The use of mobile phone data for the estimation of the travel patterns and imported Plasmodium falciparum rates among Zanzibar residents

Abstract

Background: Malaria endemicity in Zanzibar has reached historically low levels, and the epidemiology of malaria transmission is in transition. To capitalize on these gains, Zanzibar has commissioned a feasibility assessment to help inform on whether to move to an elimination campaign. Declining local transmission has refocused attention on imported malaria. Recent studies have shown that anonimized mobile phone records provide a valuable data source for characterizing human movements without compromising the privacy of phone users. Such movement data in combination with spatial data on P. falciparum endemicity provide a way of characterizing the patterns of parasite carrier movements and the rates of malaria importation, which have been used as part of the malaria elimination feasibility assessment for the islands of Zanzibar.

Data and methods: Records encompassing three months of complete mobile phone usage for the period October-December 2008 were obtained from the Zanzibar Telecom (Zantel) mobile phone network company, the principal provider on the islands of Zanzibar. The data included the dates of all phone usage by 770,369 individual anonymous users. Each individual call and message was spatially referenced to one of six areas: Zanzibar and five mainland Tanzania regions. Information on the numbers of Zanzibar residents travelling to the mainland, locations visited and lengths of stay were extracted. Spatial and temporal data on P. falciparum transmission intensity and seasonality enabled linkage of this information to endemicity exposure and, motivated by malaria transmission models, estimates of the expected patterns of parasite importation to be made.

Results: Over the three month period studied, 88% of users made calls that were routed only through masts on Zanzibar, suggesting that no long distance travel was undertaken by this group. Of those who made calls routed through mainland masts the vast majority of trips were estimated to be of less than five days in length, and to the Dar Es Salaam Zantel-defined region. Though this region covered a wide range of transmission intensities, data on total infection numbers in Zanzibar combined with mathematical models enabled informed estimation of transmission exposure and imported infection numbers. These showed that the majority of trips made posed a relatively low risk for parasite importation, but risk groups visiting higher transmission regions for extended periods of time could be identified.

Conclusion: Anonymous mobile phone records provide valuable information on human movement patterns in areas that are typically data-sparse. Estimates of human movement patterns from Zanzibar to mainland Tanzania suggest that imported malaria risk from this group is heterogeneously distributed; a few people account for most of the risk for imported malaria. In combination with spatial data on malaria endemicity and transmission models, movement patterns derived from phone records can inform on the likely sources and rates of malaria importation. Such information is important for assessing the feasibility of malaria elimination and planning an elimination campaign.

Figure 1

Zantel coverage regions in Tanzania.

Figure 2

Total ferry passenger numbers between Dar Es Salaam and Zanzibar for 2007.

Figure 3

Zantel coverage regions for Tanzania overlaid on (a) Daily Entomological Inoculation Rate (dEIR); (b) Population distribution; (c) Month of start of principal malaria transmission season; (d) Last month of principal malaria transmission season. Secondary transmission season maps shown in Additional File 1: supplemental information.

Figure 4

The proportions of Zanzibar resident users that made the majority of their non-home calls at each location.

Figure 5

The distributions of trip lengths made by Zanzibar residents to the mainland, overall and by Zantel region. Note differing y-axis limits.

Figure 6

All trips made by Zanzibar residents plotted by probability of infection acquisition, based on region population weighted mean dEIR (red line) and population weighted principal city mean dEIR (blue line).

Figure 7

Boxplots of trip probabilities of infection acquisition by Zantel region under scenarios of (a) region population weighted mean dEIR; (b) population weighted principal city mean dEIR. The central dark line in each box shows the median value, the box size shows the interquartile range, while the whiskers extend to the most extreme datapoints that are no more than 1.5 times the interquartile range from the box.