Assessing longitudinal and cross-sectional effects of age on adult obesity in an Iranian population: results from a large population-based cohort study

Obesity is a significant risk factor for Noncommunicable diseases, and it is related to many adverse health consequences. The risk of obesity commonly changes with age, which is called a longitudinal or aging effect. Also, individuals born or enter to the study of the same age have similar living conditions that may influence their obesity risk in a particular way; this is a cross-sectional effect. In the current study, an advanced statistical model is used to distinguish between longitudinal and cross-sectional effects of age on the risk of obesity for men and women. Participants are a group of 6504 Iranian adults over 35 years of age in 2001, who live in the central region of Iran. They were followed up for 12 years in a large community-based study. Various medical indexes, including Body Mass Index, were collected in 2001, 2007, and 2013. The Marginal Logistic Regression model, which includes linear and quadratic effects of the Baseline Age and its difference with current age, is used.


Results
Between 2001 and 2013, the prevalence of obesity raised from 13-18% in men and from 31-44% in women. The odds of obesity for women was approximately three times the odds of obesity for men on average adjusting for the age effects. Both cross-sectional and longitudinal effects of age were significantly associated with the odds ratio of obesity. There was a rise in the prevalence of obesity for individuals with Baseline Age 35 to 55 and a decline thereafter. Also, the odds ratio of obesity across one's life course, had about 3% increase, on average, by each year aging, regardless of the age at baseline.

Conclusions
The high rate of obesity and its fast growth is a serious public health issue among Iranians, especially in adults age , and women. In the present study, Baseline Age was more strongly associated with the risk of obesity than aging. Considering both cross-sectional and longitudinal effects of age, helps us to understand the effect of age on obesity better and to identify the related factors.

Background
Noncommunicable Diseases (NCDs) account for more than 70% of early deaths all over the world(1, 3 2). Obesity, as a significant risk factor for NCDs, has an association with life expectancy reduction (3), and it raises the risk of metabolic diseases to a considerable extend (1). Also, obesity is associated with several adverse health consequences, and it is tied to an increased risk of conditions. Furthermore, obesity leads to a reduction in the quality of life, lack of employment, lower productivity, and inadequate social points. The World Obesity Federation and other organizations have defined obesity not only as a risk factor for several diseases but also as a chronic progressive disease (3).
Obesity is defined as excessive fat accumulation that might have a negative effect on health by the World Health Organization (WHO) and is identified as Body Mass Index (BMI) over or equal to 30 kg/m 2 (4, 5).
Obesity has an association with age, gender, household income, urbanization, and lifestyle (6). It is a chronic and complex disease determined by lifestyle behaviors associated with positive energy balance, such as an inappropriate diet and a sedentary lifestyle (7,8).
The prevalence of obesity has risen all over the world to pandemic proportions during the last 50 years (9,10). Obesity and BMI prevalence vary across different countries according to the steepness of increases, slowing-down, and acceleration periods (11). The global prevalence of obesity almost tripled since 1975 and continues to grow in low-and middle-income countries, although it can be considered as a problem for high-income countries (10,12,13). A systematic review by Rahmani et al. (2015), reported the prevalence of obesity in Iran has increased between 1995 and 2010 (14). Also, Azizi et al. (2005) indicated that the prevalence of obesity was most rapidly rising in the 30 to 40year-old group in both men and women in Tehran (15). Furthermore, in a study by Sarrafzadegan et al.
on ICS (Isfahan Cohort Study) data (2001)(2002)(2003)(2004)(2005)(2006)(2007), it was reported that younger individuals gained weight more than older ones (16). In a systematic review in Iran, increased age, urbanization, low educational levels, the female gender, and being married, were indicated as some variables that are associated with inequality in obesity (17).
Due to biological and sociocultural differences, obesity distribution patterns differ by gender and age (18). The risks of many chronic diseases commonly increase with age, which is called aging or longitudinal effect. Also, people born or enter the study of the same age have similar living conditions 4 that may affect their disease risk in a particular way; this is a cohort or cross-sectional effect (19). Therefore, decomposition of cohort and aging effects could provide some valuable information. The changes in the risk of obesity were assessed using longitudinal and cross-sectional effects of age, in some studies (20)(21)(22). In many research, separation of these two effects of age on obesity is of interest. Estimates of the effect of age from cross-sectional analyses may confound Cohort effects on obesity. Longitudinal assessment is needed to determine the aging effect. In contrast with crosssectional studies, a longitudinal study is defined as a study in which individuals are measured repeatedly through time. Longitudinal studies help us to distinguish changes over time within individuals (longitudinal effect) from differences among people in their baseline levels (cross-sectional effect) (23,24).
A limited number of longitudinal analyses of the prevalence of obesity in population-based samples were done to separate aging and cohort effects and allow simultaneous assessment of aging-related changes and secular trends. In the current study, a longitudinal approach is used to distinguish between cross-sectional and longitudinal effects of age. This study is a large community-based study Regression (MLR) model was used: 1-To determine whether the prevalence of obesity changes with age in the target population and whether the patterns of change in obesity are the same for women and men. 2-To assess the cross-sectional and longitudinal effects of age. The current study is the first study in Iran that measures the longitudinal and cross-sectional effects of age on obesity, using an advanced statistical modeling approach for longitudinal measurements, the MLR model. This study is benefiting from a population-based cohort study with 12 years of follow-up and is performed among the Iranian adult population. Investigation of prevalence and trend of obesity in a population-based study provides opportunities to target subpopulations who need more care and attention in public health interventions.

Methods
Study population 5 Isfahan Cohort Study is a longitudinal ongoing population-based study, including 6504 adults aged over or equal to 35-year-old at the first measurement occasion, from three provenances in the central area of Iran (Isfahan, Arak, and Najafabad), living in urban or rural areas. The participants had been involved in the first survey of the Isfahan Healthy Heart Program (IHHP), a community trial for prevention and control of Cardiovascular Diseases (CVDs) (25,26). These three cities were selected due to their population consistency and a smaller number of migrants in comparison with the capital and other cities in Iran. Also, Isfahan is the third-largest city in Iran. The multistage random cluster sampling method was used for sampling these provinces, which represent their society's distribution of age, genders, and residential area (urban/rural). Ethical permission was issued by the Ethics Committee of Isfahan Cardiovascular Research Center. Questionnaires and anthropometric information were collected after obtaining informed written consent in 2001, 2007, and 2013 (25).
The number of individuals in the first measurement occasion was 6504, and then during the follow-up, the number decreased to 3284 and 1735 at the second and third phases, respectively. Details of the study design and ICS challenges, which led to a decrease in sample size, were described in previous papers (25)(26)(27).

Variables under study
Measurement of height was conducted using a secured metal ruler in barefoot and, weight was measured in light clothing using a calibrated scale (26). BMI was defined as the weight (Kg) divided by height squared (m 2 ) (5). A BMI over or equal to 30 indicates obesity, according to the World Health Organization definition (12). Also, waist circumference (WC) (cm) was measured horizontally. It is defined as the smallest circumference below the lowest ribs (26). In our study, although baseline measurements for all individuals were recorded at the same calendar time, age of individuals varied at the entry to the survey. Accordingly, the variations in the prevalence of obesity with age have two potential sources of information. First, the cross-sectional (or between-subject) information is about how obesity changes with age in the baseline observations obtained in 2001. Second, since individuals were measured repeatedly through the study time, the longitudinal (or within-subject) information raised. "Baseline Age" was defined as the individual age (year) at baseline and is used to 6 assess the cross-sectional effect of age. "Age -Baseline Age" was defined as the years passed since baseline and is used to assess the longitudinal effect of age.

Statistical analysis
The characteristics of the participants at the three measurement occasions (2001, 2007, and 2013) are presented as a percentage or mean and standard deviation (SD) when appropriate. Considering the outcome as a binary variable (obese or not obese), Marginal Logistic Regression Model was used (equation (1)). The parameters in the model were estimated using Generalized Estimating Equations (GEE) method.
[Please see the supplementary files section to view the equation.] (1)

Results
The study population was approximately balanced according to sex ( Table 1 In this study, most of the participants had less than 12 years of education (93. changes in the prevalence of obesity over the follow-up period, are displayed in table 2. These percentages were reported based on the available data at each occasion in each age-gender group. Table 2 In  According to the rates of obesity in table 2 and Figure 1 which show the curvilinear effect of age on the rate of obesity and higher rates of obesity for women of all ages, an MLR model including gender, linear and quadratic age, and the gender by age interactions was fitted (equation (2)). Individuals were classified as obese ( ) or not obese ( ).
[Please see the supplementary files section to view the equation.] In this model, the gender by age interactions was not significant (P-value=0.45 and 0. 35  to omit this effect also, and the model in equation (3) was considered.
[Please see the supplementary files section to view the equation.] Table 3 In table 3, Alpha1, Alpha2, and Alpha3 are pairwise log OR among measurement occasions. The pairwise log OR between adjoining occasions are almost 2.9 and 3.5, which indicate that the OR for the within-subject association is about 18 and 33, respectively (or e 2.9 and e 3.5 ). Therefore, there is a severe positive association among the indicators of obesity status at all measurement occasions.
The significance coefficient of sex shows that log OR of obesity is different between women and men.
By controlling the age effects (longitudinal and cross-sectional effects), odds of being obese for women is approximately times the odds of being obese for men on average (Gender=1 was considered for men in the dataset and reference group in the model). Hence, women are more likely to be obese (almost three times) than men. Baseline Age group (or cross-sectional) effect controlling for the longitudinal effect of age and sex (20). The log OR of obesity alters curvilinearly with Baseline Age. According to the rates of obesity in Also, according to the estimated coefficients and ORs in Table 3, in this study, the cross-sectional effect of age was more considerable than the longitudinal effect of it. In other words, the Baseline Age was more strongly associated with odds of obesity than aging (linear combination of 0.1266 and -0.0013 for cross-sectional effect vs. 0.0295 for longitudinal effect).

Discussion
The goals in this study were first, to determine whether the probability of obesity changed with age and whether patterns of change in obesity were the same for women and men, and second, to assess the cross-sectional and longitudinal effects of age. Our study was a large community-based study in a group of 6504 Iranian adults over 35 years of age who live in the central region of Iran. In our study, the prevalence of obesity (in total) increased from 22% in 2001 to 31% in 2013, while according to the WHO report, between 2000 and 2016, in all WHO regions and income groups, the global prevalence of obesity raised from 9% in 2000 to 13% in 2016 (12). It shows that the prevalence of obesity in Iran was higher than the global prevalence of obesity (almost double) in the same period.
Also, the amount of increase in the prevalence of obesity in Iran was more than global prevalence (in Iran, an 8% increase in 12 years; global average, 4% in 16 years). This massive increase is almost as large as the increase in the American population, which increased from 20% in 2000 to 29% in 2016 (12). Also, the result of our study showed that, between 2001 and 2013, the prevalence of obesity increased from 13% to 18% (5 percentage points increase) in adult men and from 31% to 44% (13 percentage points increase) in adult women, while between 1975 and 2014, the global prevalence of obesity raised from 3.2% to 10.8% in men and from 6.4% to 14.9% in women (1). Shifts in diets and eating behavior which include energy-dense foods, high in sugars and fat, higher consumption of red meat, salt, and saturated fatty acids among individuals because of the increasing economic status, and less physical activity due to the sedentary work styles, types of transportation and changes in lifestyle are combining to increase the possibility of becoming obese (4). This rate of increase is also reported by other studies (1,6,15,16,30). In both developed and developing countries and specifically in middle-east countries, the prevalence of obesity is rising at alarming rates (1,4,13).
Additionally, in our study, using the GEE model and descriptive analysis, the odds of being obese for women is more than for men. It is approximately times of odds of being obese for men on average and by controlling the age effects (aging and cross-sectional effects). In other words, we can conclude that women are more likely to be obese (almost three times) than men. In another national study in In our study, for both groups of men and women, the average BMI and WC were raised over 12 years (2001 -2013), but women had a higher average BMI and WC than men in all three phases. Among women, lower physical activity level, pregnancy, menopause, a higher rate of depression, lack of employment, lower socioeconomic status, lower educational level, and gender differences in food intake may be some of the causes for the higher prevalence of obesity in comparison with men (30,(32)(33)(34). Iranian women pay less attention to their body shape in comparison to European and Oceanic women and also, less information and knowledge about weight loss may be another cause (15) Considerable regional differences in BMI alters through time were reported in different studies. In Southern Asia (Bangladesh, Bhutan, India, Nepal, and Pakistan), in southeast Asia ( Malaysia, Indonesia, Philippines, Sri Lanka, Vietnam, and Thailand), in the Caribbean (Cuba, Belize, Jamaica, Puerto Rico, and the Dominican Republic), and in southern Latin America ( Brazil, Argentina, Paraguay, Uruguay, and Chile), an accelerated rise in BMI was reported (1). In our study, applying an MLR model, it was shown that there is a difference between the cross-sectional effect and longitudinal effect of age on obesity prevalence in the target population, controlling for the gender effect. In this study, the cross-sectional effect of age was more considerable than the longitudinal effect of it. In other words, the Baseline Age was more severely associated with the odds of obesity than aging. The cross-sectional effect of age shows differences among people in their Baseline Age. There was a growth in the prevalence of obesity for individuals with Baseline Ages from 35 to 55 and a decline thereafter. In other studies, it is reported that Iranian adults gain weight until the age of about 50-60 years old, and after this age, BMI tends to decrease. The results of our study on the association between the probability of obesity and age are close to some of the other studies (16,17,(35)(36)(37)(38).
Previous studies reported that BMI had a tendency to be higher in individuals with middle-aged in comparison with young adults and also tended to be constant or decrease in older individuals. Also, in a study in the United States of America (USA), NHANES-III, BMI had tendency to go up for ages 20 to 29 years till ages 50 to 59 years, and after the age of 59, BMI tended to decline (38). In another study in the USA, an independent cross-sectional effect of age on obesity was reported and also, a higher probability for obesity in younger Baseline Age groups in comparison to older individuals after controlling the aging effects were observed (21). With increasing age, there are alterations in food intake, energy consumption, and appetite besides bone and muscle loss that affect body composition (30). Data are presented as percentage.