Possible Modifiers of the Association Between Change in Weight Status From Child Through Adult Ages and Later Risk of Type 2 Diabetes

Even though child and adult BMI (in kilograms per meters squared) are positively associated with the risk of developing type 2 diabetes, the effects of changes in weight status between childhood and adulthood are not well understood. Previous studies indicate that children with overweight who remit from it before adulthood may reduce their risk of type 2 diabetes (1–6). Further, some studies find being persistently overweight from childhood to adulthood carries higher risk of type 2 diabetes than being overweight only in adulthood (1,2,4).

The development of childhood obesity has a strong social gradient (7), and in adult life, inverse associations between socioeconomic status and educational attainment and risks of obesity and type 2 diabetes are well established (8,9). It may be hypothesized that individuals of low socioeconomic status are more vulnerable to health risks associated with overweight. Yet, there is sparse evidence in this research area (10–12). Only one of the previous studies focusing on changes in weight status and type 2 diabetes evaluated a potential multiplicative interaction with educational attainment on these associations (4). Moreover, previous studies are limited by including only one sex (1,4,5), having a low number of diabetes case subjects (n < 237) (2,3,6), or ending follow-up in middle adulthood (3,6). Further, smoking and lower levels of leisure time physical activity (LTPA) are associated with higher risks of developing type 2 diabetes (13,14), but these were not investigated as modifying factors in the studies on changes in weight status and type 2 diabetes.

The aim of this study was to investigate the association between changes in weight status from childhood to adulthood and the risk of developing type 2 diabetes in Danish and Finnish cohorts and whether this association is modified by socioeconomic and lifestyle factors. It was hypothesized that high BMI at child and adult ages has stronger associations with type 2 diabetes among individuals with short educational attainment, those who smoked, and individuals with low levels of LTPA.

The study population was drawn from cohorts participating in the DynaHEALTH consortium that aims to build an empirical model of healthy aging (15). To be eligible for this study, we required information on weight and height at 7 and 12 years of age and in adulthood (>18 years), as well as information on educational attainment, smoking, and LTPA (Supplementary Table 1). Among the cohorts included in the DynaHEALTH consortium, 10 cohorts were eligible. Eight were subsamples of adult research studies (16–22), among whom the participants, who were born from 1930 to 1981, were included in a Danish school cohort from Copenhagen (23) (N = 19,717) (Supplementary Table 1). Additionally, the Northern Finland Birth Cohort 1966 (NFBC1966) was included. It includes individuals born in the two northernmost provinces of Finland in 1966 who participated in a clinical examination in adulthood (N = 3,985 included) (24–26). Finally, the Helsinki Birth Cohort Study (HBCS) contributed with individuals who were born between 1934 and 1944 at the Helsinki University Central Hospital, attended child welfare clinics in the city, went to school in the city of Helsinki, and attended a clinical examination in adulthood (N = 1,581 included) (27). The cohorts are described in detail elsewhere (16–24,27).

All studies were performed in accordance with the Helsinki 2 Declaration. The project was conducted on anonymous data, and it was approved by the Danish Data Protection Agency (Datatilsynet).

Body height and weight were used for calculation of BMI (weight in kilograms divided by height in meters squared) and prospectively measured at school health examinations and recorded in health records. In the Danish cohorts, BMI at the exact ages of 7 and 12 years were obtained by interpolation between successive measurements around that age or by extrapolation always within ±12 months (28). In the Finnish cohorts, BMIs from ≥6.0 to <8 years of age and from ≥11.0 to <13 years of age were included as 7-year or 12-year measurements, respectively (29). In adulthood, weight and height were measured at clinical examinations in seven cohorts (18–22,24) and self-reported in three cohorts (16,17,27). If individuals participated in more than one examination, we preferentially chose the one at the youngest adult age.

High BMI in childhood was defined as ≥85th BMI percentile at 7 or 12 years based on sex- and age-specific BMI percentiles for the Copenhagen cohorts, the NFBC1966 and HBCS, respectively (Supplementary Table 2). Obesity in adulthood was defined in accordance with the World Health Organization criteria (BMI ≥30 kg/m²) (30). Patterns of high BMI were defined as combinations of high BMI in childhood at 7 or 12 years of age and obesity in adulthood. Moreover, a BMI pattern with eight categories was defined according to combinations of high BMIs at 7 years (yes/no) and/or 12 years (yes/no) and/or adult obesity (yes/no).

At the time of the adult BMI assessment, information on educational attainment, current smoking (yes/no), and LTPA was obtained by questionnaire. Educational attainment was categorized into 0–7 (short), 8–10 (medium), or >10 (long) years of schooling. One cohort used categorizations of 7–9, 10, and >10 years due to the definition in the questionnaire (17). The available information in each cohort consisted of three to six levels of either LTPA or energy expenditure, and it was categorized into three groups corresponding to 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).

Information on type 2 diabetes was obtained by linking unique personal identification numbers (27,31) of the cohort participants to computerized and comprehensive health registers. During the follow-up period, <0.6% of the individuals emigrated, they were censored on this date, and no individuals were untraceable. Thus, loss to follow-up was minimal.

In the Danish cohorts, information on inpatient and outpatient diagnoses of type 2 diabetes (ICD-8 code 250 and ICD-10 codes E11, E13, and E14) was obtained by linkage to the National Patient Register (32) as previously described (4). In this register, although the completeness is moderate (sensitivity 64%), the positive predictive value of a diabetes diagnosis is very high (97%) when assessed against information from and verification of the diagnosis by general practitioners (33). In the NFBC1966 and the HBCS, in addition to hospital and prescription registers (34), clinical examination values (subsequent to obtaining the adult BMI) of plasma glucose ≥7 mmol/L and 2-h plasma glucose ≥11.1 mmol/L in a 2-h 75-g oral glucose tolerance test were used to identify type 2 diabetes using World Health Organization criteria (35). Furthermore, in the HBCS, HbA_1c ≥6.5% or 48 mmol/mol was used to identify type 2 diabetes. Through the combination of these methods, the sensitivity is likely very high, although a few individuals with type 2 diabetes may not be identified if they missed the clinical examination and were treated only with diet in general practice (36).

To achieve a baseline population free from diabetes, only individuals without type 2 diabetes at the adult BMI assessment were followed prospectively in registers from the age at the adult BMI measure or from 30 years of age for Danish individuals (4), whichever came last. As such, 266 individuals with type 2 diabetes were not included in this study from the Danish data, 172 from the HBCS, and 8 from the NFBC1966. The follow-up ended at the date of a type 2 diabetes diagnosis, death, emigration, loss to follow-up, or at the end of the follow-up period, which was 31 December 2013 for the HBCS and 31 December 2016 for the NFBC1966 and the Danish cohorts, whichever came first. Hence, the adult BMI measurement was collected before information on type 2 diabetes to avoid potential reverse causality from effects of type 2 diabetes on weight gain or loss.

Associations between high BMI at each age or patterns of high BMI and adult type 2 diabetes were estimated with hazard ratios (HRs) and 95% CIs using Cox proportional hazards regression. Age was used as the time scale, implying delayed entry at the time of BMI assessment, and stratified by year of birth. Specifically, researchers responsible for each cohort estimated the stratum-specific regression coefficients and the corresponding SEs by sex, and these were pooled using random-effects meta-analyses techniques. As major differences between women and men exist in the prevalence, pathophysiology, treatment response, and outcome of type 2 diabetes (37), women and men were analyzed separately.

The analyses were adjusted for age at adult BMI and additionally for educational attainment, smoking, and LTPA. Potential heterogeneity in the associations in subgroups of the following potential effect modifiers: sex, educational attainment, smoking, LTPA, birth cohort (1930–1939, 1940–1949, and 1950–1989) and study, was investigated using the I² statistic, the Cochran Q test, and meta-regression. To assess the proportional hazards assumption, we tested heterogeneity by categories of age at diagnosis divided into <70 and ≥70 years, as graphs of the cumulative hazard from one weight category versus another showed that the slope changed at around this age (data not shown). To assess the impact of self-reported BMI and reverse causality, we performed sensitivity analyses omitting the three cohorts in which BMI was self-reported and by omitting the first 3 years of follow-up after the adult BMI measure.

Additionally, we performed an analysis of adult BMI adjusted for child BMI and summarized the results in a meta-analysis. An investigation of nonlinearity by modeling the mutually adjusted associations using cubic splines with seven knot points showed changes in the slope of the associations at the ∼90th percentile for child BMIs and at 30 kg/m² for adult BMIs. Therefore, the associations were estimated using a linear spline regression with knot points inserted at the sex- and region-specific 90th percentile for child BMIs and at 30 kg/m² for adult BMIs. HRs for specific linear combinations of child and adult BMI are reported. All statistical analyses were performed using Stata (version 14.2; StataCorp, College Station, TX).

Among the 12,277 women and 13,006 men included in the study, 989 women (8.1%) and 1,370 men (10.5%), respectively, were diagnosed with type 2 diabetes from 1978 to 2016. The ages at diagnosis ranged from 30 to 85 years over an average follow-up time of 18.8 years (475,056 total person-years). We found similar associations between the BMI patterns and risk of type 2 diabetes in women and men, and these were not different in the meta-regression analyses (P > 0.33; data not shown).

Smokers, individuals with long education, or individuals with high levels of LTPA were more often in the groups who never had a high BMI or had a high BMI only as a child (Table 1). The number of individuals with type 2 diabetes by these factors is included in Supplementary Table 3. Among women and men, high child BMI and adult obesity were each positively associated with risks of type 2 diabetes (Table 2). As expected, men and smokers had higher risks of type 2 diabetes, and educational attainment and LTPA were inversely associated with risks of type 2 diabetes (Table 2).

Figure 1 shows the association between the BMI pattern and type 2 diabetes by levels of the potential effect modifiers. At higher educational levels in women and men, there was a tendency for the HRs for persistently high BMI from 7 years of age to adulthood or only high adult BMI and type 2 diabetes to be higher than those for women and men who never had a high BMI (Fig. 1A and B). These tendencies, however, were generally not supported by the subgroup meta-analysis and the meta-regression analyses (Supplementary Table 4). In women, smoking minimally influenced the associations between BMI patterns and type 2 diabetes (Fig. 1C). Nonsmoking men who only had a high BMI in childhood had a lower risk of type 2 diabetes than nonsmoking men who never had a high BMI, but associations in men with persistently high BMI or high adult BMI were positive in both smokers and nonsmokers (Fig. 1D and Supplementary Table 4). In women and men, the associations between BMI patterns and type 2 diabetes remained similar across levels of LTPA in the subgroup analyses and the meta-regression (Fig. 1E and F and Supplementary Table 4).

Women and men who had a high BMI only as a child had a similar risk of type 2 diabetes to that among women and men who never had a high BMI (women: HR 0.74 [95% CI 0.52–1.06]; men: HR 0.93 [95% CI 0.65–1.33]) (Fig. 2). Women and men with a persistently high BMI or a high BMI only as an adult had higher risks of type 2 diabetes compared with women and men who never had a high BMI (HR range: 3.78–5.27) (Fig. 2). Notably, adjustment for educational attainment, smoking, and LTPA minimally changed the estimates and the CIs. In women, the I² ranged from 0 to 41%, indicating low to moderate heterogeneity across cohorts. Among men, the I² ranged from 26.9 to 53.2% (Supplementary Appendix). Similar results were observed for weight status at 12 years of age combined with adulthood (Supplementary Fig. 1).

In subgroup analyses investigating the effects of diabetes diagnosed at <70 years, we observed stronger associations for persistently high BMI and high adult BMI with type 2 diabetes diagnosed at <70 years than after this age (Supplementary Fig. 2). However, the CIs for diagnoses after 70 years were wide, and the overall conclusions were the same. In analyses investigating potential birth cohort effects, associations between high BMIs and type 2 diabetes were stronger in later birth cohorts, but the patterns of associations were the same in all birth cohorts (Supplementary Fig. 3). Omission of the two cohorts with self-reported weight and height or restriction of follow-up time to from 3 years after the BMI assessment and onwards minimally changed the results (Supplementary Figs. 4 and 5).

Women who had a high BMI at 7 years and did not have obesity as an adult, irrespective of their childhood BMI status at 12 years, had a similar risk of type 2 diabetes as women who never had a high BMI at 7 and 12 years of age and in adulthood (Supplementary Fig. 6). In contrast, women who had a high BMI at 12 years of age but not at 7 years and who did not have obesity in adulthood had a higher risk of type 2 diabetes (HR 1.55 [95% CI 1.12–2.13]). Women who had obesity as adults had a higher risk of type 2 diabetes irrespective of their childhood BMI status. Although most results were similar for men, there was an exception. Men who had a high BMI at 12 years but not at 7 years and who did not have obesity in adulthood had a similar risk of type 2 diabetes as men who never had a high BMI. Adjustment for educational attainment, smoking, and LTPA minimally changed the results (Supplementary Fig. 6).

The regression coefficients for childhood and adult BMI from the linear spline model generally were similar across levels of education, smoking, and LTPA in the meta-analysis (Supplementary Table 5). Based upon these results, the point estimate for children who had a BMI at the 95th percentile at 7 years and an adult BMI of 30 kg/m² (corresponding to the ∼86th percentile) showed they had a fivefold risk of type 2 diabetes compared with children who had a stable BMI at the 50th percentile at 7 years and an adult BMI of 22 kg/m² (Supplementary Table 6). If a child with a BMI at the 95th percentile moved toward lower percentiles and had an adult BMI of 25 kg/m² (corresponding to the 50th percentile), the risk was much lower, although still above that of the reference group. The highest risk was observed among individuals who started at the 25th BMI percentile in childhood and ended at a BMI of 35 kg/m² (i.e., individuals who increased their degree of adiposity the most) (Supplementary Table 6). Adjustment for educational attainment, smoking, and LTPA minimally changed the results (Supplementary Table 6).

This study showed that in 10 Danish and Finnish cohorts, associations between BMI patterns and risk of type 2 diabetes were virtually not confounded or modified by educational attainment, smoking in women, and LTPA. Apart from smoking, which influenced one of the associations in men, the results were similar for men and women. This study confirmed that a high BMI in childhood combined with obesity in adulthood is associated with higher risks of type 2 diabetes, whereas a high childhood BMI combined with nonobesity in adulthood is not.

Individuals who developed obesity in adulthood had about the same risk of having type 2 diabetes as those who had a high BMI in childhood and obesity in adulthood. We also found that the BMI trajectory associated with the highest risk was the one that started at the 25th BMI percentile in childhood and ended at an adult BMI of 35 kg/m²; in other words, among individuals who increased the most in adiposity. These findings are supported by a large women-only study (38) that reported that those who were lean at 8 years and who had a sharp increase in self-reported body shape at puberty and thereafter had an almost threefold higher risk of developing type 2 diabetes compared with women whose body shape stayed in the midrange. This risk tended to be even higher compared with always having a large body shape (38).

Conversely, a decrease in BMI percentile from childhood to adulthood was associated with a lower, although still increased, risk of type 2 diabetes. This finding corresponds with those from a British study in which remission of obesity between childhood (7–16 years of age) and adulthood (23–45 years of age) was associated with a higher risk of type 2 diabetes compared with individuals who had never had obesity (6). Other studies show that remission of overweight at 8 years (5) and obesity between 4 and 19 years of age (3) and adulthood are not associated with a difference in the risk of type 2 diabetes. Together, these findings indicate that the adverse effect of a high child BMI is at least partly reversible by remission of high BMI in men and women.

Even when socioeconomic and lifestyle factors were accounted for, the associations between BMI patterns and type 2 diabetes changed little. Thus, the group with a high child BMI only did not have a higher risk at any level of education and LTPA, and those who had a persistently high BMI or developed obesity had consistently higher risk at all levels of education and LTPA. This suggests that BMI changes affect the risk of type 2 diabetes in the same way across levels of these factors. These results are in accord with our previous findings in men (4), and the current study extends these to women. We have not identified any studies reporting on potential interactions between LTPA and patterns of overweight from childhood to adulthood on risks of type 2 diabetes. In middle-aged individuals, a large case-cohort study in the European Prospective Investigation into Cancer and Nutrition cohort reported an interaction between physical activity and BMI measured at middle age only in women on the risk of type 2 diabetes (P = 0.008 in women). The higher risk of type 2 diabetes associated with lower levels of physical activity was evident in normal-weight and overweight women, but not in women with obesity (13).

Our analyses revealed a difference in the associations between the BMI pattern and type 2 diabetes in adulthood by current smoking status in men. Nonsmoking men with high child BMI only had a lower risk of type 2 diabetes than the nonsmoking men who never had a high BMI. We did not identify any other studies examining an interaction between smoking and the overweight pattern from childhood to adulthood. In adults, a large meta-analysis reported an interaction between smoking and overweight on the risk of type 2 diabetes such that the effects of smoking were stronger in overweight or obese individuals as compared with effects in normal-weight individuals (P < 0.001) (14).

When using a pattern of three BMI values, we found that men and women who had a high BMI at 7 and 12 years had a higher risk of developing type 2 diabetes only if obesity was present in adulthood. This finding is in accord with results from a study in three British birth cohorts followed into middle age (2). Our previous study yielded a slightly different result, as we found that men who had been overweight at 7 and 13 years of age but not during early adulthood had a higher risk of type 2 diabetes as compared with men who had never been overweight (4). Moreover, a large women-only study found that women with overweight (by somatotype) at 10 and 18 years of age but not at 34 years had a slightly higher risk of type 2 diabetes than women who had never been overweight (1). It is possible that these differences are due to the size of the studies as the two largest studies showed evidence of an increased risk among those with overweight in childhood and adolescence (1,4).

The strengths of this study are that we had measured weights and heights in 7 (70%) of the cohorts, limiting the potential for recall bias, we included men and women and information on lifestyle factors, and we were able to follow individuals to late adult ages. Rather than searching for published studies, we included eligible cohorts in the DynaHEALTH consortium (15), which circumvented the potential for publication bias usually encountered in meta-analysis on previously published results. The random-effect meta-analysis model allowed any potential heterogeneity in the associations across cohorts to occur around a normally distributed mean effect. Moreover, a wide range of birth cohorts was included. We found slightly stronger associations in postwar generations, but the overall patterns were the same. This suggests that the results are applicable to multiple generations, including contemporary populations, as some individuals in the study were born as recently as 1981. Further, sensitivity analyses examining the effects of self-reported weight and height and potential effects of reverse causality did not change these associations. Nevertheless, further studies are needed to investigate whether the findings are generalizable to other settings.

The study has some limitations. The sample sizes may, in some strata, have been too small to show effect modification. The estimates for some categories were imprecise, but this reflects that certain groups such as high child BMI only have a limited risk of developing type 2 diabetes. We did not have information on pubertal status, LTPA was self-reported, and we used BMI as an indicator of adiposity. BMI is a proxy for adiposity (39), so we do not know whether the changes in the risk of type 2 diabetes are due to changes in lean or fat mass or the location of the fat mass. This is important, as people without obesity also develop diabetes, dependent on their fat mass (40). Moreover, the study populations included only few children with severe obesity and, due to the racial composition of Denmark and Finland at these times, were predominantly of Caucasian descent. Whether the associations differ in groups with severe obesity, by race, or by ethnicity requires further evaluations. Lastly, similar to other studies, we did not have the age at onset of obesity.

In conclusion, in the 10 Danish and Finnish cohorts studied, a high BMI in childhood combined with obesity in adulthood was associated with higher risks of developing type 2 diabetes, whereas a high childhood BMI combined with nonobesity in adulthood was not. These associations were virtually similar across levels of educational and lifestyle factors, suggesting that BMI affects the risk of type 2 diabetes in the same way across levels of these other risk factors. Thus, public health initiatives should focus on preventing the continuation of adiposity from childhood into adulthood irrespective of educational level, and individuals with all levels of physical activity may benefit from health-promoting interventions.

Acknowledgments. The authors thank the late Drs. Aage Willumsen, Maternity Department A of the University Hospital of Copenhagen, and Bengt Zachau-Christiansen, the Pediatric Department of the University Hospital of Copenhagen, for the collection of the Copenhagen Perinatal Cohort data; all of those who initiated and/or continued the Metropolit study at the Institute of Sociology, 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 researchers who participated in the NFBC1966 and HBCS studies. The authors also acknowledge 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 Horizon 2020 research and innovation programme under grants 633595 DynaHEALTH, 733206 LifeCycle, and 824989 EUCANCONNECT; the Joint Programming Initiative: A Healthy Diet for a Healthy Life Netherlands grant agreement P75416 PREcisE; and Novo Nordisk Foundation grant NNF17OC0028338. NFBC1966 received financial support from University of Oulu grants 65354 and 24000692; Oulu University Hospital grants 2/97, 8/97, and 24301140; Ministry of Health and Social Affairs grants 23/251/97, 160/97, and 190/97; National Institute for Health and Welfare, Helsinki grant 54121; Regional Institute of Occupational Health, Oulu, Finland grants 50621 and 54231; and ERDF European Regional Development Fund grant 539/2010 A31592.

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

Author Contributions. L.G.B., T.I.A.S., M.-R.J., J.G.E., S.S., and J.L.B. conceived the study. All authors were involved in the design of the study. K.-H.H., G.B.J., E.L.M., M.O., K.O., T.S., A.T., T.I.A.S., M.-R.J., J.G.E., S.S., and J.L.B. provided data. L.G.B., N.W., R.N., and S.S. analyzed data. All authors were involved in the data interpretation. L.G.B. and J.L.B. drafted the manuscript. All authors contributed to revision and approval of the final manuscript. L.G.B. and J.L.B. are the guarantors of this work and, as such, had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. This study was presented at the 26th European Congress on Obesity, Glasgow, U.K., 28 April–1 May 2019.