Models of malaria transmission used for national strategic planning are informed by household survey data on intervention coverage, transmission intensity, and malaria burden. To set subnational intervention coverages, models rely on DHS measures of treatment-seeking rates for febrile illness among children under five, insecticide-treated nets (ITN) usage at the household level and for different age groups, and coverage of intermittent preventive treatment in pregnancy (IPTp). Modelled transmission intensity can then be calibrated to capture DHS measures of the Plasmodium falciparum parasite rate in children under the age of five (PfPR0–5).
Using the DHS to parameterize fine scale models introduces additional sources of uncertainty. To help NMCPs stratify and plan operations, models must capture data at admin-2. However, estimates of malaria prevalence and intervention coverage from the DHS are only meant to be representative at a state or provincial (admin-1) level (Figs. 1, 2) and are underpowered to measure these indicators at the admin-2 level. Modelling predictions based on parameters from DHS household cluster data would, therefore, be biased. Moreover, data collection and cleaning standards for georeferenced DHS data also increase the risk of biased admin-2 projections. Sampling errors while using GPS receivers to georeference cluster locations could lead to attribution of admin-2 data from one to another. Additionally, the displacement of cluster locations to protect participants’ confidentiality [11] and any resultant random effects from data jittering would further exacerbate the problem of misclassifying admin-2 data.
Malaria indicators captured by the DHS are subject to seasonal variations in malaria transmission and human behaviour, which limit understanding of malaria transmission intensity, ITN use, and comparability of yearly surveys. Parasite rate is typically at its maximum during the rainy or peak mosquito-biting season and trends downwards in the dry season. Individuals use ITNs during the wetter months and reduce usage in the dryer months when mosquito activity is diminished [12]. Treatment-seeking behaviour can be affected by seasonal accessibility issues and seasonal demands on parents’ time, for example, agricultural needs during the wet season. Therefore, malaria indicators from DHS surveys conducted during the dry season months do not necessarily capture parasite rate, ITN use, and case management coverage in the peak transmission season. Surveys conducted in different seasons, even within the same DHS year, are not directly comparable without adjustment for the seasonality effect. NMCPs and modellers resort to other data sources with a narrower geographic scale to capture seasonal and temporal changes in malaria transmission and accurately identify gaps in intervention coverage and areas of high prevalence.
The restriction of current questions to select age groups limit how informative the results are for driving country strategy and parameterizing models. For example, the DHS only tests children under the age of five for malaria infection, which, although important, is of limited utility for categorizing malaria transmission intensity in settings where more of the burden is in older children or adults. PfPR0-5 measured during implementation of seasonal malaria chemoprevention (SMC) may be particularly uninformative as PfPR is suppressed in this population and SMC coverage is not assessed in the DHS. Measurements of PfPR in older children can be more informative than PfPR0–5 even in high-transmission areas, as children above age two will have some immunity to clinical malaria, and hence less treatment with anti-malarials, yet limited immunity to parasitaemia itself [13]. Some models, therefore, apply standardization algorithms to convert PfPR0–5 to PfPR2–10 [14]. While such algorithms have been validated in prior work [13], the extent of bias introduced by predicted PfPR2–10, especially in fine-scale models, is unknown.
A similar issue arises with using the DHS data to evaluate case management and treatment coverage for uncomplicated malaria, where questions are restricted to children under the age of five. NMCPs, therefore, know little about access to malaria treatment in older children, where burden is increasingly shifting [15]. In the absence of case management information for uncomplicated malaria in older children and adults, modellers either assume homogeneous coverage by age or turn to site-specific research studies on treatment-seeking behaviour.
Estimating case management rates from DHS data requires analysing questions directed at a subset of DHS participants, which reduces the sample size and may introduce validity issues and inconsistencies. In the 2018 Nigeria DHS, effective case management coverage, that is the proportion of children under the age of five that received artemisinin-based combination therapy (ACT) among those that had a fever within the 2 weeks prior to the survey, was 22% at the national level. Disaggregated at the state level, ACT-related case management was remarkably low in many areas. For example, the 2018 DHS suggests that febrile children were not treated at all with ACT in Nasarawa, and only about 3 to 4% in Zamfara and Yobe (Fig. 3a).
When these estimates were discussed with the Nigerian National Malaria Elimination Programme, they indicated that the actual ACT use would likely be higher than that seen in the 2018 DHS, and the state level DHS estimates would not agree with their perceived ACT use in many parts of the country. The 2015 ACTWatch survey [16, 17] supports this view, which indicated that most outlets stocking any anti-malarials in individual states had at least one type of ACT medicine for sale (Fig. 3b). While the metrics are clearly different, the ACTWatch data suggests intense penetration of ACT across both the public and private health care sectors in Nigeria, and, together with the Nigerian programme perspective, calls into question the 2018 DHS results that suggested extremely low rates of artemisinin-based combination therapy in some areas of Nigeria. This discrepancy of trends between access (ACTWatch) and use (DHS) metrics emphasizes the limitations of the current DHS sampling strategy to capture case management coverage among febrile children, who are few in number, and the need for a strengthened DHS data collection system that builds trust and meets NMCP needs.
The gaps that we have identified within the DHS sampling strategy and questionnaires do not diminish the immense contribution of the DHS Program to evidence-based decision-making. However, when DHS measures do not adequately capture malaria indicators, or DHS data are out of concordance with institutional knowledge and beliefs of intervention and treatment access and malaria risk behaviour, deciding where to target interventions becomes more challenging and a data-driven approach nearly impossible.