Childhood and maternal undernutrition is currently the single leading cause of the global burden of disease [1]. The fraction of total global health loss attributable to undernutrition was 9.5% in the year 2000, and 14.9% in high-mortality developing regions [1]. In Cameroon, the prevalence of childhood stunting and underweight rose from 23% to 29% and from 16% to 23%, respectively, between 1991 and 1998 [2], mirroring the trends in under-5-mortality rates, which increased from 126 per thousand to 152 per thousand between 1991 and 1998 [3, 4]. Worsening child nutritional status in Cameroon during the 1990s was in line with highest burden of malnutrition in Africa [1], but was opposite to downward trends in much of the world [5], with many countries experiencing growth and nutrition transitions [6]. Understanding the determinants of malnutrition and their trends during periods of declining health is crucial for policy design.

Poor nutritional status reflects an imbalance in dietary intake and/or infectious diseases [79], and therefore is affected by multiple environmental and socioeconomic factors such as household socioeconomic status (SES), maternal education, household hygiene, and access to case management in health services [1016]. Variation by child age in the nutritional effect of household SES and the effect of community-level infrastructure (e.g. hygiene or health service delivery) on malnutrition also have been recognized, but have been investigated in only a few studies [1719]. Studies combining individual, household and community factors in a single analytical framework are still needed to provide reliable information for policy and program design.

Cameroon is characterized by a marked socio-cultural, economic and environmental diversity likely to cause variations in health and nutrition in children. However, few studies on determinants of child nutritional status have been conducted in Cameroon, and almost all have been in a single district, town or province [2022]. Further, the variables included in these studies were all at the level of the household, without the inclusion of community characteristics. Therefore the effects of geographical variables and countrywide socioeconomic factors on child nutrition have not been evaluated. Key studies on nutrition and child health conducted at the country or regional level either identified the multiple determinants of health without quantifying their effects in a multivariate framework [23], or estimated the cross-level interactions of particular household and community factors [19], but with no reference to the role of important geographical variables like region of residence, an important predictor of child nutritional status [2] and a notable confounder of the SES-malnutrition relationship in the country. Different regions in Cameroon exhibit different levels of economic development as well as variation in climatic conditions and food production likely to affect child health independently of household or neighborhood economic status [24, 25].

In this study, we use nationally representative household surveys from 1991 and 1998 and comparable measures across years to examine the role of multiple household, community, and regional socioeconomic and environmental variables in childhood undernutrition in Cameroon. We include an analysis of how the nutritional effect of household economic status may vary by child age. By considering data from two periods, between which there were important changes in national economy, we also consider how the effects of these determinants may change in response to macroeconomic factors.


Data sources

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%).

Table 1 Regional distribution of clusters, households and children


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.

Table 2 Distribution (%) of variables

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).

Statistical analysis

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=j7aInaaDaaaleaacqWGRbWAaeaacqaI5aqocqaI4aaoaaGccqWG4baEdaWgaaWcbaGaem4AaSMaemyAaKMaemOAaOgabeaakiabcQcaQiabdMha5jabdwgaLjabdggaHjabdkhaYnaaBaaaleaacqaI5aqocqaI4aaoaeqaaaGccaGLOaGaayzkaaaaleaacqWGRbWAaeqaniabggHiLdGccqGHRaWkdaaeqbqaamaabmaabaGae8hTdq2aa0baaSqaaiabdUgaRbqaaiabiMda5iabigdaXaaakiabdQha6naaBaaaleaacqWGRbWAcqWGQbGAaeqaaOGaeiOkaOIaemyEaKNaemyzauMaemyyaeMaemOCai3aaSbaaSqaaiabiMda5iabigdaXaqabaGccqGHRaWkcqWF0oazdaqhaaWcbaGaem4AaSgabaGaeGyoaKJaeGioaGdaaOGaemOEaO3aaSbaaSqaaiabdUgaRjabdQgaQbqabaGccqGGQaGkcqWG5bqEcqWGLbqzcqWGHbqycqWGYbGCdaWgaaWcbaGaeGyoaKJaeGioaGdabeaaaOGaayjkaiaawMcaaiabgUcaRiab=X7aTnaaBaaaleaacqWGQbGAaeqaaOGaey4kaSIae8xTdu2aaSbaaSqaaiabdMgaPjabdQgaQbqabaaabaGaem4AaSgabeqdcqGHris5aOGaaCzcaiaaxMaadaqadaqaaiabigdaXaGaayjkaiaawMcaaaaa@A887@


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

β k t MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaaiiGacqWFYoGydaqhaaWcbaGaem4AaSgabaGaemiDaqhaaaaa@3151@

= 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

δ k t MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaaiiGacqWF0oazdaqhaaWcbaGaem4AaSgabaGaemiDaqhaaaaa@3155@

= 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.


Descriptive and bivariate analysis

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 [3335], in addition to a lower fertility decline in rural areas compared to urban areas during this period [36, 37].

Table 3 Nutritional status in Cameroon in 1991 and 1998

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.

Table 4 Average weight-for-age and height-for-age for selected variables

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.

Figure 1a-b
figure 1

Educational gradient in weight-for-age (WAZ) and height-for-age z-scores (HAZ) in 1991 and 1998 in Cameroon. error bars represent 95% CI around the mean.

Figure 2a-b
figure 2

Effect of household economic status on WAZ and HAZ. error bars represent 95% CI around the mean.

Figure 3a-b
figure 3

Effect of MHSB on WAZ and HAZ. error bars represent 95% CI around the mean.

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.

Figure 4a-b
figure 4

Variation in WAZ and HAZ by quality of drinking water. error bars represent 95% CI around the mean.

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.

Figure 5a-b
figure 5

Regional disparities in WAZ and HAZ. error bars represent 95% CI around the mean.

Multivariate results

Effects of socioeconomic and environmental factors

The multivariate, multilevel linear regression confirmed many of the results found in bivariate analyses, but the effects of some variables declined. Controlling for economic status and MHSB showed positive effects of these variables on WAZ (table 5, model 1) and HAZ (table 6, model 1); however, the effect of MHSB completely disappeared in 1998 after additional control for maternal education, while remaining significantly positive in 1991 (table 56, model 2). Maternal education had a positive effect in both years. Model 3 shows positive effects of water, sanitation and cooking fuel, but some of these effects diminished in favor of community environmental status after additional control for this variable (model not shown).

Table 5 Multilevel linear regression estimates of the effects of socioeconomic and environmental factors on weight-for-age z-score
Table 6 Multilevel linear regression estimates of the effects of socioeconomic and environmental factors on height-for-age z-score

Because maternal education and place of residence were used as predictors in the construction of economic status and MHSB, and because of the high correlation of economic status with community environmental status (R2 = 0.74 in each year), showing that poorer households most often live in unclean neighborhoods, we estimated the effects of these variables in separate regressions after adjusting for all other variables. The effects of economic status and MHSB were still positive (tables 5–6, model 4), but the effect of economic status diminished in 1991 in favor of water, sanitation and cooking fuel after additional adjustment for these variables (tables 5–6, model 5). Also, maternal education and community environmental status still had positive effects (tables 5–6, model 6).

The full model (model 7) did not significantly change the effects of our main exposures on WAZ and HAZ. We note that either economic status or household or community environmental variables became non-significant, showing that environmental conditions are systematic modulators of the income-nutrition relationship. We also note that in general, either maternal education or MHSB became non-significant, thus highlighting the positive correlation of education with better use of primary health care facilities, and their joint positive impact on child nutrition.

The full model also showed that children living in the largest cities were still better off than those in rural areas (this is true for WAZ in both years and for HAZ in 1998), but those in the intermediate cities were eventually worse off as compared with rural areas (this is true for WAZ). This implies that the relative advantage associated with the intermediate cities over the rural areas was entirely due to the socioeconomic composition of those milieus. Regional differentials in child nutritional status remained robust to all controls for WAZ, and the relative advantage of some southern regions over the northern region increased over time, although not significantly.

Changes in the effects of independent variables during the crisis

We estimated changes over time in the effects of independent variables (table 7). The full model including all independent variables was first used for this purpose, but to avoid possible bias due to correlation between some variables as previously discussed, model 5 was used for all variables except for maternal education, place and region of residence, and community environmental index which were estimated from model 6. However the results of this exercise were very similar. We note that the positive effect of economic status (significant for WAZ and non-significant for HAZ) increased during the crisis, while the effects of environmental variables diminished in general. Maternal education also had a non-significant increasing effect during this period. The relative advantage of girls over boys also increased over time, while the advantage of living monogamously diminished, perhaps suggesting that the crisis had a greater impact on stable families.

Age-specific effects of household economic status

We estimated variation across child age groups in the nutritional effects of household economic status (table 8). We note that this variable had little effect in children under 6 months of age, but its effect was positive on length-for-age in that age group in 1998. The effect of economic status was in general significantly greater in children above 6 months, compared to younger children. This result suggests that children under 6 months of age should be distinguished from older children in the analysis of factors associated with child nutrition outcomes.

Table 7 Change in the coefficients of selected variables between 1991 and 1998
Table 8 Age-specific effects of household economic status (ES)


We examined the household and community level socioeconomic and environmental factors associated with nutritional status among children under 3 years old in Cameroon and assessed the changes in the effects of those factors between 1991 and 1998, which was a period of severe economic crisis in the country. Real GDP per capita fell from I$ 2266 in 1990 to I$ 1949 in 1998 (1996 constant price) [38]. Average weight-for-age z-score and height-for-age z-score declined respectively from -0.70 SD to -0.83 SD (p = 0.006) and from -1.03 SD to -1.14 SD (p = 0.026) during this period. The situation experienced in Cameroon during this period was opposite to global trends in malnutrition [5, 39]. The prevalence of stunting declined in Africa, Asia, Latin America and the Caribbean during the 1990s, but remained stable in Western and Eastern Africa, as well as Central America. Downward trends in malnutrition were also noted in Indonesia despite the 1997/1998 financial crisis [12, 40]. Growth experienced by many developing countries was followed by nutrition transition, often implying a rising pattern of obesity [6, 41] or a double burden of malnutrition and obesity in some households [42].

The positive effect of maternal education and health seeking behavior on child nutritional status found in our study is consistent with other studies on factors affecting child health, such as those in India and Mali [43, 44].

Economic status had a positive effect in general, but it had little effect in children aged 0–5 months, and had significantly positive effect in older ages. It is possible that the little effect of economic status in 0–5 months is due to the role of breastfeeding, which is less frequent in high-economic status mothers than lower economic status mothers due to time budget and the ability to pay for supplementation foods. The effect of economic status in children aged 0–5 months increased between 1991 and 1998 (this is particularly true for HAZ where it was positive in 1998), however, perhaps reflecting the high sensitization toward breastfeeding during the 1990s, shifting even high-economic status mothers to frequently breastfeeding their children. Unavailability of adequate data on quality of breastfeeding did not allow us to test this hypothesis. When supplementation food becomes important (after the age of 6 months), economic status is associated with improved child nutritional status because of the relative facility of high economic status mothers to afford supplementation. The age-specific effect of economic status found in our study is consistent with Sahn et al. [17], although this study used more aggregated age groups.

The positive nutritional effect of improved water and sanitation found in this study is consistent with other studies conducted in developing countries [11, 45]. Unclean water may affect nutritional status through diarrhoeal diseases [4649]. Further, we found that cleaner fuels were associated with better anthropometric indicators, consistent with the only other available study, in South India [43]. The role of clean fuels may be mediated by effects on birth weight [50], or on the risk of respiratory infections [51, 52], which may in turn influence growth. Consistently with Fotso et al. [19], we also found that better community environmental status positively affected nutritional status after other factors were adjusted, suggesting that community hygiene affects health irrespectively of individual or household characteristics.

Our study suggests that urban-rural differentials in child nutritional status in Cameroon –especially those between the intermediate cities and rural areas–are mediated by the socioeconomic composition of those areas. The relative advantage of the largest cities over the rural areas declined after all controls, but remained positive (except for HAZ in 1991), which may be attributable to some contextual factors such as better access to and quality of health care in the main cities. This is consistent to some extent with Kuate [53] who found urban advantage in survival of children under 5 years old in Cameroon to be completely mediated by hospital delivery.

Our study also reports regional disparities in child nutritional status that were robust to all controls for WAZ, but that almost disappeared for HAZ, suggesting that much of the regional gap in HAZ is mediated by differential socioeconomic conditions. Northern Cameroon, which is a region with dry climate, limited food crops and limited access to health care, consistently had the worst WAZ. The Center/South/East region had the closest outcome to the North. Such a situation may prevail because of overwhelmingly low access to food and health care in the East province, which should be distinguished from the South and Center provinces which have better access to varied food and health services [53]. Disaggregating analysis to draw this distinction was not possible, due to non-representativeness of sample sizes at the province level.

There are some limitations in the data used for this analysis. Since this work relied on survey-based data, selection and recall bias could affect the results. Using a measured anthropometric indicator of nutritional outcome, however, eliminates one of the most important sources of bias. The measure of community environmental status used in this study is only a proxy of what the real environmental conditions and infrastructure are, as factors such as water stagnation, garbage accumulation and distance to electricity grid, which capture the quality of environment, were not available in our data. The availability of those indicators may have produced a better estimate of the effect of environmental infrastructure on child nutritional health. Other limitations of this work stem from the amount of missing data. Weight-for-age and height-for-age z-scores were calculated for only 81% and 85% of children under 3 years old in 1991 and 1998 respectively. Although the fraction of observations with missing value was relatively low, we found missing data to be correlated with some factors including household economic status and hygiene, MHSB, child age, and place and region of residence, a potential source of bias.


The age-specific effect of economic status on child nutritional status found in this study suggests that children aged 0–5 months should be distinguished from older children in the analysis of factors associated with child nutrition outcomes, with an attention to the possible role of breastfeeding in early ages, as it has been proven to have great benefits to child health and survival [5456]. Socioeconomic and environmental effects may vary across age groups, motivating age-specific analysis of factors affecting health. This study shows that child nutritional health is simultaneously influenced by household and community factors, and can be addressed to some extent, but not entirely, by interventions at either level. Low nutritional status is more likely in children of low socioeconomic status, and children at risk are clustered in dry regions and within community with poor environmental status. Inclusion of household and community factors is needed in addressing the issue of child health in Cameroon, and will assist in designing more effective context-specific intervention programs.