Study design
The current study used secondary data from the 17 demographic and health surveys (DHSs) conducted in SSA. These surveys are comparable and representative (at the national level, residence level, and regional level) cluster-sample household surveys that have been conducted in more than 90 countries globally since 1984 [32]. Details about these surveys can be accessed elsewhere [33].
Data sources
We pooled 17 DHSs conducted between 2014 and 2018 in SSA. Specifically, the following countries were included: Benin (2018), Burundi (2018), Cameroon (2018), Ethiopia (2016), Ghana (2017), Guinea (2018), Lesotho (2014), Mali (2018), Malawi (2016), Nigeria (2018), Rwanda (2015), Sierra Leone (2019), South Africa (2016), Tanzania (2016), Uganda (2016), Zambia (2018), and Zimbabwe (2015) [33]. Figure 1 shows the selected SSA countries included in this study. Generally, these countries were selected and included because they had information on both DM and Hb estimates.
Data collection
In the DHS, information was collected from women aged 15–49 years with under-5 children prior to the survey using the Woman’s Questionnaire. Data on immunization coverage, DM, iron supplementation, vitamin A supplementation, anthropometric measurements and nutritional indicators, recent occurrences of diarrhoea, fever, and cough for young children, and treatment of childhood diseases were collected.
Inclusion and exclusion
We focused on pre-SAC because the DHS collects data on that age group and not on school-age children (SAC). We limited our analyses to live children aged 6–59 months as per WHO recommendation. Furthermore, children whose households were not selected for height and weight measurements or who had missing data on the other variables were excluded. This study included children whose caregivers had been interviewed and had provided consent. After applying the inclusion and exclusion criteria, which included age restriction, 50,075 children under 5 years of age were included in the analysis (Fig. 2).
Anaemia testing
In each country, anaemia testing was performed in the subsample of selected households. Blood samples for anaemia testing were collected from all children aged 6–59 months for whom consent was obtained from their parents or the adult responsible for the children. Blood samples were drawn from a drop of blood taken from a finger prick (or a heel prick from children aged 6–11 months) and collected in a microcuvette. Analysis of Hb was conducted on-site with a battery-operated portable HemoCue analyser.
Anthropometric measurements
All the DHS surveys included in this study, height, and weight were collected from children under the age of 5 years who resided in the household the previous night prior to the data collection [34]. Weight measurements of the children were collected using electronic SECA 878 flat scales while ShorrBoard® measuring boards were used to record the height measurements. For children less than 24 months old, recumbent height/length measurements were taken, while standing height/length measurements were recorded for children aged 24 months or older [35]. Information on age, height, and weight was used to calculate several childhood nutritional indices (height/length for age, weight for height/length, and weight for age).
Operationalization of the study variables
Outcome variable
The principal outcome variable of this study was anaemia (as measured by Hb). WHO recommendations were used to define anaemia in children aged 6–59 months. Children were considered to be anaemic if their Hb concentration was less than 11.0 g/dl, adjusting for altitude [36].
Main explanatory variable
The principal explanatory variable of the current study was DM. Caregivers were asked, “Was (NAME) given any drug for intestinal worms (deworming chemotherapy) in the last 6 months?” The answer to this question was categorized as “yes” if the child had received DM and “no” if the child had not received DM.
Potential confounders
Based on insights from relevant literature [37], the following characteristics were treated as potential confounders: child-level factors included sex of the child (male/female), age of the child in months (6–11, 12–23, 24–35, 36–47, and 48–59), presumed fever in the last 2 weeks (no/yes), and an episode of diarrhoea in the last 2 weeks (no/yes). An episode of diarrhoea was defined as the passage of three or more loose or liquid stools in 24 h [38]. We also considered variables that measure the nutritional status of the child, including stunting status (not stunted/stunted) and wasting status (not wasted/wasted). In addition, we considered factors at the maternal and household levels, and these characteristics included the age of the respondents in years (< 25, 25–34, ≥ 35), maternal anaemia (no/yes), education level of the respondents (no formal education, primary education, and secondary and above education), household wealth (poorest, poorer, middle, richer, and richest), amount of media exposure (0, 1, 2, and 3), type of drinking water sources (unimproved/improved), type of household sanitation facility (unimproved/improved), place of residence (urban/rural), and year of the data collection (2014, 2015, 2016, 2017, 2018, and 2019). Childhood stunting and wasting were defined as moderate and severe—that is, below minus two standard deviations (< −2SD) from median height-for-age and weight-for-age z-scores of the reference population, respectively [39].
Further, the observations for the household sources of drinking water and type of toilet facility were classified as “improved” and “unimproved” using the revised definitions specified by the WHO/UNICEF Joint Monitoring Programme (JMP) report [40]. Improved water sources included piped water into dwelling, piped water to yard/plot, public tap or standpipe, tube well or borehole, protected dug well, protected spring, and rainwater, while unimproved water sources included unprotected dug well, and unprotected spring, river, dam, lake, pond, stream, canal, and irrigation canal [40]. Improved sanitation facilities included flush toilet, piped sewer system, septic tank, flush/pour flush to pit latrine, ventilated improved pit latrine, pit latrine with slab, and composting toilet. Unimproved sanitation facilities included pit latrines without a slab or platform, hanging latrines or bucket latrines, and open defecation [40].
Maternal anaemia was defined as mothers with Hb levels of less than 12.0 g/dl, adjusting for altitude [36]. The amount of media exposure was derived from the following questions: (i) Do you read a newspaper or magazine at least once a week, less than once a week, or not at all? (ii) Do you listen to the radio at least once a week, less than once a week, or not at all? (iii) Do you watch television at least once a week, less than once a week, or not at all? Then, the amount of media exposure was calculated by summing up the reported frequency of each media if an activity was performed at least once a week. The household wealth index was constructed using data on a household’s ownership of selected assets, such as televisions and materials used for constructing the house, using the DHS wealth index [41].
Statistical analyses
Firstly, descriptive analyses were performed and are reported as frequency and percentage with their 95% confidence intervals (CI). To calculate unbiased estimates, the sampling weight, strata, and cluster were incorporated. Secondly, bivariate analyses were conducted using Rao–Scott Chi-square to test the differences between children who did and did not receive DM. Thirdly, multivariate logistic analyses were constructed using generalized linear mixed models with the binomial distribution and the logit link function, since children from the same communities/neighbourhoods/countries may present characteristics that are similar to those of individuals from different communities/countries. Thus, we adjusted for the correlated individual responses nested under a single country using models that handle correlated responses. We reported adjusted odds ratios (aORs) with their P-values and 95% CI. We also incorporated survey year fixed effects that control for common time effects across all surveys. We employed multivariable logistic regression to examine the independent factors associated with DM. A significance level of alpha equal to 5% was used to determine the statistical significance. All data entry, cleaning, and statistical analyses were conducted using SAS version 9.4 software (SAS Institute, Cary, NC, USA). We focused on only two levels (i.e. children as level 1 and country as level 2), since we were interested in examining the effects of DM on anaemia among children aged 6–59 months in selected SSA countries; multilevel models with dichotomous outcomes were employed to estimate the odds of success and the impact of various characteristics at different levels.
$$Y_{ij} = \beta_{0j} + \beta_{1j} X_{ij} .$$
(1)
Equation (1) presented above indicates a simple level 1 model with one child-level predictor, where Yij signifies the anaemia status for child i in country j, β0j signifies the intercept or the average log odds of the occurrence of anaemia in the country j, Xij is a level 1 predictor for a child i in country j, and β1j indicates the slope associated with Xij showing the link between the child-level variables and the log odds of anaemia occurrence. The regression coefficients are modelled hierarchically as follows:
$$\beta_{0j} = \gamma 00 + \gamma 01W_{j} + U_{0j} ,$$
(2)
$$\beta_{1j = } \gamma 10.$$
Equation 2 shows a simple level 2 model with one child-level predictor, where γ00 provides the log odds of anaemia prevalence among all counties, Wj is a country-level predictor, γ01 is the slope associated with this predictor, U0j is the level 2 error term representing the unique effect associated with country j, and γ10 is the average effect of the child-level predictor. Therefore, an integration of the level 1 and level 2 models produces the following.
$$Y_{ij} = \, \gamma 00 + \gamma 10X_{ij} + \gamma 01W_{j} + U_{0j} .$$
(3)