Background
Many countries are committing to nationwide malaria elimination and global eradication is once more back on the international agenda [
1‐
3]. Historically, the technical feasibility of achieving malaria elimination in a region has been conceptualized as being composed of 'receptivity' and 'vulnerability' [
4,
5]. Receptivity represents the strength of transmission in an area, while vulnerability is the risk of malaria importation [
6]. While both have been regularly discussed theoretically, neither have been quantified, nor methods for their quantification ever defined.
Quantifying imported malaria risk represents a central component for not only assessing the feasibility of malaria elimination from a region, but for planning the implementation of an elimination campaign. Malaria is constantly being exported and imported around the World, and in areas of high transmission, malaria importation is generally a minor concern. As local transmission is reduced and after malaria has been eliminated from a region, however, importation becomes a primary concern.
Zanzibar, an island group of the coast of Tanzania, is one of the territories in sub-Saharan Africa that has recently expressed its willingness to move from control towards elimination. Since 2003, the introduction of artemisinin-based combination therapy (ACT) and high coverages of long-lasting insecticide treated nets and indoor residual spraying, has reduced malaria prevalence to just 0.8% [
7,
8]. These efforts have resulted in the government of Zanzibar considering an elimination campaign and undertaking an elimination feasibility assessment. Nevertheless, proximity and high connectivity to the mainland where transmission levels remain substantially higher in many places [
9] implies that imported malaria will be a constant problem [
10].
In general, parasites can be imported into Zanzibar in one of three ways: (i) the migration of an infected mosquito, (ii) infected humans visiting or migrating from the mainland, (iii) residents visiting the mainland and becoming infected, then returning. While mosquitoes may occasionally arrive though wind-blown or accidental aircraft or ship transport, typically they will only fly short distances. Human carriage of parasites, therefore, represents the principal risk, and is to blame in many past instances elsewhere where malaria has resurged [
11‐
14]. Quantifying such movements both temporally and spatially, and the resulting imported infection risks, represents an important task if effective, evidence-based planning for elimination is to be undertaken.
Recent approaches to quantifying human mobility patterns point the way to novel insights from new data [
15,
16], especially through the analysis of mobile phone records [
17‐
19]. Anonimized phone call record data that has both the time each call was made and the location of the nearest mast that each call was routed through can be used to construct trajectories of the movements of individuals over time [
19]. Here, the potential of such data for estimating importation risk in the malaria elimination feasibility assessment for the islands of Zanzibar is demonstrated. The low market share on the mainland for the network provider restricts the focus here to those infections brought in by residents returning from mainland travel. However, the approaches put forward are sufficiently generic to be applied to alternative regions, elimination settings and phone network provider data. Moreover, this exercise aims to present the first exploration of mobile phone based approaches to the quantification of vulnerability to inform malaria elimination decisions and planning.
Discussion
Results here show that, despite data limitations, spatially and temporally referenced mobile phone usage data can provide valuable information on human movement patterns. In combination with spatial data on malaria endemicity, derived movement patterns can inform on the likely sources, risks and case numbers of imported malaria. The estimates presented represent the first quantification of the vulnerability of an area to imported malaria, a necessary quantity in determining the feasibility of achieving and sustaining elimination.
According to the Zantel data, of the 770,369 users in the entire dataset (made up of Zanzibar and mainland residents), only just over 100,000 travelled anywhere during the three-month study period. Of those Zanzibar residents that travelled, the overwhelming majority went solely to the Dar Es Salaam region (and likely to Dar Es Salaam city itself), where the population weighted average dEIR is relatively low. The majority of these trips were for just one to two days, thus posing a relatively low risk of acquiring an infection and again confirming that most trips could not have involved travel to much further beyond Dar Es Salaam city itself. If malaria prevalence levels continue to fall on the nearby mainland [
37,
38], there is reason to believe that importation risk on Zanzibar will fall simultaneously. There do however exist small mobile groups that (i) travel for extended periods to the mainland from Zanzibar (ii) travel to higher transmission areas from Zanzibar. These represent the risk groups contributing most to the imported infection numbers brought in by residents visiting the mainland. Moreover, basic analyses on mainland resident movement patterns (Additional File
1: supplemental information), suggest that similar risk groups exist among visitors to Zanzibar.
As described in the methods section, the data used here have specific limitations that prevent more comprehensive analysis. With just a 10% share of the market on the mainland and Zantel subscribers more likely to travel to Zanzibar than non-subscribers, detailed analyses were not presented based on visitors from the mainland, since the data probably exhibits significant biases. In addition, the activities of visitors to high transmission areas are unknown - in extreme scenarios, some may sleep under bed nets in air-conditioned hotels, while others may spend the night outdoors. Further, those travelling to or from further afield than Tanzania are not captured by this dataset, nor are those who switch to an alternative network provider on the mainland, nor are trips longer than three months captured. Finally, information on movement patterns on Zanzibar are also lacking, preventing an understanding of the likelihood of onwards transmission, since imported cases may play a key role in sustaining local transmission in some parts of Zanzibar. Previous work has shown however, that many mobile phone companies often have the ability to provide more precise spatial locations on data (e.g[
19]), potentially improving upon conclusions made, should similar malaria-related studies be undertaken. Moreover, additional studies are planned and should be encouraged to test the approaches presented here further and help to arrive at a clear methodology for the quantification of vulnerability. The importance of preserving the anonymity of phone users should remain the utmost priority though.
The information derived from these analyses can be used to guide strategic planning for elimination, should the Ministry of Health decide to pursue such a campaign. Typically, three principal means of reducing imported infection risk are considered: (i) Identify infected individuals and treat them promptly, ideally before or upon entry, before they can infect competent local vectors and lead to secondary cases and sustained foci of indigenous transmission; (ii) address the source of infection by directly reducing transmission in all regions that are primary sources of infected travellers; (iii) provide prophylaxis to residents visiting endemic areas. While the second method is being addressed indirectly through the scaling up of control on the mainland [
37,
38], these analyses provide baseline data to inform on the first and third approaches. Screening with rapid diagnostic tests (RDTs) or microscopy at the ports of entry and providing follow-up treatment of infected individuals may play an important role in reducing imported case numbers and outbreaks. Such an approach is being used for all individuals entering the island of Aneityum in Vanuatu [
39], while visitors from Africa were tested at the airports of Oman during its elimination campaign. Moreover, the details of all visitors to Mauritius from endemic regions are recorded and follow-up is undertaken by health surveillance officers [
40]. When movement rates are high and resources are limited however, as in the case of Zanzibar, screening all visitors at the ports or providing follow-up may be prohibitively expensive and inefficient due to the large number of low-risk trips undertaken (Figure
6).
Modelling work on achieving and maintaining elimination done for the Zanzibar malaria elimination feasibility assessment suggests that as long as effective coverage with vector control measures is higher than 80%, elimination will be achieved and can be maintained. However, once transmission is reduced to very low levels, scaling down prevention without risking resurgence will only be possible if the importation levels estimated here are lowered considerably [Moonen B, Cohen J, Smith DL, Tatem AJ, Sabot O, Msellem M, Le Menach A, Randell H, Bjorkman A, Ali A: Malaria elimination feasibility assessment in Zanzibar I: Technical feasibility. Malar Journal 2009, in preparation]. Prophylaxis for Zanzibari travellers is unlikely to be cost-effective or even practical given the high frequency of travel to mainly low risk regions. Screening on the ferries, especially of high risk groups during high risk periods of the year, might be a simpler and more cost-effective option compared to screening at the port of entry. Passengers are on the slow and fast ferries for six and two hours, respectively; enough time to administer a short questionnaire a rapid diagnostic test and treatment if necessary. However, better data are necessary to determine the Pf PR in ferry travellers to appreciate the operational consequences of such an approach.
Future work will aim to link the findings here to GIS data on travel networks in the region, and build these into stochastic metapopulation models of transmission, providing flexible tools for elimination planning. Moreover, retrospective analyses of health facility records at Zanzibar malaria early epidemic detection system sites are being undertaken at present, while surveys on the ferries are planned to corroborate and compliment findings here. This work also links into and is complemented by other datasets being gathered and analysed as part of a new research agenda initiated by the Malaria Atlas Project [
41] to quantify human movement patterns in relation to assessment of malaria elimination feasibility.
Malaria elimination requires a significant investment of resources and capacity and, as has been demonstrated twice before on Zanzibar, failure to achieve this ambitious target can lead to fatigue among donors and policymakers and subsequent devastating resurgence of malaria. As more countries across the world make progress toward malaria elimination, there is a need for evidence based and locally-tailored assessments of the feasibility of making the final step in initiating an elimination campaign. With mobile phone uptake continuing to grow around the world, this novel data source has the potential to play a key role in providing such valuable evidence. While 'vulnerability' has been discussed in relation to malaria elimination for decades, the approaches outlined here represent a first step towards finally quantifying it. Replicating and refining these approaches in other areas will enable the development of a standardized methodology for malaria importation risk assessment to aid countries that are considering and planning elimination.
Authors' contributions
AJT conceived, designed and implemented the research and wrote the paper. BM, DS, OS and YQ aided with ideas, methodological and editorial input. BM and AA provided support in data compilation. The final version of the manuscript was seen and approved by all authors.