolute insulin deficiency (1). Although autoimmunity is suggested as the predominant effector mechanism, it is thought that type 1 diabetes precipitates in genetically susceptible persons due to an environmental trigger (2,3). The prevalence of type 1 diabetes in the U.S. increased from 1.4 per 1,000 in 2001 to 1.93 per 1,000 in 2009 (4). Similar trends were observed in European countries (5). In Sweden, the incidence of type 1 diabetes increased between 1978 and 2004 but plateaued after 2005 (6).

Several studies found associations between gestational age and birth weight and type 1 diabetes, although the findings are inconsistent. In a recent meta-analysis of 14 case-control studies and four cohort studies, preterm birth (gestational age <37 weeks) was found to increase the risk of type 1 diabetes by 18% (7). The authors highlighted several limitations in the literature such as a lack of consistent adjustment for important confounders. Although the majority of studies reported estimates adjusted for several important confounders, the adjustments varied across studies. It was noted that none of the included studies provided data on birth weight for gestational age. Another recent meta-analysis reported a 17% increased risk of type 1 diabetes among macrosomic infants, while low birth weight was not significantly associated with type 1 diabetes (8). The authors noted lack of proper confounding adjustment as a limitation, since only half of the 12 included studies provided adjusted estimates of the association between birth weight and type 1 diabetes. Among studies that reported adjusted estimates, only three adjusted for gestational age, and one study adjusted for age and sex only. Significant uncertainty remains, however, about the presence and strength of any association between birth weight and type 1 diabetes (9).

We aimed to examine the effects of gestational age, small for gestational age (SGA), large for gestational age (LGA), and birth weight on risk of childhood type 1 diabetes using population-based Swedish data with information on several potential confounders. We then performed a sibling control study to assess whether any observed associations in the cohort study were likely to be causal. Sibling control analyses allow us to draw stronger causal inferences about these aspects of pregnancy (10–14). This approach is efficient in accounting for family factors, such as genetics and environmental factors that are shared by siblings. To our knowledge, this is the largest study on this topic to date and the first to use sibling control analyses to assess the effects of perinatal risk factors on type 1 diabetes.

The study used data from the Swedish national registers held by the Swedish National Board of Health and Welfare and Statistics Sweden. Each resident in Sweden is assigned a unique identifier, the Personal Identity Number, which enables data linkage from various registers and among relatives, such as parents and siblings (15). Using the Swedish Medical Birth Register, we identified almost all singleton live births in Sweden between 1 January 1973 and 31 December 2009. This register contains obstetric, maternal, and neonatal data on 96–99% of births in Sweden (16). The sibling control design study included the children with type 1 diabetes and their siblings. This is a matched case-control study—nested within the overall cohort—where the case subject has type 1 diabetes, the control subjects do not, and the case and control subjects are siblings.

Gestational age and birth weight were measured at delivery and recorded in the Medical Birth Register. Gestational age was determined at delivery and calculated using dates from early second-trimester ultrasound or calculated by menstrual dating (17) and categorized as very preterm, 22–32 weeks; preterm, 33–36 weeks; early term, 37–38 weeks; term, 39–40 weeks (reference group); and postterm, 41+ weeks. Birth weight was considered erroneous if it was recorded as <500 g or >5,500 g. Birth weight was categorized as <1,500, 1,500–2,499, 2,500–2,999, 3,000–3,999 (reference group), and 4,000–5,500 g. SGA and LGA were defined according to the Swedish weight-based growth standards (18). SGA was defined as a birth weight of 2 SD below the mean and LGA as birth weight of 2 SD above the mean of the sex-specific and gestational age distributions. Children were classified into three categories: SGA, LGA, and appropriate for gestational age (AGA) (neither SGA nor LGA).

The Swedish National Patient Register contains records of inpatient diagnoses in Sweden since 1964 (full national coverage since 1987) and outpatient diagnoses since 2001. The date of onset of type 1 diabetes was defined as the date of the first hospitalization, which led to the diagnosis of type 1 diabetes. Childhood type 1 diabetes, before 15 years of age, was defined according to the ICD-8 (250), ICD-9 (250), and ICD-10 (E10). The cohort was followed from the date of birth until onset of type 1 diabetes, 15th birthday, death, migration, or 31 December 2009 (end of the study period). The Migration Register provided the dates of migration from Sweden, while information on date of death was obtained from the Cause of Death Register.

Data on infant sex, maternal age, BMI, prepregnancy diabetes and gestational diabetes mellitus, country of origin, birth order, preeclampsia, and mode of delivery were obtained from the Medical Birth Register. Maternal BMI is measured at the first antenatal visit, which takes place before 15 weeks’ gestation in Sweden (19). Data on maternal education level were obtained from the Education Register, which contains information on the residents’ highest level of completed formal education.

The statistical analysis to examine the effects of gestational age, birth weight, SGA, and LGA on childhood type 1 diabetes was performed in two steps. First, log-linear Poisson regression with aggregated person-years was performed for each exposure variable using the entire cohort. All Poisson models were adjusted for offspring age, as a time-dependent variable; year of birth (in 1-year categories); birth order and sex; maternal age; BMI; prepregnancy diabetes; gestational diabetes mellitus; country of origin; and education level (these variables were included in the models as presented in Table 1).

The second step aimed to adjust for unmeasured familial environmental and genetic confounding factors shared by siblings using a sibling control design. Conditional logistic regression was used for the sibling control analyses with siblings identified through the Multi-Generation Register (20). The conditional logistic regression analysis included siblings where the control subject was under follow-up and type 1 diabetes free at the age that the sibling with type 1 diabetes was diagnosed. In these analyses, only siblings discordant for each exposure variable categories as well as type 1 diabetes contributed to the measure of association. However, siblings concordant for the exposure variable categories were included in the analysis, as they contribute to the potential confounders estimates. The models were adjusted for the same variables as in the Poisson models apart from maternal country of birth, which was the same for both siblings. A sensitivity analysis was performed including full siblings only in the sibling study.

Considering the complex relationship between birth weight and birth weight for gestational age with gestational age, we repeated the birth weight for gestational age models restricting the analysis to children with gestational age of ≥37 weeks and then ≥39 weeks. Similarly, we repeated the gestational age models excluding SGA and LGA children. Between 1973 and 1981, we were able to classify mode of delivery into unassisted vaginal delivery, instrumental vaginal delivery, and Caesarean section (CS), while from 1982 data on elective and emergency CS were available and were used to classify CS. Data on maternal BMI were available from 1981 onward. However, >20% of the women who delivered their children from 1981 onward had missing BMI data. Therefore, we repeated the models from 1982 onward (to coincide with the more detailed data on mode of delivery) as well as models excluding all women with missing BMI. To assess the potential impact of missing BMI data on the observed association between gestational age and type 1 diabetes, we performed two Poisson regression models including all births from 1981 onward with known maternal BMI. In the first model we removed maternal BMI and in the second model we included BMI (in addition to the other potential confounders as described in the legends to Tables 2–4). Considering that data coverage was complete at the national level from 1987, we repeated the statistical models including births from 1987 onward.

Further analyses were performed excluding 1) instrumental vaginal delivery and CS, 2) children of mothers who had preeclampsia, 3) children of mothers who had pregestation diabetes, 4) children of women who had gestational diabetes mellitus, and 5) children of women who were classified as obese (BMI ≥30 kg/m²). We performed these analyses to assess whether those factors have any influence on the observed associations, as they were found to be associated with type 1 diabetes in previous studies (19,21,22). We also repeated the Poisson models including only children. To examine the distributional assumption of the Poisson regression regarding overdispersion, we performed negative binomial regression models. There was no evidence to suggest the distributional assumption was violated; therefore, the Poisson models results are presented throughout the manuscript. In post hoc analyses, we repeated the sibling control models excluding children of women who had pregestational or gestational diabetes mellitus. Furthermore, we performed detailed post hoc analyses to examine the association between gestational age and type 1 diabetes using finer gestational age categories. We repeated the Poisson regression analysis using gestational age categorized as 22–29 weeks and in 1-week categories thereafter. Gestational age weeks 22–29 were combined in one category owing to small number of cases in each week category. Finally, we performed sensitivity analysis to examine the potential effect of unmeasured confounding. (See Supplementary Data for details.)

The study cohort consisted of 3,624,675 singleton live births with known Personal Identity Number and sex in Sweden between 1 January 1973 and 31 December 2009 (Fig. 1). During the study period (42,411,054 person-years) there were 13,944 childhood type 1 diabetes cases. Mean age at diagnosis was 8.4 years (8.5 in boys and 8.3 in girls). Mothers of children with type 1 diabetes were on average older, had a higher education level, and were more likely to have prepregnancy diabetes. The sibling control study consisted of 82% (11,403 of 13,944) of the type 1 diabetes case subjects in the cohort study (sibling pairs discordant on type 1 diabetes). The majority of the children with type 1 diabetes who were not included in the sibling analysis were only children (2,163 of 13,944; 15.5%). The maternal characteristics in the sibling control study were similar to those of the entire cohort apart from pregestation diabetes and gestational diabetes mellitus, which were more common in the sibling control study. More details are summarized in Table 1 and Supplementary Table 1.

Gestational age was significantly associated with the risk of type 1 diabetes. Very preterm (relative risk [RR] 0.67 [95% CI 0.53, 0.84]) and postterm (RR 0.87 [95% CI 0.83, 0.90]) birth subjects were less likely to develop type 1 diabetes, while preterm (RR 1.18 [95% CI 1.09, 1.28]) and early-term birth subjects (RR 1.12 [95% CI 1.07, 1.17]) were more likely to develop type 1 diabetes compared with children born at term (Table 2). All these associations remained statistically significant and almost unchanged in the sibling control study (Table 2). We further performed a log-linear Poisson regression to examine the association between gestational age (classified into 1-week categories whenever possible) and type 1 diabetes (Supplementary Table 2). These findings suggested that the risk of type 1 diabetes was largest among children born at 34 weeks’ gestation. To assess how robust the association is in relation to unmeasured confounding, we performed sensitivity analyses using different scenarios. The details of this analysis and the results are presented in Supplementary Data.

In the cohort study, there were significant associations between SGA (RR 0.83 [95% CI 0.75, 0.93]) and LGA (RR 1.14 [95% CI 1.04, 1.23]) and the risk of type 1 diabetes (Table 3). The association between SGA and type 1 diabetes remained unchanged and statistically significant in the sibling analysis, while the association between LGA and type 1 diabetes was no longer statistically significant (Table 3). Restricting the cohort analyses to early-term and term babies did not change the results of SGA or LGA materially.

Very low–birth weight (<1,500 g) children were at lower risk of type 1 diabetes compared with children with normal birth weight (RR 0.66 [95% CI 0.48, 0.91]) (Table 4). This association remained statistically significant in the sibling analysis (RR 0.50 [95% CI 0.31, 0.80]) (Table 4). Further adjustment for gestational age in the sibling control model reduced the association slightly (RR 0.59 [95% CI 0.33, 1.07]). The other birth weight categories were not associated with type 1 diabetes, with most RRs close to unity (Table 4). Of the very low–birth weight children, 95% were born before 37 weeks’ gestation (16,506 of 17,333). When the analysis was restricted to children born before 37 gestation weeks, the association between very low birth weight (RR 0.47 [95% CI 0.32, 0.67]) and low birth weight (RR 0.73 [95% CI 0.60, 0.88]) and type 1 diabetes was statistically significant with larger effect size, although the statistical interaction terms between birth weight and gestational age categories were not statistically significant (P value for interaction >0.05).

When the models were repeated for births from 1987 onwards at the cohort level, the findings were not materially changed compared with the full cohort models (data not shown). When we repeated the cohort analyses for all the exposure variables excluding birth before 1981 and then excluding all births with missing BMI data, no material change was observed. Repeating the analysis using data on full siblings only did not change the results materially. Additional cohort analyses on the three exposure variables were performed restricting to 1) only children, 2) AGA children, 3) unassisted vaginal delivery, 4) mothers with no prepregnancy diabetes, 5) mothers with no gestational diabetes mellitus, 6) nonobese mothers, and 7) mothers with no preeclampsia (Supplementary Table 3). Overall, none of these restrictions had any material effect on the findings. Restricting the Poisson models to the first two children per mother did not change the results materially. Excluding diabetes and gestational diabetes mellitus from the sibling models did not explain any of the observed associations.

This study investigated the effects of gestational age, birth weight, SGA, and LGA on the risk of childhood type 1 diabetes using a large population-based cohort including nearly all births in Sweden over four decades. We used a unique approach by applying both cohort and sibling control designs using the same population. Preterm and early-term birth were associated with an increased risk of type 1 diabetes in the cohort study, while very preterm, postterm, SGA, and very low birth weight were associated with a reduced risk of type 1 diabetes. All these associations remained significant in the sibling control study, suggesting a potential causal association.

LGA was associated with an increased risk of type 1 diabetes in the cohort analysis, but the association disappeared in the sibling analysis, which suggests that familial factors that are shared between siblings such as genetics and environmental factors might explain, at least partly, this association. There was very little evidence to support an association between birth weight of ≥1,500 g and type 1 diabetes.

The existing literature on birth weight and gestational age and type 1 diabetes is inconsistent. A recent systematic review and meta-analysis identified 18 studies on the association between gestational age and type 1 diabetes (7). Interestingly, all the identified studies were published in the last two decades, reflecting the increasing interest in the effect of perinatal risk factors on type 1 diabetes. The present finding that preterm birth increases the risk of type 1 diabetes in the cohort study is almost the same as the pooled estimate from the meta-analysis. However, the meta-analysis did not report data on very preterm, early-term, or postterm and type 1 diabetes. Another meta-analysis suggested that birth weight >4,000 g increases the risk of type 1 diabetes by 43% (8). This is inconsistent with the findings of the current study, as we found no evidence that type 1 diabetes increased risk among children born with birth weight >4,000 g. Although LGA children were found to have 14% increased risk of type 1 diabetes in the cohort study, this association was not consistent with a causal association in the sibling analysis. Our findings on low birth weight, however, are consistent with the meta-analysis, especially when we restricted the low–birth weight analysis to preterm births. Robertson and Harrild reported no association between maternal and neonatal risk factors including birth weight and gestational age and type 1 diabetes in a matched case-control study (23). It should be noted that they found increased odds of type 1 diabetes in children born preterm (<37 weeks) or with low birth weight (<2,500 g), although the results were not statistically significant. This suggests that the apparently inconsistent findings could be related to lack of adequate statistical power in Robertson and Harrild (23). Additionally, we categorized both birth weight and gestational age into more tightly defined groups including very low birth weight and very preterm birth, while they used preterm birth as <37 weeks and low birth weight as <2,500 g, which makes the comparison more complicated. The observed association between gestational age and type 1 diabetes is consistent with a large population-based study from western Australia, which found increased risk of type 1 diabetes in preterm and early term children (24). Haynes et al. (24) also reported a small association between increased birth weight and increased birth weight for gestational age and higher risk of type 1 diabetes, while we found little evidence for such an association. In another large population-based study from Northern Ireland, Cardwell, Carson, and Patterson reported an increased risk of type 1 diabetes with higher birth weight, which is not consistent with our findings (25). Comparing our gestational age finding, however, is more difficult, as Cardwell, Carson, and Patterson categorized gestational age as <39 (reference group), 39, 40, and ≥41 weeks’ gestation. Contrary to the present findings, Dahlquist, Patterson, and Soltesz (1999), reported no association between gestational age and type 1 diabetes in the EURODIAB study, which included data from seven European studies (26). They also reported reduced risk of type 1 diabetes in low–birth weight children but not SGA babies. To our knowledge, ours is the first study to perform a sibling control study nested within a cohort on this topic; therefore, there are no other studies with which to compare our results.

This study has several strengths. First, the study was based on a very large population-based data of 3.6 million children born in Sweden over four decades, which is, to our knowledge, the most comprehensive study on this topic. Second, the data obtained from the national registers were prospectively collected; therefore, the data on the outcome, exposures, and potential confounders are not subject to recall bias. Third, the type 1 diabetes diagnoses were based on ICD-8, -9, and -10 with a known and accurate date of first hospitalization, which is considered the date of diagnosis. Data on type 1 diabetes in Sweden are known to be of very high quality (27). Fourth, we were able to adjust for several potential confounders, which were adjusted for in previous studies. Finally, in addition to the conventional cohort analyses, sibling control analyses were performed. It is worth noting that the sibling design study can rule out a number of potential explanations for observed associations compared with observational designs using unrelated control subjects. Moreover, sibling design studies may play an important role in assessing potentially causal associations. In the current study, statistical models of the sibling control study allowed us to adjust for unmeasured factors that are shared by siblings such as family environment, diet, lifestyle, maternal characteristics, and genetic factors.

The following limitations should be considered when interpreting the study findings. First, we used data on almost all births in Sweden from 1973, and complete nationwide coverage on diagnoses was not achieved until 1987. However, our sensitivity analyses showed that restricting the data to births from 1987 onward was consistent with the overall results. Second, although we had access to several potential confounders, there was a lack of data on several others. For example, we had no data on maternal life style during pregnancy such as physical activity, diet, and weight gain. Excessive weight gain during pregnancy, maternal nitrite intake, and cod liver oil supplementation have been associated with the risk of type 1 diabetes (28). Furthermore, we had no data on parental and family life style such as family diet and attitude toward acquiring health care. However, the risk of residual confounding was reduced by the sibling control analyses. Sibling control statistical models are effective in adjusting for unmeasured familial characteristics that are shared by siblings, i.e., maternal and paternal factors that were fixed for each family across pregnancies. However, these methods cannot rule out unmeasured confounding factors that simultaneously vary between siblings, i.e., parental factors that are not permanently fixed for each family and could potentially be different for different pregnancies (29).

It has been hypothesized that the effect of preterm and early-term birth on the risk of type 1 diabetes may be related to fetal growth restriction (7). Our data, however, do not support this hypothesis, as the association was unchanged when SGA children were excluded from the analysis. Another hypothesis suggests that preterm children may have experienced growth restriction in early life and the catch-up growth may result in later insulin resistance (30). It is possible that preterm birth is associated with structural change within organ systems and epigenetic changes leading to higher risk of type 1 diabetes. The observed associations between preterm and early-term birth and type 1 diabetes are unlikely to be related to maternal diabetes, as this was corrected for in the cohort and sibling analyses. Notably, the results remained unchanged when women with pregestation diabetes or gestational diabetes mellitus were excluded from the analyses. Finally, preterm children have been suggested to have an altered microbiota, which may play a role in the development of type 1 diabetes (31). Kostic et al. (32) performed a study to examine the link between human gut microbiome in infancy and type 1 diabetes in 33 infants genetically predisposed to type 1 diabetes. They found a marked drop in α-diversity in type 1 diabetes progressors between seroconversion and type 1 diabetes diagnosis. It is interesting that very preterm and very low–birth weight babies have been suggested also to have an altered microbiota, but they were not associated with an increased risk of type 1 diabetes in the current study (31). Indeed, very preterm, SGA, and very low–birth weight babies appeared to have significantly lower risk of type 1 diabetes. The effects of very preterm and very low birth weight on type 1 diabetes were not studied thoroughly in the past, as such a study of rare exposure and rare outcome would require large cohorts. The biological mechanism of this association is not known, but it is possible that it is due to fetal programming where the most vital organs are protected. These findings are worth replicating in other populations, as they may generate further hypotheses on the effect of perinatal risk factors on type 1 diabetes.

We found that preterm and early-term babies are at increased risk of type 1 diabetes, while very preterm–birth, very low–birth weight, and SGA babies are at reduced risk of type 1 diabetes. The sibling study suggested that these associations, although small, were not due to familial factors shared by siblings. Further research is warranted to replicate these findings and understand the potential biological mechanisms.

Funding. This work was carried out at the Irish Centre for Fetal and Neonatal Translational Research (INFANT) and was funded in part by Science Foundation Ireland (grant no. 12/RC/2272). Financial support was provided through grants provided from Stockholm County Council (ALF projects) and the Swedish Research Council (grant no. 2011-3060) and through the Swedish Initiative for research on Microdata in the Social And Medical sciences (SIMSAM) framework (grant no. 340-2013-5867), the Swedish Heart-Lung Foundation, and the Strategic Research Program in Epidemiology at Karolinska Institutet.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. A.S.K. conceptualized and designed the study, performed the statistical analysis, and drafted the initial manuscript. L.C.K. contributed to the study design, interpreted the results, and reviewed and revised the manuscript. C.L. contributed to the study design and supervised the statistical analysis, interpreted the results, and reviewed and revised the manuscript. P.M.K. contributed to the drafting of the initial manuscript, interpreted the results, and reviewed and revised the manuscript. T.G. contributed to the study design and prepared the study cohort including performing data linkage from several registers and reviewed and revised the manuscript. R.M. advised on the statistical analysis, interpreted the results, and reviewed and revised the manuscript. C.A. conceptualized and designed the study (with A.S.K.), acquired the permission to access the data and perform the study, interpreted the results, and reviewed and revised the manuscript. A.S.K. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.