Menu
Menu
Menu
Menu
Background: Scaling up diagnostic testing and treatment is a key strategy to reduce the burden of malaria. Delays in accessing treatment can have fatal consequences; however, few studies have systematically assessed these delays among children under five years of age in malaria-endemic countries of sub-Saharan Africa. This study identifies predictors of prompt treatment with first-line artemisinin combination therapy (ACT) and describes profiles of children who received this recommended treatment. Methods: This study uses data from the most recent Demographic and Health Survey, Malaria Indicator Survey, or Anaemia and Parasite Prevalence Survey conducted in 13 countries. A Chi square automatic interaction detector (CHAID) model was used to identify factors associated with prompt and effective treatment among children under five years of age. Results: The percentage of children with fever who received any anti-malarial treatment varies from 3.6 % (95 % CI 2.8-4.4 %) in Ethiopia to 64.5 % (95 % CI 62.7-66.2 %) in Uganda. Among those who received prompt treatment with any anti-malarial medicine, the percentage who received ACT ranged from 32.2 % (95 % CI 26.1-38.4 %) in Zambia to nearly 100 % in Tanzania mainland and Zanzibar. The CHAID analysis revealed that country of residence is the best predictor of prompt and effective treatment (p < 0.001). Depending on the country, the second best predictor was maternal education (p = 0.004), place of residence (p = 0.008), or household wealth index (p < 0.001). Conclusions: This study reveals that country of residence, maternal education, place of residence, and socio-economic status are key predictors of prompt access to malaria treatment. Achieving universal coverage and the elimination agenda will require effective monitoring to detect disparities early and sustained investments in routine data collection and policy formulation.
This study uses data from the most recent Demographic and Health Survey (DHS), Malaria Indicator Survey (MIS), or Anaemia and Parasite Prevalence Survey (A&PS) conducted in each PMI priority country. Each of these surveys included a malaria module, in addition to the standard household and women’s questionnaires. All surveys were conducted between 2006 (Benin) and 2012 (Malawi), with most carried out in 2010 and 2011. The countries belong to three geographical regions of SSA: (1) west, (2) central, and (3) east (Fig. 1). Countries and household surveys included in the study All countries had adopted ACT as first-line treatment for malaria (‘effective treatment’) by the time the surveys were conducted; however, the duration between policy adoption and the survey varies from 2 years (Benin) to 7 years (Liberia, Ethiopia, and Mozambique), with an average of 5.2 years (Table 1). Duration between year of ACT policy adoption and implementation of surveys ALu artemether–lumefantrine, ASAQ arthesunate–amodiaquine Data quality checks included assessing response rates, evaluating completeness of data for key variables, and assessing reliability of birth history data. Response rates refer to the percentage of the number of people or households in the sample that completed an interview. In each country, more than 90 % of identified households and women were successfully interviewed, confirming adequate response rates. Misreporting of ages or dates of birth can affect malaria estimates because malaria infection varies significantly by a child’s age [13, 14]. In each country included in the study, at least 93 % of children had complete month and year of birth data, so no country was eliminated due to incomplete information. The reliability of maternal reporting was assessed by analysing the number of live births per year. Overall, the number of births was high for the 4 years preceding the survey but lower in the fifth year preceding the survey. This distribution did not, however, affect the overall reliability of the birth history data. Of 16 countries, 14 had information on whether anti-malarial treatment was received on the same or next day as the onset of fever (‘prompt treatment’). Two countries, Ethiopia and Mali, were excluded from the bivariate and multivariate analyses because they do not include this information. In addition, the Benin database did not have time-to-treatment for ACT and was excluded from some analyses. The analysis included two steps. The first step involved computing proportions and conducting Chi square tests for each country to identify associations between prompt treatment and selected background characteristics of children, their primary caretakers, and their households. The data was weighted to account for oversampling, undersampling, and varying response in different regions included in national household surveys. The second step consisted of pooling data from all countries and running a Chi square automatic interaction detector (CHAID) model [15, 16]. This method is a sequential fitting algorithm; at each step, the model chooses the predictor variable that has the strongest interaction with an outcome of interest: here, prompt and effective treatment. The variable with the strongest association becomes the first branch of the tree, with a leaf for each category that is significantly different. CHAID then assesses the category groupings to pick the most significant combination of variables. It is particularly useful for identifying sub-groups that are more or less likely to experience the outcome of interest and for determining the relative contributions of these groups to overall coverage in the general population. The analysis was restricted to children with fever who received any anti-malarial medicine. The covariates were the child’s sex, age group in months (<6, 6–23, 24–59), relationship to the head of household (child or stepchild, grandchild, other), maternal age in years (15–19, 20–29, 30–49), maternal education (none, primary, secondary and above), place of residence (urban, rural), household wealth quintiles, and country of residence. The output of the CHAID model is presented in a hierarchical tree structure and consists of several levels of branches: root node, parent nodes, child nodes, and terminal nodes. The root node, ‘Node 0’, comprises children who received any anti-malarial medicine. Parent nodes are upper nodes compared to lower-level child nodes. Terminal nodes are any node that does not have child nodes. For each terminal node, the CHAID model provides the following indicators: (1) demographic weight in the sample; (2) gain, the number of children who received prompt and effective treatment in the terminal node, divided by the total number of children who received any anti-malarial treatment; (3) response, the proportion of children who received prompt and effective treatment among all those within the terminal node; and, (4) gain index percentage, which represents the increased probability of prompt and effective treatment in the terminal node compared to the overall study population.
N/A
