Influence of maternal age at delivery and birth order on risk of type 1 diabetes in childhood: prospective population based family study

The Bart's-Oxford family study

The Bart's-Oxford (BOX) family study of childhood diabetes is a prospective population based study, which since 1985 has recruited more than 90% of the families of children who have developed type 1 diabetes under the age of 21 in the former Oxford Regional Health Authority area. This area has a population of 2.6 million, including 730 000 people under the age of 21 years. Ascertainment of new cases is more than 95%, and incidence up to age 15 over the period 1985–95 averaged 18.6 (95% confidence interval 17.4 to 19.8) cases per 100 000/year.14 Data, including dates of birth and history of diabetes, are collected during home visits, and further contact with the families is maintained by visits and telephone calls from the field workers, supplemented by annual postal questionnaires. More than 90% of those recruited remain under regular follow up. Diabetes is defined according to World Health Organization criteria, and cases of secondary diabetes or maturity onset diabetes of the young are excluded by examination of the clinical records.

Our analysis included 3221 offspring from 1375 families. A further 91 offspring from 49 families were excluded because the date of birth of the father or mother, or both, was unavailable. The median age of mothers at delivery of the included offspring was 27 years (range 16-45) and the median age of fathers was 29 years (range 16-74). The analysis included 1294 firstborn children, 1177 second children, 516 third children, 171 fourth children, 41 fifth children, and 22 children who were sixth or more in the birth order. Of the 1790 offspring who remained non-diabetic at the time of the analysis, 21 had been randomised into a diabetes intervention trial at median age 15.7 years (range 7-25).

Data analysis

We used Cox proportional hazards regression to assess disease free survival and hazard ratios for development of diabetes. Survival time was from birth to the date of diagnosis of the disease. Offspring who did not develop diabetes were censored at the date of last contact or date of entry into a diabetes intervention trial. The sex of each offspring, maternal and paternal ages at delivery, and order of birth of the offspring were included in the regression model. Only offspring of both parents were included in the analysis, but birth order was defined by using all the mother's offspring, including both full and half siblings of the study proband. Maternal and paternal ages were modelled in five year bands. The model was built in stages according to the general strategy for model selection suggested by Collett.15 We used univariate analysis to identify variables individually predictive of the development of diabetes (stage 1). Those found to be significant at the 5% level were then fitted together in the multivariate model (stage 2). Each variable was then excluded in turn to examine whether, in the presence of other variables, they continued to be predictive of diabetes. This procedure was continued until the exclusion of any variable would have resulted in a significant change in model fit (P<0.05, as defined by a change in −2logL). Variables excluded after stage 1 were then added to this multivariate analysis to assess whether, in the presence of other significant variables, they became significant predictors of diabetes (stage 3). We examined interactions between variables retained at the end of stage 3. Linearity of effect was examined by comparison of model fit when parental age and birth order were fitted as ordered categorical variables and as separate categories. The assumption of proportionality was evaluated graphically by using log cumulative hazard against log time plots. Standard errors have been adjusted for clustering within families. Data were analysed with STATA release 6.0. Results are expressed as relative risk (95% confidence interval).

The estimated effect of changes in distribution of maternal age on the incidence of diabetes was calculated by using demographic data for England and Wales (Office of Population Censuses and Surveys). An estimated cumulative incidence of diabetes up to age 15 of 2.5/1000 children was calculated on the basis of data on the incidence of childhood diabetes in the Oxford region in 1985.14 For each calendar year, the proportion of children born to mothers in each age group was multiplied by the percentage of children predicted to have developed diabetes by age 15, as estimated from the Cox regression model (assuming the child to be in the baseline category for all characteristics—that is, a firstborn female with a father aged less than 20 at delivery). This was summed to give a factor proportional to the cumulative incidence of diabetes in that annual birth cohort. This sum was then expressed relative to the estimate for the 1970 birth cohort.

Survival analysis

Of 3221 offspring included in this analysis, 1431 had developed diabetes at median age 10.5 years (range 0.4-28.5) and 1790 remained non-diabetic at median age 16.1 years (range 0-43). Overall, 57% (816 participants) of those with diabetes and 52% (931) of those who remained non-diabetic were male. Figure 1 shows the Kaplan-Meier survival curves for each five year band of maternal age at delivery. Cox regression analysis showed that maternal age, paternal age, birth order, and sex of offspring were independent determinants of risk. There was no evidence of significant non-linearity for parental age or birth order variables, and no significant first order interactions were found. The table shows the hazard ratios for these four variables in univariate and multivariate analyses. Risk increased with increasing maternal age at delivery and was less strongly associated with increasing paternal age. Overall risk of diabetes increased by 25% (95% confidence interval 17% to 34%) for each five year band of maternal age and by 9% (3% to 16%) for each five year increase in paternal age. Multivariate analysis showed that increasing birth order conferred some protection. The overall effect of increasing birth order was a 15% risk reduction (10% to 21%) per child. The risk in male offspring was 21% higher (8% to 35%) than in females. A small but significant inverse correlation was found between age at diagnosis and maternal age at delivery (r²=0.2, P<0.0001).

Fig 1

Diabetes free survival in offspring in relation to maternal age at delivery

• Download figure
• Open in new tab
• Download powerpoint

Estimated effect of temporal changes in maternal age distribution

Figure 2 shows the distribution of maternal age at delivery in England and Wales since 1960. The most noticeable changes were that the proportion of children born to mothers aged 20–24 years decreased from 37% to 19% between 1970 and 1996, whereas the proportion born to mothers aged 30–34 years increased from 15% to 28% during this time. Figure 3 shows the estimated risk for development of diabetes in successive annual birth cohorts compared with the 1970 cohort on the basis of the maternal age distribution and estimated disease free survival for each age group derived above. This analysis suggests that changes in maternal age distribution account for an 11.4% increase in cumulative incidence of diabetes between the 1970 and 1996 birth cohorts.

Fig 2

Maternal age at delivery in England and Wales 1970-9, 1980-9, and 1990-6

• Download figure
• Open in new tab
• Download powerpoint

Fig 3

Estimated risk of childhood diabetes in successive annual birth cohorts compared with 1970 cohort, calculated by using maternal age distribution and estimated disease free survival for each age group

• Download figure
• Open in new tab
• Download powerpoint

Conclusions

High maternal age—and possibly high paternal age—increases the risk of type 1 diabetes in offspring. Additionally, risk increases in a log linear fashion throughout the maternal age range. Higher maternal age at birth is associated with younger age of onset of type 1 diabetes in the child. Firstborn children are at greater risk than children of higher birth order. These observations, if confirmed, could account in part for the increasing incidence of diabetes and for the trend towards younger age at diagnosis. The fetal origins of type 2 diabetes, an aetiologically distinct disorder, are now well established.2 Fetal or neonatal influences could prove equally important in type 1 diabetes.