Possible Modifiers of the Association Between Change in Weight Status

1000 Diabetes Care Volume 43, May 2020

ﬁ Lise G. Bjerregaard,1 Niko Wasenius,2,3 Possible Modi ers of the Rozenn Nedelec,4,5 Line K. Gjærde,1 Lars Angquist,¨ 1,6 Karl-Heinz Herzig,7,8 Association Between Change Gorm B. Jensen,9 Erik L. Mortensen,10 Merete Osler,1,11 Kim Overvad,12 in Weight Status From Child Tea Skaaby,1 Anne Tjønneland,13,14 Thorkild I.A. Sørensen,6,11 Through Adult Ages and Later Marjo-Riitta Jarvelin,¨ 4,5,15,16 Johan G. Eriksson,2,3,17,18 Risk of Type 2 Diabetes Sylvain Sebert,4,5,19 and Jennifer L. Baker1,6 Diabetes Care 2020;43:1000–1007 | https://doi.org/10.2337/dc19-1726

1Center for Clinical Research and Prevention, Bispebjerg and Frederiksberg Hospital, Frederiksberg, Denmark 2Department of General Practice and Primary OBJECTIVE Health Care, University of Helsinki and Helsinki We investigated the association between changes in weight status from childhood University Hospital, Helsinki, Finland through adulthood and subsequent type 2 diabetes risks and whether educa- 3Folkhalsan¨ Research Center, Helsinki, Finland 4 tional attainment, smoking, and leisure time physical activity (LTPA) modify this Centre for Life Course Health Research, Faculty of Medicine, University of Oulu, Oulu, Finland association. 5

EPIDEMIOLOGY/HEALTH SERVICES RESEARCH Biocenter Oulu, University of Oulu, Oulu, Finland 6Novo Nordisk Foundation Center for Basic RESEARCH DESIGN AND METHODS Metabolic Research, Human Genomics and Using data from 10 Danish and Finnish cohorts including 25,283 individuals, Metagenomics in Metabolism, University of childhood BMI at 7 and 12 years was categorized as normal or high using age- and Copenhagen, Copenhagen, Denmark 7Research Unit of Biomedicine, Department of sex-specific cutoffs (<85th or ‡85th percentile). Adult BMI (20–71 years) was Physiology and Biocenter Oulu, Medical Research categorized asnonobese or obese (<30.0 or ‡30.0 kg/m2,respectively). Associations Center, University of Oulu, Oulu University Hos- between BMI patterns and type 2 diabetes (989 women and 1,370 men) were pital, Oulu, Finland 8Department of Gastroenterology and Metab- analyzed using Cox proportional hazards regressions and meta-analysis techniques. olism, Poznan University of Medical Sciences, Poznan, Poland RESULTS 9The Copenhagen City Heart Study, Bispebjerg and Compared with individuals with a normal BMI at 7 years and without adult obesity, Frederiksberg Hospital, Frederiksberg, Denmark 10 those with a high BMI at 7 years and adult obesity had higher type 2 diabetes risks Department of Public Health and Center for – – Healthy Aging, University of Copenhagen, Co- (hazard ratio [HR]girls 5.04 [95% CI 3.92 6.48]; HRboys 3.78 [95% CI 2.68 5.33]). penhagen, Denmark Individuals with a high BMI at 7 years but without adult obesity did not have a higher 11Section of Epidemiology, Department of Public risk (HRgirls 0.74 [95% CI 0.52–1.06]; HRboys 0.93 [95% CI 0.65–1.33]). Education, Health, Faculty of Health and Medical Sciences, smoking, and LTPA were associated with diabetes risks but did not modify or University of Copenhagen, Copenhagen, Denmark 12Section for Epidemiology, Department of Public confound the associations with BMI changes. Results for 12 years of age were Health, Aarhus University, Aarhus, Denmark similar. 13Danish Cancer Society Research Center, Copen- hagen, Denmark CONCLUSIONS 14Section of Environmental Health, Department A high BMI in childhood was associated with higher type 2 diabetes risks only if of Public Health, Faculty of Health and Medical fl Sciences, University of Copenhagen, Copenha- individuals also had obesity in adulthood. These associations were not in uenced by gen, Denmark educational and lifestyle factors, indicating that BMI is similarly related to the risk 15Department of Epidemiology and Biostatistics, across all levels of these factors. MRC-PHE Centre for Environment and Health, School of Public Health, Imperial College London, London, U.K. Even though child and adult BMI (in kilograms per meters squared) are positively 16Department of Life Sciences, College of Health associated with the risk of developing type 2 diabetes, the effects of changes in weight and Life Sciences, Brunel University London, London, U.K. status between childhood and adulthood are not well understood. Previous studies 17Department of Obstetrics and Gynaecology, indicate that children with overweight who remit from it before adulthood may re- Yong Loo Lin School of Medicine, National Uni- duce their risk of type 2 diabetes (1–6). Further, some studies find being persistently versity of Singapore, Singapore care.diabetesjournals.org Bjerregaard and Associates 1001

overweight from childhood to adulthood and in adulthood (.18 years), as well as If individuals participated in more than carries higher risk of type 2 diabetes information on educational attainment, one examination, we preferentially chose than being overweight only in adulthood smoking,andLTPA(Supplementary Table the one at the youngest adult age. (1,2,4). 1). Among the cohorts included in the High BMI in childhood was defined The development of childhood obesity DynaHEALTH consortium, 10 cohorts were as $85th BMI percentile at 7 or 12 years has a strong social gradient (7), and in eligible. Eight were subsamples of adult based on sex- and age-specific BMI per- adult life, inverse associations between research studies (16–22), among whom centiles for the Copenhagen cohorts, the socioeconomic status and educational at- the participants, who were born from NFBC1966 and HBCS, respectively (Sup- tainment and risks of obesity and type 2 1930 to 1981, were included in a Danish plementary Table 2). Obesity in adult- diabetes are well established (8,9). It may school cohort from Copenhagen (23) hood was defined in accordance with the be hypothesized that individuals of low (N 5 19,717) (Supplementary Table 1). World Health Organization criteria (BMI socioeconomic status are more vulner- Additionally, the Northern Finland Birth $30 kg/m2) (30). Patterns of high BMI able to health risks associated with over- Cohort 1966 (NFBC1966) was included. were defined as combinations of high weight. Yet, there is sparse evidence in It includes individuals born in the two BMI in childhood at 7 or 12 years of age this research area (10–12). Only one of northernmost provinces of Finland in and obesity in adulthood. Moreover, a the previous studies focusing on changes 1966 who participated in a clinical exam- BMI pattern with eight categories was in weight status and type 2 diabetes ination in adulthood (N 5 3,985 included) defined accordingtocombinations ofhigh evaluated a potential multiplicative in- (24–26). Finally, the Helsinki Birth Cohort BMIs at 7 years (yes/no) and/or 12 years teractionwith educational attainment on Study (HBCS) contributed with individuals (yes/no) and/or adult obesity (yes/no). these associations (4). Moreover, pre- who were born between 1934 and 1944 At the time of the adult BMI assess- viousstudies are limited byincluding only at the Helsinki University Central Hospi- ment, information on educational attain- one sex (1,4,5), having a low number of tal, attended child welfare clinics in the ment, current smoking (yes/no), and LTPA diabetes case subjects (n , 237) (2,3,6), city, went to school in the city of Helsinki, was obtained by questionnaire. Educa- or ending follow-up in middle adulthood and attended a clinical examination in tional attainment was categorized into (3,6). Further, smoking and lower levels adulthood (N 5 1,581 included) (27). The 0–7 (short), 8–10 (medium), or .10 (long) of leisure time physical activity (LTPA) are cohorts are described in detail elsewhere years of schooling. One cohort used cat- associated with higher risks of develop- (16–24,27). egorizations of 7–9, 10, and .10 years ing type 2 diabetes (13,14), but these All studies were performed in accor- due to the definition in the questionnaire were not investigated as modifying fac- dance with the Helsinki 2 Declaration. The (17). The available information in each tors in the studies on changes in weight project was conducted on anonymous cohort consisted of three to six levels of status and type 2 diabetes. data, and it was approved by the Danish either LTPA or energy expenditure, and it The aim of this study was toinvestigate Data Protection Agency (Datatilsynet). was categorized into three groups cor- the association between changes in weight responding to low (,2 h/week), medium status from childhood to adulthood and Assessment of Variables (light physical activity 2–4 h/week), and the risk of developing type 2 diabetes in Body height and weight were used for high (light physical activity $4 h/week Danish and Finnish cohorts and whether calculation of BMI (weight in kilograms or moderate activity $2 h/week) (HBCS: this association is modified by socioeco- divided by height in meters squared) and less than two times per week, two to four nomic and lifestyle factors. It was hypoth- prospectively measured at school health times per week, and greater than four esized that high BMI at child and adult ages examinations and recorded in health re- times per week). has stronger associations with type 2 di- cords. In the Danish cohorts, BMI at the abetes among individuals with short edu- exact ages of 7 and 12 years were ob- Assessment of Type 2 Diabetes cational attainment, those who smoked, tained by interpolation between succes- Information on type 2 diabetes was ob- and individuals with low levels of LTPA. sive measurements around that age or by tained by linking unique personal identi- extrapolation always within 612 months fication numbers (27,31) of the cohort RESEARCH DESIGN AND METHODS (28). In the Finnish cohorts, BMIs from participants to computerized and com- Study Population $6.0 to ,8 years of age and from $11.0 prehensive health registers. During the The study population was drawn from to ,13 years of age were included as follow-upperiod,,0.6%oftheindividuals cohorts participating in the DynaHEALTH 7-year or 12-year measurements, respec- emigrated, they were censored on this consortiumthat aimstobuild an empirical tively (29). In adulthood, weight and date, and no individuals were untraceable. model of healthy aging (15). To be eligible height were measured at clinical exami- Thus, loss to follow-up was minimal. for this study, we required information on nations in seven cohorts (18–22,24) and In the Danish cohorts, information on weight and height at 7and 12 years of age self-reported in three cohorts (16,17,27). inpatient and outpatient diagnoses of

18SingaporeInstitute for Clinical Sciences, Agency Received 28 August 2019 and accepted 4 Feb- © 2020 by the American Diabetes Association. for Science, Technology, and Research, Singapore ruary 2020 Readers may use this article as long as the work is 19 Department of Genomics of Complex Diseases, This article contains Supplementary Data online properly cited, the use is educational and not for ﬁ School of Public Health, Imperial College London, at https://care.diabetesjournals.org/lookup/suppl/ pro t, and the work is not altered. More infor- London, U.K. doi:10.2337/dc19-1726/-/DC1. mation is available at https://www.diabetesjournals Corresponding author: Lise G. Bjerregaard, lise .org/content/license. [email protected] 1002 BMI Changes and Type 2 Diabetes Diabetes Care Volume 43, May 2020

type 2 diabetes (ICD-8 code 250 and ICD- researchers responsible for each cohort (475,056 total person-years). We found 10 codes E11, E13, and E14) was obtained estimated the stratum-specificregres- similar associations between the BMI by linkage to the National Patient Reg- sion coefficients and the corresponding patterns and risk of type 2 diabetes in ister (32) as previously described (4). In SEs by sex, and these were pooled using women and men, and these were not this register, although the completeness random-effects meta-analyses techniques. different in the meta-regression analyses is moderate (sensitivity 64%), the posi- As major differences between women and (P . 0.33; data not shown). tive predictive value of a diabetes di- men exist in the prevalence, pathophysi- Smokers, individuals with long educa- agnosis is very high (97%) when assessed ology, treatment response, and outcome tion, or individuals with high levels of against information from and verification of type 2 diabetes (37), women and men LTPA were more often in the groups who of the diagnosis by general practitioners were analyzed separately. never had a high BMI or had a high BMI (33). In the NFBC1966 and the HBCS, in The analyses were adjusted for age at only as a child (Table 1). The number of addition to hospital and prescription reg- adult BMI and additionally for educational individuals with type 2 diabetes by these isters (34), clinical examination values attainment,smoking, andLTPA.Potential factors is included in Supplementary (subsequent to obtaining the adult BMI) heterogeneity in the associations in sub- Table 3. Among women and men, high of plasma glucose $7 mmol/L and 2-h groups of the following potential effect child BMI and adult obesity were each plasma glucose $11.1 mmol/L in a 2-h modifiers: sex, educational attainment, positively associated with risks of type 2 75-g oral glucose tolerance test were smoking, LTPA, birth cohort (1930–1939, diabetes (Table 2). As expected, men and used to identify type 2 diabetes using 1940–1949, and 1950–1989) and study, smokers had higher risks of type 2 di- World Health Organization criteria (35). was investigated using the I2 statistic, the abetes, and educational attainment and Furthermore, in the HBCS, HbA1c $6.5% Cochran Q test, and meta-regression. To LTPA were inversely associated with risks or 48 mmol/mol was used to identify assess the proportional hazards assump- of type 2 diabetes (Table 2). type 2 diabetes. Through the combina- tion, we tested heterogeneity by cate- Figure 1 shows the association be- tion of these methods, the sensitivity is gories of age at diagnosis divided into tween the BMI pattern and type 2 di- likely very high, although a few individ- ,70 and $70 years, as graphs of the abetes by levels of the potential effect uals with type 2 diabetes may not be cumulative hazard from one weight cat- modifiers. At higher educational levels in identified if they missed the clinical ex- egory versus another showed that the women and men, there was a tendency amination and were treated only with slope changed at around this age (data for the HRs for persistently high BMI from diet in general practice (36). not shown). To assess the impact of self- 7 years of age to adulthood or only high To achieve a baseline population free reported BMI and reverse causality, we adult BMI and type 2 diabetes to be from diabetes, only individuals without performed sensitivity analyses omitting higher than those for women and men type 2 diabetes at the adult BMI assess- the three cohorts in which BMI was self- who never had a high BMI (Fig. 1A and B). ment were followed prospectively in reg- reported and by omitting the first 3 years These tendencies, however, were gen- isters from the age at the adult BMI of follow-up after the adult BMI measure. erally not supported by the subgroup measure or from 30 years of age for Additionally, weperformedan analysis meta-analysis and the meta-regression Danish individuals (4), whichever came of adult BMI adjusted for child BMI and analyses (Supplementary Table 4). In women, last. As such, 266 individuals with type 2 summarizedtheresultsinameta-analysis. smoking minimally influenced the asso- diabetes were not included in this study An investigation of nonlinearity by mod- ciations between BMI patterns and type from the Danish data, 172 from the HBCS, eling the mutually adjusted associations 2 diabetes (Fig. 1C). Nonsmoking men and 8 from the NFBC1966. The follow-up using cubic splines with seven knot points who only had a high BMI in childhood ended at the date of a type 2 diabetes di- showed changes in the slope of the had a lower risk of type 2 diabetes than agnosis, death, emigration, loss to follow- associations at the ;90th percentile nonsmoking men who never had a high up, or at the end of the follow-up period, for child BMIs and at 30 kg/m2 for adult BMI, but associations in men with per- which was 31 December 2013 for the HBCS BMIs. Therefore, the associations were sistently high BMI or high adult BMI were and 31 December 2016 for the NFBC1966 estimated using a linear spline regression positive in both smokers and nonsmokers and the Danish cohorts, whichever came with knot points inserted at the sex- and (Fig. 1D and Supplementary Table 4). In first. Hence, the adult BMI measurement region-specific 90th percentile for child women and men, the associations be- was collected before information on type BMIs and at 30 kg/m2 for adult BMIs. HRs tween BMI patterns and type 2 diabetes 2 diabetes to avoid potential reverse cau- for specific linear combinations of child remained similar across levels of LTPA sality from effects of type 2 diabetes on and adult BMI are reported. All statistical in the subgroup analyses and the meta- weight gain or loss. analyses were performed using Stata (ver- regression (Fig. 1E and F and Supplemen- sion 14.2; StataCorp, College Station, TX). tary Table 4). Statistical Methods Women and men who had a high BMI Associations between high BMI at each RESULTS only as a child had a similar risk of type 2 age or patterns of high BMI and adult Among the 12,277 women and 13,006 diabetes to that among women and men type 2 diabetes were estimated with haz- men included in the study, 989 women who never had a high BMI (women: HR ard ratios (HRs) and 95% CIs using Cox (8.1%) and 1,370 men (10.5%), respec- 0.74 [95% CI 0.52–1.06]; men: HR 0.93 proportional hazards regression. Age was tively, were diagnosed with type 2 di- [95% CI 0.65–1.33]) (Fig. 2). Women and used as the time scale, implying delayed abetes from 1978 to 2016. The ages at menwithapersistentlyhighBMIorahigh entry at the time of BMI assessment, and diagnosis ranged from 30 to 85 years over BMI only as an adult had higher risks of stratified by year of birth. Specifically, an average follow-up time of 18.8 years type 2 diabetes compared with women care.diabetesjournals.org Bjerregaard and Associates 1003

Table 1—Number of individuals in the study population by patterns of BMI and educational attainment, smoking, and LTPA* Women Men Never high High child High adult Persistently high Never high High child High adult Persistently high BMI BMI BMI BMI BMI BMI BMI BMI Education† Short 1,803 (71.6) 248 (9.8) 351 (13.9) 116 (4.6) 2,365 (72.1) 286 (8.7) 448 (13.7) 180 (5.5) Medium 5,340 (75.7) 737 (10.4) 667 (9.5) 310 (4.4) 5,296 (76.3) 755 (10.9) 604 (8.7) 284 (4.1) Long 2,107 (77.9) 342 (12.6) 168 (6.2) 88 (3.3) 2,185 (78.4) 350 (12.6) 156 (5.6) 97 (3.5) Current smoking No 5,604 (75.1) 713 (9.6) 839 (11.2) 304 (4.1) 5,190 (75.2) 685 (9.9) 715 (10.4) 315 (4.6) Yes 3,646 (75.7) 614 (12.7) 347 (7.2) 210 (4.4) 4,656 (76.3) 706 (11.6) 493 (8.1) 246 (4.0) LTPA‡ Low 2,469 (73.2) 352 (10.4) 378 (11.2) 175 (5.2) 2,543 (74.6) 355 (10.4) 343 (10.1) 166 (4.9) Medium 4,463 (76.0) 627 (10.7) 553 (9.4) 227 (3.9) 4,513 (75.0) 656 (10.9) 581 (9.7) 267 (4.4) High 2,318 (76.4) 348 (11.5) 255 (8.4) 112 (3.7) 2,790 (77.9) 380 (10.6) 284 (7.9) 128 (3.6) DataareN(%).*ThepatternsofBMIweredefinedasfollows:1)neverhighBMI:,85thpercentileinchildhoodand,30kg/m2 inadulthood;2)highchild BMI: $85th percentile in childhood and ,30 kg/m2 in adulthood; 3) high adult BMI: ,85th percentile in childhood and $30 kg/m2 in adulthood; and 4) persistently high BMI: $85th percentile in childhood and $30 kg/m2 in adulthood. †Educational attainment was categorized into years of school: 0–7 years, 8–10 years, or .10 years. One cohort used categorizations of 7–9 years, 10 years, and .10 years due to the definition of the questionnaire (17). ‡LTPA was defined as low: ,2 h/week; medium: light physical activity 2–4 h/week; and high: light physical activity $4 h/week or moderate activity .2 h/week (HBCS: less than two times per week, two to four times per week, and greater than four times per week). and men who never had a high BMI (HR BMI at 7 and 12 years of age and in above that of the reference group. The range: 3.78–5.27) (Fig. 2). Notably, adjust- adulthood(Supplementary Fig.6).Incon- highest risk was observed among indi- ment for educational attainment, smok- trast, women who had a high BMI at viduals who started at the 25th BMI per- ing, and LTPA minimally changed the 12 years of age but not at 7 years and who centile in childhood and ended at a BMI estimates and the CIs. In women, the I2 did not have obesity in adulthood had a of 35 kg/m2 (i.e., individuals who in- ranged from 0 to 41%, indicating low to higher risk of type 2 diabetes (HR 1.55 creased their degree of adiposity the moderate heterogeneity across cohorts. [95% CI 1.12–2.13]). Women who had most) (Supplementary Table 6). Adjust- Among men, the I2 ranged from 26.9 to obesity as adults had a higher risk of ment for educational attainment, smok- 53.2% (Supplementary Appendix). Simi- type 2 diabetes irrespective of their ing, and LTPA minimally changed the lar results were observed for weight childhood BMI status. Although most results (Supplementary Table 6). status at 12 years of age combined with results were similar for men, there adulthood (Supplementary Fig. 1). was an exception. Men who had a CONCLUSIONS In subgroup analyses investigating the high BMI at 12 years but not at 7 years This study showed that in 10 Danish and effects of diabetes diagnosed at ,70 and who did not have obesity in adult- Finnish cohorts, associations between years, we observed stronger associations hood had a similar risk of type 2 diabetes BMI patterns and risk of type 2 diabetes for persistently high BMI and high adult as men who never had a high BMI. were virtually not confounded or modified BMI with type 2 diabetes diagnosed at Adjustment for educational attainment, by educational attainment, smoking in ,70 years than after this age (Supple- smoking, and LTPA minimally changed women, and LTPA. Apart from smoking, mentary Fig. 2). However, the CIs for the results (Supplementary Fig. 6). which influenced one of the associations diagnoses after 70 years were wide, and The regression coefficients for child- in men, the results were similar for men the overall conclusions were the same. In hood and adult BMI from the linear spline and women. This study confirmed that a analyses investigating potential birth co- model generally were similar across lev- high BMI in childhood combined with hort effects, associations between high els of education, smoking, and LTPA in obesity in adulthood is associated with BMIs and type 2 diabetes were stronger the meta-analysis (Supplementary Table higher risks of type 2 diabetes, whereas a in later birth cohorts, but the patterns of 5). Based upon these results, the point high childhood BMI combined with non- associations were the same in all birth estimate for children who had a BMI at obesity in adulthood is not. cohorts (Supplementary Fig. 3). Omission the 95th percentile at 7 years and an Individuals who developed obesity in of the two cohorts with self-reported adult BMI of 30 kg/m2 (corresponding to adulthood had about the same risk of weight and height or restriction of fol- the ;86th percentile) showed they had a having type 2 diabetes as those who had low-up time to from 3 years after the fivefold risk of type 2 diabetes compared a high BMI in childhood and obesity in BMI assessment and onwards minimally with children who had a stable BMI at the adulthood. We also found that the BMI changed the results (Supplementary Figs. 50th percentile at 7 years and an adult trajectory associated with the highest 4 and 5). BMI of 22 kg/m2 (Supplementary Table risk was the one that started at the 25th Women who had a high BMI at 7 years 6). If a child with a BMI at the 95th per- BMI percentile in childhood and ended at and did not have obesity as an adult, centile moved toward lower percentiles an adult BMI of 35 kg/m2; in other words, irrespective of their childhood BMI status and had an adult BMI of 25 kg/m2 among individuals who increased the at 12 years, had a similar risk of type 2 (corresponding to the 50th percentile), most in adiposity. These findings are diabetes as women who never had a high the risk was much lower, although still supported by a large women-only study 1004 BMI Changes and Type 2 Diabetes Diabetes Care Volume 43, May 2020

Table 2—Meta-analysis of HR and 95% CIs for the risk of type 2 diabetes for women and men with a high BMI at 7 years, 12 years, or in adulthood and for educational attainment, smoking, and LTPA* Women (N 5 12,277)† Men (N 5 13,006)‡ Variable N (%) HR (95% CI) N (%) HR (95% CI) High child BMI, 7 years No 10,436 (85) Reference 11,054 (85) Reference Yes 1,841 (15) 1.30 (1.10–1.53) 1,952 (15) 1.28 (0.97–1.69) High child BMI, 12 years No 10,435 (85) Reference 11,056 (85) Reference Yes 1,842 (15) 1.92 (1.54–2.39) 1,950 (15) 1.55 (1.36–1.78) Obesity, adulthood No 10,577 (86.2) Reference 11,237 (86.4) Reference Yes 1,700 (13.8) 5.25 (4.38–6.29) 1,769 (13.6) 4.21 (3.45–5.14) Education§ Short 2,518 (20.5) 1.17 (1.01–1.38) 3,279 (25.2) 1.19 (0.96–1.48) Medium 7,054 (57.4) Reference 6,939 (53.4) Reference Long 2,705 (22.0) 0.78 (0.54–1.12) 2,788 (21.4) 0.84 (0.68–1.04) Current smoking No 7,460 (60.8) Reference 6,905 (53.1) Reference Yes 4,817 (39.2) 1.30 (1.14–1.48) 6,101 (46.9) 1.29 (1.16–1.44) LTPA| Low 3,374 (27.5) 1.34 (1.16–1.56) 3,407 (26.2) 1.29 (1.13–1.48) Medium 5,870 (47.8) Reference 6,017 (46.3) Reference High 3,033 (24.7) 0.90 (0.76–1.06) 3,582 (27.5) 0.92 (0.80–1.05) *Theresults forhighchild BMI,overweight, and obesityareadjustedforage at adultBMI,educationalattainment,smoking,andLTPA, and the resultsfor educational attainment, smoking, and LTPA are mutually adjusted. †Among women, low heterogeneity among the cohorts was observed for the 2 associations between high child BMI or adult obesity and type 2 diabetes (I 5 0.0–34.1%; all PQ values .0.10). ‡Among men, moderate to high 2 2 heterogeneity was observed for the associations of a high BMI at 7 years (I 5 59.3%; PQ , 0.01) and obesity in adulthood (I 5 43.1%; PQ 5 0.04). §Educational attainment was categorized into years of school: 0–7 years, 8–10 years, or .10 years. One cohort used categorizations of 7–9 years, 10 years, and .10 years due to the deﬁnition of the questionnaire (17). |LTPA was deﬁned as low: ,2 h/week, medium: light physical activity 2–4h/ week, and high: light physical activity $4 h/week or moderate activity .2 h/week (HBCS: less than two times per week, two to four times per week, and greater than four times per week).

(38) that reported that those who were by remission of high BMI in men and higher risk of type 2 diabetes associated lean at 8 years and who had a sharp women. with lower levels of physical activity was increase in self-reported body shape at Even when socioeconomic and lifestyle evident in normal-weight and overweight puberty and thereafter had an almost factors were accounted for, the associ- women, but not in women with obesity threefold higher risk of developing type 2 ations between BMI patterns and type 2 (13). diabetes compared with women whose diabetes changed little. Thus, the group Our analyses revealed a difference in body shape stayed in the midrange. This with a high child BMI only did not have a the associations between the BMI pat- risk tended to be even higher compared higher risk at any level of education and tern and type 2 diabetes in adulthood with always having a large body shape LTPA, and those who had a persistently by current smoking status in men. Non- (38). high BMI or developed obesity had con- smoking men with high child BMI only Conversely, a decrease in BMI percen- sistently higher risk at all levels of edu- had a lower risk of type 2 diabetes than tile from childhood to adulthood was cation and LTPA. This suggests that BMI the nonsmoking men who never had a associated with a lower, although still changes affect the risk of type 2 diabetes high BMI. We did not identify any other increased, risk of type 2 diabetes. This in the same way across levels of these studies examining an interaction between finding corresponds with those from a factors. These results are in accord with smoking and the overweight pattern from British study in which remission of obe- our previous findings in men (4), and the childhood to adulthood. In adults, a large sity between childhood (7–16 years of current study extends these to women. meta-analysis reported an interaction be- age) and adulthood (23–45 years of age) We have not identified any studies re- tween smoking and overweight on the was associated with a higher risk of type 2 porting on potential interactions between riskoftype2diabetessuchthattheeffects diabetes compared with individuals who LTPA and patterns of overweight from of smoking were stronger in overweight had never had obesity (6). Other studies childhood to adulthood on risks of type or obese individuals as compared with show that remission of overweight at 2 diabetes. In middle-aged individuals, a effects in normal-weight individuals (P , 8 years (5) and obesity between 4 and large case-cohort study in the European 0.001) (14). 19 years of age (3) and adulthood are not Prospective Investigation into Cancer and When using a pattern of three BMI associated with a difference in the risk of Nutrition cohort reported an interaction values, we found that men and women type 2 diabetes. Together, these findings between physical activity and BMI measured who had a high BMI at 7 and 12 years indicate that the adverse effect of a at middle age only in women on the risk of had a higher risk of developing type 2 high child BMI is at least partly reversible type 2 diabetes (P 5 0.008 in women). The diabetes only if obesity was present in care.diabetesjournals.org Bjerregaard and Associates 1005

Figure1—Meta-analysisofHRsand95%CIsbylevelsofeducation(AandB),smoking(CandD),andLTPA(EandF)fortheassociationbetweentheweight pattern from7 yearsof agetoadulthood andthe risk of type2 diabetesin women andmen.A: Womenbyeducationalattainment.B: Menby educational attainment. C: Women by smoking. D: Men by smoking. E: Women by LTPA. F: Men by LTPA. A high childhood BMI is deﬁned by the 85th BMI percentile. The results are adjusted for age at adult BMI, educational attainment, smoking, and LTPA (unless stratiﬁed on the factor).

adulthood. This ﬁnding is in accord with but not during early adulthood had a of age but not at 34 years had a slightly results from a study in three British birth higher risk of type 2 diabetes as com- higher risk of type 2 diabetes than women cohorts followed into middle age (2). Our pared with men who had never been who had never been overweight (1). It is previous study yielded a slightly different overweight(4).Moreover,alargewomen- possible that these differences are due to result, as we found that men who had only study found that women with over- the size of the studies as the two largest been overweight at 7 and 13 years of age weight(by somatotype)at10and18 years studies showed evidence of an increased 1006 BMI Changes and Type 2 Diabetes Diabetes Care Volume 43, May 2020

the University Hospital of Copenhagen, for the ABcollection of the Copenhagen Perinatal Cohort data; all of those who initiated and/or continued the Metropolit study at the Institute of Soci- ology, University of Copenhagen: K. Svalastoga, E.Høgh,P.Wolf,T.Rishøj,G.Strande-Sørensen, E. Manniche, B. Holten, I.A. Weibull, and A. Ortman; and all of the cohort members and Hazard ratio (95% CI) Hazard ratio (95% CI) researchers who participated in the NFBC1966 and HBCS studies. The authors also acknowl- edge the work of the NFBC project center. Funding. The Diet, Cancer and Health study was funded by the Danish Cancer Society. This project has received funding from the European Union’s Figure 2—Meta-analysis of HRs and 95% CIs for the risk of type 2 diabetes for women (A) and men Horizon2020researchandinnovationprogramme (B) with a high BMI at 7 years, obesity in adulthood, or a high BMI at 7 years and obesity in under grants 633595 DynaHEALTH, 733206 Life- adulthood, respectively, compared with individuals with a BMI below the cutoff for high BMI at Cycle, and 824989 EUCANCONNECT; the Joint 7 years and in adulthood. A high childhood BMI is defined by the 85th BMI percentile. The results Programming Initiative: A Healthy Diet for a are adjusted for age at adult BMI (open circles) or for age at adult BMI, educational attainment, HealthyLifeNetherlandsgrant agreement P75416 smoking, and LTPA (filled circles). PREcisE; and Novo Nordisk Foundation grant NNF17OC0028338. NFBC1966 received financial support from University of Oulu grants 65354 risk among those with overweight in child- pubertal status, LTPA was self-reported, and 24000692; Oulu University Hospital grants hood and adolescence (1,4). and we used BMI as an indicator of ad- 2/97, 8/97, and24301140;Ministryof Healthand The strengths of this study are that we iposity. BMI is a proxy for adiposity (39), Social Affairs grants 23/251/97, 160/97, and 190/97; had measured weights and heights in 7 so we do not know whether the changes National Institute for Health and Welfare, Hel- sinki grant 54121; Regional Institute of Occupa- (70%) of the cohorts, limiting the poten- in the risk of type 2 diabetes are due to tional Health, Oulu, Finland grants 50621 and tial for recall bias, we included men and changes in lean or fat massor the location 54231; and ERDF European Regional Develop- women and information on lifestyle fac- of the fat mass. This is important, as ment Fund grant 539/2010 A31592. tors, and we were able to follow indi- people without obesity also develop di- Duality of Interest. No potential conflicts of interest relevant to this article were reported. viduals to late adult ages. Rather than abetes, dependent on their fat mass (40). Author Contributions. L.G.B., T.I.A.S., M.-R.J., searching for published studies, we in- Moreover,thestudypopulations included J.G.E., S.S., and J.L.B. conceived the study. All cluded eligible cohorts in the DynaHEALTH only few children with severe obesity and, authors were involved in the design of the study. consortium (15), which circumvented the due to the racial composition of Denmark K.-H.H.,G.B.J., E.L.M., M.O., K.O., T.S.,A.T., T.I.A.S., potential for publication bias usually en- and Finland at these times, were pre- M.-R.J., J.G.E., S.S., and J.L.B. provided data. L.G.B., N.W., R.N., andS.S. analyzed data.Allauthors countered in meta-analysis on previously dominantly of Caucasian descent. Whether were involved in the data interpretation. L.G.B. published results. The random-effect meta- the associations differ in groups with and J.L.B. drafted the manuscript. All authors analysis model allowed any potential het- severe obesity, by race, or by ethnicity contributed to revision and approval of the final erogeneity in the associations across cohorts requires further evaluations. Lastly, sim- manuscript. L.G.B. and J.L.B. are the guarantors of to occur around a normally distributed ilar to other studies, we did not have the this work and, as such, had full access to all of the data in the study and take responsibility for the mean effect. Moreover, a wide range of age at onset of obesity. integrity of the data and the accuracy of the data birth cohorts was included. We found In conclusion, in the 10 Danish and analysis. slightly stronger associations in postwar Finnish cohorts studied, a high BMI in Prior Presentation. This study was presented at generations, butthe overallpatternswere childhoodcombined withobesity inadult- the 26th European Congress on Obesity, Glas- – the same. This suggests that the results hood was associated with higher risks of gow, U.K., 28 April 1 May 2019. are applicable to multiple generations, developing type 2 diabetes, whereas a References including contemporary populations, as high childhood BMI combined with non- 1. Yeung EH, Zhang C, Louis GM, Willett WC, Hu some individuals in the study were born obesity in adulthood was not. These as- FB. Childhood size and life course weight char- as recently as 1981. Further, sensitivity sociations were virtually similar across acteristics in association with the risk of incident analyses examining the effects of self- levels of educational and lifestyle factors, type 2 diabetes. Diabetes Care 2010;33:1364– reported weight and height and potential suggesting that BMI affects the risk of 1369 effects of reverse causality did not change type 2 diabetes in the same way across 2. Park MH, Sovio U, Viner RM, Hardy RJ, Kinra S. Overweight in childhood, adolescence and these associations. Nevertheless, further levels of these other risk factors. Thus, adulthood and cardiovascular risk in later life: studies are needed to investigate whether public health initiatives should focus on pooled analysis of three british birth cohorts. the findings are generalizable to other preventing the continuation of adiposity PLoS One 2013;8:e70684 settings. from childhood into adulthood irrespec- 3. Juonala M, Magnussen CG, Berenson GS, et al. The study has some limitations. The tive of educational level, and individuals Childhood adiposity, adult adiposity, and cardiovascular risk factors. N Engl J Med 2011;365: sample sizes may, in some strata, have with all levels of physical activity may benefit 1876–1885 been too small to show effect modifica- from health-promoting interventions. 4. Bjerregaard LG, Jensen BW, Angquist¨ L, Osler M, tion. The estimates for some categories Sørensen TIA, Baker JL. Change in overweight from were imprecise, but this reflects that cer- childhood to early adulthood and risk of type 2 Acknowledgments. The authors thank the late diabetes. N Engl J Med 2018;378:1302–1312 tain groups such as high child BMI only Drs. Aage Willumsen, Maternity Department A of 5. OhlssonC,BygdellM,NethanderM,Rosengren have a limited risk of developing type 2 theUniversityHospitalofCopenhagen,andBengt A, Kindblom JM. BMI change during puberty is an diabetes.Wedidnothave information on Zachau-Christiansen, the Pediatric Department of important determinant of adult type 2 diabetes care.diabetesjournals.org Bjerregaard and Associates 1007

risk in men. J Clin Endocrinol Metab 2019;104: in midlife: the Metropolit 1953 Danish male birth childhood and death from coronary heart dis- 1823–1832 cohort. Int J Epidemiol 2008;37:1367–1374 ease: longitudinal study. BMJ 1999;318:427– 6. Power C, Thomas C. Changes in BMI, duration of 18. Osler M, Linneberg A, Glumer¨ C, Jørgensen T. 431 overweight and obesity, and glucose metabolism: The cohorts at the Research Centre for Prevention 30. World Health Organization. Physical Status: 45 years of follow-up of a birth cohort. Diabetes and Health, formerly ‘The Glostrup Population The Use and Interpretation of Anthropometry. Care 2011;34:1986–1991 Studies’. Int J Epidemiol 2011;40:602–610 Report of a WHO Expert Committee. WHO Tech- 7. Shrewsbury V, Wardle J. Socioeconomic status 19. Bak H, Petersen L, Sørensen TIA. Physical nical Report Series 854. Geneva, World Health and adiposity in childhood: a systematic review of activity in relation to development and main- Organization, 1995 cross-sectional studies 1990-2005. Obesity (Silver tenance of obesity in men with and without 31. Pedersen CB. The Danish Civil Registration Spring) 2008;16:275–284 juvenile onset obesity. Int J Obes Relat Metab System. Scand J Public Health 2011;39(Suppl.): 8. Stringhini S, Tabak AG, Akbaraly TN, et al. Disord 2004;28:99–104 22–25 Contribution of modifiable risk factors to social 20. Tjønneland A, Olsen A, Boll K, et al. Study 32. Andersen TF, Madsen M, Jørgensen J, inequalities in type 2 diabetes: prospective White- design, exposure variables, and socioeconomic Mellemkjoer L, Olsen JH. The Danish National hall II cohort study. BMJ 2012;345:e5452 determinants of participation in Diet, Cancer and Hospital Register. A valuable source of data for 9. Stringhini S, Batty GD, Bovet P, et al. Asso- Health: a population-based prospective cohort modern health sciences. Dan Med Bull 1999;46: ciation of lifecourse socioeconomic status with study of 57,053 men and women in Denmark. 263–268 chronic inflammation and type 2 diabetes risk: Scand J Public Health 2007;35:432–441 33. Kristensen JK, Drivsholm TB, Carstensen B, the Whitehall II prospective cohort study. PLoS 21. Appleyard M, Hansen AT, Schnohr P, Jensen Steding-Jensen M, Green A. [Validation of meth- Med 2013;10:e1001479 G, Nyboe J; The Copenhagen City Heart Study ods to identify known diabetes on the basis of 10. HeraclidesAM,ChandolaT,WitteDR,Brunner Group. The Copenhagen City Heart Study. Oster- health registers]. Ugeskr Laeger 2007;169:1687– EJ. Work stress, obesity and the risk of type 2 broundersøgelsen. A book of tables with data 1692 [ in Danish] diabetes: gender-specific bidirectional effect in the from the first examination (1976-78) and a five 34. Eriksson JG, Forsen TJ, Osmond C, Barker DJ. WhitehallIIstudy.Obesity(SilverSpring)2012;20: year follow-up (1981-83). Scand J Soc Med Suppl Pathways of infant and childhood growth that 428–433 1989;41:1–160 lead to type 2 diabetes. Diabetes Care 2003;26: 11. Lee TC, Glynn RJ, Pena~ JM, et al. Socioeco- 22. Schnohr P, Jensen GB, Lange P, Scharling H, 3006–3010 nomic status and incident type 2 diabetes melli- Appleyard M. The Copenhagen City Heart 35. International Diabetes Federation; World tus: data from the Women’s Health Study. PLoS Study: Østerbroundersøgelsen: Tables With Data Health Organization. Definition and Diagnosis One 2011;6:e27670 From the Third Examination 1991-1994. London, of Diabetes Mellitus and Intermediate Hyper- 12. Yang MH, Hall SA, Piccolo RS, Maserejian NN, Saunders, 2001 glycemia: Report of a WHO/IDF Consultation. McKinlay JB. Do behavioral risk factors for pre- 23. Baker JL, Olsen LW, Andersen I, Pearson S, Geneva, World Health Organization, 2006, p. 50 diabetes and insulin resistance differ across the Hansen B, Sørensen TIa. Cohort profile: the 36. Wikstrom¨ K, Toivakka M, Rautiainen P, socioeconomic gradient? Results from a com- Copenhagen School Health Records Register. TirkkonenH,RepoT,Laatikainen T. Electronic munity-based epidemiologic survey. Int J Endo- Int J Epidemiol 2009;38:656–662 health records as valuable data sources in crinol 2015;2015:806257 24. Jarvelin¨ MR, Sovio U, King V, et al. Early life the health care quality improvement process. 13. Ekelund U, Palla L, Brage S, et al.; InterAct factors and blood pressure at age 31 years in the Health Serv Res Manag Epidemiol 2019;6: Consortium. Physical activity reduces the risk of 1966 northern Finland birth cohort. Hyperten- 2333392819852879 incident type 2 diabetes in general and in abdom- sion 2004;44:838–846 37. Arnetz L, Ekberg NR, Alvarsson M. Sex differ- inally lean and obese men and women: the EPIC- 25. Lowry E, Rautio N, Karhunen V, et al. Un- ences in type 2 diabetes: focus on disease course InterAct Study. Diabetologia 2012;55:1944–1952 derstanding the complexity of glycaemic health: andoutcomes.DiabetesMetabSyndrObes2014; 14. Willi C, Bodenmann P, Ghali WA, Faris PD, systematic bio-psychosocial modelling of fasting 7:409–420 Cornuz J. Active smoking and the risk of type 2 glucose in middle-age adults; a DynaHEALTH study. 38. Fagherazzi G, Vilier A, Affret A, Balkau B, diabetes: a systematic review and meta-analysis. Int J Obes 2019;43:1181–1192 Bonnet F, Clavel-Chapelon F. The association of JAMA 2007;298:2654–2664 26. Tammelin T, Nayh¨ aS,HillsAP,J¨ arvelin¨ MR. body shape trajectories over the life course with 15. Sebert S, Lowry E, Aumuller N, et al. Cohort Adolescent participation in sports and adult type 2 diabetes risk in adulthood: a group-based profile: the DynaHEALTH consortium - a European physical activity. Am J Prev Med 2003;24:22–28 modeling approach. Ann Epidemiol 2015;25: consortium for a life-course bio-psychosocial 27. Yliharsil¨ a¨ H, Kajantie E, Osmond C, Forsen´ T, 785–787 model of healthy ageing of glucose homeostasis. Barker DJ, Eriksson JG. Birth size, adult body 39. Svendsen OL. Should measurement of Int J Epidemiol 2019;48:1051–1051k compositionandmusclestrengthin laterlife.IntJ body composition influence therapy for obe- 16. Schack-Nielsen L, Sørensen TIa, Mortensen Obes 2007;31:1392–1399 sity? Acta Diabetol 2003;40(Suppl. 1):S250– EL, Michaelsen KF. Late introduction of comple- 28. Baker JL, Olsen LW, Sørensen TIA. Childhood S253 mentary feeding, rather than duration of breast- body-mass index and the risk of coronary heart 40. Wang Y, Rimm EB, Stampfer MJ, Willett WC, feeding, may protect against adult overweight. disease in adulthood. N Engl J Med 2007;357: Hu FB. Comparison of abdominal adiposity and Am J Clin Nutr 2010;91:619–627 2329–2337 overall obesity in predicting risk of type 2 di- 17. Osler M, Godtfredsen NS, Prescott E. Child- 29. Eriksson JG, Forsen´ T, Tuomilehto J, Winter abetes among men. Am J Clin Nutr 2005;81:555– hood social circumstances and health behaviour PD, Osmond C, Barker DJ. Catch-up growth in 563