copy the linklink copied!5. Big data: A new dawn for public health?

5.2.1. New analytical techniques can enhance public health policy making

Traditional public health is data-poor and has traditionally lacked the key big data characteristics: high volume, high variety, and high velocity. Epidemiologic research generally relies on long-term, longitudinal, relatively small- or medium-scale cohort studies, in which data are gathered through participant questionnaires, physical examinations, and, for outcome data, health records. These data sources are often difficult to obtain and work with. Similarly, public health surveillance – defined as the ongoing systematic collection, analysis, and interpretation of data, as well as the dissemination of these data to public health practitioners, clinicians, and policy makers (Richards et al., 2017[9]) – and public health interventions have traditionally been performed through time-consuming, error-prone methods. These methods pose significant timeliness and efficiency limitations and suffer from time lags and lack of spatial resolution. Table 5.1 summarises the implications of big data for public health.

However, the development of new analytical techniques has enabled policy makers to harness existing health datasets in ways that can transform existing disparate databases into a larger set of data with key big data characteristics.

In non-communicable disease prevention, big data can help better understand and address modifiable behavioural risk factors that contribute to a large fraction of the non-communicable disease burden (e.g. diet, physical activity, tobacco use). Big data analytics (Box 5.2) can enable policy makers to more effectively assess these risk factors at the population, subpopulation, and individual levels, as well as to design better targeted interventions aimed at mitigating them. In particular, big data analytics can help understand, at a causal level, how hereditary risk factors – as well as combinations of risk factors – interact with behaviour and the physical and social environment. In communicable disease prevention, hybrid tools that combine traditional methods and big data analytics can enhance communicable disease surveillance by harnessing novel data streams to complement – rather than replace – traditional methods.

5.2.2. Big data can improve the identification of population- and person-level risk factors

The potential of big data to allow for a more precise identification of risk factors is attracting more and more interest from researchers and policy makers alike. An increasing number of large-scale population studies aim to leverage big data to pinpoint specific risk factors more precisely, using more data than traditional epidemiologic studies are able to. There is a particularly enthusiastic focus on identifying hereditary risk factors through genetic testing.

This approach, however, can backfire: the temptation to use big data to find new, exciting risk factors or more precisely measure the effects of known ones could come at the expense of engaging with “the broader causal architecture that produces population health”; in other words, the “proliferation of causal effects – typically identified through an approach that aims to isolate risk factors for particular outcomes – presents a conundrum for scientists, let alone the lay public, to synthesize and form evidence-based recommendations that can promote health.” (Keyes and Galea, 2015[13])

The capacity to leverage bigger and better data to measure the effects of precise risk factors therefore needs to be carefully weighed against what matters most for population health (Keyes and Galea, 2015[13]). Indeed, the potential of big data to improve public health lies not in better measurements of various isolated risk factors, but in the ability to analyse the complex, dynamic interactions between human behaviour/lifestyle (“behavioural phenotypes”), genetics, and the physical and social environment to determine what matters most for public health policy.

There is therefore a need to move from clinical validity (confirming robust relationships between risk factors and disease) to clinical utility: in other words, when it comes to the public health impact of big-data driven research, researchers and policy makers should address the “Who cares?” and “So what?” questions.

Furthermore, discussions of big data often go beyond the technological and analytical aspects and suggest a “mythological” dimension: “the widespread belief that large data sets offer a higher form of intelligence and knowledge that can generate insights that were previously impossible, with the aura of truth, objectivity, and accuracy” (Boyd and Crawford, 2012[14])

5.3.1. Novel data sources are uncommon as big data sources for public health

Recent technological advancements have led to new approaches to disease prediction and monitoring, with mixed results. Efforts have begun to systematically mine “virtual digital trails”, such as social media and internet query data, to assess health-related behaviours and attitudes in near-real time. Current initiatives are promising, but most are limited to one-off projects. Moreover, where used on a wider scale this approach has delivered mixed results, with a number of high-profile failures, including Google Flu Trends, pointing to the challenges implicit in using these new data sources. Scaling up these approaches in a way that also protects data privacy and security is needed.

Infodemiology refers to systematically mining, aggregating, and analysing unstructured, textual, openly accessible online information to inform public health policy (Eysenbach, 2011[18]). Infoveillance refers to using infodemiology metrics for surveillance and trend analysis (Eysenbach, 2011[18]). Crowdsourcing can also be used as an alternative to infoveillance to collect public health data using big data approaches.

Social media platforms, in particular, represent sources of rich observational data for infodemiology and infoveillance (Kim, Huang and Emery, 2016[19]). For instance, in the area of non-communicable disease prevention, the dynamics of social networks can be studied to discern patterns of how social factors influence unhealthy behaviours, such as smoking (Andreu-Perez et al., 2015[20]). Furthermore, combining insights from social media data with geolocation data can enable a better understanding of patient behaviours and social demographics; it has been used to study, for instance, influenza outbreaks and antibiotic misuse (Andreu-Perez et al., 2015[20]).

Methods for collecting, filtering, and analysing social media data need to be clearly and transparently reported. As discussed further on in this chapter, data transparency is a key driver of many successful big data initiatives, such as the “smart cities” and “smart health” initiatives. Infoveillance challenges and limitations include: privacy concerns, difficulties in establishing causal links due to lack of context, isolating signal from noise, and lack of generalisability of Internet or social media users, due to the overrepresentation of certain population groups.

Mining “virtual digital trails” – such as social media and Internet query data – offers the opportunity to assess self-reported health- related attitudes and behaviours, such as those pertaining to non-communicable diseases and their risk factors, in near real time; it can thus complement more traditional non-communicable disease surveillance data. For example, research has linked anger and stress expressed on Twitter to an increased risk of heart disease (Andreu-Perez et al., 2015[20]; Eichstaedt et al., 2015[21]). Another study was able to accurately predict the likelihood of smoking and alcohol consumption based on the user’s behaviour on Facebook, including how many and what they “Liked” on the website (Kosinski, Stillwell and Graepel, 2013[22]).

Social media infoveillance can also allow public health entities to detect whether specific communities may need certain health or social services, particularly in the case of stigmatised health issues, such as drug use. This awareness can enable more targeted surveillance and enhanced interventions. For instance, one study used Twitter data to identify online communities of illicit, recreational, and medical cannabis users connected to specific dispensary accounts (Baumgartner and Peiper, 2017[23]).

“Real-life digital trails” are “signals produced by people’s everyday actions, recorded digitally through devices and sensors measuring individuals’ movements and behaviours” (Balicer et al., 2018[24]). The ways in which an individual interacts with digital technologies – such as through texts, calls, and social media posts – are markers of his/her “digital phenotype”, which, when combined with other sources of data, such as clinical data, can allow for early disease detection and intervention (Jain et al., 2015[25]). For instance, in one study, behavioural indicators obtained from phone usage data were strongly linked to depression severity (Saeb et al., 2015[26]). Table 5.2 summarises the advantages, challenges and potential contribution to disease surveillance of sources of traditional health data, virtual digital trails and real-life digital trails.

In addition, sensor data from wearable technologies and environmental sensors can provide insights on a variety of lifestyle risk factors and aid in chronic disease management (Balicer et al., 2018[24]). Smart devices represent a promising source of crowdsourced big data that can be leveraged to track the spread of certain infectious diseases. For instance, “smart thermometers” connected to a mobile phone application provide de-identified fever data that can help track influenza activity in real time. This crowdsourced data can be used to improve influenza surveillance and forecasting by complementing traditional models, which rely on data from hospitals and clinics and which tend to lag behind real-time influenza activity. One study, which analysed data from over 8 million temperature readings generated by almost 450 000 “smart thermometers”, showed that the data were highly correlated with information obtained from traditional disease surveillance systems and could potentially predict influenza activity up to two to three weeks in advance (Miller et al., 2018[27]).

The use of such smart devices is generally limited to individual purchasers who can afford to buy them, making the data susceptible to socioeconomic biases. Given the potential of these smart devices to both capture real-time infectious disease activity at a population level and help generate improved disease forecasts, policy makers should explore strategic partnerships with the private sector that could facilitate a more widespread use of such devices – provided that they have been validated against traditional surveillance methods – and a systematic integration of their data into national communicable disease surveillance systems.

Wearable sensors, including mobile phone accelerometers, can provide valuable data about individual behaviour and lifestyle factors, such as patterns of physical activity and sleep, which are linked to a variety of health outcomes. Combined with smart technology, the data collected by these sensors can enable personalised health promotion interventions, such as mobile phone applications that prompt a user to engage in exercise if an unhealthy pattern of physical inactivity is detected.

Data from such wearable devices, however, tends to suffer from a large amount of noise (e.g. a wrist-worn accelerometer may not be able to differentiate whether different types of arm movement indicate the user is exercising or not (Gelfand, 2019[28])). Complex statistical methods are thus needed to analyse these data. Despite these issues, wearable sensor-based interventions hold the potential to enable health policy makers to implement large-scale, highly personalised, low-cost health promotion interventions, in partnership with researchers and the private sector.

5.3.2. Big data approaches can complement traditional disease surveillance methods

Traditional infectious disease surveillance is typically based on epidemiological data collected by health institutions. While these data have a high degree of veracity, they also suffer from several disadvantages, including: time lags, due to lack of human resources or problems when aggregating data from different sources; high cost; and insufficient local granularity. In contrast, big data streams – such as internet queries, social media data, and crowdsourced data – can be tracked in real time and at a local level with minimal cost, but have their own biases that need to be accounted for. (Bansal et al., 2016[7]; Simonsen et al., 2016[29])

Communicable disease surveillance is therefore one of the most exciting opportunities created by big data in the realm of public health. These novel data streams can improve the timeliness and the spatial and temporal resolution of infectious disease tracking, as well as provide access to “hidden” populations (Bansal et al., 2016[7]; Simonsen et al., 2016[29]). Some recent successes – including Health Map’s successful identification of a haemorrhagic fever outbreak in West Africa in 2014, which was subsequently identified as Ebola – point to the potential of complementing traditional surveillance methods with new approaches. Big data streams can also go beyond disease surveillance and provide information on specific behaviours and outcomes related to vaccine or drug use.

Nevertheless, while such new methods of infectious disease surveillance are promising, they may not always be mature enough and should be validated against established infectious disease surveillance systems. Policy makers and researchers must remain vigilant to avoid “big data hubris”, which refers to the assumption that “big data are a substitute for, rather than a supplement to, traditional data collection and analysis” (Lazer et al., 2014[30]) – in other words, the assumption that high-volume, high-velocity data can replace smaller, high-veracity data and traditional data analysis methods (Fuller, Buote and Stanley, 2017[12]). Some past examples of big data-driven surveillance systems that suffered from big data hubris – such as, notably, Google Flu Trends (Box 5.3) – failed to deliver reliable information.

Using digital data for public health surveillance presents a set of challenges, such as: lack of demographic information; issues of representativeness, as these data tend to represent a limited segment of the population that may not include certain age categories, or may include fewer elderly individuals; and spatial and temporal uncertainty – for instance, someone may be researching a family member's illness that occurred in a different city several weeks earlier. (Bansal et al., 2016[7])

Furthermore, before relying on these types of novel data sources, public health authorities should assess the impact of local conditions on the reliability of predictive algorithms. For instance, a dengue surveillance algorithm that used Internet query data to create an index of dengue incidence worked well in areas with high incidence of dengue and favourable vector climate conditions, but was not a reliable predictor of dengue incidence in areas with low incidence and an unfavourable climate (Gluskin et al., 2014[31]).

Novel methods should therefore complement – rather than replace – traditional methods (Bansal et al., 2016[7]). Policy makers should aim for hybrid tools that combine traditional methods and big data analytics to enhance communicable disease surveillance. Hybrid tools make use of the advantages of novel big data sources – such as timeliness, scale, and fine granularity – while maintaining a direct link to disease through traditional surveillance systems (Simonsen et al., 2016[29]). Where well-performing prediction models are already in place at the national level (e.g. the CDC flu prediction model in the US), big data analytics could be used to enhance local-level data.

Participatory surveillance systems that use crowdsourcing or crowdmapping have been growing, but have yet to be used at a large-scale by public health authorities. Examples include Influenzanet – a crowdsourcing system for self-reported flu symptoms used in ten European countries, ResistanceOpen – which aggregates publicly available data on antimicrobial resistance, and Health Map – which uses online data to track infectious disease outbreaks around the world. An extension of HealthMap is “Flu Near You”, which provides a crowdsourced, real-time “flu map” that shows influenza activity in the United States. As the integration of participatory surveillance with traditional surveillance systems can significantly improve infectious disease surveillance, policy makers should explore ways in which they can develop and contribute to both national and international efforts in this area.

Surveillance systems that track infectious diseases form the backbone of communicable disease monitoring and controlling, but tend to suffer from time lags and insufficient spatial resolution. They remain primarily based on manually collected data, which is then aggregated into national or regional reports, thus lacking local-level granularity. Novel surveillance approaches that use big data streams, including electronic health (e-health) patient records, unstructured digital data sources, and participatory surveillance should be leveraged to help strengthen infectious disease surveillance systems. (Bansal et al., 2016[7])

5.3.3. Policy makers are starting to explore big data for precision public health

Precision medicine is quickly moving to the forefront of many health systems’ vision for the future, driven by significant advances in genetic research. Combining traditional medical data with novel data and technologies from fields such as genomics, epigenomics, transcriptomics, proteomics, metabolomics, and phenomics is enabling a better understanding of disease pathogenesis and more targeted diagnoses and treatments, especially for cancer.

In 2016, the United States launched the USD 215 million Precision Medicine Initiative. This initiative includes, among other projects, the All of Us research programme, a 1-million participant study whose mission is “to accelerate health research and medical breakthroughs, enabling individualized prevention, treatment, and care” by studying “individual differences in lifestyle, environment, and biology” (National Institutes of Health, 2019[35]).

While precision medicine is seen as “an emerging approach for disease treatment and prevention that takes into account individual variability in genes, environment, and lifestyle in each person” (National Alliance Of Healthcare Purchaser Coalitions, 2018[36]), most efforts in this area have thus far focused on improving treatment, rather than prevention. The next step for health systems is to begin to use big data to transform precision medicine into precision health, by applying new insights to not only improve diagnosis and treatment, but also health promotion and disease prevention. In the United States, for example, the program Connecting Public Health Information Systems and Health Information Exchange Organizations gathered information on how public health jurisdictions use existing health information exchange (HIE) organizations as a means of sharing information with providers (The Office of the National Coordinator for Health Information Technology, 2017[37]).

Public health policy makers should leverage research discoveries to enable more targeted disease prevention strategies at a population level. Such prevention strategies can become an important part of “wellness planning” and health management, and a stepping stone towards providing “truly anticipatory “health care,” instead of the responsive “sick care” that has long been the health care system’s default” (Willard, Feinberg and Ledbetter, 2018[38])

In addition, current precision health approaches mostly focus on individual genetic variability. Most “personalised prevention” has been based on hereditary risk factors, such as cancer-causing mutations, that can be detected with genetic testing. In the United Kingdom, for example, genetic testing is made available to the family members of people with certain mutations.

In the United States, Geisinger’s MyCode/ GenomeFIRST initiative provides screening for genetic variants linked to a higher risk for certain medically actionable conditions, such as the BRCA1 and BRCA2 variants associated with an increased risk of developing breast cancer (Williams et al., 2018[39]). Based on ongoing results, it is estimated that approximately 3.5% of study participants will be found to carry risky gene variants (Trivedi, 2017[40]) (Box 5.4).

It remains to be seen whether the intervention will be cost-effective in the long run. If so, it would provide evidence that routine population level genetic screening for variants that increase the risk for medically actionable conditions should be leveraged as an important strategy for implementing precision health at a population level, particularly as the cost of genome sequencing continues to decline. Other considerations for such programs, in addition to cost effectiveness, include the potential for false positive results, as well as privacy concerns.

A precision health approach to health care, however, should move beyond just looking at genetic testing; it should take into account individual variability not only in genes, but also in environment and lifestyle, as well as their interaction. This model has been adopted by a number of large-scale research projects, including the Human Project in New York City in the United States (Box 5.5). Public-private partnerships are essential for this approach. In the United States, the State of Nevada’s “Healthy Nevada Project” aims to improve population health and personalized medicine by integrating clinical, genetic and environmental data with socioeconomic determinants to better understand the interplay between these factors; for this project, health care network Renown Health has partnered with the Desert Research Institute and Helix, a personal genomics company, with support from the State of Nevada (Renown Institute for Health Innovation, 2019[41]).

Ongoing large-scale, big data-driven population studies, like the United States 1-million participants “All of Us” study and the Danish “Harnessing the Power of Big Data to Address the Societal Challenge of Ageing” research project, also hold the potential to yield valuable insights, such as identifying which types of environments are more likely to facilitate healthier behaviours.

5.4.1. Big data approaches can help translate knowledge to practice

The development of better population profiles offers the opportunity to develop new approaches to implementing prevention strategies. Using big data to better target public health initiatives may help to improve their effectiveness. One particular area in which big data can yield valuable insights is the translation of knowledge into effective public health policy. Translation arguably represents the next frontier in public health: how to distil the vast public health knowledge yielded by public health research studies into actionable steps to inform public health policy and decision-making.

The issue of “Much is known, but little is done” is at the forefront of many countries’ public health policy discussions. The European Commission, for instance, recently formalised a new mechanism, called the Steering Group on Health Promotion, Disease Prevention and Management of Non-Communicable Diseases (the “Steering Group for Promotion and Prevention”), to facilitate the implementation of evidence-based best practices by EU countries to help prevent and manage non-communicable diseases (European Commission, n.d.[45]).

There are two main ways in which policy makers can leverage big data to address this issue. Firstly, big data can enable a faster, real-time monitoring of the impact of public health interventions. Changes in behaviour, public attitudes, public attention, or health status are often reflected in real-time online information sharing and communication patterns (Kim, Huang and Emery, 2016[19]). These data points can give decision makers valuable feedback on the effectiveness of public health interventions and thus inform policy making.

Secondly, big data can facilitate a better understanding of the interaction between human behaviour/lifestyle (“behavioural phenotypes”), genetics, and the physical and social environment. Understanding which of these interactions have the greatest causal impact on public health can inform policy making – not only in public health, but also in other areas that influence health (e.g., various socioeconomic issues). Policy makers should therefore leverage big data to help move towards a Health in All Policies approach.

5.4.2. Many promising uses of big data for public health have emerged from the municipal level

In many cases, promising uses of big data for public health have emerged not from traditional public health actors, but, for instance, from initiatives to transform urban areas into “smart cities” that use data to improve urban planning, as well as policy making more generally.

As a result of accelerating urbanisation, cities face both the challenge and opportunity of being “first responders” to key global issues, including in public health, especially given many cities’ status as hubs of data traffic and new technology applications. As centres for novel, big data-informed solutions and services in an increasingly decentralised world, “smart cities” have moved to the forefront of public health innovation.

Smart cities are “cities strongly founded on information and communication technologies that invest in human and social capital to improve the quality of life of their citizens by fostering economic growth, participatory governance, wise management of resources, sustainability, and efficient mobility, whilst they guarantee the privacy and security of the citizens.” (Pérez-Martínez, Martínez-Ballesté and Solanas, 2013[46]). In the area of public health, smart cities can leverage their ability to develop and implement innovative, data-driven public health policies informed by big data analytics and move towards “smart health”, without having to wait for national-level action – although such action can complement city-driven innovation by enhancing collaboration between cities, as discussed later in this section.

Smart health (“s-health”) is “the provision of health services by using the context-aware network and sensing infrastructure of smart cities” (Solanas et al., 2014[47]). As this definition implies, ICT, big data and big data analytics are a key driver of smart health approaches.

The City of Chicago’s food safety inspection, E. coli prediction on Lake Michigan beaches, and lead poisoning prevention programmes (Box 5.6), driven by predictive analytics and machine learning models, are examples of big data-driven, smart health solutions to public health problems. A partnership between the cities of Chicago and Las Vegas, Google, and Harvard University found that using location data and foodborne illness online searches predicted potentially unsafe restaurants better than traditional restaurant inspection methods and data-mining approaches (Sadilek et al., 2018[48]). What makes this example unique is that machine learning was used to improve accuracy, and that the linkage of these personal data also ensured that individuals remained unidentifiable.

The development and dissemination of such smart health solutions, however, has been slow. For instance, data gathered by sensors that measure environmental variables such as air pollution and airborne allergens could enhance the delivery of personalised prevention and care to asthma patients; a study that analysed data from wireless environmental sensor networks for air pollution measurement in eight cities across Europe, however, found that the performance of most of the sensors was unreliable, and they required frequent calibrations due to the interference of various environmental factors (Broday and Collaborators, 2017[49]).

As noted in the Chicago case study, open data sharing represents a key driver of big data innovation and developing smart health approaches. Open data sharing helps connect smart city innovators with the relevant data. As data collection is often the most difficult part of researching and developing a solution to a particular problem, limiting data access can result in missed opportunities for “non-insiders” to develop potentially successful solutions. Open data also needs to be organized into databases with user-friendly search tools that allow easy data filtering (Smith, 2017[54]). It is also important to ensure that when data are made widely available, it is interpreted correctly by the wider set of users. Data privacy and security are, of course, particularly important when data is made publicly available.

While the cost and effort required for such projects may seem daunting, the benefits of enabling the development of innovative solutions driven by big data analytics can far outweigh these costs. Further, in the case of data crowdsourcing projects open data sharing can serve as an incentive for the public to participate in these initiatives. Examples of open data projects include La Base Adresse Nationale (France), Trafikverket (Sweden), and Data.gov (United States).

Inter-city collaboration can significantly speed up the rate of big data solutions in public health. One of the key issues of within-city innovation is that, typically, each city designs and implements its own good practices, with other cities finding out about them at a later stage, after they are successful (if ever). But co-ordination that can scale up successful solutions more effectively is needed; for instance, countries should organise partnerships between cities to facilitate tackling common issues together. Some promising examples are emerging: the Netherlands’ “Smart City Strategy,” for instance, aims to create a “Smart City collective” that will link cities, companies, and the research sector, functioning as a catalyst of knowledge sharing and change (Institute for Future of Living, 2017[55]).

At the same time, the important role of cities in testing and implementing “smart” approaches to public health runs the risk of exacerbating rural-urban health inequities. Policy makers should ensure rural areas are included in smart health programmes.

Another key aspect needed to advance towards smart city collaborations in the area of “smart health” are cross-sector and public-private partnerships. The “quadruple helix” collaboration between government, academia, industry, and citizens is essential for smart city and smart health innovations. Inter-sectoral collaboration is another key driver, particularly in public health, given the diversity of the causes of various diseases. As an example, Chicago’s lead poisoning prevention programme involves collaboration between health care providers, lead inspectors, and housing providers, among others. Further, partnerships between cities (and other local governments) and the private sector should be explored, which can allow cities that do not yet use predictive analytics methods in-house, due to lack of resources or expertise, to contract them out.