Background
In the past 20 years, there has been significant progress in malaria control due to new technologies and increased political and financial commitment globally. Since 2000, 20 countries no longer have endemic malaria [
1], annual global malaria incidence has decreased by 36% and mortality has decreased by 60% [
2,
3]. Currently, nearly half of the world has eliminated malaria [
4] and eradication is envisioned [
5]. The last attempt at eradication failed 50 years ago, with human population movement (HPM) cited as one of the major reasons [
6].
The World Health Organization (WHO) has declared that all countries should aim for malaria elimination as their ultimate goal, regardless of their malaria burden. Subnational elimination is advocated as a preliminary step for large countries when certain areas of the country have interrupted local transmission. As most countries have diverse transmission intensity, elimination would require stratifying a national map by receptivity and transmission characteristics for targeted malaria interventions [
7]. Receptivity is the ability of an ecosystem to transmit malaria, thus determining local transmission intensity. In a non-receptive area, vulnerability (the risk posed by imported infections) would be the major concern [
8]. A receptive area would be further stratified according to whether there was transmission. An area with transmission would be stratified according to whether there was a focus or transmission was diffuse. Areas with diffuse transmission would be stratified according to the degree of transmission, while foci would be stratified by whether there was ongoing, interrupted or no local transmission [
7].
In order to accurately track transmission within and between stratified areas, increasingly granular surveillance is needed [
7]. In areas of very low transmission, surveillance should detect foci at the scale of individual villages or health facility catchment areas. Meanwhile, countries with high transmission usually stratify their subnational areas according to districts or provinces [
9,
10]. Countries require detection and response to intra and international transmission on an increasingly fine scale as countries approach elimination [
11]. Surveillance systems could, therefore, capture HPM of infected migrants as well as HPM of the malaria infected local population to quickly identify potential importation and travel-related hotspots. However, routine collection of such data as part of malaria surveillance is often not done or is limited in scope. A wide variety of alternative data sources and analytical methods have been used to quantify HPM relevant to malaria control and elimination covering different spatial and temporal scales and each with their own strengths and limitations.
This review categorizes and describes studies of HPM patterns relevant for malaria control and elimination, including the sources of data used, as well as methods for combining and analysing different datasets, to quantify varying types of HPM and identify methodological gaps for future investigation.
Discussion
Most studies combined data to provide a more comprehensive picture of HPM patterns [
50]. In addition, some studies demonstrated that different data sources could complement each other [
20,
22,
27]. However, evidence gaps along the spatiotemporal scales need to be filled with future studies.
All studies tracking neighborhood HPM included GPS as a data source, which could track not only daily HPM, but also seasonal HPM [
13‐
15]. Evidence gaps exist for daily HPM beyond the neighborhood (Table
1). Tracking daily HPM is measurement at the finest temporal scale. Routine activities in a participant’s day would usually be confined to a relatively short distance. Sampling on a larger scale (e.g. regional or intra-national) would be expensive [
12] and unlikely to detect significant patterns due to the lack of overlap in participants’ areas of travel and a small sample size. However, detecting daily cross-border movements (routine contiguous international HPM) might be feasible. Despite the frequency and large volume of contiguous international HPM travel, there is a lack of detailed HPM assessment in this aspect, which would have a significant impact on malaria elimination [
8].
All studies capturing neighbourhood HPM used GPS data-loggers and GeoODK [
13‐
15]. Although this captured movement at the finest scale, sampling and analysis were limited by this method of data collection. Studies were limited by a small sample size with non-probability sampling. As GPS data-loggers and GeoODK are highly subject to user error, children were excluded, limiting the representativeness of the population in the neighbourhood. Due to lower indoor accuracy of the devices, participants may be inaccurately classified as indoor or outdoor, thus affecting the accuracy of activity space-plots. Data analysis was mostly limited to descriptive methods, despite the addition of malaria risk maps and travel surveys [
13,
15]. Hast et al
. [
14] used statistical methods which highlighted HPM patterns among demographic groups. However, the lack of significant associations with incident parasitaemia may have been due to the limited power of the study. In high transmission settings, saturation of malaria risk may result in movement patterns not significantly predicting individual risk.
Evidence gaps exist for HPM at the regional scale, with only one study in this category, which measured periodic HPM (Table
1). Regional HPM patterns might be overlooked as using GPS would be unfeasible, yet HPM still needs to be measured at fine spatial resolution units to detect micro-epidemiological changes. Genetic epidemiologic data used to track regional periodic HPM could be analysed to detect whether any seasonality and trends existed, to build a more accurate temporal picture and fill the existing data gap.
To capture regional HPM, Knudson et al
. [
16] used genetic epidemiology to define a malaria transmission unit. This method may be suitable for areas with low to moderate malaria transmission, where hotspots become apparent. In high transmission areas, genetic epidemiology data might instead be used to evaluate interventions by analysing genetic correlates of declining transmission [
45]. In addition to defining a malaria transmission unit, Knudson et al. [
16] used genetic epidemiology to estimate the size of the asymptomatic reservoir and provide information on parasite genetics related to drug resistance and false negatives from rapid diagnostic tests. However, the use of molecular surveillance is still at an early stage. More studies are needed with larger sample sizes in different transmission settings, to decrease bias and explore how genetic and epidemiological data can best be combined to accurately track HPM.
Surveys were a commonly used source of data to capture intra-national and international HPM. Surveys provide important information on reasons for travel and for identifying hotpops (demographic groups with higher-than-average malaria prevalence) [
8,
51]. However, surveys are generally cross-sectional, prone to recall bias, may lack detail and are difficult to conduct on a large-scale, thus they are also prone to sampling bias. Sampling malaria patients from healthcare facilities may be biased by differential access to healthcare [
17,
29], while community surveys may be biased by the lack of working men and visitors who are active acquirers of infection [
18,
22]. Qualitative studies could be used to increase the granularity of survey data in a few ways: they could provide some details lacking in surveys, purposefully sample populations lacking access to healthcare and explore the reasons why, as well as overcome recall bias using diary studies, for example.
Despite the lack of routine HPM data, Guerra et al. [
18] used annual Malaria Indicator Survey [MIS] data, while Cohen et al
. [
29] used survey data from the Swaziland National Malaria Control Programme. Guerra et al
. [
18] identified hotpops and estimated importation and residual transmission. The analysis was limited by the sampling and spatial resolution of the survey. As only residents were sampled, importation rates only considered passive acquirers of infection [returning residents]. As MIS did not record the off-island destinations of travel, it was not possible to map the exact sources of malaria transmission. The case-based risk maps generated by Cohen et al. [
29] and Tatem et al
. [
21] were limited by small sample sizes, as only one year of data was used. Nevertheless, Cohen et al
. [
29] was the only intra-national study that took seasonality into account. Sinha et al
. [
17] mapped sources and sinks, hotpops and HPM patterns from survey and incidence data, which were easily collectable. The analysis was limited by the spatial resolution of the geographic data, which was at the union level instead of the village level.
Most intra-national studies included mobile phone CDRs in their analyses. Unlike surveys, mobile phone CDRs have large sample sizes. However, data is limited by cell phone tower density and sampling may be biased, as subscribers are more likely to be educated, urban and male [
52]. In addition, it cannot directly identify hotpops and cannot track cross-border HPM without tracking handset IDs and combining data from multiple countries. Despite the differences between surveys and mobile phone CDRs, Wesolowski et al
. [
22] concluded that both sources could quantify broad travel patterns, including regional differences. Used together, they could potentially complement each other to form a detailed picture of HPM. In addition to surveys and mobile phone CDRs, Chang et al
. [
19] used genetic epidemiology data. This was the only intra-national study that used genetic epidemiology data. Their genetic mixing index was not biased by incidence underestimation and was used in a transmission setting and geography where commonly used methods could not easily distinguish genetic differentiation. However, it was limited by a small sample size. Unlike Knudson et al
. [
16], only passive surveillance was used, therefore asymptomatic and subclinical infections were not sampled and results may not have been representative of the entire parasite population.
A few studies used prevalence estimates from the Malaria Atlas Project. These were limited by the lack of seasonality in the estimates [
18,
20,
22]. However, incidence data also had limitations. Incidence data from nationally reported cases to health facilities may be biased by accessibility to healthcare and representative of symptomatic cases only. In endemic areas, there may be increased immunity, leading to fewer symptomatic cases. In addition, cases were aggregated per month, which could influence the accuracy of the mathematical model [
20].
Only one study included traffic data. Le Menach et al
. [
24] included ferry traffic data, in an attempt to more accurately capture travel between Zanzibar and the mainland. Despite the inclusion, it could not fully account for informal movements via small fishing boats.
Two studies tracking international HPM included mobile phone CDRs. Tessema et al
. [
25] used mobile phone CDRs to supplement survey and genetic data. This was the only study tracking international HPM that included genetic data. Ruktanonchai et al
. [
27] used mobile phone CDRs to compare how well census-derived migratory data predicted short-term HPM and found that HPM movement patterns were strongly correlated. Therefore, census-derived migratory data was used to predict HPM patterns across Meso-America. However, as the data did not record individual-level risk, inclusion of travel history surveys might have complemented the analysis.
Two studies tracked HPM through air travel. Tatem et al
. [
26] used routine data from nationally reported statistics on imported malaria. However, the data likely represented only one-sixth of all imported cases globally. In addition, due to global differences in health systems, there would be heterogeneities in case reporting. Mixed species infections might have obscured certain species and data was pooled across a decade, in order to have sufficient data. Nevertheless, the meta-analysis detected a small number of high-traffic routes that accounted for 56% of imported malaria to non-endemic countries and the occurrence of strong spatial clustering of Plasmodium species, which could inform global malaria policy. Huang et al
. [
28] tracked passenger flows weighted by malaria prevalence and highlighted risk routes for artemisinin resistance spread in Southeast Asia using Malaria Atlas Project prevalence maps and flight schedules. However, this was limited by the lack of data on individual-level risk and the reliance on travel data when examining artemisinin resistance, thus heterogeneity in resistance throughout the region was not accounted for. Amongst studies tracking international HPM, only Saita et al
. [
30] accounted for seasonality.
All studies tracking migration used census data [
31‐
33]. However, census data did not provide fine-scale HPM data and may have missed highly mobile populations that could contribute to malaria transmission.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.