Evidence and future potential of mobile phone data for disease disaster management

Highlights

• •Actively and passively produced mobile phone data can strengthen all phases of disaster management.
• •Call detail records (CDR) and two-way SMS can increase efficiencies and improve data availability.
• •Representativeness, data quality, and legal/ethical considerations require further attention.
• •Develops a typology of mobile data sources, types, and end-uses, and framework to assist mobile data use decisions.
• •Results have relevance for the wider disaster management context.

Abstract

Global health threats such as the recent Ebola and Zika virus outbreaks require rapid and robust responses to prevent, reduce and recover from disease dispersion. As part of broader big data and digital humanitarianism discourses, there is an emerging interest in data produced through mobile phone communications for enhancing the data environment in such circumstances. This paper assembles user perspectives and critically examines existing evidence and future potential of mobile phone data derived from call detail records (CDRs) and two-way short message service (SMS) platforms, for managing and responding to humanitarian disasters caused by communicable disease outbreaks. We undertake a scoping review of relevant literature and in-depth interviews with key informants to ascertain the: (i) information that can be gathered from CDRs or SMS data; (ii) phase(s) in the disease disaster management cycle when mobile data may be useful; (iii) value added over conventional approaches to data collection and transfer; (iv) barriers and enablers to use of mobile data in disaster contexts; and (v) the social and ethical challenges. Based on this evidence we develop a typology of mobile phone data sources, types, and end-uses, and a decision-tree for mobile data use, designed to enable effective use of mobile data for disease disaster management. We show that mobile data holds great potential for improving the quality, quantity and timing of selected information required for disaster management, but that testing and evaluation of the benefits, constraints and limitations of mobile data use in a wider range of mobile-user and disaster contexts is needed to fully understand its utility, validity, and limitations.

1. Introduction

Novel sources of data coupled with new data mining techniques and data-driven logics are transforming science, business, governance and society more generally. In this age of so-called big data, there is potential for transformative change in the role of information in decision-making as we move from a mode of decision-making defined by data scarcity to a new era of data abundance (Hey et al., 2009, Miller, 2010, Miller and Goodchild, 2015). The potential of new sources of data to address global data inequities holds particular promise in low- and middle-income countries (LMIC) where conventional sources of social, environmental, and economic data are often patchy, many years out of date, or simply non-existent (Center for Global Development, 2014, Cinnamon and Schuurman, 2013, Deville et al., 2014). Kitchin (2013) describes three approaches to the way big data are produced: directed – in which a human operator focuses a data capturing technology on a person or place (e.g. surveillance camera, remote sensing); automated – in which data are passively collected via the normal operation of a system or technology (e.g. mobile phone use, Web browsing, credit card transactions); and volunteered – data which are actively or passively produced by citizens, typically via user-generated platforms including social media and crowdsourcing applications. These active and passive approaches to big data production can rapidly produce new sources of data in real or near-real time, which opens up a range of opportunities for understanding diverse social phenomena, especially for ‘data poor’ settings characterized by inadequate data infrastructures. The potential of big data to improve disaster response and management is stimulating significant interest from researchers, government, and the humanitarian community (e.g. Fadiya et al., 2014, Pu and Kitsuregawa, 2013, Shelton et al., 2014, Zoomers et al., 2016). This paper critically examines the existing evidence and explores the future potential for the use of actively and passively produced mobile phone data for managing humanitarian disasters caused by communicable disease outbreaks.

A disaster can be defined as a “serious disruption of the functioning of a community or a society involving widespread human, material, economic or environmental losses and impacts, which exceeds the ability of the affected community or society to cope using its own resources” (UN International Strategy for Disaster Reduction, 2009, p. 9). Although attention and resources are concentrated on responding to disasters, the disaster management cycle (Fig. 1 ) is commonly conceptualized as having four overlapping and interconnected phases focusing on risk management (prevention and preparedness phases) and crisis management (response and recovery phases) (Alexander, 2002, p. 6). Together these phases represent the “sum total of all activities, programs and measures which can be taken up before, during and after a disaster with the purpose to avoid a disaster, reduce its impact or recover from its losses” (Vasilescu et al., 2008, p. 46).

Fig. 1.

Phases of the disaster management cycle.

Rapid evidence-based response to events such as natural disasters, disease outbreaks, and political emergencies is essential to minimize losses and damage, and build community resilience. To decide how to distribute emergency resources in the immediate aftermath of a disaster, relief agencies need access to information about the magnitude of the event, the locations where people have been impacted, population characteristics and dynamics, and existing distribution of relief resources and infrastructure (MapAction, 2016). Data inadequacies can result in excessively slow, ineffective, and in some cases negligent disaster response and recovery as illustrated by the widely criticized response to Hurricane Katrina (Thompson et al., 2006). Poor decision-making during the response and recovery phases can be amplified by inadequate knowledge about vulnerabilities during the prevention and preparedness phases of the disaster management cycle (Pu and Kitsuregawa, 2013), highlighting the importance of access to data through all phases of disaster management.

A range of data types can support disaster management. Some directed sources of big data such as remotely-sensed satellite imagery are already relatively widely used in disaster management (Voigt et al., 2007). Use of other direct data sourcing technologies, notably unmanned aerial vehicles (UAV), is not yet standard practice but hold great promise for rapid access to high spatial and temporal resolution imagery from before and after natural disasters (Griffin, 2014) and for enhancing situational awareness in humanitarian emergencies (Aubrecht et al., 2015). Similarly, the use of volunteered sources of data in disaster management has been widely reported in recent years as part of the field of ‘digital humanitarianism’ (Burns, 2014, Meier, 2015), particularly the passive harvesting of spatial information about crisis events from social media including geotagged tweets (Shelton et al., 2014, Zook et al., 2010), and active production of data using Web-based crowdsourcing platforms (Goodchild and Glennon, 2010, Roche et al., 2013). These technologies are transforming communications during emergencies, however the representativeness, credibility, social and ethical consequences, and overall value of the data for decision-making during disasters is not well understood (Buscher et al., 2014, Crawford and Finn, 2015, Shanley et al., 2013).

Comparatively, actively and passively produced big data generated through mobile phone communications have received considerably less attention from within the academic and humanitarian communities and is an underused data source in disaster management (Madianou et al., 2015, The Economist, 2014), despite the rapid global diffusion of mobile phones in recent years. Mobile networks now cover 50% of the globe, supplying almost 8 billion mobile phone connections, and over half of the world’s population is estimated to have a mobile subscription (GSMA, 2015). Although worldwide penetration of mobile phone usage is uneven – loosely paralleling global patterns of development – access to mobile phones is one metric for which many LMIC are rapidly catching up with the rest of the world. In sub-Saharan Africa, 50% of the adult population are expected to have mobile phones by 2020, with some countries far surpassing this already, including Nigeria (89%), Senegal (84%), and Ghana (82%) where subscription rates are similar to many high-income countries, although smartphones are still relatively rare in this region (Pew Research Center, 2015). These data assume that each mobile phone owner is independent while in reality some or even many individuals may be owners of two or more mobiles, meaning coverage could be more limited than it appears. Nonetheless, there is no doubt that the number of mobile phones in use in LMIC is increasing. This technological leapfrogging to digital communication methods in LMIC has been driven by its use for a range of purposes beyond personal communications, including mobile money (digital bank accounts and financial services) which can be accessed via short message service (SMS) on basic mobile phones.

In this paper, we critically examine the current and future potential of two types of data produced via mobile phone communications: passively produced call detail records (CDRs), and active produced data through SMS. Mobile communications produce massive longitudinal datasets recorded and maintained by mobile network operators (MNO). These CDRs typically include the time a communication was made, a unique identifier for the caller, receiver’s telephone number, call duration, size of data transmitted, and the geographic location of the cellular tower the call was routed through and received from for every communication (call or SMS) made by every mobile phone user (FCS, 2012). CDRs are routinely collected by MNOs to facilitate customer billing, problem diagnostics, and network planning, and they have a massive potential to illuminate the spatio-temporal dynamics of individuals and populations at a very high resolution in near real-time (Deville et al., 2014). Researchers are starting to tap into these data sources to understand population characteristics (Douglass et al., 2015), transportation and mobility (Doyle et al., 2014), socio-spatial behaviours and interactions (Gao et al., 2013, Järv et al., 2014), urban spatio-temporal dynamics (Ahas et al., 2015) and inferred aggregate economic activity (Scepanovic et al., 2015). CDRs are thus poised to significantly advance knowledge in these areas, especially in domains that have conventionally relied on out-dated, unrepresentative, or low-resolution data sources (e.g. national census, surveys, modelled estimates).

Mobile phones can also be used to communicate and gather data directly from citizens via mobile apps and SMS. We focus on SMS, as this is available to all types of mobile phone – basic, feature, and smart – and may therefore be the most appropriate starting point for mobile-based disaster management in LMIC. SMS technology has been available since the 1990s and has been widely used as a medium for sharing information with the public during emergencies, however it is only recently that it has also seen significant use for actively collecting timely disaster information. Interactive two-way SMS communication and data collection is a survey method that works by sending questions to all or a targeted selection of mobile phone subscribers. The receiving database can be configured to trigger further questions depending on the user’s response. Two-way SMS is a reliable, acceptable, and low-cost tool for collecting data for research purposes (Whitford et al., 2012) and is a promising platform for actively engaging mobile phone users to volunteer spatio-temporal data in disaster contexts, including information on personal vulnerability and resilience, local impacts, and resource requirements, while also helping to build trust and encourage local resilience (McDonald, 2016).

These sources of “Big Mobile Data” (Ahas et al., 2014, p. 5) could be particularly effective in global health emergencies, such as communicable disease outbreaks, because of the importance of tracking population movements (Zhong and Bian, 2016) and rapidly responding to needs in this context, e.g. to minimize contact between infected and uninfected populations. However, there is a lack of clear evidence on benefits, challenges and limitations of mobile data to guide its effective and appropriate use in health emergencies and disaster contexts more generally. As relatively new data sources, this gap raises the risk of information misuse if the limitations of mobile data sources are not well-understood, and of missed opportunities to improve disaster management if the advantages of these data are poorly recognized. In this paper we seek to help address this knowledge gap by assessing the opportunities, limitations, constraints, and enablers to using mobile data in disaster contexts. While there are only a small number of early exploratory examples of CDR or SMS data use in disasters (e.g. Bengtsson et al., 2011, Yang et al., 2009), the 2013–2015 Ebola outbreak and humanitarian crisis in West Africa stimulated significant interest and debate in academic, policy, and humanitarian circles around the use of mobile phone data in disease disaster management (e.g. The Economist, 2014, The Guardian, 2014, Vayena et al., 2015, Wesolowski et al., 2014a), against the backdrop of a “growing narrative that the problem in the response effort was a lack of good information technology and, more specifically, data” (McDonald, 2016, p. 3). We systematically review the growing body of literature from previous use cases, and conduct semi-structured interviews with key informants with hands-on experience of humanitarian response, to develop a knowledge base and explore the utility, value, and challenges of CDR and SMS data for responding to and managing humanitarian disasters. These empirical findings are then used to develop a typology for mobile data in disease disaster contexts and a decision-tree that could be used to assist data use decisions. This study aims to advance discussion around the use of these data, by assembling and critically engaging with the issues that should be considered by researchers, humanitarian organizations, and policy makers who wish to make use of SMS or CDR data for disaster management. Although the focus is on disease disasters, the findings are likely to have relevance for a diverse range of disaster and humanitarian contexts.

2. Mobile phone data and communicable disease disasters: assembling the evidence

We developed a two-step methodology to assemble evidence and gather perspectives regarding mobile data use in disease disaster management. First, we followed Arksey and O’Malley’s (2005) widely used scoping review approach, which is a type of systematically-conducted literature review that aims “to map rapidly key concepts underpinning a research area and the main sources and types of evidence available” (Arksey and O’Malley, 2005, p. 21). A literature search strategy was devised which focused on academic, grey literature (e.g. organizational reports and other works not distributed by commercial/academic publishers), and media sources published between 1 January 2009 and 15 May 2015 to capture the rapid expansion of evidence and debate emanating from the Ebola crisis and the small amount of relevant literature published prior to this event. The academic search databases used were Google Scholar, Science Direct and PubMed. Media sources were identified using Nexis UK’s ‘Major World Publications’ and Google News. Google Web search engine was used to identify any grey literature including organizational reports and websites. Based on an initial scan of the literature, the research team decided to focus on extracting information relevant to five main thematic areas:

• 1.Information that can be gathered from CDR or SMS data.
• 2.Phase(s) in the disease disaster management cycle when mobile data may be useful.
• 3.Value added over previous/conventional approaches to data collection and transfer.
• 4.Barriers and enablers to use of mobile data in disaster contexts.
• 5.Social and ethical challenges relating to mobile data use.

Search terms used were Ebola, disease, health, emergency, disaster, big data, mobile phone, cell phone, SMS, call detail records, text message, and information needs. Based on these search terms, a total of 48 English language pieces were included in the review. Of these, 21 were from peer-reviewed journals (empirical studies and overview pieces), 20 were media sources (traditional and new media articles), and 7 were grey and organizational literature (e.g. reports, blog posts). A majority (25) of the pieces were explicitly or primarily about the use of mobile data to combat the 2013–15 Ebola crisis.

Next, we conducted ten in-depth, semi-structured interviews with key informants from humanitarian agencies, international aid organizations, mobile data collection service providers, and mobile data focused research teams. The respondents were selected due to their direct experience of using mobile phone data in disaster or disease management contexts, or due to their experience in information management in disease-related disaster situations. The selection involved our initial professional networks, followed by snowball sampling to purposively seek highly informed and relevant international actors in this area. The interview questions were loosely structured around the same five themes used for the review, but kept open-ended to encourage participants to share all information they felt was relevant.

3. Results: call detail records and SMS data for disease disaster management

This section presents the findings of the scoping review and interviews in thematic order.

3.1. Information that can be gathered from CDR or SMS data

CDR datasets contain information on the location and time at which a communication (call/SMS) is made along with unique identifiers for the sender and receiver. These data may be valuable for disaster management when they are used to estimate population size and density in a region or city, based on the number of phone subscribers in the coverage area of each cellular tower (Bharti et al., 2015). Information on population distribution before and after a disaster can be useful when assessing exposure risk and response needs. In a health emergency, the high spatial and temporal resolution data on population numbers inferred through CDRs could assist in identifying potential pathways of disease transmission, populations and locations at risk of disease outbreak, population-level access to and need for medical and social services, and effective allocation of humanitarian aid (Bharti et al., 2015).

Significant recent attention has focused primarily on the use of CDRs to assess spatial patterns of human travel and for tracking population mobility over time (Tatem, 2014, Tatem et al., 2014, Wesolowski et al., 2014a, Wesolowski et al., 2014b). For diseases like Ebola which are spread by person to person contact, this use of CDRs can be valuable for estimating population migration from areas of high infection to other areas, and may indicate important routes of travel (Wesolowski et al., 2014b). These population movements are derived by tracking the geographic location (i.e. based on the nearest cellular tower when a communication is made by the mobile phone user) of individual mobile numbers over a period of time, which provides information on the number of mobile users that remain in an area, or leave, and their direction of travel. Tatem et al. (2014) describe how population movements between ‘source’ and ‘sink’ (net exporter and importer of diseases) communities inferred from CDRs can be combined with disease case data to create maps of disease risk. Population location information gathered via CDRs can further be used to assess compliance with travel restrictions in disease-affected areas, or government closure orders of facilities (Oliver, 2013). When unique identifiers are included in CDR datasets, it may even be possible to assess individual-level mobility. For example, Tatem et al. (2014) describe how a ‘metric of cumulative risk’ can be calculated for individuals, based on the total amount of time spent in a disease endemic region.

SMS-based tools are, meanwhile, enabling humanitarian agencies to collect specific data on population needs when mobile phone users respond to questions sent via text message. The literature describes the use of SMS platforms for collecting information from the public and from health professionals, such as for reporting of potential disease symptoms or supply requirements (Asiimwe et al., 2011). For example, Tracey et al. (2015) describe EbolaTracks, an SMS based platform for active self-monitoring of persons who recently visited an Ebola outbreak area. The person is provided with a self-monitoring kit including thermometer, and a mobile phone with SMS capability. Twice daily over the next 21 days (Ebola incubation period), health personnel send text messages to the persons involved in the programme, instructing them to report symptoms and temperature. Any abnormal reports alert a medical officer, who proceeds to follow up the potential case and initiate an isolation procedure if required.

During a health emergency, two way SMS can also work as both a triage tool, and as a crowdsourced data collection platform for collating information from the public. Several open source two-way SMS data platforms exist, including Frontline SMS (http://www.frontlinesms.com/), and RapidPro SMS (https://community.rapidpro.io/), facilitating data collection and analysis. Trad et al. (2015) describe a proposed SMS platform that would both direct suspected Ebola cases to nearby health facilities, as well as collect information from the public about disease cases. SMS is also being used to collect other vital information needed to effectively respond to health emergencies, including availability of resources in an area. In 2014 amidst the Ebola outbreak, the World Food Programme set up a monthly SMS-based survey for all phone users in Sierra Leone, Guinea and Liberia, asking people to report food shortages and prices in their area (Anderson, 2014). The RapidPro open-source SMS system was deployed in Sierra Leone to quickly gather information on Ebola prevention supplies in schools, including washing kits and thermometers, enabling rapid response to those schools suffering shortages (Mushayi, 2015).

3.2. Phases in the disease disaster management cycle when mobile data may be useful

Fig. 2 presents a typology for describing the source, type and primary end-use of passively and actively produced mobile phone data during different phases in the disease disaster management cycle, based on information emerging from our analysis. Note that smartphone app and interactive voice response (IVR) data are not discussed here but are likely to have similar characteristics as two-way SMS data.

Fig. 2.

Typology of mobile data sources, types and uses in the disease disaster management cycle.

3.3. Value added over previous/conventional approaches to data collection and transfer

The value added of mobile data over conventional sources and methods of data collection for disaster management relate to the potential for improved data availability and accuracy, more rapid detection/response, and increased efficiency. In many LMIC in particular, mobile data could provide information that are unobtainable, incomplete, or out of date from other sources, pointing to the possibility for these data sources to fundamentally disrupt conventional approaches to disaster decision-making in settings defined by data scarcity. Census data on population size and mobility, if they exist, do not account for short term movements or visits, the category of mobility of most interest in communicable disease disaster management (Hern, 2014). The rapid growth of mobile phones in low- and middle-income countries, and the CDRs produced through their use, provide access to population figures and mobility information not available from other sources (Wesolowski et al., 2014a). CDR data are typically of high spatial and temporal resolution and broad in geographic scope, covering the operator’s network area which is often the entire country, which represents a huge value added over other population datasets derived from the census or surveys (Hern, 2014). Disaster response teams often rely on data from the International Organisation for Migration (IOM) to track population movements, which are collected by IOM personnel counting the number of people moving through official border posts. This provides a reliable source of data to humanitarian agencies – “We are very trusting of IOM methods” (Interview participant, NGO employee, 28 April 2015) – but these data are only available when IOM are in-country, are labour and time intensive to collect, and do not account for unofficial border crossings or within-country movements. In contrast, CDR data can be used for tracking within country population movements, are continuously collected, can be made available very rapidly once the right agreements are in place, and have little to no on-going data collection costs, since the data are generated automatically (Lydersen, 2014, Tatem et al., 2009), which opens up new opportunities for information generation in an emergency. With respect to the use of CDR data for understanding the spatio-temporal dynamics of populations, one participant noted “[w]e’re facing the mapping part on better more rapidly reported data” (Interview participant, academic researcher, 10 November 2015). In short, the key value added of CDR data over census or survey approaches is the potential to access current and comprehensive evidence on population size, density, and dynamics, information that is fundamentally necessary for managing any humanitarian emergency or disease-related disaster but which is often unavailable or insufficient in many settings, and especially so in most LMIC.

Rapid decision-making based on evidence is difficult without suitable data sources and adequate quality outputs. Mushayi (2015) describes how an SMS system rapidly deployed in the Ebola response in Sierra Leone enabled discussions on resource decisions to be based on evidence rather than anecdotes or guesswork. The availability of SMS data could stimulate an acceleration of knowledge and the exchange of time sensitive information such as resource needs or new disease cases during disease emergencies, which could have a positive influence on efficiency in terms of costs and resource usage (Revere et al., 2014). As one interview participant noted, “the speed and ease with which you could get some data at the country level with [SMS] would be hard to do in any other way” (mobile data platform service provider, 11 August 2015). In the aftermath of the 2008 earthquake in the Sichuan area of China, an SMS based infectious disease reporting system was developed and tested within 3 days, and was in use one week later across the region (Yang et al., 2009). Speaking about the value added of SMS over face-to-face data collection approaches, a participant who provided mobile data services highlighted the broad geographic reach and potential representativeness that could be achieved with SMS methods: “people weren’t going to be satisfied with a few focus group discussions around key parts of the response, they wanted more “robust” data. So to get that we needed data at some kind of scale, and mobile phones were the only way we could have done that” (11 August 2015). Many data collection activities in low-income countries typically use paper-based records, such as those kept by community health centres. For this, paper-based data need to be made physically accessible to the data collector, which can be a huge barrier to time sensitive decision-making in an emergency situation. Interview participants with experience in this type of data collection stressed the high value of SMS based data transfer for substantially increasing efficiencies, facilitating multi-level communication, reducing data transfer errors (even though human errors in data entry may be the same), and facilitating standardisation of data records. “I think the first thing is that your operational complexity goes down by an order of magnitude as soon as you use a mobile data collection as compared to a face to face data collection” (Interview participant, mobile data platform service provider, 26 August 2015). In short, SMS based methods can enable much faster and more widespread communication and collection of time sensitive information during emergency situations. This could accelerate the identification and isolation of potential communicable disease cases, reduce the burden on health care personnel through patient self triage, and encourage more appropriate allocation of supplies, resources and health personnel compared to conventional methods.

3.4. Barriers and enablers to use of mobile data in disaster contexts

Researchers and organizations should be aware of a range of issues that may prevent or reduce the use of CDR and SMS data in disease related disaster management. These relate primarily to data access, survey design limitations, data processing, difficultly of collecting qualitative or contextual information, problems with obtaining representative population samples and uncertain data quality and validity.

Perhaps the most significant barrier to the use of CDR data is the fundamental and highly complex challenge of data access. Government regulators have shown limited political will or interest in developing protocols and regulation for accessing CDRs (Shacklett, 2015, The Economist, 2014). Gaining access to CDR data is especially challenging in the context of tracking international migration – which is of significant interest in global health crises – because of the need to liaise with multiple mobile operators and national regulators, and users switching SIM cards as they cross borders. CDR access regulations are not internationally standardized (de Montjoye et al., 2014); as one participant described, “the regulations are changing month by month, week by week sometimes, in terms of what needs to be done to meet the national regulations on how these kind of data can be processed and accessed” (interview participant, academic researcher, 10 November 2015). There is a lack of clarity around data protection laws within countries, and laws vary greatly between countries; during the height of the Ebola outbreak centred on Guinea, Sierra Leone, and Liberia, MNOs in the first two countries released CDRs while more restrictive data protection prevented their use in Liberia (McDonald, 2016). Even if data protection regulations enable access to CDRs, MNOs may be unwilling to release these data to third parties (Lydersen, 2014, WorldPop, 2014) because of remaining concerns about customer confidentiality as well as the disclosure of proprietary information regarding their database and system design (Wesolowski et al., 2014a). CDR data access challenges were voiced by several interview participants: “Probably what most of us spend a huge amount of our time doing is talking to operators and health and national regulators, as well as operators at the group level and the country level. Data access is probably the biggest challenge at the moment.” (Interview participant, academic researcher, 10 November 2015). These challenges include getting access to the geographic information (cell tower location) essential for spatio-temporal analysis and mapping of CDR data, which is not necessarily included as standard practice by mobile operator databases: “We’ve tried to work with MNOs before on geographic info and have never had any success… We always get CDRs but I have never seen a CDR with geographic info.” (Interview participant, mobile data platform service provider, 26 August 2015). Buckee et al. (2013) describe how operators release CDRs in different formats, meaning tools have to be modified for each release. The same authors note that population mobility patterns are culture-specific, so CDR based population models and tools developed in Western settings may not apply to the mobility patterns of LMIC, such as those of seasonal pastoralists, refugees and migrant workers.

This points to one of the most significant potential enablers to mobile data use: improved protocols and guidelines for data access, processing, and sharing. The Groupe Speciale Mobile Association (GSMA) has developed guidelines for the appropriate use of CDR and SMS data in emergency situations (GSMA, 2013, GSMA, 2014), which outline among other things, best practices in terms of stakeholder agreements, access protocols, and data security measures. These initial guidelines are helping to clarify the benefits and risks of mobile data for all stakeholders, and further refinement and detail in future iterations should help to accelerate their use in emergency contexts. The mounting media attention given to CDR and its value in disaster contexts may also be helping to overcome data access barriers. For example, discussions on the social value of CDRs during emergency situations may encourage MNOs to share data and collaborate with stakeholders as part of corporate social responsibility programs. As Vayena et al. (2015) describe, health is a common good, so making CDRs available for responding to and managing health emergencies may be more acceptable to the public, governments, and MNOs than for other uses, however the issue of purpose limitation is a concern. Stakeholders are likely to be more willing to release CDRs in disease and emergency management, however they may have reservations if they feel unable to control any subsequent uses of the data.

Data access challenges are further being reduced by the growing market for mobile data service providers that facilitate data access for the NGOs and humanitarian organizations that have the funds to secure their services. There are now a number of companies that have agreements in place with MNOs in various countries and can act as intermediaries between the network and those wishing to access it to obtain CDR records or for SMS based communications. Organizations that wish to access mobile networks for these purposes can do so rapidly without the need to set up bespoke agreements with operators, and the intermediaries can also provide services to assist the organization with setting up effective surveys and messaging. However, even once access to CDRs has been agreed, there are still significant barriers to use due to the time and cost associated with data cleaning and preparation (Jacobsen, 2014). Analyzing and managing massive and complex CDR datasets requires advanced technologies and data science expertise (Asokan and Asokan, 2015, Young, 2015). Undertaking sophisticated analyses without the requisite analytical skills or domain knowledge can lead to spurious conclusions (Macharia, 2015) and deter agencies without in-house expertise from attempting to use CDRs. Even if CDRs are high quality and skills and technologies are available, decision-making models and outputs are only as good as the other datasets used, such as disease case locations which are often patchy in many settings (Bengtsson et al., 2015).

Concern over data quality and representativeness is another significant barrier to the effective use of CDRs. Although penetration has rapidly increased in many LMIC, some countries have comparatively lower rates of ownership including those affected by Ebola in West Africa (Lydersen, 2014). With phone ownership, there tends to be a bias towards the more wealthy and mobile, meaning the results may not necessarily reflect the wider population’s movements (Tatem et al., 2014), and may in fact be an overestimate (see also Xu et al., 2016, Zhao et al., 2016). Also, rural areas may not be covered or may have very large tower catchment areas, meaning the spatial resolution will be poor (Lydersen, 2014, Shacklett, 2015). However, one interview participant expressed the view that coverage should not be viewed as a major constraint with the ease at which rapid assembly mobile towers can be set up, which were successfully used in Sierra Leone by UNICEF to improve network coverage. For those communicable diseases spread by person to person contact such as Ebola, the lack of representativeness outlined above is perhaps less problematic as these diseases tend to cluster in dense populations (where penetration is typically higher), due to the mode of transmission (Buckee et al., 2013). Further, it is expected that mobile ownership and rural penetration will continue to rise significantly in the coming years, meaning demographic and geographic representativeness should steadily improve (Wesolowski et al., 2014a).

SIM card practices in LMIC are an added barrier to data quality (Wesolowski et al., 2014b). It is not uncommon in LMIC for individuals to have several SIM cards and use them at different times of day or location, complicating the tracking of multiple SIM users over time and between locations. Phone-sharing practices between friends and family members adds another level of complexity. As one interview participant remarked, “[w]e know that all we’re tracking is SIM cards and we don’t really know whose they are or how representative they are of the population and then it’s another step beyond that to link them to disease and infections moving around” (Interview participant, academic researcher, 10 November 2015).

In comparison to CDR, use of SMS data is currently more common in the humanitarian sector, although there remain a number of barriers to their use. Similar to CDRs, access is a challenge when trying to set up SMS data collection platforms. Telecommunication infrastructure requirements for SMS based data collection can prevent access to these data in time sensitive emergency situations: “I think the issue with the emergency situations is that it is very difficult to deal with telcos [telephone companies] to set up the infrastructure, and so in an emergency situation you need to move quick, and basically if you don’t have it set up already, you probably have no hope of getting it set up” (Interview participant, mobile data platform service provider, 26 August 2015). Here, infrastructure refers to a centralized data Center set up to receive or send data through mobile line connections, and the software that enables this information to be managed. While it is not essential to collaborate with MNOs to set up SMS data collection systems, this participant explained that doing so has many advantages, not the least the ability to access the phone numbers of potential SMS survey participants in the region of interest. Indeed, humanitarian use of the networks to broadcast information and gather data without usage costs requires agreements between the mobile operator, governments and humanitarian agencies, which may be time consuming (O’Donovan and Bersin, 2015).

SMS are limited to 160 characters per message and relying on simple or truncated communications can have significant limitations for both communicating and gathering vital and time sensitive health information (Revere et al., 2014). This points to a wider issue in that SMS based data collection requires a relatively simple survey design and this may pose a challenge for users accustomed to more sophisticated survey designs. “People are really in a mindset of, like, I need to ask someone 200 questions or it’s not worth it. And you know, 200 questions is pretty hard to do over the phone.” (Interview participant, mobile data platform service provider, 26 August 2015). SMS is best for collecting basic closed-answer questions, for instance using a Likert scale or binary yes/no structure. As one participant described “open questions are really difficult to do, we have done them in the past, and the quality really, really ranges … from absolutely useless to really in depth” (interview participant, mobile data platform provider, 11 August 2015). In depth qualitative information is thus challenging to acquire via SMS, and can lead to the collector receiving an incomplete picture or failing to interact with a participant in a sufficiently sensitive manner. Subtle information provided by a respondent’s tone of voice or body language is not captured in an SMS survey, while this can be systematically recorded during face-to-face interactions with individuals, or questions that allow open-ended responses. In addition several participants described how the presence of multiple spoken languages as well as low rates of literacy and numeracy in some settings can inhibit the potential of SMS data collection and should be carefully considered before conducting SMS-based surveys.

The increasing recognition of crowdsourcing and citizen science in society could be seen as an enabler of SMS based data collection (Gilpin, 2014), however this also presents a major barrier to the effective use of these data for decision-making. Relying on often-anonymous members of the public or health workers to provide reports on personal disease symptoms, potential disease cases, or resource shortages may produce data of unknown quality and veracity, which presents a significant limitation of SMS data. Speaking about the anonymity of participants in a distributed SMS-based survey, one interviewee questioned, “how robust is that data, how much faith and trust can you put in it?” (Interview participant, mobile data platform service provider, 11 August 2015). Data produced by members of the public may not only be incorrect, it may also be biased which could negatively affect or put other members of the public at risk (Tracey et al., 2015). One interview participant explained that during a phone-based campaign to identify Ebola cases in Sierra-Leone, many SMS messages were received from concerned people seeking information rather than from people with Ebola symptoms: “the problem was how do you [quickly] sift out that kind of stuff from the real information that needs to be known right now” (Interview participant, NGO volunteer, 28 April 2015).

In LMIC where mobile networks may already be weak, networks may get overloaded by a rapid increase in communications during a disease emergency (Muah et al., 2014), which may limit the potential of SMS based data collection, however transmitting SMS is typically more stable and reliable than phone calls during an emergency (Revere et al., 2014). Similarly, one interview participant was concerned that poor mobile connectivity particularly in rural areas could mean that resources are misdirected because “Where you get random social media and SMS reporting of people expressing need, it tends to be the people that has the technology and can use it and shout the loudest the will come back asking for help. And they’re not necessarily the ones in most need.” (Interview participant, NGO employee, 28 April 2015).

Similar to CDRs, a major challenge to using data received by SMS in disaster response is the time it takes to collate, clean and convert these data into formats useful for mapping, analysis, and subsequent decision-making. Data cleaning and processing is a time consuming task that needs to be completed prior to decision-making, and SMS based surveys do not always produce immediately actionable data, especially due to the anonymity of participants. As one participant noted, “with these epidemics it has to be clean from the source, because you have no time to clean data” (Interview participant, international aid organization, 7 May 2015). However, these SMS data processing challenges were refuted somewhat by another respondent: “You get the [SMS] data, then you need to do some kind of analysis on it, and then really all you need is someone that understands something about statistics and analysing data….Which honestly should not be that difficult if you have monitoring and evaluation folks, or statistics folks, or economists, or researchers” (Interview participant, mobile data platform service provider, 26 August 2015).

In general, raising awareness among the humanitarian sector about how, when and why to collect data using mobile phones would facilitate uptake. As one participant mentioned, “There are few people in humanitarian or development organizations that I think actually understand the benefits that mobile data collection could bring to them and the limitations” (Interview participant, mobile data platform provider, 26 August 2015). Yet widespread uptake of mobile data usage will require active support from donor, private sector (particularly mobile operator) and government (particularly mobile regulators) organizations to pay for training and project management, set up secure contracts to protect mobile users and mobile operators, provide incentives for mobile operators to provide access to mobile data or their networks in disaster contexts, and develop universal standards for data collection, processing and sharing.

3.5. Social and ethical challenges related to mobile data use

In addition to the issues outlined in the previous section, there are a number of associated ethical concerns that should be considered carefully by any organization wishing to use mobile data for disease disaster management. Perhaps most significantly, concerns about individual privacy were widely discussed in the literature and raised by interview participants, especially pertaining to the use of CDR data since these illustrate the locations, spatio-temporal movements, and communication patterns of individual phone users. Although the actual content of the communications are not included in CDR datasets, it is possible to identify sensitive details about individuals’ activities by matching phone numbers from CDR datasets (if they are included) to publically available directories (Google, Yelp) and analyzing patterns in the time of day and the locations from and to which calls were made (Mayer and Mutchler, 2014). When made available to researchers, CDRs are normally pseudonymized (names and numbers removed and replaced with a unique ID) and can be aggregated to the cell tower area to preserve privacy (Tatem et al., 2014, Wesolowski et al., 2014a), and encrypted using mathematical algorithms (GSMA, 2014). However, anonymization makes re-identification more difficult but still feasible by triangulation with online data that links individuals to geographic locations at specific times, such as geo-tagged social media (Cecaj et al., 2016, de Montjoye et al., 2014). Faced with these challenges in the Ebola response, GSMA released guidelines for mobile phone data use advising that CDRs remain secured and encrypted on the operator’s server; all analysis completed within their premises; no attempts to de-anonymize IDs occur, and; only outputs (e.g. maps, models, aggregate statistics) should be shared with third parties (GSMA, 2014).

SMS data also raises significant privacy and confidentiality concerns. These systems are frequently designed to collect potentially sensitive information from participants, which – as in any survey or data collection exercise involving individuals – generally seek to retain participant anonymity. For SMS systems designed to collect information from people on resource needs or disease cases, location privacy will likely be compromised. For systems designed to collect symptom and vital sign data from the public, contributions will be highly sensitive personal information; as one participant described, the data collected via SMS-based systems for monitoring disease symptoms, although highly valuable for containing future outbreaks, is “essentially an extension of a patient’s medical records” (interview participant, international organization employee, 29 April 2015). Yet, data transparency is crucial in these situations, and so knowing the persons and places in which contributions are coming from will be vital for decision-makers to establish its validity and credibility (Asokan and Asokan, 2015). For both SMS and CDR data, aggregation and anonymization are necessary privacy protections, however, they can seriously curtail their potential value. One participant noted, “it limits us in terms of not having any information on who the people are … [w]e know for instance that certain segments of the population are much more likely to be malaria parasite carriers than others, and there are certain segments of the population that are much more mobile, and having that information could be very valuable to disease control programs and to responses to outbreaks” (Interview participant, academic researcher, 10 November 2015).

The possibility of malevolence also has to be taken seriously, if systems are designed to rely on anonymous contributions. Humanitarian emergencies, and certainly communicable disease outbreaks, can be accompanied by a climate of fear and mistrust. Anonymous SMS reporting systems within these contexts could not only result in poor quality data, but could also provide a platform for fuelling community disharmony; as one participant noted, “around something like Ebola, there is huge amounts of fear, there is huge amounts of you know, [people] looking over their shoulder, people reporting neighbours for things” (Interview participant, mobile data platform provider). Similarly, cultural practices related to mobile phone use should be carefully considered such as gender issues, for instance; as one participant identified; “we’ve heard stories if you’re randomly [texting] people in communities, husbands will get jealous and angry … a woman gets a text on her phone from someone she doesn’t know, you could be causing chaos” (Interview participant, mobile data platform service provider, 11 August 2015).

The potential lack of representativeness of mobile data outlined in the previous section is more than just a barrier to use; there are also ethical issues that arise due to the uneven distribution of phones geographically and in populations. While those without a phone will not be subject to the potential privacy incursions outlined above, their voices will not be heard and they may miss out on the ability to shape agendas and resource decisions, or benefit from shared information (Buckee et al., 2013, Shacklett, 2015). This is especially the case for SMS campaigns that provide critical information to users, such as the locations of nearby treatment facilities (Trad et al., 2015).

There is considerable hype surrounding mobile technologies and big data, which present a challenge for those in positions of power and a concern for any organization that wishes to use them. Especially, expectations must be managed when communicating findings to stakeholders and the public: “when people see the products that come out of these things, that gee-whiz factor sort of overwhelms and you know people will start focusing in on that. Unless you bring it into the context of everything else that’s going on in an emergency, it can kind of swamp the other messages that need to get out” (Interview participant, NGO volunteer, 28 April 2015). Aside from managing expectations, data users need to be aware of negative feedbacks that can arise from the introduction of a new data collection technique, especially in settings that have severely limited resources for disaster response. In the context of health emergencies, Higgins (2014) cautions that investing resources in new technologies and data sources can sometimes strip resources away from basic epidemiological analysis and disaster response. These examples suggest that, before employing SMS or CDR based data transfer systems, it is important to assess potential trade-offs and be upfront about any expected benefits or losses that might accrue to individuals or disaster response efforts.

Care must be taken when using CDR data and reporting findings to ensure information is not misleading or inaccurate. Sophisticated algorithms and data mining techniques are required to make sense of big data, and even then, conclusions are often made on the presence of correlations, of which the causal certainty is not well known. Although touted as the ‘end of theory’ – a new approach to science in which data speak for themselves (Anderson, 2008) – big data analytics should be informed by theoretical and specific domain knowledge and should involve human not just algorithmic and data-driven decision making (Miller and Goodchild, 2015), in order to make responsible and context relevant decisions: “there is a lot of thinking that these data just on their own can solve a lot of problems but the importance of linking it with other data sources is incredibly valuable to provide that context and understand the data better” (Interview participant, academic researcher, 10 November 2015). The effects of poor data analytics can be devastating in an emergency context. Vayena et al. (2015) describe how false identification of outbreaks and spread trajectories could result in inappropriate resource allocation decisions or harmful travel bans. However, the novelty of CDRs and big data analytics more generally means that the development of tools, software, and algorithms are in their infancy, and researchers need to develop and compare the accuracy, sensitivity and robustness of different CDR data analysis techniques to minimize the risk that these data can be misinterpreted.

In addition to individual level ethical concerns, the use of mobile phone data in disease disaster contexts could have serious societal consequences. The use of CDRs for surveillance and control as part of state security apparatuses cannot be ignored here, and their use in disaster management could be seen as an example of ‘function creep’, a process of normalization through data repurposing or recontextualization (Custers and Uršič, 2016) that serves to accelerate and strengthen the global corporate-government surveillance assemblage (Haggerty and Ericson, 2000, Murakami Wood, 2013, Taylor, 2015, Taylor and Broeders, 2015). Similarly, the surveillance potential of SMS needs to be better understood, including the ways in which this technology might be used for ‘social sorting’, such as the categorization of people according to risk, which can have serious real-world consequences (Lyon, 2003). In disaster situations, using citizen-generated SMS data to define population groups based on risk categories is clearly problematic; when this process initiates different levels of assistance or intervention there is the possibility of engendering significant social harm or conflict. Arguably different rules apply in disaster situations, however the value of CDRs for monitoring movements and SMS for gathering information from citizens must be carefully considered against threats to freedom and societal control (see Buscher et al., 2013, Taylor, 2015). As Lyon (2007, p. 3) has argued, surveillance is a neutral concept (neither inherently good or bad); each specific implementation exists “somewhere on a continuum between care and control” (italics in original), and so it is vital that the potential for societal harm is carefully considered despite the ostensibly good intentions of disaster management and digital humanitarianism activities.

We use the results presented above to develop a decision-tree for mobile data use in disease-disaster management (Fig. 3 ), which could help government, humanitarian and other organizations seeking information to inform disease disaster management decisions decide whether mobile phone data is an effective method to obtain the information they require in a specific institutional, socio-economic and geographic context.

Fig. 3.

Decision-tree for mobile phone data use in disease disaster management.

4. Discussion and conclusion

“Huge potential, not yet fully refined, I think is the conclusion” (Interview participant, mobile data platform service provider, 11 August 2015)

Against the backdrop of potentially globalized health risks, including recurring communicable disease outbreaks such as Ebola and emerging threats such as Zika and Chikungunya, actively and passively produced mobile phone data could help information managers to fill information gaps, enabling more effective decision-making in all phases of disaster management, from prevention to recovery. This study attempts to fill knowledge gaps and advance our understanding regarding mobile phone data use in this context, and develops a typology and decision making tool to enable researchers and organizations involved in disease disaster management to ascertain how and if mobile data could be useful in future disaster decision-making. Evidence emerging from this research is likely to be relevant to the wider disaster context in many settings, not only to communicable disease emergencies.

In general, this study illustrates the rather different potential of the two mobile data types considered. While SMS data are limited in terms of the depth of data that can be obtained due to the constraints of the format and user-accepted survey length, SMS approaches can be very effective at rapidly extracting diverse types of information that is vital for disaster management, from resource needs to disease cases and symptoms. Combining the data collection functionality of SMS with the ability to broadcast vital information to dispersed citizens makes SMS a very powerful platform that could be effectively utilized by organizations and researchers of diverse technical ability, providing its limitations are properly understood and minimized. Organizations that are able to team up with MNOs to access their network and customer phone numbers – either directly or via an intermediary – may be better able to rapidly set up an SMS data collection/communication system. New tools such as RapidPro SMS are specifically designed to allow simple and quick development of SMS-based surveys at a range of scales.

Our results suggest that CDRs are limited in terms of the types of information that they can generate, yet their potential value for understanding population characteristics and dynamics in time-sensitive situations is unparalleled. Analysis of CDRs in disaster contexts can add huge value to existing data on population characteristics and dynamics. This is especially true for LMIC, which typically have inadequate or non-existent data infrastructures, yet which are frequently at the epicenter of communicable disease outbreaks. The current use of CDRs is, however, largely restricted to researchers and organizations that have big data analysis expertise, and the ability to engage in complex data access negotiations with regulators and MNOs. The future of CDR use in disease outbreaks and other humanitarian emergencies will therefore depend on the outcome of current discussions and negotiations aimed at accelerating access to these data, and on the transfer and uptake of technical knowledge within and across the humanitarian community. This includes establishing guidelines and protocols to ensure access and use of CDRs is considered acceptable by both MNOs and their customers.

Aggregated CDRs are unlikely to have truly disruptive potential in these contexts however (see Kitchin, 2014a). As McDonald (2016) describes, an unprecedented use of CDRs was undertaken in June 2015 in South Korea in the midst of a Middle East Respiratory Syndrome (MERS) outbreak; the government extracted personal information from mobile phone databases to pre-emptively quarantine and restrict the movements of 17,000 people based on algorithmically calculated infection probabilities. Disaggregated, de-anonymized CDRs could be transformative for such purposes as quarantine and contact tracing – vital activities that are often undertaken during contact-based disease outbreaks such as Ebola to contain spread – however more attention to the trade-off between social harm and the public good is needed around the use of CDRs in this format. This represents a crucial question for further academic and policy debate, despite some observers claiming that “data privacy and security … are largely irrelevant in the face of an epidemic outbreak” (Koch, 2016, p. 5).

Irrespective of the validity of this claim, the accumulation and use of big data has further societal costs beyond issues of privacy that must be more fully understood; highly relevant here are the ways in which power is being oriented away from citizens and governments towards corporations (boyd and Crawford, 2012, Thatcher et al., 2016). The discussion of mobile data stakeholders here focuses on governments, MNOs, researchers, and disaster response organizations, yet omits an important group, the phone users themselves who actively or passively produce these data. Despite the data being explicitly about or produced by them, phone users are a little recognized stakeholder group in mobile data use debates, paralleling broader processes of disempowerment of ‘data subjects’ who neither control nor are well-informed about the uses of their data (see Kennedy and Moss, 2015, Leszczynski, 2015, Taylor, 2015, Zwitter, 2014). Shifting power relations are negatively impacting governments as well. The use of mobile phone data in LMICs presents significant opportunities for societal advancement through international development, humanitarian work, and disaster management, however the role of MNOs as ‘data gatekeepers’ is highly concerning, in an era of big data in which “power accrues to those who hold the most data” (Taylor and Broeders, 2015, p. 229). As mobile and other types of big data come to replace or stand in for datasets produced by public or non-governmental bodies (e.g. census or migration data), the growing ability of these private corporations to control how and what they are used for is a concern with significant ethical and democratic implications. Although this power shift has been documented in other areas of big data, notably the growing influence of IBM and other providers of ‘smart city’ solutions in shaping urban governance (Kitchin, 2014b, Townsend, 2013), the significant realized and potential value of mobile phone data highlighted in this paper illustrates the need for critical attention to similar powers of governance afforded to mobile network operators, an issue that may be especially problematic in LMIC.

Similarly, mobile data collection and communication implementations during disaster situations are frequently technocratic, top down, and centralized, which can produce uneven and damaging power relations between disaster relief organizations and distant ‘needy’ victims (see Burns, 2015). For the 2013–15 Ebola outbreak in West Africa, for example, there have been significant critiques of the communication strategies employed and their assumptions of behaviour and embedded social structures. Chandler et al. (2015) argue that many direct communication campaigns in West African countries sought to improve biomedical understanding of risks by targeted populations (e.g. hygienic practices) but that many campaigns were ineffective because they assumed local practices were deficient and were contributing to disease transmission. While it is unclear if these claims are substantiated, such critiques highlight the potential limitations of standardized protocols in some settings, as well as over-reliance on a single, centralized medium of delivery in engaging populations in managing disasters of many kinds. Hence, we stress the need for cultural and contextual awareness and sensitivity, especially in the face of complex emergency situations where poorly constructed campaigns can at best, waste resources, and at worst, undermine the disaster response effort.

Based on empirical findings from interviews and a scoping review, this paper develops a typology of mobile data sources, types, and end-uses and a decision-making tool which researchers and organizations involved in disease disaster management can use to ascertain at what phases of disaster management mobile data could meet their information needs, and what practical and ethical issues need careful attention before mobile data collection methods are attempted. The evidence suggests mobile data sourcing has a real potential to support disaster management efforts and there is growing interest among the humanitarian community in exploring its uses. Yet a range of issues still need to be addressed to secure widespread acceptance, access to, and informed use of mobile phone data in disease outbreaks and disaster management more generally. We recommend that future research efforts focus on: (i) facilitating development of global mobile data usage guidelines, regulations and standards to provide rapid data access to approved organizations and for prescribed purposes, such as humanitarian crises; (ii) documenting, comparing and contrasting existing tools and platforms that can assist in collection, management and dissemination of mobile data, in terms of their capabilities (input and output formats, computational options, compatibility with other information management tools), and usability (telecommunications infrastructure, skill type and level, time, accessibility, availability of training materials) as a minimum; (iii) testing the validity and effectiveness of mobile data use in a wide range of technological and cultural contexts, including through in-depth monitoring and evaluation of active use cases; and (iv) develop quick and effective methods to identify and overcome gaps, anomalies and biases in mobile datasets.

Our findings show that sourcing data through mobile phones could significantly strengthen existing methods of data collection and transfer in all phases of disease-related disaster management. SMS and CDR data from mobile phones provide a means to rapidly gather baseline and situational information in data-scarce environments, and can significantly increase efficiencies, consistency and ease of data capture and transfer for institutions involved in disaster management. This study also finds that these benefits must be weighed against prevailing uncertainties regarding SMS and CDR data validity and representativeness, user privacy, and broader concerns regarding the potential for decisions based on mobile phone data to deepen societal power imbalances. Further research to match mobile data use options (data source, data type, end-use) to specific social, cultural, technological and disaster contexts is needed to accelerate the integration of mobile data sources into humanitarian response effort while ensuring these data complement and do not undermine other valuable data collection approaches.