Access the full text.
Sign up today, get DeepDyve free for 14 days.
Loading next page...
/lp/sage/individual-benefits-and-collective-challenges-experts-views-on-data-lCJNvv32o4
- Publisher
- SAGE
- Copyright
- Copyright © 2022 by SAGE Publications Ltd, unless otherwise noted. Manuscript content on this site is licensed under Creative Commons Licenses.
- ISSN
- 2053-9517
- eISSN
- 2053-9517
- DOI
- 10.1177/20539517221092653
- Publisher site
- See Article on Publisher Site
Abstract
Healthcare provision, like many other sectors of society, is undergoing major changes due to the increased use of data- driven methods and technologies. This increased reliance on big data in medicine can lead to shifts in the norms that guide healthcare providers and patients. Continuous critical normative reflection is called for to track such potential changes. This article presents the results of an interview-based study with 20 German and Swiss experts from the fields of medi- cine, life science research, informatics and humanities of digitalisation. The aim of the study was to explore expert opi- nions regarding current challenges and opportunities related to data-driven medicine and medical research and to provide a methodological framework for empirically grounded, continuous normative reflection. To this end, we developed a heuristic tool to map and structure empirical findings for normative analysis. Using this tool, our interview material points to a polarisation between individualistic and collectivistic orientated argumentations among experts. The study shows that a multilevel analysis is required to deal with complex normative implications of data-driven approaches in medical research and healthcare. Keywords Data-driven methods, digital health, medical research, artificial intelligence, mobile health devices, empirical-normative analysis, interview study, experts, Germany ‘personalised’ medicine has been promoted since the late Introduction 1990s and 2000s in the context of human genetics There is an increased use in medical research and healthcare (Prainsack, 2017: 89). Thus, the data-driven approach, of data-driven methods and technologies. Data-driven which nowadays includes not only biomedical data but methods include in various combinations digital technolo- also increasingly data from mobile health devices gies as well as mathematical and computational techni- (‘mHealth’) such as wearables for clinical monitoring ques. Whereas terms such as ‘digital health’ and ‘digital (Lucivero and Jongsma, 2018), goes hand in hand with medicine’ remain rather broad, vague and normatively the promise of ‘personalised’ and ‘precision’ medicine. neutral, the term ‘data-driven’ puts the accent on the meth- Demand is high for the generation and permanent storage odological approach. Following Shah and Tenenbaum of more personal biomedical and health data to feed (2012), Torkamani et al. (2017) and Hummel and Braun algorithm-driven data analysis. With ever-expanding (2020), we understand data-driven medicine to be ‘medical research and care that rests upon consideration of large amounts of data and deploys algorithmic tools to Department of Medical Ethics and History of Medicine, University guide prediction, prevention, diagnosis and treatment, for Medical Center Göttingen, Germany example in attempts to steer towards precision medicine Faculty of Economics, Law and Social Sciences, University of Erfurt, Germany […]’ (Hummel and Braun, 2020: 2, italics in original). As this working definition indicates, the data-driven approach Corresponding author: in medicine and medical research is closely linked to a Lorina Buhr, Department of Medical Ethics and History of Medicine, specific goal: the implementation of so called ‘precision’ University Medical Center Göttingen, Göttingen, Germany. or ‘personalised’ medicine. The concept of ‘precision’ or Email: lorina.buhr@med.uni-goettingen.de Creative Commons CC BY: This article is distributed under the terms of the Creative Commons Attribution 4.0 License (https:// creativecommons.org/licenses/by/4.0/) which permits any use, reproduction and distribution of the work without further permission provided the original work is attributed as specified on the SAGE and Open Access page (https://us.sagepub.com/en-us/nam/open-access-at-sage). 2 Big Data & Society troves of biomedical data from genomics, molecular genet- temporal and spatial index, for example, indicating an over- ics, digital patient healthcare and other sources in clinical view of issues at a certain time within a particular spatial and non-clinical contexts, massive data repositories are context. In sum, we consider CCNR an approach to be con- now available to developers of personalised medicine ducted collaboratively by different scholars, discourses and (PM). Those massive data and data-driven methods are studies such as biomedical ethics, data ethics, science and expected by some observers to give rise to a new scientific technology studies (STS), sociology and philosophy of paradigm in medicine (Moerenhout et al., 2018). Others technology, and technology assessment (TA), bundling speak of the ‘datafication of health’, expressing a more individual-centred and collective-oriented perspectives sceptical stance regarding data-driven healthcare and arguments together. (Ruckenstein and Schüll, 2017: 262). From this perspective, the ‘datafication of health’ is associated with a kind of Aim and outline ‘dataism’, ‘dataveillance’ or ‘data-driven capitalism’ (Sadowski, 2019). These terms imply that data analysis is The primary objective of this article is to contribute to taking on a role ‘between scientific paradigm and ideology’ CCNR by reporting an interview study we conducted (van Dijck, 2014). Whereas commentators from politics and with 20 experts from Germany and Switzerland. These indi- economics as well as media figures in medical informatics viduals have been recognised for their leading expertise on are strongly enthusiastic about this development, ‘evidence digital society, interdisciplinary research and medicine in is scare for successful implementation of products, algo- the German speaking context. The German context is pecu- rithms and services arising that make a real difference to liar as it has been identified as ‘front runner’ country among clinical care’ (Car et al., 2019: 1). other European countries and WEIRD states with regard to data protection regulation (Custers et al., 2017: 241) – a fact that might be relevant when it comes to data-driven medical Continuous critical normative reflection research and healthcare. The purpose of our study was to As with every new technology, normative ethics should explore the perceptions, appraisals and attitudes of this subject the data-driven approach to continuous critical nor- group regarding data-driven methods and technologies in mative reflection (CCNR) about its advantages, disadvan- clinical practice, medical research and regarding ‘PM’ gen- tages, opportunities, risks, limitations and potential erally. The results allow us to draw a picture of a selection alternatives. This applies all the more because new digital of current normative issues and their assessment with a technologies and methods are being developed at a rapid focus on the German-speaking expert community on data- pace. In the following we provide a short outline about driven methods in medicine. Since the topic is very our understanding of CCNR. CCNR should not only iden- broad, the methodological challenge of our analysis was tify and evaluate current ethical, political and social issues, how to integrate heterogeneity productively while reducing impacts and concerns (‘normative oversight’) – all induced the complexity of the analytical and normative dimensions by the development and implementation of new technolo- to be addressed. To do so, we developed a heuristic tool gies – but also always demand ‘Ethical Foresight allowing us to map, categorise and evaluate complex and Analysis’ (Floridi and Strait, 2020). Furthermore, CCNR diverse statements in a systematic way. The introduction should be sensitive to two further dimensions: identifying and application of this heuristic tool, which we consider a shifts in basic assumptions and concepts and addressing suitable method to support collaborative CCNR, is a sec- public communities’ and stakeholders’ perspectives (c.f. ondary aim of this article. Schicktanz et al., 2012) to capture differing national, cul- We proceed in four steps below. First, we start with the tural and historical contexts, and mentalities to ensure epi- description of six subject areas, which we consider as cur- stemic justice and normative contexualism. For this, rently being intensively discussed in the biomedical data literature review and empirical-participatory studies ethics literature, and which we focused on in our interview- engaging with stakeholders’ and public communities’ study. Second, we explain our methodological approach. views are considered complementary approaches enriching Third, we present our mapping of empirical findings on each other within the context of CCNR. For continuous six ‘levels of normative analysis’ (LoNA), inspired by the reflection, both approaches ought to be conducted at mapping approach of Morley et al. (2020), which itself is regular intervals. A crucial objective of CCNR, at least to inspired by the method of ‘levels of abstraction’ (LoA) by our understanding, is to identify and categorise key norma- Luciano Floridi (Floridi, 2008). Finally, we discuss our tive areas where technological and respective methodo- main findings: two overarching normative argumentation logical developments are progressing. CCNR can provide patterns that show (a) how individual and collective consid- a ‘big picture’ or ‘landscape’ of phenomena and normative erations and goals are addressed and (b) there is a consider- issues as well as principles and values at different normative able stress on collective challenges among experts’ levels (from the individual to the collective) involved. statements. We conclude with some reflection on how this Within the scope of CCNR, those big pictures bear a might require going beyond established schemes in Buhr and Schicktanz 3 bioethics primarily focussing on an individual-centred partnerships in healthcare and research’ and (f) ‘cross- perspective. sectoral big data’ (Xafis et al., 2019: 234). Normative landscape of data-driven approaches Background For our interview study above cited studies by Mittelstadt Various attempts have been made in the last years to sketch et al. (2016), Mittelstadt and Floridi (2016) and Xafis out overviews of the ‘ethical landscape’ of data-driven et al. (2019) functioned as a starting point for selecting rele- approaches. The term ‘ethical landscape’, widely used in vant areas to be addressed in our German interview study. A ethical analysis, indicates the diversity of normative literature research was additionally conducted to integrate issues arising in a specific subject of inquiry. The landscape recent trends, including the German context. After a metaphor also underscores that the subject matter can be second stage of condensation, we decided that eight norma- transformed, (re-)designed and further explored, so bears tive topics are crucial for our study as they cover general, a temporal index. recent topics: (1) the notion and vision of ‘PM’, (2) the use of digital mobile measurement devices (wearables and health apps), (3) the practice or goal of linking medical Previous work on mapping the normative issues and non-medical data, (4) the use of algorithmic systems Two widely quoted examples for ‘mapping studies’ are pro- and artificial intelligence technologies for clinical decision- vided by Mittelstadt et al. (2016) and Mittelstadt and Floridi making, (5) the eventual paradigm shift in medical knowl- (2016). Mittelstadt et al. (2016: 2) aimed to ‘map the ethical edge and the concepts of health and disease and (6) require- problems prompted by algorithmic decision-making’ in ments for regulation and social transformation (see also data-intensive approaches. In their literature review, they Table 2). Two more additional topics in the interviews identified ‘six types of ethical concerns raised by algo- were (7) ‘consent issues in data ethics’ and (8) ‘citizen rithms’: (a) ‘inconclusive evidence’, (b) ‘inscrutable evi- science’. For our article here, we focus on topics (1) to dence’, (c) ‘misguided evidence’, (d) ‘unfair outcomes’, (6), discussed in order below. Topics (7) and (8) proved (e) ‘transformative effects’ and (f) ‘traceability’ (p. 4, to be so extensive and complex that we decided to Figure 1). The first three ‘concerns’ are characterised as present and discuss our results separately elsewhere. ‘epistemic concerns’, the last three as ‘normative concerns’. ‘Personalised medicine’ is the vision of tailoring diagno- In their systematic meta-analysis of academic bioethics and sis and treatment of diseases to the individual patient’s ‘big data’-related literature, Mittelstadt and Floridi (2016) genetic and molecularorganic dispositions. It stands in identified ‘five key areas of concern’: ‘(a) informed opposition to traditional ‘one size fits all’-medicine, by consent, (b) privacy (including anonymisation and data pro- which for example the same drug and dosage is prescribed tection), (c) ownership, (d) epistemology and objectivity to patients who have a similar diagnosis although they have and (e) “Big Data Divides”’ (p. 303). Six additional different characteristics and medical backgrounds. topics relevant in the near future were identified by con- However, the concept remains vague and has been disputed sulted experts: ‘(f) the dangers of ignoring group-level for years. For some, it remains more a buzzword than a ethical harms, (g) the importance of epistemology in asses- concept, ‘[p]romising to make health care more effective sing the ethics of Big Data, (h) the changing nature of fidu- and efficient’ (Schleidgen et al., 2013: 9). For others, per- ciary relationships that become increasingly data saturated, sonalisation in medicine indicates the already implemented (i) the need to distinguish between “academic” and “com- strategy of stratifying patients into medical subgroups (e.g., mercial” big data practices in terms of potential harm to of types of immunological reactions to tumours) with the data subjects, (j) future problems with ownership of intel- consequence that treatments are differentiated by sub- lectual property generated from analysis of aggregated data- groups. Suggestions to modify the term ‘personalised’ sets and (k) the difficulty of providing meaningful access with ‘precise’ or ‘precision’, ‘predictive’ or ‘participative’ rights to individual data subjects who lack necessary (c.f. Flores et al., 2013; Hood and Friend, 2011) are resources’ (p. 303). A third widely quoted overview of nor- attempts to resolve concerns about conceptual deficiency. mative issues and relevant values was provided by Xafis An important question remains whether personalised medi- et al. (2019); this was developed by a working group of cine has the potential to become a general approach in experts with feedback cycles. They identified ‘six key healthcare. domains of big data in health and research’ for which The use of digital mobile measurement devices such as they proposed ‘an Ethics Framework’. These are: (a) ‘open- wearables and health apps allow patients and research par- ness in big data and data repositories’, (b) ‘precision medi- ticipants to be monitored in clinical contexts as well as cine and big data’, (c) ‘real-world data to generate evidence during daily activities. The use of mobile digital technolo- about healthcare interventions’, (d) ‘AI-assisted decision- gies, also called ‘mHealth’, generates massive amounts of making in healthcare’, (e) ‘big data and public-private individual health, location, social and other data. Since 4 Big Data & Society mHealth intensifies the amount and diversity of data in a necessary for it to do so? Samerksi and Müller (2019) substantial and new way, some speak of a ‘mHealth revolu- note in this context the claimed need for a special kind of tion’ (Ganasegeran et al., 2017; Lucivero and Jongsma, data literacy, termed as ‘digital health literacy’, for man- 2018; Waegemann, 2010). Key ethical issues are privacy, aging the digital transformation healthcare system autonomy, data ownership, storage, codification, account- successfully. ability, security, bystanders, third-party use, the role and power of tech companies and transparency about the data Methods collected by these mobile devices (Schmietow and Marckmann, 2019). As previously described, the three frequently quoted Data linkage of heterogenous, individual health and mapping and framework studies and our own literature non-health data follows the idea of combining different review informed the development of an interview guideline types of data and data sources to obtain denser datasets for expert interviews. We employed a semi-structured for analysis at the individual and population level. As expert interview design as a form of ‘exploratory expert Harron et al. (2014: 1) note that in research ‘the success interview’ (Bogner et al., 2014), since the field is currently of such data linkage depends on data quality, linkage changing rapidly and is influenced by various disciplines. methods and the ultimate purpose of the linked data’. Our study is a part of a larger German research project While data linkage has a large potential as a tool in obser- tasked with evaluating, from an ethical and political per- vational research (Bohensky et al., 2010), it can also be spective, a research initiative of the German Federal employed by health insurance companies and private tech- Ministry of Education and Research (BMBF) that is actors to construct predictive models for individual risk pro- intended to strengthen medical informatics in German files. As data linkage uses personal data from different data clinics and university medical centre hospitals. This initia- sources and thus involves heterogenous data actors (such as tive comes late relative to other industrialised countries tech firms, research institutions), two key concerns are such as Great Britain, where biobanks were established in privacy and consent (Boyd, 2007; Xafis, 2015). 2007, or Sweden, where the electronic health record was The use of algorithmic systems and artificial intelligence established in 2012. Thus, our interviews allowed us to technologies such as machine learning for clinical decision- explore expert views and attitudes regarding a still-nascent making regarding diagnosis and treatment raises a plethora transformation process as currently occurring in the of epistemic as well as ethical questions of reliability, German-speaking countries. Ethics approval was obtained explainability, opacity, paternalism, automation bias, trust- from the Local Ethics Committee on Human Research at worthiness, the potential traceability of supposedly University Medical Centre Göttingen (No. 28/7/18). anonymous data, fairness and moral responsibility (Floridi et al., 2018; Goddard et al., 2012; Grote and Berens, Population and recruitment 2020; Mittelstadt et al., 2016). From an ethical point of view, crucial questions revolve around what role algorith- Experts were selected on the basis of our own assessment of mic systems ought to play in clinical decision-making, their prominence in public and expert discussions (e.g. and the consequences of integrating (opaque) algorithmic members of public policy advisory boards). We selected systems into the doctor–patient relationship (McDougall, individuals covering different disciplines: natural sciences, 2019). medicine informatics, data sciences, philosophy and ethics, The application of data-driven methods may lead to a sociology and law. Additionally, we searched for represen- paradigm shift in the generation of medical knowledge. tatives of civic organisations or advocacy groups with wide This raises epistemological questions, with normative competency in digital society, digital medicine and health. implications, because the use of data-driven methods may The experts were identified through Internet and literature change our concepts of health and disease. One key contro- research. Thirty-three potential interview partners were versy points to the potential of an overall shift from caus- contacted. Three could not participate due to conflicting ation to correlation in our expectations about what appointments or illness, four referred us to other persons, scientific knowledge is, with a corresponding transform- one person was not interested and four did not respond. ation of the goals, methods and implications of scientific All interviewees are experts insofar as their practice and inquiry (Skopek, 2018). knowledge centres on in data-driven approaches and appli- Taking all these subject areas into consideration, overall cations in medical research, clinical practice or digital questions of the requirements for legal regulation and society (Grundmann, 2017). The interviews were con- social transformation arise. What changes in society are ducted between August 2019 and March 2020 (Table 1). necessary such that we are able to anticipate and manage All interviewees received the same invitation letter and the rapid digitalisation of the healthcare system? To what signed a written consent form. If requested, the question- extent should data-driven developments in medicine and naire was sent in advance. The interview guide consisted medical research follow societal needs, and what is of four questions regarding professional background, the Buhr and Schicktanz 5 Table 1. Characteristics of the interviewees. Participants n = 20, Length (min) Ø = 61 Minutes Sex Female 8 Male 12 Nationality German 18 Swiss 2 Academic background* Philosophy 4 Social sciences 1 Journalism 1 Law 2 Medicine, psychology 5 Public health, health sciences 2 Natural sciences 4 Mathematics 2 Informatics, data sciences & engineering 5 Current main function Scientist 10 Representative of a civil organisation (non-healthcare) 3 Health association 5 Individual expert 2 * Some interviewees have academic background in two fields. aforementioned six topics (see above) and four questions Our analysis revealed a wide range and complexity of concerning consent models in medical research and normative positions on each of the main topics. To encom- citizen research, but these two latter topics are evaluated pass the various ‘dimensions’ of normative analysis, we elsewhere. The interviews were conducted by telephone paraphrased experts’ statements and collated them into the or, in three cases, in person. The interviews lasted 61 min six main topics and into analytical levels. In the sense of on average and were held in German. Afterwards, the inter- the ‘mapping review’ or ‘systematic map’ (Grant and views were transcribed and interviewees’ personal informa- Booth, 2009: 97–98), this approach can be characterised tion removed. as ‘mapping analysis’. Analytical levels were drawn from the typology proposed by Morley et al. (2020: 7). Referring to Floridi’s (2008) ‘Method of Levels of Abstraction (LoA)’ and to Grant and Booth’s (2009) Content analysis explication of the ‘mapping review’, they proposed We analysed the transcripts of the interviews by methods of six different ‘LoA’: (a) individual, (b) interpersonal, (c) qualitative content analysis drawing on Mayring and Fenzl group, (d) institutional, (e) sectoral and (f) societal. (2019) and Kuckartz (2018). We focused deductively on six Morley et al. (2020: 2–3) describe LoAs ‘as an interface thematic codings (‘categories’) representing the six subject that enables one to observe some aspects of a system ana- areas mentioned above. The categories were differentiated lysed, while making other aspects opaque or indeed invis- systematically by subcodes (i.e. opportunities and chal- ible’. We kept this methodological matrix, but in the light lenges) and complemented with inductive codes. of our normative analysis, we call the stratified classifica- Additionally, thematic categories were supplemented by tions ‘LoNA’ instead of ‘LoA’.Wedefine the distinct the category of academic background of the interviewee. levels, starting from the individual and personal level and The coding list was checked for validity and consistency covering several collective levels as follows. by two peers using parallel peer coding (Table 2). �‘ Individual’ addresses the individual as person with fun- damental rights and moral autonomy, and as a subject The method of mapping the LoNA taking a certain social role (here: patient, physician) Our empirical-normative analysis consisted of two steps. �‘ Interpersonal’ addresses the relationship between two First, we mapped our empirical findings from the interviews persons (here: doctor–patient relationship) using a grid that deploys different Levels of Normative �‘ Group’ addresses the social collective constituted by Analysis (‘LoNA grid’). Second, we highlighted the individuals sharing a certain characteristic (e.g. chronic most relevant findings with regard to a normative analysis patients, elderly people and medical professions) (see Supplemental Material 1) and illustrated them with � ‘Institutional’ addresses commonly shared rules, regula- selected quotations from expert interviews. tions and practices 6 Big Data & Society �‘ Sectoral/organisational’ addresses aspects of social yet, and some even doubted that PM is being used in sectors (e.g. healthcare, insurance) and the activity of Germany at all, calling it a ‘vision for the day after tomor- large organisations (e.g. companies, big tech firms) row’ (D92_21:48). Some experts remarked also the wide � ‘Societal’ addresses the society as a whole or general gap between the promise of PM and its actual implementa- aspects of society tion. Many stressed that the concept and goals of PM are limited in two specific respects. First, the individual ana- By additionally subdividing the LoNA according to lysis of molecular genetics makes sense, if at all, only for ‘opportunities’ or ‘benefits’, ‘challenges’ and ‘risks’ (or a limited range of diseases and fields, such as oncology ‘requirements’ for the topic of regulatory and social and pharmacology. In those limited domains, PM might requirements), we obtained a grid for each subject area help patients and doctors to optimise individual therapeutic that compiles the empirical findings in a clearly structured decisions. Second, the term ‘personalised’ is misleading way. When interviewees repeated their appraisals regard- because the underlying methodological approach is based ing certain topics within the interview, it is registered as on statistical analyses of data from large numbers of one statement to avoid bias due to verbal repetitions. patients, not on ‘individual’ data belonging to one person. However, the grid not only mapped the empirical findings One interviewee put it this way: but also prepared and contributed to the empirical- normative analysis. Thus, the grid itself served as a tool I would always prefer to use the term ‘individualised medi- for normative analysis and reflection. As the grid was cine’, because it’s about the human person. ‘Personalised’ applicable for every subject area, we named it ‘topical does have ‘person’ in it, but actually it’s the mask, so to LoNA grids’ (see Supplemental Material 1). At the same speak, where something comes through. What was always time, the subdivision into six levels allowed a more meant by ‘personalised’ medicine in the first years was to detailed mapping and analysis, which is particularly be as genomically informed as possible…. (D101_16:40) appropriate to the complex field of data-driven medicine. Others see what one could call a ‘personalising’ effect in Results medical practice that arises when personal health data Most of the interviewed experts were highly sensitised to from digital mobile devices are included in clinical differences among various ethical issues and subject decision-making. In sum, there were substantial reserva- areas. This high level of differentiation led to highly reflect- tions and scepticism about the concept and programme of ive evaluations of data-driven medicine; adamantly pro or PM in general, but the advent of a more personal medicine contra positions were not evident. This held true across dis- for individual patients due to the availability of more per- ciplinary and stakeholder boundaries. sonal health data was considered possible (see After mapping the empirical findings using the LoNA Supplementary Material 1, Table 1). grid, we created six tables that provide an overview of exemplary interviewee statements in each subject area Use of digital mobile measurement devices (DMD) (see Supplementary Material 1). The essential results for each area are presented below. In view of the richness of With regard to the application of DMD, a number of issues the material, we decided to focus at maximum on four were identified by our interviewees. A widespread view main findings. We illustrate central findings with experts’ among them was that DMD have clear benefits for health- quotations, which are translated by the authors from the care on the individual and on the group level, that is, for German. individual patients, physicians and patient groups. However, these benefits are associated with new challenges for everyone involved and for the institutional arrangement Concept of PM of clinics and research institutes, as the findings presented below indicate. Strong reservations about the concept of PM as a general approach for clinical practice. The promise of the PM approach was seen to lie in the more precise diagnosis and treatment of previously neglected groups and sub- DMD have much potential to empower patients and support groups through the analysis of molecular and genetic physicians. For the patient, the possibility of self-monitoring data. However, many interviewees articulated reservations, via DMD would, so the expressed view, increase health specifically in terms of feasibility of PM as a general, non- awareness and could improve chronic patients’ self- stratified account in medical practice. PM was widely regulating disease management. It could also motivate judged to be not applicable for all patients and for all type patients in general and certain patient groups to adopt of diseases and thus was not considered a valuable approach healthier lifestyles. The opinion was repeatedly articulated at the institutional level of clinical practice. A number of that the use of DMD ‘activates’ patients and strengthens interviewees stated that PM is hardly being implemented their role in healthcare. Buhr and Schicktanz 7 [S]o the role changes from the somewhat passive role of the Establishing data literacy as a collective practice as condition for patient to a more active role. The patient becomes, so to success. A common view among interviewees was that the speak, his own co-therapist through these possibilities. He plethora of MHD is creating more epistemic and cognitive not only has the…data, he also intervenes concretely. uncertainty, which has to be countered by strengthening There are very different functions. (D91_09:27) data literacy. Strengthening data literacy was seen as a prag- matic response to uncertainty. It was considered to be crit- A common perception was that patients gain control and ically necessary that literacy of MHD accompany the ‘empowerment’ (D99_07:02, D74_:16:42; D1_01:31:51) overall approach of data-driven medicine and healthcare. by being enabled to ‘manage’ and administer their mobile To formulate this positively: if health data literacy is health data (MHD). The prerequisite for this, however, is high, MHD could support the physician in diagnosing and that they have access to their own data. The empowering treating a disease and evaluating its course, thus having par- effect on the individual patient, some experts argued, ticular benefits at the individual level. However, concerns could reduce informational asymmetry in the relationship about data literacy were expressed in manifold ways. between doctor and patient at the interpersonal level. An Improving health data literacy means not only – as advantage for physicians was also mentioned: MHD several interviewees stated – a call for changes on the indi- could help them improve diagnosis, monitoring and vidual and inter-personal level but also challenges for treatment. healthcare institutions and for the healthcare sector. Certain deep-rooted conventions and professional cultures would have to change because establishing data literacy Dealing with massive amounts of health data as challenge on requires the ability to generate individual and shared inter- several levels. Many expected benefits associated with the pretations, strategies for dealing with deviations and appro- application of DMD in healthcare were counterbalanced priate, effective communication of results. For this, as by a number of perceived challenges on the individual multiple interviewees stated, strong interdisciplinary and and interpersonal level. For example, many experts consid- interprofessional collaboration would be necessary. One ered correct and appropriate interpretation of MHD to be a interviewee put it this way: major challenge. The absence of routines and experience with massive amounts of health data could lead to depend- This cooperation has already taken place but can certainly ency, uncertainty or even ‘actionism’, that is, taking action take place much more strongly in the future and perhaps only for the sake of acting, as one interviewee called it new professions or new training courses must also be (D74_22:21). This could negatively affect the patient, the created at this interface in order to ensure this translation doctor and the doctor–patient relationship. Insecurity and of expectations. This is sometimes very, very difficult, uncertainty were seen as becoming increasingly acute, because you come from different ways of thinking and and the more evaluation and assessment of data become you have a different language and you have different the focus of medical decision-making. The following research methods, and it is very challenging to derive a comment illustrates this point: common denominator. (D102_17:56) The doctor-patient relationship changes. Because it makes Interviewee statements expressed the idea that for health both of them more dependent on data collected by others, data literacy to be successful, it needs to be performed as evaluated by others and which could also represent a kind a collective and collaborative practice. For one interviewee, of independent or supposedly independent authority. And data literacy in general should be valued as a ‘cultural asset with that, both are unsure about what the interpretation, […] like the ability to write’ (D83_03:37). This view was the evaluation, the consequences of these data sets could echoed by another participant who suggested that health lit- be. (D74_01:36) eracy and health data literacy should be a part of school and professional curricula (D102_17:30). Several experts pointed out that important factors for the situational interpretation of data are the preparation of data via interfaces and dashboards on the one hand and Challenge for the medical sector due to undetermined role of interoperability at the level of data infrastructures on the big tech companies. The advantages of health data literacy other hand. One expert emphasised that in the design of are less likely to be realised, however, if no high-quality dashboards power structures and guiding factors for inter- data are available, if data repositories are insecure and if pretation are always embedded. The precise governance data storage and use are not transparently regulated. and design of data infrastructures and devices constitute, Addressing the last point, a recurrent concern in the inter- as we would summarise those statements, challenges at views was the lack of transparency at the organisational the institutional and sectoral level in terms of transparency, level, concretely about the role and power of DMD provi- security, and quality of MHD. ders (see Supplemental Material 1, Table 2). 8 Big Data & Society Table 2. Normative subject areas, main topics in questionnaire and categories ( = first-order codes) for deductive coding and analysis. Subject area Main topics covered by questionnaire Category for deductive coding 1 Personalised medicine Attitude towards the concept of personalised medicine Opportunities and challenges concerning (including the evaluation of the state of implementation) the concept of personalised medicine (question no. 10 in questionnaire) (PM) PM_Chances, Challenges 2 Use of digital mobile Evaluation of the use of digital mobile measurement devices Opportunities and challenges of digital measurement devices (i.e. ‘wearables’) for medical research and in clinical mobile measurement devices (DMD) practice (affects the role of patients, doctors and DMD_Chances, Challenges further stakeholders such as technology providers and clinical entrepreneurs; mobile health data) (no. 5) 3 Data linkage medical and Evaluation of the idea and practice of linkage of data from Opportunities and challenges of data non-medical data medical and non-medical contexts for medical research and linkage of heterogenous data practice (the benefit for patients, risks and beneficiaries Linkage_Chances, Challenges of data linkage) (no. 6) 4 Algorithmic systems and AI Evaluation of the use of algorithm-based assistive systems Opportunities and challenges of use of technologies (including artificial intelligence technologies) in clinical algorithm-based assistive systems practice (particular challenges) (no. 7) (AAS) AAS_Chances, Challenges 5 Medical knowledge and Description and evaluation of the effects on medical Medical knowledge (building) building on concepts knowledge generation (underlying methods and data-driven methods methodological decisions, conceptions of health and Meth_Chances, Problems disease) (no. 8) 6 Requirements for regulation Significance of regulation and societal transformation (i.e. Need for legal regulation and societal and social transformation education and literacy) (no. 9) transformation Society Academic background of interviewee Key terms and ideas broader data repositories, which, in turn, could be of Data linkage benefit to individual patient care at the end. In their remarks about the linkage of medical and non- Strong reservations, however, were expressed about data medical data, opinions differed regarding which stake- linkage conducted by third parties such as health insurances holder groups might profit the most from it. or tech companies, such as Google or Apple, providing the technical infrastructure for data storage, e.g., Apple’s Data linkage holds the promise of benefits for individuals and iCloud. First, there is always the risk that individuals may research but comes with risks. As with DMD, the advantages become traceable within such data sets. Second, by making and disadvantages mentioned spanned the individual, organisa- it possible to link data of specific behaviours to risk-dependent tional and sectoral levels. Several experts assessed that data insurance pricing, there is the risk that the principle of solidar- linkage will mainly bring advantages in clinical practice (diag- ity – foundational to the German public health system – could nosis, treatment, prevention) for individuals, since an enriched be undermined. While the former risk would steadily increase data base would enable doctors and patients to make better due to improved algorithmic data analysis systems (e.g. decisions. As one interviewee put the patient’sbenefit: machine learning), the latter risk would become a reality when health insurance companies start ‘personalising’ their So of course, if you have a comprehensive database, you insurance contracts. Pricing based on personal risk would can always say that the benefit for the patient is there. decrease costs for socially advantaged groups, so it would Because you have a larger decision-making basis. Of possibly benefit some individuals and groups, but because it course, that is an advantage. (D86_05:10) would also increase costs for the socially disadvantaged, it would disadvantage others. This risk must be seen in the To expand on this point: New insights into individual national and historical context of the German health insurance courses of disease and hidden variables would be enabled system, as one interviewee considered: by linking individuals’ clinical health data with data on their daily real-world and social-media activities. At the sectoral level, some experts argued that the domain of It is important that the principle of solidarity in health insur- health-services research would benefit from systematic ance not be undermined. And we have to think very care- linkage of medical and non-medical data providing much fully about what forms of insurance we want to decouple Buhr and Schicktanz 9 [from the principle of solidarity] and what forms we don’t performance. A few interviewees even attributed to AAS want to decouple. This is speaking very, very generally the potential of compensating for human fallibility. As one and is naturally a German way of seeing things. interviewee said: (D83_03:16) I believe that in order to work around these deficits we need To sum up, data linkage conducted in contexts of medical AI and Big Data very urgently, because I have seen, more care and research was considered to be of possible advan- than once, mistakes made because someone did not per- tage to the individual patients, when data-driven research ceive or wrongly perceived certain things, which through methods indeed lead to new and valid insights. On the Big Data and AI possibly would have been avoided. So, I hand, data linkage performed by third-parties would think if we say yes to digitalisation in diagnostics and create serious challenges or even dangers for individuals therapy, we also have to say yes to digitalisation in the (traceability) and – at the sectoral level – for the ethical evaluation of these findings. (D85_04:25) pillar of the solidarity-based insurance system (see Supplemental Material 1, Table 3). With regard to the perceived overall benefit, a slight ambiguity is present in the answers. Some interviewees Algorithmic-based assistive systems (AAS) stated that AAS could support conversation and dialogue When the interview partners were asked about opportunities between physicians and patients, in this way having a ‘per- and risks associated with the application of algorithmic sonalising’ effect, and provide a benefit for the interper- assistive systems in clinical practice, one distinct advantage sonal level. Doctors would be able to give more accurate and a number of critical issues were identified. advice on the individual level, and the interpersonal rela- Interestingly, concerns expressed about AAS replicated tionship would profit from using assistive systems. and reinforced the interviewees’ concerns about digital Others considered the application of ASS as factor for mobile measurement devices, MHD and data linkage. ‘automation bias’ (D95_20:24) at the individual level stemming from over-reliance on standard models to the neglect of a patient’s physical and personal individuality. AAS substantiate concerns regarding digitalisation. The posi- Automation bias was explained by one interviewee as tive side of AAS, in the judgement of many interviewees, follows: is the potential for supporting clinical decision making, whether it be in diagnostics or in health prediction. It was […] [T]he problem of automation bias […] means the ten- stated that AAS would in particular support physicians in dency of people to rely on what is displayed to them diagnostics by providing a kind of second opinion and without necessarily understanding it (D95_20:25). accelerating the diagnostic process. A number of interview partners mentioned that the machine learning systems currently in use for pattern recognition (in dermatology The interviewee added that the costs associated with reject- and radiology for example) have demonstrated good ing a recommendation made by software (D95_20:28) Table 3. Results: current normative benefits and challenges across normative levels of analysis and along six subject areas. Subject area Results (1) Personalised medicine Strong reservations regarding the concept of personalised medicine. However, possible benefits for individual patients and a few groups of patients in selected domains (e.g. oncology). (2) Use of digital mobile measurement devices DMD have much potential to empower patients and support physicians at the DMD individual level. Uncertainty due to massive amounts of potentially invalid health data. Data literacy at the individual, institutional and sectoral level as a condition for success. (3) Linkage of medical and non-medical data Data linkage holds the promise of benefits for health service research and in turn with individual treatment but comes with risks at the institutional and organisational level. (4) Algorithmic systems and AI technologies Use of algorithmic-based assistive systems substantiate concerns regarding digitalisation. Support for physicians (individual level). (5) Medical knowledge and concepts Data-driven methods as a new resource for medical knowledge production (institutional level), however built on shaky foundations. (6) Need for regulation and social transformation Urgency of broad public discourse on data-driven healthcare and medical research, thus data-driven approaches as challenges for society as whole. 10 Big Data & Society would be of regulatory concern. Thus, the risk of automa- Requirements for regulation and social tion bias plays out at the individual level but could be rein- transformation forced or mitigated by institutional governance structures. One dominant theme of almost all interviews was the need Many experts also mentioned concerns at the institu- for broad societal deliberation on digitalisation in general tional, respective regulatory level. First, there is a substan- and public discussion about the effects and governance of tial problem with quality and control of training data used the digitalisation of medicine and healthcare in particular. for training and preparing ASS before implementation in clinical practice. Second, ASS are still prone to considerable error rates related to specificity and sensitivity (false positive and Urgency of public discourse on data-driven medicine. The need false negative results). These limit the performance of for juridical and legislative regulation was often mentioned, ASS. In sum, not only digitisation and the explosion of such as registration and authorisation for health devices and health data were seen as a challenge for all stakeholders (international) standardisation of health data quality and and professionals in healthcare, but also the computational measurements. Beyond consideration of technical aspects methods of their analysis (see Supplemental Material 1, such as encryption, interoperability and disclosure of appli- Table 4). cation programming interfaces (APIs), many experts emphasised the need for an intense public discourse about data-driven medicine. Public deliberation should deal with Medical knowledge and concepts questions such as data ownership, profit and transparency. One participant described these concerns as follows: When interview partners were asked about potential changes in the way medical knowledge is produced with But in fact, for me this is not at all a purely legal question or support of data-driven methods, some interviewees empha- a question of data security. If one wants to advance this Big sised that these methods could help to discover previously Data model in medicine, I really see a responsibility for undiscovered correlations. Others noted that such correla- society as a whole. […] Because the sober, legal view tions cannot replace knowledge about causal relationships alone doesn’t get us any further at this point. As a lawyer and should be used with caution. you might say, well, since the patient has been transparently informed, [data gathering] must be ok. But the question is New resource for knowledge generation but on shaky what does that do to our society? What does that do to foundations. A few interviewees saw epistemological the individual patient? (D86_05:15) value in correlations and models computed by machine learning methods. However, the majority of our interview This view was echoed by another interviewee who stressed partners stated that data-driven methods would change the necessity of a ‘dialogue across society’ about artificial knowledge production profoundly. The interviewees eval- intelligence and where it should be applied or permitted. uated the actual and potential impact of data-driven This should be undertaken not only for narrow clinical con- methods on the conceptualisation of health and disease dif- texts, but also, for example, for the national healthcare ferently. Overall, there was a sense that a change in the system as a whole (D94_08:53). conceptualisation of ‘health’ and ‘disease’ had already One interviewee stated that data-driven medicine should started with human genomics and had led to a kind of trans- not be taken ‘as a purely medical phenomenon […]but forming the dichotomical arrangement of the concepts of rather, [as] a social process […] which also, and rather health and disease into a polar continuity. As one inter- belatedly, is gaining a foothold in medicine’ (D85_19:30). viewee said: The interviewee added that for him ‘the ethical discussion about this has long reached the public and politics’, and ques- I think that a change is also occurring maybe now not only tions on ‘regulations for medicine’ are not in the ‘foreground’ through Big Data but generally through technology, that but rather are embedded in general debates on the ‘possibil- you start thinking of being healthy as only for the time ities’ that ‘AI and big data offer us [and] that we have to being, being latently ill. (D102_17:22) learn to deal with’ (D85_19:40). Another interviewee stated that is important to reconsider the whole subject of data-driven Since the German labour system and health insurance approaches in medicine in the light of societal justice and systems are – as one interviewee pointed out – built related issues of discrimination and inequalities: on the dichotomy of health and disease, shifts in the conceptualisation of health and disease in the light of Questions of how all these digital technologies change acces- data-driven methods entail organisational implications sibility, what new forms of discrimination or bias there are, and crucial challenges (see Supplemental Material 1, and how existing injustices or inequalities in the health care Table 5). system [can] be remedied if necessary (D95_40:13). Buhr and Schicktanz 11 In sum, these results show that experts were unanimous in solutions. In terms of our matrix, these benefits would be the view that data-driven approaches in medical research located at the institutional and sectoral levels. Ethical reflec- and care and the overall aim of ‘PM’ requires detailed tion in this strand of discussion is primarily directed at iden- and differentiated analysis, assessing transformative tifying and problematising ethical implications and dangers effects and norms and values at stake (Table 3). for the individual (patient, user). These individual chal- lenges and risks are examined in the light of the classical Discussion principles of medical ethics (Beauchamp and Childress, 2012), namely ‘respect for autonomy’, ‘beneficence’, ‘non- We turn now to two observations that stand out clearly from maleficence’ and ‘justice’. The data ethics literature the interview material. The first observation is that the argu- recently added a fifth principle, ‘explicability’ (Floridi mentation patterns regarding potential opportunities and and Cowls, 2021). The expected emphasis would be then problems follow a classical methodological and mostly on the individual-related principles of respect for social-ontological division in political theory: the individual the autonomy of the patient (translated as the imperative versus the collective. of respect for individual privacy and self-determination) and individual harm. Interestingly, the argumentation of our interview part- Individualistic and collective orientations in the ners showed exactly the opposite tendency: potential bene- argumentation fits were mainly identified on the individual level, while for Our mapping analysis revealed that the German experts inter- the collective level, significant challenges – and fewer ben- viewed showed a tendency to focus on the individual, institu- efits – were noted in all subject areas. In general, their eva- tional and sectoral/organisational levels in their statements. luations were predominantly oriented towards collective The societal level was also addressed, but less often. and institutional concerns. Interestingly, the interpersonal (doctor–patient relations) and The German author Julie Zeh’snovel Corpus Delicti the group levels were addressed less often. Especially regarding (2009) illustrates well the inversion of perspectives digital mobile measurement devices (i.e. wearables) and MHD, observable in our German interview material. In interviewees focused on the individual level (see Table 2 in Corpus Delicti, which is now taught in schools in Supplemental Material 1), mentioning individual benefits for Germany, a futuristic, benevolent state dictates indivi- the patient and the doctor. Furthermore, our findings indicate duals’ behaviour and bodies to achieve the common a remarkable tendency among the participating experts to good of a healthy collective. The system is viewed critic- point out problematic issues and possible negative conse- ally by Mia, a biologist, who comes to the realisation that quences on the institutional and organisational levels. So, the state not only invades privacy but also makes mis- there are two normative orientations present in the experts’ takes at the cost of human lives and individual rights. answers. First, the interviewees engaged in individual-centred The novel can be seen as dystopian comment on the col- argumentation and evaluation of benefits, chances and draw- lective objectives of health, data sharing and a misinter- backs for the individual (e.g. ‘empowerment’ of the patient, preted solidarity that overrides individual interests, physician support in diagnostics). On the other hand, intervie- rights and benefits. Hence, the novel presents a classic- wees’ arguments focused in substance on the impacts on and ally liberal warning about the collective benefits pro- challenges for institutional and organisational structures and vided by technology: they necessarily collide with their normative pillars (i.e. solidarity, justice and equality). individual benefits and rights. In contrast, our expert This collectivity-orientated argumentation was in a few cases study points in another direction, namely that collective explicitly underlined by references on the ethical and political benefits are in no way inevitable. principles of ‘solidarity’, ‘justice’ and ‘public discussion’.To In sum, our findings indicate a need for bridging the our understanding, these political-ethical principles as well as assessment of individual benefits and drawbacks and the the focus on the individual served as background for problema- assessment of institutional, organisational and societal tising the long-term implications of data-driven medicine. impacts when reflecting on the various facets of data-driven medicine (c.f. Williams, 2005). This result broadly supports the relevance of work that considers the collective implica- Individual benefits and collective challenges tions, for example, ‘community solidarity’ in healthcare In bioethics and data ethics literatures, there seems to be a (among many others, see Biller-Andorno and Zeltner, tendency to follow those health policy and technology- 2015; Davies and Savulescu, 2019). It is therefore not sur- focussed campaigns that claim that data-driven methods prising that in recent bioethics literature, PM has itself will improve healthcare and offer great potential for become a target of a critique that affirms the foundational medical research. Unquestioned are the underlying assump- – and not merely the institutionalised – role of the principle tions that society needs high-performance digital technol- of solidarity in many European healthcare systems ogy and tech companies need to produce new healthcare (Martinovic, 2019; Prainsack, 2017). ́ 12 Big Data & Society data-intensive devices (i.e. wearables) as well as AAS (i.e. Common goods and goals machine learning and AI systems) in diagnostics and treat- To our understanding, the classical liberal–communitarian ments, patients’ individual health status and norms might be debate in political ethical theory shines through here (c.f. overlooked due to an overreliance on statistical models. The Christensen, 2012; Taylor, 2003). In brief, this debate temptation is strong to give statistically ‘normal’ values the deals with the question of whether society can and should status of medical norms, thereby overriding individual agree on a substantive formulation of the common good values and subjective perceptions regarding health or body in order to form a just and peaceful social life (a communi- status. Third, public debate should influence the fields of tarian position would say yes, a liberal position would say application, the roles and the purposes of high-performance no). When applying this perspective to our topic, it raises technologies in medical care and research. the following questions. First, is there some common or col- Our interview-based study was designed to help to trace lective good to be realised through using data-driven medi- current normative assessments in order to make an empiric- cine? Second, if so, how we should weight individual goods ally based contribution for the ongoing critical examination (e.g. empowerment and self-determination) relative to col- of data-driven approaches in medicine. The method of the lective goods (e.g. saving costs in public health or being LoNA grid helped in describing and mapping the internationally successful in medical research)? Finally, complex normative landscape in six subject areas within how should we deliberate about these questions (cf. van the broad field of data-driven medicine and along six Beers et al., 2018)? Our interviewees clearly expressed LoNA, revealing underlying patterns and orientations in their perception of a need for societal deliberation to interviewees’ arguments. Our LoNA- grids revealed a clarify the concrete goals of data-driven medicine and tech- strong motivation among experts to reflect on the conse- nologies such as artificial intelligence. Furthermore, issues quences of the data-driven method for the individual, as such as (health) data literacy were conceptualised by the well as to reflect on transformative effects in (medical) experts both as individual empowerment and as collabora- science, in the organisations of healthcare and health insur- tive practice and thus as a kind of collective good. ance and in society as whole. Furthermore, our study Individualistic ethics, which builds on the premise of the showed that there is high demand among experts for autonomous individual, having a focus on individual bene- public deliberation about subjects concerning data-driven fits, and which discounts the effects of social connected- medicine far beyond legal regulations and ethical recom- ness, reaches its limits in the confrontation with mendations on data security and technical standardisation. inherently collective phenomena (cf. Callahan, 2003; We conclude with a methodological proposal. We Etzioni, 2011). Future normative approaches should suggest that CCNR on data-driven medicine and healthcare strengthen the theoretical conception of individual benefits would benefit from the following three directives (see also and costs linked to the instrumental and enabling powers of Supplemental Material 2): collective actors (Beier et al., 2016). 1. Reflect on basic notions and contested concepts that pre-structure various lines of argumentation. In our Conclusions study, the focus was on PM and the concepts of After a decade of discussion of the ethical and social implica- health and disease, but it also includes a political dimen- tions of PM, and after about half a decade debating the data- sion: which actors make which concepts practically driven approach and various digital technologies in use now, relevant and to what extent do certain conceptualisa- we need a more refined assessment of their normative dimen- tions express certain interests? sions. In recent contributions to social science and social 2. Cover and integrate several subject areas related to theory, the advance of new forms of normativity has been technological and methodological developments. In identified as ‘digital normativity’ (Fourneret and Yvert, our study, we covered six subject areas in the context 2020) or ‘algorithmic normativities’ (Grosman and of data-driven medical research and healthcare. Reigeluth, 2019). Interestingly, many interviewed experts Subjects that stretch across two or more areas should expressed the similar idea that data-driven medicine would be addressed as well. produce new kinds or accents of normativity in the medical 3. Elaborate different LoNA allows one to identify underlying and health context. Against this backdrop and to round off (social) ontological patterns and normative frameworks in our contribution for CCNR, we propose transferring argumentation. We suggested six levels – individual, inter- Fourneret and Yvert’s concept of ‘digital normativity’ to the personal, group, institutional, organisational and societal – medical sphere as ‘digital health normativity’. The results of and we identified individualistic as well as collective and our interview study indicate what it covers. First, there is societal orientated argumentation patterns. the risk of automation bias as described and explained by one interviewee in situations of clinical decision-making. The LoNA grid provides a heuristic tool to approach sys- Second, there is a risk that with the application of tematically normative analysis with these three rules of Buhr and Schicktanz 13 Beier K, Jordan I, Wiesemann C, et al. (2016) Understanding col- analysis. Bioethics and data-ethics research should critically lective agency in bioethics. Medicine, Health Care and and continuously examine existing and imminent develop- Philosophy 19(3): 411–422. DOI: 10.1007/s11019-016-9695- ments in the field of data-driven methods and technologies. Biller-Andorno N and Zeltner T (2015) Individual responsibility Acknowledgements and community solidarity–The Swiss health care system. The The interview study was conducted within the Work Package New England Journal of Medicine 373: 2193–2197. DOI: ‘Ethics & Stakeholders’ as a part of the HiGHmed: Heidelberg – 10.1056/NEJMp1508256. Goettingen – Hannover Medical Informatics consortium Bogner A, Littig B and Menz W (2014) Interviews mit Experten: (HiGHmed). Scott Stock Gissendanner deserves a special thank Eine praxisorientierte Einführung. Qualitative for his thorough language editing. Sozialforschung. Wiesbaden: Springer VS. Bohensky MA, Jolley D, Sundararajan V, et al. (2010) Data linkage: A powerful research tool with potential problems. Funding BMC Health Services Research 10(346): DOI: 10.1186/ This work was supported by the HiGHmed consortium, which is 1472-6963-10-346. funded by the German Federal Ministry for Education and Boyd KM (2007) Ethnicity and the ethics of data linkage. BMC Research (Bundesministerium für Bildung und Forschung, Public Health 7(318). DOI:10.1186/1471-2458-7-318. BMBF) (Grant No. 01ZZ1802B). We acknowledge support by Callahan D (2003) Individual good and common good: A commu- the Open Access Publication Funds of the Göttingen University. nitarian approach to bioethics. Perspectives in Biology and Medicine 46(4): 496–507. DOI: 10.1353/pbm.2003.0083. Car J, Sheikh A, Wicks P, et al. (2019) Beyond the hype of Declaration of conflicting interests big data and artificial intelligence: Building foundations The author(s) declared no potential conflicts of interest with for knowledge and wisdom. BMC Medicine 17(143). DOI: respect to the research, authorship and/or publication of this 10.1186/s12916-019-1382-x. article. Chen CP and Zhang C-Y (2014) Data-intensive applications, chal- lenges, techniques and technologies: A survey on big data. ORCID iDs Information Sciences 275(10): 314–347. DOI: 10.1016/j.ins. 2014.01.015. Lorina Buhr https://orcid.org/0000-0002-0718-6026 Christensen E (2012) The Re-emergence of the liberal- Silke Schicktanz https://orcid.org/0000-0001-9627-752X communitarian debate in bioethics: Exercising self- determination and participation in biomedical research. The Supplemental material Journal of Medicine and Philosophy: A Forum for Bioethics and Philosophy of Medicine 37(3): 255–276. DOI: 10.1093/ Supplemental material for this article is available online. jmp/jhs012. Custers B, Dechesne F, Sears AM, et al. (2017) A comparison of Notes data protection legislation and policies across the EU.ID 3091040, SSRN Scholarly Paper, 20 December. Rochester, 1. The terms ‘data-driven’ and ‘data-intensive’ are used inter- NY: Social Science Research Network. DOI: 10.2139/ssrn. changeably in this paper. A useful overview of data-intensive applications is provided by Chen and Zhang (2014). Davies B and Savulescu J (2019) Solidarity and responsibility in 2. WEIRD (western, educated, industrialised, rich and health care. Public Health Ethics 12(2): 133–144. DOI: 10. democratic). 1093/phe/phz008. 3. Based on a systematic literature review on definitions of perso- Etzioni A (2011) On a communitarian approach to bioethics. nalised medicine, Schleidgen et al. (2013: 23) provide the fol- Theoretical Medicine and Bioethics 32: 363–374. DOI: 10. lowing synthetical definition: ‘PM [Personalised Medicine] 1007/s11017-011-9187-8. seeks to improve stratification and timing of healthcare by uti- Flores M, Glusman G, Brogaard K, et al. (2013) P4 medicine: lising biological information and biomarkers on the level of How systems medicine will transform the healthcare sector molecular disease pathways, genetics, proteomics as well as and society. Personalized Medicine 10(6): 565–576. DOI: metabolomics’. 10.2217/pme.13.57. 4. The Medical Informatics Initiative (MII) started in 2016 and Floridi L (2008) The method of levels of abstraction. Minds and will have received 300 million EUR by 2026 in funding: Machines 18: 303–329. DOI: 10.1007/s11023-008-9113-7. https://www.medizininformatik-initiative.de/en/start (accessed Floridi L and Cowls J (2021) A unified framework of five princi- on 30 March 2022). Our study was a part of the Ethics & ples for AI in society. In: Floridi L (ed.) Ethics, Governance, Stakeholder project of HiGHmed, which is one of four constor- and Policies in Artificial Intelligence. Philosophical Studies tia under the umbrella of the MII. Series. Cham: Springer International Publishing, 5–17. DOI: 10.1007/978-3-030-81907-1_2. References Floridi L, Cowls J and Beltrametti M et al. (2018) AI4People—An Beauchamp TL and Childress JF (2012) Beauchamp, T: Principles ethical framework for a good AI society: Opportunities, risks, of Biomedical Ethics, 7th edition. New York, Oxford: Oxford principles, and recommendations. Minds and Machines 28(4): University Press. 689–707. DOI: 10.1007/s11023-018-9482-5. 14 Big Data & Society Floridi L and Strait A (2020) Ethical foresight analysis: What it is Engineering Ethics 22(2): 303–341. DOI: 10.1007/s11948- and why it is needed? Minds and Machines 30: 77–97. DOI: 015-9652-2. 10.1007/s11023-020-09521-y. Mittelstadt BD, Allo P, Taddeo M, et al. (2016) The ethics of algo- Fourneret E and Yvert B (2020) Digital normativity: A challenge for rithms: Mapping the debate. Big Data & Society 3(2): 1–21. human subjectivation. Frontiers in Artificial Intelligence 3: 27. DOI: 10.1177/2053951716679679. DOI: 10.3389/frai.2020.00027. Moerenhout T, Devisch I and Cornelis G (2018) E-health beyond Ganasegeran K, Renganathan P, Rashid A, et al. (2017) The technology: Analyzing the paradigm shift that lies beneath. m-health revolution: Exploring perceived benefits of Medicine, Health Care and Philosophy 21: 31–41. DOI: 10. WhatsApp use in clinical practice. International Journal of 1007/s11019-017-9780-3. Medical Informatics 97: 145–151. DOI: 10.1016/j.ijmedinf. Morley J, Machado CCV, Burr C, et al. (2020) The ethics of AI in 2016.10.013. health care: A mapping review. Social Science & Medicine Goddard K, Roudsari A and Wyatt JC (2012) Automation bias: A 260: 113172. DOI: 10.1016/j.socscimed.2020.113172. systematic review of frequency, effect mediators, and mitiga- Prainsack B (2017) Research for personalised medicine: Time for tors. Journal of the American Medical Informatics solidarity. Medicine and Law 36(1): 87–98. Association 19(1): 121–127. DOI: 10.1136/amiajnl-2011- Ruckenstein M and Schüll ND (2017) The datafication of health. 000089. Annual Review of Anthropology 46(1): 261–278. DOI: 10. Grant MJ and Booth A (2009) A typology of reviews: An analysis 1146/annurev-anthro-102116-041244. of 14 review types and associated methodologies. Health Sadowski J (2019) When data is capital: Datafication, accumula- Information & Libraries Journal 26(2): 91–108. DOI: 10. tion, and extraction. Big Data & Society 6(1): 1–12. DOI: 1111/j.1471-1842.2009.00848.x. 10.1177/2053951718820549. Grosman J and Reigeluth T (2019) Perspectives on algorithmic Samerksi S and Müller H (2019) Digital health literacy – Thesen normativities: Engineers, objects, activities. Big Data & zu Konzept und Förderungsmöglichkeiten. In: Pfannstiel MA, Society 6(2): 1–12. DOI: 10.1177/2053951719858742. Da-Cruz P and Mehlich H (eds) Digitale Transformation von Grote T and Berens P (2020) On the ethics of algorithmic Dienstleistungen im Gesundheitswesen VI: Impulse für die decision-making in healthcare. Journal of Medical Ethics Forschung. Wiesbaden: Springer, 35–50. 46(3): 205–211. DOI: 0.1136/medethics-2019-105586. Schicktanz S, Schweda M and Wynne B (2012) The ethics of Grundmann R (2017) The problem of expertise in knowledge soci- ‘public understanding of ethics’—Why and how bioethics eties. Minerva 55(1): 2548. DOI: 10.1007/s11024-016-9308-7. expertise should include public and patients’ voices. Harron K, Wade A, Gilbert R, et al. (2014) Evaluating bias due to Medicine, Health Care and Philosophy 15(2): 129–139. data linkage error in electronic healthcare records. BMC DOI: 10.1007/s11019-011-9321-4. Medical Research Methodology 14(36). DOI: 10.1186/1471- Schleidgen S, Klingler C, Bertram T, et al. (2013) What is perso- 2288-14-36. nalized medicine: Sharpening a vague term based on a system- Hood L and Friend SH (2011) Predictive, personalized, prevent- atic literature review. BMC medical Ethics 14(1): 1–12. DOI: ive, participatory (P4) cancer medicine. Nature Reviews 10.1186/1472-6939-14-55. Clinical Oncology 8(3): 184–187. DOI: 10.1038/nrclinonc. Schmietow B and Marckmann G (2019) Mobile health ethics and 2010.227. the expanding role of autonomy. Medicine, Health Care and Hummel P and Braun M (2020) Just data? Solidarity and justice in Philosophy 22(4): 623–630. DOI: 10.1007/s11019-019- data-driven medicine. Life Sciences, Society and Policy 16(8): 09900-y. 1–18. DOI: 10.1186/s40504-020-00101-7. Shah NH and Tenenbaum JD (2012) Focus on translational bio- Kuckartz U (2018) Qualitative Inhaltsanalyse. Methoden, informatics: The coming age of data-driven medicine: Praxis, Computerunterstützung. Weinheim and Basel: Beltz Translational bioinformatics’ next frontier. Journal of the Juventa. American Medical Informatics Association 19(e1): e2–e4. Lucivero F and Jongsma KR (2018) A mobile revolution for DOI: 10.1136/amiajnl-2012-000969. healthcare? Setting the agenda for bioethics. Journal of Skopek JM (2018) Big data’s epistemology and its implications for Medical Ethics 44(10): 685–689. DOI: 10.1136/medethics- precision medicine and privacy. In: Cohen G, Fernandez Lynch 2017-104741. H, Vayena E and Gasser U (eds) Big Data, Health Law, and Martinovic A (2019) Personalised medicine in health care Bioethics. Cambridge: Cambridge University Press, 30–41. systems and EU law: The role of solidarity? In: Taylor C (2003) Cross-purposes: The liberal-communitarian debate. Bodiroga-Vukobrat N, Rukavina D, Pavelic K, et al. (eds) In: Taylor C (ed.) Philosophical Arguments.Cambridge:Harvard Personalized Medicine in Healthcare Systems.Cham: University Press, 181–203. Springer, 181–196. Torkamani A, Andersen KG, Steinhubl SR, et al. (2017) Mayring P and Fenzl T (2019) Qualitative Inhaltsanalyse. In: Baur High-definition medicine. Cell 170(5): 828–843. DOI: 10. N and Blasius J (eds) Handbuch Methoden der Empirischen 1016/j.cell.2017.08.007. Sozialforschung. Wiesbaden: Springer, 633–648. DOI: 10. van Beers B, Sterckx S and Dickenson D (2018) Personalised 1007/978-3-658-21308-4_42. Medicine, Individual Choice and the Common Good McDougall RJ (2019) Computer knows best? The need for value- (Cambridge Bioethics and Law). Cambridge: Cambridge flexibility in medical AI. Journal of Medical Ethics 45(3): 156– University Press. DOI: 10.1017/9781108590600. 160. DOI: 10.1136/medethics-2018-105118. van Dijck J (2014) Datafication, dataism and dataveillance: Big Mittelstadt BD and Floridi L (2016) The ethics of big data: Current data between scientific paradigm and ideology. Surveillance and foreseeable issues in biomedical contexts. Science and & Society 12(2): 197–208. DOI: 10.24908/ss.v12i2.4776. Buhr and Schicktanz 15 Waegemann CP (2010) Mhealth: The next generation of telemedi- justifications. BMC Medical Ethics 16(79): 1–16. DOI: 10. cine? Telemedicine and e-Health 16(1): 23–25. 1186/s12910-015-0070-4. Williams G (2005) Bioethics and large-scale biobanking: Xafis V, Schaefer GO, Labude MK, et al. (2019) An ethics frame- Individualistic ethics and collective projects. Genomics, Society work for big data in health and research. Asian Bioethics and Policy 1(50): 1–17. DOI: 10.1186/1746-5354-1-2-50. Review 11(3): 227–254. DOI: 10.1007/s41649-019-00099-x. Xafis V (2015) The acceptability of conducting data linkage Zeh J (2009) Corpus Delicti: Ein Prozess. 18. Aufl. btb 74066. research without obtaining consent: Lay people’s views and München: btb.
Journal
Big Data & Society – SAGE
Published: Apr 26, 2022
Keywords: Data-driven methods; digital health; medical research; artificial intelligence; mobile health devices; empirical-normative analysis; interview study; experts; Germany