Risk factors for anemia in children under 6 years of age in Ethiopia: analysis of the data from the cross-sectional Malaria Indicator Survey, 2007

Abstract

Background:

Malaria is a leading cause of morbidity in Ethiopia. However, its transmission varies in both space and time, and large areas of the country are hypoendemic and epidemic-prone. The Ethiopia National Malaria Indicator Survey 2007 is a cross-sectional, nationally-representative household survey. The objective of the analyses presented here were to use the survey’s data to identify factors associated with anemia presence in children under 6 years of age (U6); specifically, investigate the association between malaria and anemia; and discuss using anemia as a malaria proxy biomarker in the Ethiopian hypo-endemic transmission setting.

Methods:

The survey sampled 4185 households in 347 enumeration areas ≤2500 m above sea level. Primary outcome was increasing anemia severity in sampled children: no anemia (Hb: ≥11g/dl); mild anemia (Hb: ≥8g/dl and <11g/dl); and moderate-severe anemia (Hb: <8g/dl). Secondary outcomes were positive malaria rapid diagnostic test (RDT) or blood slide microscopy.

Results:

The analysis included 6054 (92.0%) children U6 in 3962 households. The proportion of children with no anemia, mild anemia, and moderate-severe anemia was 63.6%, 31.3%, and 5.1%, respectively. The overall prevalence of anemia (Hb <11g/dl) was 36.4% (95% CI 34.4–38.4). Factors independently associated with reduced relative odds of anemia categories were age (0R=0.7, 95% CI 0.7–0.7) and female sex (0R=0.9, 95% CI 0.8–1.0); malaria RDT positivity was associated with increased relative odds of a more severe anemia category (0R=5.8, 95% CI 3.7–9.2).

Conclusions:

We conclude that at altitudes ≤2500 m malaria appears to be a significant risk factor for anemia; potentially anemia could be used as a useful proxy biomarker for malaria and its control in Ethiopia.

Introduction

Malaria is a leading cause of morbidity and mortality in Ethiopia. In 2009–2010, malaria was the top cause of outpatient visits and admissions with 12.0% of all visits and 9.9% of admissions.¹ Climatic factors such as temperature, rainfall and humidity, partly as a function of altitude,^2–4 cause malaria transmission to vary greatly in both space and time in Ethiopia. In general, areas below 2000 m elevation are considered at risk for malaria and are targeted to receive malaria interventions, including indoor residual spraying of households with insecticide (IRS), long-lasting insecticidal nets (LLINs) and confirmatory diagnosis and anti-malarial drugs.⁵

Anemia, i.e. the condition of having fewer than the normal number of red blood cells or less than the normal quantity of hemoglobin (Hb) in the blood, is the most frequent hematological condition of infancy and childhood.⁶ In 2010, global anemia prevalence was 32.9%, causing 68.4 million disability adjusted life years.⁷ Although any reduction in Hb concentration from normal levels (i.e. >11 g/dl) decreases the oxygen-carrying capacity of the blood, anemia is characterized clinically by pallor of the skin and mucous membranes and hematologically when the Hb level falls below 7–8 g/dl.⁸ Anemia has a multifactorial etiology, primarily nutritional (e.g. lack of iron intake)⁸ and parasitogical (e.g., hookworm, schistosomiasis, and malaria).^9–11

Ethiopia undertook a Malaria Indicator Survey (MIS) in 2007 to evaluate the status of the country’s malaria prevention and control efforts.¹² The MIS showed that since 2005 Ethiopia increased IRS coverage, improved access to anti-malarial drugs, and, in particular, significantly scaled-up ownership of LLINs. In priority targeted areas ≤2000 m, prevalence of moderate–severe (i.e. Hb <8 g/dl) anemia in children under 5 years of age (U5) was shown to be high, 6.6%, while prevalence of malaria infection, as measured by microscopy, was shown to be very low, 1.0%. Prevalence of malaria by sampled enumeration area (EA) varied from 0 to 33.3%, confirming the spatial and temporal heterogeneity of malaria transmission in Ethiopia.^2,13–15 However, the survey also called into question the usefulness of tracking malaria prevalence in a cross-sectional study in a setting of comparatively low endemicity and whether tracking anemia prevalence would not be more useful.

In the current analyses we include all available relevant data from the MIS 2007, i.e., all altitudes ≤2500 m and children up to age 6 years (U6). The objectives of the analyses presented here were to identify factors associated with presence of anemia in children U6; more specifically, investigate the association between malaria and anemia; and discuss the possibility of using anemia as a proxy biomarker for malaria burden in a hypo-endemic transmission setting such as Ethiopia.

Materials and methods

Study area and survey

The MIS is a cross-sectional nationally representative survey. It was conducted in the peak malaria transmission season of 2007, from October-December, and was carried out according to Roll Back Malaria Monitoring and Evaluation Reference Group (RBM-MERG) guidelines,¹⁶ adapted to the local context. Although the main objectives of the survey were to measure the coverage, use of and access to malaria interventions as well as the prevalence of fever and malaria parasitemia,¹² various socio-demographic characteristics and anemia in children U6 were also measured.

The details of the survey have been described in full previously, including sample size calculations.¹² Briefly, the survey was of a stratified two-stage cluster sample design with census enumeration areas (EAs; comprising approximately 200 households) as primary sampling units, stratified by several domains (e.g. altitude and degree of urbanization). To fulfill sample size calculations, 347 EAs ≤2500 m were randomly selected; then, in each EA 25 households were randomly selected. Personal digital assistants (PDAs) were used for the random sampling of households and recording of the interview, malaria rapid diagnostic test (RDT) and Hb results. The questionnaires, as well as the household listing, sampling and navigation programs were integrated and installed using a Visual Basic program developed by the U.S. Centers for Disease Control and Prevention (CDC). The household questionnaire was administered to the head of the household (or if absent, to another adult) to assess socio-economic characteristics and household coverage of malaria interventions. Women aged 15–49 years living in the household completed the women’s questionnaire, which included their background characteristics, reproduction history, knowledge, attitude, and practices around malaria, and fever history in children U6. Note, all children U6 had been included in the original survey protocol to ensure that no children U5 were missed, as children U5 represent one of the target population groups for various malaria intervention indicators. Here we include biomarker results for all children U6 tested.

Anemia and malaria parasite testing

Blood samples were taken from all children U6 and from all household members in every fourth household. Anemia testing followed the recommendations of the RBM-MERG,¹⁷ with Hb concentrations measured using a portable spectrophotometer (HemoCue^®, Anglom, Sweden). The following anemia classification was used: mild (Hb: ≥8 g/dl and <11 g/d); moderate (Hb: ≥5 g/dl and <8 g/d); and severe (Hb: <5 g/dl).¹⁷ The malaria diagnostic tests included RDTs and blood slides for microscopic examination. The RDT used (ParaScreen^®, Zephyr Biomedical Systems, Verna, Goa, India) is a HRP2/pLDH-based antigen test detecting both Plasmodium falciparum and other Plasmodium spp. (in Ethiopia most likely Plasmodium vivax). Sensitivity and specificity of the test in operational conditions in Ethiopia have, depending on parasite species, been estimated to be 82.5–85.6% and 92.0–97.2%, respectively.¹⁸ Two blood slides, thick and thin films (in duplicate), were taken for each participant by a laboratory technician as per standard WHO-approved protocol,¹⁹ and processed as described before.¹² The specimen processing was organized in such a way that all three tests (RDT, anemia, slide) were performed simultaneously from a single finger prick. As per RBM-MERG MIS protocol, children with moderate anemia were given iron folate tablets;¹⁷ children with severe anemia were referred to the hospital. Similarly, children with malaria diagnosis confirmed by RDTs received anti-malarial treatment.

Data analysis

We analyzed the data of all children U6 who were eligible for hemoglobin testing. Children were assigned an ordinal outcome of increasing anemia severity based on the above-mentioned three categories of Hb levels: no anemia (Hb: ≥11 g/dl); mild anemia (Hb: ≥8 g/dl and <11 g/dl); and moderate-severe anemia (Hb: <8 g/dl). There were very few children with Hb levels <5g/l and these were included in the moderate-severe category. The explanatory variables explored included: age; sex; blood slide result; RDT result; reported use of any mosquito net or LLIN the night prior to the survey; the number of sleeping spaces in the household; household net density; reported status of residual spraying of households with insecticide in the last 12 months; terciles of household wealth index; maternal malaria knowledge mean score; altitude; and rural or urban location of clusters. Net density was calculated by dividing the number of nets in a household by the number of people in the household to account for the potential correlation between these variables. The household wealth index was derived from relevant household characteristics using principal components analysis as previously described,⁴ with terciles defining poorest, middle, and wealthiest strata. The malaria knowledge score was derived based on methods previously described by Hwang et al.²⁰ In brief, from the malaria knowledge questionnaire, a composite malaria knowledge score was calculated for each woman where every correct answer received a single point. To account for households that had more than one woman completing the malaria knowledge questionnaire (7.5% of households), a mean malaria knowledge score was generated for every household and categorized into three levels (0–1, 2–3, and ≥4).

Statistical analysis was conducted using Stata 8.2 (Stata Corporation, College Station, TX, USA). Descriptive statistics were used to examine the characteristics of the sample, and prevalence of outcomes and explanatory factors. To account for differences in the sampling design, prevalence estimates were adjusted for sampling weights. To investigate the association between ordinal categories of anemia and explanatory factors, hierarchical regression models were developed using generalized linear latent and mixed models (GLLAMM).²¹ The multilevel structure of GLLAMM allowed for non-independence of the household variables, enabled clustering of LLIN observations within households and clusters, and allowed for variability at LLIN, household and cluster levels.

Ordinal logistic regression models were fitted to study associations between anemia categories and potential risk factors.²² This model allowed for analysis of a polytomous ordinal response on a set of predictors and computed ORs of having a more severe category of anemia compared to a less severe one. Univariate analysis was conducted for each potentially explanatory factor. Multivariable models were then developed by stepwise regression analysis for model selection. This involved starting with a null model then proceeding in a sequential fashion of adding/deleting explanatory variables if they satisfied the entry/removal criterion which was set at 5% significance level using a log-likelihood ratio test. Sensitivity analyses were conducted using conventional logistic regression.

Ethical approval

The MIS protocol received ethical clearance from the Emory University Institutional Review Board [IRB#6389], the CDC Ethical Review [IRB#990132], the Program for Appropriate Technology in Health (PATH) Ethical Committee, and the Ethiopian Science and Technology Agency. Verbal informed consent was obtained from the heads of the household and women aged 15–49 years of age to participate in the household and women’s questionnaire, respectively; for blood sample collection from children U6 verbal informed consent was sought from their parent or guardian.

Results

Characteristics of the sample and prevalence of anemia

The individual and household characteristics of the sample population are summarized in Table 1. A total of 6584 children U6 in 4185 households were identified. The analysis included 6054 (92.0% of eligible) children in 3962 households. A total of 530 children in 223 households were excluded from analysis due to missing anemia, blood slide or RDT results (491) and missing household data (39). Among the children U6 included in the analysis, the mean age (standard deviation) was 2.8 (1.6) years and males comprised 50.9%. Of the households included in the analysis, a substantial proportion had a net density of <0.5 per person (77.1%), were located within an altitude of ≥1000 m to ≤2000 m (58.7%), and were in rural EAs (82.1%).

Table 1.

Socio-demographic characteristics as well as prevalence of anemia and malaria in 6054 children sampled nationally in 347 enumeration areas in Ethiopia, September–December 2007

Factors	n	%
Individual characteristics (n=6054)
Sex
Male	3081	50.9
Female	2973	49.1
Ordinal anemia categories
No anemia (Hb: ≥11 g/dl)	3850	63.6
Mild anemia (Hb: ≥8 g/dl and <11 g/dl)	1895	31.3
Moderate anemia (Hb: ≥8 g/dl and <8 g/dl)	291	4.8
Severe anemia (Hb: <5 g/dl)	18	0.3
Prevalence of anemia (Hb: <11 g/dl)	2204	36.4
Malaria results
Blood slide microscopy positivea	36	0.6
RDT positive	103	1.7
Slept under a net the previous night
Any net	2276	37.6
LLIN	2173	35.9
Household characteristics (n=3962)
Net density
<0.5	3055	77.1
≥0.5 to <1.0	709	17.9
≥1.0	198	5.0
Household sprayed with insecticide within the last 12 months	515	13.0
Wealth index terciles
Poorest	1331	33.6
Middle	1327	33.5
Richest	1303	32.9
Malaria knowledge mean scoreb
0–1 (least knowledgeable)	1260	31.8
2–3	784	19.8
≥4 (most knowledgeable)	1918	48.4
Altitude
<1000 m	345	8.7
≥1000 to ≤2000 m	2326	58.7
>2000 m	1292	32.6
Enumeration areas
Rural	3253	82.1
Urban	709	17.9

LLIN: long-lasting insecticidal net; RDT: rapid diagnostic test.

^a5676 children in 3720 households with blood slide results.

^b5731 children in 3711 households with known score.

The overall prevalence of anemia (i.e. Hb: <11 g/dl) was 36.4% (95% CI 34.6–38.4), with prevalence varying considerably between EAs (median: 26.7%; range: 0–100%); the prevalence of mild and moderate-severe anemia was 31.3% and 5.1%, respectively.

Associations between anemia and explanatory factors

Univariate ordinal logistic regression analysis of the associations of anemia categories and explanatory variables are shown on Table 2. Factors associated with reduced odds of moderate-severe anemia were increasing age (per additional year) OR=0.7; 95% CI 0.7–0.7), female sex (0R=0.9; 95% CI 0.8–1.0) and increasing maternal malaria knowledge (ptrend=0.043). Net density >1.5 per household was strongly associated with reduced odds of moderate-severe anemia (0R=0.4; 95% CI 0.1–1.3); however, this was not statistically significant (p=0.117). Increased odds of a more severe anemia category were associated with malaria positivity by blood slide (OR=6.5; 95% CI 3.2–13.1) and by RDT (OR=4.8; 95% CI 3.0–7.1) (Table 2; Figure 1). There was no association between ordinal anemia categories and individual net use the previous night, IRS, wealth index, altitude, or rural/urban areas.

Table 2.

Univariable ordinal logistic regression analysis of association between ordinal anemia categories (no anemia, mild anemia and moderate–severe anemia) and explanatory variables in 6054 children sampled nationally in 347 enumeration areas in Ethiopia, September–December 2007

Factors	Total children n=6054	No anemia (Hb: ≥11 g/dl)		Mild anemia (Hb: ≥8 g/dl and <11 g/dl)		Moderate-severe anemia (Hb: <8 g/dl)		OR	95% CI	p-value
		n	%	n	%	n	%
Individual characteristics
Age (per additional year)								0.7	0.7–0.7	<0.001
Age (years)
<1	667	306	45.9	314	47.1	47	7.0	1.0		Ptrend <0.001
1	933	469	50.3	395	42.3	69	7.4	0.7	0.6–0.9
2	1106	620	56.1	399	36.1	87	7.9	0.6	0.5–0.7
3	1079	722	66.9	308	28.5	49	4.5	0.3	0.3–0.4
4	1046	773	73.9	243	23.2	30	2.9	0.2	0.2–0.3
5	1223	958	78.3	238	19.5	27	2.2	0.2	0.1–0.2
Sex
Male	3080	1910	62.0	995	32.3	176	5.7	1.0
Female	2974	1939	65.2	901	30.3	134	4.5	0.9	0.8–1.0	0.017
Malaria positive by blood slidea
No	5640	3581	63.5	1771	31.4	288	5.1	1.0
Yes	36	7	19.4	19	52.8	10	27.8	6.5	3.2–13.1	<0.001
Malaria positive by RDT
No	5950	3814	64.1	1856	31.2	280	4.7	1.0
Yes	104	35	33.7	42	40.4	27	26.0	4.6	3.0–7.1	<0.001
Slept under a net the previous night
No	3774	2419	64.1	1155	30.6	200	5.3	1.0
Yes	2280	1430	62.7	741	32.5	109	4.8	1.0	0.9–1.1	NS
Slept under a LLIN the previous night
No	3878	2490	64.2	1183	30.5	206	5.3	1.0
Yes	2176	1358	62.4	714	32.8	104	4.8	1.0	0.9–1.2	NS
Household characteristics
Number of sleeping spaces in household (per additional space)	3962							1.0	0.9–1.0	NS
Net density
<0.5	4680	2986	63.8	1451	31.0	239	5.1	1.0
≥0.5 to <1.0	1071	665	62.1	352	32.9	54	5.0	1.0	0.9–1.3	NS
≥1.0 to <1.5	286	182	63.6	90	31.5	14	4.9	1.0	0.8–1.3	NS
≥1.5	17	13	76.5	4	23.5	0	0.0	0.4	0.1–1.3	NS
IRS within last 12 months
No	5279	3357	63.6	1652	31.3	269	5.1	1.0
Yes	775	491	63.4	244	31.5	40	5.2	1.0	0.8–1.2	NS
Wealth index terciles
Poorest	2033	1279	62.9	644	31.7	108	5.3	1.0
Middle	2039	1291	63.3	646	31.7	104	5.1	1.0	0.8–1.1	NS
Richest	1982	1276	64.4	606	30.6	97	4.9	0.9	0.8–1.1	NS
Malaria knowledge mean scoreb
0–1 (least knowledge)	1823	1158	63.5	580	31.8	88	4.8	1.0		Ptrend <0.043
2–3	1137	692	60.9	379	33.3	66	5.8	1.1	0.9–13
≥4 (most knowledge)	2771	1790	64.6	837	30.2	144	5.2	0.9	0.7–1.0
Altitude (in meters)
<1000	528	328	62.1	168	31.8	32	6.1	1.0
≥1000 to ≤2000	3551	2273	64.0	1108	31.2	170	4.8	1.0	0.8–1.2	NS
>2000	1975	1248	63.2	620	31.4	107	5.4	1.0	0.8–1.3	NS
EA characteristics
Rural	4984	3155	63.3	1565	31.4	264	5.3	1.0
Urban	1070	692	64.7	333	31.1	45	4.2	1.0	0.7–1.1	NS

EA: enumeration area; IRS: indoor residual spraying of household with insecticide; LLIN: long-lasting insecticidal net; NS: not significant; RDT: rapid diagnostic test.

^a5676 children in 3720 households with blood slide results.

^b5731 children in 3711 households with known score.

Figure 1.

Association between anemia and malaria infection as measured by microscopic examination of blood slides (BS) and rapid diagnostic tests (RDTs). Hg: hemoglobin.

Table 3 shows the multivariable associations between ordinal anemia categories and explanatory variables. Malaria results by blood slide were not included in the final model because of missing blood slide results and correlation with RDT results (rho=0.5, p<0.0001) (Figure 1). Factors independently associated with reduced relative odds of severe anemia categories were age (OR=0.7; 95% CI 0.7–0.7) and female sex (OR=0.9; 95% CI 0.8–1.0). Malaria positivity by RDT was a strong independent predictor of increased relative odds of more severe anemia category (OR=5.8; 95% CI 3.7–9.2). For all anemia cases, the sensitivity and specificity of anemia of predicting RDT positivity was 66.3% and 64.1%, respectively. Corresponding figures for moderate and severe anemia were 54.5% and 67.3% and 43.5% and 93.1%, respectively.

Table 3.

Multivariable ordinal logistic regression analysis of association between ordinal anemia categories (no anemia mild anemia and moderate–severe anemia) and explanatory variables in 6054 children sampled nationally in 347 enumeration areas in Ethiopia September–December 2007

Risk factors	OR	95% CI	p-value
Age (per additional year)	0.7	0.7–0.7	<0.001
Sex	0.9	0.8–1.0	0.038
Malaria positive by RDT	5.8	3.7–9.2	<0.001

RDT: rapid diagnostic test

Sensitivity analysis by conventional logistic regression and multinomial logistic regression revealed ORs consistent with those presented in the ordinal analysis and no new risk factors became apparent (data not shown).

Discussion

Using a large, nationally representative dataset, we show that prevalence of childhood anemia (defined as Hb <11 g/dl) in Ethiopia was 36.4%, which is similar to the figures commonly found in other large population samples in Sub-Saharan Africa,²³ and indeed Ethiopia.^14,24–26 In Ethiopia children obtain their iron from plant (non-haem) sources, mainly teff (Eragrostis tef). which is the main cereal in the Ethiopian diet and which is rich in iron and which is consumed by almost all children at least once a week. However, meat and other foods rich in ascorbic acid, which improve the absorption of non-haem iron,⁸ tend to be rarely consumed. Parasitic diseases that will cause anemia are common in Ethiopia, with a recent survey reporting 33.5%, 2.9% and 0.6% of school-aged children infected with soiltransmitted helminths, schistosomiasis and malaria.¹⁴ Other factors that have commonly been found associated with anemia, are related to environment and health services (e.g. unsafe drinking water), family caring capacity (e.g. mother being ill), food intake (e.g. specific diet), food security (e.g. family not having food reserves), socio-economic status (e.g. income below the poverty line).^24,27,28

In this study both univariate and multivariate analysis showed that anemia is associated with age and sex as well as (despite the low prevalence of infection) with malaria RDT positivity. The observed decreased prevalence of anemia with age would support the view that the association between young age and anemia could possibly be due to age-related differences of cumulative malaria exposure leading to acquisition of immunity to malaria.²⁹ Alternatively, this may be due to maternal or child related nutrition factors and breastfeeding practices or age-related maturation of the immune system favoring higher Hb concentrations at greater ages, independent of cumulative exposure of parasitic infections.³⁰ Anemia affected more male than female children U6 for as yet unclear reasons. A similar observation was made in malaria-endemic regions of India,²⁷ Malawi³¹ and Ghana;³² the sex differences in childhood anemia present a potential area for further studies.

Similar to many other studies,^33–35 malaria diagnostic test positivity was shown to be associated with anemia, confirming that even in the hypo-endemic transmission setting of Ethiopia, malaria appears to be an important contributing cause of anemia. One of the challenges in low transmission settings has been the reliance on microscopic examination of slides or RDTs for nationally representative estimates of malaria burden in cross-sectional surveys. Point prevalence surveys may, however, underestimate period prevalence as well as the spatial and temporal heterogeneity of infection in such settings; alternative approaches able to measure the long-term immune response to infection (e.g. serology)³⁶ or infections with very low parasite densities (e.g. PCR)³⁷ have yet to be fully standardized to be used in mass-screening surveys. Anemia has consistently been shown to be a major factor associated with malaria infection (and vice versa) and could be used as a proxy for malaria infection as well as an indicator of the impact of control measures. Even in this low transmission setting, malaria infection remains a risk factor for anemia.

Unlike other studies in Ethiopia,^24–26 we did not observe an association between altitude or poverty (as estimated by socio-economic strata) and anemia. Controlling or eliminating malaria in Ethiopia is likely to have the effect of concomitantly and significantly reducing anemia. Since 2007 interventions for malaria as well as other health issues (e.g. pneumonia management, neglected tropical diseases) have been scaled-up and/or sustained throughout Ethiopia, but of note is that the status for malaria and anemia indicators, as measured in nationally representative, cross-sectional surveys, has remained unchanged. Thus, in the recent MIS in 2011,³⁸ malaria prevalence was shown to be 1.3% and 4.5% by microscopy and RDTs in enumeration areas <2000 m, respectively; similarly, the proportion of children U5 with no anemia was shown to be 47.1% in the MIS 2011³⁸ and 55.8% in the Ethiopian Demographic and Health Survey 2011.³⁹

Limitations

A number of caveats have to be noted. First, the dataset used did not allow us to comprehensively assess the effects of other common causes of anemia, including dietary deficiencies and parasitic infections other than malaria; hemoglobinopathies (e.g. sickle cell disease, thalassemia) are believed to be rare in Ethiopia.⁴⁰ Second, we did not collect data on household members’ residency time, and cannot specify whether children acquired anemia when living within or outside of the place where they were sampled. This could have affected associations between anemia and household-level characteristics. However, we note that in previous studies anemia prevalence was not shown to be significantly greater in immigrants than local residents. Third, the number of regression analyses that were carried out will also have increased the odds of finding a significant association between explanatory variables and anemia. Nonetheless, both univariate analyses and multivariate analyses were consistent in identifying age, sex and positive malaria RDT as risk factors for anemia.

Conclusion

The analysis of the MIS 2007 dataset confirms that prevalence of anemia in children U6 in Ethiopia remains high. Moreover, we show that, despite the low malaria prevalence, malaria appears to be a significant explanatory variable for the levels of anemia seen, with age and sex being less strongly associated. In the absence of adequate diagnostic approaches to correctly estimate the burden of malaria in nationally representative surveys, anemia should be considered as a diagnostic marker and a proxy indicator for the success of malaria control activities in low transmission settings such as Ethiopia.