Controversies surrounding data use, storage, and sharing illustrate the important ethical questions that emerge when data collection and analyses are applied to new ends. Examples in the news abound, for instance, the rise of direct-to-consumer genetic testing for diseases such as cancer, seen recently through the example given on National Public Radio of an uninsured American woman concerned about her risk of breast cancer [6]. Upon reading her results from 23andMe, the woman admitted feeling less urgency about getting additional testing or mammograms with her physician—something that geneticists worry could pose problems for individuals carrying variants undetected by tests offered by commercial sources, or for those who receive summary advice from individuals without proper training, possibly leading to clinical harm in the future. Other disputes have emerged when health technologies are applied to new ends, such as the recent identification of the Golden State Killer in California, United States, in April 2018. Detectives were able to identify the perpetrator by matching crime scene evidence with a family member’s DNA profile that the family member had uploaded to a genealogy website. The incident, and subsequent admission that private companies have shared access to their database with law enforcement to find potential suspects, spurred controversy among experts and the public over the legitimacy of using the personal data of volunteers who had not consented to such law enforcement applications [7,8]. Controversies such as these—as well as others surrounding privacy and, for example, the hacking of medical devices [9], or matters of justice and fairness in algorithms [10]—point to the centrality of data at the heart of negotiations over the public good; the status of data generated outside of official forums of science and medicine; and central ethical questions of privacy, consent, and benefit that are emerging in new configurations [11,12].

By data work, we are referring broadly to the forms of technological, analytical, and emotional work undertaken by all actors within the health care system that is necessary to make data clinically and personally meaningful. Here, we focus on the emerging forms of data work undertaken by patients and health professionals. This work is already occurring, for example in the interpretation of direct-to-consumer genetic tests [13], efforts to improve patient understanding of broad consent in biobanking [14], or as researchers define proteomic markers of risk, such as for ovarian cancer [15], albeit in an often unrecognized and patchwork manner. Although science studies scholarship has explored various determinants and conditions of data production in the health sphere [16-18], the types of work that are necessary to make diverse forms of health data actionable in daily life by patients and health professionals have not been systematically addressed or conceptually analyzed [19]. Data work is ongoing and constitutes a formidable yet underresearched challenge in the era of data-rich medicine. But what kinds of work does this entail, and for whom? What divisions of work or tools would be necessary for addressing ethical and equitable applications of data in everyday life?

Empirical studies examining the organization and structure of medical work from a sociological perspective [20-24] have been helpful to draw attention to the often invisible contributions that patients and their family members make to all aspects of health care. However, conceptualizations of such patient work in the era of data-driven medicine are, as of yet, largely missing [25]. As debate grows in medicine over how to best actualize voluminous and diverse data for better outcomes in health care [26,27], many of the biggest challenges are of a social, rather than technical, nature [1]. In this context, more systematic attention to the ways in which professional and nonprofessional actors within the health care system help, for example, to create and interpret data, would fill an important gap. In the following section, we outline and describe three areas of emerging forms of work that have accompanied the turn toward big data in medicine, identify who does this work, and sketch potential ways of addressing concerns that arise in connection with this work. For each area of data work, we offer one vignette to illustrate the forms of data work that are already ongoing. Although the boundaries between these different types of data work are fluid, we posit that there is analytic value in drawing out the key features that characterize each activity to see what challenges they pose and how we might address these.

Advances in mobile devices have changed how health information and support services are being accessed, communicated, monitored, and acted upon [28], offering potential gains ranging from clinical oncology [29] to improving health outcomes for low-income populations [30]. As a result, patients create and engage with health data not only in medical institutions but also in their homes and in other places outside the clinic, via wearable or portable devices, or other tools. Patients and health care professionals alike are faced with ever wider types and larger volumes of data that could potentially be relevant for health care, without a clear understanding of the implications of specific forms of personal data [31]. In the domain of mobile apps, one form of emergent data work is the work done by patients who search through the fine print of Terms of Services of new devices and apps to decide whether or not to use them. This is often not easy to do; for instance, the interests of a company providing a digital health device or platform might be hard to fathom for a user, posing potential concerns for individuals who are consenting to data use agreements for a health app, or uploading their medical history to a Web portal for a rare-disease patient community.

Furthermore, the ability to learn about genetic traits—which can now be done with ever lower expense on the internet —raises profound ethical challenges. As Kung and Wu ask, “if we discover certain genetic risk factors in our genome sequences, do we (or our health care providers) have a responsibility to inform our family members who might have similar genetic risks?” [32]. Privacy matters, and the effects of new health technologies on future generations all become important concerns with which individuals have to grapple, while very little or no guidance may be available. The work that people are doing when navigating the landscape of available offers, and in deciding what test they should take and what behavior they should track in an attempt to maintain or increase their health, should not be trivialized. There are increasing expectations that individuals make informed decisions as responsible managers of their health, and now also as owners (morally or legally) of their data.

In addition to such new data work for patients, another novel form of data work emerges for health professionals. This consists of assisting patients and their families in navigating the landscapes of available offers for tests, devices, and services, and helping them to decide whether they should datafy certain aspects of their lives and bodies in the first place. This data work includes engaging patients in conversations about the implications of their potential data contributions before patients have had practical experience with these digital practices, and about whether and how they should consider engaging in certain activities. Steering patients through the multitude of options is an important yet complex task. Recent studies have also shown that socioeconomic status, age, English literacy, and digital literacy all play important roles in the uptake of new mobile technologies such as health apps [28,29,33] in engaging in Web-based participatory medical research [34] and in efforts to counter the digital divide [35-37]. Importantly, these differences also influence whose data are missing from the broader evidence base upon which future decisions in medicine might be made [25]. This points to the growing need to ensure that such digital health practices and technologies do not exacerbate existing inequalities in society or health and the critical role that health professionals are called upon to play in mediating digital engagements.

Looking forward, we thus anticipate that the data work of professionals in this space will include not only assisting patients in navigating this digitalized network of health-relevant services but also assisting those who cannot, or choose not to, engage digitally [38]. As noted, people who do not make use of digital tools to collect, view, and share data and information about themselves can become missing bodies in today’s health care environments, meaning that their bodies, needs, and behaviors remain unaccounted for in decisions made on the basis of new digital health sources [25,39]. Especially when the stakes are so high, neither offering guidance on patient use of digital tools and new health products nor understanding the advantages and disadvantages of the many new products on the market every day is intuitive. To be effective, these activities require time and appropriate training, which are in very short supply in today’s time-starved health care environment [40-43].

One possibility, as we have argued elsewhere, to better support both patients and providers in the era of data-rich medicine would be the creation of a new, intermediary profession entirely, which we have termed health information counselors (HICs) [44]. With a broad knowledge of various kinds of health data and data quality evaluation techniques, as well as analytic skills in statistics and data interpretation, our vision is that HICs would be trained also in interpersonal communication, health management, insurance systems, and medico-legal aspects of data privacy. Operating as a clinical consultancy, HICs would have the ability to translate the complex language of data into intelligible and actionable information for both patients and physicians. The creation and implementation of such a specialty would enable patients to make educated, truly autonomous choices about how these novel forms of health data can inform their personal care decisions. Although certainly not the only option for addressing the aforementioned concerns, the creation of this new specialty would go a long way in assisting individuals such as G from our opening vignette, as well as health care professionals, to consider their options and make more informed choices about how increasing amounts of health data and information can or should inform health care.

Testing practices such as the one described in this vignette have become a means through which individuals understand themselves and their relationship to society. For some patients, the quantified self can allow people to see new patterns or make changes in their lives: counting steps might lead one to take the stairs, and tracking sleep patterns might lead another to try and get an extra hour of sleep. For others, finding out the percentages of one’s global ancestry or likelihood that they could be a carrier for a genetic condition represents personally significant information. Yet, the effects of health-related data and information are often difficult to anticipate and understand. Randomized controlled trials have studied the clinical impact of patients’ use of mobile and digital health tools, such as the effectiveness of smartphone apps for weight loss and self-applied therapies [45-47]. Other studies have shown the necessity of looking at patient experience of digital tools to understand how mobile health affects self-management of chronic conditions or changes in well-being [48-50]. In some cases, certain forms of health information may have personal utility for some people even if they lack clinical utility [51]. Overall, such research shows that further work—such as prescreening or offering hands-on assistance and consultation—is needed to turn a health app or Web-based service such as direct-to-consumer testing into a meaningful tool for an individual patient [52].

Data science holds the potential to offer important predictive and diagnostic information that can be used to improve decisions taken by clinicians to reduce error or support estimates, such as the likelihood of medication adherence or organ rejection [53,54]. Yet, from body temperature to steps taken, heartbeats, and hydration levels, it is not yet clear what the biometric data collected via devices such as wearables or smartphones will mean for medical practice and health practitioners. The same is true for nonmedical grade testing services. Both the quality of the data and the possibilities of data interpretation are relevant here. Commercial devices are often not calibrated to the standards of medical grade devices, particularly if not used exactly as intended, which means that data collected through them cannot be used as reliable evidence for health care decisions. Internet communities and apps that offer peer-to-peer support can also be problematic when inaccurate or purely anecdotal information is shared, for example, how-to-hack Web-based tutorials or the increasing use of YouTube as a platform for disseminating misleading health information or offering problematic interpretations of existing data on conditions such as anorexia and bulimia [55-58].

The complex task of discerning irrelevant, unreliable, or misleading health information from relevant, valid, and clinically actionable personalized health resources and then interpreting and contextualizing these for specific patients and their families is emerging as a significant, and time-consuming, activity for health care providers. In our survey of health professionals working in the region of Schleswig-Holstein, Germany, providers expressed repeated concerns about the increasing amount of time devoted in patient encounters to explaining why data from a Web-based genetic test are not relevant, or why a novel therapy reported on a patient community website is not the best choice for a family member [59]. These findings are echoed by recent reports that have pointed to the need for new and improved decision aids to situate the most personally relevant and high-quality digital tools for patients [28,60]. Although some standardization work regarding this issue is currently undertaken by groups such as the Consumer Technology Association, the creation of new devices, apps, and programs and the demands these pose regarding data interpretations and contextualization continues to exceed regulatory processes and physician workloads.

In this context, data work includes deciding which data or information are reliable and relevant for a given context of a specific patient—including contexts outside of the clinic—to decide which intervention, tool, or device might be appropriate or helpful in a given situation, or in future. Again, this is a complex task. For example, discerning whether data brought in by patients derived from commercial or hacked devices can be clinically relevant involves researching devices, analyzing the information they collect, and deciding if, and how, the information generated could be used to inform individual case decisions. In some instances, such data work could include contacting the company producing the device for more information, or seeking out additional resources to evaluate the reliability of the data generated. The same is true for commercially available genetic testing, or the results derived from nonstandard forms of research occurring on patient platforms, such as in some citizen science initiatives [61].

The work of contextualization also increasingly extends to the analysis of the algorithms used to produce data in the health care context. Algorithms are neither ‘objective’ nor intrinsically neutral and they can exacerbate societal inequities. Biases—regarding race, gender, educational status, body mass index, and so on—are programmed into systems, and the characteristics of datasets that these systems use to learn might reproduce inequities [10,62]. As more and more parts of our lives are being datafied, there is an increasing need for contextualization of the health data gained through Web-based tests, mobile, and digital technologies [63]. This includes making the context of data explicit, and asking questions such as: What data was collected, from whom, and how? What do these data represent, and what do these leave out? How has it been made legible for computation, and what has been lost or gained in the process? Such questions are increasingly necessary given the growing ubiquity of domains of everyday life being understood through computational practices. All of the above forms of evaluation require a significant degree of analytical and computational literacy and reflection on whether a particular process of meaning-making relies on evidence that is accurate and reliable in a technical sense, if it is mostly personal and social, or if it is indeed faulty or misleading [64].

Patients, in addition to health care professionals, are also increasingly participating in specific forms of work, including outside of clinical settings. This is the case, for example, when patients do internet searches and seek assistance in making sense of reports or articles found on the internet, thus engaging in the work of sorting, interpreting, and analyzing diverse and often competing sources of information. Often this type of work is undertaken by family members or caregivers to support a patient’s health care choices. The work of contextualization will remain a persistent challenge in years to come as more devices, apps, health-related services are offered to individuals outside the supervision of medical professionals. As an area that is in need of robust investigation and public debate, it would be productive to have greater involvement by scientific and academic societies in conducting and sharing analysis of how data can and should be used. Although some of this work is already ongoing, such as recent reports addressing the opportunities, risks, and ethical questions associated with use of good artificial intelligence (AI) in health care, or developing specific suggestions that can be taken up by stakeholders and policy makers at national and international levels [65,66], further work is needed on different aspects of the use of big data in medicine. By fostering greater debate, and providing material that is available for lay readership to engage with the stakes of their data engagement, academic scholarship can better support digital literacy in this area.

The prior areas of data work that we have outlined have emphasized the need for a strong awareness of what new data, tests, and technologies are available and how they work. Data-rich medicine highlights a number of ethical issues [11], not least of which is the cross-cutting work of addressing different aims, goals, and interests. As data are increasingly accessible, distributed, revealing, and reidentifiable, ethical concerns pertaining to digital health, large datasets, and precision medicine are multiplying, including issues of consent, protecting participant privacy concerns, and maintaining public trust [68]. Given that many of data-driven practices track new territory in health, questions of power asymmetries and social-economic value are emerging with new relevance [12,69]. An important form of data work thus involves fostering conversations with and across stakeholder groups around these concerns.

As precision medicine moves away from one size fits all approaches to treatment, machine learning approaches are increasingly improving the ability to target patients for specific treatments, such as in the use of DNA methylation to subclassify tumors of the central nervous system [70]. The potential of this work to improve personalized therapies through the use of mathematical models is great, yet both the perceived benefits and the social, economic, and health-related concerns vary by actor [71]. In other words, a provider will likely have a different set of investments in the technology, research, and treatment outcomes than a given patient, a hospital chief executive officer, a pharmaceutical company, or an interested member of the public. A patient might be most concerned about loss of privacy, discrimination, or stigmatization (albeit also interested in disease prevention and better treatment), whereas company representatives might be uneasy about losing exclusive access to datasets and find themselves at odds with community members committed to principles of open access. Thus, a central aspect of data work is creating the spaces for interaction and facilitating conversations between differently motivated parties, such as assisting one actor to understand the concerns of another, or finding novel ways to address specific concerns around discrimination, privacy, or equity.

In the digital era, privacy concerns take on a different configuration than in the paper age [72]. Data work in the context of privacy is not limited to simply informing patients of what happens with their data and information once it has been collected but includes moving beyond the widely accepted ethical principle of respecting patient autonomy [73] to including patients in decisions over what type of information will be collected about them in the first place, and to what end. The General Data Protection Regulation (GDPR) introduces protections that began in 2018 across the European Union (EU; including the United Kingdom), but outside of the EU, there is little agreement on regulatory standards for digital health tools or data protection in research, databanks, and big data [61,74-77]. Despite the overall objective of European harmonization, the GDPR gives member states leeway, for instance, in determining whether patient consent is required for secondary data use in medical research, and in which form [74,78]. These national differences have various practical and normative consequences, most of which have not yet been fully analyzed, as well as different implications for research practice across member states. Legislation in countries where data protection is sector specific, rather than general, such as Health Insurance Portability and Accountability Act (HIPAA) in the United States, has addressed data privacy and security concerns relating to medical information since 1996. Subsequently, the HIPAA omnibus rule of 2013 modified the Act to meet guidelines set by the Health Information Technology for Economic and Clinical Health in 2009. Such efforts have expanded the extent of HIPAA beyond providers and insurance companies to also consider the role of business associates. However, even though concerns surrounding patient privacy and the reuse of health information have long been an important topic, the ability of existing regulation such as the GDPR or HIPAA to fully address the concerns emerging in the age of big data remains unknown [79]. We highlight here that the forms of data work we identify can pose particular challenges for privacy, including: the rapid rate of digital innovation; that decisions need to be made on both on the individual and societal level about which aspects of everyday life should be captured by data in the first place; that harm can occur from data use that is not necessarily illegal [80]; as well as broader concerns about data privacy protection legislation.

How to effectively engage a range of stakeholders, including patients, providers, researchers, and insurance companies in these data work concerns, is an ongoing discussion in both clinical practice and biomedical research [81-83]. One critical area of data work for health care providers and researchers is holding conversations with patients about data collection and privacy to better understand the impact of collecting anonymized patient health data in research [14,84]. Data work includes ensuring that patients are party to the decisions about what information will be included in their records, who the gatekeepers for this information are, and for which goals and for whose benefit this information will be used beyond the realm of individual-level health care decisions. It is critical that these discussions include reflections on how data could potentially be reused in the future, for example, the use of predictive health and frailty scores by insurance companies as mentioned in the vignette, as well as the identification of potential protections to guard against uses of data that could be harmful or exclusionary to patients. Specific conditions of access, reuse, and reidentification need to be identified and continually updated in light of new digital advances.

In particular, digital technologies raise important questions over the access of personal information. Each patient’s needs and interests are influenced by their human, natural, and artifactual environments. An individual’s decision to access his or her electronic health records or use a Web-based genetic testing service is not just a choice made by an atomistic individual but an act shaped by the person’s family ties and social relations, his or her connection to others, and the country in which he or she lives [85]. For example, an individual may want to share and discuss this health information with his or her partner or children [82]. This decision to share and discuss information received is not an afterthought but may well have shaped the decision to obtain information in the first place [86]. This layer of dyadic or multilateral forms of decision making can vary significantly across cultural contexts.

In sum, joining distinct datasets from different types, locations, and ethical standards adds additional layers of deliberation to well-rehearsed ethical considerations. Recognition and fostering dialog around aims and goals and the more complex, potentially shared nature of decision making in the era of big data is a critical form of data work. However, how this can be achieved when data are held in dispersed locations and are diverse in nature is entirely unclear. It will require close communication between the patient and the health care provider to ensure that the built-in decisional pathways offered by data-driven practices do not eclipse individual priorities. One potential way of addressing this concern is to reconsider existing methods for ensuring patient privacy and protection and addressing them through regulatory measures, for example through the GDPR in Europe. According to the GDPR, for personal data to be processed lawfully, either individual consent is required, or a legal authorization has to apply. The most relevant legal authorization in the medical context is the research exemption (Article 89). However, particularly in view of international research collaborations, further work is necessary on how GDPR is implemented across individual countries. To provide an example, in line with Article 89, Germany now allows data processing of pseudonymized data for scientific or historical research purposes or for statistical purposes, at least prima facie, without requiring individual consent. However, neither clear guidance exists as of yet for how these purposes are exactly delineated nor have studies been conducted on how this new legal provision has penetrated research practice and how effects differ from countries that are more restrictive. Countries that have long-term experience with more permissive approaches, such as broad or blanket consent (eg, the United Kingdom) and the processing of genetic data should help to anticipate the implications of the novel practice and to raise the standards for how informed consent can be better operationalized in light of the concerns of big data—also in areas outside of Europe [87].