Testing the Accelerator Hypothesis: The relationship between body mass and age at diagnosis of type 1 diabetes

The study was conducted retrospectively on 94 children (49 boys and 45 girls) aged 1–16 years presenting for management of acute-onset type 1 diabetes to the James Cook University Hospital, Middlesbrough, U.K., during the period 1980–2000. All but two patients were white, and twin births (n = 4) were excluded. The criteria for inclusion were those diagnostic of type 1 diabetes: high blood glucose, presence of ketoacidosis, and/or requirement of insulin. Islet-related antibodies were not tested. The relationships between fatness and age at diagnosis of diabetes were examined in the context of birth weight, weight change since birth, weight at diagnosis, and BMI at diagnosis, all expressed as the standard deviation score (SDS) adduced from the 1990 U.K. growth standards (33,34). Measurements were made by a single clinic team on cantilever scales and a Harpenden stadiometer. The children were reviewed weekly after presentation for 6 weeks, and “at diagnosis” relates to height and weight recorded at week 6, when rehydration is likely to have been complete. We also compared the data “at diagnosis” with BMI SDS recorded 12 months later, when stable growth trajectories are likely to have been reestablished. We also recorded the duration of symptoms and identified those presenting with ketoacidosis.

The study had sufficient power to show a correlation of r = 0.29 with 80% power and 95% certainty. Proportions were compared using the χ² test. Simple correlations were calculated between age at diagnosis and birth weight SDS, weight change SDS, weight SDS, height SDS, and BMI SDS at diagnosis and 12 months later. Stepwise regression was performed to determine which combination of independent variables best predicted age at diagnosis (the dependent variable). Partial correlations (correlations controlled for the effect of the model variables) were also calculated for variables not selected in the regression model. The children were grouped into age-at-diagnosis quartiles. Mean birth weight SDS, weight SDS change since birth, weight SDS, and BMI SDS at diagnosis were calculated for each age-at-diagnosis quartile and were compared using ANOVA. Correlations and comparisons were also conducted on BMI recorded 12 months after diagnosis.

All ages at onset from 1 to 16 years were represented in this study (Fig. 1). A total of 22 of the 94 children were ketoacidotic at presentation. The proportion of diabetic ketoacidosis in the youngest quartile (10 of 23) was significantly higher than the second (3 of 25), third (4 of 23), and oldest (5 of 23) quartiles (P < 0.05). Importantly, there were no differences between the quartiles in the duration of symptoms recorded and year of diagnosis (P = 0.95).

The mean weight SDS at birth, weight SDS and BMI SDS at diagnosis, and weight SDS change since birth according to age-at-diagnosis quartile are shown in Table 1. Change in weight SDS since birth and BMI SDS at diagnosis were each greater in those who developed diabetes at a younger age. Putting the two observations together, it might be concluded that those who gained the most weight developed diabetes at the youngest ages. However, weight SDS change was highly correlated with weight SDS at diagnosis (r = 0.73, P < 0.001); those who gained the most weight also ended up among the heaviest.

Table 2 shows the correlations between age at diagnosis, birth weight SDS, weight SDS change since birth, weight SDS at diagnosis, height SDS at diagnosis, and BMI SDS at diagnosis and 1 year later. There were inverse and statistically significant relationships for weight SDS change since birth, weight SDS at diagnosis, and BMI SDS at diagnosis but not birth weight. The inverse relationship between age at diagnosis and BMI SDS remained when the boys and girls were analyzed separately (boys r = −0.44, P = 0.002; girls r = −0.33, P = 0.025). Stepwise regression selected BMI SDS at diagnosis only for the group as a whole (r = −0.43, P < 0.001). The multiple regression analysis shown in Table 3 indicates that the other variables were all either co-correlates of BMI SDS or less predictive of age at diagnosis. None of these variables, singly or in combination, better predicted age at diagnosis, although the partial correlation of birth weight SDS with age at diagnosis, controlling for BMI SDS, became closer to significance (r = 0.20, P = 0.06).

The boys presented at a significantly younger age than the girls (6.74 years in boys, 8.32 years in girls; P = 0.049). However, the boys were significantly fatter than the girls at diagnosis (BMI SDS 0.56 in boys, −0.08 in girls; P = 0.006), and the difference in age at presentation was lost when corrected for BMI SDS (boys 7.12, girls 7.90, P = 0.31).

We compared the data at diagnosis with recordings of BMI made 12 months later. The correlation between the BMI SDS of individuals on the two occasions was strong (r = 0.76, P < 0.001), despite a substantial increase in mean BMI SDS from +0.18 at diagnosis to +0.94 12 months later (P < 0.001). Importantly, we also looked at the weight gain according to age to infer whether the younger children may simply have lost less weight before diagnosis. The change in BMI SDS between diagnosis and 12 months later was not related to the age of the subject, whether analyzed by age-at-diagnosis quartile (P = 0.16) or by regression (r = 0.16, P = 0.14). The gradient from regression was small: an increase of 0.03 SDS in the difference between the BMI at diagnosis and the BMI 12 months later for every 1-year increase in age at diagnosis. Finally, the relationship between age at diagnosis and BMI SDS at 12 months remained unchanged from what it had been at 6 weeks (r = −0.40, P < 0.001).

The data suggest that the age at diagnosis of type 1 diabetes is a function of body mass. Three lines of evidence presented here lead to that conclusion. First, weight SDS and BMI SDS, whether analyzed for differences in distribution according to age-at-onset quartile, or in a simple regression, are significantly related to age at presentation. Second, the relationship between age at diagnosis and weight SDS change since birth was almost as strong. All children gain weight, but weight SDS change is a measure of the excess weight gained or centiles crossed (21). Third, the boys were significantly fatter than the girls and presented with diabetes at a significantly younger age. The difference in age at diagnosis disappeared, however, when adjusted for body weight, suggesting that weight or something related to it was the factor responsible.

However, the study has a number of weaknesses. First, there was a lack of genetic and serological data with which to type the children immunogenetically. Nevertheless, we believe the children were likely to have been type 1 diabetic (autoimmune) for three reasons. All required insulin to normalize blood sugar levels, none of the children were particularly obese at presentation (maximum BMI SDS 2.73), and type 2 diabetes presenting before 16 years of age has been rare in the U.K. Second, it is possible that greater weight loss because of a longer prodrome in the older child could be misconstrued as weight-driven acceleration of the disease in those of younger onset. Type 1 diabetes is certainly associated with weight loss, and the prodrome is often more aggressive and of shorter duration in the younger child. Predisease growth charts would have provided the best evidence of true growth trajectory but were not available. Instead, we compared the data at diagnosis with recordings made 12 months later, when it is likely that metabolic health would have been restored and growth trajectories reestablished. There was a strong correlation in BMI SDS between diagnosis and 12 months later, despite a substantial mean weight adjustment. More importantly, age had no significant impact on the adjustment in BMI SDS, suggesting that, had a longer prodrome been associated with diabetes in older children, it was not responsible for the younger onset of diabetes in heavier children. Indeed, age at diagnosis was correlated as well with BMI SDS at 12 months as it was at diagnosis, suggesting that, although there are substantial weight changes around diabetes onset, children tend to retain their growth trajectories relative to each other, much as they do after chronic illness. It seems likely that BMI SDS after diagnosis is a sufficiently reliable index of premorbid body mass for group analysis. We also lacked direct measures of insulin resistance at and before development of diabetes. Insulin resistance is a well-known feature of type 1 diabetes at presentation (4–8), but the premorbid measures related to body mass, i.e., 1 year before onset, would provide the strongest evidence of cause and effect. Finally, the numbers in this study were small by epidemiological standards but nevertheless had sufficient power to reveal with clear statistical significance the relationships on which its conclusions were based.

Earlier studies reporting a higher risk of type 1 diabetes in children who were heavier as toddlers needed a control group with which to compare the diabetic children (22–26). Although such reports predicted greater risk among heavier infants, the Accelerator Hypothesis goes further and predicts earlier onset, without necessarily a change in lifetime risk (35). Here, the analysis is one of intragroup correlation, rather than intergroup comparison, and a control group was not needed.

Contrary to the present series, many studies have recorded an earlier diagnosis of type 1 diabetes in girls. However, there is no inconsistency here, as girls are generally fatter (and more insulin-resistant) than boys (21). Puberty is associated with a sharp increase in insulin resistance, occurs earlier in girls than in boys, and has to be considered in the relationship between body mass and age at onset of diabetes. Tanner staging for pubertal development is not routine in the diabetes clinic, but sexual maturity is unlikely to have had a material impact on the question we addressed in this study. We used sex-specific SDS charts for height and weight with which to adjust for the differences between boys and girls, and the youngest (prepubertal) children contributed as much to the relationship between body mass and age at onset as the older (pubertal) ones. Importantly, the relationship remained true for both boys and girls when analyzed separately, and BMI, which can overestimate fatness in younger (shorter) children (36), did not seem to be an artifact because the relationship applied as much to simple weight SDS as to BMI SDS.

We did not find this cohort of diabetic children to be systematically heavier at birth, as might have been expected from the reports referred to above where type 1 diabetic people were heavier than their peers as young children. Only the birth weights of those in age-at-diagnosis quartiles 3 and 4, in whom onset of diabetes was least accelerated, lay above the mean for the whole group. Indeed, the younger the age at diagnosis, the lower the birth weight. Those in quartiles 1 and 2, in whom diabetes was most accelerated, lay at or below mean birth weight, not above. The combined influence on insulin resistance of lower birth weight and rapid weight gain in early childhood has been attributed to so-called “catch-up” growth (16,19,20,37). However, as we have shown previously in the EarlyBird study (21), weight “catch-up” is closely correlated with current weight and probably not, as such, a mechanism for insulin resistance. Indeed, the children from quartiles 1 and 2 in the present study were close to mean birth weight, and it is arguable that the group in whom the onset of diabetes was most accelerated were not so much children of low birth weight as those who had gained most during their early years to reach the highest BMI.

Diabetes is the outcome of a process; it is not the process itself. The process is one of progressive β-cell loss that may take years. Although insulin resistance is widely regarded to be the factor responsible in pre–type 2 diabetes, epidemiologists have until now viewed pre-type 1 diabetes as a separate process and have searched in vain for separate immunologic triggers. Among those suggested are viruses, dietary nitrosamines, and cow’s milk protein (38). During the ∼35 years since diabetes was first perceived histologically to be of two types, the one metabolic and the other autoimmune (39), the prevalence of obesity in the industrialized world has more than doubled (40). So has that of type 2 and type 1 diabetes (41). Indeed, wherever in the world there has been an increase in type 2 diabetes, there has been a parallel increase in childhood type 1 diabetes (42).

The data reported here are not proof that the increasing prevalence of obesity in childhood is the cause of the increasing incidence of type 1 diabetes. It is indeed possible that the increase in weight is a maladaptation to pre-diabetes accelerated by another cause. The data are only consistent with a hypothesis that will ultimately best be tested by intervention with either weight reduction or insulin-sensitizing medication. It is now possible to predict type 1 diabetes with some precision (43), and given the encouraging data from lifestyle prevention studies in pre-type 2 diabetes (44,45), it may arguably be worth considering lifestyle intervention trials to reduce insulin resistance in people with pre-type 1 diabetes before embarking on further immunomodulatory drug trials. Insulin resistance is the predictable and inevitable result of weight increase to which whole populations are now subject, and the data reported here lend support to the hypothesis that type 1 and type 2 diabetes are one and the same disorder of insulin resistance set against different genetic backgrounds.

We thank Roche Products, Smith’s Charity (London), the Child Growth Foundation, and the London Law Trust for their generous support.