Decomposition of factors associated with housing material inequality in under-five deaths in low and middle-income countries

Background: Low-and Medium-Income Countries (LMIC) continue to record a high burden of under-five deaths (U5D). There is a gap in knowledge of the factors contributing to housing materials inequalities in U5D. This study examined the contributions of the individual- and neighbourhood-level factors to housing materials inequalities in influencing U5D in LMIC. Methods: We pooled data from the most recent Demographic and Health Surveys for 56 LMIC conducted between 2010 and 2018. In all, we analysed the data of 798,796 children living in 59,791 neighbourhoods. The outcome variable was U5D among live births within 0 to 59 months of birth. The main determinate variable was housing material types, categorised as unimproved housing materials (UHM) and improved housing materials (IHM) while the individual-level and neighbourhood-level factors are the independent variables. Data were analysed using the Fairlie decomposition analysis at α = 0.05. Results: The overall U5D rate was 53 per 1000 children, 61 among children from houses built with UHM, and 41 among children from houses built with IHM (p < 0.001). This rate was higher among children from houses that were built with UHM in all countries except Malawi, Zambia, Lesotho, Gambia, Liberia, Sierra Leone, Indonesia, Maldives, Jordan, and Albania. None of these countries had significant pro-IHM inequality. The factors explaining housing inequalities in U5D include household wealth status, residence location, source of drinking water, media access, paternal employment, birth interval, and toilet type. Conclusions: There are variations in individual- and neighbourhood-level factors driving housing materials inequalities as it influences U5D in LMIC. Interventions focusing on reducing the burden of U5D in households built with UHM are urgently needed. Data for this cross-sectional study were obtained from ICF, the primary owner and implementer of Demographic and Health Surveys (DHS) across countries worldwide. These surveys are nationally representative household surveys conducted in LMIC. This study used data from 56 recent DHS conducted between 2010 and 2018 and available in the public domain as of 15th of September 2020 when the data were harvested from dhsprogram.com. The DHS uses a multi-stage, clustered and stratified sampling design with households as the sampling frame using an individual country’s most recent census as a sampling frame. Typically, the upper stages include states or districts, or regions depending on country-specific administrative nomenclatures. The last stage of the multi-stage sampling is the selection of clusters from which households are selected. The clusters are referred to as the primary sampling units (PSUs) [13, 14]. All eligible women (aged 15 to 49 years) and men (aged 15–65 years) within each sampled household were interviewed. The surveys were not self-weighting. Therefore, sampling weights were calculated for all participants to account for unequal selection probabilities as well as for non-response. The application of the sampling weights ensures that survey findings represent the full target population. The DHS data includes household data, women’s data, birth recode data and children recode data and men’s data. However, the present study made use of the children recode data. Further details of the sampling methodologies are available at dhsprogram.com. In all, we pooled the data of 798,796 children, living in 59,791 neighbourhoods nested within 56 countries. The outcome variable in this study is under-five death (U5D). Under-5 death is a death among live births within the first five years of life. It is defined as all deaths within 0 to 59 months of birth [13, 15]. To ensure the completeness and correctness of this variable, mothers were asked to name all their livebirths within five years preceding the date of the study. They were then asked if each of those children were alive or dead. The dates of death, the ages at death for the dead children and survival status were then used to determine U5D. Under-5 death was coded as a dichotomous variable: Alive or Died before the 5th birthday. The main determinate variable is the quality of the materials used to build houses where children under 5 years of age live. It was derived from the three sub-variables. These are the materials used for the house floor, wall and roof. During the survey, the enumerators personally observed these housing materials. The DHS provided a guideline for the classification of materials used for any of these three parts of the house as either improved or not. The classifications are (i) the improved floor materials are cement, ceramic tiles, vinyl asphalt strips, parquet and polished wood while the unimproved floor materials are earth, sand, dung, rudimentary, wood planks, palm, bamboo, and others; (ii) the improved wall materials include cement, stone with lime/cement, cement blocks and bricks while the unimproved wall materials are no wall, cane/palm/trunks, dirt, rudimentary, bamboo with mud, stone with mud, uncovered adobe, plywood, and others (iii) the improved roof materials are cement and roofing shingles while the unimproved materials consist of no roof, thatch/palm leaf, sod, rudimentary, rustic mat, palm/bamboo, wood planks, cardboard, wood, and others [15–18]. Each of the improved floor, wall and roof materials were scored “1” while unimproved materials were scored “0”. We then summed these scores. Houses with > = 2 scores out of the maximum obtainable 3 scores were classified as houses built with improved housing materials (IHM) while houses with < 2 scores were categorized as houses built with unimproved housing materials (UHM). The independent variables consist of individual-level and neighbourhood-level factors identified in the literature to be associated with childhood deaths. The children characteristics, mothers’ characteristics and the households’ characteristics constitute the individual-level factors. The children characteristics are sex (male, female), weight at birth (average+, small and very small), birth interval (firstborn, < 36 months and > =36 months) and birth order (1, 2, 3 and 4+), a child is a twin (single, multiple (2+). The maternal characteristics: maternal education (none, primary or secondary plus), maternal age [15, 19–23], marital status (never, currently and formerly married), maternal and paternal employment status (working or not working), health insurance (yes /no). The household characteristics include the sex of the head of the household (male or female), access to media (at least one of radio, television, or newspaper), sources of drinking water (improved or unimproved), toilet type (improved or unimproved), cooking fuel (clean fuel or biomass), housing materials (improved or unimproved) and household wealth index (poorest, poorer, middle, richer and richest), place of residence (rural or urban). Neighbourhood was operationalized as the clustering of children. The DHS uses “clusters” as the PSU. People of the same cluster are very likely to share similar contextual factors [13, 14]. We regard children as “neighbours” if they belong to the same cluster. In this study, we computed neighbourhood socioeconomic status (SES) as a neighbourhood-level from the proportion of mothers within the same clusters without education, belonging to a household in the two lowest wealth quintiles, has no media access and unemployed using the principal component factor method. The analytical approach for this study included descriptive statistics, bivariable analysis and multivariable decomposition analysis. Descriptive statistics to show the distribution of the children’s background characteristics as well as the distribution of U5D among the children from houses with IHM and UHM by countries and characteristics. The bivariable analysis was conducted using the Z-test to determine the equality of proportions of U5D among the children from houses with IHM and UHM within each country and region (Table 1). Charts were used for visualization. The spatial distribution of under-five deaths per 1000 livebirths among children in houses with improved and unimproved housing materials are shown in Fig. ​Fig.1.1. The maps were built in Microsoft Projects 2020. Distribution of sample characteristics by countries, regions and prevalence of under-five deaths in LMIC by the quality of housing material, 2010–2018 *significant at 5% test of equality of proportions We calculated the risk differences (RD) in U5D among the children from houses with IHM and UHM. A risk difference greater than 0 suggests that U5D are higher among the children from houses built with UHM than those from IHM (pro-unimproved housing material). Conversely, a negative RD indicates under-5 deaths are higher among the children from houses with IHM than those from UHM (pro-improved housing material). We carried out a country-level meta-analysis of the prevalence of U5D in each of the countries by computing the risk difference in the development of U5D between U5C from houses with improved and unimproved housing materials and presented the results in Fig. 2. A random-effects meta-analysis was used on the assumption that each country is estimating a study-specific true effect. We implemented the meta-analysis in R software by specifying the summary measure (SM) as risk difference (RD), the number of deaths in houses with improved and unimproved housing materials as well as the numbers of participants for each country, grouped by regions using the “metabin” command in R. We built a 95% confidence interval (CI) around the RDs to determine their significance. a Spatial distribution of under-five deaths among children in houses with unimproved housing materials in the LMIC studied. 2b Spatial distribution of under-five deaths among children in houses with improved housing materials in the LMIC studied The Mantel-Haenszel (MH) Odds Ratio (OR) and tests of heterogeneity of ORs were conducted to ascertain that the countries are different with regards to the odds ratio of U5D among children from houses with IHM and UHM and a test of homogeneity of ORs among all the countries with a significant odds ratio of U5D to determine if the odds of having U5D in those countries are homogenous. Finally, the Fairlie decomposition analysis (FDA) techniques using logistic models was applied. Sampling weights were applied in all the analyses in this study to adjust for unequal cluster sizes, stratifications and to ensure that our findings adequately represent the target population. Multicollinearity among the independent variables was tested using the “colin” command in Stata version 16. The command provided the variance inflation factor (VIF). The VIF is approximate of the 1/(1-R2) ranging from 1 to infinity. The R2-value is obtained by regressing tjth independent variable on other independent variables. All variables with VIF > 2.5 were removed from the regression analysis. Literature has shown concerns about VIF > 2.5 [24]. The FDA technique is an offshoot of the well-known Blinder-Oaxaca decomposition analysis technique that was originally developed for linear models [25–27]. FDA was developed following the inefficiency of the Blinder-Oaxaca decomposition analysis technique in handling non-linear outcomes such as logit or probit models [19, 20, 28–30]. The FDA was developed for non-linear regression models and used in the quantification of the contributions to differences in the prediction of an outcome of interest between two groups [31]. This technique is a counterfactual method with an assumption that “what the probability of under-5 death would be if children from houses built with UHM had the same characteristics as the children whose houses are built with IHM?” The FDA allows for the decomposition of the difference in an outcome variable between 2 groups (children from houses with IHM and UHM) into 2 components. The first component is the “explained” (also referred to as the “compositional” or “endowments”) portion of that gap that captures differences in the distributions of the measurable characteristics. The explained part is the portion of the gap in U5D attributable to the differences in observable, measurable characteristics between children from houses with IHM and UHM. This method helps to quantify how much of the gap between the children from houses with UHM and the children from houses with IHM is attributable to these differences in specific measurable characteristics. The second component of the model is the “unexplained” (also referred to as the “structural” component or the “coefficient”) part. The unexplained part is the portion of the gap due to the differences in the estimated regression coefficients and the unmeasured variables between the two groups. The Fairlie decomposition technique works by constraining the predicted probability between 0 and 1 as available in a logit model. The coefficients (β) estimated by the logit regression technique with the probability of under-5 deaths conditioned on the independent variables (X) is obtained as We carried out an FDA analysis by calculating the difference between the predicted probability for Group A (children from houses with UHM) using the Group B (children from houses with UHM) regression coefficients and the predicted probability for under-5 deaths among Group B using its regression coefficients [19]. Fairlie et al. showed that the decomposition for a nonlinear equation Y = F(X), can be expressed as: Where NA is the sample size for group J [32]. In equation (1), Y¯ is not necessarily the same as FX¯β^, unlike in BODA where F(Xiβ) = Xiβ. The 1st term (explained) is the part of the gap in the binary outcome variable that is due to group differences in distributions of X, and the 2nd term (unexplained) is the part due to differences in the group processes determining levels of Y (under-5 deaths). The 2nd term also captures the portion of the binary outcome variable gap due to group differences in unmeasurable or unobserved endowments. The estimation of the total contribution is the difference between the average values of the predicted probabilities. Using coefficient estimates from a logit regression model for a pooled sample, β^∗, the independent contribution of X1 and X2 to the group, the gap can be written as and respectively. The contribution of each variable to the gap is thus equal to the change in the average predicted probability from replacing the group B distribution with the group A distribution of that variable while holding other variables constant. Other detailed numerical of this approach have been reported in the literature [19, 20, 30, 32, 33]. We implemented the FDA in STATA 16 (StataCorp, College Station, Texas, United States of America) using the “Fairlie” command.