Introduction

Wasting, as an indicator of acute malnutrition, is defined by a weight-for-height/length z-score (WHZ) of more than -2 standard deviations below the World Health Organization (WHO) growth standard [1]. Wasting represents thinness and signifies recent failure to gain optimal body weight or loss of body weight, which is often associated with persistent hostile environments, chronic disease or acute starvation [2–4]. Wasting is regarded as a leading contributor to under 5 child mortality and morbidity in low and middle income countries [5–7], with the case fatality rates for wasted children around 5–60% [8–10], in part because wasting acts as a precursor to common childhood illnesses such as diarrhea and pneumonia. Adverse effect on long term physical and cognitive development have also been suggested [8].

Though the prevalence of malnutrition in Asia is decreasing, Bangladesh and its neighbors still have the highest cumulative number of malnourished children [11–13]. The significant reduction in prevalence of stunted and underweight Bangladeshi children in the past 10 years, highlights the considerable effort of the government and other national and international stakeholders in nutrition and holistically portrays the success of the country’s health system. The stagnancy in the prevalence of wasting, which has continued at around 15%, a figure that the World Health Organization (WHO) class as ‘critical’ [3], may not then, be due to any systematic failure of the health system. Some assessment of the many intervention programs oriented towards the reduction of incidence and prevalence of wasting in the country along with the associated risk factors is then needed to understand the point at which continued improvement can be found.

Factors associated with wasting are diverse [4,14,15], but can be broadly categorized into proximal maternal and child functions and distal socio-economic determinants [16]. The socio-economic factors consist of resources needed for adequate food security, hygienic environment and optimal child care [15]. Despite having no general consensus among the scientific community about which indictors appropriately contribute to a composite socio-economic variable, cited literature suggested women’s education, access to safe water and sanitation and household economic status can be used as proxy indicators [15].

The effects of socio-economic status (SES) on WHZ is well documented [4,17]. Many studies have also identified associations between poor Water, Sanitation and Hygiene (WASH) practices with wasting [18–21], which is critical given that WASH is a practical intervention in the malnutrition disease cycle. Socio-economic disparity in context of optimal hygiene and WASH practices has been the salient finding of many studies [22–24]. Optimal WASH practices require improved education, infrastructure and communal improvements to water and sanitation, all of which is reflected by the socio-economic conditions of the households.

Many studies [12,25–29] have used variables related to safe water consumption, optimal sanitation and hygiene practices to construct a domain commonly referred as WASH. The critical role of optimal WASH practices for better health is well established [30–32] and with regards wasting in particular [13]. It should be noted that optimal WASH practices invariably includes the use of soap [33,34]. Handwashing using soap removes or kills transient flora present on the skin surface [35,36], for example bacteria such as P. pyocyanea and strains of Salmonella, Shigella and E. coli [37] that are all associated with diarrheal disease [38,39]. On other aspects of WASH, a study determining the factors associated with childhood malnutrition in 36 low and middle income countries suggested that the use of pit latrine and flush toilets have significantly positive effect on WHZ [15], which can also be linked to reduced rate of gut or enteric infections [32]. However, a comprehensive review which included analysis on the effect of improved water quality, sanitation and hygiene practices on WHZ of 4,322 under 5 children score found no such significant association [12]. Similar statistically non-significant associations between SES determinants and WHZ were observed in nationally representative samples of under 5 Bangladeshi children [14,17,32]. The inconsistent effect of WASH and SES on WHZ may be due to statistical methods used to measure and represent SES and WASH in relevant studies, for example the number and identity of other regressors included in analyses and the inherently overlapping variance explained by individual SES and WASH indicators. More importantly, for WASH interventions not to work effectively in resource poor settings, like that of Bangladesh, is the intervention components itself. Existing WASH interventions showed to hinder fecal-oral transmission of pathogens, which is responsible for causing undernutrition among children [29]. Common WASH interventions in Bangladesh failed to provide improved environmental hygiene for children as the interventions were not effective to moderate risky behaviors such as crawling and picking objects from contaminated soil and surfaces [29] and pica [40].

Therefore, to explicitly address the relationships between SES and WASH and their joint relationships with WHZ, a Structural Equation Model (SEM) using the causal mediation analysis approach [41,42] is presented. The SEM offers an integrated approach with increased validity as the construction of latent variables captures the multifactorial properties of inter-related domains of which no single observable variable would be representative [42,43].

To formally address the interrelationships of SES and WASH and their respective contribution to the prevalence of WHZ, we hypothesize that SES affects WHZ, but that SES is positively associated with WASH and that this acts indirectly on WHZ as well. The objective of this study is to assess the causal change in WHZ by SES and WASH.

Materials and methods

Latent constructs

Socio-economic variables.

A household socio-economic status (SES) latent variable was constructed from: ownership of homestead land, agricultural land, household presence of electricity, solar panel, radio/television, telephone/mobile phone, ceiling fan, khat/chawki (bed/cot), almirah (wardrobe)/showcase, refrigerator, table/chair, clock/watch, bicycle, motorcycle/scooter/tempo (automatic tricycle), animal drawn cart, car/truck, country boat, motorized boat, rickshaw/van, power tiller, shallow machine (primarily used as water pump), Uninterrupted Power Supply device (UPS)/electric generator, fishing net; cane/palm/trunks, dirt, bamboo with mud, tin, cement and bricks and planks/shingles as major construction material of house’s walls; wood, thatch/palm leaf and tin as major construction material of the roof; wood, straw/grass/leaves/agricultural waste, animal dung and other materials (kerosene/charcoal/bio-gas/natural gas/LPG gas etc.) as fuel used while cooking; ownership of cow/buffalo, sheep/goat/pig, chickens/duck/geese and small game (rabbits, pigeons, etc.). Additionally, food security status of the household (whether or not the house was food secure) as indicated by Household Food Insecurity Access Scale (HFIAS) [45]; income generating status of the mother of the child; education status of the mother (at least one year of formal schooling); and household income (previous month’s income).

Water supply, sanitation and hygiene (WASH).

The latent WASH variable was constructed using responses from mothers on the use of soap while washing clothes, child’s bottom, child’s hand and washing own hand after defecating, after cleaning a child, before feeding a child, before preparing food and before eating and during bathing (child and self). Other variables included were the use of hygienic latrine by household members and availability of a safe source of water for drinking. Safe water sources included rain or tubewell water and if it was covered stored for later use.

Covariates.

Three covariates- child’s age, sex and residence type were adjusted for. Child’s age was regressed on WHZ only; child’s sex was regressed on WHZ and WASH; residence type (urban/rural) was regressed on SES, WASH and WHZ.

Anthropometric measurements.

The weight of children was measured to the nearest 0.1 kg using a portable electronic weighing scale (TANITA Corporation Japan, model HD-305). Recumbent length of children less than two years of age and height for children aged between 2 to 5 years was measured to the nearest 0.1 cm using a locally made height and length board. All anthropometric measurements were performed following WHO guidelines [46]. WHZ was calculated according to the WHO growth standards [1].

Data analysis.

Exploratory Factor Analysis (EFA) was used to calculate the factor loading values of the measured SES variables onto the latent SES construct. Variables with strong loading value (> 0.5) were considered for the final latent construct of SES in order to have a solid factor [47]. Variables that were used for the final SES construct were: presence of electricity in the household, possession of radio/television, ceiling fan, almirah (wardrobe), refrigerator, clock/watch, major constructing materials of a household, household food security status and monthly income. Measurement model technique was not applied for constructing the latent variable WASH as the number of variables was limited and all the variables used were included.

An expert-defined SEM was then constructed to evaluate (Fig 1): (i) whether SES has any direct statistically association with WHZ; (ii) whether WASH itself has any direct association with WHZ; and (iii) whether WASH mediates the relationship between SES and WHZ. The measurement component (i.e. the description of the latent variables) of the model included the WASH and SES constructs, whilst the structural component of the model included a direct pathway from SES to WHZ and an indirect pathway from SES to WHZ via the mediator latent variable WASH. The model also included child’s age, sex and location of residence (rural/urban) to control for their effects on WHZ and SES (excluding age given that household SES was assumed to be time invariant).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Conceptual model of causal pathways between SES, WASH and WHZ.

https://doi.org/10.1371/journal.pone.0172134.g001

Reliability of the loadings on the latent variables was assessed using Chonbach’s alpha. Model fit was assessed using standard SEM diagnostics (chi-square, Root Mean Squared Error of Approximation (RMSEA), Comparative Fit Index (CFI) and Bentler-Bonett Nonnormed Fit (NNFI)). Standardized factor loading values (coefficients) with 95% CI along with p-value are reported.

All data were analyzed using STATA 14 (StataCorp. TX. USA).

Results

Descriptive statistics

Among 10,478 children, the mean WHZ was -0.79 and the prevalence of wasting was 10.4% (95% CI: 9.84%, 11.0%). Over half (52%) of children included were male and mean age of the study children was 29.7 months. Around 92% of the study population was from the rural area and the average number of family members was 5.4. Twenty-six percent households had monthly income more than 10,000 Taka (approximately 128 USD). Day laboring was the main profession for majority of the main earners (42.6%). Nearly one fifth of mothers were involved in income generating activities and around 18% mothers had no formal schooling.

Soap was mostly used during washing clothes (81.7%), washing respondent’s own body (87.6%) and while washing own hands after defecation (55.9%). Prevalence of hand washing with soap was 1.6% before feeding a child, 4.4% before preparing food for the family and 4.9% before eating. Use of soap while washing children, while washing child’s bottom, while washing child’s hand and while washing own hand after cleaning the child were reported by 21%, 11.8%, 3.4% and 18.2% mothers/caretakers respectively. Use of a hygienic latrine was found in 15.5% households.

Half of the study population had government provided electricity in their households and 9% reported to use personal solar panels. Thirty-three percent of the households had television or radio and 83% households had mobile phone or telephone. Around 45% households had tin as construction material of external wall. Eighty-nine percent households had tin as construction material of the roof. The main construction material of the floor was sand/clay for majority of the household (80.4%). Wood (41.3%) and grass/straw (42.3%) were used as cooking fuel in most of the households. Tube well was the chief source of drinking water for majority (91%) of the families and around 84% families used safe water for drinking (tube well or rain water and covered if stored). Nearly two third (62%) of the families were found to be food secure. All other descriptive statistics are mentioned in Table 1.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Descriptive statistics.

https://doi.org/10.1371/journal.pone.0172134.t001

Measurement model and latent variables

The factor loading values for the observed SES variables were truncated at 0.5 during exploratory factor analysis. Eight variables out of 34 variables were retained for the latent variable SES: ‘household food insecurity’ and households to ‘have electricity’, ‘have radio/TV’, ‘have fan’, ‘have almirah’, ‘have refrigerator’, ‘have watch/clock’and ‘have brick and cement walls’. The factor loading values are mentioned in Table 2.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Factor loading values for SES.

https://doi.org/10.1371/journal.pone.0172134.t002

The eigenvalue for the 1^st factor was 4.78 and the proportion of variance it explained was 28%. The agreement (alpha) value of the eight variables was 0.817. All thirteen water, sanitation and hygiene variables were used to construct the latent WASH variable as we found all of them to be critical to define WASH. All standardized coefficient values of the observed variable on SES and WASH respectively and p-value are shown in Fig 2.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Structural Equation Model: Direct and Indirect Effects.

https://doi.org/10.1371/journal.pone.0172134.g002

Discussion

The relationship between SES and wasting has long been scrutinized, however the interrelationship between SES and WASH has often been ignored despite WASH representing an actionable target to reduce the prevalence of wasting. Here, we explicitly examining the interrelationship between these three variables. The SEM model showed that SES had statistically significant direct effect on WHZ, but crucially that there is a substantial indirect relationship that is mediated through WASH practices.

Our finding suggests that approximately 60% of the association between SES and the WHZ is mediated though WASH. This framework has been used to hypothesize that unimproved sanitation and failure to wash hand after defecation cause fecal contamination of home surroundings, which leads to fecal ingestion and contamination causing enteropathy [29], and ultimately results in childhood wasting. Similarly, contaminated water risks ingestion of fecal material more directly [12].

Participant’s compliance or motivation to follow WASH intervention protocols is couched in terms of their socio-economic circumstances along with their personal and communal belief, education status, perceived need, previous experience and visibility of the program’s impact [48]. Acting on WASH cannot be sustainably achieved without reference to the SES context [30], though equally projects to improve sanitation do not necessarily permeate to the level of behavioral practice in the home.

Two prior studies [17,32] found no evidence for SES or WASH impacts on wasting respectively, however neither examined the other driver or indeed their relationships. A secondary analysis of Bangladesh Demographic and Health Survey data of 2007, however found an independent association of maternal education but not socio-economic status with wasting [4]. Maternal education has been, however, a constituent part of defining SES in other studies [49] and may capture a wider set of conditions that shares variance with other SES metrics.

In a global context, evaluation of a three-year water treatment and intervention in rural Guatemala which assessed many handwashing practices that are common to our analysis along with safe water supply and hygienic latrine use, did not find any association between WASH interventions and WHZ from 929 under 5 children [25]. However, improved hand washing was found to be positively and significantly associated with WHZ during initial months but found to be diminished later on among Nepalese children [27].

Previous studies have included proxy indicators and used multiple regression to quantify the independent contribution of each variable to wasting. Such analyses fail to capture the multifaceted nature of SES and WASH domains or to address the multicollinearity of multiple indicators of the same theme. In contrast, the use of SEM deliberately represents WASH and SES as constructs of many components and allows explicit testing of their structural relationships. Additionally, regression of possible confounders such as age, sex and strata in the analysis reduced bias in the expression of the independent effect of the variables of interest.

In concordance with our hypothesis, findings from a Nepalese study [27] suggested that optimal hand washing practices can be a necessity for good health of the population living in crowded and contaminated areas but nonetheless, it cannot inhibit subclinical infections which is associated with child malnutrition. The investigators concluded that the ultimate factor adversely affecting child nutrition status is poor living conditions associated with poverty. In order for WASH interventions to sustain and have a significant impact on childhood recurrent infections, it is necessary to address the underlying poor socio-economic condition.

Conclusion

In Bangladesh, the reductions in the prevalence of wasting has stalled despite dramatic early progress. Here we show that WASH factors account for 60% of the association between SES and wasting. Improvements to SES may result in sustainable reductions in the prevalence of wasting, but these are likely to be achieved through the implementation of WASH initiatives. We expect our findings to enhance the existing evidence base for the efficacy and effectiveness of WASH interventions. Translation of our findings might contribute to policy formulation in terms of justifying the inclusion of WASH components in nutrition and livelihood improvement strategies. Moreover, research uptake by the stakeholders in nutrition may aid in designing value-for-money and effective interventions for preventing child undernutrition especially wasting in Bangladesh.

Limitations

It should be acknowledged that the data collection team for surveillance could not visit very remote hard-to-reach areas and therefore data were not collected from some hard to reach areas. Much of the information was collected through maternal responses which might have introduced recall bias. No handwashing demonstration was observed. The data analysis was constrained by the number of variables originally selected for the food security and nutrition survey. The data is country specific and therefore special consideration must be taken on extrapolating the findings to other contexts.

Supporting information

Acknowledgments

This manuscript was made possible by the committed contribution and partnership of James P Grant School of Public Health (JPGSPH), BRAC University (BU) and icddr,b. The manuscript is based on data collected and shared by FSNSP, implemented by JPGSPH, BU, Helen Keller International, Bangladesh and Bangladesh Bureau of Statistics and funded by EU. This manuscript has been reviewed by FSNSP technical reviewers for scientific content and consistency of data interpretation with previous FSNSP publications and significant comments have been incorporated prior to submission for publication. We acknowledge the following donors for providing unrestricted support to icddr,b’s effort and advancement to its strategic plan: Government of the People’s Republic of Bangladesh; Canada (Global Affairs, Canada-GAC), Sweden (Sida), and the United Kingdom (DFID).