Distinct Body Mass Index Trajectories to Young-Adulthood Obesity and Their Different Cardiometabolic Consequences

Supplemental Digital Content is available in the text. Objective: Different body mass index (BMI) trajectories that result in obesity may have diverse health consequences, yet this heterogeneity is poorly understood. We aimed to identify distinct classes of individuals who share similar BMI trajectories and examine associations with cardiometabolic health. Approach and Results: Using data on 3549 participants in ALSPAC (Avon Longitudinal Study of Parents and Children), a growth mixture model was developed to capture heterogeneity in BMI trajectories between 7.5 and 24.5 years. Differences between identified classes in height growth curves, body composition trajectories, early-life characteristics, and a panel of cardiometabolic health measures at 24.5 years were investigated. The best mixture model had 6 classes. There were 2 normal-weight classes: normal weight (nonlinear; 35% of sample) and normal weight (linear; 21%). Two classes resulted in young-adulthood overweight: normal weight increasing to overweight (18%) and normal weight or overweight (16%). Two classes resulted in young-adulthood obesity: normal weight increasing to obesity (6%) and overweight or obesity (4%). The normal-weight-increasing-to-overweight class had more unfavorable levels of trunk fat, blood pressure, insulin, HDL (high-density lipoprotein) cholesterol, left ventricular mass, and E/e′ ratio compared with the always-normal-weight-or-overweight class, despite the average BMI trajectories for both classes converging at ≈26 kg/m2 at 24.5 years. Similarly, the normal-weight-increasing-to-obesity class had a worse cardiometabolic profile than the always-overweight-or-obese class. Conclusions: Individuals with high and stable BMI across childhood may have lower cardiometabolic disease risk than individuals who do not become overweight or obese until late adolescence.

T he number of adults worldwide with obesity increased from ≈100 to 671 million between 1975 and 2016, with an additional 1.3 billion in the overweight range. 1 This epidemic is strongly related to morbidity and mortality rates, particularly due to diseases affecting the circulatory and endocrine systems. 2 Adults with obesity do not, however, form a single homogenous group with the same health profiles, and a substantial part of the heterogeneity is likely explained by different childhood body mass index (BMI) trajectories. 3 The seminal publication of Abraham et al 4 in 1971 showed that rates of some cardiovascular diseases were the highest among individuals who were overweight in adulthood but below average weight in childhood. Numerous articles have replicated this type of analysis. 5,6 Others have investigated how BMI trajectories and growth traits differ between subgroups of adults with obesity, which are often classified using crude definitions of metabolic health. 7,8 Fewer studies have, however, used growth mixture modeling to identify latent groups or classes of individuals who share similar individual BMI trajectories and then subsequently investigated how adulthood cardiometabolic health measures or events differ between them. [9][10][11] This is a powerful approach, which does not involve arbitrary categorization, conditioning on an outcome, or imposing linear constraints between growth traits and an outcome. The 2018 European Heart Journal article by Buscot et al 9 used data from 2631 individuals in the Cardiovascular Risk in Young Finns Study and is arguably the best published example. Even this article, however, had serious limitations, including the individual BMI trajectories spanning different age ranges for different birth cohorts (eg, 6-37 years if born in 1974; 18-49 years if born in 1962) and multiple categorical outcomes being related to classes with few individuals (eg, n=33). In addition, no previous study has investigated differences in height growth and body composition trajectories between the classes they identified based on serial BMI data. This is fundamentally important to understand and interpret not only the classes but also their associations with outcomes (eg, one class might have high BMI, yet a healthy cardiometabolic profile due to high lean mass).
Using data from the larger and deeply phenotyped ALSPAC (Avon Longitudinal Study of Parents and Children), we aimed to develop a growth mixture model that properly captured heterogeneity in childhood to youngadulthood BMI trajectories. We further aimed to extensively describe each identified latent class (in terms of height growth, body composition trajectories, and early-life characteristics) and quantify between-class differences in young-adulthood cardiometabolic health measures. We hypothesized that there would be 2 distinct groups of individuals who developed young-adulthood obesity and that the more favorable cardiometabolic health profile would be observed in the group of individuals who were consistently heavy across childhood.

MATERIALS AND METHODS
The ALSPAC data are available to scientists on request via the following website, which also provides details and distributions of the study variables: http://www.bristol.ac.uk/ alspac/researchers/access/. The statistical code for the analyses in this article has been placed in GitHub-the openaccess online repository: https://github.com/tomnorris1988/ ALSPAC-BMI-Mixture-model-. Please also see the Major Resources Table in the Data Supplement. Sample ALSPAC is a prospective birth cohort study. 12,13 Pregnant women living in the defunct county of Avon in England with an expected delivery date between April 1991 and December 1992 were invited to take part in the study. The total number of pregnancies is 15 454, representing 15 589 fetuses of which 14 901 were alive at 1 year of age. The 688 (ie, 15 589-14 901) who did not survive includes miscarriages and fetal loss/death, as well as (a low incidence of) neonatal death. Follow-up has included parent-and child-completed questionnaires, links to routine data, and clinic attendance at 10 sweeps at target ages of ≈7, 8,9,10,11,12,13,15,17, and 24 years. Ethical approval for the study was obtained from the ALSPAC Ethics and Law Committee and the local research ethics committees (and conformed to the Declaration of Helsinki). Informed consent for the use of data collected via questionnaires and clinics was obtained from participants following the recommendations of the ALSPAC Ethics and Law Committee at the time. Consent for biological samples has been collected in accordance with the Human Tissue Act (2004

Cardiometabolic Health Data
The 24-year sweep included detailed phenotyping and data collection using REDCap. 14 Resting systolic and diastolic blood pressure was measured, after a 2 minute rest, using an Omron Healthcare (Kyoto, Japan) M6 upper arm blood pressure monitor. The average of 2 or 3 measurements was used. A fasting blood sample was obtained from which cholesterol, HDL (highdensity lipoprotein) cholesterol, LDL (low-density lipoprotein) cholesterol, triglycerides, insulin, glucose, and CRP (C-reactive protein) were assayed. Carotid-femoral pulse wave velocity was assessed using the oscillometric method (Vicorder; Skidmore Medical, Bristol, United Kingdom). Bilateral carotid intima-media thickness was measured using established ultrasound machine techniques with a 13.5-MHz linear array broadband transducer (CardioHealth; Panasonic, Tokyo, Japan). 15 End-diastolic measurements were recorded in the far wall of the right and left common carotid arteries (1 cm proximal to the carotid bifurcation) and were averaged for analysis.
Approximately 1 in 2 participants attending the 24-year clinic sweep were recruited for detailed cardiac phenotyping. An echocardiogram, using a Philips Medical Systems (North Andover, MA) EPIQ 7G Ultrasound System equipped with a X5-1 transducer, was performed using the American Society of Echocardiography protocols. 16 Based on previous literature and to limit multiple comparisons, 5 clinical parameters were computed. Relative wall thickness and left ventricular mass indexed to height in m 3 are principally indicators of cardiac structure. Peak mitral annular velocity in systole measured by pulsed Doppler (s′) is principally an indicator of systolic function. Early/late mitral inflow velocity (E/A ratio) and early mitral inflow velocity/mitral annular early diastolic velocity (E/e′ ratio) are principally indicators of diastolic function.

Potential Confounder Data
Potential confounding variables of the relationship between BMI trajectories and cardiometabolic health measures in adulthood included sex, ethnicity, birth weight, gestational age, gestational hypertension, diabetes in pregnancy, smoking and alcohol consumption during pregnancy, parental age at birth of baby, maternal BMI at 12 weeks of gestation, paternal BMI at 7-year sweep, breastfeeding duration, parental educational qualifications, parental occupation, household income, and a family adversity index.

Statistical Analyses
Using Mplus, a growth mixture model (sexes combined) was developed to identify distinct groups of individuals who had similar BMI trajectories across the 10 sweeps ( Figure I in the Data Supplement). The data used in, and diagnostics obtained from, these mixture models are included in Tables II through VI in the Data Supplement and Figure II in the Data Supplement.
As a sensitivity analysis, we refitted our final growth mixture model as a multigroup model, in which the growth parameters (ie, intercept, slope, quadratic, and cubic terms) of each trajectory class were allowed to differ by sex. Two figures were produced: one showing the average fitted trajectories (with 95% CIs) between 7.5 and 24.5 years, and the individual observed trajectories, for each class and one showing the average trajectories superimposed on the International Obesity Task Force ranges (averaged across sex) for obesity, overweight, normal weight, and thinness. 17 To compare the height growth curves for each latent class, we used the Super-Imposition by Translation and Rotation model. 18 Class-specific models were developed and used to plot the average height distance (cm) and velocity (cm/ year) curves between 7.5 and 17.5 years for each class, incorporating estimates of age at and magnitude of peak height velocity. Multilevel models were developed to describe how BMI indicates body mass index; cIMT, carotid intima-media thickness; CRP, C-reactive protein; DBP, diastolic blood pressure; HDL-C, high-density lipoprotein cholesterol; IQR, interquartile range; LDL-C, low-density lipoprotein cholesterol; LVMI, left ventricular mass indexed to height in m 3 ; PWV, pulse wave velocity; and SBP, systolic blood pressure. body composition, assessed via dual-energy x-ray absorptiometry, changes over age in each class. A single model was developed for each of the 6 outcomes: whole-body fat and lean masses, trunk fat and lean masses, and peripheral (ie, arms+legs) fat and lean masses. Adjustments for height at each sweep were made. Using the models, we estimated trajectories between 9.5 and 24.5 years, showing how body composition in each of the 5 classes differed compared with a referent class.
Descriptive statistics for each potential confounder, stratified by class, were produced. General linear regression was used to estimate differences in each cardiometabolic health measure between the classes, with adjustment for all potential confounders. The transformation y=100 log(e)× was used for skewed   outcomes, and the resulting estimates are symmetrical percentage differences. These analyses were performed using outcome and confounder data that were multiply imputed 100× using chained equations and were weighted by the posterior probabilities of most-likely class membership (using importance weights). The missing data patterns are shown in Tables VII and VIII in the Data Supplement. As sensitivity analyses, the regression models were rerun 4×. First, stratified by sex; second, not including the weights; third, not including any confounder adjustment; and finally, using a dataset in which only the potential confounders (and not the outcomes) were imputed. Full details regarding the derivation of echocardiogram outcomes, selection of confounding variables, implementation of the mixture modeling, Super-Imposition by Translation and Rotation analysis, multilevel body composition modeling, and multiple imputation have been published in GitHub (repository URL: https://github.com/tomnorris1988/ALSPAC-BMI-Mixture-model-).

RESULTS
Descriptive statistics of the sample are presented in Tables 1 and 2. Briefly, this is a predominantly White British (96.1%) sample of individuals born with a mean birth weight of 3.4 kg, to relatively well-educated (≈50% of parents educated to A level or beyond) parents from relatively high social classes (>55% fathers in the highest 2, of 7, social class strata). At age 24 years, the average BMI of the sample was 23.7 kg/m 2 and the average fat mass was 20.5 kg.

Latent BMI Classes
A mixture model with 6 classes and entropy 0.706 provided the best representation of the serial BMI data and the most plausible solution. Figure 1 shows the average    • Class 5 comprised 35% of the sample and had a nonlinear average trajectory in the normal-weight range, which matches what we would expect due to changes in growth velocity during puberty. This class is referred to as normal weight (nonlinear). • Class 2 comprised 21% of the sample and had a nearlinear average trajectory in the normal-weight range. This class is referred to as normal weight (linear). • Class 3 comprised 18% of the sample and had an average trajectory that increased from normal weight to overweight. This class is referred to as normal weight increasing to overweight. • Class 1 comprised 16% of the sample and had an average trajectory that approximated the International Obesity Task Force overweight threshold. This class is referred to as normal weight or overweight. • Class 6 comprised 6% of the sample and had an average trajectory that increased from normal weight to obesity. This class is referred to as normal weight increasing to obesity. • Class 4 comprised 4% of the sample and had an average trajectory that approximated the International Obesity Task Force obesity threshold. This class is referred to as overweight or obesity.
As shown in Table IX in the Data Supplement, there was no pattern or relationship between the number of serial measurements a child had and the class they were assigned to. Results estimated using multiply imputed data and weighted by posterior probabilities of most likely class membership. BMI indicates body mass index; and CSE, certificate of secondary education. Refitting the growth mixture model as a multigroup model did not provide evidence that the growth parameters characterizing each class' trajectory differed by sex (Table X in the Data Supplement; Figure IX in the Data Supplement).

Height and Body Composition Trajectories
Compared with the normal-weight (nonlinear) class, the 5 other classes were initially taller (on average) but transitioned to being shorter by 17.5 years of age, due to an earlier and lower magnitude of peak height velocity (Figure 3). These differences were most pronounced in the normalweight-increasing-to-obesity and overweight-or-obesity classes. After accounting for differences in height, we observed large differences in fat mass trajectories between the classes but much smaller differences in lean mass trajectories ( Figure 4). The overweight-or-obesity class, for example, was estimated to have 15 to 21 kg more total body fat than the normal-weight (nonlinear) class between 9.5 and 24.5 years but only 2 to 6 kg more lean mass.

Early-Life Characteristics
The percentage of women in the normal-weight-increasing-to-overweight class, and in the normal-weightincreasing-to-obesity class, was ≈9% higher than in the full sample (Table 3). Conversely, the percentage of men in the normal-weight (nonlinear) class was ≈7% higher than in the full sample. Table 3 also shows a clear, and expected, patterning of estimated levels of potential confounders across the classes. For example, maternal BMI increased across the classes, from 22.8 kg/m 2 in the normal-weight (nonlinear) class to 27.6 kg/m 2 in the normal-weight-increasing-to-obesity class, as did rates

Cardiometabolic Outcomes
Estimated differences in each cardiometabolic health measure between the classes are presented in Tables 4  and 5. The normal-weight (linear) class had 20.6% (95% CI, 17.1-24.1) more trunk fat, and higher blood pressures, left ventricular mass indexed to height in m 3 , and insulin, compared with the normal-weight (nonlinear) referent class ( Table 4). The normal-weight-or-overweight and normal-weight-increasing-to-overweight classes fared even worse. Despite the average BMI trajectories for these classes converged at ≈26 kg/m 2 at 24.5 years (Figure 2), the normal-weight-increasing-to-overweight class had worse levels of blood pressures, HDL cholesterol, left ventricular mass indexed to height in m 3 , E/e′ ratio, trunk fat mass, and insulin compared with the normal-weight-or-overweight class ( Table 5). The overweight-or-obesity and normal-weight-increasing-toobesity classes had worse levels of nearly all health measures compared with the normal-weight (nonlinear) class ( Table 4). The normal-weight-increasing-to-obesity class, for example, had 84.4% (72.5-96.2) higher insulin and 110.7% (89.5-131.8) higher CRP. The average BMI trajectory for the normal-weight-increasing-to-obesity class was considerably higher during early adulthood than that for the overweight-or-obesity class (Figure 2), and the former had worse levels of some health measures than the latter (Table 5). s% estimates are symmetrical percentage differences. Results estimated using confounder-adjusted regression models applied to multiply imputed data and weighted by posterior probabilities of most likely class membership. BMI indicates body mass index; cIMT, carotid intima-media thickness; CRP, C-reactive protein; DBP, diastolic blood pressure; HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol; LVMI, left ventricular mass indexed to height in m 3 ; PWV, pulse wave velocity; and SBP, systolic blood pressure. Tables XI through XV in

DISCUSSION
This article provides the most detailed investigation in the literature of latent-childhood to young-adulthood BMI trajectory classes and their cardiometabolic consequences. We found and describe 6 classes in the ALSPAC sample: 2 classes had BMI trajectories that resulted in youngadulthood normal weight, 2 classes had BMI trajectories that resulted in young-adulthood overweight, and 2 classes had BMI trajectories that resulted in youngadulthood obesity. The classes demonstrated remarkable differences in fat mass, but not lean mass, trajectories. Average levels of most cardiometabolic health measures, including echocardiogram measures of heart structure and function, were worse in the classes that led to overweight or obesity. Our key finding, however, is that classes with average BMI trajectories that were high and stable had better cardiometabolic health profiles than classes with average BMI trajectories that demonstrated rapid gain and transitioned in late adolescence from normal weight to overweight or obesity.
A recent systematic review found 14 studies that had used mixture modeling to investigate BMI trajectories from birth onward, 19 and additional articles have considered different age windows in childhood or adulthood. [20][21][22] In our literature search, only 3 publications had applied growth mixture modeling to capture heterogeneity in age-related BMI trajectories spanning childhood into adulthood and subsequently related the latent s% estimates are symmetrical percentage differences. Results estimated using confounder-adjusted regression models applied to multiply imputed data and weighted by posterior probabilities of most likely class membership. BMI indicates body mass index; cIMT, carotid intima-media thickness; CRP, C-reactive protein; DBP, diastolic blood pressure; HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol; LVMI, left ventricular mass indexed to height in m 3 ; PWV, pulse wave velocity; and SBP, systolic blood pressure. classes identified to adulthood cardiometabolic health measures. The studies by Hao et al 10 and Oluwagbemigun et al 11 both had a sample size of only ≈650 individuals and discovered uninformative low, middle, and high BMI trajectory classes. These studies implemented a group-based or latent-class trajectory model, which is a heavily constrained type of mixture model that (unrealistically) assumes no individual variation in growth within each class. This introduces potential problems of meaning and utility of the derived classes. 23 The Buscot et al 9 study of 2631 individuals in the Cardiovascular Risk in Young Finns Study presents a stronger analysis but is not without limitations. They also identified 6 classes, but the average BMI trajectories (between 6 and 49 years) were different compared with those presented in the present article, at comparable ages. This must at least partly be due to the different BMI measurement schedules and age ranges. The authors reported that trajectories of worsening or persistent obesity were associated with increased risk of outcomes including type 2 diabetes and hypertension, but these estimates were based on small numbers (eg, 7 individuals had type 2 diabetes in the persistent increasing overweight/obese class).
Over half of our sample demonstrated 1 of the 2 trajectories that persisted in the normal-weight BMI category. Individuals in these normal-weight trajectory classes subsequently displayed, on average, the most optimal cardiometabolic profiles in young adulthood, thus providing further evidence of the benefits associated with the management of a healthy weight in childhood and adolescence. Approximately 6% of our sample belonged to a class with an average trajectory that started in the normal-weight range at 7.5 years but finished in the obesity range at 24.5 years. This class had the highest carotid intima-media thickness, a subclinical marker of atherosclerosis and a surrogate end point for coronary artery disease, 24 the highest left ventricular mass indexed to height in m 3 , a risk factor for coronary heart disease and heart failure, 25,26 and the highest E/e′ ratio, a correlate of left ventricular diastolic filling pressure that is related to cardiac events. 27 This class also had 50% higher CRPan inflammatory marker related to risk of coronary heart disease, stroke, and vascular mortality 28 -than the class whose average BMI trajectory both started and finished in the obesity range. These differences clearly demonstrate the adverse consequences of rapid adolescent BMI gain. They are also in agreement with a populationbased cohort study in Sweden, which reported (relevant to a normal weight group) a higher hazard ratio for cardiovascular mortality in boys who became overweight during puberty (hazard ratio, 2.39) than in boys who were consistently overweight (hazard ratio, 1.85). 29 The reasons why individuals who become obese during adolescence have worse cardiometabolic health prospects than individuals who have been obese for longer since childhood are not well understood. One explanation, supported by work by Sachdev et al, 30 is that adolescent BMI gains are more strongly related (than childhood BMI gains) to increases in visceral adiposity. 30,31 This was observed here, with the normal weight increasing to obesity class having 28% more trunk fat than the overweight or obesity class. Thereafter, it is well known that centrally stored adipose tissue is associated with increases in blood pressure, low-grade inflammation, insulin resistance, and impaired glucose tolerance. 31,32 Alternatively, this may also reflect a greater adulthood lean mass in those who had a high and stable BMI in childhood compared with those who did not become obese until late adolescence, as it has been shown that before adolescence, increases in BMI are driven primarily by increases in fat-free mass. 33 By modeling height growth curves for each class, we found that the most deleterious classes tended to be taller than their peers initially but transitioned to being shorter than their peers in adulthood, due to an earlier and lower magnitude of peak height velocity. This pattern of growth has been documented before in children with obesity, 34 and both earlier puberty and shorter adulthood height are related to increased risk of cardiometabolic disease-related morbidity and mortality. 35,36 Relevant to the archetypical normal-weight group, the childhood height advantage was the least pronounced and the final height deficit was the most pronounced, in the two classes with average BMI trajectories that demonstrated rapid gain. This suggests that it might be rapid BMI gain rather than obesity per se which results in earlier puberty and shorter adulthood height.
The ALSPAC cohort is predominantly White British, with low levels of socioeconomic deprivation, which may limit generalizability of the findings. Approximately 25% of the initial ALSPAC cohort participated in the 24-year clinic sweep, and differential selection into our sample may have biased results. 37 A pragmatic decision was made to analyze data from both sexes together, as was done by Buscot et al 9 and others. 19 While there are systematic differences in childhood BMI between boys and girls, this is not a reason to hypothesize that there should be a different number of latent classes for each sex. Further, stratifying analyses by sex would have led to smaller classes and reduced the power of our regression models investigating relationship with cardiometabolic outcomes. However, as sensitivity analyses, we refitted our final growth mixture model as a multigroup mode to investigate whether, in each trajectory class, growth parameters differed between the sexes. We did not find strong evidence to suggest that trajectories differed between the sexes. We also ran sex-specific regression models to investigate whether the relationship between the trajectory classes and cardiometabolic health measures differed by sex, and we observed similar patterns of results to those observed with sexes combined. In this second step of the analysis, we did not adjust for BMI or height at 24 years because this would have biased the estimates due to conditioning on a mediator. The carotid intima-media thickness, pulse wave velocity, and echocardiogram outcomes had large amounts of missing data, but results were comparable in the main analysis (which accounted for missing data using multiple imputation) and in a sensitivity analysis in which the outcomes were not imputed.
In conclusion, this article demonstrates how the relationship of young-adulthood overweight or obesity with cardiometabolic health is dependent on the process by which a child becomes overweight or obese. Individuals who have high and stable BMI across childhood may have lower cardiometabolic disease risk than individuals who do not become overweight or obese until late adolescence. At a time when the focus of obesity and related-disease prevention is increasingly targeting early childhood, this is an important consideration for researchers, clinicians, and public health officials.