Weight tracking in childhood and adolescence and type 2 diabetes risk

Abstract

Aims:

To examine the associations of average weight and weight velocity in three growth periods from birth through adolescence with type 2 diabetes incidence.

Methods:

Child participants representing three growth periods- pre-adolescence (birth to ~8 years), early adolescence (~8 to ~13 years), and late adolescence (~13 to ~18 years) -were selected from a 43-year longitudinal study of American Indians. Age-sex-height-standardized average weight z score and weight z score velocity (change/year) were computed in each period. Participants were followed for up to 25 years from the end of each growth period until they developed diabetes. Associations of average weight z score or weight z score velocity with diabetes incidence were determined with sex-, birth date- and maternal diabetes-adjusted Poisson regression models.

Results:

Among 2,100 participants representing the pre-adolescence growth period, 1,558 representing the early adolescence period, and 1,418 representing the late adolescence period, there were 290, 315, and 380 incident diabetes cases, respectively. During the first 10 years of follow-up, diabetes incidence rate ratio was 1.72 (95% CI 1.40–2.11)/standard deviation (SD) of log₁₀weight z score mean in pre-adolescence, 2.09 (1.68–2.60)/SD of log₁₀weight z score mean in early adolescence, and 1.85 (1.58–2.17)/SD of log₁₀weight z score mean in late adolescence. Diabetes incidence rate ratio was 1.79 (1.49–2.17)/SD of log₁₀weight z score velocity pre-adolescence, 1.13 (0.91–1.41)/SD log₁₀weight z score velocity in early adolescence, and 1.29 (1.09–1.51)/SD log₁₀weight z score velocity in late adolescence. There were strong correlations in the average weight z scores and weak correlations in the weight z score velocities between successive periods.

Conclusions:

Higher weight and accelerated weight gain in all growth periods associates with increased type 2 diabetes risk. Importantly, higher weight and greater weight velocity pre-adolescence jointly associate with the highest type 2 diabetes risk.

Introduction

The incidence and prevalence of obesity are increasing among children and adolescents in the United States^1,2 and coincide with an increasing type 2 diabetes incidence among youth and young adults.^3,4 Excessive body weight in childhood or adolescence often persists into adulthood,⁵ thereby increasing the risk of early-onset type 2 diabetes.^6,7 Given the significant long-term complications and mortality associated with early-onset type 2 diabetes,^8,9 careful monitoring of growth in childhood and adolescence is important. However, available evidence on the impact of variability in weight gain at critical growth periods in childhood and adolescence on type 2 diabetes development is limited by a paucity of serial anthropometric data for weight tracking from early childhood or of prospective biochemical data for ascertaining type 2 diabetes incidence.^6,10–19

Pima Indians living in the Southwestern United States have a high prevalence of obesity and type 2 diabetes, especially among children and young adults,^20,21 with considerable long-term burden associated with these comorbidities.²² In the present study, we use longitudinal data obtained from Pima children and adults to identify critical growth periods during childhood and adolescence that are associated with type 2 diabetes incidence. We hypothesized that both higher average weight and greater weight velocity in childhood and adolescence associate with greater type 2 diabetes risk, and that these associations differ by the growth period.

Methods

Study design and participants

A longitudinal open cohort study of diabetes and related complications was conducted from 1965 through 2007 among residents of the Gila River Indian Community (GRIC) in Arizona aged ≥ 5 years. Diabetes in this community is thought to be exclusively type 2, as available data from both child and adult residents of this community with diabetes suggest no autoimmunity (i.e. islet cell antibodies including GAD65Ab) typical of type 1 diabetes or insulin dependence even in those with very young onset.^23,24

Participants were invited to undertake standardized research examinations approximately every two years. Although participants in the longitudinal study were aged ≥5 years, data on events preceding participation in the longitudinal study (e.g. weight at birth) were obtained from secondary sources (e.g. medical records and birth certificates) when available. Because participants entered the biennial study at different ages and the time points at which research examinations occurred were not uniform in all participants, there were limited serial data in childhood. Therefore, to maximize the available data, the present study used data obtained at four time points in childhood (Fig. 1): birth, age ~8 years (range 7–9 years at the last birthday, median 8.4), age ~13 years (range 12–14, median 13.2), and age ~18 years (range 17–19, median 18.3). If a participant had more than one examination in the window defining a time point, only the examination closest to the midpoint age was used. These time points were used to define three sequential growth periods: pre-adolescence (birth to ~8 years), early adolescence (~8 to ~13 years), and late adolescence (~13 to ~18 years; Fig. 1).

Fig. 1.

Study design and calculation of average weight z scores and weight velocity z scores. Age ~8 years ranged from 7–9, median 8.4; Age ~13 years ranged from 12–14, median 13.2; Age ~18 years ranged from 17–19, median 18.3; wtz is the log₁₀weight z score at a time point.

^a Given that analyses were conducted separately for each growth period, a participant representing more than one growth period could be followed for varying lengths of time as the start point for follow-up varies by growth period.

Although the period of adolescence is typically defined as being between 10–19 years of age, this growth period varies across populations and contexts.²⁵ Data are limited on age at puberty among American Indians. Therefore, we selected the age range of 7–9 years as the endpoint for pre-adolescence given that the mean age at menarche among Pima Indian girls is 12 years²⁶ and that menarche is 2–3 years after the onset of the first signs of puberty. Coincidentally, most of the available data points in early childhood were concentrated in this age range. We selected age ranges 12–14 years and 17–19 years as end points for early and late adolescence respectively to maximize available data points in adolescence while reflecting the variability in growth at this stage of childhood.

Participants were included in the analysis for a growth period if they had measurements of height and weight at the beginning and the end of the period; if they did not have diabetes during the period (ascertained using OGTTs at the research examinations including examinations defining the growth periods); and if they had at least one research examination after the period. Participants could be included in the analysis for more than one growth period. The study was approved by the Institutional Review Board of the National Institute of Diabetes and Digestive and Kidney Diseases. Minors (aged <18 years) gave assent, and written informed consent was obtained from their parents. Participants aged ≥18 years provided written informed consent.

Data collection

At each research examination, height was measured to the nearest centimeter and weight to the nearest kilogram while participants were dressed in light clothing without shoes. Birthweights were obtained from birth certificates and review of medical records. A 75-gram oral glucose tolerance test (OGTT) was performed at each examination. Diabetes was diagnosed by World Health Organization criteria when a 2-hour plasma glucose concentration was ≥11.1 mmol/l (200 mg/dl),²⁷ or when a diagnosis was documented during routine clinical care between research examinations. More recently adopted tests for diabetes, such as the fasting plasma glucose and glycated hemoglobin, were not used in this analysis because they were not used throughout the longitudinal study that began in 1965. The same methods were used to ascertain maternal diabetes prior to the birth of the participant, i.e. diagnosis was by OGTT or documented clinical diagnosis rather than by self-report. A subgroup of 2,630 participants was assessed for functional mutation of melanocortin-4 receptor (MC4R), a common genetic cause of both obesity and type 2 diabetes in childhood and adolescence in this study population.²⁶ The MC4R variants have a larger effect on obesity traits among Pima Indians than other known genetic variants associated with obesity in childhood and adolescence.

Statistical analysis

Characteristics of participants were summarized with descriptive statistics, including median, 25th (Q1) and 75th (Q3) centiles for non-normally distributed variables. The objective of this analysis was to estimate the associations with diabetes incidence of two growth parameters at different points in childhood - standardized weight at a point and standardized weight velocity at an interval leading to that point. To achieve this, we modeled weight-for-height rather than body mass index (BMI). This is because modeled weight-for-height is more appropriate for modeling growth in children than BMI as the latter leaves residual confounding by height.²⁸ Therefore, weights in childhood and adolescence were standardized to the study population to obtain z scores by (1) identifying sex-specific age- and height-adjusted regression models that provide the best fit for weight, i.e. log₁₀weight=β₀ + β₁age + β₂height for males and females; and (2) standardizing residuals from each model to a mean=0 and standard deviation (SD)=1. Because data on gestational age and length at birth were unavailable, birthweight was standardized to have a mean=0 and SD=1, separately by sex.

Standardized measures of weight at a point and of weight velocity during an interval leading to that point are strongly correlated thereby limiting the ability to clearly delineate their individual and joint associations. Given that this correlation can be considerably minimized by using the average and the difference of two scores,^29,30 we analyzed both the average weight z score and the weight z score velocity of each participant for each growth period (Fig. 1). The average weight z score was the z score of the mean of log₁₀weight z scores at the points defining a growth period, while the weight z score velocity was the z score of the difference in log₁₀weight z scores divided by the length (in years) of the growth period (Fig. 1). Growth was modeled using this approach because of limited data at regular age intervals in childhood in this study.

Although there is some correlation introduced by using some of the same measures to calculate the average weight z score or weight velocity z score velocity for the growth periods (Fig. 1), these overlapping measures were not included in the same model. In other words, analyses of diabetes incidence were conducted separately for each growth period, and thus, separately for each group of participants representing the growth periods. Each person was followed from the examination that defined the end of the corresponding growth period until diabetes incidence or the last research examination, whichever came first (Fig. 1). Although some participants had a follow-up period >25 years, analyses were restricted to up to 25 years of follow-up because of limited (<1000 person-years) follow-up afterwards which may reduce the precision of the incidence rate estimates; and to ensure better balance in the length of follow-up between groups than when follow-up is extended beyond 25 years. Given that analyses were conducted separately for each group, a participant representing more than one group could be followed for varying lengths of time as the start point for follow-up varied by growth period.

Diabetes incidence rate was calculated per 1000 person-years of follow-up. Poisson regression was used to model diabetes incidence rate ratios (IRRs) for the average weight z score or weight z score velocity in a growth period with the offset term, i.e. log (person-years), accounting for different length of follow-up. Models were adjusted for sex, birth date and maternal diabetes. For each growth period, average weight z score and weight z score velocity were analyzed in separate models (to estimate their individual associations) and then jointly in the same model (to estimate their joint associations). In some models (particularly those estimating the IRR for average weight z scores), associations of the average weight z score or weight z score velocity with diabetes IRR, were non-proportional (i.e. non-constant) over time. Therefore, time was split into 5-year intervals follow-up and IRRs were estimated with increasingly longer 5-year intervals follow-up. For example, a participant who was censored at the 18th year of follow-up will be included in the analysis for <5 years, <10 years, <15 years, and <20 years.

All models were tested for potential interactions of the exposure variables with sex, birth date or maternal diabetes by including product terms. Models were also tested for any quadratic (non-linear) associations of the weight variables with type 2 diabetes incidence.

A bootstrap analysis was performed to account for the partial overlap of participants representing the growth periods and for any potential imbalance of covariates between groups and for any correlation arising from using some of the same measures to calculate the average weight z score or weight velocity z score for the growth periods. Briefly, we drew 10,000 random samples of participants in the longitudinal analysis with replacement to obtain a bootstrapped estimate of the standard error of the difference in risk estimates between the growth periods; this standard error was used to assess the significance of the difference. This approach assumes that the bootstrapped statistic approximates a normal distribution.³¹

A few sensitivity analyses were performed. To use all available data points in childhood in these participants, we estimated weight z score velocities using the linear spine mixed-effects model.³² This multilevel approach involved using piecewise linear splines with knots at 9 years (at the last birthday) and 14 years (at the last birthday). The positions of the knots were selected to obtain linear slopes of log₁₀weight z scores corresponding to the growth periods analyzed in the initial (simple) growth modeling approach, i.e. birth-9 years, 9–14 years, and 14–19 years. The estimated slopes, which varied between participants, were standardized to mean=0 and SD=1 and were subsequently used as continuous variables in Poisson regression models to estimate diabetes IRRs. The slope in a growth period was adjusted for the average weight z score in that period and other covariates, such as, sex, birth date and maternal diabetes. For more details on these analyses, please see the SAS syntax in the methods section of the ESM.

To test whether risk estimates were influenced by exposure to maternal diabetes, models were restricted to participants without such exposure. Similarly, models were also restricted to participants without functional mutations in the MC4R gene to test any influence of such mutations. Finally, models were adjusted for birthweight in analyses restricted to participants with data on birthweight.

All analyses were performed using SAS (version 9.4; SAS Institute, Cary, NC). A two-sided p <0.05 was considered statistically significant.

Results

Characteristics of participants

This study comprised 2,100 child participants representing the pre-adolescence growth period, 1,558 representing the early adolescence period, and 1,418 representing the late adolescence period. Overall, the present analysis included 3,107 participants with some participants representing more than one growth period. Participants representing the late adolescence were more often females (63.3%) than those representing the pre-adolescence period (52.2%) and the early adolescence period (55.2%; Table 1). Exposure to maternal diabetes was similarly common among participants representing the growth periods. There is balance in the median follow-up period to the end point for this analysis. However, participants representing pre-adolescence had a greater median number of follow-up visits (n=6) than those representing older growth periods (n=3; Table 1). The percentage of subsequent diabetes diagnoses made in follow-up at research examinations (as opposed to from clinical data) was comparable for the different growth periods (59.3% for pre-adolescence, 56.8% for early adolescence and 54.2% for late adolescence).

Table 1.

Participant characteristics by growth period

	Pre-adolescence a (n=2100)	Early adolescence a (n=1558)	Late adolescence a (n=1418)
Female (%)	1097 (52.2)	860 (55.2)	898 (63.3)
Birthweight (kg) b	3.45 (3.15, 3.75)	3.45 (3.15, 3.75)	3.44 (3.13, 3.75)
Mean ± SD weight (kg)	19.2 ± 5.7	49.8 ± 14.3	72.6 ± 19.4
Mean ± SD weight velocity (kg/year)	3.7 ± 1.2	6.5 ± 2.4	3.6 ± 2.5
Maternal diabetes (%)	264 (12.8)	196 (12.6)	143 (10.1)
Median (Q1, Q3) years of follow-up
Up to the end of the longitudinal study	12.7 (6.4, 30.0)	14.6 (6.1, 24.9)	14.3 (5.8, 23.2)
When analyses were restricted to up to 25 years	11.2 (6.1, 22.1)	11.5 (5.3, 20.8)	9.8 (4.3, 18.5)
Median (Q1, Q3) follow-up visits
Up to the end of the longitudinal study	6 (3, 11)	3 (2, 6)	3 (2, 6)
When analyses were restricted to up to 25 years	5 (3, 10)	3 (2, 5)	3 (1, 5)

Q, quartile; SD, standard deviation

^aPre-adolescence is birth to ∼8yr, early adolescence is ∼8yr to ∼13yr, and late adolescence is ∼13yr to ∼18yr (see Methods for precise definitions of the three growth periods)

^bData are available for 1244 participants representing the early adolescence period and 990 participants representing the late adolescence period.

Among participants representing more than one growth period, there were significant correlations in the average weight z scores between successive periods, i.e. r=0.67 between weight z scores in pre-adolescence and early adolescence, r=0.92 between early adolescence and late adolescence, and r=0.49 between pre-adolescence and late adolescence (Fig. 2). By contrast, only weight z score velocities in early and late adolescence showed significant but weak negative correlation (r=−0.19).

Fig. 2.

Correlations of pre-, early, and late adolescence age-sex-height-standardized average weight z score (a-c) and weight z score velocity (d-f).

r, Pearson’s correlation coefficient; p, p-value of the correlation coefficient. number of participants analyzed: 1244 for (a) and (d), 725 for (b) and (e), and 595 for (c) and (f). Pre-adolescence is birth to ~8yr, early adolescence is ~8yr to ~13yr, and late adolescence is ~13yr to ~18yr (see Methods for precise definitions of the three growth periods).

Diabetes incidence

During up to 25 years of follow-up from the end of the growth periods, there were 290 incident cases of diabetes among participants representing the pre-adolescence growth period (10.1 per 1000 person-years), 315 in those representing the early adolescence growth period (15.2 per 1000 person-years), and 380 in those representing the late adolescence growth period (22.0 per 1000 person-years; Table 1). Overall, there were 607 unique incident diabetes cases of which 63% were females. At the first research examination with diabetes, the mean age was 28 years and 89% were obese (BMI ≥30 kg/m²; ESM Table 1). Incidence rate estimates became less precise when analyses were extended beyond 25 years of follow-up because fewer participants had >25 years of follow-up (ESM Table 2).

Adjusted for sex, birth date, and maternal diabetes, greater average weight z score in each growth period associated with increased diabetes incidence during the first 10 years of follow-up (Fig. 3a; ESM Table 3), i.e. IRR 1.72 per SD of log₁₀weight z-score pre-adolescence (95% CI 1.40–2.11), IRR 2.09/SD of log₁₀weight z-score in early adolescence (CI 1.68–2.60), and IRR 1.86/SD of log₁₀weight z-score in late adolescence (CI 1.58–2.17). Although these associations declined steadily over time (ESM Fig. 1a–c), they persisted for up to 25 years of follow-up (Fig. 3b). Comparing associations across the growth periods, there was no significant between-growth period difference in the association of the average weight z score with diabetes incidence (p ≥0.05).

Fig. 3.

Sex-, date of birth- and maternal diabetes-adjusted individual association by growth period of average weight z score with type 2 diabetes incidence during (a) the first 10 years and (b) up to 25 years of follow-up; and of weight z score velocity with type 2 diabetes incidence during (c) the first 10 years and (d) up to 25 years of follow-up.

IRR, incidence rate ratio.

Pre-adolescence is birth to ~8yr, early adolescence is ~8yr to ~13yr, and late adolescence is ~13yr to ~18yr (see Methods for precise definitions of the three growth periods).

Weight z score velocity pre-adolescence significantly associated with diabetes incidence during the first 10 years of follow-up (IRR 1.79/SD of log₁₀weight z score velocity, 95% CI 1.49–2.17; Fig. 3c; ESM Table 3), with this significant association persisting for up to 25 years of follow-up (Fig. 3d; ESM Fig. 1d). Weight z score velocity in early adolescence had no significant effect on subsequent diabetes incidence during the first 10 years of follow-up (IRR 1.13/SD of log₁₀weight z score velocity, 95% CI 0.91–1.41; Fig. 3c) but had a significant but relatively weak association when incidence was extended to up to 25 years of follow-up (IRR 1.16/SD of log₁₀weight z score velocity, CI 1.05–1.30; Fig. 3d; ESM Fig. 1e). Similarly, compared to the weight z score velocity pre-adolescence, the cumulative association of weight z score velocity in late adolescence with diabetes incidence was weak during the first 10 years of follow-up (IRR 1.29/SD of log₁₀weight z score velocity, CI 1.09–1.51) and up to 25 years of follow-up (IRR 1.19/SD of log₁₀weight z score velocity, CI 1.07–1.32; Fig. 3c–d; ESM Fig. 1f). Overall, weight z score velocity in pre-adolescence had a significantly greater association with diabetes incidence than those in early adolescence (p<0.001) and late adolescence (p=0.016).

For each growth period, the pattern and magnitude of individual associations of the average weight z score and weight z score velocity with diabetes incidence (Fig. 3; ESM Table 3) were similar to those of their joint associations (ESM Table 4). There were no significant interactions of sex, birth date, or maternal diabetes with either the average weight z score or weight z score velocity in any analyses. Similarly, there was no significant non-linear association between the weight variables and type 2 diabetes incidence.

Sensitivity analysis

Findings from the linear spine mixed-effects model, adjusted for average weight z score, sex, birth date and maternal diabetes, were generally similar to those of the initial (simple) modeling approach with weight z score velocity pre-adolescence more strongly associated with type 2 diabetes incidence than weight z score velocities in early and late adolescence (ESM Tables 5). However, the linear spine mixed-effects model, unlike the simple model, showed a significant association between weight z score velocity in early adolescence and type 2 diabetes incidence during the first 10 years of follow-up (IRR 1.34/SD of log₁₀weight z score, CI 1.07–1.68) and up to 25 years of follow-up (IRR 1.25/SD of log₁₀weight z score, CI 1.07–1.37).

Results were substantially similar when models were restricted to participants without exposure to maternal diabetes (ESM Table 6) or when models were restricted to participants with no known MC4R deficiency (data not shown). Similarly, no meaningful difference was observed between results of analyses adjusted for birthweight compared to models not adjusted for this covariate (ESM Table 7); and in results of models comprising weight z scores (Fig. 3) compared with those comprising BMI z scores (ESM Fig. 2).

Discussion

Statement of principal findings

In this longitudinal analysis, higher weight in all growth periods significantly associated with greater type 2 diabetes incidence, as did accelerated weight gain in pre- and late- adolescence. Importantly, higher weight and greater weight velocity pre-adolescence jointly associated with the highest type 2 diabetes risk; these associations were similar, but weaker, for higher weights and greater weight velocities in early and late adolescence.

Strengths and weaknesses of the study

The strength of this study is the prospective ascertainment of diabetes using OGTTs. In addition, all anthropometric and biochemical measurements were performed at the same center and in the same laboratory for the entire 43-year study. Importantly, our longitudinal data permit the identification of critical and clinically meaningful growth periods.

Our study did not adjust birthweight for length at birth and gestational age, or adjust analyses for age at puberty, diet, physical activity, or socioeconomic status because of insufficient data on these variables. Similarly, due to limited serial research examination in childhood, growth was modelled using a simple approach that used weight measures from selected data points rather than all available data points. However, conclusions were not substantially affected when all available data were used in a linear spline mixed-effects model. We were unable to use other analytic approaches such as growth curve modelling (requiring more data) or age-period-cohort analysis (requiring larger sample sizes at age intervals). There is potential for selection bias given the greater proportion of females among participants representing the late adolescence period, although there is no evidence that risk was modified by sex. However, our estimates should be interpreted in the context that administrative censoring limited the proportion of participants completing all growth periods, i.e. children who were younger in the early years of the longitudinal study had longer potential follow-up than those who entered the study later, closer to the end of the study. Pima Indians have a high prevalence of obesity and early-onset type 2 diabetes.^19,20 However, prior findings on childhood obesity and type 2 diabetes in this population have generally been replicated in other populations.

Strengths and weaknesses in relation to other studies

Our findings on the association between childhood weight and type 2 diabetes incidence agree largely with results of previous studies, although direct comparisons are difficult given the inconsistencies in the growth periods examined, lack of longitudinal data on type 2 diabetes incidence, and differences in analytical approach. Similar to our findings, greater BMI z score at ages 4 to 19 years consistently associated with higher prevalence of prediabetes or type 2 diabetes in adulthood in the Bogalusa Heart study.¹² Moreover, being overweight at age 13 years (BMI ≥21.8 kg/m²) and persistent overweight at ages 7 years (BMI ≥17.4 kg/m²) and 13 years associated with greater type 2 diabetes among adult Danish men (30–60 years).⁶ However, contrary to our findings, being overweight at 7 years unadjusted for subsequent weight was not associated with type 2 diabetes incidence in the Danish study. The contrasting results may be due to differences in the analytic method as BMI was categorized and diabetes incidence was assessed from age ≥30 years in the Danish study. In the present study, weight was analyzed as a continuous variable while diabetes incidence was assessed from childhood. In a larger and gender diverse Danish study,¹³ although there were no significant interactions of childhood BMI with covariates, analyses were stratified by BMI, sex, birth cohort, and age at diagnosis, thereby complicating the interpretation of findings. We previously showed the additive effects of post-natal BMI, birthweight, maternal diabetes, and genotype on type 2 diabetes incidence.³³ In the present analysis, birthweight, maternal diabetes, and genotype did not significantly add to or modify the association between the child weight variables and type 2 diabetes incidence. Similarly, no such interactions were reported in other studies.^6,12,13

The observed association between greater weight gain in childhood and type 2 diabetes risk, especially the relatively stronger association of weight gain in early childhood, agrees with findings from a study of female nurses in the United States, although, in that study, childhood adiposity was assessed by recall of somatotypes or pictorial diagrams.¹⁴ By contrast, our findings differ from those of the Bogalusa Heart study in which weight gain in adolescence (aged 10–19 years) and not in pre-adolescence (aged 5–9 years) significantly associated with the prevalence of prediabetes or type 2 diabetes in adulthood.¹² In a Swedish study, weight change from ages 2–11 years significantly associated with type 2 diabetes (odds ratio 1.20) but not impaired glucose tolerance in adulthood (1.06), and the association was weaker when these outcomes were combined (1.12).¹⁵ Similarly, there was no significant association between weight gain in early childhood and the prevalence of hyperglycemia in other studies.^16,17 The contrasting results may be due to the difference in analytic approach. In the present study, risk was evaluated from childhood based on the time at risk of diabetes. By contrast, in other studies,^16,17 risk was evaluated based on diabetes prevalence in adulthood. Given that individuals with greater weight gain in early childhood are more likely to develop type 2 diabetes before adulthood, the association between weight gain in childhood and type 2 diabetes risk is more efficiently estimated if based on time at risk. In some studies,^15,17–19 weight was adjusted for or weight gain was calculated from adult weight obtained close to the time of type 2 diabetes ascertainment rather than measures that clearly preceded type 2 diabetes onset, thereby making reliable comparison with our estimates difficult. Concurrent weight, being closer to diabetes diagnosis, usually accounts for most of the variability in type 2 diabetes incidence thereby weakening the association of weight variables preceding diabetes incidence.

Meaning of the study: possible explanations and implications for clinicians and policy makers

This study identifies higher weight and greater weight change in pre-adolescence to be associated with the greatest type 2 diabetes incidence. There may be several possible explanations for this finding. Children who attained a higher weight, or gained weight excessively during pre-adolescence, are exposed to adiposity induced chronic inflammatory state for a longer duration, thereby increasing their metabolic risk. Indeed, for participants representing more than one growth period, the high correlations in their longitudinal weights suggests that participants who were heavier in pre-adolescence were also heavier in adolescence, i.e. weight tracked from younger to older ages, in agreement with a previous report in this population.²¹ Longer duration of obesity increases type 2 diabetes incidence both in the current study population³⁴ and in a nationally representative sample.³⁵ Therefore, the strong association of excess weight, regardless of the period attained, with greater type 2 diabetes incidence coupled with the weaker association of weight velocities in early and late adolescence with later type 2 diabetes risk suggest that excess weight-related type 2 diabetes risk is driven largely by greater weight velocity pre-adolescence and the effect of continuing weight gain thereafter.

Preadolescence is also a period when physiological “adiposity rebound” occurs with a second rise in BMI. Although we did not assess the age of adiposity rebound in the current study, others have shown an association between early onset of adiposity rebound and metabolic dysfunction including diabetes.^36,37 Pre-pubertal adiposity could also elicit an early rise in insulin resistance before the expression of physical and biochemical changes typically associated with puberty, thereby increasing the risk for type 2 diabetes.³⁸

Unanswered questions and future research

Pre-adolescence is an important growth period for preventing and managing obesity and reducing the incidence of type 2 diabetes. Lifestyle interventions aimed at weight loss and increased physical activity reduce the incidence of type 2 diabetes in adults.³⁹ This suggests that similar interventions for weight management in children and adolescents might be effective in preventing type 2 diabetes, although this hypothesis has not yet, to our knowledge, been tested.

Table 2.

Incidence rates of diabetes by growth period and by 5-year follow-up intervals

Length of follow-up	Early childhood a (n=2100)				Early adolescence a (n=1558)				Late adolescence a (n=1418)
	Cases	n	PY	I	Cases	n	PY	I	Cases	n	PY	I
< 5 years	32	2100	9820.6	3.3	33	1558	7116.7	4.6	59	1418	6361.5	9.3
5 to < 10 years	45	1711	7199.0	6.3	48	1220	5216.9	9.2	96	1060	4447.6	21.6
10 to < 15 years	49	1196	5104.1	9.6	68	893	3856.7	17.6	90	742	3189.4	28.2
15 to < 20 years	78	870	3768.5	20.7	87	662	2777.6	31.3	76	528	2113.1	36.0
20 to < 25 years	86	652	2713.8	31.7	79	450	1726.7	45.8	59	325	1191.4	49.5
Total	290		28606.0	10.1	315		20694.6	15.2	380		17303.0	22.0

n, number of participants followed; PY, person-years in a 5-year interval of follow-up; I, diabetes incidence rate in cases per 1000 person-years.

^aPre-adolescence is birth to ∼8yr, early adolescence is ∼8yr to ∼13yr, and late adolescence is ∼13yr to ∼18yr (see Methods for precise definitions of the three growth periods).

Research in context.

What is already known about this subject?

• Excessive body weight in childhood and adolescence increases the risk of early-onset type 2 diabetes and associated complications.

• Limited evidence exists on critical growth periods in childhood and adolescence that predispose to the development of type 2 diabetes.

What is the key question?

• What are the critical growth periods during childhood and adolescence that are associated with type 2 diabetes incidence?

What are the new findings?

• Prospective analyses revealed that higher weight and accelerated weight gain in all growth periods significantly associated with a greater risk of type 2 diabetes.

• Higher weight and greater weight velocity pre-adolescence jointly associated with the highest risk for type 2 diabetes.

Impact on clinical practice

• Pre-adolescence is an important growth period for preventing and managing obesity and reducing type 2 diabetes incidence.

Abbreviations

MC4R

Melanocortin-4 receptor

OGTT

Oral glucose tolerance test

IRR

Incidence rate ratio

Data availability

Owing to privacy concerns of the small population of volunteers, data are not available for distribution.