Background
Although millions are affected by clinical malaria each year, the last 15 years have seen unprecedented gains from international efforts made to control this disease. The establishment of the Roll Back Malaria initiative and the Millennium Development Goals in the year 2000 were followed by a nearly 20-fold increase in international funding for malaria control [
1]. The scale-up of interventions that followed has resulted in a 40 % decline in
Plasmodium falciparum clinical incidence in Africa while prevalence of the infection has nearly halved since the year 2000 [
2]. This and marked reductions in malaria-associated deaths [
3] were largely attributed to increased coverage of the insecticide-treated bed nets (ITNs). Second to ITNs, which were the most widespread intervention, access to artemisinin-based combination therapy (ACTs) has been found to greatly impact the incidence of disease [
2].
Prompt diagnosis and treatment of clinical malaria is the mainstay of all control or elimination programs [
4‐
6]. The primary aim of treatment with ACT is to curtail clinical disease in patients, though access to effective treatment also impacts onward transmission to the wider community by reducing the infectious reservoir while at the same time containing the spread of drug resistance [
2]. Treatment coverage has been assessed across Africa [
7], but understanding how access to treatment varies throughout the malaria endemic world is essential to evaluating the true public health impact of treatment coverage. This is all the more significant during the current transition from control of clinical disease towards regional focuses on malaria elimination and the post-2015 future is shaped as progress towards the Millennium Development Goals is evaluated [
8].
In assessing care-seeking behaviours, it is also important to consider what proportion of care is sought at government-based facilities. Government facilities are more likely to comply with recommended diagnostic and treatment schedules [
1,
9], and to have their routine records integrated into the national health data management system. Even in areas with strong surveillance systems, reports of passively detected malaria cases will capture only a certain fraction of all malaria cases [
10‐
13] and so must be adjusted by a number of parameters before use as official burden estimates [
1,
14]. These include (1) treatment-seeking behaviour (representing the proportion of cases not attending health facilities and thus being omitted from aggregated case reports [
15‐
22] as well as the proportion seeking treatment outside the public health system); (2) malaria diagnoses made presumptively without parasitological confirmation (leading to reported case numbers including non-malaria illnesses [
23,
24]); and (3) incomplete reporting (which leads to cases being lost from reported data) [
1,
25]. Here we aim to improve the evidence-base of the first parameter to enable refined estimates of the true clinical burden of malaria disease.
Treatment-seeking rates vary widely between countries and greatly affect final burden estimates. Where available, these parameters are drawn from nationally representative, cross-sectional, household surveys such as Demographic and Health Surveys (DHS) [
26], Malaria Indicator Surveys (MIS) [
27] and UNICEF Multiple Indicator Cluster Surveys (MICS) [
28]. However, not all malaria-endemic countries (MECs) have such survey data available, resulting in these important parameters being either assumed or omitted and making comparisons of access to treatment across all malaria endemic regions and effective estimation of the burden of disease difficult.
The aim of this study was an exhaustive assembly of the available data on treatment-seeking for all MECs, sub-nationally where possible. For MECs lacking national survey data, predictive models were built to estimate use of public facility treatment as well as treatment of any kind. These estimates, including measurements of their uncertainty, will allow for improved understanding of how health services in endemic countries are routinely accessed and facilitate more accurate disease burden estimation.
Discussion
Information regarding the treatment-seeking behaviours of fever cases in MECs is essential to assessing the feasibility and success of malaria control and elimination programs. Malaria is a treatable disease and while effective therapies exist, population health-seeking behaviours may limit the extent to which they are utilized. Understanding care-seeking rates in MECs also helps to quantify the scale of the malaria burden. Routine surveillance data are used to measure the burden of clinical disease and only those cases who seek care at a government-based facility are likely to be included in regional estimates. Quantification of the proportion of cases that are severe enough to seek care (clinical), but are missed from passive surveillance, is therefore necessary to more accurately estimate ‘true’ case numbers.
Data obtained from national surveys (DHS and MICS) could be assembled for the majority of the MECs (78 %,
n = 76). Treatment seeking rates specific to fevers were available from all DHS surveys and MICS 5, while only cough-based treatment rates were available from the earlier MICS surveys. Precedent exists for using care-seeking for respiratory infection as a proxy for fever treatment [
14,
25], and this was evidenced by comparable treatment-seeking rates for cough and fever in the observed data gathered here (Additional file
1: Figure S2). Therefore, all available treatment-seeking data were used to inform the model. However, because fever is a primary symptom of malaria and because DHS data are georeferenced and could be mapped sub-nationally, DHS data were given precedence when generating final mapped outputs and summary estimates of the data gathered. Sub-national predictions and mapping of the treatment-seeking outcomes using cluster-level data to produce smooth surfaces like those produced by the Malaria Atlas Project for prevalence was also explored [
2,
53,
54]. However, there were not sufficient covariate data available at smooth resolutions at the time of this analysis to support this. The quantity and quality of higher resolution sub-national covariate data that can be used in geostatistical analyses continue to improve with time and there may be greater potential for this type of analysis in the future [
55].
Gathering treatment-seeking data for all MECs was the primary aim of this analysis, with the intent to also show sub-national rates where possible. Data that corresponded to the covariate variables identified in the literature review of key factors determining treatment seeking behaviour were readily available from the World Bank and produced models to show that government treatment-seeking and any treatment-seeking could be predicted at the national level from a limited set of covariate variables: year, WHO region, percent of pregnant women that receive prenatal care, immunization rates, primary education completion rate, GDP growth, and national health expenditure (public and total). The percentage of women receiving prenatal care was a strong indicator of fever treatment-seeking that likely drove the low model error values observed. The differing drivers of seeking care from either source were evidenced through the difference in the best models for each treatment-seeking outcome. Government-based treatment seeking was predicted primarily by other health-seeking indicators such as the childhood immunization (DPT) and prenatal care. Access to any treatment, on the other hand, was also influenced by country wealth and education. Educated individuals in more economically stable countries are therefore more likely to spend money on health care outside of the government system.
From the assembled observed data and modelled missing data, there emerged geographic patterns in both the outcomes and the certainty of the predictions. There were areas with low access or use of public treatment facilities in all regions of the malaria-endemic world. Figure
3 highlights areas such as Central Africa and Indian sub-continent. Accessing treatment of any kind was inherently higher in all countries because government facilities are included in that metric. However, the any treatment data revealed that treatment-seeking in some endemic areas, such as India, Pakistan and Afghanistan, was largely pursued outside the public sector. The CI ranges (Fig.
2) show that outcomes were well predicted in the Americas, but less so in Asia and least accurate in the Eastern Mediterranean countries. This implies that models and indicator variables were better suited to the Central and South American countries. Future predictions of this nature may be improved by including additional covariates or, following further research into treatment-seeking indicators, parameters that are tailored to each region.
The comparison of the two treatment-seeking outcomes and the measures of uncertainty provide valuable information as countries define control and elimination goals. The predicted measurements resulted in the greatest uncertainty, and signal the need for treatment-seeking to be formally assessed in those regions. Most notably, this data assembly and analysis reveals parts of the malaria endemic world where treatment is primarily sought outside of government programs. In light of concerns regarding the spread of antimalarial resistance [
56,
57], it is essential for countries to ensure that cases are being diagnosed and treated properly using approved and legitimate drugs [
58]. If treatment is most commonly sought outside of government-based facilities, control programs must consider how best to monitor treatment safety and efficacy as well as numbers of cases presenting for treatment.
Conclusion
Information on treatment-seeking behaviours in malaria endemic countries can be readily assembled from national survey data. Where data on treatment-seeking behaviours were not available from national surveys, modelling techniques using freely available data were applied to fill data gaps. Both the results and methods presented have potential application beyond those described here and may inform the control and burden of other febrile diseases. However, in this context, data on treatment-seeking for fever are essential to understanding the efficacy with which malaria cases are treated and detected. Gathering and visualizing these data for all MECs, sub-nationally when possible, is of use to estimate the burden in areas of low endemicity where passive surveillance is the primary tool through which cases are monitored. These results will facilitate downstream efforts to produce a hybridized burden estimation approach that employs both surveillance-based and cartographic techniques in an effort to more accurately quantify the global burden of falciparum and vivax malarias and provide immediate feedback regarding parts of the malaria endemic world where treatment for malaria is not readily accessed or is more commonly sought beyond the government or control programme sectors.
Authors’ contributions
KEB and PWG conceived the study and oversaw its implementation, with input from SIH and REH. KEB wrote the first draft of the manuscript and assembled treatment-seeking and indicator data with assistance from HG, DW, BM, and UD. DB assisted the design of the modelling framework with input from SB and EC. All authors read and approved the final manuscript.