Integrating rapid risk mapping and mobile phone call record data for strategic malaria elimination planning

Significant progress is being made in reducing the morbidity and mortality attributed to malaria globally [1‐10], and the Global Malaria Action Plan (GMAP) [11] articulates a long-term vision for malaria eradication through shorter-term local efforts to eliminate malaria. A total of 36 of the 107 malaria-endemic countries have declared they have a national policy for malaria elimination or are pursuing spatially progressive elimination within their borders [11‐14].

Achieving elimination requires a re-orientation away from the sorts of universal prevention and treatment measures that define a control programme towards targeted operations, such as identifying residual transmission foci, focusing vector control or parasite-based attack measures to high-risk areas, identifying and curing both asymptomatic and symptomatic infections, and managing importation risk [15]. Many of these operational requirements can be facilitated by accurate and timely creation of risk maps. Such maps can help elimination programmes understand the epidemiology of a disappearing disease, and may allow proactive deployment of vector control measures to high risk areas to prevent local transmission and onward spread to other receptive areas, or suggest areas where active case detection may be used to identify and treat remaining parasite reservoirs [16]. Parasite rate-based maps for malaria have now been constructed [17, 18], but infection prevalence is a poor metric for measuring malaria at very low levels of endemicity (below 5% parasite prevalence) due to the large sample size surveys required for precise measurement in such contexts [19]. In very low transmission environments, diagnostically confirmed malaria incidence provides a more useful measure than prevalence, and elimination-focused programmes are building capacity to rapidly provide such information, including the place of residence of cases [20]. Such a surveillance system is a crucial component of an elimination strategy, but achieving and maintaining elimination will require finding and curing infections that may be asymptomatic or may never come into contact with reporting health facilities [15]. Such infections can be identified through intensive proactive surveillance, but the generation of case-based risk maps at high spatial resolution has the potential to remotely identify regions in which transmission is likely to be occurring more quickly and at substantially lower cost.

Risk maps are essential for knowing where to attack malaria, but they are insufficient for a strategic elimination plan. Attacking strategically requires deploying the right measures in the right places, and doing so in a way that gains are not lost due to movement of people and parasites. For example, an attempt to eliminate malaria in Haiti in the 1960s through mass drug administration combined with DDT-spraying failed because the highly mobile population continually reintroduced parasites into areas that had just been cleared [21]. Understanding human movement, which can provide connections between disparate high-risk areas, is critical to designing appropriate elimination strategies and avoiding resurgence in post-elimination settings [22, 23]. However, data on human movement patterns in malaria-endemic regions have been difficult to obtain, and often restricted to local travel history surveys or census-derived migration data [22]. The rapid global proliferation of mobile phones has presented unprecedented opportunities for measuring and understanding human movement dynamics. The retrospective analysis of billions of call detail records (CDRs), whereby temporal sequences of phone tower locations through which user communications were routed are converted into movement trajectories [24‐27], providing information on human travel for sample sizes of millions and at scales of entire countries. Previous studies have demonstrated the value of such data when combined with parasite prevalence maps in providing quantitative guidance to malaria programmes [25, 28, 29], and mapping ‘source’ and ‘sink’ areas of net infection export or import [30]. However, in elimination settings where infection prevalence is an inappropriate measure and where case-based malaria maps are of greater utility, such approaches have yet to be applied.

Here, the potential of integrating mobile phone CDRs with rapid case-based mapping in providing a dynamic evidence base to support malaria elimination planning in low transmission settings is demonstrated, using Namibia as an example. Between 2004 and 2011, scale up of vector control and case management interventions in Namibia contributed to a remarkable decline in reported malaria cases from 610,800 to 14,400 [31]. Namibia is rapidly scaling up its malaria programme, with significant strengthening of its diagnosis and surveillance systems planned over the next five years, focused on achieving elimination by 2020. While the country has a clear strategic plan and recently drafted national elimination policy in place [32], achieving its goals will require a clearly defined strategy to deploy resources to optimal effect. The integration of movement data with case-based risk maps for Namibia provides a dynamic framework for understanding the connectivity between existing and potential malaria risk areas and defining ‘source’ and ‘sink’ regions, where relatively larger numbers of parasites may be exported than imported through travel, and vice-versa. Targeting aggressive attack measures to source communities will reduce malaria both at their locations and throughout the wider region to which it exports parasites. At the same time, sustainable measures to reduce receptivity in sink regions will be important to limit onwards transmission from imported infections.

As many countries reap the success of recent investments in malaria control with reported cases declining significantly, and re-orientate strategies towards elimination, parasite carriage by human travellers is rising up national and global agendas [14, 22, 48, 49]. In elimination settings, the importation of malaria from outside a country becomes the focus of a malaria control programme, but intranational human population and malaria parasite movement is an important part of achieving elimination. Understanding this movement should be a critical component of the design of an elimination strategy, since it enables programmes to target resources in the most efficient way, plan attack strategies and ensure that context-adapted intervention strategies are employed across all high-risk areas. Past difficulties in quantifying and gaining a better understanding of human movement patterns are being overcome through new technologies [24, 50] and here the potential of one of these, mobile phones, is outlined in providing valuable information that can be integrated with rapid case-based malaria risk mapping [16] to guide the design of disease control and elimination strategies.

The analyses presented here illustrate heterogeneities that exist in terms of both malaria risk and mobility across Namibia. The case-based risk mapping results (Figures 1, 2, 3 and 4) reveal the consistency in driving factors of the probability of cases at the spatial scales examined here between the three regions for which data were available (Additional file 1), as well as the accuracy with which risk factors and areas can be distinguished from the lower risk ‘background’ conditions (Additional file 1). Through integrating such high resolution risk mapping with CDRs, the targeting of elimination activities through identifying aspects of risk analogous to both the ‘hotspots’ and ‘hotpops’ concepts [49] could be undertaken if system flexibility and costs of undertaking this allow, enabling the focused deployment of limited resources in an attempt to focus surveillance activities and maximize impact (Figure 13). In planning an attack strategy, thinking spatially and accounting for mobility could be critical – with a mass drug administration (MDA) or mass screen and treat (MSAT) approach, reducing receptivity in high transmission risk sinks could be a focus through encouraging bed net use, while high transmission sources are attacked (Figures 1, 2, 3, 4, 8, 9, 10, 11 and 12). Such an approach will likely be much less costly and operationally difficult than trying to achieve blanket high coverage of MDA/MSAT in all high-risk areas (Figure 13). In post-elimination settings, the framework presented here provides guidance for targeting surveillance by highlighting how areas that are climatically, ecologically and demographically receptive to transmission are connected by human movement (Figures 8 and 9, Additional file 2) and through examining likely sources and onward movements from local outbreaks. It is clear that the exportation of parasites to other locations is not always problematic if the destination is not receptive, and the approaches presented here enable the separation of these ‘dead-end’ movements from possible problematic movements to receptive areas. The design of strategic plans for controlling, eliminating and preventing malaria re-establishment should, therefore, ideally account for human and, in turn, likely parasite movement patterns, and the analyses presented here show that tools built on the integration of datasets that are readily collected and stored by control programmes, satellite operators and mobile phone network providers can provide this valuable information for prioritizing efforts.

Whilst the analyses presented of the connectivity between risk areas in a malaria elimination setting go beyond previous work, it is clear that a range of uncertainties remain. Many of those crossing the border into Namibia will be captured by phone data due to SIM card switching, but clearly one of the biggest drawbacks of such data for mobility analyses is the lack of cross-border movement rate quantification. Infection importations from Angola and other neighbouring countries likely play a role in the epidemiology of malaria in Namibia [48], and if the community detection analyses could include cross-border movements they would likely highlight the north-central regions as being in the same community as south-central Angola and Caprivi joined with its surrounding countries, with many economic and family ties across the border prompting significant movements [51] and collaboration in control being vital if elimination is to be achieved [14, 48]. While phone ownership and usage is high in Namibia, only a certain percentage of the population is being represented by the CDRs used here, and these are partially biased towards specific age groups and the richer and more mobile segments of the country [30, 41] (Additional file 2). Moreover, the demographics and daily activities of network subscribers remain relatively unknown (Additional file 2), with different groups and activities likely presenting significantly greater risks of infection acquisition than others [22, 47, 52]. However, recent analyses on similar data in Kenya suggest that this is not likely to present a substantial bias in mobility estimates [43].

In terms of the risk mapping undertaken, it remains clear that the approach identifies broad areas of suitability for finding cases based on ecological, climatic, physical and demographic indicators, which provides no guarantee of finding ongoing transmission. However, the cross-validation undertaken suggests good performance in terms of identifying areas where cases have occurred (Additional file 1), providing a valuable tool for prioritizing areas for surveillance and further investigation. Ideally, alternative metrics of transmission, such as serological markers [53] should also be incorporated as more stable measures of transmission and to identify asymptomatic infections, thus, better quantifying true hotspots of transmission, but such measures are not yet routinely collected. The utilization of training data from just three districts here, where also spatial differences in treatment seeking rates remain unknown, results in uncertainties in risk predictions elsewhere, though the accuracies in predictions and consistency in variables selected as top predictors across the three districts suggests that the drivers of transmission remain relatively consistent countrywide (Additional file 1). Moreover, broad similarities of the outputs to the most recent surveillance data [31] also suggests accurate mapping prospectively. Assessment of the sensitivity of outputs presented here to variability in quality of surveillance system data should represent an area of future work, however. Ideally, information on the receptivity (the propensity to result in onward transmission following an imported case) of each area should form a valuable additional metric to improve assessments of local transmission risks from case introductions. Pre-control era prevalence data have been used to define this for the 1969-92 period for Namibia [54, 55], but significant development, population growth and urbanization over recent years [42, 56] have likely changed receptivity substantially. Finally, the lack of travel histories in the case data used raises the possibility that some infections were acquired away from their location of residence, though the strong clustering of cases is indicative of local transmission and removal of isolated cases left outputs unchanged (Additional file 1).

The continued upgrade of the Namibia surveillance system, as well as those in other elimination countries, will begin to provide more in-depth information on cases, enabling the separation of likely local versus imported cases, as well as the travel histories of imported cases [57]. These improvements in type, quality and quantity of surveillance data will in turn present opportunities for the application of improved space-time statistical mapping approaches and mathematical transmission models to quantify and account for uncertainties, as well as the estimation of post-elimination risks of resurgence [23]. As data become more regularly reported, a central repository in the form of an online mapping tool is likely to be an important asset for elimination programs [58, 59]. Integrating into such a tool rapid case-based risk mapping that can be dynamically updated as new data are reported, to account for seasonal and interannual variations [16], would provide useful prioritization for further investigations and surveillance activities. The linkage to phone data would then provide valuable information on mobility and connectivity. Further, combining the CDRs with other forms of movement data, such as census, survey and satellite [22, 50, 60], could inform on the demographics, drivers and seasonality of movements, as well as cross-border data, all of which are lacking in phone data. Finally, many of the methods outlined here are not restricted to malaria elimination scenarios, with issues such as artemisinin resistance spread [61, 62], vaccine-preventable childhood illnesses [63], and the elimination of other diseases [64] also reliant on an understanding of movement dynamics.