Diagnostic testing of pediatric fevers: Meta-analysis of 13 national surveys assessing influences of malaria endemicity and source of care on test uptake for febrile children under five years

Background: In 2010, the World Health Organization revised guidelines to recommend diagnosis of all suspected malaria cases prior to treatment. There has been no systematic assessment of malaria test uptake for pediatric fevers at the population level as countries start implementing guidelines. We examined test use for pediatric fevers in relation to malaria endemicity and treatment-seeking behavior in multiple sub-Saharan African countries in initial years of implementation. Methods and Findings: We compiled data from national population-based surveys reporting fever prevalence, care-seeking and diagnostic use for children under five years in 13 sub-Saharan African countries in 2009-2011/12 (n = 105,791). Mixedeffects logistic regression models quantified the influence of source of care and malaria endemicity on test use after adjusting for socioeconomic covariates. Results were stratified by malaria endemicity categories: low (PfPR2-10<5%), moderate (PfPR 2-10 5-40%), high (PfPR2-10>40%). Among febrile under-fives surveyed, 16.9% (95% CI: 11.8%-21.9%) were tested. Compared to hospitals, febrile children attending non-hospital sources (OR: 0.62, 95% CI: 0.56-0.69) and community health workers (OR: 0.31, 95% CI: 0.23-0.43) were less often tested. Febrile children in high-risk areas had reduced odds of testing compared to low-risk settings (OR: 0.51, 95% CI: 0.42-0.62). Febrile children in least poor households were more often tested than in poorest (OR: 1.63, 95% CI: 1.39-1.91), as were children with better-educated mothers compared to least educated (OR: 1.33, 95% CI: 1.16-1.54). Conclusions: Diagnostic testing of pediatric fevers was low and inequitable at the outset of new guidelines. Greater testing is needed at lower or less formal sources where pediatric fevers are commonly managed, particularly to reach the poorest. Lower test uptake in high-risk settings merits further investigation given potential implications for diagnostic scale-up in these areas. Findings could inform continued implementation of new guidelines to improve access to and equity in point-of- care diagnostics use for pediatric fevers. © 2014 Johansson et al. National population-based cross-sectional surveys from DHS, MICS, MIS and ACT Watch conducted in sub-Saharan Africa since 2008 were systematically reviewed for inclusion in this study (Figure 1). 84 surveys were conducted in sub-Saharan African countries between 1 January 2008 and 1 June 2013; 40 datasets were publicly available by 1 June 2013 or were made available by the implementing organization. All datasets were included if they measured the outcome according to RBM guidelines [35], and main covariates as described below. 11 surveys did not collect the outcome measure, or data were collected using non-standard methods. 14 surveys did not collect information to measure main covariates (source of care or malaria endemicity). Two datasets were excluded because a more recent survey was available for the country. 13 DHS and MIS met inclusion criteria, which spanned the period 2009–2011/12 (Table 1). Survey methods are described elsewhere [36]. All surveys with one exception were conducted after national policies were changed to recommend parasitological diagnosis for all age groups prior to treatment, although countries were at different stages of operationalizing these policies at the time of survey fieldwork [40]. For this reason, country-level results are included as a supplement to this paper (Table S1). Malaria diagnostic test use is measured by asking caregivers of children under five with reported fever in the past two weeks if “At any time during the illness did (name) have blood taken from his/her finger or heel for testing?” This question does not differentiate between diagnostic tests, and is assumed to refer to either microscopy or mRDT. There were two main covariates: source of care and malaria endemicity. Source of care is measured by asking caregivers of febrile children if they sought advice or treatment for the illness, and if so, from where care was sought. Multiple responses are allowed, and response categories are standardized across countries with some modifications to account for different health system structures. This covariate was categorized as: (1) hospital (2) non-hospital formal medical (3) community health worker (CHW) (4) pharmacy (5) other (6) no care sought. Hospital, CHW, and pharmacy include any such listed response. Non-hospital includes any formal health system source that is not a hospital or CHW, including health centers or posts, outreach or mobile clinics, and private doctors. Some countries include additional sources for this category, such as maternities or municipal clinics. Other includes shops, traditional practitioners, relatives, and non-specified sources. ‘Hospital’ and ‘non-hospital’ categories were further dichotomized into public or private sources to analyze test uptake across different managing authorities. The questionnaire does not explicitly record where testing occurred, but plausibly happened where care was sought. If the child visited multiple sources (e.g. hospital and pharmacy), it was assumed testing occurred at the highest level attended and the covariate was coded using a hierarchical stepwise approach. We conducted a sensitivity analysis by comparing adjusted odds ratios with a covariate constructed by excluding febrile children visiting both hospital and non-hospital sources. In this approach, 732 febrile children visited multiple sources in 13 countries; 367 were excluded that visited both hospital and non-hospital sources. No significant difference was found between approaches (data not shown). Malaria Atlas Project estimates of malaria endemicity were included in the model, which are described elsewhere [41]. Briefly, the geographical limits of malaria transmission were estimated using routine reporting data and biological models of transmission-limiting aridity and temperature conditions. Within these limits, parasite prevalence survey data were assembled, geolocated, and used within a Bayesian geostatistical model to interpolate a continuous space-time posterior prediction of age-standardized Plasmodium falciparum parasite rate in 2–10 year olds (PfPR2–10) for every 5×5-km pixel for the year 2010. Malaria endemicity estimates were linked to survey datasets through geocoded PSUs. All individual observations were assigned their PSU-level malaria risk value, which was then categorized into one of five malaria endemicity classes: malaria free; unstable transmission; and low (PfPR2–10<5%), medium (PfPR2–10 5%–40%), and high (PfPR2–10>40%) stable endemic transmission. Socioeconomic covariates associated with child survival intervention uptake were incorporated in the model. These included child’s age and sex, maternal age and education, household wealth and density, and residence [42], [43]. Child’s age was categorized as 0–5, 6–11, 12–23, 24–35, 36–47, 48–59 months. Maternal age was categorized as 15–24, 25–29, 30–34, 35–39, 40–49 years. Maternal education was categorized as no education, primary and at least secondary education attendance. A household wealth index is pre-specified in datasets and described elsewhere [44]. Household density was categorized as 1–4, 5–8, 9–12 and 13 or more household members [45]. Residence was dichotomized as urban or rural. Among 29,245 febrile children under five surveyed in 13 countries, 300 had missing values for the outcome, 312 for source of care, 752 for malaria endemicity, and one for maternal education. 36 had missing values for two or more variables. Listwise deletion was used to exclude observations with any missing value from the analysis. Mixed-effects logistic regression models were used to quantify the influence of covariates on malaria test use in pooled and individual country datasets. PSUs were nested within country identifiers and normal distribution of the random effects was assumed. Covariates were included as categorical fixed effects nested within PSUs. Crude odds ratios of main covariates (malaria endemicity and source of care) were initially estimated for their effect on the outcome. Main covariates were then included simultaneously in one model and, subsequently, odds ratios were adjusted for the effect of all covariates, as listed above. We tested for an interaction between maternal education and malaria endemicity in the final model, and separately for an equivalent interaction between maternal education and residence. Results were stratified by malaria endemicity categories and separately by residence to examine effect differences across contexts. The level of statistical significance was set to 0.05. National point estimates were tabulated using sample weights pre-specified in datasets, and proportions for the pooled dataset were estimated using meta-analytical methods. Standard error estimation accounted for data clustering in survey designs. Stata 12 (STATA Corp, College Station, TX) was used for all analyses. We also crudely estimated total pediatric fevers attending and tested at different sources of care in 2010 across studied countries to further contextualize findings in our discussion. This was done by applying proportions tested from our analysis to published estimates of total pediatric fevers updated to 2010 [46]. This crude analysis helps visualize the rough magnitude of tested and untested pediatric fevers at different sources of care in order to further inform discussion of results.