Introduction

Top Abstract Introduction Subjects and methods Results Discussion References

Type 1 diabetes mellitus is caused by immune mediated destruction of the pancreatic  cells. Both genetic and non-genetic factors are involved in the pathogenesis, but the factors initiating the destructive process are largely unknown. Environmental risk factors may have a role early in life, possibly in utero.¹

Several studies have investigated the relation between maternal age at delivery or birth order and risk of childhood onset type 1 diabetes.^2-13 Many of these studies were relatively small, used census data as controls or siblings as controls, and inappropriately analysed the data as if they arose from a cohort study. ^{2 3 7 13 14} Some studies found a weak increase in risk of type 1 diabetes in children born to older mothers, although others found no significant association.

Studies of birth order have given particularly inconsistent results. Maternal age and birth order are correlated, and the inconsistent findings may be explained by differences in adjustment for maternal age. Furthermore, interaction between maternal age and birth order may exist. Results from studies of association between paternal age at delivery and risk of type 1 diabetes have also been inconsistent. ^{5 7 8 12} The objective of this study was to estimate the associations of maternal and paternal age at delivery and of birth order with the incidence of type 1 diabetes and the interactions between these variables.

	Subjects and methods

Top Abstract Introduction Subjects and methods Results Discussion References

In Norway, all newly diagnosed cases of type 1 diabetes in children under 15 years old have been prospectively registered in the national childhood diabetes registry since 1 January 1989.¹⁵ We designed a cohort study linking records in the medical birth registry with those in the national childhood diabetes registry through the unique personal identification number assigned to all residents of Norway. We linked 1824 of the 1863 cases of type 1 diabetes diagnosed between 1 January 1989 and 31 December 1998. The study was approved by the regional ethics committee and the National Data Inspectorate.

All live births in Norway between 1974 and 1998 contributed time under observation from birth to diagnosis of type 1 diabetes (from 1989 to 1998), age 15 years, death in the first year of life, or 31 December 1998, whichever occurred first. Since registration of type 1 diabetes started in 1989, time under observation was counted only from 1 January 1989 for those born before this date. This means that even if some children developed type 1 diabetes before 1989, it would not influence the results of this study. We did not have information on deaths occurring between age 1 and 15 years, but these are rare and would not influence our results.

We calculated maternal and paternal age as the difference between the birth year of the index child and the birth year of the mother and father, respectively. The child's birth order was inferred from the number of previous births (including stillbirths) reported by the mother at the time of birth of the index child (1. 9% of mothers who had at least one previous birth reported at least one previous stillbirth). We excluded 0.1% of the cohort because of missing data on birth order, including 12 who developed diabetes. In analyses of paternal age, we excluded 7.6% with missing data, including 118 who developed diabetes. Parental age and birth order were divided into five categories (see table 1).

We calculated the number of incident cases and person years under observation in each exposure category using Datab in the Epicure package, version 1.8w.¹⁶ We calculated incidence as the number of incident cases divided by the person years under observation in each category. We estimated rate ratios with 95% confidence intervals by Poisson regression analyses. Exposure variables were also entered as continuous variables with five levels corresponding to the mean in each category.¹⁶ We calculated percentage differences in incidence associated with given changes in exposure by subtracting one from the rate ratios and multiplying by 100. We used the likelihood ratio test for exposures entered as continuous variables to test for trend.

We included calendar period of birth in five year categories and age group in three year categories to adjust for possible period effects. We also entered maternal diabetes mellitus diagnosed before or during the index pregnancy, pre-eclampsia, bleeding during pregnancy, caesarean delivery, sex, and birth weight (as a continuous variable)¹⁷ to evaluate confounding. To assess whether the associations of parental age and birth order with type 1 diabetes were the same for different levels of other exposures, age groups, sex, and calendar periods, we inspected the results of stratified analyses and tested the significance of the respective interaction terms.

	Results

Top Abstract Introduction Subjects and methods Results Discussion References

A total of 1 382 602 individuals contributed 8 166 731 person years under observation between 1989 and 1998. The mean time from birth to censoring or type 1 diabetes was 10.2 (SD=5.0) years, and the mean time under observation after 1 January 1989 was 5.9 (3.3) years. The mean age at diagnosis among the 1824 who developed type 1 diabetes was 8.6 (3.7) years.

We found a weak crude association between maternal age at delivery and incidence of type 1 diabetes. This became somewhat stronger and significant after year of birth and age group were adjusted for (table 1). There were no significant crude associations between birth order and incidence of type 1 diabetes (table 1). When both maternal age and birth order were included in the model, the rate ratios for maternal age were hardly changed but birth order became weakly negatively associated with type 1 diabetes (test for trend; P=0.06). However, after stratification, an interaction between maternal age at delivery and birth order appeared.

We found no association between maternal age at delivery and incidence of type 1 diabetes among firstborn children, but among second or later born children there was a positive association. The strength of the association increased with birth order (figure). In a regression model with maternal age and birth order entered as continuous variables, their multiplicative interaction was highly significant (P=0.004). Table 2 shows the estimated change in incidence for each five year increase in maternal age from regression models stratified by birth order. For instance, among fourthborn children each five year increase in maternal age was associated with a 43.2% increase (95% confidence interval 6.4% to 92.6%) in incidence of type 1 diabetes.

View larger version (17K): [in this window] [in a new window]   Maternal age at delivery and incidence of childhood onset type 1 diabetes by birth order. Data points are crude incidences in the maternal age groups <20, 20-24, 25-29, 30-34, and 35 years, with the points placed at the mean maternal age within each category. There were no diabetic people with certain rare combinations of maternal age and birth order (birth order 3 with maternal age <20 and birth order 4 with maternal age <25)

Compared with firstborn children, the incidence of type 1 diabetes was lower for second or later children when maternal age was low (figure). For instance, among children born to mothers aged 20-24 years at delivery, each increase in birth order was associated with a 17.9% reduction in the incidence of type 1 diabetes (3.2% to 30.4%). The association was weaker or non-existent when maternal age at delivery was 30 years or more (table 2).

There was a weak association between incidence of type 1 diabetes and paternal age at delivery. This was significant after adjustment for calendar period of birth and age group (table 1). The same patterns that were found for maternal age were also found for paternal age. However, the strength of the associations and the interaction with birth order were weaker than those for maternal age and the P values were larger (data not shown). After the effects of maternal age and birth order were adjusted for, there was no significant association between paternal age and type 1 diabetes. The effects of maternal age and birth order, however, were essentially unchanged after adjustment for paternal age.

The incidence of type 1 diabetes showed a small non-significant decrease during the study and was slightly higher among boys than girls. We previously found a weak but significant positive association between birth weight and incidence of type 1 diabetes,¹⁷ but in this study there was no association between the other potential confounding factors evaluated and incidence of type 1 diabetes. All main results were essentially independent of potential confounders evaluated. The associations were similar for each sex, year of birth, and age group, although the main effect of maternal age was slightly stronger in the earlier birth periods than later periods (P=0.03 for interaction between maternal age and birth period in model with only maternal age and period and in model including age group, birth order, and interaction term between maternal age and birth order).

	Discussion

Top Abstract Introduction Subjects and methods Results Discussion References

We found that the association between maternal age and childhood onset type 1 diabetes increased with rising birth order. Furthermore, any association between paternal age and type 1 diabetes was secondary to maternal age at delivery. The advantages of this study were the large sample size and the fact that the data were based on computerised registries with nearly complete coverage. We cannot, however, exclude confounding by unmeasured factors such as socioeconomic status of the children's families, maternal smoking during pregnancy, or whether delayed or early childbearing and number of births were intentional.

Other studies
A previous study which tested for an interaction between maternal age and birth order found no significant interaction.¹³ One problem with this study was that the authors used cohort analysis of data arising from a case-control study with siblings as controls.¹³ This probably biased their results.¹⁴ Furthermore, results from families with type 1 diabetes cannot readily be generalised to the total population. On the other hand, we cannot exclude the possibility that the effects of maternal age and birth order on risk of type 1 diabetes are different in different populations.

Birth order and paternal age
Associations of birth order and paternal age with risk of type 1 diabetes have been particularly inconsistent in previous studies. Both significantly higher risk ^{5 7} and lower risk¹¹ among firstborn children compared with second or later born children have been found, and there was evidence of heterogeneity between centres in two of these studies. ^{5 11} Some studies have not found any significant association between birth order and type 1 diabetes. ^{9 10 12} Paternal age at delivery has been negatively associated⁷ and positively associated with risk of type 1 diabetes.⁵ One study found a U shaped relation,⁸ but others found no significant association.¹² Our results indicate that these inconsistencies may be due to small sample sizes, no adjustment for relevant confounders, and lack of stratification by maternal age and birth order.

Possible explanations for the effect
Maternal age and parity are associated with various sociodemographic and biological factors. Delayed childbearing is associated with longer maternal education, complications in pregnancy, lower birth weight, fetal loss, and perinatal mortality. ^{19 20} Some of these associations depend on maternal parity.¹⁹ Pregnancies at older maternal ages are probably a mix of intentional and non-intentional pregnancies. Parous women who give birth at older ages may have had short or long intervals between previous pregnancies and may have had varying number of previous abortions or stillbirths.²⁰ Increasing age of the mother may be a marker for accumulated exposures such as infections or environmental toxins. Fetomaternal immune responses may also change with each pregnancy,²¹ and this could partly explain our results.