Effect of national wealth on BMI: An analysis of 206,266 individuals in 70 low-, middle- and high-income countries

Background This study explores the relationship between BMI and national-wealth and the cross-level interaction effect of national-wealth and individual household-wealth using multilevel analysis. Methods Data from the World Health Survey conducted in 2002–2004, across 70 low-, middle- and high-income countries was used. Participants aged 18 years and over were selected using multistage, stratified cluster sampling. BMI was used as outcome variable. The potential determinants of individual-level BMI were participants’ sex, age, marital-status, education, occupation, household-wealth and location(rural/urban) at the individual-level. The country-level factors used were average national income (GNI-PPP) and income inequality (Gini-index). A two-level random-intercepts and fixed-slopes model structure with individuals nested within countries was fitted, treating BMI as a continuous outcome. Results The weighted mean BMI and standard-error of the 206,266 people from 70-countries was 23.90 (4.84). All the low-income countries were below the 25.0 mean BMI level and most of the high-income countries were above. All wealthier quintiles of household-wealth had higher scores in BMI than lowest quintile. Each USD10000 increase in GNI-PPP was associated with a 0.4 unit increase in BMI. The Gini-index was not associated with BMI. All these variables explained 28.1% of country-level, 4.9% of individual-level and 7.7% of total variance in BMI. The cross-level interaction effect between GNI-PPP and household-wealth was significant. BMI increased as the GNI-PPP increased in first four quintiles of household-wealth. However, the BMI of the wealthiest people decreased as the GNI-PPP increased. Conclusion Both individual-level and country-level factors made an independent contribution to the BMI of the people. Household-wealth and national-income had significant interaction effects.

regression analyses of mean, population BMI regressed against GDP per capita, a monotonically increasing relationship between national income and BMI up to a GDP of about USD $30,000 per capita. Unfortunately, there is no way to know how individual wealth within each country affected individual BMI-the issue of the ecological fallacy [14,15].
While the earlier analysis of WHS data by Nandi and colleagues [11] did not include highincome countries, data from these countries are available. This study explores the relationship between BMI and national wealth and the cross-level interaction effect of national wealth and individual household wealth using multilevel analysis. It was hypothesised that there would be a cross-level interaction effect, and this would extend into the high-income countries.

Methods
Data were from the World Health Survey (WHS) conducted in 2002-2004, which was launched by the World Health Organization (WHO) to provide nationally representative, valid, reliable and comparable information across 70 low-, middle-and high-income countries from all world regions. WHS is a unique comparable dataset available for 70 countries representing the low-, middle-and high-income countries. In each country, the target population was adults aged 18 years and over, living in private households. Participants were selected using multistage, stratified cluster sampling. This study was approved by the Monash University Human Ethics Committee.
BMI was used as the outcome variable, self-reported height and weight were used to estimate individual level BMI, calculated as weight in kilograms divided by height in meters squared. Several individual and country-level factors were included in the analysis as potential determinants (co-variates) of individual level BMI including, participants' sex, age, marital-status, education, occupation as well as household wealth and location (rural/urban). Age was measured in years and was centered at mean age of 41.1 years. Marital status was classified as married (including those co-habiting), never married and previously married (separated, divorced and widowed). Education was grouped into three categories; primary school or less, secondary school or college, and higher. Household wealth was determined using a wealth index which classified households based on their ownership of a range of permanent income indicators (household assets). The household items included in the index were the number of rooms in the home, the number of cars, the number of chairs, the number of tables, the presence of electricity; and household ownership of a: bicycle, bucket, washing machine for clothes, washing machine for dishes, refrigerator, fixed line telephone, mobile / cellular telephone, television, computer and clock [16,17]. A principal components analysis (PCA) was carried out separately for each country to determine the weights to create an index of the asset variables. The weights for the first component were then applied to each person's data giving a continuous asset index measure [16,18]. Four break points were created from the PCA score that define wealth quintiles as: Quintile 1(poorest), Quintile 2 (lower-middle), Quintile 3 (middle), Quintile 4 (higher-middle), and Quintile 5 (wealthiest). Occupation was grouped into four categories following the Goldthorpe schema [19]: High (Legislator, Senior Official, Manager, Professional and armed forces), medium (Technician, Associate Professional, Clerk, Service or sales worker), low (Agricultural, fishery worker, Craft, trades worker, Plant/machine operator or assembler) and elementary (elementary workers).
The principal country-level, economic factor used was national income measured by Gross National Income adjusted for purchasing power parity (GNI-PPP) from World Bank data for 2003 [20]. GNI-PPP was centered at mean GNI-PPP at USD 8840. Income inequality was included as a potential confounder. Income inequality was measured using the Gini index based on World Bank data [20]. The Gini index varies from 0 (perfect equality) to 100 (perfect inequality) [21].
Survey analytic procedures were used to account for the complex survey design (stratification and clustering) and incorporate sampling weights to generate population-level estimates and standard errors for each specific country. The R statistical environment (R-3.1.0) was used for the analyses [22]. Lumley's survey package was used for all design-based analyses, [23] and Bates' linear mixed-effect package, lme4 was used for all multilevel, linear regression analyses [24]. The primary analysis was conducted to test the global null hypothesis for all the predictor variables using a two-tailed test and a significance of 0.05. F-values and p-values were calculated for each predictor variable.
A two-level random-intercepts and fixed-slopes model structure with individuals nested within countries was fitted, treating BMI as a continuous outcome. The fixed-and randomparameter estimates for the two-level regression model were calculated using the full maximum likelihood estimation method, as implemented in lme4. Multilevel modeling incorporating survey design features is a matter of ongoing debate with no agreement about the analytic strategy and in any case, unavailable in lme4 [23].
Our modeling strategy was first to estimate the null model (Model 0) and then to include explanatory variables gradually into the model. All individual-level factors were included as explanatory variables in Model 1. Country-level factors (GNI-PPP and Gini index) were included as explanatory variables in subsequent models. The cross level interaction effect between individual level wealth and national income was assessed in final model.

Results
A sample of 206,266 people from 70 countries was included in this study. Sample size and response rate for each country is given in Table 1. Weighted and unweighted descriptive analysis of individual level variables for is presented in the Table 2. The weighted mean BMI and standard error (SE) in these 70 countries was 23.90 (4.84). Weighted mean age (SE) of the participants was 41.1 (0.17).
To analyse the pattern of BMI in across all 70 countries, the design-based mean BMI (with 95% confidence intervals) were calculated (Fig 1). Most low-income countries were at the lower end of the mean BMI, and high-and middle-income countries, at the higher end of the mean BMI. All the low-income countries had a mean BMI below 25.0 and most of the highincome countries had a mean BMI above 25.0. Middle-income countries were scattered in this spectrum from low to high mean BMI.
Model 0, the null model or the variance component model for BMI is shown in Table 3. The fixed part of the model is represented by the coefficient for the constant, which is 24.3 with standard error of 0.20. The random part is given under the heading "Random effect" for variance of level 1 residuals. The estimate of the between-countries variance was 2.75 and the estimate of within-countries variance was 20.07. These estimates for random effect were used to calculate the intra-class correlation coefficient (ICC = 0.125). The ICC suggests that the proportion of total variance that occurs between countries is 0.125. That is to say, 12.5% of the variance in the individual level BMI was between countries and remaining 87.5% of the variation in the individual level BMI was within countries.
The combined effect of all individual level variables on BMI was tested in model 1 (Table 3). Age was positively associated with BMI, every 10 years increase in age was associated with a 0.34 units increase in BMI. Gender was not significantly associated with BMI. On average people with secondary education had higher BMI then people with primary education. Married people had significantly higher BMI than never married and previously married people. Household wealth was also significantly related to BMI. All wealthier quintiles had higher scores in BMI compared with the lowest quintile when, holding all the other variables constant. Professionals and elementary workers did not have significantly different BMIs; however, people in the low occupation category had significantly lower mean BMI than professionals. People living in the rural areas had an average BMI significantly lower than people living in urban areas.
To assess the effect of country level factors on BMI after controlling all individual variables, country level variables including national income and income inequality were added in model 2. First, the association of GNI-PPP with BMI was tested. This showed a 0.4 unit increase in BMI with each USD10,000 increase in GNI-PPP. Later, in model 3 national income and income inequality were added together controlling for all individual level variables. In model 3, the regression coefficient for GNI-PPP remains significant but the regression coefficient for Gini index was not significant. All the individual level variables had similar relationship as in model 2. This model explained 28.1% of country level, 4.9% of individual level and 7.7% of total variance in BMI across the 70 countries.
The cross-level interaction effect between national income and individual level income was modelled to measure the effect of national income on the relationship of individual level income and BMI (Table 4). These results showed a significant interaction effect between all the individual level wealth quintiles and GNIPP except quintile 2. To make results of this model more interpretable I graphically present the interaction effect in Fig 2. This graph shows that as the national income increases people in the first four quintiles show increasing BMI with    increasing national wealth. However, the wealthiest quintile shows the reverse pattern. The BMI of the wealthiest people decreases as the national income increases.

Discussion
The major finding of this study is that there is a significant cross-level interaction effect between individual wealth and country level wealth and individual BMI. In the four poorest quintiles, increasing individual wealth and increasing national wealth are associated with increasing individual BMI. However, in the wealthiest quintile, BMI steadily decreases with increasing levels of national wealth (Fig 2). This is a potentially important result for several reasons. First, it suggests that earlier results from equivalent studies using exclusively LMIC data, are likely to extend into high-income country data [9]. Second, the results suggest that as countries become wealthier, the BMI outcomes for the bottom 80% of the population become increasingly worse, while the BMI outcomes for the wealthiest 20% of the population become increasingly better. This alone, could have significant consequences for the equity declaration of the Sustainability Development Goals, "to leave no one behind". The results of this study showed a strong positive association between individual income/ wealth and obesity: BMI increased with increases in income/wealth, after adjusting for national income and national income inequality. This global association is similar to the obesityincome/wealth relationship in low-and middle-income countries. In low-and middle-income countries people with higher income/wealth had a higher prevalence of obesity. The majority of the studies, which used income or wealth as an SES indicator showed that wealthier people were more likely to be obese in low-and middle-income countries [10]. An important reason for this trend in 70 WHS countries is that most of the WHS countries were low-and middleincome countries. As most of the countries were low-and middle-income the overall analysis showed a similar pattern to other low-and middle-income country studies.
There are various reasons for the positive association of BMI and wealth in low-and middle-income countries. Household wealth/income enhances the household assets, including owning a car, washing machines, that significantly increased the risk for obesity. Additionally, it has been established that a better economic standing primarily affects obesity in terms of the resources available to buy more food. Therefore, as income increases, households and individuals increase their consumption of food and reduce their energy expenditure, and consequently BMI increases [16,25].
A shift in income from low to high usually associated with the nutrition transition characterized by a shift towards an unhealthy diet of higher fat and calories and decreased physical National income and BMI activity at work or leisure [17]. In the transition, peoples' daily diets rely more on animal food sources, and their lifestyles are increasingly sedentary, with less physical activity. Moreover, it could also be linked to excessive consumption of higher calories and fat condensed food (such as animal foods and processed food [17]. In addition, high-income people were at increased risk of snacking and shifting away from traditional healthy cooking patterns to less healthy cooking patterns and less healthy food [18]. Hence, people with higher income and more wealth may increase their risk of obesity. This study showed a clear gradient in the national income and obesity relationship, where people in poor countries have lower BMI than people in high-income countries. After keeping all other things equal, low-and middle-income countries on average have a lower BMI than high-income countries. Every 10,000 USD increase in GNI-PPP is associated with 0.3 unit increase in BMI. These results are in agreement with previous cross-national studies identifying a positive association between obesity and national income [9,[26][27][28]. A positive correlation between national income and BMI exists, with the prevalence of obesity being greater in developed countries than less developed countries, and obesity rates increasing as per capita incomes increases [29,30]. However, some previous studies showed no association of BMI and national income, but the majority of these studies were based only on high-income countries [31].

Model 4 Global null hypothesis
The positive associations between high national income and higher BMI or obesity are attributed to differences in lifestyle behaviours that accompany economic development and urbanization (e.g., alterations in the quantity and sources of caloric intake, and changes in physical activity). While its main proximate cause has been identified as a surge in extra-meal snacking and secondary eating consumption (including eating more, and buying more entertainment and energy saving devices), a decline in physically demanding labour [32], changes in food production technologies and prices have all been found to contribute to obesity development [33,34].
While income inequality was treated as a potential confounding variable in this study, and was not the specific focus of any hypothesis, there has been sufficient research looking at National income and BMI income inequality as a putative cause of health inequality that the results are worthy of consideration here [26,27,31,35,36]. After controlling for national income, household income, and other individual level factors, national income inequality was not significantly associated with BMI. It is possible that the effects of the income inequality are already subsumed by income/ wealth at the individual-level and national income at the country-level [16]. It is also possible that the country level income inequality is simply not associated with BMI. It is perhaps the absolute income of a person and the absolute income of a country that makes unhealthy/health food accessible or unhealthy/health lifestyle accessible.
These results are in contrast with the majority of the earlier literature on income inequality and health [31,33]. The positive correlation between income inequality and obesity prevalence was observed in most developed countries including the U.S. [13], Europe [26], and OECD countries [31]. Many studies by Wilkinson and colleagues reported the detrimental effect of income inequality on health (mortality, morbidity and self-reported health status) in the OECD countries [37]. As this evidence was predominantly from high-income countries, it is National income and BMI possible that the positive association between income inequality and poor health reported by Wilkinson and colleagues only have effect in the high-income countries where Gini is low, but not for the low-and middle-income countries. However, the inverse Gini effect on obesity has also been observed for some developing countries such as China and India [38]. On the other hand, there are studies that found no significant relationship between income inequality and health [27,35,39,40].
Most of the countries included in this study had good response rates of more than 60%, with the exception of Bangladesh and Ethiopia. Achieving high response rates in national surveys is always challenging, especially for low-and middle-income countries. Lack of information on non-respondents and exclusion of these non-respondents for weight or height is a limitation of this study. However, the extent of the bias, if any, which could have been introduced could not be assessed.

Conclusion
Both individual-level and country-level socioeconomic factors make an independent contribution to the BMI of the population. In the view of income inequality, household income and national income have independent, albeit unequal effects on obesity. The pattern is consistent, regardless of the other individual level factors. Meanwhile, the association between income inequality and obesity risk warrants further investigation.

Author Contributions
Conceptualization: MM DR.

Key Points
• The major finding of this study is that there is a cross-level interaction between individual wealth and country level wealth and individual BMI.