Past, present, and future trends of overweight and obesity in Belgium using Bayesian age-period-cohort models

Background Overweight and obesity are one of the most significant risk factors of the twenty-first century related to an increased risk in the occurrence of non-communicable diseases and associated increased healthcare costs. To estimate the future impact of overweight, the current study aimed to project the prevalence of overweight and obesity to the year 2030 in Belgium using a Bayesian age-period-cohort (APC) model, supporting policy planning. Methods Height and weight of 58,369 adults aged 18+ years, collected in six consecutive cross-sectional health interview surveys between 1997 and 2018, were evaluated. Criteria used for overweight and obesity were defined as body mass index (BMI) ≥ 25, and BMI ≥ 30. Past trends and projections were estimated with a Bayesian hierarchical APC model. Results The prevalence of overweight and obesity has increased between 1997 and 2018 in both men and women, whereby the highest prevalence was observed in the middle-aged group. It is likely that a further increase in the prevalence of obesity will be seen by 2030 with a probability of 84.1% for an increase in cases among men and 56.0% for an increase in cases among women. For overweight, it is likely to see an increase in cases in women (57.4%), while a steady state in cases among men is likely. A prevalence of 52.3% [21.2%; 83.2%] for overweight, and 27.6% [9.9%; 57.4%] for obesity will likely be achieved in 2030 among men. Among women, a prevalence of 49,1% [7,3%; 90,9%] for overweight, and 17,2% [2,5%; 61,8%] for obesity is most likely. Conclusions Our projections show that the WHO target to halt obesity by 2025 will most likely not be achieved. There is an urgent necessity for policy makers to implement effective prevent policies and other strategies in people who are at risk for developing overweight and/or obesity. Supplementary Information The online version contains supplementary material available at 10.1186/s12889-022-13685-w.

Because of the large increase in cases in the past decade, overweight and obesity are now both considered as a global epidemic [10]. Belgium is no exception to this situation: obesity prevalence rose from 9.0% in 1978 to 15.0% in 1993 in a subsample of working, middle-aged men [11], and kept rising since. Consequently, a global obesity target to halt obesity at 2010 levels by 2025 was adopted during the World Health Assembly in 2013 [12], whereby consensus was reached to prevent a further growth in the number of men and women with obesity above the observed prevalence in 2010.
As obesity has been recognized as a major public health problem in Europe, research on its likely evolution in European countries is necessary. To this end, a variety of statistical models has been applied to the European context, such as linear extrapolations and wave functions. The diversity in methodologies employed differed substantially, and not all were able to yield realistic obesity estimates for the near future [13][14][15][16][17][18][19][20][21][22]. To better comprehend the complexity underlying the overweight and obesity epidemic, and obtain reliable and valid projections, it is important to consider the contributing risk factors. These factors may include socio-demographic factors such as age, sex, and education, but also lifestyle habits such as sedentary lifestyles and the consumption of highcalorie foods [1,[23][24][25]. For example, there is a known sex gradient in the impact of overweight and obesity, whereby the risk for developing asthma [26] and diabetes [27] is different between men and women. In addition, it is important to consider the so-called generation effect, i.e. the degree to which individuals are receptive to societal and social changes. These effects specifically feature the common exposures of people from the same birth cohort, such as common nutritional habits, smoking habits, and others [28][29][30][31]. Hence, the birth of age-periodcohort (APC) analysis, whereby the age effect reflects the association between age and weight status, the period effects reflects the evolution of weight over time, and the birth cohort effect reflects the degree to which an individual's weight is receptive to societal and social changes.
An important issue in trend analysis is the limited availability of longitudinal data at regular time intervals [32]. Instead, epidemiologists often have to rely on repeated cross-sectional data to obtain long-term trends in overweight and obesity, whereby only basic information on age and sex is available for future populations based on population projections. Here is where the APC analysis excels as its estimated projections consider disentangled trends in age, period and cohort (i.e. generation) for overweight and obesity.
To date, only a limited number of studies have undertaken APC analyses to the prevalence rates of overweight and obesity, among which the number of European studies is scarce [10,24,[33][34][35]. In addition, many of these studies did not consider the complex sampling process of the population microdata at hand, or have built further on the recent critiques regarding APC analysis [36,37]. Lastly, the current pool of available studies mainly focused on estimating the APC effects, but the APC-analysis could be expanded by including other specific effects such as sex, and education. Therefore, the current study aims to (1) evaluate the age-period-cohort and other important effects based on the past and current data, and (2) estimate case projections using a APC modelling approach for a period of 10 years. To this end, a flexible Bayesian hierarchical APC model will be applied to population microdata available through the Belgian Health Interview Survey.

Data
Statistical analyses were performed using the datasets of the Belgian Health Interview Survey (BHIS), which was organized for the first time in 1997 by Sciensano, the Belgian Institute for Health, and contains a series of repeated cross-sectional sample surveys [38][39][40]. So far, six national health surveys have been carried out -in 1997, 2001, 2004, 2008, 2013 and 2018. The goal was to collect information on lifestyle and chronic diseases for approximately 10,000 individuals in each survey round. The survey applies a stratified multistage, clustered sampling method for each survey year. The combination of the large net sample size and the elaborate sampling methodology ensures that each sample is representative for the Belgian population, and that any trend found using the subsequent health surveys can be generalized to the larger Belgian population. More details on the sampling procedure have been published elsewhere (Demarest et al., 2013). The sole inclusion criteria to be eligible for the health interview study is a registration in the national register. For the current study, participants were only included if they had an age larger than or equal to 18. The survey was carried out in line with the Belgian privacy legislation and approved by the ethical committee of Ghent University.
In the current study, information on age, sex, education level, migration background, urbanization level, income level, and self-reported height and weight of participants was included from the different waves of BHIS [39].
The education level was used as a proxy indicator of the socio-economic status of the household and all its members. This indicator is based on the highest education level of the reference person or his/her partner and allocated to each member of the household. Possible values are "primary or no degree", "secondary inferior", "secondary superior", and "superior education" following the ISCED-11 classification, whereby superior education includes all obtained degrees higher than secondary superior [41].
The income level is calculated based on the 'total available income of the household' , for which an equivalent scale is applied [42]. This allows comparing incomes of different households taking their size and composition into account. The different members of the household receive a specific weight: 1.0 for the first adult member of the household, 0.5 for each additional adult (18+ years) and 0.3 for each child (< 18 years). The total available income of the household is divided by the sum of the weights of all the members of the household to calculate the equivalent income. The income levels (quintiles) include "< 750 euro", "750-1000 euro", "1000-1500 euro", "1500-2500 euro", and "> 2500 euro", which are hereafter referred to as quantiles 1 to 5.
The level of urbanization was determined based on morphological and functional characteristics of the municipalities. Two morphological criteria are used to classify the municipalities: the population density and the area of habitation. Three criteria are used to describe the functional characteristics of the municipalities: the commercial function, the educative function and the employment rate. Based on these attributes, municipalities are labelled as "Big cities and dense agglomerations", "Suburban", "Urbanized municipalities", or "Rural".

Data analysis
To model the number of cases with overweight and obesity using age, period, and cohort effects, an age-periodcohort (APC) analysis was performed. We have built two separate models, one with overweight and one with obesity as dependent variable.

Independent variable
As explanatory variables, the models considered sex, education level, migration background, urbanization level, and income level as fixed independent categorical effects, and the categorical age, period and cohort effects as random effects. In the forecasting model, fixed effects that were not available as strata in the demography projections from the Belgian Federal Planning Bureau were excluded.

Dependent variable
The dependent variable was obesity and overweight, which were dichotomized based on the BMI of each individual.

Statistical model
One of the major issues in classical APC models is the linear dependency between the age effect, the period effect, and cohort effect. To tackle this linear dependency, the hierarchical age-period-cohort (HAPC) model has been introduced, which can include a mix of fixed and random effects [45][46][47]. Applying these HAPC models in a Bayesian framework allows for a direct interpretation of future trends in terms of credibility (e.g., how likely will overweight and obesity increase by at least 10%), whereas in the frequentist setting the projected uncertainty intervals cannot be interpreted as credibility. Hence, Bayesian HAPC models have been applied more frequent to forecast future trends in prevalence and incidence of cancer [48,49], but to our knowledge, no study has applied the Bayesian APC model to forecast future trends in the prevalence of overweight and obesity.
Using the Bayesian framework, APC models [48] were fitted using the INLA package (Version 20.3.17) [50]. INLA stands for Integrated Nested Laplace Approximation, a novel approach that makes Bayesian inference faster compared to the computer-intensive Bayesian Markov chain Monte Carlo methods. More information on the INLA package can be found elsewhere (https:// www.r-inla. org/). The number of overweight or obese individuals, y ij , in age group i and period j was modelled as a Binomial process with the mean equal to the product of the population at risk, N ij , and the estimated prevalence. The logit of the prevalence, η ij , was estimated as a linear combination of the age, period and cohort effects, respectively α i , β j and γ k , where k = M(I − i) + j is the birth cohort, M is the number of periods per age group and I is the number of age groups.
Second-order random walk priors (RW2) were applied to the age, period and cohort effects. These RW2 effects are particularly well-suited to model unequal time intervals in the APC effects [51]. Log-gamma priors were applied to the precision parameters with scale and shape parameters of 1 and 0.00005 for each of the age, period and cohort effects. Modelling priors were based on the methodology from Cameron & Baade (2019) and Riebler & Held (2010). Model selections for factors and priors was based on information criteria: DIC, Deviance information criterion, and WAIC, Watanabe-Akaike information criterion. Projected temporal trends were described as the median and 95% credibility intervals (CrI) around the median were constructed based on the 2.5 and 97.5% quantile of the posterior distribution.
In addition, based on the marginal posterior distribution of the modelled count, we calculated the probability, P(y t + x > y t + m × y t ), that a projected rate in year t + x is greater than the modelled value in the final year of observed data (t = 2018), by some margin m (expressed as percentage increase, %). All statistical analyses were performed in R 4.1.0 [52].

Model validation
Model fit and predictive accuracy were assessed by fitting the model to the repeated cross-sectional sample surveys [38]. The explained variance (R 2 ) and root mean squared error (RMSE) between the observed and projected prevalence counts were calculated for different priors and models. More details on the model validation are provided in Supplementary material A.

Population
In total, the sample included 73,681 participants across all surveys. More information on the final sample that was selected for the analysis sample is depicted in the flowchart (Fig. 1). The raw socio-demographic, and health-related characteristics of each cohort are listed in Table 1. The average age across the different crosssectional cohorts ranged from 44 to 51 years. The majority of participants were females with proportions ranging from 52 to 54%. Over time, the proportion of participants with a higher education has increased from 29% in 1997 to 41% in 2018. The level of urbanization has remained stable over time with the majority of participants living in big cities (47%), and a minority living in a suburban (13%) or rural area (15%) in 2018. The income distribution shows an increase of incomes in the higher quintiles (Quintile 4 and 5) and a decrease in the lower quintiles (Quintile 1 and 2). The number of non-EU immigrants has doubled from 5.2% in 1997 to 11.0% in 2018.  Table 3.  N = 8471 a 2001, N = 9949 a 2004, N = 11297 a 2008, N = 9637 a 2013, N = 9051 a 2018, N = Figures 2 and 3 depicts the disentangled age, cohort, and period effects by sex for the occurrence of overweight and obesity in the Belgian population. The prevalence of overweight and obesity is the highest for the middle-aged groups in both men and women. In addition, the prevalence of overweight and obesity has increased over time in men and women. In men, the observed increase was stronger compared to the increase in women. Lastly, there was a clear increase in the prevalence of obesity and overweight among the oldest generations. However, since the generations born from 1915 and onwards the cohort effect has remained relatively stable.

Future trends in overweight and obesity
Modelled and projected rates with their 95% credibility interval (CrI) for overweight and obesity are given by sex in Fig. 4. The explained variance in prevalence expressed as the coefficient of determination (R 2 ) by the overweight model equalled 80.6 and 76.0%, and the explained variance in prevalence by the obesity model equalled 76.7 and 58.3% for men and women, respectively.
Temporal trends in rates for overweight in men showed a steady state to a slight decrease from 54 Probabilities for exceeding a certain threshold m (> 0%, > 5%, > 10%, > 25%, > 50%, and > 100%) based on the marginal posterior distribution of the modelled counts are listed in Table 4. In men, there is a 43.3% probability for an increase in the rate of overweight by the year 2025 and 2030. The probability for an increase in the number of men with obesity by the year 2025 and 2030 are much higher in comparison, with estimated probabilities of 86.7% and 84.1%, respectively. In women, there is a 56.0% and 57.4% probability for an increase in the rate of overweight by the year 2025 and 2030, respectively. The probability for an increase in the number of women with obesity by the year 2025 and 2030 are much higher in comparison, with estimated probabilities of 54.6% and 56.0%, respectively.
As depicted in Fig. 5, the age structure of men and women with overweight and obesity has changed over time. In 2018, 57.7% and 61.8% of men, respectively with overweight and obesity, had an age between 36 and 65 years. Similarly, 51.5% and 53.9% of men, respectively with overweight and obesity, had an age between 36 and 65 years. The current projections estimate that these proportions will drop in 2030 to 53.6% and 50.4% in men, and 43.8% and 47.8% in women, whereby the proportion of men and women with obesity will mainly increase in the youngest (< 36 years) age category.

Discussion
Our analyses based on nationally representative data collected over six large scale health interview surveys, covering a period spanning over 20 years from 1997 to 2018, showed a disturbing picture of the future overweight and obesity epidemic in Belgium, assuming similar future increase rates as observed in the past. In men, an increase in overweight is less likely, whereby the probability of an increase by 2030 equals 43.3%. In contrast, an increase in obesity by 2030 is very likely with a probability of 84.1%. In women, an increase in overweight and obesity will likely be reached by 2030 with a respective probability of 57.4% and 56.0%.

Past and current trends in overweight and obesity
The age, period and cohort effects are similar to those reported in previous reports [1,35], whereby the  number of men and women with overweight and obesity has increased over time among all generations. The current trends indicate that overweight and obesity do not occur among specific generations, but rather affect all generations with an increasing trend over time among all generations. In addition, higher prevalence rates of overweight and obesity were demonstrated among middle-aged women and men. More specifically, among Belgians, a middle-age man, born before 1970 with no higher education and a middle income has the highest risk for being overweight and obesity at present. Among these risk factors, the highest were sex and education. It is well-known that sex plays an important role in the metabolic and genetic predisposition of overweight and obesity, whereby overweight and obesity do not only occur more frequent among men, but also yield an increased risk for developing overweight-related disorders among men [53,54].
Socio-economic factors also play an important role in the occurrence of overweight, whereby healthier behaviour could be driven by a higher health-literacy, which in its turn relates to higher education levels [55].

Projections in overweight and obesity
Our projections indicate that Belgium is unlikely to meet the global obesity target to halt obesity "at 2010 levels" by 2025, which was adopted during the World Health Assembly in 2013 [12]. The current projected prevalence rates for overweight and obesity will likely be higher in both sexes compared to their respective 2010 prevalence estimates. In another study, it was estimated that the global rate of overweight in established economic markets would rise to 36.3% by 2030 following population projections and a steady prevalence rate of 2005. The same study estimated an increase to 30.0% when modelling the prevalence based on past data. Likewise, the obesity prevalence was estimated at 22.1% and 36.2% [21]. These estimates are lower compared to ours, however, the observed prevalence in 2018 already exceeded the projection for overweight by 10.6%, which indicates that these previous estimates were rather too conservative. Similar to our results, most study reports agree that the growth in cases with obesity is faster compared to the growth in cases with overweight [13][14][15][16][17][18][19][20][21][22]. The increase in overweight and obesity likely results from a complex interaction between changes in the food environment, physical activity, socioeconomic, environmental, and genetic factors [1]. For example, the number of low quality away from home food consumptions, a known environmental risk factor for overweight and obesity [56], has grown over the years. Assuming trends in behaviour are likely to evolve similarly to the past, the number of cases with overweight and/or obesity are very likely to increase further. Efforts have been made to halt the rise of overweight, but prevention is a complex issue and requires collective efforts from the governments, the scientific and the medical communities, the industry, and various social organizations towards the changing of dietary and lifestyle habits.

Policy implications
We argue, in accordance with the current evidence, that a multi-faceted approach will be necessary to halt a further increase in cases. This approach should include policy guidelines and legislation that focuses on prevention, but also include treatments that are considered effective for people with overweight and obesity. The main pillar to reach a halt in cases is the prevention of overweight and obesity. Prevention should already start in childhood and early adolescence by implementing health promotion in schools including offering healthy snacks and meals, and promoting sufficient levels of physical activity [57]. Further prevention can be achieved by implementing the prioritized food environment policies [58]. If prevention fails, healthcare workers can rely on effective treatment strategies, which mainly focus on lifestyle and behavioural changes in nutrition and physical activity, or -in the worst case -pharmacotherapy or bariatric Surgery [59][60][61]. Next, it is important to consider the socio-demographic gradient in overweight and obesity. It is known that some groups are more prone to develop overweight or obesity. Therefore, it is important implement policies that target this socio-demographic gradient in cases [62]. Without taking action, the rise in cases will likely follow the projected trajectory, whereby the rates of disease burden and associated healthcare cost of non-communicable diseases will also rise [63,64]. The fact that environmental and behavioural forces fuelling the obesity epidemic are unlikely to be modified overnight, and even effective prevention programs may take years to show a significant impact [13], further emphasized the importance of prioritizing overweight and obesity on the healthcare agenda.
Lastly, the current forecasted rate of increase might have been exacerbated by the COVID-19 pandemic, as government introduced a myriad of measures such as cancellation of small gatherings, quarantine, lockdowns, and individual movement restrictions to combat the spread of the coronavirus [65]. Consequently, decreases in physical activity and increases in sedentary behaviours during lockdowns have been reported by different studies [66,67], resulting in a potential increase of overweight and obesity in the current generations [68].

Strengths and limitations
APC models account for trends in risk factors, without requiring measurements of exposures. Projected counts are hence susceptible to unforeseeable changes, which might result in an over-or underestimate the actual future prevalence. The key assumption in the APC models is, as with any projection model, a continuation of the Despite the wide uncertainty in projections beyond 5 or 10 years, governments require longer term projections for effective planning and policy development. The increased uncertainty provides a more realistic picture of the accuracy of projections, particularly when unforeseen interventions and changing circumstances (e.g. COVID-19) that could take place are considered. Since the cohort is a linear combination of age and period, the effects are not identifiable and cannot provide statistical evidence of change in prevalence [69]. Nevertheless, the models provide information on the prevalence and distribution across the population. The current study includes a large population of individual data over a time-span of 10 years. These data were modelled using a welldescribed modelling strategy including the incorporation of weights according to the sampling procedure, which results in realistic projections of future scenarios.
BMI has been defined according to the WHO definition, and based on self-reported estimates for weight and height. Consequently, the actual number of participants with obesity and overweight might have been higher as BMI is on average underestimated by 0.97 kg/m 2 [70].
In a next step, future research could focus on integrating policy-decisions based on a scenario-based modelling approach. For example, including information on the level of physical activity or nutritional habits, and associated policy interventions could be expressed in terms of the changes in projected rate.