Perinatal risk factors for type 1 diabetes revisited: a population-based register study

Studies on the natural history of childhood-onset type 1 diabetes mellitus show that markers of beta cell autoimmunity appear several years before the onset of clinical disease, with islet autoantibodies being detectable during infancy [1‐3]. Persistent positivity to multiple such autoantibodies is strongly associated with clinical disease [4]. It is therefore of interest to search for exposures that may start the process early in life, leading to type 1 diabetes later during childhood. An early suggestion was that intrauterine viral infections may be important, since the prospective follow-up of a cohort of individuals with rubella embryopathy showed they had a dramatically increased risk of diabetes later in life [5]. Since 1974, all girls (and later all children) in Sweden have been vaccinated against rubella and the disease has become exceedingly rare; the last domestic case of rubella embryopathy was reported in 1985 [6]. Still, the reported incidence of childhood-onset type 1 diabetes in Sweden doubled in the 30 years from 1978 to 2007 [7].

A series of population-based studies originating from prospectively recorded perinatal data from either registries or hospital records have suggested associations between the risk of type 1 diabetes and factors such as maternal pre-eclampsia, delivery by Caesarean section, older maternal age, increased birthweight by gestational age and neonatal jaundice induced by maternal–fetal blood-group incompatibility [8‐13]. Some of these have been confirmed in meta-analyses [14‐17], whereas others remain controversial [18,19]. Socioeconomic factors may confound some of these risk factors.

The Swedish Childhood Diabetes Register (SCDR) has recorded virtually all incident cases of childhood-onset type 1 diabetes in Sweden since 1 July 1977 [7]. In the current study, we matched four controls to each case in the register and linked data from official health registers, providing a larger database than that used by many meta-analyses, as well as individual-level information on a number of health and socioeconomic exposures. Compared with our own previous analyses of perinatal risk factors for type 1 diabetes using data from the SCDR [8‐11], the current study uses around five times as many cases covering a large timespan. Thus, we aimed to use this substantially larger database to conduct an in-depth study of early factors influencing the trajectory of type 1 diabetes development.

Our study population consisted of incident type 1 diabetes cases from the SCDR from 1 January 1978 to 31 December 2013 and four controls per case from the Swedish Total Population Register, matched for age (birth year and month) and municipality at the time of diagnosis of the incident case. Statistics Sweden, Stockholm, performed the matching.

For cases and controls, we extracted data from the Swedish Medical Birth Register (MBR) on the type of delivery (vaginal or Caesarean section), the age of the mother at the birth of the child, and the weight, gestational length and gestational age of the child at birth. Furthermore, we obtained data on maternal illnesses (i.e. diabetes, hypertension, pre-eclampsia, eclampsia, urinary tract infections [UTIs], premature rupture of membranes [PROM], pyrexia and other infections during labour). The MBR has recorded data on prenatal, delivery and neonatal care since 1973 [20, 21], and it is mandatory for all healthcare providers to report to the register. Hence, we discharged those born before 1973 (n = 13,905). The MBR has been validated and found to be a valuable source of information for research on perinatal events [21]. Very little systematic error was found, and missing data in the register affects prevalence estimates but has little effect on risk estimates. We also collected data from the National Patient Register (NPR) [22], which includes all inpatient diagnoses since 1987 [23]. The register uses the WHO ICD coding system [24] (see Electronic Supplementary Material [ESM] Table 1). The combined use of the MBR and the NPR was suggested from the validation of the MBR in order to decrease problems with missing data for infant diagnoses [21]. From both the MBR and the NPR [22], we collected data on neonatal jaundice, intrauterine and birth asphyxia and respiratory distress of the newborn, neonatal aspiration syndromes, various neonatal infections and whether the child had been treated with phototherapy for neonatal jaundice.

From 1992, data in the MBR on first-trimester maternal height and weight was complete enough to allow calculation and classification of maternal BMI [21]. Normal weight was classified as BMI less than 25 kg/m², overweight as BMI 25–30 kg/m² and obesity as BMI more than 30 kg/m².

To limit the presence of known confounders we restricted our study population to singleton births, children of non-diabetic mothers and children of mothers born in the Nordic countries. The number of individuals in the study population by the mother’s country of birth is displayed in ESM Table 2. We conducted univariate and multivariate analyses for various factors and the risk of type 1 diabetes using conditional logistic regression models, thus considering the matched case–control design. For the univariate analyses of perinatal variables and diagnoses, we also stratified the data by sex and age at diabetes onset. Assuming that the OR approximates the RR, we estimated interactions on an additive scale by calculating the relative excess risk due to interaction (RERI) between sex and the perinatal variables [25‐27]. This was not feasible for the interaction between age at onset and diagnoses, since age at onset was a matching variable in the initial case–control design. To avoid bias when assessing the total effect of each risk factor, we use directed acyclic graphs as tools for confounder selection in the multivariate analyses [28‐30]. We illustrated the assumptions of underlying mechanisms by causal models in directed acyclic graphs and selected confounders accordingly [31].

Premature birth was defined as delivery before 259 days of gestation, which was then subclassified into moderate to late preterm (224–258 days’ gestation) and very preterm (<224 days’ gestation) [32]. We followed the WHO’s definition of the perinatal period as commencing at 154 days’ gestation and ending at 7 completed days after birth. In our context, the perinatal period contains both maternal and infant variables. Neonatal was defined as relating to or affecting the newborn during the first month of life. Birthweight z scores were calculated from intrauterine growth charts [33]. Furthermore, we stratified the birthweight z scores to compare ORs between different categories of birthweight z scores. Analyses were made by excluding missing variables; however, there were generally few missing variables, except for maternal BMI (ESM Table 3).

To evaluate possible changes in risk factors over time, we also conducted stratified analyses for four different birth cohorts (birth years 1973–1982, 1983–1992, 1993–2002 and 2003–2013). Three different WHO ICD coding systems [24] were used during the study period: ICD 8 (until 1986; http://​www.​wolfbane.​com/​icd/​icd8.​htm), including 4921 cases and 18,739 controls; ICD 9 (1987–1996; www.​icd9data.​com/​2007/​Volume1), including 5535 cases and 20,478 controls; and ICD 10 (from 1997; http://​apps.​who.​int/​classifications/​icd10/​browse/​2016/​en), including 4493 cases and 16,497 controls. In addition, we investigated the association between risk factors and age of onset of type 1 diabetes mellitus.

All analyses were performed using the statistical software R, version 3.4.2 [34] with the package ‘survival’ and the logit function for conditional logistic regression. For the RERI analyses we used bootstrap [26] to calculate CIs using the R package ‘boot’. Collection of matched controls and data linkage were performed at Statistics Sweden and only coded data were delivered back to the research group. The study was approved by the Regional Ethics Review Board at Umeå University and the ethics committee at Statistics Sweden, Stockholm, according to Swedish law on research ethics, and performed in accordance with the Declaration of Helsinki.

After excluding twins, children born to mothers with diabetes and children with mothers born outside of the Nordic countries, we analysed data from 14,949 cases and 55,712 controls (Table 1). There were 34,095 females (7024 cases) and 36,566 males (7925 cases), and the mean age at diagnosis of type 1 diabetes was 8.29 ± 3.87 years.

The present study confirms that the risk of developing type 1 diabetes during childhood is associated with a number of perinatal factors. Birthweight z scores of more than 1 increased the risk and birthweight z scores of less than –1 decreased the risk. For those with higher birthweight, these results support our earlier findings as well as those of some more recent meta-analyses. Both Harder et al [35] and Cardwell et al [15] reported a 10–17% increase in the risk for type 1 diabetes in children with birthweight >4000 g. In contrast to our findings, both meta-analyses also reported non-significant decreases in the risk for children with birthweight <2500 g. Cardwell et al [15] differentiated between cohort and case–control studies and did show that, in the cohort studies, birthweight of <2500 g was associated with a lower risk for diabetes (OR 0.79, 95% CI 0.67, 0.92), but the combined estimate was not statistically significant (OR 0.98, 95% CI 0.84, 1.13). A probable explanation for the different results may be that the meta-analyses did not account for the effect of gestational age, which was possible in the present study. In the substudy, including data from 1992–2013, we could also see that the risk for later type 1 diabetes increased in a dose-dependent fashion with increasing maternal BMI. This finding is in line with a 2015 Swedish study that reported similar correlations between higher maternal BMI and offspring type 1 diabetes [36]. However, that study included only a few perinatal risk factors, thus omitting the combined effects of some other factors on the risk of childhood type 1 diabetes. The association between high birthweight and high maternal BMI and type 1 diabetes support the so called overload or acceleration hypothesis, suggesting that a high pre- or postnatal growth rate may accelerate an ongoing beta cell destruction by nutrient overload [37, 38].

Previous studies have indicated that being born before 37 full weeks of gestation increases the risk of type 1 diabetes in the child [39]. In the present study, we show that prematurity is an independent risk factor for later type 1 diabetes, but the increased risk only applies to children born between 32 and 36 weeks of gestation, whereas birth before week 32 was protective against the later development of type 1 diabetes. Algert et al have shown similar effects of moderate to late prematurity on the risk of type 1 diabetes, but had an insufficient sample size to show significant associations for very preterm infants [40]. Prematurity has been linked to a number of alterations in immune function [41], which may have long-term consequences. Shorter gestational age has been associated with HLA categories conveying a higher risk for type 1 diabetes [42]. Other effects of premature birth include in or ex utero growth restriction, causing impaired glucose regulation later in life and changes in organ function, including endocrine feedback mechanisms or epigenetic effects [43], but how these exactly contribute to the risk of type 1 diabetes is unclear. A protective effect of very preterm birth has not been reported previously in the case of type 1 diabetes, but a recent cohort study from Sweden found that very preterm birth was associated with a lower risk for food allergies [44]. From currently available data, it is not possible to say if the protective effect against type 1 diabetes is conveyed by being very preterm per se, or by other factors leading to the preterm delivery or by the neonatal care provided to the infants.

The rate of Caesarean sections has increased worldwide over many decades, including in Sweden [45], paralleling the increase in the incidence of type 1 diabetes [7]. Caesarean section has also been associated with a higher risk for immune-mediated diseases [46], including type 1 diabetes [17], and is suggested to be linked through differences in gut microbiota composition [47, 48], although this has been recently disputed [49]. A Danish nationwide register study published in 2016 found no association between type 1 diabetes and Caesarean section [50]. In the present study, we found that birth by Caesarean section conveyed a small but statistically significant increase in the risk for type 1 diabetes. However, after allowing for maternal BMI, as we could do for those born after 1992, the association with Caesarean section disappeared. Delivery by Caesarean section is more common in older mothers [51] and in women with a higher BMI [52]. This would support the notion that it is the factors leading to Caesarean section (e.g. higher maternal age and higher maternal BMI) and not the delivery mode itself that is associated with the risk for type 1 diabetes.

When exploring the effects of other diseases during perinatal life, specifically infectious disorders, we found associations between later risk for type 1 diabetes and UTIs during pregnancy and PROM during labour, but not for any other serious infections in the mother or newborn. In animal models, perinatal antibiotic treatment of the mother has been shown to affect the risk of type 1 diabetes in the offspring, but the risk differed depending on which type of antibiotic was used [53]. It has been noted that up to 7% of women are diagnosed with a UTI, including asymptomatic bacteriuria, during pregnancy [54]. We found a very low prevalence of UTIs at only 3–4 per 1000, indicating that the birth and patient registers we used may underestimate the true prevalence, but this would only mean that our results represent a more conservative risk estimate. UTI during pregnancy has previously been associated with adverse perinatal outcomes, including an increased risk for birthweight below 2.5 kg, premature birth, maternal hypertension and pre-eclampsia and amnionitis [55]. Of these, we have shown that moderately preterm birth and high birthweight z score, but not maternal hypertension or pre-eclampsia, are associated with later risk for type 1 diabetes. However, the exact mechanism by which UTIs are associated with type 1 diabetes is unexplained and must be further investigated. PROM has been reported to complicate around 8% of deliveries in the USA [56], which is similar to the prevalence reported in the present study. PROM is associated with preterm birth [56], which we have shown is linked to an increased risk for type 1 diabetes.

We could not confirm earlier findings of an association between pre-eclampsia or eclampsia in the mother and type 1 diabetes in the child [9, 10]. A meta-analysis of published results [14] showed significant heterogeneity among studies and concluded that there is little evidence for a substantial effect of pre-eclampsia on the risk of type 1 diabetes. That meta-analysis, as well as the present nationwide register study, covered a substantial time span of birth cohorts. One explanation for the different findings may be that new, more active treatments during pregnancy have eradicated a possible effect of pre-eclampsia on the fetal immune system.

Neither could we confirm the repeatedly reported and more evident effect (OR between 1.5 and 3) of a relationship between later type 1 diabetes and maternal–child blood-group incompatibility or neonatal jaundice, as previously shown in Sweden and other European countries and in meta-analyses [9, 10, 14]. Again, this could be due to changing perinatal practices; for example, Rh isoimmunisation in the newborn has long been prevented by the active prenatal immune treatment of Rh-negative mothers. Moreover, we found that decreasing proportions of children had a jaundice diagnosis or ABO isoimmunisation in more recent calendar periods of birth.

The sample size of our study is larger than that of all [14‐17, 35, 50] but one [39] of the recent meta-analyses, and thus allowed us to use individual-level data to thoroughly investigate the disputed relationships between perinatal factors and risk for childhood-onset type 1 diabetes. The large dataset made it possible to minimise confounding using exclusion criteria and multivariate analyses and to analyse time trends. Data were prospectively, routinely collected and stored in official registers, similarly among cases and controls. Thus, any over- or underreporting (e.g. of diagnoses) will not have led to type 1 errors, but rather to bias towards unity.

The drawbacks of a large sample size are that it may detect minor and possibly clinically less interesting associations (e.g. maternal age). A further weakness of the study is that three different ICD coding systems were used with clear difficulties in converting between them, even with lists for translation provided by Swedish National Board of Health and Welfare. Again, the same errors should be distributed equally among cases and controls since exposure data were registered long before the diabetes diagnosis. We also performed stratified analyses using the different ICD coding systems for key outcomes without finding significant differences (data not shown).

Even if the use of a large register database has great methodological advances compared with meta-analyses, the conclusions from the present data can only be drawn for Sweden. We are also aware that some of the unexpected results, such as a possible effect of UTIs in the mother, could be due to multiple comparisons and need to be further confirmed in future studies.

In conclusion, our comparably very large prospective and nationwide register study covering 36 years of birth cohorts confirms that perinatal factors (i.e. high birthweight z score, being born between 32 and 36 weeks of gestation, UTI during pregnancy and PROM during labour) are associated with childhood-onset type 1 diabetes. Some of these, such as high birthweight z score and high maternal BMI, lend support to the theory that a high pre- or postnatal growth rate might accelerate ongoing beta cell destruction by nutrient overload, whereas the others are mainly associated with preterm birth. Of the former, some factors, such as high maternal BMI, may be amendable with improved antenatal care.