Socioeconomic patterning of Overweight and Obesity between 1998 and 2015: Evidence from India Shammi Luhar Supervised by Lynda Clarke & Prof Sanjay Kinra The London School of Hygiene and Tropical Medicine September 29, 2017 1 Introduction Overweight (OW) and obesity (OB) are increasingly threatening health in transitioning economies [1][2], and are responsible for 3.4 million deaths per year globally[4][5][6][7]. In India, between 1998 and 2006, the prevalence of OW and OB among women (15-49) years increased from 10.6 to 12.6% [13][14]. Since then, the number of obese women has doubled[12]. OW/OB prevalence is associated with a high socioeconomic position (SEP) in developing countries, due to richer diets and more sedentary lifestyles compared to the relatively poor. In rapidly developing countries, the risk of OW/OB increases among the poor, in part due to cheaper costs of high calorie food, making the positive socioeconomic gradient in OW/OB become less positive and eventually turn negative. Studies also find a higher prevalence of OW/OB among women, in addition to a rise in OW/OB among poor women at earlier stages of development, compared to poor males [3]. Although some studies have attempted to explain variation in OW/OB in India using nationally representative data, there is little understanding of the socioeconomic patterning of OW/OB since 2005-06 sub-nationally, the level at which health policy is dictated. With the aim of identifying the groups currently most a ff ected by OW/OB, and understanding how the socioeconomic patterning of OW/OB is evolving, we aim to address the following questions: • Is the overweight/obesity-SEP association in India in 2014-15 less positive than in 1998-99 (2005-06) for women (men)? • Is the 2014-15 overweight/obesity-SEP association in India less positive among women, compared to men? • Is the overweight/obesity-SEP association less positive in 2014-15, compared to the initial period, only in high GDP per capita (pc) states? 2 Data and Methods Nationally representative data from NFHS waves 2, 3 and 4 (1998-99; 2005-06; 2014-15) will be used for the study. In waves 2 and 3 health and sociodemographic data was collected on 90,303 and 124,385 women respectively, and 74,369 males in NFHS-3. The forthcoming 4th wave will contain data on 628,826 women and 94,324 men. Body Mass Index of ever-married individuals has been be used to create outcome variables of OW/OB ( > 22.99 kg/m 2 ) and OB ( > 27.49 kg/m 2 ) as per guidelines for South Asian populations[8][9]. The following socioeconomic variables will be used as the key exposures. • A time comparable Wealth index has been created and split into three tertiles, as per the method by Rutstein and Staveteig (2014). The NFHS Wealth Index aims to capture household economic status that mirrors their expenditure and income position based on ownership of particular assets and access to services. [17]. 1
• Respondent’s education has been categorised as individuals with no education, primary, secondary and higher education (based on the number of completed years of schooling) [10]. • Residence is defined as either rural or urban, based on the census bureau’s definition. Logistic multilevel regressions will be adopted for multivariate analysis in order to account for the clustered nature of the data and avoid standard error underestimation. Three level random intercept models will be used for the national level analysis, with individuals nested in PSUs, nested in states. Two level random intercept model will be used for subnational analysis, with individuals nested in PSUs. To examine a changing association between survey waves, the primary exposure will be interacted with a categorical variable representing the survey wave. In a fully adjusted model, controlling for all exposures at the same time, we would expect some covariates to lie on the pathway between the key exposure and ‘Overweight/Obesity’ odds [18] (Figure 1). For instance, the association between higher education and ‘Overweight/Obesity’ may be partially mediated by higher standard of living (Wealth Index). Therefore we provide results of minimally (adjusted for age and parity) and fully adjusted models, and expect the true odds ratios to lie between them. To investigate variation in the association by state-level development, the data will be subsetted to the following two groups, and analysis carried out on them separately: • High GDP states:Gujarat, Maharashtra, Tamil Nadu, and Kerala (average GDP pc 2013-14 = US $1,694.57) [19]. • Low GDP states: Uttar Pradesh, Bihar, Madhya Pradesh, and Assam (average GDP pc 2013-14 = US $633.02) [19]. 3 Preliminary Results We currently await the release of NFHS-4 survey data to complete the analysis. Preliminary results show that in NFHS-2 and 3 across India, for men and women, the odds of OW/OB is highest for those with higher education, residents of urban areas, and those in the highest wealth tertile (Figures 2 and 3). As initially hypothesised, the social gradient in 2005-06 is less positive than in 1998-99 for women when considering ‘Wealth index’ as the primary exposure. No change in the OW/OB-SEP association was observed when considering education or residence as the main exposure. In the 2014-15 data, we expect to find an even smaller odds of OW/OB among women in the highest wealth tertile, relative to women in the lowest, in addition to observing a less positive OW/OB-SEP association when using education and residence as the exposure of interest. As only one time period is currently available with which to analyse the association among men, any evidence of a less positive social gradient can only be ascertained upon release of the 2014-15 survey data. Each bar in Figure 4 shows the di ff erence between the predicted probability of OW/OB of the most and least advantageous strata of the three primary socioeconomic variables. A bar exceeding a value of zero indicates a positive OW/OB-SEP association, whereby the predicted probability is higher amongst the highly educated, the top wealth tertile or urban residents, compared to those with no education, those from the lowest wealth tertile, and rural residents, respectively. Initial results indicate a positive association between OW/OB and SEP, in high and low-GDP states among females in both waves, and males in wave 3. Some evidence is provided of a more positive social gradient in OW/OB in 2005-06, compared to 1998-99 for women in low-GDP states, irrespective of the exposure. Conversely, there is some evidence of a less positive OW/OB-SEP association in high-GDP pc states, when using residence or education as the primary exposures, and a stagnation in the association when considering the wealth index. As high-GDP pc states have continued to develop at a faster pace than low-GDP pc states in the decade since 2005-06, we expect to find a smaller positive association in NFHS 4, and a larger positive OW/OB-SEP association in low-GDP pc states. 4 Conclusion/Expected findings with NFHS-4 Using NFHS-4 survey data, we expect to find a persisting stronger positive social gradient in OW/OB in males compared to females, however, an overall decline relative to NFHS-2 and 3. Our findings thus far are similar to those of Sengupta et al [11], who find the strongest increase in OW/OB among women of the lowest band of standard of living index and rural areas in a handful of states defined by a high prevalence of OW. To our knowledge, this is the first study to examine the evolution of the socioeconomic patterning of OW/OB among both males and females using nationally representative Indian data post 2005-06. Identifying diverging trends between states with di ff erent levels of 2
economic development, highlights the limitations of such analysis for India as a whole. Given the doubling of the number of individuals classified as obese in the last decade, understanding the most a ff ected groups in society will crucial to developing e ff ective combative policy. Figure 1: Framework of potential pathways socio-economic variables can a ff ect overweight/obesity in a developing country setting (based on Samal et al’s 2015 [18] framework) Figure 2: Association between socioeconomic characteristics and OW/OB for women 15-49 in NFHS Waves 2, 3, and 4 3
Figure 3: Association between socioeconomic characteristics and OW/OB for men 15-54 in NFHS Waves 3 and 4 Figure 4: Di ff erence between fully adjusted predicted probabilities of the most and least advantageous strata of SEP in each of the three NFHS waves that measure BMI (by state development and sex) 4
Recommend
More recommend
