Exploring Disparities for Obesity in Korea Using Hierarchical Age-Period-Cohort Analysis With Cross-Classified Random Effect Models

Obesity represents a significant global public health issue, imposing substantial burdens on individuals and healthcare systems.1 In South Korea, a nation experiencing rapid socioeconomic transformation, there has been a notable increase in the prevalence of obesity.2 It is essential to understand the factors contributing to this rise to devise effective public health strategies. Age-period-cohort (APC) analysis is useful for elucidating the age, period, and birth cohort effects on the prevalence of obesity within the South Korean population.

APC analysis is a well-established, comprehensive epidemiological method that explores temporal trends and interplay of age, period, and birth cohort effects in shaping health outcomes. In the realm of public health research, the hierarchical APC cross-classified random effects model (HAPC-CCREM) enhances the traditional APC models by integrating a hierarchical structure that assesses interactions among APC effects and variations associated with lifestyles and socioeconomic factors.3 HAPC-CCREM acknowledges that age, period, and birth cohort are distinct dimensions that differentially impact health outcomes. Age effects may stem from the biological aging process or cumulative exposures, period effects from specific yearly factors related to measurement or exposures affecting the entire population at a particular time, and birth cohort effects from commonalities among those born in the same era or uneven distribution of environmental influences.4

Despite increasing acknowledgement of APC analysis as a valuable approach, its application in studying obesity in South Korea has been limited. A previous study focusing on health examination of male civil servants did not account for lifestyle or socioeconomic influences obesity.5 Moreover, unique features of obesity in Asian populations,6 which might differ from those in Western populations, and lack of consideration for lifestyle or socioeconomic changes within cohorts in other East Asian studies underscore existing research gaps.7, 8

Therefore, in this study, we employed the Korea National Health and Nutrition Examination Survey (KNHANES) to perform an HAPC-CCREM on obesity prevalence in South Korea. This approach also investigates the potential interactions among these factors, shedding light on the mechanisms underlying the obesity epidemic in South Korea. By identifying age-specific patterns, and temporal trends, and by recognizing disparities in obesity prevalence related to lifestyle and socioeconomic factors, we aim to identifying critical periods and vulnerable birth cohorts that might benefit from targeted public health interventions.

All estimated prevalences of obesity in the figures are derived from Model 4 (Supplementary Table 3). Fig. 1 illustrates the combined age and birth cohort effects on obesity in the final model. The prevalence of obesity, as influenced by life course, varied by sex and birth cohort. Among men, more recent birth cohorts exhibited a higher peak prevalence of obesity, appearing at younger. In contrast, the lowest peak prevalence of obesity for women occurred in the 1980 birth cohort, with gradual increases in subsequent birth cohorts. The crude prevalence of obesity is shown in Supplementary Fig. 1.

Fig. 1
Adjusted age and birth cohort effects according to sex.

• Click for larger image
• Download as PowerPoint slide

Fig. 2 displays the adjusted effects of age (Fig. 2A), period (Fig. 2B), and birth cohort (Fig. 2C and D). The prevalence of obesity peaked during middle age for both sexes (Fig. 2A), and the disparity in obesity prevalence between sexes narrowed after middle age. The period effect was statistically significant (P for random terms of survey years < 0.001), showing a negative period effect in 2014, which turned positive in 2020 and 2021 (Fig. 2B). The birth cohort trajectory for obesity prevalence showed a crossover around the 1935–1940 birth cohort, after which prevalence was higher in men than in women (Fig. 2C). For both sexes, adjusted obesity prevalence declined after the 1980 birth cohort. The negative random effects on obesity were significantly lower in the 1955–1965 birth cohort (Fig. 2D), though these effects are minor relative to the temporal trends in birth cohorts (Supplementary Fig. 2).

Fig. 2
Adjusted age (A), period (B), and fixed birth cohort (C) effects by sex, and random birth cohort effect (D). (B) The horizontal solid line indicates the adjusted average prevalence of obesity.

• Click for larger image
• Download as PowerPoint slide

Fig. 3 illustrates the significant interaction between the age effect and various covariates. Among lifestyle and socioeconomic factors, smoking history (Fig. 3A,P for interaction = 0.020) and physical activity (Fig. 3B,P for interaction < 0.001) showed statistically significant interactions with the age effect. The interaction between the age effect and other covariates is detailed in Supplementary Fig. 3. In both sexes, the highest obesity prevalence was observed in current smokers, followed by never-smokers and ex-smokers, before reaching 40 years of age. Conversely, after this age, ex-smokers exhibited the highest obesity prevalence, followed by never-smokers and current smokers. The variation among smoking histories became more pronounced with age (Fig. 3A). Moreover, in both sexes, individuals who engaged in physical activity showed higher obesity prevalence until middle age, but exhibited lower prevalence thereafter compared to those who were inactive (Fig. 3B).

Fig. 3
Interaction between age and covariates. The interaction effects of smoking (A) and physical activity (B).

• Click for larger image
• Download as PowerPoint slide

Fig. 4 reveals the significant interaction between the birth cohort effect and covariates. Among lifestyle and socioeconomic factors, smoking history (Fig. 4A,P for interaction = 0.020), physical activity (Fig. 4B,P for interaction < 0.001), and place of residence (Fig. 4C,P for interaction = 0.005) demonstrated statistically significant interactions with the birth cohort effect. For both sexes, the adjusted obesity prevalence according to the covariates crossed over during 1955–1965 birth cohort, with the gap widening in subsequent years. Post-1960 birth cohort of 1960, the adjusted prevalence of obesity was higher in ex-smokers than in never or current smokers (Fig. 4A). Additionally, after the 1960 birth cohort, physically active individuals had a lower adjusted obesity prevalence compared to inactive individuals (Fig. 4B). The adjusted obesity prevalence was also higher in individuals residing in urban areas as opposed to those in rural areas from the 1960 birth cohort onward, with the disparity increasing over time (Fig. 4C). Supplementary Fig. 4 displays the interaction between the cohort effect and other covariates.

Fig. 4
Interaction between birth cohort and covariates. The interaction effects of smoking (A), physical activity (B), and residence (C).

• Click for larger image
• Download as PowerPoint slide

We conducted an investigation on obesity disparities using HPAC-CCREM using a nationwide survey in South Korea. Our results indicate that disparities in obesity, particularly with respect to birth cohort effects, have exhibited a significant widening trend between urban and rural areas beginning with the 1960 birth cohort. Additionally, we found that certain lifestyle factors, such as smoking history and physical activity, modify the influence of both age and cohort effects on obesity. Notably, the period effect on obesity showed a significantly negative association in 2014, which shifted to a positive association in 2021 and 2022.

In our epidemiological study examining the APC effect on obesity in South Korea, we noted a higher prevalence of obesity among individuals in rural areas compared to urban areas post-1960 birth cohort. The disparity in obesity prevalence between urban and rural regions has grown over time. Previous cross-sectional studies have shown inconsistent urban-rural differences in obesity. In the United States, obesity prevalence was higher in rural than urban areas.12 However, in Europe, differences in obesity between rural and urban areas were not statistically significant.13 In contrast, urbanicity was positively correlated with obesity in Southeast Asia,14 and this correlation was stronger in a lower income countries.

Although no study has evaluated the interaction between birth cohort and residence, our finding align partially with the global ecological study results.15 Globally, the increase in BMI was more pronounced in rural than in urban areas. Moreover, in the representative industrialized region and in high-income Western and Asia Pacific regions, urban women had a lower mean BMI compared to rural women, with this disparity growing over time. This trend may be linked to the urbanization of rural setting, marked by agricultural mechanization and increased availability of processed foods. Furthermore, economic and social disadvantages in rural area might also play a role in this discrepancy. In South Korea, urban-rural differences in obesity trends reverse following industrialization and are exacerbated by rural challenges. Despite the implementation of several nationwide health initiatives targeting chronic disease prevention and management, such as the Korean community-based Registration and Management Program for Hypertension and Diabetes,16 the Non-communicable Diseases Prevention and Control Program,17 and the Primary Care Project,18 healthcare service utilization still shows significant variation with urbanicity, indicated by the National Health Insurance Service.16 Therefore, to effectively address this urban-rural gap in obesity rates, tailored interventions for rural areas, including telehealth and web-based programs, are essential.19

Several previous studies have evaluated the APC effect on obesity or BMI in East Asia.5, 7, 8 Only one previous APC study on obesity has targeted in South Korea.5 This research focused on male civil servants who underwent health examinations in 1992, 1996, and 2000, showing that obesity prevalence increased over time and with successive birth cohorts, with a stronger birth cohort effect than period effect. However, unlike our study, which uses nationwide survey data, the findings from this specific group of male civil servants may not be generalizable. In contrast, studies in China and Japan show different trends. In the Chinese General Society Survey (CGSS),7 there was an inverted U-shaped association between age and obesity prevalence. However, unlike our findings that obesity was consistently more prevalent in men, CGSS found that obesity was more prevalent in women than in men after the age of 60. Furthermore, the National Health and Nutrition Survey (NHNS) in Japan8 demonstrated a positive association between age and BMI until the age of 40 for both sexes. Beyond that age, there were no significant changes in men, while BMI gain decreased in women.

We observed a negative birth cohort effect on obesity in the post-war era (1955–1965). Both our study and other studies included cohorts born at post-war period, yet the cohort effects differed. Similar to our findings, in France, the cohort born during the Second World War and the subsequent two decades showed a decelerating trend in obesity prevalence.20 In contrast, the CGSS identified a positive birth cohort effect on obesity among Chinese men during the Chinese Civil War (1920s and early 1930s) and the reform era (late 1970s).7 Additionally, the NHNS reported a consistent increase in mean BMI by birth cohort among men but a decrease among women from the 1930s to the 1970s.8 These discrepancies may reflect social changes or variations in individual composition within each country, potentially influencing cohort or period effects on obesity. Differences in statistical models or covariates used in these studies necessitate further evaluation to clarify the APC effect on obesity across countries.

Our findings revealed a significant period effect on obesity in 2014 and 2020–2021. The increase in BMI during this latter period is likely influenced by the coronavirus disease 2019 (COVID-19) pandemic. This is consistent with previous longitudinal studies21, 22, 23 that suggested that disaster situations can lead to increased BMI. Moreover, the rise in BMI during the COVID-19 pandemic is also linked to lifestyle changes, including reduced physical activity and altered dietary habits.24, 25, 26

The period effect remained statistically significant even after adjusting for physical activity in our study, suggesting residual effects. This could be because physical activity was modeled as a dichotomous variable. Furthermore, our study did not examine dietary habits owing to limitations in statistical power. Including total energy intake, as estimated through food frequency questionnaires or 24-hour recalls, would have led to the exclusion of at least 10,000 participants due to missing total energy intake values. Nonetheless, we excluded total energy intake to maintain sufficient statistical power, given that our primary objective was to construct an APC model. We did not identify any public health issues that could influence the prevalence of obesity in South Korea in 2014. Interest in obesity interventions might have risen following the 2013 classification of obesity as a disease by the American Medical Association.27 However, additional evaluation is necessary to assess the negative period effect on obesity in South Korea in 2014.

We observed that ex-smokers had a higher obesity prevalence than others in the newer birth cohorts, which may be due to weight gain after quitting smoking. Smoking cessation groups have shown a greater increase in obesity prevalence compared to groups that continued smoking in randomized control trials and observational studies.28, 29, 30 The larger disparity in obesity prevalence among smoking groups in newer birth cohorts reflect variations in smoker characteristics by birth cohort, such as the degree of nicotine dependence31 and comorbidities at the time of smoking cessation.32, 33 Further evaluation is required as KNHANES did not include data on these variables.

Our research identified a complex interplay among age, smoking history, and physical activity in shaping obesity prevalence. It is critical, however, to recognize the inherent limitations of cross-sectional data, especially in assessing longitudinal age effects. A significant difference in obesity prevalence was apparent among individuals with diverse smoking histories, particularly after the age of 40. After this age, there was a noticeable increase in obesity prevalence among former smokers, which may suggest weight gain following smoking cessation.28, 29, 30 Conversely, in individuals younger than 40, obesity was more prevalent among current smokers. This discrepancy may be attributed in part to age-dependent variations in the effects of nicotine on body weight. According to the Health Outcomes and Measures of the Environment Study,34 postnatal cotinine levels were associated with higher BMI. In contrast, findings from the National Health and NHANES showed the relationship between BMI at the time of smoking and daily cigarette use with weight changes post-cessation.31 Additionally, the pattern of high prevalence of obesity associated with physical activity before the age of 40, shifting to a lower prevalence afterwards, supports the idea that aging promotes weight gain, while physical activity can mitigate this effect.35, 36 Age-related increases in body fat percentage also need to be considered when examining BMI trends across different age groups.37 Therefore, future longitudinal studies incorporating body composition analyses are warranted.

This study substantially enhances our understanding of the APC effect on obesity in South Korea using data from a nationwide survey. We found that differences in obesity prevalence related to residential and lifestyle factors were particularly marked in more recent birth cohorts. Additionally, the APC effects on obesity documented here provide a robust foundation for subsequent APC studies assessing the contributions of the age, period, and cohort elements to other health outcomes, such as cancer incidence and mortality.38

Despite its strengths, our study has several limitations. First, the KNHANES is not a panel data, potentially resulting in inadequately control over individual variability. Specifically, panel data would allow for a more nuanced analysis of age trajectories in relation to factors like smoking history and physical activity. Second, our statistical approach depended heavily on the statistical significance and the goodness-of-fit of the model used, which might limit their biological plausibility. Future studies should therefore consider alternative modeling approaches that better integrate biological concepts when examining period or cohort effects on obesity in Korea.

These findings have significant implications for public health strategies focused on the prevention and management of obesity. They will inform the development of evidence-based policies, programs, and initiatives designed to lessen the health impact of obesity on individuals and alleviate its social and economic burden, thereby addressing a major public health issue.

Supplementary Table 1

Weighted baseline characteristics of study population according to survey years in men

Supplementary Table 2

Weighted baseline characteristics of study population according to survey years in women

Supplementary Table 3

Model parameter estimates

Supplementary Fig. 1

Age- (A) and birth cohort- (B) specific crude prevalence of obesity.

Supplementary Fig. 2

Combined fixed and random effects of birth cohort; solid and dashed lines represent the fixed and random effects of birth cohort, respectively.

Supplementary Fig. 3

Interaction between age and covariates. Interaction effect of alcohol consumption (A), household income (B), educational attainment (C), marital status (D), and residence (E).

Supplementary Fig. 4

Interaction between birth cohort and covariates. Interaction effect of alcohol consumption (A), household income (B), educational attainment (C), and marital status (D).