Household and community socioeconomic and environmental determinants of child nutritional status in Cameroon

Demographic and Health Surveys (DHS) [26] were conducted in Cameroon in 1991 and 1998, designed to be representative at the national, urban-rural and regional level. A two-stage probabilistic sampling technique was used to select clusters at the first level and households at the second level (table 1). The household response rate was 83% in 1991 and 94% in 1998. The survey included a household module, as well as a questionnaire administered to women aged 15 to 49 years, comprising a birth history, information on individual characteristics and health behaviors, and details on their children.

For children alive at each survey (those under age 5 years in 1991 and under age 3 years in 1998), weight and height were measured and used to calculate anthropometric indicators. For the purpose of cross-year comparability, we restricted our study to children under 3 in both surveys. Of the 1966 children born after 1988 and surviving to 1991, anthropometric data were available for 1587 (80.7% of the weighted sample), and of the 2260 children born after 1995 and surviving to 1998, anthropometric data were reported for 1923 (85%).

Child nutritional status is measured by weight-for-age z-scores (WAZ) and height-for-age z-scores (HAZ) using the United States National Center for Health Statistics/WHO international reference population. WAZ has been used in many epidemiological studies of undernutrition and child mortality, including in the latest systematic review and meta-analysis [27], and is suitable for the analysis of multiple determinants of child health, including socioeconomic determinants [12]. HAZ is an indicator for linear growth and reflects cumulated and chronic child health conditions. At the individual and household level (referred to as level 1), independent variables included mother's characteristics (education, health seeking behavior, age at child birth, and marital status), household variables (economic status, source of drinking water, sanitation and cooking fuel), and child characteristics (age, sex, size at birth, breastfeeding status and preceding birth interval). Individual and household variables were considered at the same level because there was less than one child per household in our data (table 1). Details on the definitions and distributions of these variables appear in table 2.

Household economic status and maternal health-seeking behavior (MHSB) are index variables constructed based on the statistical model developed by Ferguson et al. [28]. The model was designed to measure economic status based on possession of household consumer durables such as electricity, television, bicycle and car. The basic premise of the model is that wealthier households are more likely to own any given set of assets; and that some assets are likely to be consumed at relatively low levels of economic status (e.g. radio or bicycle), while others will be owned only at higher levels (e.g. television or car). The model postulates a continuous level of economic status (unobserved) predicted by a series of socio-demographic covariates (age and sex of the head of the household, mother's education and occupation, and urban or rural place of residence), with observed ownership of each asset captured in a set of indicator variables. The inclusion of certain assets presumed to be owned at roughly the same level on an internationally comparable economic status scale allows comparisons across countries or over time. A similar approach was used to estimate levels of MHSB, with predictive covariates including mother's education and occupation, and place of residence, and dichotomous indicator variables for prenatal visit, tetanus injection during pregnancy, medical assistance at delivery, knowledge of oral rehydration solutions (ORS) and possession of a health card for the child.

Increasing awareness of the effects of community or neighborhood on health beyond individual and household-level influences has produced a vast literature [29, 30]. In developing countries, community-level factors that may influence health include community economic development, climatic conditions (these two factors are often captured by region of residence) and environmental hygiene. Children residing in clean neighborhoods may have better health than similar children living in unclean neighborhoods. In our study, community statistical units were the DHS clusters sample for the years 1991 and 1998, and community-level independent variables (referred to as level 2 variables) included place of residence, region of residence and environmental status (table 2). Community environmental status is a continuous variable built using pooled 1991 and 1998 datasets and principal components analysis on 5 variables including the proportion of children in households with access to water, sanitation, electricity, using electric stove or gas as cooking fuel, or having finished floor in each cluster. This variable therefore represents access to clean environmental conditions at the community level, affected by both household resources and the broader community-level infrastructure (e.g. waste disposal infrastructure and distance to water source or electricity grid).

Analysis of variance (ANOVA) was used to assess the bivariate association between average WAZ and HAZ and selected independent variables, including child sex and age, maternal education, MHSB, economic status (the latter two recoded into 5-level variables using the quintile cut-off points of the pooled 1991 and 1998 data), water, sanitation, cooking fuel, and place and region of residence. A two tailed t-test was also performed to test the statistical significance of the change in the average WAZ and HAZ between 1991 and 1998 (table 4).

In multivariate analysis, we pooled the 1991 and 1998 data and used a two-level random intercept model to estimate the specific effect of household and community independent variables on WAZ and HAZ, using the Stata 9 statistical software. This model takes into account the hierarchical sample selection design characterizing the DHS surveys, and further adjusts for spatial correlation and heteroskedasticity, as the nutritional status of children living in the same neighborhoods (hereafter 'cluster') may be correlated due to common neighborhoods influences (e.g. access to water, electricity, etc.) This model also allows estimation of the fraction of variance of the dependent variable occurring at each level of the analysis [31, 32], and is specified (for WAZ) as follows.

WAZ i j = β 0 + β 1 y e a r 98 + ∑ k ( β k 91 x k i j * y e a r 91 + β k 98 x k i j * y e a r 98 ) + ∑ k ( δ k 91 z k j * y e a r 91 + δ k 98 z k j * y e a r 98 ) + μ j + ε i j ( 1 ) MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqqGxbWvcqqGbbqqcqqGAbGwdaWgaaWcbaGaemyAaKMaemOAaOgabeaakiabg2da9GGaciab=j7aInaaBaaaleaacqaIWaamaeqaaOGaey4kaSIae8NSdi2aaSbaaSqaaiabigdaXaqabaGccqWG5bqEcqWGLbqzcqWGHbqycqWGYbGCdaWgaaWcbaGaeGyoaKJaeGioaGdabeaakiabgUcaRmaaqafabaWaaeWaaeaacqWFYoGydaqhaaWcbaGaem4AaSgabaGaeGyoaKJaeGymaedaaOGaemiEaG3aaSbaaSqaaiabdUgaRjabdMgaPjabdQgaQbqabaGccqGGQaGkcqWG5bqEcqWGLbqzcqWGHbqycqWGYbGCdaWgaaWcbaGaeGyoaKJaeGymaedabeaakiabgUcaRiab=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X7aTnaaBaaaleaacqWGQbGAaeqaaOGaey4kaSIae8xTdu2aaSbaaSqaaiabdMgaPjabdQgaQbqabaaabaGaem4AaSgabeqdcqGHris5aOGaaCzcaiaaxMaadaqadaqaaiabigdaXaGaayjkaiaawMcaaaaa@A887@

where

WAZ _ij = weight-for-age z-score for a child i in cluster j

β ₀ = intercept

year _t = dummy indicator for the year t (t = 1991, 1998)

β ₁ = coefficient on the year 1998

x _kij = value of variable x _k for a child i in cluster j (x _k is an individual or household level variable)

x _kij * year _t = interaction term between variable x _k and year t (t = 1991, 1998) evaluated for a child i in cluster j

= coefficient on variable x _k in period t (t = 1991, 1998), representing increase in WAZ due to a unit increase in x _k if x _k is continuous, or the differential effect of x _k on WAZ relative to a reference category if x _k is a dummy indicator representing a category of a categorical variable

z _kj = value of variable z _k for a cluster j (z _k is a community level variable)

z _kj * year _t = interaction term between variable z _k and year t (t = 1991, 1998) evaluated for a cluster j

= coefficient on variable z _k in period t (t = 1991, 1998), interpreted similarly as β k t MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaaiiGacqWFYoGydaqhaaWcbaGaem4AaSgabaGaemiDaqhaaaaa@3151@

μ _j = clusters residuals assumed to be independent and normally distributed

ε _ij = within-cluster individuals residuals assumed to be independent and normally distributed.

The residual terms μ _j and ε _ij are assumed to have zero mean, with respective variances σ μ 2 MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaaiiGacqWFdpWCdaqhaaWcbaGae8hVd0gabaGaeGOmaidaaaaa@3146@ and σ ε 2 MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaaiiGacqWFdpWCdaqhaaWcbaGae8xTdugabaGaeGOmaidaaaaa@3137@, which are the variance of WAZ occurring at the cluster and individual level, respectively, after netting out the effects of independent variables. Only coefficients β k 91 MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaaiiGacqWFYoGydaqhaaWcbaGaem4AaSgabaGaeGyoaKJaeGymaedaaaaa@31D0@ and β k 98 MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaaiiGacqWFYoGydaqhaaWcbaGaem4AaSgabaGaeGyoaKJaeGioaGdaaaaa@31DE@ estimating simultaneously the effects of these independent variables for the years 1991 and 1998, respectively, were reported in addition to the individual and cluster level variances σ ε 2 MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaaiiGacqWFdpWCdaqhaaWcbaGae8xTdugabaGaeGOmaidaaaaa@3137@ and σ μ 2 MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaaiiGacqWFdpWCdaqhaaWcbaGae8hVd0gabaGaeGOmaidaaaaa@3146@ (table 5).

To estimate changes across years in the effects of independent variables, we used the following equation:

WAZ i j = α 0 + α 1 y e a r 98 + ∑ k ( α k 98 − 91 x k i j * y e a r 98 + α k 91 x k i j ) + ∑ k ( ϕ k 98 − 91 z k j * y e a r 98 + ϕ k 91 z k j ) + μ j + ε i j ( 2 ) MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqqGxbWvcqqGbbqqcqqGAbGwdaWgaaWcbaGaemyAaKMaemOAaOgabeaakiabg2da9GGaciab=f7aHnaaBaaaleaacqaIWaamaeqaaOGaey4kaSIae8xSde2aaSbaaSqaaiabigdaXaqabaGccqWG5bqEcqWGLbqzcqWGHbqycqWGYbGCdaWgaaWcbaGaeGyoaKJaeGioaGdabeaakiabgUcaRmaaqafabaWaaeWaaeaacqWFXoqydaqhaaWcbaGaem4AaSgabaGaeGyoaKJaeGioaGJaeyOeI0IaeGyoaKJaeGymaedaaOGaemiEaG3aaSbaaSqaaiabdUgaRjabdMgaPjabdQgaQbqabaGccqGGQaGkcqWG5bqEcqWGLbqzcqWGHbqycqWGYbGCdaWgaaWcbaGaeGyoaKJaeGioaGdabeaakiabgUcaRiab=f7aHnaaDaaaleaacqWGRbWAaeaacqaI5aqocqaIXaqmaaGccqWG4baEdaWgaaWcbaGaem4AaSMaemyAaKMaemOAaOgabeaaaOGaayjkaiaawMcaaaWcbaGaem4AaSgabeqdcqGHris5aOGaey4kaSYaaabuaeaadaqadaqaaiab=v9aQnaaDaaaleaacqWGRbWAaeaacqaI5aqocqaI4aaocqGHsislcqaI5aqocqaIXaqmaaGccqWG6bGEdaWgaaWcbaGaem4AaSMaemOAaOgabeaakiabcQcaQiabdMha5jabdwgaLjabdggaHjabdkhaYnaaBaaaleaacqaI5aqocqaI4aaoaeqaaOGaey4kaSIae8x1dO2aa0baaSqaaiabdUgaRbqaaiabiMda5iabigdaXaaakiabdQha6naaBaaaleaacqWGRbWAcqWGQbGAaeqaaaGccaGLOaGaayzkaaGaey4kaSIae8hVd02aaSbaaSqaaiabdQgaQbqabaGccqGHRaWkcqWF1oqzdaWgaaWcbaGaemyAaKMaemOAaOgabeaaaeaacqWGRbWAaeqaniabggHiLdGccaWLjaGaaCzcamaabmaabaGaeGOmaidacaGLOaGaayzkaaaaaa@9D79@

Note that equation (2) is algebraically equivalent to equation (1) in that α k 91 MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaaiiGacqWFXoqydaqhaaWcbaGaem4AaSgabaGaeGyoaKJaeGymaedaaaaa@31CE@ = β k 91 MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaaiiGacqWFYoGydaqhaaWcbaGaem4AaSgabaGaeGyoaKJaeGymaedaaaaa@31D0@, ϕ k 91 = δ k 91 , α k 98 − 91 = β k 98 − β k 91 MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaaiiGacqWFvpGAdaqhaaWcbaGaem4AaSgabaGaeGyoaKJaeGymaedaaOGaeyypa0Jae8hTdq2aa0baaSqaaiabdUgaRbqaaiabiMda5iabigdaXaaakiabcYcaSiab=f7aHnaaDaaaleaacqWGRbWAaeaacqaI5aqocqaI4aaocqGHsislcqaI5aqocqaIXaqmaaGccqGH9aqpcqWFYoGydaqhaaWcbaGaem4AaSgabaGaeGyoaKJaeGioaGdaaOGaeyOeI0Iae8NSdi2aa0baaSqaaiabdUgaRbqaaiabiMda5iabigdaXaaaaaa@4D57@ and ϕ k 98 − 91 = δ k 98 − δ k 91 MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaaiiGacqWFvpGAdaqhaaWcbaGaem4AaSgabaGaeGyoaKJaeGioaGJaeyOeI0IaeGyoaKJaeGymaedaaOGaeyypa0Jae8hTdq2aa0baaSqaaiabdUgaRbqaaiabiMda5iabiIda4aaakiabgkHiTiab=r7aKnaaDaaaleaacqWGRbWAaeaacqaI5aqocqaIXaqmaaaaaa@4133@. Therefore only coefficients and ϕ k 98 − 91 MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaaiiGacqWFvpGAdaqhaaWcbaGaem4AaSgabaGaeGyoaKJaeGioaGJaeyOeI0IaeGyoaKJaeGymaedaaaaa@34E6@ estimating respectively the cross-year changes in the effects of individual/household and community independent variables were reported (table 7). Equation (1) and equation (2) were also estimated for height-for-age z-score by replacing WAZ by HAZ (tables 6–7).

Summarizing what precedes, while equation (1) estimates the effects of independent variables for the years 1991 and 1998 simultaneously, equation (2) estimates changes in the effects of these variables between the two years.

The testing strategy used in the multivariate analysis estimated first the effects of some socioeconomic factors (tables 5–6, models 1–2) and environmental factors (tables 5–6, model 3) on nutritional status in separate equations; because maternal education and place of residence were used as predictors in the construction of economic status and MHSB, the adjusted effects of these variables were estimated separately (tables 5–6, models 4–6). We also estimated the full model including all independent variables (tables 5–6, model 7). Finally, we estimated the age-specific effects of household economic status to check variations in these effects by child age (table 8). Multilevel analyses were not weighted.

Average WAZ and HAZ in children younger than 3 years old in Cameroon declined respectively from -0.70 standard deviations (SD), i.e. 0.70 SD below the reference median value, to -0.83 SD (p = 0.006), and from -1.03 SD to -1.14 SD (p = 0.026) between 1991 and 1998. During this period, the prevalence of underweight (defined as WAZ<-2 SD) and stunting (% HAZ<-2 SD) increased respectively from 16% to 22% (P < 0.0001), and from 23% to 29% (p < 0.0001). Average weight-for-height z-score (WHZ) also deteriorated, mirroring increase in the prevalence of wasting (%WHZ<-2 SD) (table 3). Because of a shift of the sample toward rural children during this period (see tables 1 and 2), trends in nutritional indicators were adjusted for place of residence and child sex and age, and we still found evidence of a decline in nutritional status. It should be noted that this sample shift might reflect a massive urban-to-rural migration flow during the 1990s economic crisis [33‐35], in addition to a lower fertility decline in rural areas compared to urban areas during this period [36, 37].

Results for bivariate analyses are presented in table 4. The decline in nutritional status occurred mostly in boys, children aged 12–23 months, those born to uneducated mothers, and those of low economic status. Child sex was not significantly associated with nutritional status in 1991, but girls had higher average WAZ and HAZ compared to boys in 1998 due to uneven declines. Child age was also a significant predictor of nutritional status.

In both years, nutritional status improved with maternal education (table 4, figures 1a-b). Further, the advantage associated with education increased between 1991 and 1998, as decline in WAZ and HAZ was concentrated in children of uneducated mothers, although this advantage was not statistically significant when clustering of observations was taken into account (see unadjusted model in table 7). A positive economic status gradient was also noted in 1991 and 1998 (table 4, figures 2a-b). Moreover, the gap between the richest and the poorest economic groups increased during this period, although not significantly. Maternal health seeking behavior also had a significantly positive effect on nutritional status in both years (table 4, figures 3a-b). Further, the difference in HAZ between children born to mothers with the highest MHSB and those born to mothers with the lowest MHSB increased during the crisis, but not significantly.

Improved water and cleaner cooking fuels were also associated with better nutritional status in 1991 and 1998 (table 4; also see figures 4a-b for water). Similar results were found for sanitation. Children with flush toilets in their houses had lower nutritional status than those with improved pit latrines, but this anomaly was not statistically significant.

At the community level, child nutritional status was higher in urban areas in each year (table 4). Average WAZ was -0.16 SD in the largest cities Yaounde and Douala and -0.82 SD in rural areas in 1991 (p < 0.001); it was respectively -0.31 SD and -0.91 SD in 1998 (p < 0.001). Average HAZ was -0.57 SD in Yaounde/Douala and -1.15 SD in rural areas in 1991 (p < 0.001), and was respectively -0.75 SD and -1.22 SD in 1998 (p < 0.001). In both years, the advantage of children in intermediate cities versus rural areas was not statistically significant for either nutritional indicator. Significant regional variation in child nutrition was also observed in the country (table 4; also see figures 5a-b). Children living in the West or Littoral province had the highest nutritional status after those living in the largest cities; in contrast, children in northern Cameroon had the lowest average WAZ and HAZ.