Two decades of malaria control in Malawi: Geostatistical Analysis of the changing malaria prevalence from 2000-2022 [version 1; peer review: awaiting peer review]

Abstract

malaria for the last two decades to understand the past transmission and set the scene for the elimination agenda. Methods A collation of parasite prevalence surveys conducted between the years 2000 and 2022 was done. A spatio-temporal geostatistical model was fitted to predict the yearly malaria risk for children aged 2-10 years (PfPR 2-10) at 1×1 km spatial resolutions. Parameter estimation was done using the Monte Carlo maximum likelihood methods. District level prevalence estimates adjusted for population are calculated for the years 2000 to 2022. Results A total of 2,595 sampled unique locations from 2000 to 2022 were identified through the data collation exercise. This represents 70,565 individuals that were sampled in the period. In general, the PfPR2_10 declined over the 22 years. The mean modeled national PfPR2_10 in 2000 was 43.93 % (95% CI:17.9 to 73.8%) and declined to 19.2% (95%CI 7.49 to 37.0%) in 2022. The smoothened estimates of PfPR2_10 indicate that malaria prevalence is very heterogeneous with hotspot areas concentrated on the southern shores of Lake Malawi and the country's central region.

Conclusions
The last two decades are associated with a decline in malaria prevalence, highly likely associated with the scale up of control interventions. The country should move towards targeted malaria control approaches informed by surveillance data.

Introduction
Malaria is a disease of public health importance affecting many communities to date 1 . In 2019, it was estimated that there were 215 million malaria cases globally, 94% of which were in sub-Saharan Africa 2 . Malaria infection is caused by protozoa of the genus Plasmodium that has five known species that are responsible for human infection 3 : Plasmodium falciparum, P. vivax, P. ovale, and P. malariae, more recently, P. knowlesi,. The vector responsible for human transmission is the female anopheles mosquito 4 . In Sub-Saharan Africa, malaria is mainly caused by Plasmodium falciparum and is one of the leading causes of morbidity and mortality especially in children under five years 5 . Other high-risk groups include pregnant women 6 and immunologically naïve persons like travelers coming from non-endemic places 7 .
In 2016, The World Health Organisation (WHO) released the global technical strategy (GTS) for malaria 2016-2030 to help countries accelerate progress toward malaria elimination. The strategy targets reducing global malaria incidence and mortality rates by at least 90% by 2030 and eliminating in at least 35 countries by 2030 8 . Malawi has aligned its malaria strategic goals to GTS, and the country reflected its commitment to the GTS in the 2016-2022 Malaria strategic plan 9 and will plan to continue implementing the strategy beyond 2022.
Since 2010, malaria control efforts in Malawi have scaled up substantially through multiple control measures that include bednets, Artemisinin Combination Therapies (ACTs) and malaria Rapid Diagnostic Tests (mRDTs). With this, malaria transmission has reduced by 44% 10,11 , and clinical case data has become much more accurate as over 95% of government-provided treatments are now based on mRDT results 12 . As malaria transmission declines, its heterogeneity will increase, and transmission will be increasingly driven from 'hotspots.' At lower transmission levels, targeted control efforts becomes essential to maximize available resources' impact and further reduce the burden 13 . In 2019, the Malawi National Malaria Control Programme (NMCP) was on track against its Malaria Strategic Plan (MSP) for 2017-2022 to reduce malaria incidence by 50% by 2022 from an initial baseline of 386 per 1,000 population in 2015 (2019 NMCP programme review). If progress continues, sub-district targeted control will become crucial for the next seven-years (MSP 2023(MSP -2030.
The Malawi NMCP priorities are in line with WHO guidance for countries to regularly analyse their key malaria indicators to predict, respond, and monitor the malaria situation in-country in terms of intervention delivery, coverage, and disease burden. This includes the ability to detect local malaria hotspots to guide control programmes with timely evidence-informed responses 8 . Malaria risk mapping has a long legacy in Africa, including Malawi. Mosquito breeding site maps became available in the 1950s, soon after the discovery of mosquitos as the malaria vector by Sir Ronald Ross 14 . Early European settlers did early risk maps for Malawi in their attempts to provide cartographic information on climate, agriculture, and mosquito breeding sites. These maps provided control agencies with a plan for larval control, environmental management, and mass drug administration targets 15 .
Application of the more recent methods of model-based geostatistical (MBG) methods to map malaria risk in Malawi only started to be used in 2006. Kazembe and colleagues in their work utilised data on malaria prevalence from 73 sampled survey locations, where children aged 1-10 years had been sampled between 1970 and 2001 16 . Their work was directly used in the Malawi malaria programme review in 2010 and the national strategic plan 2011-2015 to highlight the hyper-endemic nature of malaria transmission in the country, with variations in higher altitude areas.
Efforts to model spatio-temporal heterogeneity in malaria have focused on parasite prevalence (infection) data from household surveys, because of concerns over the quality and completeness of routine clinical malaria case data from the district health information system (DHIS2). In 2019, Chipeta and colleagues used MBG methods to describe the changing malaria transmission in Malawi between the years 2010 and 2017 17 .
There have been several exciting developments in malaria control in the period between 2017 and 2021. In 2019, the country started piloting the RTS,S malaria vaccine in 11 districts. The RTS,S/AS01 is a leading malaria vaccine candidate developed to prevent diseases caused by Plasmodium falciparum. The results of a phase 3 trial of this vaccine confirmed moderate protection with overall efficacy estimates of 46% (95% CI 42,50) against clinical malaria and 38% (95% CI 18, 53) against severe malaria by 18 months after dose 3. A fourth dose, given 18 months after dose 3, increased efficacy against clinical malaria from 26% (95% CI 21, 31) to 39% (95% CI 34,43) and from -2 (95% CI -31,20) to 31.5% (95% CI 9,48) against severe malaria. Vaccine efficacy was also confirmed against malaria hospitalization (37%, 95%CI 27-48.5), all-cause hospitalization (15%,  and severe anaemia (62%, 95%CI 26.5-81) in children who received 4 doses of RTS, S/AS01 18,19 . Following a successful pilot, the WHO has recommended the vaccine for wider use. Another success has been the mass net distribution campaigns. In 2018, the country introduced piperonyl butoxide (PBO) net in addition to the regular long-lasting insecticide treated bed nets 20 . Additionally, Malawi also implemented annual indoor residual spraying (IRS) in some districts in 2018 and annually afterwards.
The same period has been faced with several challenges: The COVID-19 pandemic reduced the health care seeking for most febrile illnesses including malaria. There were also associated health system disturbances because of the pandemic.
Other health system disruptions included tropical storms in 2019 and 2021, measles outbreak in 2017, polio outbreak in 2021 and cholera outbreak in 2020.
Carrying forward the successes in the last five years, the NMCP is poised to develop the revised malaria strategic plan (MSP) in 2023-2030. Informed by strengthened surveillance systems, the NMCP is transitioning from blanket interventions across the country to exploring options for targeted intervention strategies. This is already reflected in the five-year integrated vector control strategy (IVCS) plan for the period 2020-2024 that informs net distribution and targeted indoor residual spraying (IRS). Similarly, the NMCP is evaluating the expansion of community-level malaria case management in over five year olds for hard-to-reach and high burden areas. Targeted interventions will become a key part of the 2022-2027 MSP, which is scheduled to be developed in 2022/2023. It is for this reason why it is important to describe the malaria epidemiology in both space and time. The aim of this analysis is to describe the malaria epidemiology prevalence for the last two decades including description of the subnational disease risk to guide target interventions.

Study area
Malawi is a small country in sub-Saharan Africa. It neighbours Tanzania in the north, Mozambique in the south, southwest and east and Zambia to the west. The country has five major lakes (Lakes Malawi, Malombe, Chilwa, Chiuta, Kazuni and Kaulime) that contribute to 21% of the country's 118,484 km 2 territorial surface area. The country consists of four main geographical regions: the Great Rift Valley, the central plateaus, the highlands, and the isolated mountains with the rift valley forming the most striking topographic feature that runs for the entire length of the narrow country. It passes through Lake Malawi in the Northern and Central Regions and stretches to the Shire Valley in the south. This contributes to the variable temperature patterns in Malawi that average 144° to 32° Celsius based on altitude and proximity to the lake 17,21 . There are three weather seasons hot-wet, hot-dry and cool-dry. From May to August, the weather is cool and dry, becomes hot in September and October, and the rainy season begins in October or November, continuing until April.
The country is divided into three regions namely Northern, Central, and Southern regions. There are 28 districts in the country: 6 districts in the Northern region, 9 in the Central region, and 13 in the Southern region ( Figure 1). The country has a population of 17.6 million 18,22 and a GDP per capita of 320 USD.

Data gathering
The collation of malaria survey data into a single geocoded repository followed a cascaded approach. The first step followed a more traditional peer-reviewed publication search in various data bases: PubMed, Google Scholar, the WHO Library Database and African Journals Online. The key words used for this search were "malaria" and "Malawi". The last electronic search was completed in June 2022. The next step was a data request from the Malawi NMCP for the national Malaria indicator survey data for the surveys done in 2010, 2012, 2015, 2017 and 2021. The last step was to reach out to the research community within Malawi for any unpublished survey data. This included data collected as part of the malaria vaccine pilot implementation programme.
From each of the identified survey reports the minimum required data fields for each record were: date and location, age range information about blood examination (number of individuals tested and number positive for Plasmodium infections by species), the methods used to detect and, the lowest and highest age in the surveyed population (decimal years).

Model description
The spatio-temporal variation in PfPR2-10 was modeled using geostatistical methods to borrow strength of information across time and space.
Let Y i denote the number of individuals that test positive for plasmodium falciparum at survey cluster location x i and time t i And that the survey team went to the sampled clusters given by x i and sampled m i : i = 1….n at time t individuals at risk in the cluster and recorded the outcome of every person that tests positive and negative for plasmodium falciparum malaria.
Then standard geostatistical spatial-temporal model, then assumes that:~( Binomial distribution with m i trials and probability of a positive test P(x i , t i ) specified in the binomial geostatistical model below: where mA and MA are the min and max age among the sampled individuals at location x. TSI represents temperature suitability index at location x and time t. S(x,t) to denote the variation in malaria risk between communities (e.g. variation due to different other environmental conditions) and Z(x,t) the variation within communities (i.e. genetic and behavioural traits). In statistical jargon, S(x,t) and Z(x,t) are so-called random effects that are used in a model in order to capture the effects of unmeasured malaria risk factors. A stationary and isotropic Gaussian process for the spatio-temporal random effects is assumed S(x, t), with an exponential correlation function given as where φ and ψ are scale parameters which regulate the rate of decay of the spatial and temporal correlation for the increasing distance and time separation, respectively; u = ||x − x^'|| is the distance in space between the location of any two communities, one at x and the other at x^'; ν = |t − t^'| is the time separation in years between any two surveys.
The model parameters were estimated via maximum likelihood in the R software environment (version 3.4.1) using logit-transformed prevalence. The targets for the predictions were PfPR2-10 over the 1 x 1 km regular grid surface covering the whole of Malawi. Maps of malaria risk were generated for the years 2000-2022 in QGIS Version 3.2 Uncertainty of the prevalence estimates was addressed using the traditional approach of confidence intervals generated from standard errors of the estimates. Additionally, a more novel approach of Exceedance Probabilities was explored. It has been argued that Exceedance Probabilities, could be an approach that is more relevant to policy makers, than the traditional approach of using confidence intervals. Exceedance Probabilities (E.P.) method sets policy relevant thresholds, that sets the probability that the prevalence is exceeds a policy relevant threshold. The E.P. can be formally expressed as: where t is the prevalence threshold. This approach in this work was only exploratory and do not present it in this paper

Model validation
To test whether there was any evidence against spatial correlation in the data, empirical variogram methods was used. A simulation of 1000 empirical variograms around the fitted model were ran and these were used to compute 95% confidence intervals at any given spatial distance of the variogram.
A conclusion was reached that there is spatial correlation in the data if the empirical variogram obtained from the data fell outside the 95% tolerance bandwidth.

Ethical considerations
This is a secondary data analysis, and therefore exempt from ethical approval. The original study participants from which the data was obtained consented to participate in the surveys. The data used in the analysis was collated such that identity or exact location of the human subjects could not be ascertained directly or through identifiers linked to the subjects. Permission to use the dataset was obtained from the either the investigators, or from the National Malaria Control Programme, through the set data request procedures for the institutions.

Results
A total of 2,595 sampled unique locations from 2000 to 2022 were identified through the data collation exercise. This represents 70,565 individuals that were sampled in the period. The distribution of the sampled locations across the years is shown in Figure 2 below.
The sampled locations were distributed across the entire surface of Malawi as shown in Figure 3 below.
There was a decline in the prevalence of malaria ages 2-10(PfPR2_10) in the sampled locations over the period 2000 to 2022. The period 2000 to 2005 was associated with sampled locations having higher prevalence which was followed by points having a reducing PfPR2_10. The year 2021 has the greatest number of locations with the least PfPR2_10.
In general, the PfPR2_10 was declining over the 22 years. The mean modelled national PfPR2_10 in 2000 was 43.93% (95%CI: 17.92%-72.84) and declined in the subsequent years. Table 1 below shows the national PfPR2_10 estimates with their associated confidence intervals (CI). Figure 4 below shows the national PFPR2_10 estimates and their associated 95% confidence intervals for the years 2000 to 2022. The same estimates are presented in Table 1 below.

Figure 3. Spatial temporal distribution of the sampled locations 2000-2022.
While we show the national estimates above, the main objective of this analysis is to map the malaria prevalence estimates and their associated uncertainty for the whole surface of Malawi at higher resolution (1x1 Km grids and district level) for the period of interest as these estimates are not available with the traditional MIS. The maps in Figure 5 below indicate the modelled PfPR2_10 at a spatial resolution of 1km by 1 km grid. The predictions have been made for the years 2000 to 2022.
The model outputs confirm the heterogeneous nature of malaria transmission in Malawi, with central and southern lake shore areas having higher PfPR2_10 than other parts of the country. Comparison of the high-resolution maps shows the spatial-temporal decline in malaria prevalence especially in Northern parts of Malawi across the years.
Malawi has a decentralized health system, where decision making is at district level. For that reason, we present district  level prevalence estimates for the period 2016 to 2022, that coincides with the most recent strategic plan. These estimates are shown in the map in Figure 6 bellow. There is remarkable decline in PfPR2_10 in 2017 as compared to the previous years. Higher malaria transmission is in the districts on the central southern tip of Lake Malawi over the years, though the prevalence has been decreasing. In 2021, there was an increase in PfPR2_10 in some districts. This include one district (Chitipa) in the north and seven districts in the central region, with prevelence estimates in the 11-20% category. District level PfPR estimates for the period 2000 to 2022 are shown in Table 2.

Model validation
Using variogram based techniques described above, the model above was tested for evidence for spatial corelation.
The results of this process are shown in the Figure 7 below. Since the empirical semi-variogram (solid line) falls within the 95% confidence intervals (grey envelope), then this shows that the model is valid; the model for malaria prevalence is therefore compatible with the data.

Discussion
The past two decades have been characterised by a substantial scale-up of available malaria control tools in Malawi. Historically, the country has been known to be ahead of many other African countries with regard to its malaria policies 23 and is usually among the first to respond to new or available malaria interventions. For example, in 1993, the country was the first in Africa to change its first-line therapy for uncomplicated malaria from chloroquine to sulfadoxine-pyrimethamine (S.P.) 24 . With increasing evidence of a reduced cure rate of S.P., to as low as 82%, the country was again first to change the first-line treatment for uncomplicated malaria from S.P. to artemether-lumefantrine (A.L.) in 2007 for all adults and children over the age of 5 who test positive for malaria using a rapid diagnostic test 23 . Within this period between 2000-2022, there have been four strategic malaria plans that have been implemented 20 . Under these strategies, several key things have happened including the introduction of rapid diagnostic tests that scaled up the testing and treatment of malaria, introduction of artemisinin combination therapies as the first line treatment for uncomplicated malaria, use of indoor residual spraying, mass net distribution of long-lasting insecticide treated nets including PBO nets that were introduced in 2018 and more recently, the roll out of the malaria vaccine. Indeed, the last two decades have been an evolving period for malaria control. Based on what we know and validated by the present analysis, malaria transmission is decreasing and becoming more heterogeneous at subnational levels. There is a need for more robust tools to guide targeted control efforts in the remaining hotspot areas.
The current Malawi malaria strategic plan that started in 2016 ended in December 2022. The strategy 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. Based on the 2019 malaria programme review, the NMCP are on track with their indicators. The subsequent 2023 to 2030 malaria strategic plan is crucial to the elimination of malaria. To directly inform the next strategic plan, we analysed the last 20 years of available prevalence data to understand the changing transmission patterns in the 21-year period between 2000 and 2021. Specifically, we aimed to map the malaria prevalence estimates and their associated uncertainty for the whole surface of Malawi at higher resolution  From the analysis, within this 22-year period, we have demonstrated that malaria transmission in Malawi is becoming more heterogeneous. There are hotspots of high transmission and areas of very low transmission. This is due to varied climatic conditions, vector and parasite resistance, conducive environmental factors in urban and rural areas and, varied intervention uptake in the different parts of the country. The years between 2000 and 2010 were associated with minimal malaria funding available to the country 25 , which is evidenced by the high prevalence of malaria in 2006, based on this present analysis.
In the follow up years, between 2010 and 2015, massive donor investments and local support for malaria control were made available leading to an increase in the coverage of interventions and a decline in prevalence 26 . In the absence of known climate anomalies that could have led to a sudden decline in malaria infection during this period, it can be interpolated that the decline that was observed in this period was due to the expansion of malaria control initiatives in the country 17 .
Understanding and predicting patterns of transmission risk forms an essential component of an effective elimination campaign, allowing limited resources for control and elimination to be targeted cost-effectively. Cognizant of this, WHO recently updated its guidance to view malaria transmission as a continuum within countries, encouraging countries to use surveillance as one of the core interventions and to incorporate malaria early warning systems that can predict outbreaks or unexpected and short-term disease changes to effectively allocate resources 13 .
In the present analysis, we use Temperature Suitability Index (TSI) for malaria. This measure has been used in various fields, including ecology, agriculture, and public health, to estimate the suitability of environmental conditions, in this case temperature, for a specific organism, crop, or disease vector. In the context of malaria, TSI is an indicator of how suitable the ambient temperature is for the transmission and development of the malaria parasite within its mosquito vector. Temperature plays a crucial role in the life cycle and transmission dynamics of malaria parasites and their mosquito vectors. The development rate of the parasite within the mosquito (sporogonic cycle) and the mosquito's life span are both temperature-dependent. There is an optimal temperature range that allows the malaria parasite to complete its development within the mosquito, and for the mosquito to survive long enough to transmit the parasite to a new host.The TSI for malaria transmission takes into account these temperature-dependent relationships and can be used as a covariate in models to predict the spatial and temporal distribution of malaria risk. By incorporating TSI into a model, researchers can better understand how environmental factors, such as temperature, contribute to the observed patterns of malaria transmission and prevalence.
A key strength in the current analysis is that we have leveraged data from multiple surveys, which in turn improves the predictions of prevalence, as opposed to using single survey data which often time may contain data that is sparse for high resolution predictions. So far, in Malawi, efforts to model spatio-temporal heterogeneity in malaria have focused on parasite prevalence (infection) data from household surveys, because of concerns about the quality and completeness of routine clinical malaria case data from the national District Health Information System (DHIS2). National malaria prevalence surveys, however, are costly and are only conducted every 2-4 years and, while parasite prevalence reflects transmission, it does not necessarily align with the disease burden in higher transmission settings. The funding landscape is also becoming increasingly unpredictable. The 2019 Malawi malaria indicator household survey was not conducted due to funding constraints and COVID-19 delays, interfering with national progress tracking. Keen to maximise the use of routine data, the NMCP has made substantial investments in the DHIS2 data since 2016, resulting in steady data quality improvements in terms of timeliness, completion rates and data accuracy of routine data. The 2019 WHO Malawi mid-term review confirmed this achievement that routine case data can be confidently used for surveillance and decision-making. Future descriptions of subnational malaria risk may have to utilise routinely collected data and consider using composite measures. Employing these models in a situation of reliable routinely collected data may be less ideal. Despite this, to date this is the most detailed/ up to date description of malaria prevalence for the last two decades in Malawi.

Conclusions & recommendations
Malaria remains a public health concern, especially in Malawi. The last two decades have been characterized by a scale-up of available control tools leading to a reduction in prevalence.
Decreasing malaria transmission has however contributed to "heterogeneous" transmission landscapes in different parts of countries. The prevalence estimates from the modelling need to be triangulated with routinely collected data. Efforts to control malaria beyond 2022 should focus on targeting of control measures in areas of highest need.