Trends and determinants of obesity among women of reproductive age - - PDF document

Trends and determinants of obesity among women of reproductive age in Bangladesh 2004-2014: A multivariate decomposition analysis

M Sheikh Giash Uddin, Department of Statistics, Jagannath University, Dhaka Mohammed Ahsanul Alam, NIPORT, Dhaka, Bangladesh Abstract: To assess the current levels, trends and gaps between socioeconomic groups of obesity in Bangladeshi women. The study also identified correlates that effect on obesity. Anthropometric data associated with socio-demographic characteristics among ever-married women, were extracted from the Bangladesh Demographic and Health Surveys conducted in 2004 to 2014. Households’ socioeconomic status was measured using principal component analysis. Logistic regression analyses were used to identify the determinants of obesity and logit-based decomposition analysis conducted for factor contributing to recent changes. It is found that prevalence of overweight increased (2004: 8.8%; 2014: 23.7%). The risk of being overweight was higher among women who were older and of higher socioeconomic status. The rich-poor gap in obesity is also significant. Wealth index, place of residence, use of contraception and the women’s education are identified as important correlates of women’s obesity. Women with upper quintile were 6.6 times more likely to be overweight compared with lowest quintile (OR 6.6; 95% CI: 5.6-7.8). Introduction Bangladesh has made commendable achievements on several Millennium Development Goals (MDG) set by the United Nations in the year 2000. These include reducing under five mortality two- third by 2015, taking primary education enrolment up to nearly 98 percent (NIPORT et al 2016). Maternal mortality rate and reducing prevalence of underweight children below 5 years are on

• track. For any populous country like Bangladesh, nutrition is of prime concern (NIPORT et al.,

2012). Though the problem of underweight children aged below five has been tackled, there are still

the more than a third of our children who suffer from stunting. The prevalence of overweight women is also a growing concern in developing countries. Obesity has significant health and economic consequences. In adults, they are associated with an increased risk

• f developing various non-communicable diseases (NCDs), including hypertension, coronary heart

disease, diabetes, stroke and some forms of cancer (Brown et al 2000; Sturmand and Hattori 2013;

(Dixon, 2010; Vucenik and Stains, 2012; Ojeda et al., 2014)). In developing countries, maternal

underweight is a leading risk factor for preventable death and diseases (Friedrich, 2002; Popkin et al., 2012). Overweight individuals are predisposed to a wide range of health problems such as diabetes and heart disease as well as poor birth outcomes for women (Dixon, 2010). In many countries, though, chronic energy deficiency, characterized by a BMI of less than 18.5 among adults remains the predominant problem, leading to low work productivity and reduced resistance to illness. In contrast, it has also been found that fats and oils constitute a large proportion of the daily diet of people with higher SES and that most of the population does not consume adequate fruits and

• vegetables. Another cause is urbanization which has been reported to be associated with a shift of

the BMI distribution of a population towards higher values, which is related to changes of the diet as well as lifestyle, in particular a reduction of physical activity. The projection suggests that by 2035 about half of the Bangladesh population will be urban, which will have impacts on the requirements for provision of basic needs, including the health care services. Rapid urbanization accompanied by rural urban migration is one challenge that must be addressed in improving maternal health and reducing risk of non-communicable diseases. In the light of above, the main aim is to assess the current levels, trends and gaps between socioeconomic groups of obesity in Bangladeshi women. The study also identified correlates that affect on obesity. Materials and Methods The study is based on four data sets of the Bangladesh Demographic and Health Surveys (DHS) carried out in 2004, 2007, 2011 and 2014 [NIPORT et al., 2005; NIPORT et al., 2009; NIPORT et al., 2013; NIPORT et al, 2016]. These surveys were conducted by the National Institute of Population Research and Training (NIPORT). All DHS are nationally representative and apply a common methodology across countries. Data for anthropometric measurements associated with socio- demographic characteristics among ever-married women, were extracted from these surveys. Body mass index (BMI) is calculated as weight in kilograms divided by height in meters squared. Obese is defined whose BMI is greater than or equal to 25. The present analysis included 10716, 10164, 16022 and 16534 women from last four surveys. The analysis excluded pregnant women and women with a birth in the preceding 2 months. Statistical Analysis

This study employed descriptive and trend analysis of prevalence of obesity, examination of the determinants of use, and decomposition of changes in percent of obesity. The trend in

• besity was analyzed using descriptive analyses, stratified by region, urban-rural residence,

and selected socio-demographic characteristics. The trend was examined separately for the periods 2004, 2007, 2011 and 2014. Logistic regression analysis was also done to identify the determinants of obesity among young married women, using data from the 2014 BDHS. Complex sample survey methodology was considered during analysis. Hence, the study adjusted for the effects of clustering due to sampling procedures and non-response. We used Blinder-Oaxaca (Oaxaca

Blinder)- decomposition (Blinder 1973; Oaxaca 1973), or as multivariate decomposition, decomposition techniques, component analysis, shift-share analysis or regression decomposition as detailed by Powers and Yun (2009),this approach provides a way to analyze the outcome of two different groups.

Multivariate decomposition analysis of change in obesity was employed to answer the major research question of this study. The analysis was a regression decomposition of the difference in percent in obesity between two surveys. The purpose of the decomposition analysis was to identify the sources of changes in the obesity in the last decade. Both changes in population composition and population behavior related to obese (effect) are

• important. This method is used for several purposes in demography, economics, and other
• fields. The present analysis focused on how obesity responds to changes in women’s

characteristics and how these factors shape differences across surveys conducted at different times. The technique utilizes the output from a logistic regression model to parcel out the

• bserved difference in prevalence of obesity into components. This difference can be

attributed to compositional changes between surveys (i.e. differences in characteristics) and to changes in effects of the selected explanatory variables (i.e. differences in the coefficients due to changes in population behavior). Hence, the observed difference in body mass index between different surveys is additively decomposed into a characteristics (or endowments) component and a coefficient (or effects of characteristics) component.

The model can be presented as follows: ∆Y2014-2004 =(X2014 – X2004) β2014 + X2004(β2014 - β2004) + [(X2014 - X2004)( β2014 - β2004)] Where ∆Y: Difference in mean prediction between 2014 and 2004, i….Xk: Different characteristics and βi….βk: estimated regression coefficients; (X2014– X2004) β2014: represent the difference due to endowment; X2004(β2014 - β2004): represent the difference due to coefficients, [(X2014 - X2004)( β2014 - β2004)]: represent the difference in interaction between endowment and coefficients. The Blinder-Oaxaca decomposition outputs provide details on endowments, coefficients, and interaction between the two time periods. As in the results, the interaction part is not statistically significant at the level of 5%, only the two major parts of the results will be presented: Endowments- part of the changes in occurrence of obesity due to differences in characteristics and Coefficients- part of the changes in occurrence of obesity due to effects of explanatory variables. In this analysis, socio-economic status was assessed by constructing a household wealth index (WI) based on principal components analysis. The DHS WI is an asset-based index that reflects the relative socioeconomic status of the household and is widely used in low- and middle-income countries to quantify inequalities and to control the confounding effect of socioeconomic variables. Descriptive statistics were use such as mean, standard deviations (SD) or percentages where appropriate. To measure the significance association among variables bivariate analysis were used with chi-square

• test. Logistic regression model was used to identify the proximate correlates on obesity among

women. Results The mean BMI for ever-married women age 15-49 years were 20.0 to 22.3, which falls in the normal BMI classification. During 2004–2014, a decreasing trend in the prevalence of thinness and an increasing trend in overweight were detected. Over the period, the prevalence of obesity steadily increased from 8.8 to 23.8% among women. Among urban women, prevalence of obesity increased considerably from 19.7% in 2004 to 36.4% in 2014. The prevalence of obesity increased at 1.7 percentage point/year in urban areas and 1.3 percentage-point/years in rural areas. The bi-vatiate analysis shows that the proportion of overweight women increases with age (Table 1). Urban

women are two times more likely to be overweight or obese (BMI>=25.0) than rural women (36.4 percent and 18.6 percent respectively). Among the divisions, the proportion of obese women was high in Dhaka and Khulna divisions. As educational attainment and household wealth rise, the proportion of overweight or obese women increases. Bangladeshi women from the highest wealth quintile are seven times more likely to be overweight or obese compared with women from the lowest wealth quintile. Ever-married women from households with food insecurity are much more likely to be thin than those from households in which food is more secure. Logistic regression analyses revealed that rural women aged 25 years or older were 3.0 times (95% CI: 2.6-3.4) more likely to be overweight compared with women aged <25 years (Table 2), and among the rural women, this risk was 25% (OR 0.76; 95% CI: 0.69-0.83) less, comparing the urban women (Table 2). Higher SES was significantly associated with over nutrition. The women of richest quintile were 6.6 times more likely to be overweight (95% CI: 5.6-7.8) than the women of poorest

• quintile. Women with secondary and above schooling had a 1.4 times higher risk of being
• verweight compared with their non-educated peers. There was about a 1.2 fold increase in the risk
• f being overweight among women who lived with long acting and permanent method.

Decomposition analysis Table 3 reports the mean prediction of obesity in 2004 and in 2014 and also shows how much of the difference attributable is to changes in women’s characteristics (endowments), variation attributable to the effects of these characteristics (coefficients), and their interaction. Overall from 2004 to 2014, there has been an impressive increase in obesity. The mean prediction has increased three times, from 0.089 to 0.237, resulting in an increase in prediction of 0.148. It is clear that the gap explained by the effects of selected explanatory variables is more important (0.122) than the gap explained by the changes in these characteristics (0.022). The interaction term (0.004) is not significant. However, even though the overall increase explained by the effects of the coefficients is higher than the gap explained by the endowment, the contribution of independent variables varies substantially from one variable to another and, according to categories of within variables (Table 5). In regard to the overall increase in obesity between 2004 and 2014 attributable to the changes in coefficients, the most important independent variables that provide significant contribution are income inequality (wealth quintile), age of respondent women, place of residence, accounting for 66%, 37.8% and 8.6% of the total difference between 2004 and 2014, respectively.

Results show that the contribution of changes in effects of socio-economic development is the most important, accounting for 66% of changes due to coefficient. The change in obesity between 2004 and 2014 is mainly explained by the categories of women with richest quintiles and those with richer and middle quintiles, at 19% and 12.8%, respectively. The compositional effects related to women’s educational level are important at 59% as well as higher order birth (12%). It should be noted that the intercept (-0.025) accounts for more than one-fifth of the change due to coefficients (-20.5%). This suggests that the model fit presented some limitations in explaining the increase in obesity between 2004 and 2014. Discussions A number of previous studies have reported that the emergence of overweight and obesity in developing countries like Bangladesh (Nsour et al, 2013; Sharma et al, 2016, Ghorbani et al 2015). Our study revealed that the prevalence of obesity increased considerably from 8.8 to 23.8 among women in Bangladesh. The prevalence of obesity in our study is remarkably low compared to the prevalence of obesity among U S women of childbearing age (Vahratian 2009). The results showed that different socio-economic factors are associated with women being obese. In multivariate analysis, women’s age, their education, number of living children, place of residence, food security, use of contraceptive methods, currently working status, and wealth index had a significant association with prevalence of obesity among Bangladeshi women. The findings of our study further revealed that higher age of women was significantly associated with higher odds of women being obese compared to women in lower age group. A previous study in Bangladesh has also demonstrated this association, with increased age as a significant predictor (Sharma et al, 2016). Same results were found in India and Jordan. In both, the association between

• verweight and obesity increased significantly with the increase in age (Nsour et al, 2013, Gouda

2014) . Women’s education had a positive and significant association with women being overweight or

• bese. Higher education opens better employment opportunities for women and leads to be self-

dependent and for further improvement in socio-economic status. This possibly helps women live a life which involves less physical activity and helps access energy-dense food which is considered to cause overweight or obesity (Sharma et al, 2016, Gouda 2014).

Women with higher parity are more overweight or obese in Bangladesh (Sharma et al, 2016) . Same result was found in another study in Jordan (Nsour et al, 2013). Overweight and obesity were found to be significantly associated with women’s place of residence. Urban women was significantly associated with higher odds of women being obese compared to women in rural areas (Nsour et al, 2013) . This study found that food security status had significant impact on women being obese. This findings of this study is consistent with previous studies. We also found contraceptive using was significantly associated with women being obese. Contraceptive users were more likely to being

• bese than non-users (Sharma et al, 2016).

The results of the study showed that women of wealthier households had higher odds of being

• verweight or obese compared to women of less wealthy households. This finding is consistent with

previous studies (Sharma et al, 2016; Ghorbani et al 2015). Possible reasons for the positive association between increase wealth and being obese are changes in dietary behavior with changes in income. Findings of the study suggest that, with the increase in income, the intake of higher energy and fat, and consumption of animals and processed foods increases, all of which are associated with overweight and obesity. For the compositional differences in groups, it appears that most of the endowment in obesity between 2004 and 2014 may be attributable to differences for women with wealth, women in urban, and had higher order birth. The government and stakeholders embarked on this issue. Conclusion The current study shows a consistently strong relationship between the obesity and the socio- economic and demographic factor of women. The concurrent evidence on the trend of the obesity would be a growing concern among women parallel to thinness. Given the double burden of malnutrition (undernourished and obese), the results revealed the importance of national prevention efforts for obesity and to reduce the risk of non-communicable diseases. In addition, research is needed to revisiting the life style and food habit of women to identify the increasing prevalence of overweight and obesity. Limitations of the Study

The analysis is limited for not being able to measure many other important factors, for example, those related to nutrition and food habit, which also affect obesity and malnutrition. Also, by pooling datasets, the study could not analyze some important variables that were not available in all the

• datasets. While decomposition analysis is a promising tool to analyze contributions of various factors

to changes in outcome, our model is constrained by limited availability of data to explain the difference. References Blinder A.S. 1973. Wage Discrimination: Reduced Form and Structural Estimates. The Journal of Human Resources 8: 436–455. Brown CD, Higgins M, Donato KA, Rohde FC, Garrison R, Obarzanek E, et al. (2000). Body mass index and the prevalence of hypertension and dyslipidemia. Obes Res. 2000; 8:605–619. Dixon, J.B.(2010).The effect of obesity on health outcomes. Mol. Cell. Endocrinol. 316, 104– 108.doi:10.1016/j.mce.2009.07.008 Friedrich, M.J.(2002).Epidemic of obesity expands its spread to developing countries. JAMA 287, 1382–1386.doi:10.1001/jama.287.11.1382-JMN0320-2-1 Ghorbani R, M Nassaji , J Jandaghi , B Rostami ,N Ghorbani (2015) . Overweight and Obesity and Associated Risk Factors among the Iranian Middle-Aged Women. International Journal of Collaborative Research on Internal Medicine & Public Health, 2015, 7: 120-131. Gouda J, R K Prusty (2014). Overweight and Obesity among Women by Economic Stratum in Urban

• India. Journal of Health Population and Nutrition, 2014, 32(1): 79-88.

National Institute of Population Research and Training (NIPORT), Mitra and Associates, and ORC Macro.2005.Bangladesh Demographic and Health Survey 2004.Dhaka, Bangladesh, and Calverton, Maryland: NIPORT, Mitra and Associates, and ORC Macro. National Institute of Population Research and Training (NIPORT), Mitra and Associates, and Macro

• International. 2009. Bangladesh Demographic and Health Survey 2007. Dhaka, Bangladesh, and

Calverton, Maryland, USA: NIPORT, Mitra and Associates, and Macro International. National Institute of Population Research and Training (NIPORT), MEASURE Evaluation, and icddr, b.

• 2012. Bangladesh Maternal Mortality and Health Care Survey 2010.Dhaka, Bangladesh: NIPORT,

MEASURE Evaluation, and icddr,b. National Institute of Population Research and Training (NIPORT), Mitra and Associates, and ICF

• International. 2013. Bangladesh Demographic and Health Survey 2011. Dhaka, Bangladesh and

Calverton, Maryland, USA: NIPORT, Mitra and Associates, and ICF International. National Institute of Population Research and Training (NIPORT), Mitra and Associates, and ICF

• International. 2016. Bangladesh Demographic and Health Survey 2014. Dhaka, Bangladesh and

Calverton, Maryland, USA: NIPORT, Mitra and Associates, and ICF International.

Nsour M. A, Gh. Al Kayyali and S. Naffa (2013). Overweight and obesity among Jordanian women and their social determinants. Eastern Mediterranean Health Journal, 19: 1014-1019. Ojeda,E.,Lopez,S.,Rodriguez,P.,Moran,L.,Rodriguez,J.M.,andDelucas,P. (2014). Prevalence of sleep apnea syndrome in morbidly obese patients. Chest 145(3 Suppl.),601A.doi:10.1378/chest.1781090 Oaxaca, R. 1973. Male-Female Wage Differentials in Urban Labour Markets. International Economic Review 14: 693-709. Popkin, B. M., Adair,L.S.,andNg,S.W.(2012).Global nutrition transition and the pandemic of obesity in developing countries. Nutr. Rev. 70, 3–21.doi: 10.1111/j.1753-4887.2011.00456.x Powers, D.A., and M.S. Yun. 2009. Multivariate Decomposition for Hazard Rate Models. Sociological Methodology 39(1): 233-263. Powers, D.A., and W.T. Pullum. 2006. Multivariate Decomposition for Nonlinear Models. The Stata Journal 11(4): 556–576. Sturmand R , A. Hattori . 2013). Morbid obesity rates continue to rise rapidly in the United States,” International Journal of Obesity, vol. 37, no. 6, pp. 889–891. Sarma, N Saquib, M Hasan, J Saquib, A S Rahman, J Rahman, J Uddin, Mark R. Cullen, and T Ahmed. (2016) Determinants of overweight or obesity among ever-married adult women in Bangladesh. BMC Obesity, 2016, 3:13. Vahratian A (2009). Prevalence of Overweight and Obesity among Women of Childbearing Age: Results from the 2002 National Survey of Family Growth. Matern Child Health Journal, 2009, 13(2):268-273 Vucenik,I.,andStains,J.P.(2012).Obesityandcancerrisk:evidence,mechanisms, and recommendations.

• Ann. N. Y. Acad. Sci. 1271, 37–43.doi:10.1111/j.1749- 6632.2012.06750.x

34 30 24 19 9 12 17 24 5 10 15 20 25 30 35 40 BDHS 04 BDHS 07 BDHS 11 BDHS 14

Percent

Figure 1: Trend in (2004-2014) Proportion of women with overweight and Undernourished

Undernourished Obese

.1 .2 20 40 60 20 40 60

BDHS 2004 BDHS 2014

Fraction Normal BMI

Body Mass Index (BMI)

Graphs by Survey Figure 2: Body Mass Index (BMI) Gaps between BDHS 2004 and BDHS 2014

Table 1: Prevalence of obesity (BMI ≥25) among women of reproductive age by Background characteristics Background characteristics BDHS 2004 BDHS 2007 BDHS 2011 BDHS 2014 Age in Year <25 4.2 5.5 8 13.3 >=25 11.1 14.6 19.8 27.8 p<0.01 p<0.01 p<0.01 p<0.01 Region Barisal 7 8.3 12.6 21.6 Chittagong 9.8 11.5 18 26.8 Dhaka 11.4 14.3 18 25.2 Khulna 10.3 12.1 19.8 27.6 Rajshahi 5.8 10.5 15.6 22.3 Rangpur

• 10.7

16.8 Sylhet 6.9 8.4 13 14.8 p<0.01 p<0.01 p<0.01 p<0.01 Type of place of residence Urban 19.7 24.2 28.7 36.4 Rural 5.8 8.2 12.1 18.6 p<0.01 p<0.01 p<0.01 p<0.01 Level of education No education 4.7 7.1 11.1 16 Primary 8 9.6 13.6 20.5 Secondary+ 16.1 18.3 22.2 29.5 p<0.01 p<0.01 p<0.01 p<0.01 Number of living children None 7 8.6 12.4 15.8 One 7.5 10.9 14.6 20.4 Two 10 14.4 19.5 28.6 Three or more 9.3 11.4 16.1 23.6 p<0.01 p<0.01 p<0.01 p<0.01 Region East 9 10.7 16.8 23.6 West 7.2 11 15.4 22 p<0.01 p<0.01 p<0.01 p<0.01

Table 1 (Continued): Prevalence of obesity (BMI ≥25) among women of reproductive age by Background characteristics Background characteristics BDHS 2004 BDHS 2007 BDHS 2011 BDHS 2014 Wealth index Poorest 2.1 3.3 5 8.6 Poorer 2.9 3.9 6.7 13 Middle 5 7.4 11.1 19.8 Richer 9.6 11.7 20.3 27 Richest 24.8 31.2 36.3 46.4 p<0.01 p<0.01 p<0.01 p<0.01 Respondent currently working No 9.6 13.4 16.3 25.4 Yes 6.6 8.7 17.4 19.8 p<0.01 p<0.01 p=0.343 p<0.01 Use of Contraceptive Methods No 7.9 10.6 16.3 22.9 Temporary 9.7 13 16.3 23.7 LAPM 9.1 11 17.9 25.9 p<0.01 p<0.01 p<0.01 p<0.01 Food Security Score Food secure 20 Mild food insecurity 11.1 Moderate food insecurity 7.4 Incomplete Severe food insecurity 6.9

Table 2: Logistic regression analysis on Obesity with selected demographic and socioeconomic characteristics of women (Dependent variable: Whether a woman is obese or not), BDHS 2014 Independent variables Odd ratio (95% CI of OR) Age in Year <25 1 ≥25 2.99 (2.64-3.38) Level of Education No education 1 Primary 1.38 (1.23-1.54) Secondary+ 1.74 (1.55-1.95) Number of living children None 1 One 1.07 (0.90-1.28) Two 1.34 (1.11-1.61) Three or more 1.20 (1.00-1.45) Type of place of residence Urban 1 Rural 0.76 (0.69-0.83) Use of Contraceptive Methods No 1 Temporary 0.94 (0.87-1.03) LAPM 1.15 (0.99-1.29) Respondent currently working No 1 Yes 0.79 (0.72-0.86) Region East 1 West 1.15 (1.10-1.25) Wealth index Poorest 1 Poorer 1.58 (1.34-1.87) Middle 2.35 (2.00-2.75) Richer 3.37 (2.89-3.94) Richest 6.59 (5.59-7.77) Constant 0.04

• 2 Log likelihood

15665.5 Model Chi-Square 2279.4

Table 4. Mean values of obesity predicted for 2004 and 2014

Mean prediction 2014 0.237*** Mean prediction 2004 0.089*** Total Difference 0.148*** Difference due to Endowments 0.022*** Difference due to Coefficients 0.122*** Difference due to Interaction 0.004 *** p<0.01; ** p<0.05

Table 5. Contribution of explanatory variables to the difference in obesity between 2004 and 2014

Endowment %

Coefficient %

Interaction % Age in Year <25 >=25 0.005*** 20.6 0.046*** 37.8 0.003*** 93.5 Level of Education No education Primary 0.000 0.2 0.006* 5.1 0.000 1.1 Secondary+ 0.013*** 58.8 0.004 2.9 0.002 55.4

S/Total

0.013 59.0 0.010 8.0 0.002 56.5 Number of living children None One 0.000 1.2 0.003 2.1 0.001 15.7 Two 0.001 4.0 0.010** 8.5 0.003** 70.0 Three or more

• 0.001
• 2.3

0.015** 12.1

• 0.003
• 75.5

S/Total

0.000 1.7 0.025 20.6 0.000

• 5.5

Type of place

• f residence

Rural Urban 0.004*** 17.2 0.011 8.6

• 0.001
• 21.9

Region West East 0.001** 2.7

• 0.019***
• 15.6
• 0.002***
• 43.9

Contraceptive Use No Temporary 0.000

• 0.3
• 0.005
• 4.4

0.000

• 12.7

LAPM 0.000 0.0 0.001 0.8 0.000

• 0.2

S/Total

0.000

• 0.3
• 0.004
• 3.7

0.000

• 12.8

Currently Working No Yes

• 0.003***
• 11.2
• 0.004*
• 3.4
• 0.002*
• 54.2

Wealth index Poorest Poorer 0.000 0.2 0.008*** 6.8 0.000

• 10.8

Middle 0.000 0.1 0.016*** 12.8 0.000 5.3 Richer 0.000 1.3 0.024*** 19.3 0.001 25.8 Richest 0.002** 7.5 0.033*** 27.1 0.002** 52.4

S/Total

0.002 9.1 0.081 66.0 0.003 72.7 Constant

• 0.025 -20.0

Total 0.022 100.0 0.122 100.0 0.004 100.0

*** p<0.01; ** p<0.05; ** p<0.10