Childhood Size and Life Course Weight Characteristics in Association With the Risk of Incident Type 2 Diabetes

The Nurses' Health Study II (NHSII) is an ongoing prospective study of U.S. female nurses aged 25–42 years. Follow-up is conducted using biennial questionnaires from 1989 to 2005. A total of 109,172 participants remained after exclusion for diagnosis of any type of diabetes, cancer, or cardiovascular disease at baseline (2%) or for missing information on childhood body shape (2.5%), BMI at age 18 years (0.7%), or baseline BMI (0.3%). This study was approved by the institutional review board of the Partners Health Care System (Boston, MA).

At baseline, body fatness at ages 5, 10, and 20 years of age was assessed by recall of somatotypes or 9-level pictorial diagrams developed by Stunkard et al. (9) representing sizes ranging from extreme thinness (category 1) to obesity (category 9). The use of recalled somatotypes has been validated in both older (10) and younger (11) women by comparison with childhood records. Somatotypes at ages 5, 10, and 20 years correlated moderately with recorded BMI (r = 0.60, 0.65, and 0.66, respectively) (10). The validity did not differ by adult BMI at the time of report (10). Overweight was defined as somatotype ≥5. (12).

Weight at age 18 and adult height and weight were self-reported. Weight change was the difference between weight at age 18 years and weight at baseline in 1989. The correlation between recalled weight at age 18 and documented weight from college or nursing school records was 0.84, and the correlation between self-reported and technician-measured adult weight was 0.96 (13).

A supplementary questionnaire was mailed to confirm the self-report of diabetes diagnosis and to distinguish different types of diabetes (14). The National Diabetes Data Group diagnostic criteria were used for cases reported through 1997 and required confirmation of at least one of the following: 1) one or more symptoms (weight loss, hunger, thirst, or polyuria) and elevated glucose (fasting ≥7.8 mmol/l [140 mg/dl] or random plasma or 2-h glucose ≥11.1 mmol/l [200 mg/dl]), 2) no symptoms but occurrence of elevated plasma glucose as described above on at least two different occasions, or 3) use of insulin or oral hypoglycemic medication (15). For cases occurring after 1998, the cutoff of fasting plasma glucose was changed to 7.0 mmol/l (126 mg/dl) in accordance with revised criteria (16). We excluded women classified as having only gestational diabetes mellitus or type 1 diabetes. In the Nurses' Health Study, the supplemental questionnaire was highly reliable regarding diabetes diagnosis (14). In a random sample of 84 women classified as being a case subject, medical records were available for 62 of these women and an endocrinologist confirmed the diagnosis in 61 women (98%) (14).

Age was calculated as months from the reported birth date to date of questionnaire return. Race, smoking status, birth weight, prematurity, multiple gestation birth, age of menarche, being breast-fed, alcohol consumption, and family history of diabetes were self-reported at baseline or in 1991. Parity and age at first birth were measured biennially. Physical activity, in MET units, was derived from the average time spent in certain activities (e.g., jogging or running) in 1989, 1991, 1997, and 2001.

Differences in participant characteristics by childhood somatotype at age 10 years were compared using χ² or linear regression. Person-years were calculated based on date of return to date of diagnosis, death, or 1 July 2005, whichever came first. Multivariate Cox proportional hazard models were used to estimate the relative risk (RR) for the associations of somatotypes at ages 5, 10, and 20 years and BMI at age 18 years with type 2 diabetes risk. Somatotypes were categorized by 1 to ≥6. We used category 2 as the reference because most women had reported this category at both ages 5 and 10 years. We collapsed the uppermost categories because few women reported those categories.

In one multivariate model, estimates were adjusted for age (continuous), race (African American, Hispanic, Asian, or white), parity (0, 1–2, or ≥3) in combination with age at first birth (>24 years), family history of diabetes (maternal, paternal, or both), smoking (current, past, or never), and physical activity (quintiles). In a second model among participants not missing birth weight information (n = 87,349, case subjects = 2,771), estimates were adjusted for characteristics influencing childhood size including age, race, family history of diabetes, birth weight (<5.5, 5.5–6.9, 7.0–8.4, 8.5–9.9, or ≥10 pounds), prematurity or multiple gestation birth status of the woman (yes or no), and age of menarche (<12, 12, 13, 14, or >14 years). To test for significant trends, linear models were fitted using the median values of each category of exposure (e.g., 17.0, 19.0, 21.0, 23.5, 28.0, and 35.0 for BMI at age 18 years).

The differences in somatotype categories (between ages 5 and 10 and between ages 10 and 20) or in weight (between age 18 years and baseline) were used to assess whether change in size was associated with diabetes development. Models were additionally adjusted for starting size (e.g., adjustment for age 5 somatotype in evaluating the difference between ages 5 and 10), because where a woman began in size may reflect weight gain or loss (i.e., heavier women have the possibility to lose more weight).

The cumulative effect of overweight across the life course before reported diabetes was evaluated using the combination of somatotype ≥5 at age 10, BMI ≥25 kg/m² at age 18, and BMI ≥25 kg/m² at baseline (i.e., to represent adulthood). The reference group comprised women who reported no overweight at any of those time points. Analyses were performed using SAS (version 8.2; SAS Institute, Cary, NC).

There were 3,307 incident cases of type 2 diabetes over 16 years of follow-up, for an incidence rate of 197 cases per 100,000 person-years. Table 1 shows the characteristics for all women and by reported childhood somatotype at age 10 years categorized from 1 to ≥6. Women with somatotype 1 at age 10 years were more likely to be of minority race, to have birth weight <5.5 pounds, to have been born prematurely, and to have had a later age at menarche; whereas women with somatotype ≥6 were more likely to have birth weight >10 pounds, to actively smoke, to be nulliparous, and to have had a positive parental history of diabetes. Similar associations were found at age 5 years. Somatotypes at ages 5, 10, and 20 years were associated with each other. Of the women reporting somatotype ≥6 at age 10 years, 41% reported somatotype ≥6 at age 5 years, and 28% at age 20 years. Weight and BMI were also positively associated with somatotype; a 1–2 kg/m² increase in adolescent or adult BMI was observed per increase in somatotype category.

Somatotypes at ages 5, 10, and 20 years and BMI at age 18 years (Table 2) were all significantly and positively associated with diabetes risk (all P_trend < 0.0001). Compared with having a somatotype of 2, the RR of diabetes associated with a somatotype ≥6 at age 5 was 2.19 (95% CI 1.79–2.67), at age 10 years was 2.57 (2.20–3.01), and at age 20 years was 5.67 (4.92–6.54). The RR among women who reported a BMI at age 18 >30 kg/m² was 8.72 (7.58–10.02) compared with women with a BMI of 18–19 kg/m². Among a subgroup not missing birth weight information (n = 87,175), adjustment for childhood factors including birth weight did not substantially change these estimates.

Type 2 diabetes risk increased among women who reported increases in size whether by somatotype between ages 5 and 10 and between ages 10 and 20 or by weight gain since age 18 (Table 3). Compared with women reporting no change in somatotype, one or more unit increases in somatotype at these ages were associated with approximately twice the risk, whereas decreases in somatotype at these ages were associated with reduced risk. These associations were strengthened after adjustment for earlier somatotype. Weight gain since age 18, which is the difference between baseline weight and weight at age 18, was also significantly associated with risk of diabetes. Compared with women who had little change in weight (i.e., ±4.9 pounds), even a weight gain of 5–8 pounds doubled risks, whereas weight gain of >25 pounds increased risk by >20 times. Weight loss was associated with increased risk until after adjustment for BMI at age 18 to account for greater weight loss being associated with larger adolescent body size. In analyses stratified by BMI at age 18 years, weight loss of ≥10 pounds was significantly associated with a reduced risk of diabetes among women who were overweight (RR 0.45 [95% CI 0.22–0.91]) or obese (0.45 [0.28–0.72]) in adolescence but not among those who were lean (1.72 [0.76–3.90]).

The prevalences of overweight by somatotype ≥5 at ages 5, 10, and 20 years were 7, 12, and 11%, respectively. Most of the cases (83%) in the cohort occurred among women who were overweight (BMI ≥25 kg/m²) at baseline with an incidence rate of 562 per 100,000 person-years. Among these women, the prevalences of overweight by somatotype ≥5 were 12, 21, and 23%, respectively.

To evaluate the cumulative effect of being overweight across the life course, women were jointly categorized as being overweight using somatotype at age 10, BMI at age 18 years, and BMI at baseline (with mean age of 34 years) (Fig. 1). Women who were overweight only as an adult (BMI >25 kg/m²) had an adjusted RR of 8.23 (95% CI 7.41–9.15) compared with women who were never overweight at any age. The adjusted RR increased to 15.10 (13.21–17.26) for women who were also overweight at age 10 (somatotype ≥5) and age 18 (BMI >25 kg/m²). However, women who were overweight at age 10 but came off the trajectory and became lean in adulthood (i.e., not being overweight in adulthood) did not have a significantly increased risk of diabetes (1.02, 0.74–1.40).

Among 109,172 women followed for >16 years, somatotypes at ages 5 and 10 years were positively associated with the risk of incident type 2 diabetes after adjustment for major risk factors. However, women who were overweight at age 10 but were lean in adulthood did not have an increased risk of diabetes associated with childhood overweight, underscoring the importance of continued efforts to control adiposity among overweight children.

Limited studies have investigated the long-term consequences of childhood overweight and the risk of type 2 diabetes in adulthood, with inferences hindered by small numbers of case subjects and/or lack of differentiation of diabetes type (2,,,,,–8). In addition, some studies included case subjects with young age at onset (<30 years old) (3,5,7). One study using data based on records (185 cases) showed that the cumulative incidence of diabetes was doubled (from 4.2 to 8.4%) among women in the largest BMI group at age 11 years (>17.4 kg/m²) compared with those in the smallest BMI group (≤15.3 kg/m²) (4). We found a similar increase in risk using recalled somatotypes. In contrast, a study of older French women using recalled body shape, found an association between childhood thinness and diabetes risk (8). However, this birth cohort vastly differed from the NHSII as demonstrated by >60% of their case subjects having reported extreme leanness (somatotype 1) at age 8 years, possibly due to their nutritional status having been affected by World War II (8). Thus, our findings among a more recent birth cohort may be more relevant.

In addition to absolute size, longitudinal changes through different ages are also relevant to the risk of diabetes (17,18). Under normal development, infants lose weight after 6 months of age and continue doing so until ∼5 years of age when their adiposity rebounds (18). Although we were unable to assess age at adiposity rebound, earlier rebound has been associated with increased risk of type 2 diabetes (17), and there is some suggestion that age at rebound matters more than the absolute size of the child at any point in time because it corresponds to weight gain. Our findings of increased risk associated with increases in somatotype and weight gain support the importance of change in size in addition to absolute size. These findings offer support for continued weight reduction efforts across the life span.

Moreover, in analyses of the cumulative effect of overweight through childhood, adolescence, and adulthood, the risk for type 2 diabetes was greatest among women reporting overweight by all three measures, even though the difference was marginal when compared with adolescent overweight. However, women who became lean in adulthood did not have an increased risk of diabetes associated with childhood overweight. Similar observations have been made from other studies of youth overweight and type 2 diabetes or related traits, with findings becoming nonsignificant after accounting for adult BMI (6,19,20). Tracking of metabolic risk factors (e.g., HDL and triglycerides) has also been observed after 21 years of follow-up from childhood (21), supporting the fact that cumulative overweight is linked with prolonged exposure to metabolic irregularities, putting children on the path to β-cell dysfunction earlier. It should be noted that although there is tracking of overweight to obesity from childhood to adulthood, the trajectory is not fixed from youth (18,22). It has been shown that ∼10–30% of overweight children do not go on to be overweight as adults with conflicting evidence as to whether there is greater tracking in girls than in boys (22). Conversely, the majority of adults who are overweight were not in childhood (22) as confirmed here with only 12–20% of the overweight women reporting childhood overweight.

Our study had some limitations. Although experts have recommended BMI for the measurement of childhood size (18,23), we used recalled somatotype. Misclassification of childhood size is inevitable. However, two previous studies have shown that the accuracy of recall between childhood somatotypes and recorded childhood BMI did not differ by adult BMI (10,11). The misclassification of childhood size is thus more likely to be nondifferential, which cannot explain the observed positive association between childhood size and type 2 diabetes risk. In addition, our findings were in agreement with those of other studies that used childhood measures of weight and height (4,20). The generalizability of our findings may be limited to white women who comprised >95% of our study population. More research is needed in minority populations who have higher rates of overweight and diabetes (1,24). Last, we cannot rule out the possibility of residual confounding by unmeasured confounders because of the observational nature of the study.

As for strengths, the NHSII included a large number of women (>3,000 case subjects) and adjusted for many risk factors. Follow-up for previous studies required recontacting participants in adulthood with low to moderate success rates (20–60%) (5,7,19,20), whereas we evaluated somatotypes at baseline and >90% of the women remained for follow-up.

In summary, our findings demonstrate that the importance of childhood overweight stems largely from adult overweight. Women who do not continue to be overweight in adulthood do not have increased risks. It remains important then to promote lifestyle changes from youth so that the adverse trajectory could be avoided. Multiple interventions to address childhood overweight have been suggested (23), but these remain to be fully tested.

This study was funded by research grants CA50385 and DK58845 from the National Institutes of Health. E.H.Y., C.Z., and G.M.B.L. were supported by the Intramural Research Program of the Eunice Kennedy Shriver National Institute of Child Health and Human Development, National Institutes of Health. Funding sources had no role in data collection, analysis, interpretation, or article submission, and researchers acted independently from funders.

No potential conflicts of interest relevant to this article were reported.

Parts of this study were presented in abstract form at the 27th annual scientific meeting of The Obesity Society, Washington, D.C., 24–28 October 2009.