Diabetologia (2020) 63:519–527

https://doi.org/10.1007/s00125-019-05058-7

ARTICLE

Age of obesity onset, cumulative obesity exposure over early

adulthood and risk of type 2 diabetes
Juhua Luo 1 & Allison Hodge 2,3 & Michael Hendryx 4 & Julie E. Byles 5

Received: 10 October 2019 / Accepted: 31 October 2019 / Published online: 20 December 2019
# Springer-Verlag GmbH Germany, part of Springer Nature 2019

Abstract
Aims/hypothesis Obesity is a risk factor for type 2 diabetes, yet little is known about how timing and cumulative exposure of
obesity are related to disease risk. The aim of this study was to examine the associations between BMI trajectories, age of onset of
obesity and obese-years (a product of degree and duration of obesity) over early adulthood and subsequent risk of type 2 diabetes.
Methods Women aged 18–23 years at baseline (n = 11,192) enrolled in the Australian Longitudinal Study on Women’s Health
(ALSWH) in 1996 were followed up about every 3 years via surveys for up to 19 years. Self-reported weights were collected up
to seven times. Incident type 2 diabetes was self-reported. A growth mixture model was used to identify distinct BMI trajectories
over the early adult life course. Cox proportional hazards regression models were used to examine the associations between
trajectories and risk of diabetes.
Results One hundred and sixty-two (1.5%) women were newly diagnosed with type 2 diabetes during a mean of 16 years of
follow-up. Six distinct BMI trajectories were identified, varying by different initial BMI and different slopes of increase. Initial
BMI was positively associated with risk of diabetes. We also observed that age at onset of obesity was negatively associated with
risk of diabetes (HR 0.87 [95% CI 0.79, 0.96] per 1 year increment), and number of obese-years was positively associated with
diabetes (p for trend <0.0001).
Conclusions/interpretation Our data revealed the importance of timing of obesity, and cumulative exposure to obesity in the
development of type 2 diabetes in young women, suggesting that preventing or delaying the onset of obesity and reducing
cumulative exposure to obesity may substantially lower the risk of developing diabetes.

Keywords Age of obesity onset . Cumulative obesity . Diabetes . Obese-years . Obesity . Type 2 diabetes . Weight trajectory

Abbreviations GMM Growth mixture model

BIC Bayesian information criterion MET Metabolic equivalents
hs-CRP High-sensitivity C-reactive protein

* Juhua Luo
[email protected]

1
Department of Epidemiology and Biostatistics, School of Public
Health, Indiana University, 1025 E 7th Street,
Introduction
Bloomington, IN 47405, USA
2
Cancer Epidemiology Division, Cancer Council Victoria,
The prevalence of type 2 diabetes has increased dramatically
Melbourne, VIC, Australia during the past several decades. Currently, more than 500
3
Centre for Epidemiology and Biostatistics, Melbourne School of
million adults are living with diabetes worldwide [1]. The
Population and Global Health, University of Melbourne, increase in type 2 diabetes prevalence is closely related to
Parkville, VIC, Australia the upsurge in obesity. At present, more than 2.1 billion people
4
Department of Environmental and Occupational Health, School of (nearly 30% of the world’s population) are either obese (BMI
Public Health, Indiana University, Bloomington, IN, USA ≥30 kg/m2) or overweight (25 kg/m2≤ BMI <30 kg/m2) [2]. In
5
Research Centre for Generational Health and Ageing, University of 2017, almost two-thirds (63%) of adults in Australia were
Newcastle, Newcastle, NSW, Australia overweight or obese, and 28% were obese [3].
520 Diabetologia (2020) 63:519–527

Although obesity is a well-established risk factor for type 2 Methods

diabetes [4], little is known about the relationships between age
of onset of obesity and cumulative exposure to obesity and risk of ALSWH
diabetes, especially among young adults. In most previous studies,
exposures (obesity status or weight change) have relied on only The ALSWH is a longitudinal population-based study of
one or two measures of weight [5]. As a result, researchers have Australian women that was launched in 1996. It was designed
been unable to account for the dynamic changes of weight over to examine the relationships between biological, psychologi-
time, or to assess whether timing of weight change over the life cal, social and lifestyle factors and women’s physical health,
course makes a difference to the risk of developing type 2 diabetes. emotional wellbeing, and their use of and satisfaction with
A few studies have examined the association between the healthcare [11]. This study recruited a nationally representa-
duration of general or central obesity and the risk of diabetes tive sample of more than 40,000 women in three age cohorts,
but results are mixed [6–9]. Some studies reported that duration born in 1921–1926 (aged 70–75 years at baseline), 1946–
of obesity was associated with the risk of diabetes independently 1951 (aged 45–50 years) and 1973–1978 (aged 18–23 years).
of baseline BMI [7, 9], while others reported that the association Details of the design, recruitment, implementation and prog-
between diabetes and duration of obesity was largely determined ress of the ALSWH study are described elsewhere [11–15].
by current BMI [6, 8]. The study was approved by the Universities of Newcastle and
Obese-years is a construct that accounts for both how long Queensland Ethics Review Committees. All participants
a person has been obese and the magnitude of their obesity provided written informed consent.
(no. of BMI units above 30 kg/m2 × no. of years at that BMI)
and has been shown to be a better predictor of diabetes risk
than duration of obesity or level of BMI alone [10]. However, Study population
few studies have examined the relationship between the
cumulative exposure to obesity and risk of diabetes. The current study was based on data from the 1973–1978 birth
We used data from the Australian Longitudinal Study on cohort (aged 18–23 years at recruitment, n = 14,247).
Women’s Health (ALSWH) to identify BMI trajectories over Participants were excluded from the analysis for the following
the early adult life course. We then examined the relationship reasons: they reported a history of diabetes at enrolment or
between distinct BMI trajectories and risk of type 2 diabetes. reported having type 1 diabetes during follow-up (n = 247);
Further, we investigated the associations between timing of they answered only the baseline survey (n = 2309); they had
obesity onset, obese-years and type 2 diabetes. We hypothesise missing values for major covariates at baseline including
that earlier age at onset of obesity and more obese-years over physical activity, income and stress (n = 133); or they had no
follow-up would be more strongly associated with type 2 diabe- BMI data (n = 366). After exclusions, 11,192 women
tes risk than BMI alone. remained in the analytical cohort.
Diabetologia (2020) 63:519–527 521

Follow-up across items in specific life domains: own health, health of

other family members, work/employment, living arrange-
Participants completed survey 2 in 2000 (4 years after base- ments, study, money, relationship with parents, relationship
line) and have been followed up every 3 years thereafter up to with partner/spouse, relationships with children and relation-
19 years and a total of seven survey administrations. Surveys ship with other family members. The score has been demon-
were completed by mail, and an on-line option has been avail- strated to be reliable and valid [15].
able since survey 6 in 2012. For the 1973–1978 cohort, the
retention rate was low (69%) from wave 1 to wave 2, due to Statistical analysis
high levels of mobility and inability to contact participants at
this stage, but then remained relatively stable thereafter with First, descriptive statistics were calculated. Second, BMI
retention rates of 57–64% [15]. growth curves (trajectories) over time were examined. We
fitted the data by using both linear and quadratic models and
Measurements compared which of these models fit the data better by looking
at whether the z score test was significant for the quadratic
Outcome: incidence of type 2 diabetes during follow-up growth factor mean. Since the quadratic growth factor mean
Incident type 2 diabetes was defined via self-report of new was significant (p < 0.05), we used quadratic models for
doctor-diagnosed diabetes during follow-up. Self-reported growth curves.
diabetes in the ALSWH has been validated for middle-aged Third, a growth mixture model (GMM) was used to iden-
and older cohorts by comparing self-report data with admin- tify distinct BMI trajectories over time. The GMM allows
istrative hospital records [16]. There was substantial agree- modelling variables with partially missing data [18]. The opti-
ment for diabetes (κ > 0.70) for both cohorts [16], indicating mal number of groups was assessed by the Bayesian informa-
that self-reported diabetes in the ALSWH may be a reliable tion criterion (BIC) [19]. A lower BIC indicates better fit to the
indicator of diagnosed diabetes. data. Once the best-fitting model was identified, participants
were assigned to the trajectory groups to which their posterior
Exposure: self-reported weight and height measures across membership probability was largest. Cox proportional hazards
surveys In the ALSWH, height was collected at baseline and regression models were then used to evaluate the associations
weights were self-reported at baseline and at each follow-up between the trajectories and risk of type 2 diabetes. Survival
survey. BMI was calculated as weight (kg) divided by height time was defined as time between baseline and the date of
(m) squared. Only measures of BMI before the date of diag- newly diagnosed diabetes, censoring due to missing data,
nosis of type 2 diabetes were included in this analysis. death, or end of follow-up at survey 7, whichever came first.
Accuracy of BMI estimated from self-reported height and For women who reported newly diagnosed diabetes on a
weight has been validated in the ALSWH [17]. There was survey, we assumed the point of diagnosis was halfway
substantial agreement (84%) between BMI categories derived through the period between that survey and the previous
from self-reported and measured height and weight data; survey. The BMI trajectory would end at the previous survey.
agreement for healthy-weight women was 94%. This suggests In the multivariate-adjusted models, potential confounders
that the self-reported data on weight and height obtained from were included as covariates. We also examined relationships
mailed surveys in the ALSWH can be used to estimate BMI between sociodemographic characteristics, lifestyle and stress
with reasonable accuracy [17]. and BMI trajectories.
Fourth, since the GMM approach enables us to
Covariates Participants’ sociodemographic characteristics model BMI trajectories with random variation among
(age, educational attainment, income adequacy), lifestyle individuals within a group, we allowed each individual
factors (physical activity, smoking, alcohol use), gestational to have her own BMI trajectory; namely, we predicted
diabetes history, and psychosocial factors (a stress score) BMI trajectory (including intercept, slope and quadratic)
measured at baseline were considered as potential for each individual. Using the predicted curve for each
confounders. Table 1 presents detailed information about individual, we estimated the age of onset of obesity if
covariate categories and scores. Income adequacy was any, degree of obesity (measured by difference in BMI
assessed by an item asking ‘how do you manage on the above obesity [≥30 kg/m2]), and duration in years of
income you have available?’ Physical activity data was obesity. Obese-years was estimated by calculating the
collected by asking for the amount of time spent in the last AUC that was above 30 kg/m2 by trapezoid rule [20].
week on different activities including walking briskly, moder- Age of onset of obesity and obese-years were treated as
ate leisure activity and vigorous leisure activity and time-varying covariates in Cox models.
converting to metabolic equivalents (MET)/week (continu- We conducted a sensitivity analysis using baseline-only
ous). Perceived stress was measured continuously as a sum BMI categories to examine HRs for diabetes, with control
 522

Table 1 Comparison of baseline characteristics across BMI trajectories

Variable Overall NW-low stable NW-low modest NW-mid stable NW-mid rapid increase OW-rapid increase OB-rapid increase p value
(N=11,192) (13%; n=1558) increase (21%; n=2329) (14%; n=1563) (28%; n=3118) (21%; n=2335) (3%; n=289)

Age, years 20.8 ± 1.4 20.7 ± 1.5 20.7 ± 1.4 20.8 ± 1.4 20.8 ± 1.5 20.8 ± 1.5 20.9 ± 1.4 0.2749
Physical activity, MET 19.0 ± 14.1 20.3 ± 14.8 18.5 ± 14.1 20.4 ± 14.2 18.7 ± 13.9 18.4 ± 13.6 17.6 ± 13.6 <0.0001
Mean stress score 0.90 ± 0.6 0.86 ± 0.5 0.88 ± 0.5 0.90 ± 0.5 0.89 ± 0.6 0.94 ± 0.6 0.92 ± 0.6 0.0003
Smoking <0.0001
Never 5828 (59.8) 975 (65.7) 1415 (63.5) 690 (63.0) 1610 (56.8) 990 (53.3) 148 (61.7)
Former 1630 (16.7) 243 (16.4) 315 (14.1) 177 (16.1) 487 (17.2) 366 (19.7) 42 (17.5)
Current: <10 cigarettes/day) 1006 (10.3) 138 (9.3) 232 (10.4) 116 (10.6) 296 (10.4) 208 (11.2) 16 (6.7)
Current: 10–19 cigarettes/day 749 (7.7) 88 (5.9) 158 (7.1) 69 (6.3) 252 (8.9) 162 (8.7) 20 (8.3)
Current: ≥20 cigarettes/day 529 (5.4) 40 (2.7) 109 (4.9) 44 (4.0) 189 (6.7) 133 (7.2) 14 (5.8)
Alcohol intake 0.03
Low-risk, non or rarely drinker 10,499 (94.7) 1595 (95.7) 2400 (95.5) 1193 (93.7) 3066 (94.6) 1977 (93.6) 268 (95.0)
Risky drinker (3–4 drinks per day) 523 (4.7) 65 (3.9) 104 (4.1) 71 (5.6) 154 (4.8) 115 (5.4) 14 (5.0)
High-risk drinker (≥5 drinks per day) 64 (0.6) 6 (0.4) 9 (0.4) 9 (0.7) 20 (0.6) 20 (0.9)
Education <0.0001
High school 7809 (69.8) 1199 (71.3) 1727 (68.2) 881 (68.5) 2273 (69.4) 1535 (72.0) 194 (67.8)
or lower
Technical training 1993 (17.8%) 219 (13.0%) 454 (17.9%) 189 (14.7%) 645 (19.7%) 410 (19.2%) 76 (26.6%)
University degree or higher 1390 (12.4) 263 (15.6) 352 (13.9) 216 (16.8) 357 (10.9) 186 (8.7) 16 (5.6)
Income adequacy <0.0001
Difficult always 1921 (17.2) 214 (12.7) 378 (14.9) 177 (13.8) 592 (18.1) 495 (23.2) 65 (22.7)
Difficult sometimes 3665 (32.7) 497 (29.6) 815 (32.2) 418 (32.5) 1078 (32.9) 748 (35.1) 109 (38.1)
Not too bad 4100 (36.6) 653 (38.8) 981 (38.7) 489 (38.0) 1205 (36.8) 684 (32.1) 88 (30.8)
It is easy 1506 (13.5) 317 (18.9) 359 (14.2) 202 (15.7) 400 (12.2) 204 (9.6) 24 (8.4)
Gestational diabetes, yes 275 (2.5) 25 (1.5) 39 (1.5) 15 (1.2) 104 (3.2) 80 (3.8) 12 (4.2) <0.0001

Values are expressed as n (%) for categorical variables or mean±SD for continuous variables
Participants were grouped according to baseline BMI (NW, normal BMI at baseline; OB, obesity at baseline; OW, overweight at baseline; low, at low end of the category; mid, at middle of the category;
high, at high end of the category) and BMI trajectory (growth curves over time: stable, slope <1; modest increase, slope = 1 to <2; rapid increase, slope ≥2)
p value comparisons across class categories are based on χ2 test for categorical variables; p values for continuous variables are based on ANOVA
Diabetologia (2020) 63:519–527
Diabetologia (2020) 63:519–527 523

for the same covariates, to compare HRs to those obtained for In the study population, we observed that two classes had
weight trajectories. stable BMI trajectory over time (27% of women), including
Finally, we performed a sensitivity analysis by further NW-low stable and NW-mid stable classes; 21% of women
adjusting for diet data from a food frequency questionnaire with normal weight at baseline had modest BMI increase, and
[21], including information about total energy intake, fibre more than half of the women had a rapid increase of BMI over
intake, dietary glycaemic index and glycaemic load. time, including those in the NW-rapid increase, OW-rapid
Statistical analyses were performed using Mplus 8.2 [22] increase and OB-rapid increase classes. The largest proportion
and SAS Version 9.4 (SAS Institute, Cary, NC, USA). (28%) of women belonged to the NW-mid rapid increase
class. The next most frequent categories were the OW-rapid
increase group (21%) and NW-low modest increase group
(21%). A small group (3%) of women belonged to the OB-
Results rapid increase class, with initial BMI around 33 kg/m2.
Table 1 shows baseline characteristics across the six BMI
Among 11,192 women who were free of diabetes at baseline, trajectories. Based on mean baseline BMI from low to high,
162 women (1.5%) developed type 2 diabetes during a mean compared with the lower ranked class, women in the higher
of 16 years of follow-up. The mean BMI increased from ranked classes were more likely to be physically inactive, have
22.8 kg/m2 at baseline to 26.9 kg/m2 at survey 7. The obesity higher stress scores, smoke more, be more risky drinkers, have
prevalence rate increased from 6.5% at baseline to 25.7% at lower education level, have more difficulty in managing on
survey 7, and the mean obesity-years was 8.4 years (SD 5.2) available income and were more likely to have a history of
among the 2008 women who became obese. gestational diabetes. Baseline age across classes was not
We examined different quadratic models with varying different (Table 1).
number of classes (trajectories) from two to seven, and iden- Estimated intercept, slope and quadratic parameters
tified a six-class solution based on the lowest BIC value. We for each class are listed in Table 2. Specific growth
ranked the six classes (Fig. 1) based on their corresponding trajectories are presented in Fig. 1. Compared with
mean of BMI at baseline from low to high. We then women in the NW-low stable class, two groups had
categorised the four classes with baseline BMI <25 kg/m2 as significantly higher risk of type 2 diabetes (HR 4.75
normal weight, and labelled them according to their baseline [95% CI 2.43, 9.28] for OW-rapid increase class; HR
BMI and slope: NW-low stable (slope <1); NW-mid stable; 10.06 [95% CI 4.69, 21.58] for OB-rapid increase class]
NW-low modest increase (slope 1 to <2); NW-rapid increase after adjustment for potential confounders. Initial BMI
(slope ≥2). The remaining two classes were overweight at was significantly and positively associated with risk of
baseline, with BMI 25 to <30 kg/m2 and with a rapid increase type 2 diabetes. Slope was borderline significantly asso-
in BMI (OW-rapid increase), or obese (BMI ≥30 kg/m2) with ciated with the risk (p = 0.07) but the quadratic term was
a rapid increase (OB-rapid increase). not associated with risk (Table 2).

Fig. 1 BMI trajectories over 19 40

years of follow-up. The estimated 38

mean BMI in each trajectory is
36
shown. Participants were grouped
according to baseline BMI (NW, 34
normal BMI; OB, obese; OW, 32
BMI (kg/m2)

overweight; low, at low end of the

30
category; mid, at middle of the
category) and BMI increase 28
(growth curves over time: stable, 26
slope <1; modest increase, slope 1
to <2; rapid increase, slope ≥2) 24

22

20

18
0 2 4 6 8 10 12 14 16 18 20
Follow−up (years)

NW-low stable (13%; n=1558) NW-low modest increase (21%; n=2329)

NW-mid stable (14%; n=1563) NW-mid rapid increase (28%; n=3118)

OW−rapid increase (21%; n=2335) OB−rapid increase (3%; n=289)

524 Diabetologia (2020) 63:519–527

Table 2 Estimated means of intercept, slope and quadratic parameters for each BMI class, and HRs for type 2 diabetes incidence

Class/parameter No. of cases Intercept Slope Quadratic Multivariable-adjusted model,

HR (95%CI)a

Class
NW-low stable (13% of study population) (n = 1558) 10 19.92 0.13b 0.16 1.00
NW-low modest increase (21% of study population) (n = 2329) 21 20.27 1.61 0.07b 1.34 (0.63, 2.85)
NW-mid stable (14% of study population) (n = 1563) 11 22.46 0.03b 0.23b 1.37 (0.50, 3.22)
NW-mid rapid increase (28% of study population) (n = 3118) 32 22.61 3.72 −0.44 1.49 (0.73, 3.04)
OW-rapid increase (21% of study population) (n = 2335) 67 26.56 5.51 −0.79 4.75 (2.43, 9.28)
OB-rapid increase (3% of study population) (n = 289) 21 33.28 7.69 −2.32 10.06 (4.69, 21.58)
Quadratic model parameter
Intercept 1.12 (1.09, 1.15)
Slope 1.06 (0.99, 1.13)
Quadratic 0.97 (0.83, 1.14)

Participants were classed according to baseline BMI (NW, normal BMI at baseline; OB, obesity at baseline; OW, overweight at baseline; low, at low end
of the category; mid, at middle of the category; high, at high end of the category) and BMI trajectory (growth curves over time: stable, slope <1; modest
increase, slope = 1 to <2; rapid increase, slope ≥2)
a
Model adjusted for age (continuous), physical activity in (MET/week, continuous), smoking (never, former, current smoker [<10 cigarettes/day, 10–19
cigarettes/day, ≥20 cigarettes/day]), alcohol (low-risk, non or rarely drinker, risky drinker, high-risk drinker), education (high school or lower, technical
training, university degree or higher), income adequacy (impossible, difficult always, difficult sometimes, not too bad, it is easy), stress score (contin-
uous) and history of gestational diabetes (no, yes)
b
These estimates were not significant; p<0.05 for all other estimates

Among the 10,521 (94%) women who were not obese at The sensitivity analysis using only baseline BMI revealed that
baseline, we observed that women who became obese during individuals with baseline obesity (BMI ≥30 kg/m2) had an HR of
follow-up had a threefold (95% CI 2.01, 4.50) risk of type 2 7.07 (95% CI 4.73, 10.57), which is lower than the HR of 10.06
diabetes compared with women who did not become obese. (95% CI 4.69, 21.58) shown in Table 2 for individuals in the
There was a negative association between age at onset of obese with rapid increase group. Similarly, individuals shown in
obesity and risk of type 2 diabetes among women who became Table 2 who were overweight with rapidly increasing BMI had
obese during follow-up (HR 0.87 [95% CI 0.79, 0.96]) an HR of 4.75 (95% CI 2.43, 9.28), which was higher than the
(Table 3). Compared with women who did not become obese baseline-only overweight group (HR 2.33 [95% CI 1.53, 3.55]).
during the follow-up, women who became obese and had A sensitivity analysis that further adjusted for the number
obese-years of <10, 10 to <30 and ≥30 had an HR (95% CI) of children and dietary intake (including total energy intake,
of 2.18 (1.25, 3.81), 3.01 (1.53, 5.91) and 5.88 (3.15, 10.97), fibre intake and dietary glycaemic index or glycaemic load)
respectively (p for trend <0.0001) (Table 3). found similar results (data not shown).

Table 3 Associations between

age at onset of obesity, obese- Cases/population HR (95% CI)
years, and risk of type 2 diabetes
incidence among 10,521 (94%) Became obese during follow-up
women who were not obese at No 79/8513 1.00
baseline Yes 38/2008 3.01 (2.01, 4.50)
Age at onset of obesity among women who became obese 38/2008 0.87 (0.79, 0.96)a
Obese-years
Not obese 79/8513 1.00
<10 15/916 2.18 (1.25, 3.81)
10 to <30 10/494 3.01 (1.53, 5.91)
≥30 13/598 5.88 (3.15, 10.97)
p for trend 117/10,521 <0.0001
a
For a 1 year increment in age of obesity onset
Diabetologia (2020) 63:519–527 525

Discussion of BMI at one point in time, measurement of obese-years

account for the cumulative effect of obesity. A previous study
From this population-based prospective cohort of Australian compared different measures of obesity as predictors of diabe-
young adult women with 19 years of follow-up, we identified tes risk and reported that obese-years was a better predictor for
six distinct BMI trajectories varying by initial BMI and slope. type 2 diabetes than duration of obesity or BMI values alone
More than half of the women experienced a rapid BMI [10]. This may be because obese-years represent the cumula-
increase from early (18–23 years old) to middle adulthood tive damage to the body caused by obesity. It is well-accepted
(37–42 years old). Our data confirmed that BMI in young that excess adiposity can have deleterious metabolic effects
adulthood played an important role in the subsequent risk of such as increased insulin resistance or increased levels of
developing type 2 diabetes during adulthood. We also proinflammatory cytokines [33, 34]. Prolonged duration of
observed that women who were non-obese at baseline but excess adiposity may result in additional metabolic changes,
became obese during follow-up had a higher risk of type 2 leading to the development of diabetes [34, 35]. These results
diabetes relative to women who stayed non-obese; the youn- indicate the importance of both duration of obesity and the
ger the age at onset of obesity or the greater obese-years, the degree of obesity in the development of type 2 diabetes.
higher the risk of type 2 diabetes. We also observed that older age at onset of obesity was
A few previous studies have investigated associations associated with lower risk of type 2 diabetes when compared
between BMI trajectories and diabetes incidence [23–26], with younger age at onset. Our finding indicates that early-
diabetes-related metabolic markers, or inflammatory markers onset obesity may be particularly deleterious for future diabe-
[27–30]. Findings from these studies indicate that risks for tes risk in women. We only identified one study that examined
diabetes incidence, diabetes-related metabolic markers (i.e. age at onset of obesity in relation to the risk of type 2 diabetes
glucose, insulin and HOMA-IR), or elevated hs-CRP (high- [36]. This analysis, based on the Framingham Heart study,
sensitivity C-reactive protein) [23, 27–30] are associated with found that age of onset may be a useful, practical addition to
steeper weight gain trajectories, consistent with our findings. current BMI in the elderly. One explanation for the finding of
Two studies [27, 28] based on a cohort in a similar age group higher risk of type 2 diabetes among women who became
to our study also identified six similar trajectories and obese at a younger age may be the prolonged duration of
observed that a steeper weight gain trajectory was associated obesity, because women with early onset of obesity will likely
with higher risks of metabolic markers of diabetes or greater also have longer duration of obesity. However, it is possible
risk of elevated hs-CRP compared with more moderate weight that obesity during young age may be more deleterious for
grain trajectories. A recent study using repeated measure- insulin resistance and diabetes than obesity during older age
ments of BMI from childhood to adulthood observed that [37, 38]. Another study has also demonstrated that diabetes
the linear slope of BMI change between ages 10 and 19 years risk was particularly high in individuals who were obese as
was positively associated with adult hyperglycaemia [31]. adolescents relative to those with adult-onset obesity [38].
Another study based on retrospectively recalled weights from These data highlight the need for early diabetes prevention
age 20 years and weight in middle-aged Japanese men and efforts to prevent or delay the onset of obesity.
women found that long-term weight change slope was signif- Strengths of the study include the long-term follow-up in a
icantly associated with risk of type 2 diabetes [32]. That is, the large prospective cohort. Several limitations deserve mention.
steeper slope, the higher the risk of type 2 diabetes. In our First, weight was self-reported. There is evidence from a
study, the association between slope and risk of type 2 diabe- previous validation study that self-reported height and weight
tes was only borderline significant. This may be due to the are reliable [17]; however, the reliability of self-reported
small number of women diagnosed with type 2 diabetes over weight is better for normal weight women than obese women.
time among this young cohort. Thus, potential misclassification in trajectory assignment may
Our data indicated that baseline BMI among young women still exist and the fact that the degree of under-reporting in
was significantly associated with risk of developing type 2 weight is differential with respect to their true weight may lead
diabetes, with women in the overweight or obese BMI trajec- to an overestimate of the associations between trajectories and
tories having significantly higher risk of type 2 diabetes diabetes risk. Second, although self-reported diabetes in the
compared with women in the NW-low stable group. The cohort has been validated for middle-aged and older cohorts
results highlight the importance of overweight or obesity in and has been demonstrated to be a reliable indicator of diag-
early adulthood as risk factors for adult diabetes, indicating nosed diabetes [16], a relatively larger degree of under-
that weight control starting before early adulthood is critical diagnosis in the young cohort is possible, which may bias
for reducing type 2 diabetes risk in later life. our findings. For example, if diabetes is more likely to be
In addition to initial BMI, we also observed that a higher under-diagnosed among obese women than non-obese
number of obese-years was associated with higher risk of type women, our results may be under-estimated. On the other
2 diabetes in a dose–response fashion. Unlike measurements hand, if diabetes is less likely to be under-diagnosed for obese
526 Diabetologia (2020) 63:519–527

women than non-obese women our results may be over-esti- Contribution statement All the people listed as authors fulfil all the
following three criteria: (1) substantial contributions to conception and
mated. The prevalence of diabetes we observed (1.5%) was
design, acquisition of data, or analysis and interpretation of data; (2)
similar to self-reported data from the Australian Bureau of drafting the article or revising it critically for important intellectual
Statistics (ABS) 2014–15 National Health Survey, where the content; and (3) final approval of the version to be published. JL designed
prevalence of diabetes was 1.5% among women aged 18– the study and drafted the manuscript. JEB acquired the data. JL is the
guarantor of this work.
44 years [39]. Third, the low incidence of diabetes in our
young study population during the follow-up period may be
a concern for the Cox models, given these can produce unre-
liable estimates when the outcome is rare. Fourth, BMI data
References
were only collected from baseline; thus, we were unable to
1. Kaiser AB, Zhang N, van der Pluijm W (2018) Global prevalence
calculate the timing of obesity for women who were already of type 2 diabetes over the next ten years (2018-2028). Diabetes
obese at baseline. These women had to be excluded from 67(Suppl 1):202
analyses for age at onset and obese-years. This may reduce 2. Ng M, Fleming T, Robinson M et al (2014) Global, regional, and
national prevalence of overweight and obesity in children and
study power despite the fact that these women accounted for
adults during 1980-2013: a systematic analysis for the Global
only 6.5% of the study population. Another limitation is that Burden of Disease Study 2013. Lancet 384:766–781
our analysis does not consider the uncertainty of the classifi- 3. Australian Institute of Health and Welfare (2017) A picture of over-
cation rather than forcing individuals into their most likely weight and obesity in Australia Available from https://www.aihw.
gov.au/getmedia/172fba28-785e-4a08-ab37-2da3bbae40b8/aihw-
trajectory patterns. Finally, the study was conducted among
phe-216.pdf.aspx?inline=true. Accessed 10 Jun 2019
young Australian women. Caution should be taken when 4. CDC (2019) Risk factors for diabetes. Available from https://www.
generalising our findings to other populations. cdc.gov/diabetes/basics/risk-factors.html. Accessed 6 May 2019
In conclusion, our data have shown that more than half of the 5. Kodama S, Horikawa C, Fujihara K et al (2014) Quantitative rela-
tionship between body weight gain in adulthood and incident type 2
women experienced a rapid BMI increase from early to middle
diabetes: a meta-analysis. Obes Rev 15:202–214
adulthood, suggesting the importance of monitoring weight 6. Bouchard DR, Porneala B, Janssen I et al (2013) Risk of type 2
change over time. Persons who experience rapid weight gain diabetes and cumulative excess weight exposure in the
are at increased diabetes risk independently of baseline BMI Framingham Offspring Study. J Diabetes Complications 27:214–
status. In particular, our data indicate the importance of timing 218
7. Abdullah A, Stoelwinder J, Shortreed S et al (2011) The duration of
of obesity and cumulative exposure to obesity measured by obesity and the risk of type 2 diabetes. Public Health Nutr 14(1):
obese-years in relation to diabetes risk in young women. The 119–126. https://doi.org/10.1017/S1368980010001813
results highlight the importance of preventing or delaying the 8. Hu Y, Bhupathiraju SN, de Koning L, Hu FB (2014) Duration of
onset of obesity and reducing cumulative exposure to obesity obesity and overweight and risk of type 2 diabetes among US
women. Obesity (Silver Spring) 22(10):2267–2273. https://doi.
to substantially lower the risk of developing diabetes. We recom- org/10.1002/oby.20851
mend that people self-monitor weight change over time, and that 9. Reis JP, Hankinson AL, Loria CM et al (2013) Duration of abdom-
healthcare providers attend to weight change in addition to static inal obesity beginning in young adulthood and incident diabetes
weight as another risk factor for diabetes. through middle age: the CARDIA study. Diabetes Care 36(5):
1241–1247. https://doi.org/10.2337/dc12-1714
10. Abdullah A, Wolfe R, Mannan H, Stoelwinder JU, Stevenson C,
Acknowledgements We are grateful to the women who provided the Peeters A (2012) Epidemiologic merit of obese-years, the combi-
survey data. The authors also thank G. Giles of the Cancer nation of degree and duration of obesity. Am J Epidemiol 176(2):
Epidemiology Division, Cancer Council Victoria, for permission to use 99–107. https://doi.org/10.1093/aje/kwr522
the Dietary Questionnaire for Epidemiological Studies (Version 2), 11. Brown WJ, Bryson L, Byles JE et al (1998) Women’s Health
Melbourne: Cancer Council Victoria, 1996. Australia: recruitment for a national longitudinal cohort study.
Women Health 28:23–40
Data availability Data may be made available to collaborating 12. Brown WJ, Dobson AJ, Bryson L, Byles JE (1999) Women’s
researchers where there is a formal request to make use of the material. Health Australia: on the progress of the main cohort studies. J
Permission to use the data must be obtained from the Data Access Womens Health Gend Based Med 8(5):681–688. https://doi.org/
Committee of ALSWH (https://www.alswh.org.au/how-to-access-the- 10.1089/jwh.1.1999.8.681
data/alswh-data). 13. Dobson AJ, Hockey R, Brown WJ et al (2015) Cohort Profile
Update: Australian Longitudinal Study on Women’s Health. Int J
Funding The research on which this paper is based was conducted as part Epidemiol 44:1547 1547a-1547f
of the ALSWH by the University of Queensland and the University of 14. Lee C, Dobson AJ, Brown WJ et al (2005) Cohort Profile: the
Newcastle. We are grateful to the Australian Government Department of Australian Longitudinal Study on Women’s Health. Int J
Health for funding. This study was also supported by a Pilot and Epidemiol 34(5):987–991. https://doi.org/10.1093/ije/dyi098
Feasibility Award within the Center for Diabetes and Metabolic 15. Women’s Health Australia (2019) Australian Longitudinal Study on
Diseases (CDMD) at Indiana University School of Medicine, NIH/ Women s Health: sample. Available from https://www.alswh.org.
NIDDK Grant Number P30 DK097512. au/about/sample. Accessed 25 Jan 2019
16. Navin Cristina TJ, Stewart Williams JA, Parkinson L, Sibbritt DW,
Duality of interest The authors declare that there is no duality of interest Byles JE (2016) Identification of diabetes, heart disease, hyperten-
associated with this manuscript. sion and stroke in mid- and older-aged women: Comparing self-
Diabetologia (2020) 63:519–527 527

report and administrative hospital data records. Geriatr Gerontol Int 29. Walsh EI, Shaw J, Cherbuin N (2018) Trajectories of BMI change
16:95–102 impact glucose and insulin metabolism. Nutr Metab Cardiovasc Dis
17. Burton NW, Brown W, Dobson A (2010) Accuracy of body mass 28(3):243–251. https://doi.org/10.1016/j.numecd.2017.12.003
index estimated from self-reported height and weight in mid-aged 30. Liu R, Brickman WJ, Christoffel KK et al (2012) Association of
Australian women. Aust N Z J Public Health 34(6):620–623. adiposity trajectories with insulin sensitivity and glycemic deterio-
https://doi.org/10.1111/j.1753-6405.2010.00618.x ration: a longitudinal study of rural Chinese twin adults. Diabetes
18. Curran PJ, Obeidat K, Losardo D (2010) Twelve Frequently Asked Care 35(7):1506–1512. https://doi.org/10.2337/dc11-2060
Questions About Growth Curve Modeling. J Cogn Dev 11(2):121– 31. Zhang T, Xu J, Li S et al (2019) Trajectories of childhood BMI and
136. https://doi.org/10.1080/15248371003699969 adult diabetes: the Bogalusa Heart Study. Diabetologia 62:70–77
19. Nagin DS, Tremblay R (2005) Developmental trajectory groups: 32. Zhang Y, Yatsuya H, Li Y et al (2017) Long-term weight-change
fact or a useful statistical fiction. Criminology 43:873–904 slope, weight fluctuation and risk of type 2 diabetes mellitus in
20. Tai MM (1994) A mathematical model for the determination of total middle-aged Japanese men and women: findings of Aichi
area under glucose tolerance and other metabolic curves. Diabetes Workers’ Cohort Study. Nutr Diabetes 7:e252
Care 17(2):152–154. https://doi.org/10.2337/diacare.17.2.152 33. Hansen E, Hajri T, Abumrad NN (2006) Is all fat the same? The role
21. Women s Health Australia (2019) Australian Longitudinal Study on of fat in the pathogenesis of the metabolic syndrome and type 2
Women s Health: Food frequency information / dietary question- diabetes mellitus. Surgery 139(6):711–716. https://doi.org/10.1016/
naire for epidemiology studies. Available from https://www.alswh. j.surg.2005.10.018
org.au/for-researchers/data/15-for-researchers/74-food-frequency-
34. Kahn SE, Hull RL, Utzschneider KM (2006) Mechanisms linking
information. Accessed 5 Aug 2019
obesity to insulin resistance and type 2 diabetes. Nature 444(7121):
22. Muthen LK, Muthen BO (1998–2016) Mplus User’s Guide
840–846. https://doi.org/10.1038/nature05482
(Seventh Edition). Muthen & Muthen, Los Angeles, CA
23. Fagherazzi G, Vilier A, Affret A, Balkau B, Bonnet F, Clavel- 35. Felber JP, Golay A, Jequier E et al (1988) The metabolic conse-
Chapelon F (2015) The association of body shape trajectories over quences of long-term human obesity. Int J Obes 12(5):377–389
the life course with type 2 diabetes risk in adulthood: a group-based 36. Tanamas SK, Wong E, Backholer K et al (2016) Age of onset of
modeling approach. Ann Epidemiol 25(10):785–787. https://doi. obesity and risk of type 2 diabetes. Aust N Z J Public Health 40(6):
org/10.1016/j.annepidem.2015.06.002 579–581. https://doi.org/10.1111/1753-6405.12593
24. Mano Y, Yokomichi H, Suzuki K et al (2015) Do body mass index 37. Moran A, Jacobs DR Jr, Steinberger J et al (1999) Insulin resistance
trajectories affect the risk of type 2 diabetes? A case-control study. during puberty: results from clamp studies in 357 children. Diabetes
BMC Public Health 15:718 48(10):2039–2044. https://doi.org/10.2337/diabetes.48.10.2039
25. Peter RS, Keller F, Klenk J, Concin H, Nagel G (2016) Body mass 38. The NS, Richardson AS, Gordon-Larsen P (2013) Timing and dura-
trajectories, diabetes mellitus, and mortality in a large cohort of tion of obesity in relation to diabetes: findings from an ethnically
Austrian adults. Medicine 95:e5608 diverse, nationally representative sample. Diabetes Care 36(4):865–
26. Heianza Y, Arase Y, Kodama S et al (2015) Trajectory of body mass 872. https://doi.org/10.2337/dc12-0536
index before the development of type 2 diabetes in Japanese men: 39. Australian Institute of Health and Welfare (2019) Diabetes snap-
Toranomon Hospital Health Management Center Study 15. J shot. Available from https://www.aihw.gov.au/reports/diabetes/
Diabetes Investig 6(3):289–294. https://doi.org/10.1111/jdi.12308 diabetes-snapshot/contents/how-many-australians-have-diabetes.
27. Gordon-Larsen P, Koehler E, Howard AG et al (2014) Eighteen Accessed 12 Jun 2019
year weight trajectories and metabolic markers of diabetes in
modernising China. Diabetologia 57:1820–1829 Publisher’s note Springer Nature remains neutral with regard to jurisdic-
28. Thompson AL, Koehler E, Herring AH et al (2016) Weight gain tional claims in published maps and institutional affiliations.
trajectories associated with elevated C-reactive protein levels in
Chinese adults. J Am Heart Assoc 5:e003262