Background
Malaria is one of the most important causes of morbidity and mortality in Malawi. Since 2007, mass distribution of insecticide-treated nets (ITNs) to the demographically most vulnerable population groups has been a major part of Malawi’s vector control efforts. This, in combination with improved diagnosis and treatment of cases, has resulted in a 36% reduction in malaria cases, from an estimated 5.6 million cases in 2010 to 3.6 million in 2016 [
1]. Despite this progress, continued investment in malaria control has not led to a further decrease in cases [
1,
2] with an estimated 4.3 million cases still occurring in 2020.
The Malawi National Malaria Control Programme (NMCP) aimed to reduce malaria incidence by at least 50% from a 2016 baseline of 386 per 1000 population to 193 per 1000, and reduce malaria deaths by at least 50% from 23 per 100,000 population to 12 per 100,000 population by 2022 [
3]. To reach these targets, it is necessary to understand the impact of current control activities and tailor future vector control to local epidemiological and entomological dynamics [
4] in the context of a fast growing population. The Malawian population has grown almost 30% during from 2012 to 2022 to 20.4 million [
5]. To control the malaria vectors, in 2007, the government started distributing long-lasting insecticide-treated nets (LLINs). It is generally accepted that ITN coverage has helped decrease malaria prevalence in Malawi [
6‐
9]. However, retrospective studies have also found that a 13% increase in bed net access from 2012 to 2014 was not associated with a reduction in malaria burden in children [
10]. This was corroborated by a similar study investigating overall trends within the 2012 and 2014 malaria indicator survey (MIS) data [
11]. Comparison of data from 2004 and 2016 also showed that community malaria prevalence was not related to ITN use [
12]. No personal, and limited community, protection from ITNs was found in a field study from 2012 [
13]. These inconsistent conclusions suggest that ITN access and use may have a heterogeneous impact on malaria prevalence.
One of the main challenges for malaria elimination is the heterogeneity of the current malaria landscape [
1,
14,
15]. It is not well understood why this heterogeneity has emerged, although the possible varied efficacy of ITNs could be partly responsible. ITNs can impact areas differently due to different vector population compositions and behaviours, climate variation and the presence of resistance [
16‐
18]. Furthermore, gaps may exist between policy and implementation [
19], with human behaviour one of the most complex variables involved in malaria transmission. The local population might accept, but not use and maintain nets, use nets for other purposes or migrate to areas with higher malaria risk [
20‐
22]. Further reasons include social factors, such as autonomy in health care decisions [
23], bed net integrity and insecticide degradation [
24].
The best method for measuring the efficacy of ITNs directly is randomized controlled trials, ideally conducted across different parts of a country. No such randomized control trials have been conducted in Malawi due to the unethical nature of withholding nets from a proportion of the population. However, an important alternative source of information exists in the MIS. A search on the National Library of Medicine identifies more then 2,000 papers that have used MIS data in some capacity (search date 16–02-2022). These routine, large-scale household surveys are designed to produce snapshots of the malaria situation at national, regional and urban/rural levels. In Malawi, national MIS were conducted in 2010, 2012, 2014 and 2017 [
25‐
27]. These MIS, together, capture the malaria situation and dynamics throughout the country and represent an essential source of information for policy development [
1]. For example, MIS studies helped recognize the possible ineffectiveness of ITNs [
10,
11], which led to the recent addition of pyrethroid-Piperonyl butoxide (PBO) nets to the vector control programme. The latest MIS survey was conducted in Malawi in April 2021. Data are currently being analysed and have not been released to the public.
Even with the recent addition of PBO and dual active ingredient nets to the vector control programme, traditional permethrin nets will likely remain an important part of control efforts in Malawi due to the uncertain durability of the next generation bed nets[
28], their additional costs and unknown acceptance by the local population. In this study, the spatial relationship between ITN access and use, and malaria prevalence in children was investigated using MIS data from 2012, 2014 and 2017.
Discussion
This study presents an important, key modelling approach that allows malaria control programmes to spatially unravel the relationship between child malaria prevalence and ITN distribution. The non-stationary generalized linear model helped visualize variation in the relationship between malaria cases and bed net indicators. Increasing household access to and ownership of ITNs resulted in a decrease in malaria cases. However, this did not occur homogeneously across the country. The ITN coverage and use increased from 2012 to 2017, while child malaria prevalence decreased only in some areas. The BGGLM regression using MIS survey data showed that child malaria prevalence had a negative association with ITN population access and a positive association with ITN use. However, it is challenging to identify cause and effect without temporally linked data. The MIS data are currently the best information available to understand the impact of ITNs on malaria prevalence in Malawi. The large variation in odds ratio reflects the high uncertainty in the data and the need for further localized data that takes into account any potential confounder, to spatially disentangle the relationship between ITN indicators and malaria prevalence.
A negative association between malaria prevalence and ITN population access in all regions of Malawi was found. Areas with high population access to nets had lower malaria prevalence than areas with low access. This corresponds to earlier work in Malawi, where ITN ownership was found to be protective against malaria parasitaemia in children [
7,
47]. The non-stationary spatial analysis highlighted heterogeneity in the relationship with some areas showing no negative association. The assumption was made that nets maintained physical integrity and bioefficacy of insecticides at least one year post-distribution. Yet, the lack of a discernible relationship could indicate that the roll out was slow, new nets from the campaigns were not used (still in packaging), nets were not used consistently or nets were ineffective. The nets are unlikely to be entirely ineffective, as even with high insecticide resistance the physical barrier of nets and sublethal effect of the insecticides results in a negative relationship between malaria prevalence and ITNs [
47‐
49]. It is likely that other factors also affected prevalence, creating artefacts in the relationship between the outcome and predictor. For example, behavioural resistance, with mosquitoes biting when people are not protected by the nets, could play a role [
50,
51]. The quality of the nets has also not been considered, while studies have shown that using a mosquito bed net that is more than one year old is a risk factor for malaria in Malawi [
52]. Moreover, extreme droughts the year prior could have impacted overall malaria prevalence dynamics as the model does not account for key environmental factors [
53].
There was a positive relationship between malaria prevalence and ITN use in all Malawian regions, with high net use occurring in areas with high child malaria prevalence. The question ‘did you sleep under a bed net last night?’ was used as a proxy for overall ITN use. The question focuses on ‘last night’ and does not capture the overall ITN use behaviour. The positive association found here between malaria prevalence and net use is likely due to the large numbers of malaria cases motivating the population to sleep under nets more regularly in the short term. The MIS household surveys are sensitive to recall bias and social desirability factors [
54]. Additionally, nightly variables, such as temperature and mosquito density, could also greatly impact the use of nets [
55]. For ITN use throughout the malaria season to be related to malaria prevalence, thus identifying its protective ability, temporal studies are necessary where fieldworkers independently confirm bed net use over both space and time throughout the season.
This study revealed spatial heterogeneity in the relationship between malaria prevalence and the ITN indicators, with child malaria prevalence higher in rural than urban areas. It is widely accepted that malaria dynamics differ between rural and urban areas due to differences in demographics, socioeconomics, housing, drainage and access to health care [
22,
56‐
59]. Generally, in Malawi malaria is more prevalent in rural areas, although high case numbers have also been reported in urban areas [
10,
11,
22,
56,
58,
60]. Malaria control likely has a different impact in the urban and rural habitats, due to environmental differences and population movements. Furthermore, the dichotomization of the factors areinconsistent across the research field (the definition of what an urban or rural area is differs), with a more nuanced definition of urban malaria risk and prevention efforts necessary in Malawi to control adequately for the contextual factors that drive malaria prevalence [
61]. The results from this study further substantiate the complexity of urban/rural malaria dynamics [
22,
56] and show the importance of including this factor in the analysis to investigate the impact of vector control for sustainable intervention measures.
Many studies have shown that the scale-up of ITNs has protected the Malawi population from malaria [
6‐
9]. This study shows that this relationship varies across space. Heterogeneity in malaria cases has been identified previously in Malawi [
14,
52,
62]. Malaria epidemiology is a complex dynamic between many factors at the individual, household and community levels [
62]. Confounders can have a large impact on this heterogeneity, and they are challenging to identify. The introduction of vector control in this environment can have very different outcomes, with the decline rate in malaria prevalence known to be very different across the country [
14]. One possible explanation for spatial heterogeneity in the relationship between malaria prevalence and ITN use is that it captures the heterogeneity in malaria prevalence, with net coverage homogeneously high throughout the country. A small group of households can account for a majority of cases, which results in spatial heterogeneity in even small geographical areas [
63]. Another possibility is that insecticide and behavioural resistance of vector species is impacting net efficacy. Little information about the spatial distribution of malaria vectors and their resistance status is available, but reports do indicate heterogeneity in its spread with high resistance reported in the south and around Lake Malawi [
3]. It is also important to note the potential effect of other malaria control interventions on the heterogeneity of the relationship between malaria prevalence and ITNs, including the use of IRS, larval source management and house improvements [
34,
64] on malaria prevalence, which have not been considered in this study. Understanding the spatial dynamics will help increase the effectiveness of vector control campaigns, for example, by changing to bed nets that kill resistant mosquitoes more effectively or shifting to different control tools altogether in specific areas. The fine-scale spatial and temporal heterogeneity and their causes need to be investigated further by implementing field studies designed to answer these specific questions.
In concurrence with the recommendations from the Ministry of Health in Malawi, and in alignment with prior Malawian MIS studies [
10,
11], blood smear data were used as the sole proxy for malaria prevalence. Both the RDT and blood smear malaria testing methods have strengths and limitations [
65]. As shown here, the RDT generally produces higher positivity rates than blood smears, as it measures antigens that are detectable in the blood up to four weeks after parasite clearance [
66]. This is contrary to blood smear tests, which measure the physical presence of malaria parasites. From a modelling perspective, instead of choosing one indicator, joint distribution modelling of RDT and blood smear results could help improve inference. Recently, Amoah et al.[
67] developed a geostatistical framework to combine spatially referenced disease prevalence data from multiple diagnostics. Joint distribution models draw benefit from the combination of different diagnostics, although particular care needs to be taken for diagnostics with large discrepancies in sensitivity and specificity.
It is important to note that malaria prevalence is not the only way to measure malaria burden in a country. Studies have found mortality reduction without a decrease in malaria prevalence, and the other way around [
68]. Furthermore, malaria prevalence here is focused on children aged 6-to-59 months, while school-aged children are both at higher risk of infection and asymptomatic infection [
69,
70]. The study could, thus, be improved by using malaria prevalence in the entire population as an outcome variable, which is not possible with MIS data. Excluding this important risk group from the analysis could have skewed the data and results. Additionally, the MIS data are a snapshot and do not include the seasonality of malaria. Although surveys are planned during peak malaria season, as shown by Chirombo et al.[
15], this peak differs yearly in Malawi. Whether data are collected during the peak season or two weeks later hugely impacts malaria prevalence estimates. If snapshots of malaria prevalence are compared between the different years without a clear understanding of the seasonal dynamics in these different years, this will greatly influence the analysis and incorrect conclusions can be made. It is important to place malaria prevalence from MIS data in the context of the country and investigate how this is linked to mortality and other malaria indicators.
The MIS data are currently the only data available in Malawi to investigate the relationship between malaria prevalence and ITN use. As the MIS has not been designed for this purpose, caution is advised when interpreting the results. Studies designed to investigate the impact of ITNs are indispensable to understand the efficacy of ITNs [
68]. Until these studies are available, creative solutions are necessary to analyse MIS data. Two studies have previously used MIS data from 2012 and 2014 to investigate the relationship between child malaria prevalence and ITN use. Contrary to this study, both found that the number of bed nets per household was not significantly associated with malaria morbidity [
10,
11]. Both studies focused on socio-demographic characteristics, while this study included the spatial coordinates of the data within the analysis. This study shows that the impact of ITNs differs geographically. For vector control programmes to make informed decisions about future control activities, it is essential to have access to both country-wide and spatially disaggregated analyses of ITNs impact. This spatial disaggregation is especially important with an increase in IRS activities since 2019 and a combination of PBO and dual active ingredient bed nets distributed during mass ITN campaigns in 2018 and 2021 [
33].
The non-stationary generalized linear model visualized the geographically changing relation between malaria prevalence and the ITN indicators. Although the geostatistical mixed regression model presented the overall spatial relationship between child malaria prevalence and bednet indicators, it did not allow for visualization of variation of these relationships across Malawi. The maps produced by the non-stationary model can help vector control programme spatially disentangle the impact of interventions on malaria prevalence for vector control programs. The non-stationary model has some limitations. For example, the bandwidth is optimized based on accurate prediction of the response variable, not on accurate estimation of the coefficients [
46]. Especially when the regression model is fitted within a small kernel or with limited data, collinearity can be a problem [
71]. Nevertheless, for spatially clustered data such as the MIS data, this method can be appropriate to provide insights into how estimated relations vary across the country. Although relationships are not causal, they could highlight underlying covariates that have been left unmeasured. It is an important tool for eco-epidemiological studies [
72‐
74], especially for a disease such as malaria, that is so closely related to the environment and the sociodemographic dynamics of the population. Yet, spatially non-stationary models have only rarely been used in malaria research.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.