Why We Need Crowdsourced Data in Infectious Disease Surveillance

Abstract

In infectious disease surveillance, public health data such as environmental, hospital, or census data have been extensively explored to create robust models of disease dynamics. However, this information is also subject to its own biases, including latency, high cost, contributor biases, and imprecise resolution. Simultaneously, new technologies, including Internet and mobile phone based tools, now enable information to be garnered directly from individuals at the point of care. Here, we consider how these crowdsourced data offer the opportunity to fill gaps in and augment current epidemiological models. Challenges and methods for overcoming limitations of the data are also reviewed. As more new information sources become mature, incorporating these novel data into epidemiological frameworks will enable us to learn more about infectious disease dynamics.

Global patterns of disease burden are constantly shifting. Recent studies of the emergence of novel infectious diseases have indicated numerous drivers, including the shift of populations to urban centers, increased mobility, and evolving human–animal interactions [1, 2]. Understanding disease dynamics in populations provides the best opportunity for understanding, controlling, and predicting disease spread. Spatiotemporal models based on public health surveillance data have been extensively explored for this purpose, elucidating patterns and processes by which infectious diseases diffuse across regions. These models traditionally rely on official or government sources, such as environmental, hospital, or census data [3, 4]. Although these data sets are robust and validated and attempt to report on entire populations and their collection is facilitated by intermediaries, they suffer from inherent limits resulting from latency, high cost, contributor biases, and imprecise demographic and geographic resolution [5, 6]. Additionally, studies have indicated areas of deficiency in traditional health systems, including timeliness and financial barriers to care [7].

Simultaneously, new technologies, including Internet tools such as social media or mobile devices, all coupled with global positioning systems, enable a new form of infectious disease information to be garnered directly from citizens. These crowdsourced data evade potentially constraining infrastructure costs and regulations, can be generated in real time, and can be used to fill in gaps in health information due to barriers in health-seeking behaviors through traditional systems [8–10]. Furthermore, these tools can now be deployed at scales that enable information to be garnered at a population level.

Generally, crowdsourcing is the process of obtaining services, ideas, or other information via a large group from the public, rather than a specific set of people (such as government institutions or hospitals). From crisis management to bioinformatics and ecology, information from individuals is providing disparate views and solutions, supplementing existing systems in normal or interrupted use [11–14]. In infectious disease surveillance, crowdsourcing offers the opportunity for collection of symptom and related information right from the point of care [15].

Although considered “gold standards,”, the prerequisite acquisition, aggregation, and validation steps in traditional clinical data sets naturally incur limitations. For example, the United States Centers for Disease Control and Prevention’s (CDC) influenza-like illness (ILI) surveillance system has been the primary metric for measuring national influenza activity. Yet because of differences in laboratory practices and patient populations seen by different providers, comparison of the CDC data between regions and across seasons is not straightforward [16]. Furthermore, temporal trends in the CDC data can be driven by multiple factors that are difficult to disentangle (Fig. 1); during holiday weeks, there could be a higher percentage of ILI visits based on increased disease activity and/or changes in health-seeking behavior, since there are fewer patient visits to sentinel sites overall at these times [17].

Fig. 1.

a Average percentage of visits to CDC sentinel sites for ILI by week. b Average number of patients seen at sentinel sites by week. Data are for seasons 2000–2011, pandemic seasons and those with 53 weeks excluded. Holiday weeks (shaded areas: 46–48, Thanksgiving and 51–1, New Years) show both an increase in %ILI visits and a decreased amount of patient visits

On an international scale, the World Health Organization (WHO) field reports of infectious disease outbreaks come from technical institutions and organizations that have the capacity to contribute to international outbreak alert and response. The WHO’s network provides some access to information from affected regions but is limited to organizationally obtained information and their reach [18]. Additionally, the data collection process can be affected by unequal selection whereby larger outbreaks are more likely to be detected, so that estimates of transmissibility may be biased upward [19]. Filling some of these gaps, news media have proven useful, in aggregate, for providing early information of epidemiological value for population-level disease surveillance and have decreased time to outbreak detection substantially [20]. More than 60% of all initial outbreak reports come from unofficial informal sources, such as news media [21]. However, Internet-based news is also subject to distinct limitations based on credibility, detection speed, reach to isolated populations, and geographic coverage of areas where media are restricted or limited. Figure 2 demonstrates the differences in these data sources, illustrating HealthMap [22] disease alerts by continent from 2006 to 2009, in contrast to WHO disease reports for the same time period. These pervasive limitations of current data sources hinder our understanding of disease dynamics. For instance, seasonality of infection risk in malaria is poorly understood [23], and domestically, we have weak understanding of temporal and spatial variation in influenza incidence as described above.

Fig. 2.

Disease events by continent via news reports 2006–2009, compared with WHO disease reports for the same time period

Crowdsourcing offers a real-time picture of disease by harnessing information as individuals are diagnosed or even before [8, 24]. These temporal advantages are especially vital since increased ease of mobility decreases the time for infectious diseases to spread globally to the scales of hours or minutes, much quicker than even the serial interval of many diseases [25]. Additionally, these tools can spatially augment information in places that current surveillance sites do not cover [9, 26]. Another benefit of working directly with the public is that it augments engagement and enables individuals to become more aware of and involved in their own health, as anecdotal evidence has shown [10]. Thus, this approach can provide an avenue for targeted health education and rapidly measuring responses to public health interventions. Finally, through crowdsourcing infectious disease information, we can learn about aspects of disease dynamics that are not accessible through traditional data, such as contact patterns and aspects of the social environment [27, 28].

Simultaneously, crowdsourced data present their own challenges. There are issues of validation, which current studies are addressing by bringing reported data together with diagnostic or other clinical measures, such as emergency room crowding [29]. Additionally, low specificity, 1– P(false alarm), can result from confounding factors such as media events [9, 30] or demographic biases [31, 32]. Although more work is needed, some studies have uncovered demographic or temporal factors shaping use of the tools [30–32].

Every data source includes biases and challenges that must be robustly understood before the data can be used to study disease dynamics. Further studies of crowdsourced data should continue to focus on addressing issues of population representativeness, reporting bias, and validation in order to demonstrate how the data can be used as a complement to existing epidemiological sources. As crowdsourcing data types and sources become more ubiquitous, we expect these data to serve as a vital component of global disease surveillance efforts.