Birth weight and childhood onset type 1 diabetes: population based cohort study

Participants

Since the beginning of 1989 all newly diagnosed cases of type 1 diabetes diagnosed in children aged up to 15 years in Norway have been prospectively registered with a high level of ascertainment in the National Childhood Diabetes Registry.15 We designed a cohort study by record linkage of the Medical Birth Registry of Norway and the childhood diabetes registry through the unique personal identification number assigned to all residents of Norway. Out of 1863 cases of type 1 diabetes diagnosed between 1 January 1989 and 31 December 1998, 1824 were linked. All live births in Norway between 1974 and 1998 contributed time under observation from birth to diagnosis of type 1 diabetes, age 15 years, or 31 December 1998, whichever occurred first. As registration of cases started in 1989, the time under observation was counted only from 1 January 1989 for those born before this date. Deaths in the first year of life were censored, but we did not have information on deaths between age 1 and 15 years of age. A total of 1 382 602 individuals contributed 8 184 994 person years of observation between 1989 and 1998. The mean (SD) time from birth to censoring was 10.2 (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. The study was approved by the regional ethics committee and the national data inspectorate.

Data analysis

From the entire cohort we excluded from analysis 2468 individuals (0.2%) with missing data on birth weight, including three who developed type 1 diabetes. Gestational age was calculated from the first day of bleeding in the last menstruation. Those with missing data on gestational age (6.9%) were included. We calculated the number of incident cases (D_j) and person time under observation (T_j) in each exposure category (j) using DATAB in the EPICURE package, version 1.8w.16 Incidence rates were calculated as D_j/T_j. To assess a “dose-response” relation we plotted incidence rates against median birth weight in seven categories and against median gestational age in six categories and a category for missing data. We calculated confidence intervals for the incidence rates based on the Poisson assumption and using a log transformation. We used the AMFIT program of EPICURE to fit Poisson regression models, providing estimated rate ratios with 95% confidence intervals. We use the likelihood ratio test for birth weight entered as a continuous variable to test for trend.

We considered sex, maternal age at delivery (<25, 25-30, and ≥30 years), maternal parity (0, ≥1), caesarean section, maternal pre-eclampsia, attained age (0-4.9, 5-9.9, and 10-14.9 years), and calendar period of birth (1974-83, 1984-90, and 1991-8) as potential confounders. We did stratified analyses and included potential confounders in the regression models, both one by one and all covariates together to assess confounding. The effect of excluding multiple births and maternal diabetes mellitus (any type) diagnosed before or during the index pregnancy was also investigated. To test whether the association of birth weight and type 1 diabetes was homogenous in different age groups and in different categories of gestational age, we tested the significance of the respective interaction terms. We regarded a two sided P value less than 0.05 or a 95% confidence interval excluding the value 1.0 for the rate ratio as significant.

Comparison with previous studies

Our data support the results of some relatively large studies, 3^–5 20 as well as a study in experimental animals.21 In a large population based study from Sweden, however, Dahlquist et al found an association between birth weight relative to the average for any particular gestational age and type 1 diabetes but no significant association with birth weight per se.3 This apparent discrepancy is difficult to explain. Dahlquist et al adjusted for maternal smoking during pregnancy in a subset of their data, indicating that maternal smoking did not influence their result.3 Consistent with our observation, an increase in both mean birth weight and the incidence of type 1 diabetes has been observed in many Western countries in the past two to three decades. 22 23 However, the magnitude of the association between birth weight and type 1 diabetes was estimated as a 1.7% increase in incidence rate per 100 g increase in birth weight, which is probably not sufficient to explain these secular trends in type 1 diabetes. Detection of such a relatively weak association requires a large sample size. For instance, 2500 cases and 2500 controls are needed to attain 80% power to detect a significant association between birth weights of 2500 g and more versus less than 2500 g and type 1 diabetes in a case-control study if the true odds ratio is 1.5 and 5% of the population have birth weights less than 2500 g. This indicates that most of the published case-control studies may have had far too little power to detect such an association.

Possible explanations

Both birth weight and gestational age are influenced by several factors, such as maternal diabetes during pregnancy, nutrition, and parity, and may be viewed as markers of the combined effect of many factors. Potential factors explaining the present results would have to be associated with both increased birth weight and increased risk of type 1 diabetes over a wide range of birth weights. Maternal diabetes is one such candidate but is not likely to explain the observed association as the association persisted after exclusion of data from children whose mother had diabetes in pregnancy. As only a very small proportion of the children had a mother with diabetes diagnosed before or during pregnancy (0.5%), even a hypothetical underreporting of maternal diabetes is unlikely to explain the observed association. This indicates that other rare conditions probably do not explain the observed relation either. We cannot exclude the possibility that some genetic factors predispose to both higher birth weight and type 1 diabetes. A common genetic variant of the insulin gene region has been associated with lower birth weight in one study, but the effect was weak and significant only in a subgroup.24 The same variant is associated with an increased risk of type 1 diabetes.25 Allelic variation in the insulin gene region is therefore likely to diminish the association between birth weight and type 1 diabetes.

We can only speculate about biological mechanisms that could explain a hypothetical casual effect of birth weight on risk of type 1 diabetes. Insulin is the most important fetal growth factor late in pregnancy.26 In vitro studies have shown that pancreatic β cells actively secreting insulin express more antigens associated with diabetes27 and are more susceptible to insults by interleukin 128 compared with less active β cells. Increased growth in utero may therefore lead to an increased risk of later immune mediated destruction of the pancreatic β cells. Low birth weight has been associated with impaired cellular immune competence up to 5 years of age29 and with increased mortality from infectious diseases in the age group 1-15 years. 17 30 Perhaps an impaired cellular immune response makes children less prone to immune mediated destruction of the β cells? It has been speculated that the decrease in pancreatic β cell mass associated with reduced birth weight is relevant in the association between reduced birth weight and increased risk of type 2 diabetes.31 Our findings indicate that a possible reduction in pancreatic β cell mass associated with reduced birth weight is not relevant for type 1 diabetes.

Conclusion

We found a relatively weak but significant association between birth weight and increased risk of childhood onset type 1 diabetes consistent over a wide range of birth weights. It is possible that perinatal factors influence the risk type 1 diabetes, but the mechanisms are unknown.