Time trends in body mass index distribution in the general population in Denmark from 1987 to 2021

The rise in the prevalence of obesity in past decades, is a growing public health concern [1]. At the population level, monitoring of weight status over time is mostly operationalised by grouping the BMI into categories of overweight and obesity as defined by the World Health Orgainsation’s cut-offs. This approach is intuitive, simple and facilitates comparison of developments within and across populations. However, important details of how the entire BMI spectrum develops is not illustrated by this approach [2].

In principle, when the prevalence of obesity increases (i.e. the fraction of individuals with a BMI above a certain cutpoint of the right tail of the distribution), two underlying mechanisms may be occuring. In one scenario, the whole distribution shifts towards higher values, increasing both the median and all other percentiles [2-4]. In this scenario, everybody in the population is experiencing a higher BMI – everybody is simply gaining weight. In the second scenario, the shape of the BMI distribution changes, by widening, having a longer tail or getting more skewed. This affects the prevalence of those in the high end of the distribution but not so much the median. Such movements actualise a situation in which mostly a subset of people in the population is becoming heavier. Last and naturally, any combination of the two scenarios is possible and likely, and the rate and nature of change of the distribution may vary among sub-groups, including socioeconomic groups [2, 5].

Understanding these mechanisms is important as they can help identify whether specific population groups are at a higher risk of an increasing BMI, or whether the population as a whole is experiencing weight gain. Different public health and clinical responses are required depending on the scenario experienced [2, 6], e.g., population-based strategies or high-risk targeted subgroup strategies.

Here, we present a detailed description of how the BMI distribution changes in Danish adults overall and by educational level and local area wealth. For this purpose we use data from nationally representative surveys conducted repeatedly and similarly from 1987 to 2021.

METHODS

We used data from the Danish Health and Morbidity Surveys, which are representative of the adult Danish population. These surveys were conducted in 1987, 1994, 2000, 2005, 2010, 2013, 2017 and 2021. A total of 95,471 participated among whom 91,684 had full information on height and weight [7, 8]. General characteristics of the study populations for each survey year are presented in Supplementary Table A.

BMI was calculated as weight (kg) divided by squared height (m). Obesity was defined as BMI ≥ 30 kg/m². We used data from the Danish Health Examination Survey to adjust for errors in self-reported height and weight [9]. In short, calibration equations were derived from information of self-reported and objectively measured height and weight in 15,692 participants. Weight was generally reported with a high accuracy in men and women in all age groups, whereas height was over-reported in both men and women and increasingly so with advancing age.

We calculated the prevalence of obesity overall and by sex and per age group, separately for each examination. As no significant difference was observed in the prevalence by sex, the overall prevalence of obesity was presented. BMI distributions for five selected survey years were presented in a density plot and with a cumulative distribution plot. We used quantile regression to estimate selected BMI percentiles (10th, 25th, 50th = median, 75th and 90th) by survey year, adjusted for age and sex. Analyses also included an analytical weight indicating the response probability for each individual with the purpose of increasing the degree of representativity of each survey. The Danish Health and Morbidity Surveys are nationally representative and independent cross-sectional surveys. Thus, any overlap is not accounted for and treated as random.

The quantile regression was presented for the study population overall, by educational level and by municipality wealth. Lastly, we repeated the quantile regression using weight measurements instead of BMI values, stratified by sex and adjusted for height. STATA version 17 was used for all analyses.

Trial registration: not relevant.

RESULTS

The prevalence of obesity as defined by a BMI ≥ 30 kg/m² increased from 6.1% in 1987 to 18.4% in 2021 (Figure 1). The distribution of BMI in selected years (1987, 1994, 2000, 2010, and 2021) is shown in Figure 2A. Overall, a shift to the right from one survey wave to the next was observed, indicating a general, higher BMI for the whole population. This was confirmed by the cumulative quantile regression (Figure 2B).

This visual impression was confirmed by quantile regression of the 10th, 25th, 50th, 75th and 90th percentiles of the BMI distribution (Figure 3). All the percentile lines exhibited increasing trends from 1987 to 2021. However, the increase was steeper in the higher percentile levels, demonstrated by a gradually larger linear slope from the lowest percentile to the highest. Hence, the lower percentiles had a small increase in BMI (0.8 BMI points in the 10th percentile), whereas the higher percentiles had a larger increase (3.4 BMI points in the 90th percentile). An increased dispersion of the BMI distribution over time was also demonstrated by a larger range between the 10th and the 90th percentile in 2021 (10.5 BMI points) compared with 1987 (7.9 BMI points). When modelling weight measurements adjusted for height, the same pattern as for BMI values was seen for the percentiles (Supplementary Figure A).

Analyses of BMI were repeated, stratified by educational level (short, medium and long) (Figure 4). For all educational levels and for all percentiles (10th, 50th and 90th), a steady increase in BMI was observed from 1987 to 2021. For the 50th and 90th percentiles, a clear separation was observed of BMI values by educational level, whereas for the 10th percentiles, the three lines were entangled. Additionally, the linear slope was less steep for long educational level than for short or medium educational level for both the 50th and 90th percentile. For example, BMI in the long educational level increased by 1.05 (0.93-1.16) kg/m² per ten years, whereas the increase was 1.24 (1.14-1.34) and 1.23 (1.07-1.40) for medium and short educational level. Lastly, the increase in the dispersion of the BMI distribution over time was larger for shorter educational level, illustrated by greater changes in 10th-90th percentile range from 1987 to 2021 with shorter educational level (3.4, 3.1, and 2.9 BMI points for short, medium and long educational level, respectively).

In strata of municipality wealth as defined by the median municipal income, similar tendencies were observed. However, these were less pronounced (Supplementary Figure B). A tendency was seen towards a steeper increase in BMI in the lowest tertile compared with the middle and highest tertiles in all BMI percentiles (10th, 50th, and 90th).

DISCUSSION

This study presents trends in BMI variation in Denmark from 1987 to 2021, overall and by socioeconomic groups. We observed a shift in the distribution curve of BMI to the right illustrated by positive linear slopes for high and low BMI percentiles alike. Additionally, a change in the shape of the BMI distribution was observed with a broader range and lower peak in the most recent survey (also illustrated by steeper slopes for the 90th percentile). The prevalence of obesity in Denmark increased from 6.1% in 1987 to 18.4% in 2021. Thus, the above-described trends indicate that the increased prevalence in obesity may be explained by an increase in BMI among all BMI categories with a somewhat higher increase in the higher BMI categories.

Our findings are in line with international studies [2]. Additionally, specific studies from Norway and the USA showed similar trends [10, 11]. In Norway, a general shift towards higher body weight in all weight categories with larger increases in obesity class I and II was reported [10]. In the USA, changes in median BMI contributed to 75% of the increase in obesity from 1980 to 2000 [11].

Several mechanisms have been proposed to explain the higher increase in BMI in the higher BMI categories. Interaction between the individuals’ genetic susceptibility and environmental factors has been suggested by many [12-15]. Additionally, weight bias and discrimination may lead to an increase in hormone levels, including cortisol, food intake and negative emotions, decreased physical activity and, consequently, weight gain [16]. This may potentially be a contributing factor to the increased dispersion in BMI. Thus, in an obesogenic environment, the positive skewing of the distribution curve of BMI (meaning that the tail of a distribution curve is proportionally longer on the right side or the upper end than at the lower end) may increase over time as heavier individuals gain more weight than lighter individuals. However, the increasing BMI dispersion is likely to be multifactorial, and its nature is not yet fully understood.

An important finding was that increases in and dispersion of BMI occurred for all educational groups, although the dispersion over time was larger for individuals with short education, suggesting a trend towards increasing social inequality in BMI. This is consistent with well-known social inequality in the prevalence of obesity in Denmark and comparable populations [17]. The larger increase in BMI among individuals with short education suggests that ongoing inequalities in weight gain will further put lower socioeconomic groups at a higher risk of obesity and its comorbidities. It has been proposed that socioeconomic inequalities in obesity may, to some extent, be explained by different exposure to obesogenic environments, suggesting that low-socioeconomic areas have fewer healthy choices, either in relation to food or physical activity [18, 19]. However, we did not observe large differences in the 10th, 50th or 90th percentile BMI between tertiles of municipality wealth, and no differences were recorded in the dispersion of BMI between tertiles. These findings suggest that social factors, such as educational level, rather than the physical environment, drive the inequality in dispersion of BMI – at least in Denmark. However, it is plausible that inequalities in the environment within municipalities are not captured as they are equalised when measuring municipality wealth on average.

Increasing population BMI with a right shift of the whole distribution has implications for public health and suggests that public health measures to prevent obesity should be targeted at the general population, e.g., by introducing structural public health programmes.

A major strength of this study is that the the Danish National Health and Morbidity Surveys are nationally representative and cover a period of 34 years. However, the response rate has varied across the survey years and overall decreased from 79.9% in 1987 to 45.4% in 2021 [7, 8]. Although the response rates were similar in other studies, they may be influenced by degree of health, including BMI and sociodemographic factors. In the Danish Health and Morbidity Surveys, non-respondents were generally more likely to be young men, have another ethnic background than Danish, be unmarried and have lower socioeconomic status (e.g., educational level and income). Further details have been described elsewhere [7, 8, 20]. To account for this, sampling weights based on sociodemographic information of non-respondents were applied. Previous studies have shown that self-reported information on weight and height have been biased in such way that individuals are likely to overestimate their height and underestimate their weight, resulting in an underestimation of BMI. However, in a Danish validation study with a similar study population and setting as used in this study, the degree of BMI underreporting was found to be low [9]. Also, the applied calibrations have been validated in a sub-sample of 15,692 participants who both filled out questionnaires and participated in a health examination with weight and height measures [9].

The difference in the distribution indicators were not tested for statistical significance as the purpose of this study was to describe distribution trends. How much of the increasing obesity prevalence and mean BMI may be attributed to a right shift and changes in the distribution cannot be concluded. However, we believe quantifying this is less important and that the important finding is that both mechanisms are in play and that the population as a whole is affected by increasing BMI. This is an important consideration when planning public health measures to prevent obesity as such initiatives may include public health measures to the broader public as well measures targeted at individuals in high BMI categories.

CONCLUSIONS

We observed a shift to the right in the BMI distribution curve from 1987 to 2021 with gradually higher values in all BMI percentiles and in all socioeconomic strata. Thus, our results show that, generally, everyone in the Danish population is gaining weight. Continued monitoring of population BMI trends is needed for identification, implementation and evaluation of public health policies and evidence-based interventions.