Sustained underweight in rural areas and emergence of overweight in urban Ethiopian women: a multivariate analysis of EDHS data 2000–2016

A growing body of evidence indicates the emergence of overweight/obesity in developing countries before the battle against undernutrition has been won. We conducted this study to quantify the reduction of underweight and the emergence of overweight among Ethiopian women from 2000 to 2016 and evaluate factors explaining the progress. We used the four Ethiopian Demographic and Health Surveys (2000–2016) to analyze body mass index (BMI) trends among women. Data from 43,815 non-pregnant, non-puerperal reproductive-age women was used to evaluate the linear change in BMI and changes in the percentage of overweight and underweight over time. Using multivariate decomposition analysis of change in underweight and overweight percentages, we identified sources of change in BMI in the past 16 years of the survey periods. The BMI of Ethiopian reproductive-age women increased by 0.88 kg/m2 from 2000 to 2016. The increment was pronounced in urban areas with 1.46 kg/m2. There has been a significant reduction in underweight women since 2000 (p-value < 0.001), and 87.62% of the changes were attributed to behavioral changes toward weight management. And there was a significant upswing in overweight women from 2000 to 2016 (p-value < 0.001) as well. A compositional change of factors including region, women’s age, women’s educational status, religion, type of place of residence, and use of contraceptives contributed to 57.51% of the observed increment in the percentage of overweight women. A relatively slow decrease in underweight and an increment in overweight have been observed. This progress can be disaggregated into persistent underweight in the rural and poorest, and swift development of overweight in the urban and richest communities. Targeted nutrition interventions for both underweight and overweight women are mandatory. Nutritional interventions in Ethiopia should focus on behavioral change to reduce hunger and malnutrition as well as to avert the emergence of overweight or obesity in the affected communities.

www.nature.com/scientificreports/ in BMI was evident in many developing regions 2 .BMI, with or without other variables, can provide reliable evidence in the prediction of overall mortality 3 .
BMI and overall mortality have a U-shaped relationship.Mortality is lowest among those with a BMI of 25-27.4kg/m 2 .Women with a BMI < 22.5 kg/m 2 had significantly higher overall mortality.Additionally, women with a BMI < 20 kg/m 2 displayed an increased risk of cardiovascular mortality 4 .High BMI is frequently linked with non-communicable diseases (NCDs).Overweight and obesity are major risk factors for a number of chronic diseases, including cardiovascular diseases such as heart disease and stroke, which are leading causes of death worldwide 5 .There is a rising concern over higher BMI which led to the inclusion of adiposity as global NCD target 6,7 .Not only higher BMI, the slimmest (BMI < 18 kg/m 2 ), although had a lower risk for cardiovascular disorders, the overall mortality was higher 8 .
The two extremes of BMI, which are addressed as a double burden of malnutrition in much literature, impose a great deal of problems at individual, societal, and intergenerational levels, especially in women 9 .The emerging dual burden of malnutrition impacts the health outcomes of mothers and offspring related to childbirth in lower-and middle-income countries.For example, in India, persistently short adult stature and adult obesity are associated with increased cesarean delivery 10 .Furthermore, evidence is compelling on the metabolic complications of the double burden of malnutrition.The recognized cardio-metabolic risks of being overweight are exacerbated among stunted mothers 11 .
In Africa, an analysis of prevalence and time trends in overweight and obesity in women in 24 countries indicated an increasing trend in overweight and obesity in all countries in the last three decades 12 .The prevalence doubled or tripled in half of the 24 countries in urban areas.Addressing the consequences of underweight or undernutrition is the primary agenda for African countries.While battling the burden of undernutrition, these countries are facing the other edge of the sword, overweight/obesity.
In Ethiopian women, there is growing evidence indicating the emergence of overweight/obesity especially in urban areas 13,14 .A recent cross-sectional study in southern Ethiopia gave us a glimpse of the coming uncertainty, that, more than a third of the women were either overweight or obese 14 .In the capital (Addis Ababa) a steeper rise in urban women's obesity was recorded from successive demographic and health surveys 13 .
As BMI is the current metric for anthropometric characteristics in adults, the knowledge of population-based trends has several advantages 15 .Besides the common interpretation of BMI as indicator of an individual's fatness, it is widely used as a risk factor for morbidity and mortality 16 and is widely used in formulating public health policies.The current study tried to show the progress Ethiopia has made in managing undernutrition and the neglected fact that we are facing as a developing country: the burden of overweight and obesity.In this study, we determined the trend in BMI from four Ethiopian Demographics and Health Survey data and the factors responsible for the change during the same period.The result of the current study also indicated the progress of underweight and overweight among Ethiopian women.Due to the paucity of such data, this study will provide concrete input for the current nutritional intervention strategies.

Study design and setting
Ethiopia is the second-most populous country in Africa next to Nigeria with a population of more than one hundred million.Administratively, Ethiopia is divided into eleven geographical regions (Tigray, Afar, Amhara, Oromia, Somali, Benishangul-Gumuz, SNNPR, Gambella, and Harari, as well as newly added Sidama and South West Ethiopia regions) and two administrative cities, Addis Ababa and Dire Dawa.Ethiopia shares the boundaries in the North with Eritrea, in the South with Kenya and Somalia, in the West with South Sudan and North Sudan, and in the East with Djibouti and Somalia.
The current study uses the four Ethiopian Demographics and Health Surveys (EDHS) conducted in 2000, 2005, 2011, and 2016.These surveys were conducted with nationally representative samples from all regions of the country.Details of sampling design, including sampling framework and sample implementation, and response rates are provided on respective EDHS reports (http:// www.measu redhs.com).The four EDHS data were used to determine the level and trend in BMI in reproductive-age women.These four DHS data were used to identify factors responsible for the change and to indicate the shift in variables.

Data source
The dataset used in this study was obtained from the MEASURE DHS database at http:// dhspr ogram.com/ data/ after getting the approval letter from the DHS program office.The EDHSs conducted from 2000 to 2016 have a population of 61,635 reproductive-age women.For the current study, we used 43,815 (sampling weights applied) women excluding pregnant mothers, mothers at puerperium, and unmeasured BMI.The survey was conducted with nationally representative samples from all of the country's regions with multiple enumeration areas.

Sampling procedures
The DHS utilizes a two-stage cluster sampling procedure; where a group of adjacent households serves as a primary sampling unit.In the first stage, a stratified sample of enumeration areas (EAs) was selected with probability proportional to size: in each stratum, a sample of a predetermined number of EAs is selected independently with probability proportional to the EA's measure of size.In the selected EAs, a listing procedure was performed such that all dwellings/households were listed.
In the second stage, after a complete household listing is conducted in each of the selected EAs, a fixed (or variable) number of households is selected by equal probability systematic sampling in the selected EAs.In each selected household, a household questionnaire is completed to identify women aged 15-49.The detailed www.nature.com/scientificreports/sampling procedure for DHS is available at https:// dhspr ogram.com/ publi catio ns/ publi cation-dhsm4-dhs-quest ionna ires-and-manua ls.cfm.
For example: during 2016 EDHS, in the first stage, a total of 645 clusters (202 in urban areas and 443 in rural areas) were selected based on the 2007 population and housing census.A household listing was carried out in all of the selected clusters from September to December 2015 and served as a sampling frame for the selection of households in the second stage.
The representative samples of 18,008 households were selected for the 2016 EDHS.From the interviewed households, 16,583 eligible women were identified for individual interviews.Interviews were completed with 15,683 women, yielding a response rate of 95% 16 .In this study, a total of 43,996 mothers were included from the four EDHS datasets (EDHS 2000 = 12,685, EDHS 2005 = 5523, EDHS 2011 = 13,282, and EDHS 2016 = 12,505).The steps followed in identifying eligible mothers are summarized in Fig. 1.

Study variables
The outcome variable of the study was a change in BMI across different times of the EDHS.BMI is calculated by dividing an individual's weight in kilograms by an individual's squared height in meters.The variable was classified as "underweight" (BMI < 18.5), "normal" (18.5 ≤ BMI < 25), overweight (25 ≤ BMI < 30), and "obesity" (BMI ≥ 30) 17 .However, for the sake of computational feasibility, the last two categories were merged as overweight.
Again, for the feasibility of our statistical analysis, multivariate nonlinear decomposition analysis with dichotomous outcome was applied.First we analyzed the change of underweight over time by comparing with normal and overweight individuals.In this case, we have dichotomous underweight versus others (

Statistical analysis
The data were cleaned and analyzed using STATA Version 16.0.First, descriptive statistics were performed to summarize the background characteristics of women after applying women's individual sampling weights.Second, the trend of BMI among total women, rural as well as urban women from 2000 to 2016 was computed after applying sampling weights.Percentage point differences were calculated to indicate positive and negative changes over different EDHS phases.
The study used Multivariate Decomposition Analysis to investigate changes in BMI among women.Multivariate Decomposition Analysis is an effective tool for comprehending the origins of disparities and shifts in outcomes among different groups or over time.By separating the contributions of various factors and breaking down observed differences in outcomes, such as BMI, between two groups or time periods, Multivariate Decomposition Analysis provides valuable insights for policymakers to tackle inequalities and enhance outcomes.
Our analysis considered both changes in population composition (endowment) and population behavior (coefficient) in relation to BMI.The endowment effect refers to the portion of the difference attributable to variations in explanatory variables, while the coefficient effect pertains to the differing impact of these explanatory variables across groups.Through the application of Multivariate Decomposition Analysis, the study identified and quantified these effects, offering a clearer understanding of the factors driving changes in BMI among women.
In the present case, Multivariate Decomposition Analysis was used to identify sources of change in BMI in the past 16 years of the survey periods.Decomposing the overall gap into these components, helps to understand the relative contribution of each factor.Decomposition analysis are used to understanding differences in nutritional and health outcomes that can shed light on whether observed gaps are due to starting point differences (endowments) or how effectively groups utilize their resources (coefficients) 18 .If the endowments effect is dominant, policies might focus on equalizing opportunities or resources.Conversely, a large coefficients effect might suggest a need for targeted training or mentorship programs.
We used logit-based decomposition analysis to identify factors contributing to the change in BMI over time.The observed difference in BMI between different surveys is additively decomposed into a characteristic (E) component-variation in the population composition across the survey and a coefficient (C) component-change in the behavior of the population.
The observed difference was decomposed in two components.The first component was difference due to variation in population composition across the survey.The second was due to change in the behavior of the population.The change in the outcome was due to either change in population composition or change in behavior of the population.The model structure for decomposition analysis was:

Ethical approval and consent to participate
Procedures and questionnaires for standard DHS surveys have been reviewed and approved by ICF Institutional Review Board (IRB).Additionally, country-specific DHS survey protocols are reviewed by the ICF IRB and typically by an IRB in the host country.ICF IRB ensures that the survey complies with the U.S. Department of Health and Human Services regulations for the protection of human subjects (45 CFR 46), while the host country IRB ensures that the survey complies with laws and norms of the nation (https:// dhspr ogram.com/ Metho dology/ Prote cting-the-Priva cy-of-DHS-Survey-Respo ndents.cfm).
Our study doesn't involve collection of information from subjects.We sent a one-page proposal abstract of the study to DHS program office.They gave a permission to access the data.

Respondents description
A total of 43,815 reproductive-age women (after applying sampling weights) were included in this study.The number of respondents (weighted) per DHS was 1246, 5368, 13,362, and 12,648 from earlier to recent surveys respectively.Table 1 presents the characteristics of respondents in each of the four surveys.As indicated, the percentage of women with higher education increased from 0.63% in 2000 to 5.99% in 2016.Another notable Vol.:(0123456789)

Trends in BMI and urban-rural disparities
The BMI of non-pregnant, non-puerperal Ethiopian reproductive-age women increased by 0.88 kg/m 2 from 2000 to 2016.The increment for the total women was pronounced in urban dwellers with 1.46 kg/m 2 , while the rise in BMI in rural women was relatively slowly moving with 0.63 kg/m 2 .Trends in BMI, as well as urban-rural disparities, are given in Fig. 2.
The percentage of underweight women has decreased, and the percentage of overweight women showed an increment, especially in urban areas.Table 2 and Fig. 3 indicate the urban-rural trends in BMI as classified by WHO.Of note is that the percentage of normal (BMI 18.5-24.9kg/m

Multivariate decomposition analysis
The percentage point differences in underweight and overweight women are given in Table 3.Generally, the table indicated the overall decrement in the percentage of underweight women and the increment in the percentage of overweight women.Regional variations exist in percentage changes; for example, little improvement has been observed in the reduction of underweight mothers in the Tigray region, whereas, the Benishangul Gumz region showed remarkable progress in a similar outcome (18.78 percentage points).The Amhara region registered the lowest percentage point increment in overweight women and Addis Ababa had the highest percentage point increment over the study period (12.83).As shown in Table 3, there was decrement in the percentage of overweight women who attended higher education, the percentage point difference was − 4.35.Vol:.( 1234567890)  3), where behavioral change towards BMI plays a significant role for the two regions.
Compositional factors contributed to 12.38% of the observed change in underweight percentage among Ethiopian women.Covariates in this category involve region and type of place of residence.As shown in Table 2, the percentage of underweight women dropped from 31.25 to 23.18% in 2016, this observed change was attributed to the compositional change in place of residence and other significant variables (shown in Table 4).

Overweight
There was a significant increase in women overweight from 2000 to 2016 (p value < 0.001) in Ethiopia.Table 5 depicts the contribution of both compositional (Endowment) and behavioral (coefficient) factors to the rise of overweight.Compositional change of factors contributed to 57.51% of the observed increment in the percentage of overweight women.Region, women's age, women's educational status, religion, type of place of residence, and use of contraceptive were among the significant covariates for the compositional change observed in overweight mothers.
As shown in Table 5, keeping the compositional factors constant, 42.49% of the change in the percentage of overweight women was attributed to behavioral changes toward BMI.These factors include the sex of the household head.As shown in Table 3, female-headed households had a speedy drive towards overweight-5.19%than male-headed households-3.86%.

Discussion
In 16 years of the survey period, Ethiopian women increased 0.88 kg/m 2 of BMI.This figure was significant and the urban women contributed much to the national BMI rise, which was 1.47 kg/m 2 .The percentage of underweight women dropped from 31.25% in 2000 to 23.18% in 2016, during this time the percentage of overweight/ obese women increased from 3.49 to 7.74%.Despite the decreased percentage of underweight women in urban areas of Ethiopia, the number of women having normal BMI decreased from 65.2% in 2000 to 63.58% in 2016, signifying that the drop in the percentage of underweight was overwhelmed by the accelerated development of overweight.
Global trends regarding BMI indicated the age-standardized mean BMI increased from 22.1 kg/m 2 in 1975 to 24.4 kg/m 2 in 2014 among women 19 .During this period global prevalence of underweight decreased from 14.6 to 9.7% and the prevalence of overweight increased from 6.4 to 14.9% 19 .This trend has also been observed in lower-and middle-income countries since the 2000s.Most of such countries decreased the prevalence of underweight, while countries like Senegal, Madagascar, and Mali failed to do so, and the prevalence in the three countries was increasing 20 .While this trend of increased overweight and decreased underweight was consistent across sub-Saharan African countries 12,21 , the ratio of underweight to overweight can be used as a proxy to measure countries' progress in reducing underweight while maintaining a stable percentage of overweight.In the Ethiopian context, the ratio of underweight to overweight decreased from 8.95 in 2000 to 2.99 in 2016 nationally.This is achieved by decreasing underweight and increasing overweight, and in fact, underpins the decrement of underweight was faster.In urban women, the ratio decreased from 2.06 in 2000 to 0.72 in 2016, which indicates the development of overweight was accelerating while 15.24% of the women are still underweight.In most developing countries, there is an emergence of overweight/obesity before the battle against underweight has been won, as observed in Ethiopia 22 .
Our finding disclosed that geographical endowment differences (regional differences or place of residence) contributed to the reduction in prevalence of underweight.Urban women are more likely to be overweight, and rural women are more likely to be underweight.In rural areas, women mostly engaged in agricultural activities, which involves physical exertion and therefore unlikely to gain as much weight as the urban women.Moreover, women in rural areas are less exposed to the western lifestyle and the nutritional transition in the urban areas www.nature.com/scientificreports/ is not likely to be prevalent in a rural setting 23 .Furthermore, it is likely that whereas urban women have more than enough to eat, women in rural have less.
Women's age contributed to the increase in prevalence of overweight.Body composition does not remain the same throughout life.Reduction of muscle mass begins in the 30s' and concomitant increase in adipose tissue will take place 24 .In addition various hormonal changes like ovarian steroids, growth hormone, insulin (impaired B cell function) coupled with decreased physical activity in a later age, all push towards overweight and obesity 24 .
Modern contraceptive use is associated with increased prevalence of overweight.Since most contraceptives are synthetic steroids, they tend to cause fat deposition in a different part of the body leading to overweight and obesit 25 .
Although, behavioral factors play the lion's share in significantly reducing the percentage of underweight, both compositional and behavioral factors were responsible for the increment of overweight over the past 16 years.While the current study showed behavioral factors had stronger effect on the change in underweight, a study among adolescent girls in Ethiopia pointed out compositional changes were major drivers of change for reduction of thinness 26 .These changes could related to "modernization" including increasing GDP per capita, urbanization, and women's empowerment 27 .Findings from other countries indicated both the compositional and behavioral factors were important elements as drivers change for overweight and obesity 28 .
To address both underweight and overweight in Ethiopian women, multisectoral approach, promoting diversified diet including animal source foods for underweights, promote healthy snacks and meals, reduce the consumption of soda, diversified diet including fruits and vegetables for overweight, and implementation of nutrition education are important strategies 29 .
This study used nationally representative samples of the country from DHS.The DHS data is worldwide data implemented across various countries and at multiple points in time within a country.It provides comparability, consistency, and the best-quality survey results.It strengthens the quality of data through probability sampling, full coverage of the target population, use of suitable sample size and household listing, and pre-selection of households 30 .
Besides, this study showed the trend of both underweight and overweight at the same time.In countries like Ethiopia, where malnutrition is a two-edged sword, the analysis of both under and over-nutrition was critical.Sub-Saharan Africa is experiencing the double burden of malnutrition; high levels of undernutrition on one side and a growing burden of overweight/obesity on the other side 31 .Undernourishment in sub-Saharan Africa increased between 2010 and 2016 32 .Meanwhile, overweight/obesity is increasing in all age groups, with girls and women being more affected than boys and men 32 .The introduction of cheap processed foods and non-alcoholic beverages into African markets drives the consumption of unhealthy diets 31 .Besides, sedentary lifestyle has role in the development of obesity 33 .Progress toward the goal of ending hunger and malnutrition by 2030 requires intensified efforts to reduce undernutrition and focused action on the reduction of obesity and diet-related NCDs 31 .
For statistical feasibility, we dichotomize BMI 34 .First, to evaluate underweight, normal BMI and overweight was merged.And in the second scenario, the outcome variable (y = 1), if BMI ≥ 25 kg/m 2 , and y = 0 otherwise.Associated with this procedure, although, obesity was not such a burning issue in Ethiopia; the study didn't show the change in obesity.Equally, to its strength of showing underweight and overweight trends; (1) the overweight category was added to the obese, and (2) the study didn't report decomposition analysis on the obesity category.We ran the analysis to understand the status of obesity in Ethiopian women, and there was an increment from 2000 to 2016 (shown in the supplementary material, Supplementary Tables 1 and 2).In 2000 1.3% of the women were obese, this figure increased to 3.9% in 2016 nationally.The increment in the proportion of obese women was largely attributed to the urban women, where, urban obesity increased from 4.57 to 12.6% in the same period.Little change has been observed in rural women regarding obesity.Similar trends in obesity have been observed, in several sub-Saharan African countries 35 .One critical issue in this analysis was the failure to analyze BMI changes based on the wealth index.This was raised from the unavailability of wealth index in 2000 EDHS data, but, is present in subsequent reports.We analyzed BMI trends and the contribution of wealth index from 2005 to 2016 (given in the supplementary material, supplementary Table 3).The mean BMI of the poorest was 19.52 kg/m 2 in 2005 and decreased to 19.39 kg/m 2 in 2016.Generally, the poor, the middle, and the rich had shown an increment in BMI, the increment of BMI in the rich quantile was greater than poor and the middle (the two were comparable).As expected, the richest had the highest increase in BMI-which was 1.01 kg/m 2 in the decade.Surprisingly, the mean BMI ratio of the richest/ the poorest was 1.09 in 2005, this ratio increased to 1.16 in 2016.It was a significant compositional factor for decrement in underweight in the richest quartile, and an increase in overweight in the rich and richest quartiles from 2005 to 2016, given in the supplementary material (supplementary Tables 8 and 9).
In developing countries both individual and country-level wealth were important factors 36,37 .Individuallevel wealth index and country-level average national income were associated with increased BMI.The BMI of the wealthiest quantile was increasing fast compared with the low and stable poorest quantile.This signifies the poorest stay thinner and the richest grow fatter.The association between BMI and wealth was positive in many developing countries was not affected by survey periods, and remained so recently 37 .

Conclusion
The percentage of underweight women decreased from 2000 to 2016.The decrement was pronounced in urban areas.Meanwhile, there was increment of overweight women nationally.Generally, the progress can be disaggregated as persistent underweight in rural and the poorest, and emergence of overweight in the urban and richest.Change in the population composition across surveys contributed to the increment of overweight.Behavioral changes largely contributed to the reduction of underweight.Nutritional interventions in Ethiopia should focus on the reduction of underweight in rural areas and lowest wealth quantile, and weight management in urban and highest wealth quantile.

Figure 2 .
Figure 2. Mean BMI changes in Ethiopian women from 2000 to 2016.It also displayed mean BMI changes in urban and rural women.Women in the urban Ethiopia had higher mean BMI at 2000 than the rural women (20.82 kg/m 2 and 19.48 kg/m 2 ), and the changes towards 2016 were also higher in urban women.

Figure 3 .
Figure 3.The trends in percentage of women having normal BMI, underweight and overweight in Ethiopia 2000 to 2016.The percentage of women underweight had decreased and overweight's increased.

Table 1 .
Description of the respondents in EDHS 2000 to 2016.NA: data not available.

Table 2 .
Trends in percentage of women underweight/overweight as classified by WHO, EDHS 2000 to 2016, urban-rural disparities.

Table 3 .
The percentage point differences in underweight and overweight Ethiopian women 2000 to 2016.

Table 4 .
Multivariate decomposition analysis of percentage changes in underweight among Ethiopian women 2000-2016.CI; confidence interval, pct; percentage *p value ≤ 0.05, **p value ≤ 0.01.As indicated in the table the change in the population composition contribute to 12.38% of the change in underweight.Region and place of residence were significant factors in this category.On the other hand, after controlling the effect of compositional factors 87.62% of the change in underweight was due to the effects of characteristics.

Table 5 .
Multivariate decomposition analysis of percentage changes in overweight among Ethiopian women 2000-2016.CI; confidence interval, pct; percentage *p value ≤ 0.05, **p value ≤ 0.01.This table depicts decomposition changes of overweight among reproductive age women in Ethiopia.As shown, keeping the effects of coefficients variation in the population composition across surveys contributed to 57.51% of the change in overweight.These variables are region, women's age, women's educational status, religion, type of place of residence, and use of contraceptive.