Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Digital phenotyping and the (data) shadow of Alzheimer's disease:

Digital phenotyping and the (data) shadow of Alzheimer's disease: In this paper, we examine the practice and promises of digital phenotyping. We build on work on the ‘data self’ to focus on a medical domain in which the value and nature of knowledge and relations with data have been played out with particular persistence, that of Alzheimer’s disease research. Drawing on research with researchers and developers, we consider the intersection of hopes and concerns related to both digital tools and Alzheimer’s disease using the metaphor of the ‘data shadow’. We suggest that as a tool for engaging with the nature of the data self, the shadow is usefully able to capture both the dynamic and distorted nature of data representations, and the unease and concern associated with encounters between individuals or groups and data about them. We then consider what the data shadow ‘is’ in relation to ageing data subjects, and the nature of the representation of the individual’s cognitive state and dementia risk that is produced by digital tools. Second, we consider what the data shadow ‘does’, through researchers and practitioners’ discussions of digital phenotyping practices in the dementia field as alternately empowering, enabling and threatening. Keywords Data shadow, digital phenotype, data double, digital health, ageing, Alzheimer’s disease This article is a part of special theme on Digital Phenotyping. To see a full list of all articles in this special theme, please click here: https://journals.sagepub.com/page/bds/collections/digitalphenotyping disease research. Such data, derived from genomics, elec- Introduction tronic health records and increasingly from digital In this paper, we draw on work with researchers and devel- sources, is intended to enable the detailed characterisation opers engaged in digital health to explore how digital phe- of individual patients that lies at the heart of ‘precision’ notyping represents the ageing body, and, further, how medicine. While Engelman (2020) argues that there is these representations act as entities in their own right. We both an epistemic and sociological naivety to the belief build on sociological and anthropological work on the that clear disease will emerge from the mining of data, it data self and focus on a medical domain in which the remains a powerful driving impetus, particularly in fields value and nature of knowledge and relations with data like Alzheimer’s disease research in which the connections have been played out with particular persistence, that of Alzheimer’s disease research. As a leading cause of ill-health in high-income countries, 1 Engagement and Society, Wellcome Connecting Science, Hinxton, UK dementia has been the focus of considerable policy and Cambridge Public Health, University of Cambridge, Cambridge, UK research attention since the mid-1980s. Much of this atten- Department of Sociology, Goldsmiths, University of London, London, UK tion has concentrated on Alzheimer’s disease, the most Corresponding author: common cause of dementia. As is the case across biomedi- Richard Milne, Engagement and Society, Wellcome Connecting Science, cine, the collection, sharing and analysis of large volumes Hinxton CB10 1SA, UK. of data is seen as central to the future of Alzheimer’s Email: rm23@sanger.ac.uk 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 between the normal and the pathological, the biological and (2017), Insel, the former director of the US National the clinical, remain profoundly contested (cf Lock, 2013). Institute of Mental Health, argues for acknowledgement A shift in Alzheimer’s disease research and in diagnostic of the limitations of the growing dominance of a focus on criteria in the last two decades has placed biomarkers at the the biological correlates of mental health. He describes a centre of what constituted ‘disease’ and created a drive to risk that recent psychiatric research and diagnostic practice better understand the ‘pathway’ of biomarker change over have been dominated by genomics, pharmacology and neu- time. It has driven a focus on the thresholds at which an roscience at the expense of behavioural assessments and individual can be detected as moving between stages on detailed clinical interactions. Instead, Insel and other expo- this pathway from ‘normal’ to ‘biomarker positive’,or nents of a digital phenotyping approach suggest that it from healthy to preclinical, prodromal and on to sympto- enables researchers and clinicians to use portable tools matic dementia. Research increasingly focuses on under- and informatic techniques to capture the ‘extended mind’ standing the pathology, natural history and epidemiology (Raballo, 2018). of the disease to prevent or delay the process of cognitive Digital phenotyping approaches are based around either decline. The hope is that this move to prevention will active user engagements with tests or assessments, or on counter the long-standing status of the field as a ‘graveyard ‘passively’ collected data generated in everyday life and of drug development’ (Hawkes, 2016). It frequently engagements with the world. The central promise is ‘the involves, however, the identification of people deemed to possibility of continuous measurements. Use of apps, be at ‘high-risk’, those who might be most eligible for clin- phone calls, typing speed, and voice features can be moni- ical trials (Milne, 2018). tored unobtrusively every second over a lifetime, with real- Since the identification in 1993 of the first gene asso- time algorithms checking for alarming transformations’ ciated with susceptibility to Alzheimer’s disease, the (Ebner-Priemer and Santangelo, 2020: 298). As with the ApoE e4 allele, an enduring bioethical debate has taken wider field of psychiatry, neurodegenerative disorders and place about whether and how information about risk dementias including Parkinson’s (Trister et al., 2016) and states should be made available to individuals, and what Alzheimer’s disease (Kourtis et al., 2019) are important the consequences would be of doing so (Post, 1996). As sites of hope and promissory investment. Alzheimer’s disease research and in many cases clinical The development of digital phenotyping tools draws into care have become increasingly dominated by assessments focus the intersection between data practices of disease of the biological state of the brain (Lock, 2013), these detection and those associated with the extraction of the debates have been taken forward into discussions around ‘behavioural surplus’ of surveillance capitalism (Zuboff, the use of biomarkers such as beta-amyloid or tau to 2019). These come together explicitly in work from com- assess an individual’s risk of future dementia (Karlawish, panies including Microsoft, Google and Intel. Microsoft, 2011; Schicktanz et al., 2014). This has spurred a for example, has explored the possibility of using search growing body of work on the experience of living with engine data to develop ‘web search digital phenotypes’ information about Alzheimer’s disease risk (Largent et al., that can be used to detect neurodegenerative disorders 2020; Milne et al., 2018). It has also, however, prompted (White et al., 2018)). Patents for the Google Home device critical commentary on the constraints of biology-centred meanwhile describe a possible use of the device to representations of current and future health that marginalise analyse ‘the unique signatures of the occupants’ for patterns the collection or use of information about environmental indicative of Alzheimer’s (Fadell et al., 2018: 252). These exposures or lived experiences (Brayne and Kelly, 2019; ‘unique signatures’ draw discussion of digital phenotyping Leibing and Schicktanz, 2020; Lock, 2013). into dialogue with the wider body of work on the configur- Questions about the adequacy of representation and the ation and exploitation of the ‘digital subject’ (Goriunova, experience of a life lived in relation to information about 2019). In the following section, we draw on this literature future health are made all the more pressing and relevant to consider the forms of representation and the relation by the emergence of ‘digital phenotyping’ in medical between individuals and data described in the development research and accompanying promissory narratives (Birk of digital tools for the assessment of cognition. et al., this issue). Digital phenotyping approaches are based around the fundamental promise that an individual’s Doubles and shadows – life with data experience of health, ‘is expressed in the digital traces that a person leaves behind’ (Birk and Samuel, 2020: 1873). In The idea that distinctive features of our identity might be the context of psychiatry, leading proponents of the reassembled using the digital tracks and traces of our digital phenotyping approach position it as a response to lives – our ‘unique signatures’– has generated a rich and the concern of many psychiatrists that the field was continuing conversation on the formation of digital subjects moving from being ‘brainless’ to ‘mindless’ (Insel, 2017: and its consequences. Schüll highlights the ‘creative voca- 1215). In his JAMA article setting out his programme for bulary’ (2018: 35) introduced by scholars aiming to digital phenotyping as ‘a new science of behaviour’ capture the intensive datafication of life in western societies. Milne et al. 3 These include the use of terms such as ‘data doubles’ or through everyday interactions and encounters with data col- data selves to describe emergent and temporary virtual/ lection and storage (Zook et al., 2004). Such ‘thick data informational profiles that are aggregated from different shadows’ are not ‘just a way of describing data itself, and data sources, sliced and circulated and through which our increased prowess in measuring, mapping, analyzing, selves become both objects and subjects of power in and visualizing, but a meme that speaks to and produces digital worlds (Douglas-Jones, 2021; Green and new ways of establishing truth’ (Graham and Shelton, Svendsen, Haggerty and Ericson, 2000; Lupton, 2019). 2013: 257). However, as Leonelli and colleagues point Thus, confronted by an advertising company’s elaboration out, the concept also draws attention to ‘an ambiguity and of the use of data to target a specific consumer – herself – a strategic relationality to shadowing processes that paral- Goriunova asks lels the relational nature of data and the multiplicity of motives, goals, and conditions through which data may be “What exactly is this digital entity that she identified as me? construed as (in)significant, partial or complete, (un)intelli- What relation does it have to me? How do I relate to it? gible, or (in)accessible’ (Leonelli et al., 2017: 194). This How is it able to stand in for me and construct a me that focus draws critical attention to how representational appa- attracts advertisements and thus alters me, while still ratuses are assembled, how patterns of illumination and being reliant on my activity? How is it produced outside obscurity are authorised and legitimised, how and why of my awareness, mobilized, and recruited?” (2019: 126) some aspects of the data self are drawn into the light, seen as ‘available, portable, and/or meaningful’ (2017: The mapping of persons to digital traces captured over time 194) and other data made ‘missing, unavailable, or invisi- produces a particular kind of subject formation. However, ble’ (2017: 191), and how this changes over time (cf as Goriunova here suggests, this digital entity acts and McGoey, 2017). alters the subject to whom it relates – as Lupton puts it First, we suggest that the idea of the data shadow extends ‘people and their data make each other’ (Lupton, 2018: discussion of the representational nature of the data object, 5). In this, data doubles echo the discussion of Frank, following Goriunova (2019) in complicating representa- who draws attention to the multiple images and codings tions of data subjects as 1:1 depictions of persons or through which the body is doubled and redoubled in con- subject positions and emphasising the strategic relationality temporary medicine, such that the ‘image on the screen highlighted by Leonelli et al. (2017). Although Gorinuova becomes the “true” patient’ and ‘initial certainty of the treats ‘data doubles’ and ‘shadows’ as interchangeable real (body) becomes lost in hyperreal images that are notions of the indexical digital subject, we argue that the better than the real body’ (in Nettleton, 2004: 669). relational nature of shadows offers possibilities that have Lupton, engaging with the concept of the data double in not yet been explored. Thus, while the notion of ‘doubling’ the case of health data, describes how such abstracted ‘data- concentrates attention on duplication and replication, the doubles’ both categorise and identify ‘at-risk’ individuals, nature of the shadow usefully and intuitively captures the and become materially forceful, ‘feeding back information dynamic and distorted (cf Green and Svendsen, this issue) to the user and encouraging the user’s body to act in nature of data representations. It draws attention to the cir- certain ways’ (Lupton, 2012: 237). As she continues, the cumstances of the production of the data self: the relations data double is part of ‘a continual loop of the production between the properties of an object or figure, how that of health-related data and response to these data’ (Lupton, figure relates to a light source, the ground against which 2012: 237). In her work with members of the quantified it stands and on which the shadow is cast, and the place self-movement, ‘pioneers in the art of living with and of the observer. It is this configuration, and the epistemic through data’ (Schüll, 2018: 35) who represent the contem- strategies that define it that enables a shadow to be porary apogee of self-tracking and monitoring practices, produced. Schüll captures how data becomes part of a ‘loop of reflex- Second, the shadow has a cultural resonance that gives ive recomposition’ (2018: 35) and digital tools provide self- the notion of the data shadow value beyond representational trackers the freedom to engage in projects of self- concerns. As an artistic, literary and cinematic trope, the transformation. This ‘loop’ of (self-) representation and shadow has value for its ability to capture the unease and transformation is both recognised and valued by those concern associated with encounters between individuals involved in the development and capitalisation of data- or groups and data about them (Stoichita, 1997). In these driven tools. It is central to the premise of fitness and ‘well- contexts, the shadow is not simply a representation of an ness’ devices that aim to encourage reflexive interactions absent subject so much as an entity in its own right, a with the data self, whether for health or wealth ‘reality to your consciousness‘ (Bergson, 1913: 54). As (Douglas-Jones, 2021; Lupton, 2019). such, it can form its own object of study, a menacing In this paper, we explore the nature of the ‘data self’ other rather than a represented self. Consequently, ‘we using the metaphor of the ‘data shadow’. The data produce our own data shadow, but do not have full shadow refers to the counterpart to the individual produced control over what it contains or how it is used to represent 4 Big Data & Society us’ (Zook et al., 2004: 169). Such shadows have value and reported here consisted of a domain mapping of academic can be traded in networks of exchange, creating situations and commercial activity and research in the field, incorpor- in which a subject may feel ‘over-shadowed’ by circulating ating published academic and patent literature and presenta- information about them – for example, as they attempt in tions of tools in websites, company webinars and press the contexts of employment, finance or insurance. These releases and conference presentations, covering 30 tools entities are not simply archives capable of revealing under commercial development. This mapping was fol- ‘hidden patterns of action at play in our day-to-day lives’ lowed by 26 semi-structured interviews in 2019 and (Schüll, 2018: 5). Our ‘othered’ data selves here pose 2020. Participant was drawn from companies involved in potential threats to the life chances and choices of indivi- the development of digital tools (n= 7), academic neu- duals, including when they are considered to be predictive roscience and data science researchers (n= 8), clinicians of future problems (Eubanks, 2018; Lyon, 2014; Ruppert, in neurology or old age psychiatry (n= 3) and policy and 2012). research officers working in non-profit organisations Individual lives are thus not simply captured in data, but involved in developing or funding digital phenotyping lived in relation to these data and the futures they ‘fore- tools for dementia (n= 9). Participants were based in the shadow’. In this sense, the concept of the data shadow UK, Europe and North America. Interviews asked about allows us to interrogate how access to services, from health- respondents’ hopes and expectations around the tools they care to insurance, is shaped by the future selves cast for- were involved in developing, their concerns and their wards by data. The absence of such foreshadowing, awareness of ethical considerations related to the use of though, can also be problematic. As Beer puts it ‘When digital tools for the early detection of cognitive decline. we are informational persons – that is to say, when we The second study involved three focus group discussions have become our data – the deletion of our data amounts held in the UK in 2020. Ten UK-based researchers partici- to the erasure of our identities’ (Beer, 2021: 390). pated, drawn from academic neuroscience and data science, Elaborating on this, the lack of a data shadow may limit clinical research, industry and clinical old-age psychiatry access to future-oriented domains such as insurance. The and neurology. All were involved in the development of a ‘active’ life of our data shadows, and the material conse- large project aimed at the development of markers of quences of their absence thus emphasise both the conse- Alzheimer’s disease progression for use in clinical trials. quences of living in relation to and without our data self. While not all were directly involved in the development In juxtaposing such representational and material, of digital tools, the project as a whole aimed to incorporate ‘virtual’ and ‘vital’ elements of the data shadow, we do these into clinical trial practice. These focus groups again not intend to sharpen the distinction between the digital concentrated on expectations and concerns around the and lively, biological processes – between knowledge of future role of digital tools in the assessment of cognitive life and life itself. Such a distinction is far from clear, decline. perhaps least so amongst those people producing the tools In the following sections, we explore how this data can and knowledge to phenotype complex neurological condi- provide insight into how the value of digital phenotyping tions such as dementia. Rather, we aim to explore the differ- is being imagined, and the challenges associated with ent relationships that are forged between living subjects and this. We draw on the perspectives of interviewees who the informational traces of ageing bodies and minds. In this are, in the main, involved in articulating and giving way, we are more interested in the messy relationships impetus to the possible futures of digital health. As such, between knowledge of life and life itself; and in digital phe- we are aware that this focus on hopeful narratives, often notyping as only one iteration of the data shadow, with its oriented towards the creation of promissory value, creates own particular qualities, distortions and effects. our own ‘data shadows’ and absences (cf Leonelli et al., In the following sections, we move on to extend and 2017), not least those associated with alternative ways of illustrate our discussion through our empirical data. We seeing, doing and living Alzheimer’s disease (Leibing, consider what the data shadow ‘is’ in relation to ageing 2014). data subjects, and the nature of the representation of the individual’s cognitive state and dementia risk that is pro- Representing cognition: Casting shadows duced by digital tools. Second, we consider what the data shadow ‘does’, through researchers and practitioners’ dis- In this section, we explore how our interviewees, many of cussions of digital phenotyping practices in the dementia whom are deeply invested in the development of digital field as alternately empowering and threatening. phenotyping and the scientific and clinical exploitation of Our discussion draws on two research studies. The first, the data shadow, conceptualise the promise of their an empirical study of ethical challenges associated with approach and the value, nature and status of the representa- digital detection in dementia, involves work with both tions they are involved in producing. We describe how, in domain experts, reported here, and older adults around interviews, publications and corporate material, the their use of, and experience with, digital health. The work promise of digital phenotyping tools is established Milne et al. 5 through three closely related discourses of epistemic value. emphasised, it may be that a critical feature of digital phe- These discourses emphasise uniqueness and the detail of notyping is not simply the accumulation of information digital representations of individuals; extension or the from passively collected data. Instead, it is in the extraction ability to track an individual over time; and ecological of value through inference, in a manner that echoes – and validity, or the possibility of ‘real world’ measurement. indeed explicit references the wider ‘digital exhaust’ These discourses, we argue, allow us to understand the stra- (Hirschtritt and Insel, 2018). As one interviewee put it: tegic relationality of digital phenotyping’s data shadow. Interviewees and other proponents of digital approaches “We know language changes in later stages without being to capturing cognitive and behavioural states claim that detected. So, that’s one of the reasons why we’re looking these offer the possibility of capturing data that is distinct- that earlier. … But then … the literal data and words you ively representative of an individual. As a consequence, say is not what we care about. It’s what insight that gives they suggest, such tools are able to depict an individual’s us into the change that’s going on in your brain” (Ethics brain health in a manner that is both potentially both officer, non-profit) more profound and more able to matter to people. Thus, for example, interviewees described how the use of Another senior company interviewee described how digital devices is unique to each individual and thus such functional measures – and their algorithmic interpreta- uniquely identifying. As one interviewee put it ‘the way tion – can provide truly ‘personal’ assessments: that people interact with their phone, is like a fingerprint for them’ (staff member, non-profit organisation). This fin- “We have a beta version that we are testing internally … gerprint generated in interaction with devices, in turn, has and the data is fascinating because it shows that it’s abso- scientific and potential clinical value for developers in lutely personal, for myself I can see sleep has a huge excess of its uniquely identifying capacity; for example, impact on my performance. Some other colleagues that through its use to generate a ‘cognitive fingerprint’ do this, they can see that the days that they have physical (researcher, non-profit). As another interviewee described, exercise … you can see a huge improvement in how their such fingerprints may be based on the aggregation of multi- brain capacity improves. We have done this in a small ple digital streams that capture the multiple aspects of a group of people over a six month period, we’re basically ‘neurological condition … like slight behavioural training the AI to understand what factors we need to con- changes, executive function changes, the syntax of your sider.” (CEO, UK-based digital health company) language’ (Researcher, non-profit). As in wider discussions of digital phenotyping (Birk and This promise of better representations of an individual’s Samuel, 2020), the ability of digital representations to act as world, produced by repeated measurement over time, is measures of an individual’s brain health is not universally reflected in the Delphic corporate slogan of the leading accepted amongst our respondents, as we discuss below. digital cognitive testing company Altoida to ‘Know However, for interviewees involved in elaborating the Thyself’, and in the motivations of researchers working promise of digital approaches, only part of their value lies across the field. As one interviewee described: in their individualisation and a persistent ‘fingerprint’.In fact, the value of this representation lies in its instability. “It was our understanding that some of these people were Unlike its physical counterpart, the digital ‘fingerprint’ is saying that they had subjective cognitive complaints, mutable and changes over time. What the fingerprint ‘is’ saying that in their opinion their cognitive ability was in this case is defined by what it can technically ‘do’, declining, but on their standard tasks they were doing just record and follow change, track the process of an individual fine, so a clinician couldn’t say that they had any kind of becoming different from themselves and extrapolate from deficits or impairments … so we tried to understand why this to infer future health. For our respondents, it is this these people were actually talking about their cognition focus on individual fluctuation and change that distin- declining even though the tests that we were using guishes digital phenotyping: weren’t picking up on that, so we just tried to make tests as close to real life” (Researcher, non-profit) “Digital tools enable you to detect change, and will put more emphasis on the role of this fluctuation and this For this interviewee, digital tests have the potential to decline before you actually need to reach a threshold to provide more accurate representations of cognitive state get any intervention.” (Company researcher) that may more closely align with lived experience and ‘real life’. This extract, and its emphasis on individuals’ The move away from thresholds and towards subliminal experiences of decline, highlights the temporal ambitions fluctuations in a continuously reproduced image of the of those developing digital tools, to track change over self has consequences for the forms and content of the time. It also, however, introduces a third key feature of data that make up this picture. As other interviewees digital phenotyping that draws attention to situated 6 Big Data & Society change. The unique fingerprints described by our respon- ‘require subjects to perform a task outside of the context dents and their change over time are conceptualised as of everyday behaviour’ (Dagum, 2018: 1), and emphasise reflecting an individuals’‘real life’ engagement with the the value of collecting data using smartphones “in a world. natural environment” (Dagum, 2018: 2). Dagum’s discus- Being able to track the process of becoming different is sion here closely Insel’s perspective on the development part of the possibility of a new form of knowledge. As one of the ‘new science of behaviour’– and indeed, the two company CEO described to us how: co-founded the mental health and cognitive testing company Mindstrong (Reardon, 2017). ““I think what’s important here is that we have no good bio- In our study, researchers using virtual and augmented markers for the brain … nothing that really is measuring reality approaches to assessment particularly emphasised functional capacity, nothing that’s, we’re measuring the possibility of generating representations of cognition illness but we’re not really measuring the impact that it’s created by ‘making something real life’ (company having on someone’s ability to function, to use their brain researcher). They suggest that digital approaches offer the to solve a task and live a healthy life, so that’s important” possibility of a wider shift towards new ways of seeing: (CEO, US-based digital health company) “I kind of get the sense that sometimes in medicine it’s just, For this individual, articulating their vision for the technol- you know, checking the box. They’ve a cognitive assess- ogy and its potential, digital phenotyping allows the ‘real ment. And they don’t really care what the cognitive assess- measurement’ of the impact of illness through an indivi- ment is, they’re more interested in the biology of the dual’s lifetime. For them, this remedies the limits and blind- disease, because that’s what the treatment is going to spots of existing measures, moving the field towards have an effect. That’s why it’s almost as if seeing the capturing the lived experience of ‘using the brain’ to ‘live pathology and the disease go down is what you want to a health life’. see. And if that has a cognitive effect, then fine. Good. This commitment to continuous measurement in ‘real It’s almost always been secondary to the cognitive change life’ forms the core of what the developers and researchers has always been secondary to the biomarker change. And we interviewed refer to as ‘ecological validity’, a central I think that’s shifting now.” (Researcher, non-profit) epistemological premise of the digital phenotyping enter- prise. For actors in the field, this claim to ecological validity The transition this researcher describes and urges, away posits that digital tools’ ability to provide access to the ‘real from a biomarker-driven approach to ‘seeing the pathology world’ makes their data representations qualitatively differ- go down’ towards a focus on cognitive change is reflected ent from existing ways of seeing and ‘doing’ Alzheimer’s in the approaches of a number of companies and projects disease. As one interviewee put it, ‘it tells us about what operating in the field – the digital health companies it is to be a human being with this brain interacting with Winterlight and MyndYou, for example, echo the chief the world.’ (clinical researcher). Another described how: scientificofficer of Applied Cognition (Dagum, 2018) in describing how their analyses of voice data give ‘eco- “Technology can give you the real-world scenario, continu- logical’ measurements of cognitive state. ous data, understanding a little bit more about how, really, As the extracts above suggest, the development of digital people function and behave and how their condition is.” tools is taking place amid an existing marketplace of (Company researcher) approaches to representing the ageing brain. In positioning their field respondents suggest distinctions or even As this researcher makes clear, the promise of technology sequences in these approaches. The future applications of for them is both temporal, in the form of ‘real-time’ con- digital phenotyping they describe explore relations of com- tinuous data, and ethological, in the form of real behaviour. monality, complementarity and conflict between digital and This suggestion, repeated across interviews, that digital biological ways of knowing and representing brain health. phenotyping captures how people ‘really’ function recapi- For some interviewees, the shadows cast by digital and bio- tulates questions in the wider psychology (and sociology) logical data were complementary, and associated with dif- literature of the relationship between experimental evidence ferent aspects of disease – as one put it: produced in laboratory settings and behaviours ‘in the wild’ (Cicourel, 1982, 1996; Orne, 1962). Indeed, the transfer- “I mean digital data can measure definitely your activities of ability of evidence beyond the laboratory or the clinic has daily living. So I think it’s good for the people who are not I been a persistent question for the neuropsychological mean, who are a little bit, they’re not very serious who are research on which many digital measures of cognition slowly going towards maybe the cognitive impairment build (Chaytor and Schmitter-Edgecombe, 2003; phase. And then, yeah, then the dementia phase. So for Kvavilashvili and Ellis, 2004). Those working on digital early disease prediction I think this is the case that these tools thus suggest the limits of measures and tasks that patients are given the digital data. But if you think that a Milne et al. 7 person [has dementia] you have to somehow take brain The data encounter: Living with shadows images of the patient. That cannot be done with a digital As introduced above, digital phenotyping produces data readout, because to really see if this patient has dementia representations that are both contemporaneous and predic- or not, then you have to look into the brain of the person tive, existing alongside the user while suggesting possible and you have to see that something has happened.” futures. As a result, encounters with data shadows involve (Clinical researcher, emphasis added) both current and future health and illness. This encounter with foreshadowed illness, the possibilities it offers and the harms it may cause, have been a repeated site of For this interviewee, the data shadow produced by digital ethical contestation in the Alzheimer’s disease field phenotyping both ‘definitely’ measures daily life, and (Karlawish, 2011; Post, 1996). In this section, we consider casts forward the future health of individuals ‘slowly how our interviewees described the nature of this encounter. going towards’ cognitive impairment and dementia. The In doing so, we consider how our respondents conceptualise data shadow’s ability to represent, however, is partial – it encounters with data selves as entities that empower and cannot sufficiently illuminate the current state of the brain enable, or that threaten. and, in order to align the clinical data shadow with the disease model articulated in dominant diagnostic definitions of disease, has to be accompanied by measurement of the The encounter that empowers and enables changes in biological markers of brain health. The representations discussed above, produced by and For the interviewee above, the representation of the through encounters between individuals and digital ageing body produced by digital phenotyping is comple- devices are primarily oriented towards the early detection mentary to the biological. For others, the biological model of cognitive change, and the prediction of an individual’s of Alzheimer’s disease meant that the shadows cast by risk of future dementia. For our respondents, the relation- digital phenotyping were emphatically the ‘wrong’ type ship between an individual data subject and these represen- of picture. As one senior clinical researcher put it: tations was often couched in terms that emphasise the possibilities afforded by this future orientation. Across digital psychiatry, developers emphasise the ‘biomedical “I think digital biomarkers are a new way of measuring the virtue’ of empowering or enabling ‘self-care’ (Pickersgill, wrong thing … Alzheimer’s disease is a brain disease, it is 2019). In our study researchers described how datafied not a cognitive disorder. If you want to measure the brain representations of cognitive health produced by digital disease, you measure it directly with biomarkers and tools could give the subject ‘a window on their self’: imaging, you do not measure it on cognition.” (Clinical researcher, emphasis added) “I think giving some data as a window into their own health, physical and mental, is definitely something we would like to see down the line, whether it’s someone that’s older and In criticising digital approaches, this researcher negates the healthy or older and has some cognitive or neuropsycholo- claim to ecological validity – reducing it to an outmoded gical problems.” (Senior academic researcher) focus on cognition, and suggesting that the data captured by digital phenotyping is distorted and unhelpful. They con- As others described, the ability to provide users with the trast this misguided approach to measurement with that ability to ‘see through’ this window to look at their data focus on ‘brain disease’, suggesting that the data produced self is the key step in giving them the power to take action: by assessments such as brain imaging allow direct access to the causes of disease. This speaks to the epistemic and poli- “we want to be able to collect that data and give a better tical commitments that shape the ground on which data assessment of where the patient is, and also by making shadows are cast: the picture of ‘brain disease,’ as that, visualising that data for the user, not necessarily a opposed to ‘cognitive disorder’ underpins the project of patient … we can help them to do things that can help identifying early biological markers, in order to develop them become healthier” (CEO, UK digital health drugs that make Alzheimer’s potentially treatable, even company 1) before cognitive signs of dementia emerge. Such approaches are often tied to the development of therapies, This positive potential of the data shadow to empower reca- in which the object to be illuminated through a clinical pitulates both the promises of digital psychiatry and, assessment may be determined by the target of a drug. indeed, of digital health more broadly (Lupton, 2012). The context in which data shadows are produced is thus Thus, another interviewee explicitly situated the goal of full of coexisting and sometimes competing for projects their company to use routine digital assessments to enable and perspectives. older adults to maintain their cognitive health firmly 8 Big Data & Society within existing data relationships. In this vision, digital cog- Here, we see the possibility that the enabling possibili- nitive assessments simply fit into the new ways of seeing ties of the digital phenotyping data shadow derive from and being associated with digital health: their integration into an architecture of insurance-supported healthcare. This vision, elaborated by researchers in both academic and corporate spheres, is evidenced in corporate “The penetration of Fitbit … for example is really quite development models in the field, in which the challenge advanced into the 60 plus market. So that really is the rise (and uncertain financial return) of ‘health system’ tools vali- of acceptance of personalised data, and I think that increas- dated to regulatory standards for medical devices is accom- ingly people do just kind of by second nature understand panied by ‘lifestyle’ products, which may incorporate the that this is something that is part of your day to day life.” same technical basis, but are oriented towards the more (CEO, UK digital health company) accessible market of health insurance. Narratives of self-care embedded in insurance-driven For this interviewee the relationship between the data healthcare situations situate digital phenotyping firmly subject and a personalised data self, formerly the domain within a particular structure of healthcare provision. In of the self-tracker, has become commonplace. Our respon- that sense, they connect digital health to political discourses dents’ elaboration of the relationship in turn attempts to of preventive health that shift the burden of action away reinforce this idea of digital phenotyping as a quotidian, from the state towards the individual (cf Lupton, 2012) producing data shadows as unremarkable objects that we and seek to embed digital tools for cognitive assessment live with and alongside. In turn, these data shadows, they within technology-enabled projects of self-transformation. suggest, do not simply empower, but enable – they can These visions of empowerment and enablement also, both contribute to improving health and facilitate wider however, emphasise the threat of being absent from data. aspects of life such as employment or insurance. In the In addition to being provided with information they can Alzheimer’s disease case, this is closely tied to the shift use themselves, in the narratives presented by developers, in emphasis towards the early detection of, and intervention the ‘measured’ are enabled to access healthcare and insur- in, disease described earlier. As a company researcher ance services in ways that are not available to the ‘unmea- described: sured’. In this vision of future application, digital phenotyping presents possibilities – as the data shadow becomes supportive of health and facilitative of access to “I think a lot of the traditional tools … use thresholds of the a particular vision of healthcare – but also threatens and output, the score. If you’re below this then you are clinically restricts, in both its presence and absence. classified as this whereas, for a lot of people, they might not be below that threshold yet but they will have experienced a lot of change. They would have fallen from where they The encounter that threatens were previously but according to the standard, traditional, The threats to self and to individual autonomy posed by the tools they don’t deserve any clinical attention because data shadow were raised by a number of our interviewees they haven’t passed that threshold yet.” (Company and, in many ways, return our analysis towards those con- researcher) cerns raised in the bioethics literature around the return of results in the context of Alzheimer’s disease, and the rela- Here, digital tools offer new possibilities for people who tionship between discussions of risk and those of affective experience concerns about changes in their thinking or encounters between individuals and data. Thus, the memory, enabling them to become ‘deserving’ of clinical researcher with whose interview we closed the preceding attention. Such considerations firmly integrate digital phe- section continued by elaborating the need for control over notyping into existing practices of clinical research, and one’s data in light of the potential of this data entity to the spaces of commercial possibility associated with the cause harm if allowed to move. As they put it, early detection of disease (cf Dumit, 2012). However, enab- ling visions of digital phenotyping also connect with, and “If you let people test your cognition for you, then it prob- raises comparable questions to, the role of the data ably waived your privacy goodbye very quickly. If you can shadow in ‘representing’ or standing in for somebody (cf do that independently then you can claim your data and pre- Goriunova, 2019; Lupton, 2019). One key context for this serve your privacy.” (Academic researcher) is insurance, as another interviewee described: In this quote, the researcher emphasises that the relation- “If I put the clock forward 20 years, I would envisage this. If ship between self and data should be just that – avoiding you’ve got health care insurance, whether it be state or intervention or mediation from third parties, with the indi- private, part of your annual assessment will be your perfor- vidual data subject retaining control over both her data mance on cognition tests” (Academic researcher) and the futures it may foreshadow. However, as Milne et al. 9 interviewees repeatedly raised, this intimate relationship is commercial development (cf Dumit, 2012; Milne, 2018), also one that comes with an existential threat to the data for example, in the case of pharmaceutical innovation or subject herself. One respondent described this, drawing insurance markets. However, for some, the threat posed on their experience providing feedback as part of a research by the encounter with the Alzheimer’s data shadow in study: this context has emerged as a barrier to their research and development: “[W]e had a number of people write in and say ‘Participating in a study is too much for me, I see my “the first day that we pitched the idea in [X] we had one of scores going down week by week, I see that my medicine the major investors in [X] went on a crusade to stop us from is not making any difference, this is a lot’, ‘having access developing our technology, and we’ve had that over the last to my data’– because we gave our participants access to seven years as well, keep coming back at us is if there’sno their data –‘having access to my data is just depressing treatment there’s no point in diagnosis, and what you do is me; having the ability to do some things now and unethical. I think I’ve expressed that enough and I still knowing that I’m not going to have the ability to do them believe as a researcher, as a patient, that’s not the right later, is crushing me, and being reminded of this three approach but it is a stance some people take. (CEO, UK times a day is overwhelming, I have to quit’, and so that digital health company) was some very sobering feedback (Staff member, non-profit organisation) The intensity of concerns around risk, and the ability of these concerns to shape the development of digital tools, is A researcher from a company similarly described how a prominent feature of Alzheimer’s disease. It characterises users of their tool will ‘often … reach out and say, like, scientific, clinical and popular discussions of early detection hey, I want to talk to somebody about this’. The concerns and risk prediction in a way that can become forceful in that these respondents highlight, in which people feel over- technological development, as previously in the case of whelmed and emotionally affected by an intimate, pharmacogenetics and biomarker development (Boenink, one-to-one encounter with data, both encapsulates the 2018; Hedgecoe, 2008). In the case of digital phenotyping, threatening nature of the data shadow and draws discussion researchers described the problem of risk communication as back towards that of the bioethical discourses with which ‘one of the kind of big ethical concerns’ (company we started. As another researcher at a non-profit described, researcher), which, for their company, currently precludes this relationship revolves around a core dilemma the possibility of any direct relationship between the indivi- dual and their data shadow. Further, the menace posed by “In terms of detection, if you tell them that they are at this relationship re-emphasises the need for digital pheno- increased risk. OK, you’ve told me this. What can I do typing to capture and be embedded in everyday life, to then to reduce that risk? And obviously in this space we ‘appropriately support people through this’ (company can’t really do that. We can just say there are some risk researcher). Such considerations emphasise that, for indivi- factors that have been associated, but we can tell you how duals and organisations, data shadows cast long into the to reduce your risk by 20 percent or something like that. future, and living with them is a long-term commitment. Can you cope with that? … So, it’s quite a hard topic to discuss with them.” (Staff member, non-profit organization) Working with shadows As this extract highlights, and as introduced earlier, discus- In the preceding sections, we have introduced two dominant sion around the ‘data encounter’ in Alzheimer’s disease – threads in how the relationship between data subjects and and with risk information collected from genetics, family their data selves is conceptualised by researchers and devel- history, or lifestyle factors – has long revolved around the opers working in the field of digital tools for the early detec- potential for psychological harm in the absence of an effect- tion of Alzheimer’s disease. We have used the metaphor of ive therapy (Brayne and Kelly, 2019; Lock, 2013; Post, the data shadow to explore two elements of digital pheno- 1996). This threat persists even as developers attempt to typing and its products in this context – their representa- establish an ‘enabling’ data relationship – and, as this tional function and their role as enabling or threatening respondent describes, remains a ‘hard topic’, despite materials. increasingly well-established clinical approaches to risk For our interviewees – personally, scientifically or finan- communication (e.g. Harkins et al., 2015). Indeed, some cially invested in the futures of digital health – digital phe- respondents suggested that this potential for harm may be notyping approaches draw from the wider promise of data intensified in the context of cognitive and behavioural the possibility of a different way of understanding and data that are seen as ‘a bit of your identity’ in a way that bio- representing the ageing brain. For these researchers, as for logical markers are not (Digital health researcher). Risk and advocates of digital psychiatry such as Insel, the potential its communication serves as an engine for scientific and of the data shadows of Alzheimer’s disease is that of an 10 Big Data & Society individualised, longitudinal and ‘ecological’ mode of repre- early detection do not necessarily indicate the current sentation. The construction of this way of seeing illustrates state of the brain, instead creating data shadows that are how the representational apparatus of Alzheimer’s disease cast forward in time. In the context of research that is is being assembled and which aspects of the disease are moving away from a focus on symptomatic diagnosis of being exposed to, or concealed from, view. It also, disease towards early detection and the identification of however, shows how this process reflects and reproduces the risk of future dementia, analyses of digital traces are a wider discourses around the possibilities and perils asso- means of knowing both the current and (possible) future ciated with inferences made through our digital traces. self. Thus, our respondents highlight how digital phenotyping These considerations point to the emerging complexities and the elaboration of digital data shadows intersects with of digital phenotyping as developers attempt to integrate the existing practices of data collection and analysis, as well collection of digital ‘biomarkers’ by active or passive as the ontological and epistemological commitments of means into existing clinical domains. Our use of the data contemporary Alzheimer’s disease research. Participants shadow to explore these themes, and their representational draw attention to the potential for complementarity but and material articulations, points to the ways in which also tensions between digital data and those that emphasise digital phenotyping requires continued attention to inter- biology, biomarkers and the therapeutic models that target twined ethical and epistemological considerations. The them. These relations further gesture towards the commer- empowering, enabling and threatening aspects of the cial arrangements and health system architectures asso- shadow highlight the challenges associated with controlling ciated with different ways of seeing, and emphasise the or containing the Alzheimer’s disease data self. In addition, persistent forms of strategic relationality through which they suggest the importance of continuing to revisit what is, data shadows are cast, bodies made visible and visualisable and what is not, made visible in the process of representing or left in obscurity (cf Leonelli et al., 2017). the ageing brain, what is gained or lost with the commit- In addition, however, in our respondents’ discussions of ment to longitudinal, situated representation, and what fea- digital phenotyping the ‘data shadow’ is more than repre- tures of ageing continue to remain unseen. sentational, playing a material role in the emergence and promise of new technologies and tools. The Alzheimer’s Acknowledgements disease data shadow has the potential to be materially force- The authors would like to thank the editors of the special issue and ful both in the lives of data subjects and to empower and to the other panellists at the 2020 4S/EASST session for their com- enable – but also to threaten. For many of our respondents, ments on this paper, and particularly the reviewers for their helpful this is captured in both the opportunities the data shadow and considered feedback. We would also like to thank the partici- pants in our interviews and focus groups for their time and affords for ‘self-care’ and the facilitative role that the data contribution. shadow can play as it stands in for the data subject in their interactions with healthcare and insurance. Declaration of conflicting interests Conversely, our respondents highlight the threat this shadow poses to the individuals living ‘in the shadow’ of The author(s) declared no potential conflicts of interest with future dementia, connecting emerging discussions around respect to the research, authorship, and/or publication of this article. digital phenotyping with longstanding bioethical debates (Karlawish, 2011; Post, 1996). This threat looms large in Funding a field in which the question of ‘would you want to know’ is a recurrent feature of popular and policy discus- The author(s) disclosed receipt of the following financial support sions around the early detection of disease. For our respon- for the research, authorship, and/or publication of this article: dents, this threat is to both those living with such shadows, RM and AC’s work was supported by Welcome Trust grants 213579 and 206194. NB’s work was supported by an ESRC and to those looking to establish a space for diagnostic Postdoctoral Fellowship and grant (MR/N029941/1) from the innovation and negotiate the challenges of ‘innovating National Institute for Health Research (NIHR) and the Medical with care’ in Alzheimer’s disease (Boenink et al., 2016). Research Council (MRC). Further, the potential to enable – within a particular political and economic configuration of preventative medicine – also ORCID iDs suggests the potential challenges associated with the Richard Milne https://orcid.org/0000-0002-8770-2384 absence of data, or living without a data shadow as digital Alessia Costa https://orcid.org/0000-0002-3761-9080 phenotyping practices become embedded in healthcare or Natassia Brenman https://orcid.org/0000-0002-6567-2129 insurance markets. The ability of data shadows to empower, enable and References threaten, and the consequences of their absence, draws par- ticular attention to the way shadows are situated in time. As Beer D (2021) A history of the data present. History of the Human our respondents described, digital phenotyping tools for Sciences 34(3–4): 385–398. Milne et al. 11 Bergson H (1913) Time and Free Will: An Essay on the Immediate Hedgecoe A (2008) From resistance to usefulness: Sociology Data of Consciousness. London: George Allen and Company. and the clinical use of genetic tests. BioSocieties 3(2): 183– Birk R and Samuel G (2020) Can digital data diagnose mental 194. health problems? A sociological exploration of ‘digital pheno- Hirschtritt ME and Insel TR (2018) Digital technologies in psy- typing’. Sociology of Health & Illness 42(8): 1873–1887. chiatry: Present and future. Focus 16(3): 251–258. Boenink M (2018) Gatekeeping and trailblazing: The role of bio- Insel TR (2017) Digital phenotyping. JAMA 318(13): 1215. markers in novel guidelines for diagnosing Alzheimer’s Karlawish J (2011) Addressing the ethical, policy, and social chal- disease. Biosocieties 13(1): 213–231. lenges of preclinical Alzheimer disease. Neurology 77(15): Boenink M, van Lente H and Moors E (2016) Diagnosing 1487–1493. Alzheimer’s disease: How to innovate with care. In: Boenink Kourtis LC, Regele OB, Wright JM, et al. (2019) Digital biomar- M, van Lente H and Moors E (eds) Emerging Technologies kers for Alzheimer’s disease: The mobile/wearable devices for Diagnosing Alzheimer’s Disease. Dordrecht: Springer, opportunity. npj Digital Medicine 2(1): 9. pp.263–275. Kvavilashvili L and Ellis J (2004) Ecological validity and twenty Brayne C and Kelly S (2019) Against the stream: Early diagnosis years of real-life/laboratory controversy in memory research: A of dementia, is it so desirable? BJPsych Bulletin 43(3): 123– critical (and historical) review. History and Philosophy of 125. Psychology 6: 59–80. Chaytor N and Schmitter-Edgecombe M (2003) The ecological Largent EA, Harkins K, van Dyck CH, et al. (2020) Cognitively validity of neuropsychological tests: A review of the literature unimpaired adults’ reactions to disclosure of amyloid PET on everyday cognitive skills. Neuropsychology Review 13(4): scan results. PLOS ONE 15(2): e0229137. 181–197. Leibing A (2014) The earlier the better: Alzheimer’s prevention, Cicourel AV (1982) Interviews, surveys, and the problem of eco- early detection, and the quest for pharmacological interven- logical validity. The American Sociologist 17(1): 11–20. tions. Culture, Medicine and Psychiatry 38(2): 217–236. Cicourel AV (1996) Ecological validity and ‘white room effects’: Leibing A and Schicktanz S (eds) (2020) Preventing Dementia?: The interaction of cognitive and cultural models in the prag- Critical Perspectives on a New Paradigm of Preparing for matic analysis of elicited narratives from children. Old Age. New York: Berghahn Books. Pragmatics & Cognition 4(2): 221–264. Leonelli S, Rappert B and Davies G (2017) Data shadows: Dagum P (2018) Digital biomarkers of cognitive function. npj Knowledge, openness, and absence. Science, Technology, & Digital Med 1(1): 1–3. Human Values 42(2): 191–202. Douglas-Jones R (2021) Bodies of data: Doubles, composites, and Lock M (2013) The Alzheimer Conundrum. Princeton: Princeton aggregates. Journal of the Royal Anthropological Institute University Press. 27(S1): 159–170. Lupton D (2012) M-health and health promotion: The digital Dumit J (2012) Drugs for Life: How Pharmaceutical Companies cyborg and surveillance society. Social Theory & Health Define Our Health. Durham: Duke University Press. 10(3): 229–244. Ebner-Priemer U and Santangelo P (2020) Digital phenotyping: Lupton D (2018) How do data come to matter? Living and becom- Hype or hope? The Lancet Psychiatry 7(4): 297–299. ing with personal data. Big Data & Society 5(2): Engelmann L (2020) Into the deep – AI and total pathology. 2053951718786314. Science as Culture 29(4): 625–629. Lupton D (2019) Data Selves: More-than-Human Perspectives. Eubanks V (2018) Automating Inequality: How High-Tech Tools Cambridge: Polity. Profile, Police, and Punish the Poor. New York: St Martin’s Lyon D (2014) Surveillance, Snowden, and big data: Capacities, Press. consequences, critique. Big Data & Society 1(2): Fadell AM, Matsuoka Y and Sloo D, Veron M, and Google LLC. 2053951714541861. (2018) Monitoring and reporting household activities in the McGoey L (2017) The elusive Rentier Rich: Piketty’s data battles smart home according to a household policy. US9872088B2. and the power of absent evidence. Science, Technology, & Available at: https://patents.google.com/patent/ Human Values 42(2): 257–279. US9872088B2/, accessed 10 June 2021. Milne R (2018) From people with dementia to people with data: Goriunova O (2019) The digital subject: People as data as persons. Participation and value in Alzheimer’s disease research. Theory, Culture & Society 36(6): 125–145. BioSocieties 13(3): 623–639. Graham M and Shelton T (2013) Geography and the future of big Milne R, Diaz A, Badger S, et al. (2018) At, with and beyond risk: data, big data and the future of geography. Dialogues in Expectations of living with the possibility of future dementia. Human Geography 3(3): 255–261. Sociology of Health & Illness 40(6): 969–987. Haggerty KD and Ericson RV (2000) The surveillant assemblage. Nettleton S (2004) The emergence of E-scaped medicine? The British Journal of Sociology 51(4): 605–622. Sociology 38(4): 661–679. SAGE Publications Ltd. Harkins K, Sankar P, Sperling R, et al. (2015) Development of a Orne MT (1962) On the social psychology of the psychological process to disclose amyloid imaging results to cognitively experiment: With particular reference to demand characteris- normal older adult research participants. Alzheimer’s tics and their implications. American Psychologist 17(11): 776. Research & Therapy 7(1): 1–9. DOI: 10.1186/ Pickersgill M (2019) Digitising psychiatry? Sociotechnical expec- s13195-015-0112-7 tations, performative nominalism and biomedical virtue in Hawkes N (2016) Sixty seconds on … solanezumab. BMJ 355: (digital) psychiatric praxis. Sociology of Health & Illness i6389–i6389. 41(S1): 16–30. 12 Big Data & Society Post SG (1996) On not jumping the gun: Ethical aspects of APOE and Human Augmentics, Artificial Intelligence, Sentience. gene testing for Alzheimer’s disease. Annals of the New York Abingdon: Routledge, pp.25–38. Academy of Sciences 802: 111–119. Stoichita VI (1997) Short History of the Shadow. London: Raballo A (2018) Digital phenotyping: An overarching framework Reaktion Books. to capture our extended mental states. The Lancet Psychiatry Trister AD, Dorsey ER and Friend SH (2016) Smartphones as new 5(3): 194–195. tools in the management and understanding of Parkinson’s Reardon S (2017) Former US mental-health chief leaves google disease. npj Parkinson’s Disease 2(1): 1–2. for start-up. Nature. DOI: 10.1038/nature.2017.21976. White RW, Doraiswamy PM and Horvitz E (2018) Detecting neu- Ruppert E (2012) The governmental topologies of database rodegenerative disorders from web search signals. npj Digital devices. Theory, Culture & Society 29(4–5): 116–136. Medicine 1(1): 1–4. Schicktanz S, Schweda M, Ballenger JF, et al. (2014) Before it is too Zook M, Dodge M, Aoyama Y, et al. (2004) New digital geog- late: Professional responsibilities in late-onset Alzheimer’s raphies: Information, communication, and place. In: Brunn S, research and pre-symptomatic prediction. Frontiers in Human Cutter S and Harrington JJr. (eds) Geography and Technology. Neuroscience 8: 21. London: Kluwer Academic Publishers, pp.155–176. Schüll ND (2018) Self in the loop: Bits, patterns, and pathways in Zuboff PS (2019) The Age of Surveillance Capitalism: The Fight for a the quantified self. In: Papacharissi Z (ed) A Networked Self Human Future at the New Frontier of Power. London: Profile Books. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Big Data & Society SAGE

Digital phenotyping and the (data) shadow of Alzheimer's disease:

Big Data & Society , Volume 9 (1): 1 – Jan 11, 2022

Loading next page...
 
/lp/sage/digital-phenotyping-and-the-data-shadow-of-alzheimer-s-disease-f08PrUvIUE

References (57)

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
10.1177/20539517211070748
Publisher site
See Article on Publisher Site

Abstract

In this paper, we examine the practice and promises of digital phenotyping. We build on work on the ‘data self’ to focus on a medical domain in which the value and nature of knowledge and relations with data have been played out with particular persistence, that of Alzheimer’s disease research. Drawing on research with researchers and developers, we consider the intersection of hopes and concerns related to both digital tools and Alzheimer’s disease using the metaphor of the ‘data shadow’. We suggest that as a tool for engaging with the nature of the data self, the shadow is usefully able to capture both the dynamic and distorted nature of data representations, and the unease and concern associated with encounters between individuals or groups and data about them. We then consider what the data shadow ‘is’ in relation to ageing data subjects, and the nature of the representation of the individual’s cognitive state and dementia risk that is produced by digital tools. Second, we consider what the data shadow ‘does’, through researchers and practitioners’ discussions of digital phenotyping practices in the dementia field as alternately empowering, enabling and threatening. Keywords Data shadow, digital phenotype, data double, digital health, ageing, Alzheimer’s disease This article is a part of special theme on Digital Phenotyping. To see a full list of all articles in this special theme, please click here: https://journals.sagepub.com/page/bds/collections/digitalphenotyping disease research. Such data, derived from genomics, elec- Introduction tronic health records and increasingly from digital In this paper, we draw on work with researchers and devel- sources, is intended to enable the detailed characterisation opers engaged in digital health to explore how digital phe- of individual patients that lies at the heart of ‘precision’ notyping represents the ageing body, and, further, how medicine. While Engelman (2020) argues that there is these representations act as entities in their own right. We both an epistemic and sociological naivety to the belief build on sociological and anthropological work on the that clear disease will emerge from the mining of data, it data self and focus on a medical domain in which the remains a powerful driving impetus, particularly in fields value and nature of knowledge and relations with data like Alzheimer’s disease research in which the connections have been played out with particular persistence, that of Alzheimer’s disease research. As a leading cause of ill-health in high-income countries, 1 Engagement and Society, Wellcome Connecting Science, Hinxton, UK dementia has been the focus of considerable policy and Cambridge Public Health, University of Cambridge, Cambridge, UK research attention since the mid-1980s. Much of this atten- Department of Sociology, Goldsmiths, University of London, London, UK tion has concentrated on Alzheimer’s disease, the most Corresponding author: common cause of dementia. As is the case across biomedi- Richard Milne, Engagement and Society, Wellcome Connecting Science, cine, the collection, sharing and analysis of large volumes Hinxton CB10 1SA, UK. of data is seen as central to the future of Alzheimer’s Email: rm23@sanger.ac.uk 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 between the normal and the pathological, the biological and (2017), Insel, the former director of the US National the clinical, remain profoundly contested (cf Lock, 2013). Institute of Mental Health, argues for acknowledgement A shift in Alzheimer’s disease research and in diagnostic of the limitations of the growing dominance of a focus on criteria in the last two decades has placed biomarkers at the the biological correlates of mental health. He describes a centre of what constituted ‘disease’ and created a drive to risk that recent psychiatric research and diagnostic practice better understand the ‘pathway’ of biomarker change over have been dominated by genomics, pharmacology and neu- time. It has driven a focus on the thresholds at which an roscience at the expense of behavioural assessments and individual can be detected as moving between stages on detailed clinical interactions. Instead, Insel and other expo- this pathway from ‘normal’ to ‘biomarker positive’,or nents of a digital phenotyping approach suggest that it from healthy to preclinical, prodromal and on to sympto- enables researchers and clinicians to use portable tools matic dementia. Research increasingly focuses on under- and informatic techniques to capture the ‘extended mind’ standing the pathology, natural history and epidemiology (Raballo, 2018). of the disease to prevent or delay the process of cognitive Digital phenotyping approaches are based around either decline. The hope is that this move to prevention will active user engagements with tests or assessments, or on counter the long-standing status of the field as a ‘graveyard ‘passively’ collected data generated in everyday life and of drug development’ (Hawkes, 2016). It frequently engagements with the world. The central promise is ‘the involves, however, the identification of people deemed to possibility of continuous measurements. Use of apps, be at ‘high-risk’, those who might be most eligible for clin- phone calls, typing speed, and voice features can be moni- ical trials (Milne, 2018). tored unobtrusively every second over a lifetime, with real- Since the identification in 1993 of the first gene asso- time algorithms checking for alarming transformations’ ciated with susceptibility to Alzheimer’s disease, the (Ebner-Priemer and Santangelo, 2020: 298). As with the ApoE e4 allele, an enduring bioethical debate has taken wider field of psychiatry, neurodegenerative disorders and place about whether and how information about risk dementias including Parkinson’s (Trister et al., 2016) and states should be made available to individuals, and what Alzheimer’s disease (Kourtis et al., 2019) are important the consequences would be of doing so (Post, 1996). As sites of hope and promissory investment. Alzheimer’s disease research and in many cases clinical The development of digital phenotyping tools draws into care have become increasingly dominated by assessments focus the intersection between data practices of disease of the biological state of the brain (Lock, 2013), these detection and those associated with the extraction of the debates have been taken forward into discussions around ‘behavioural surplus’ of surveillance capitalism (Zuboff, the use of biomarkers such as beta-amyloid or tau to 2019). These come together explicitly in work from com- assess an individual’s risk of future dementia (Karlawish, panies including Microsoft, Google and Intel. Microsoft, 2011; Schicktanz et al., 2014). This has spurred a for example, has explored the possibility of using search growing body of work on the experience of living with engine data to develop ‘web search digital phenotypes’ information about Alzheimer’s disease risk (Largent et al., that can be used to detect neurodegenerative disorders 2020; Milne et al., 2018). It has also, however, prompted (White et al., 2018)). Patents for the Google Home device critical commentary on the constraints of biology-centred meanwhile describe a possible use of the device to representations of current and future health that marginalise analyse ‘the unique signatures of the occupants’ for patterns the collection or use of information about environmental indicative of Alzheimer’s (Fadell et al., 2018: 252). These exposures or lived experiences (Brayne and Kelly, 2019; ‘unique signatures’ draw discussion of digital phenotyping Leibing and Schicktanz, 2020; Lock, 2013). into dialogue with the wider body of work on the configur- Questions about the adequacy of representation and the ation and exploitation of the ‘digital subject’ (Goriunova, experience of a life lived in relation to information about 2019). In the following section, we draw on this literature future health are made all the more pressing and relevant to consider the forms of representation and the relation by the emergence of ‘digital phenotyping’ in medical between individuals and data described in the development research and accompanying promissory narratives (Birk of digital tools for the assessment of cognition. et al., this issue). Digital phenotyping approaches are based around the fundamental promise that an individual’s Doubles and shadows – life with data experience of health, ‘is expressed in the digital traces that a person leaves behind’ (Birk and Samuel, 2020: 1873). In The idea that distinctive features of our identity might be the context of psychiatry, leading proponents of the reassembled using the digital tracks and traces of our digital phenotyping approach position it as a response to lives – our ‘unique signatures’– has generated a rich and the concern of many psychiatrists that the field was continuing conversation on the formation of digital subjects moving from being ‘brainless’ to ‘mindless’ (Insel, 2017: and its consequences. Schüll highlights the ‘creative voca- 1215). In his JAMA article setting out his programme for bulary’ (2018: 35) introduced by scholars aiming to digital phenotyping as ‘a new science of behaviour’ capture the intensive datafication of life in western societies. Milne et al. 3 These include the use of terms such as ‘data doubles’ or through everyday interactions and encounters with data col- data selves to describe emergent and temporary virtual/ lection and storage (Zook et al., 2004). Such ‘thick data informational profiles that are aggregated from different shadows’ are not ‘just a way of describing data itself, and data sources, sliced and circulated and through which our increased prowess in measuring, mapping, analyzing, selves become both objects and subjects of power in and visualizing, but a meme that speaks to and produces digital worlds (Douglas-Jones, 2021; Green and new ways of establishing truth’ (Graham and Shelton, Svendsen, Haggerty and Ericson, 2000; Lupton, 2019). 2013: 257). However, as Leonelli and colleagues point Thus, confronted by an advertising company’s elaboration out, the concept also draws attention to ‘an ambiguity and of the use of data to target a specific consumer – herself – a strategic relationality to shadowing processes that paral- Goriunova asks lels the relational nature of data and the multiplicity of motives, goals, and conditions through which data may be “What exactly is this digital entity that she identified as me? construed as (in)significant, partial or complete, (un)intelli- What relation does it have to me? How do I relate to it? gible, or (in)accessible’ (Leonelli et al., 2017: 194). This How is it able to stand in for me and construct a me that focus draws critical attention to how representational appa- attracts advertisements and thus alters me, while still ratuses are assembled, how patterns of illumination and being reliant on my activity? How is it produced outside obscurity are authorised and legitimised, how and why of my awareness, mobilized, and recruited?” (2019: 126) some aspects of the data self are drawn into the light, seen as ‘available, portable, and/or meaningful’ (2017: The mapping of persons to digital traces captured over time 194) and other data made ‘missing, unavailable, or invisi- produces a particular kind of subject formation. However, ble’ (2017: 191), and how this changes over time (cf as Goriunova here suggests, this digital entity acts and McGoey, 2017). alters the subject to whom it relates – as Lupton puts it First, we suggest that the idea of the data shadow extends ‘people and their data make each other’ (Lupton, 2018: discussion of the representational nature of the data object, 5). In this, data doubles echo the discussion of Frank, following Goriunova (2019) in complicating representa- who draws attention to the multiple images and codings tions of data subjects as 1:1 depictions of persons or through which the body is doubled and redoubled in con- subject positions and emphasising the strategic relationality temporary medicine, such that the ‘image on the screen highlighted by Leonelli et al. (2017). Although Gorinuova becomes the “true” patient’ and ‘initial certainty of the treats ‘data doubles’ and ‘shadows’ as interchangeable real (body) becomes lost in hyperreal images that are notions of the indexical digital subject, we argue that the better than the real body’ (in Nettleton, 2004: 669). relational nature of shadows offers possibilities that have Lupton, engaging with the concept of the data double in not yet been explored. Thus, while the notion of ‘doubling’ the case of health data, describes how such abstracted ‘data- concentrates attention on duplication and replication, the doubles’ both categorise and identify ‘at-risk’ individuals, nature of the shadow usefully and intuitively captures the and become materially forceful, ‘feeding back information dynamic and distorted (cf Green and Svendsen, this issue) to the user and encouraging the user’s body to act in nature of data representations. It draws attention to the cir- certain ways’ (Lupton, 2012: 237). As she continues, the cumstances of the production of the data self: the relations data double is part of ‘a continual loop of the production between the properties of an object or figure, how that of health-related data and response to these data’ (Lupton, figure relates to a light source, the ground against which 2012: 237). In her work with members of the quantified it stands and on which the shadow is cast, and the place self-movement, ‘pioneers in the art of living with and of the observer. It is this configuration, and the epistemic through data’ (Schüll, 2018: 35) who represent the contem- strategies that define it that enables a shadow to be porary apogee of self-tracking and monitoring practices, produced. Schüll captures how data becomes part of a ‘loop of reflex- Second, the shadow has a cultural resonance that gives ive recomposition’ (2018: 35) and digital tools provide self- the notion of the data shadow value beyond representational trackers the freedom to engage in projects of self- concerns. As an artistic, literary and cinematic trope, the transformation. This ‘loop’ of (self-) representation and shadow has value for its ability to capture the unease and transformation is both recognised and valued by those concern associated with encounters between individuals involved in the development and capitalisation of data- or groups and data about them (Stoichita, 1997). In these driven tools. It is central to the premise of fitness and ‘well- contexts, the shadow is not simply a representation of an ness’ devices that aim to encourage reflexive interactions absent subject so much as an entity in its own right, a with the data self, whether for health or wealth ‘reality to your consciousness‘ (Bergson, 1913: 54). As (Douglas-Jones, 2021; Lupton, 2019). such, it can form its own object of study, a menacing In this paper, we explore the nature of the ‘data self’ other rather than a represented self. Consequently, ‘we using the metaphor of the ‘data shadow’. The data produce our own data shadow, but do not have full shadow refers to the counterpart to the individual produced control over what it contains or how it is used to represent 4 Big Data & Society us’ (Zook et al., 2004: 169). Such shadows have value and reported here consisted of a domain mapping of academic can be traded in networks of exchange, creating situations and commercial activity and research in the field, incorpor- in which a subject may feel ‘over-shadowed’ by circulating ating published academic and patent literature and presenta- information about them – for example, as they attempt in tions of tools in websites, company webinars and press the contexts of employment, finance or insurance. These releases and conference presentations, covering 30 tools entities are not simply archives capable of revealing under commercial development. This mapping was fol- ‘hidden patterns of action at play in our day-to-day lives’ lowed by 26 semi-structured interviews in 2019 and (Schüll, 2018: 5). Our ‘othered’ data selves here pose 2020. Participant was drawn from companies involved in potential threats to the life chances and choices of indivi- the development of digital tools (n= 7), academic neu- duals, including when they are considered to be predictive roscience and data science researchers (n= 8), clinicians of future problems (Eubanks, 2018; Lyon, 2014; Ruppert, in neurology or old age psychiatry (n= 3) and policy and 2012). research officers working in non-profit organisations Individual lives are thus not simply captured in data, but involved in developing or funding digital phenotyping lived in relation to these data and the futures they ‘fore- tools for dementia (n= 9). Participants were based in the shadow’. In this sense, the concept of the data shadow UK, Europe and North America. Interviews asked about allows us to interrogate how access to services, from health- respondents’ hopes and expectations around the tools they care to insurance, is shaped by the future selves cast for- were involved in developing, their concerns and their wards by data. The absence of such foreshadowing, awareness of ethical considerations related to the use of though, can also be problematic. As Beer puts it ‘When digital tools for the early detection of cognitive decline. we are informational persons – that is to say, when we The second study involved three focus group discussions have become our data – the deletion of our data amounts held in the UK in 2020. Ten UK-based researchers partici- to the erasure of our identities’ (Beer, 2021: 390). pated, drawn from academic neuroscience and data science, Elaborating on this, the lack of a data shadow may limit clinical research, industry and clinical old-age psychiatry access to future-oriented domains such as insurance. The and neurology. All were involved in the development of a ‘active’ life of our data shadows, and the material conse- large project aimed at the development of markers of quences of their absence thus emphasise both the conse- Alzheimer’s disease progression for use in clinical trials. quences of living in relation to and without our data self. While not all were directly involved in the development In juxtaposing such representational and material, of digital tools, the project as a whole aimed to incorporate ‘virtual’ and ‘vital’ elements of the data shadow, we do these into clinical trial practice. These focus groups again not intend to sharpen the distinction between the digital concentrated on expectations and concerns around the and lively, biological processes – between knowledge of future role of digital tools in the assessment of cognitive life and life itself. Such a distinction is far from clear, decline. perhaps least so amongst those people producing the tools In the following sections, we explore how this data can and knowledge to phenotype complex neurological condi- provide insight into how the value of digital phenotyping tions such as dementia. Rather, we aim to explore the differ- is being imagined, and the challenges associated with ent relationships that are forged between living subjects and this. We draw on the perspectives of interviewees who the informational traces of ageing bodies and minds. In this are, in the main, involved in articulating and giving way, we are more interested in the messy relationships impetus to the possible futures of digital health. As such, between knowledge of life and life itself; and in digital phe- we are aware that this focus on hopeful narratives, often notyping as only one iteration of the data shadow, with its oriented towards the creation of promissory value, creates own particular qualities, distortions and effects. our own ‘data shadows’ and absences (cf Leonelli et al., In the following sections, we move on to extend and 2017), not least those associated with alternative ways of illustrate our discussion through our empirical data. We seeing, doing and living Alzheimer’s disease (Leibing, consider what the data shadow ‘is’ in relation to ageing 2014). data subjects, and the nature of the representation of the individual’s cognitive state and dementia risk that is pro- Representing cognition: Casting shadows duced by digital tools. Second, we consider what the data shadow ‘does’, through researchers and practitioners’ dis- In this section, we explore how our interviewees, many of cussions of digital phenotyping practices in the dementia whom are deeply invested in the development of digital field as alternately empowering and threatening. phenotyping and the scientific and clinical exploitation of Our discussion draws on two research studies. The first, the data shadow, conceptualise the promise of their an empirical study of ethical challenges associated with approach and the value, nature and status of the representa- digital detection in dementia, involves work with both tions they are involved in producing. We describe how, in domain experts, reported here, and older adults around interviews, publications and corporate material, the their use of, and experience with, digital health. The work promise of digital phenotyping tools is established Milne et al. 5 through three closely related discourses of epistemic value. emphasised, it may be that a critical feature of digital phe- These discourses emphasise uniqueness and the detail of notyping is not simply the accumulation of information digital representations of individuals; extension or the from passively collected data. Instead, it is in the extraction ability to track an individual over time; and ecological of value through inference, in a manner that echoes – and validity, or the possibility of ‘real world’ measurement. indeed explicit references the wider ‘digital exhaust’ These discourses, we argue, allow us to understand the stra- (Hirschtritt and Insel, 2018). As one interviewee put it: tegic relationality of digital phenotyping’s data shadow. Interviewees and other proponents of digital approaches “We know language changes in later stages without being to capturing cognitive and behavioural states claim that detected. So, that’s one of the reasons why we’re looking these offer the possibility of capturing data that is distinct- that earlier. … But then … the literal data and words you ively representative of an individual. As a consequence, say is not what we care about. It’s what insight that gives they suggest, such tools are able to depict an individual’s us into the change that’s going on in your brain” (Ethics brain health in a manner that is both potentially both officer, non-profit) more profound and more able to matter to people. Thus, for example, interviewees described how the use of Another senior company interviewee described how digital devices is unique to each individual and thus such functional measures – and their algorithmic interpreta- uniquely identifying. As one interviewee put it ‘the way tion – can provide truly ‘personal’ assessments: that people interact with their phone, is like a fingerprint for them’ (staff member, non-profit organisation). This fin- “We have a beta version that we are testing internally … gerprint generated in interaction with devices, in turn, has and the data is fascinating because it shows that it’s abso- scientific and potential clinical value for developers in lutely personal, for myself I can see sleep has a huge excess of its uniquely identifying capacity; for example, impact on my performance. Some other colleagues that through its use to generate a ‘cognitive fingerprint’ do this, they can see that the days that they have physical (researcher, non-profit). As another interviewee described, exercise … you can see a huge improvement in how their such fingerprints may be based on the aggregation of multi- brain capacity improves. We have done this in a small ple digital streams that capture the multiple aspects of a group of people over a six month period, we’re basically ‘neurological condition … like slight behavioural training the AI to understand what factors we need to con- changes, executive function changes, the syntax of your sider.” (CEO, UK-based digital health company) language’ (Researcher, non-profit). As in wider discussions of digital phenotyping (Birk and This promise of better representations of an individual’s Samuel, 2020), the ability of digital representations to act as world, produced by repeated measurement over time, is measures of an individual’s brain health is not universally reflected in the Delphic corporate slogan of the leading accepted amongst our respondents, as we discuss below. digital cognitive testing company Altoida to ‘Know However, for interviewees involved in elaborating the Thyself’, and in the motivations of researchers working promise of digital approaches, only part of their value lies across the field. As one interviewee described: in their individualisation and a persistent ‘fingerprint’.In fact, the value of this representation lies in its instability. “It was our understanding that some of these people were Unlike its physical counterpart, the digital ‘fingerprint’ is saying that they had subjective cognitive complaints, mutable and changes over time. What the fingerprint ‘is’ saying that in their opinion their cognitive ability was in this case is defined by what it can technically ‘do’, declining, but on their standard tasks they were doing just record and follow change, track the process of an individual fine, so a clinician couldn’t say that they had any kind of becoming different from themselves and extrapolate from deficits or impairments … so we tried to understand why this to infer future health. For our respondents, it is this these people were actually talking about their cognition focus on individual fluctuation and change that distin- declining even though the tests that we were using guishes digital phenotyping: weren’t picking up on that, so we just tried to make tests as close to real life” (Researcher, non-profit) “Digital tools enable you to detect change, and will put more emphasis on the role of this fluctuation and this For this interviewee, digital tests have the potential to decline before you actually need to reach a threshold to provide more accurate representations of cognitive state get any intervention.” (Company researcher) that may more closely align with lived experience and ‘real life’. This extract, and its emphasis on individuals’ The move away from thresholds and towards subliminal experiences of decline, highlights the temporal ambitions fluctuations in a continuously reproduced image of the of those developing digital tools, to track change over self has consequences for the forms and content of the time. It also, however, introduces a third key feature of data that make up this picture. As other interviewees digital phenotyping that draws attention to situated 6 Big Data & Society change. The unique fingerprints described by our respon- ‘require subjects to perform a task outside of the context dents and their change over time are conceptualised as of everyday behaviour’ (Dagum, 2018: 1), and emphasise reflecting an individuals’‘real life’ engagement with the the value of collecting data using smartphones “in a world. natural environment” (Dagum, 2018: 2). Dagum’s discus- Being able to track the process of becoming different is sion here closely Insel’s perspective on the development part of the possibility of a new form of knowledge. As one of the ‘new science of behaviour’– and indeed, the two company CEO described to us how: co-founded the mental health and cognitive testing company Mindstrong (Reardon, 2017). ““I think what’s important here is that we have no good bio- In our study, researchers using virtual and augmented markers for the brain … nothing that really is measuring reality approaches to assessment particularly emphasised functional capacity, nothing that’s, we’re measuring the possibility of generating representations of cognition illness but we’re not really measuring the impact that it’s created by ‘making something real life’ (company having on someone’s ability to function, to use their brain researcher). They suggest that digital approaches offer the to solve a task and live a healthy life, so that’s important” possibility of a wider shift towards new ways of seeing: (CEO, US-based digital health company) “I kind of get the sense that sometimes in medicine it’s just, For this individual, articulating their vision for the technol- you know, checking the box. They’ve a cognitive assess- ogy and its potential, digital phenotyping allows the ‘real ment. And they don’t really care what the cognitive assess- measurement’ of the impact of illness through an indivi- ment is, they’re more interested in the biology of the dual’s lifetime. For them, this remedies the limits and blind- disease, because that’s what the treatment is going to spots of existing measures, moving the field towards have an effect. That’s why it’s almost as if seeing the capturing the lived experience of ‘using the brain’ to ‘live pathology and the disease go down is what you want to a health life’. see. And if that has a cognitive effect, then fine. Good. This commitment to continuous measurement in ‘real It’s almost always been secondary to the cognitive change life’ forms the core of what the developers and researchers has always been secondary to the biomarker change. And we interviewed refer to as ‘ecological validity’, a central I think that’s shifting now.” (Researcher, non-profit) epistemological premise of the digital phenotyping enter- prise. For actors in the field, this claim to ecological validity The transition this researcher describes and urges, away posits that digital tools’ ability to provide access to the ‘real from a biomarker-driven approach to ‘seeing the pathology world’ makes their data representations qualitatively differ- go down’ towards a focus on cognitive change is reflected ent from existing ways of seeing and ‘doing’ Alzheimer’s in the approaches of a number of companies and projects disease. As one interviewee put it, ‘it tells us about what operating in the field – the digital health companies it is to be a human being with this brain interacting with Winterlight and MyndYou, for example, echo the chief the world.’ (clinical researcher). Another described how: scientificofficer of Applied Cognition (Dagum, 2018) in describing how their analyses of voice data give ‘eco- “Technology can give you the real-world scenario, continu- logical’ measurements of cognitive state. ous data, understanding a little bit more about how, really, As the extracts above suggest, the development of digital people function and behave and how their condition is.” tools is taking place amid an existing marketplace of (Company researcher) approaches to representing the ageing brain. In positioning their field respondents suggest distinctions or even As this researcher makes clear, the promise of technology sequences in these approaches. The future applications of for them is both temporal, in the form of ‘real-time’ con- digital phenotyping they describe explore relations of com- tinuous data, and ethological, in the form of real behaviour. monality, complementarity and conflict between digital and This suggestion, repeated across interviews, that digital biological ways of knowing and representing brain health. phenotyping captures how people ‘really’ function recapi- For some interviewees, the shadows cast by digital and bio- tulates questions in the wider psychology (and sociology) logical data were complementary, and associated with dif- literature of the relationship between experimental evidence ferent aspects of disease – as one put it: produced in laboratory settings and behaviours ‘in the wild’ (Cicourel, 1982, 1996; Orne, 1962). Indeed, the transfer- “I mean digital data can measure definitely your activities of ability of evidence beyond the laboratory or the clinic has daily living. So I think it’s good for the people who are not I been a persistent question for the neuropsychological mean, who are a little bit, they’re not very serious who are research on which many digital measures of cognition slowly going towards maybe the cognitive impairment build (Chaytor and Schmitter-Edgecombe, 2003; phase. And then, yeah, then the dementia phase. So for Kvavilashvili and Ellis, 2004). Those working on digital early disease prediction I think this is the case that these tools thus suggest the limits of measures and tasks that patients are given the digital data. But if you think that a Milne et al. 7 person [has dementia] you have to somehow take brain The data encounter: Living with shadows images of the patient. That cannot be done with a digital As introduced above, digital phenotyping produces data readout, because to really see if this patient has dementia representations that are both contemporaneous and predic- or not, then you have to look into the brain of the person tive, existing alongside the user while suggesting possible and you have to see that something has happened.” futures. As a result, encounters with data shadows involve (Clinical researcher, emphasis added) both current and future health and illness. This encounter with foreshadowed illness, the possibilities it offers and the harms it may cause, have been a repeated site of For this interviewee, the data shadow produced by digital ethical contestation in the Alzheimer’s disease field phenotyping both ‘definitely’ measures daily life, and (Karlawish, 2011; Post, 1996). In this section, we consider casts forward the future health of individuals ‘slowly how our interviewees described the nature of this encounter. going towards’ cognitive impairment and dementia. The In doing so, we consider how our respondents conceptualise data shadow’s ability to represent, however, is partial – it encounters with data selves as entities that empower and cannot sufficiently illuminate the current state of the brain enable, or that threaten. and, in order to align the clinical data shadow with the disease model articulated in dominant diagnostic definitions of disease, has to be accompanied by measurement of the The encounter that empowers and enables changes in biological markers of brain health. The representations discussed above, produced by and For the interviewee above, the representation of the through encounters between individuals and digital ageing body produced by digital phenotyping is comple- devices are primarily oriented towards the early detection mentary to the biological. For others, the biological model of cognitive change, and the prediction of an individual’s of Alzheimer’s disease meant that the shadows cast by risk of future dementia. For our respondents, the relation- digital phenotyping were emphatically the ‘wrong’ type ship between an individual data subject and these represen- of picture. As one senior clinical researcher put it: tations was often couched in terms that emphasise the possibilities afforded by this future orientation. Across digital psychiatry, developers emphasise the ‘biomedical “I think digital biomarkers are a new way of measuring the virtue’ of empowering or enabling ‘self-care’ (Pickersgill, wrong thing … Alzheimer’s disease is a brain disease, it is 2019). In our study researchers described how datafied not a cognitive disorder. If you want to measure the brain representations of cognitive health produced by digital disease, you measure it directly with biomarkers and tools could give the subject ‘a window on their self’: imaging, you do not measure it on cognition.” (Clinical researcher, emphasis added) “I think giving some data as a window into their own health, physical and mental, is definitely something we would like to see down the line, whether it’s someone that’s older and In criticising digital approaches, this researcher negates the healthy or older and has some cognitive or neuropsycholo- claim to ecological validity – reducing it to an outmoded gical problems.” (Senior academic researcher) focus on cognition, and suggesting that the data captured by digital phenotyping is distorted and unhelpful. They con- As others described, the ability to provide users with the trast this misguided approach to measurement with that ability to ‘see through’ this window to look at their data focus on ‘brain disease’, suggesting that the data produced self is the key step in giving them the power to take action: by assessments such as brain imaging allow direct access to the causes of disease. This speaks to the epistemic and poli- “we want to be able to collect that data and give a better tical commitments that shape the ground on which data assessment of where the patient is, and also by making shadows are cast: the picture of ‘brain disease,’ as that, visualising that data for the user, not necessarily a opposed to ‘cognitive disorder’ underpins the project of patient … we can help them to do things that can help identifying early biological markers, in order to develop them become healthier” (CEO, UK digital health drugs that make Alzheimer’s potentially treatable, even company 1) before cognitive signs of dementia emerge. Such approaches are often tied to the development of therapies, This positive potential of the data shadow to empower reca- in which the object to be illuminated through a clinical pitulates both the promises of digital psychiatry and, assessment may be determined by the target of a drug. indeed, of digital health more broadly (Lupton, 2012). The context in which data shadows are produced is thus Thus, another interviewee explicitly situated the goal of full of coexisting and sometimes competing for projects their company to use routine digital assessments to enable and perspectives. older adults to maintain their cognitive health firmly 8 Big Data & Society within existing data relationships. In this vision, digital cog- Here, we see the possibility that the enabling possibili- nitive assessments simply fit into the new ways of seeing ties of the digital phenotyping data shadow derive from and being associated with digital health: their integration into an architecture of insurance-supported healthcare. This vision, elaborated by researchers in both academic and corporate spheres, is evidenced in corporate “The penetration of Fitbit … for example is really quite development models in the field, in which the challenge advanced into the 60 plus market. So that really is the rise (and uncertain financial return) of ‘health system’ tools vali- of acceptance of personalised data, and I think that increas- dated to regulatory standards for medical devices is accom- ingly people do just kind of by second nature understand panied by ‘lifestyle’ products, which may incorporate the that this is something that is part of your day to day life.” same technical basis, but are oriented towards the more (CEO, UK digital health company) accessible market of health insurance. Narratives of self-care embedded in insurance-driven For this interviewee the relationship between the data healthcare situations situate digital phenotyping firmly subject and a personalised data self, formerly the domain within a particular structure of healthcare provision. In of the self-tracker, has become commonplace. Our respon- that sense, they connect digital health to political discourses dents’ elaboration of the relationship in turn attempts to of preventive health that shift the burden of action away reinforce this idea of digital phenotyping as a quotidian, from the state towards the individual (cf Lupton, 2012) producing data shadows as unremarkable objects that we and seek to embed digital tools for cognitive assessment live with and alongside. In turn, these data shadows, they within technology-enabled projects of self-transformation. suggest, do not simply empower, but enable – they can These visions of empowerment and enablement also, both contribute to improving health and facilitate wider however, emphasise the threat of being absent from data. aspects of life such as employment or insurance. In the In addition to being provided with information they can Alzheimer’s disease case, this is closely tied to the shift use themselves, in the narratives presented by developers, in emphasis towards the early detection of, and intervention the ‘measured’ are enabled to access healthcare and insur- in, disease described earlier. As a company researcher ance services in ways that are not available to the ‘unmea- described: sured’. In this vision of future application, digital phenotyping presents possibilities – as the data shadow becomes supportive of health and facilitative of access to “I think a lot of the traditional tools … use thresholds of the a particular vision of healthcare – but also threatens and output, the score. If you’re below this then you are clinically restricts, in both its presence and absence. classified as this whereas, for a lot of people, they might not be below that threshold yet but they will have experienced a lot of change. They would have fallen from where they The encounter that threatens were previously but according to the standard, traditional, The threats to self and to individual autonomy posed by the tools they don’t deserve any clinical attention because data shadow were raised by a number of our interviewees they haven’t passed that threshold yet.” (Company and, in many ways, return our analysis towards those con- researcher) cerns raised in the bioethics literature around the return of results in the context of Alzheimer’s disease, and the rela- Here, digital tools offer new possibilities for people who tionship between discussions of risk and those of affective experience concerns about changes in their thinking or encounters between individuals and data. Thus, the memory, enabling them to become ‘deserving’ of clinical researcher with whose interview we closed the preceding attention. Such considerations firmly integrate digital phe- section continued by elaborating the need for control over notyping into existing practices of clinical research, and one’s data in light of the potential of this data entity to the spaces of commercial possibility associated with the cause harm if allowed to move. As they put it, early detection of disease (cf Dumit, 2012). However, enab- ling visions of digital phenotyping also connect with, and “If you let people test your cognition for you, then it prob- raises comparable questions to, the role of the data ably waived your privacy goodbye very quickly. If you can shadow in ‘representing’ or standing in for somebody (cf do that independently then you can claim your data and pre- Goriunova, 2019; Lupton, 2019). One key context for this serve your privacy.” (Academic researcher) is insurance, as another interviewee described: In this quote, the researcher emphasises that the relation- “If I put the clock forward 20 years, I would envisage this. If ship between self and data should be just that – avoiding you’ve got health care insurance, whether it be state or intervention or mediation from third parties, with the indi- private, part of your annual assessment will be your perfor- vidual data subject retaining control over both her data mance on cognition tests” (Academic researcher) and the futures it may foreshadow. However, as Milne et al. 9 interviewees repeatedly raised, this intimate relationship is commercial development (cf Dumit, 2012; Milne, 2018), also one that comes with an existential threat to the data for example, in the case of pharmaceutical innovation or subject herself. One respondent described this, drawing insurance markets. However, for some, the threat posed on their experience providing feedback as part of a research by the encounter with the Alzheimer’s data shadow in study: this context has emerged as a barrier to their research and development: “[W]e had a number of people write in and say ‘Participating in a study is too much for me, I see my “the first day that we pitched the idea in [X] we had one of scores going down week by week, I see that my medicine the major investors in [X] went on a crusade to stop us from is not making any difference, this is a lot’, ‘having access developing our technology, and we’ve had that over the last to my data’– because we gave our participants access to seven years as well, keep coming back at us is if there’sno their data –‘having access to my data is just depressing treatment there’s no point in diagnosis, and what you do is me; having the ability to do some things now and unethical. I think I’ve expressed that enough and I still knowing that I’m not going to have the ability to do them believe as a researcher, as a patient, that’s not the right later, is crushing me, and being reminded of this three approach but it is a stance some people take. (CEO, UK times a day is overwhelming, I have to quit’, and so that digital health company) was some very sobering feedback (Staff member, non-profit organisation) The intensity of concerns around risk, and the ability of these concerns to shape the development of digital tools, is A researcher from a company similarly described how a prominent feature of Alzheimer’s disease. It characterises users of their tool will ‘often … reach out and say, like, scientific, clinical and popular discussions of early detection hey, I want to talk to somebody about this’. The concerns and risk prediction in a way that can become forceful in that these respondents highlight, in which people feel over- technological development, as previously in the case of whelmed and emotionally affected by an intimate, pharmacogenetics and biomarker development (Boenink, one-to-one encounter with data, both encapsulates the 2018; Hedgecoe, 2008). In the case of digital phenotyping, threatening nature of the data shadow and draws discussion researchers described the problem of risk communication as back towards that of the bioethical discourses with which ‘one of the kind of big ethical concerns’ (company we started. As another researcher at a non-profit described, researcher), which, for their company, currently precludes this relationship revolves around a core dilemma the possibility of any direct relationship between the indivi- dual and their data shadow. Further, the menace posed by “In terms of detection, if you tell them that they are at this relationship re-emphasises the need for digital pheno- increased risk. OK, you’ve told me this. What can I do typing to capture and be embedded in everyday life, to then to reduce that risk? And obviously in this space we ‘appropriately support people through this’ (company can’t really do that. We can just say there are some risk researcher). Such considerations emphasise that, for indivi- factors that have been associated, but we can tell you how duals and organisations, data shadows cast long into the to reduce your risk by 20 percent or something like that. future, and living with them is a long-term commitment. Can you cope with that? … So, it’s quite a hard topic to discuss with them.” (Staff member, non-profit organization) Working with shadows As this extract highlights, and as introduced earlier, discus- In the preceding sections, we have introduced two dominant sion around the ‘data encounter’ in Alzheimer’s disease – threads in how the relationship between data subjects and and with risk information collected from genetics, family their data selves is conceptualised by researchers and devel- history, or lifestyle factors – has long revolved around the opers working in the field of digital tools for the early detec- potential for psychological harm in the absence of an effect- tion of Alzheimer’s disease. We have used the metaphor of ive therapy (Brayne and Kelly, 2019; Lock, 2013; Post, the data shadow to explore two elements of digital pheno- 1996). This threat persists even as developers attempt to typing and its products in this context – their representa- establish an ‘enabling’ data relationship – and, as this tional function and their role as enabling or threatening respondent describes, remains a ‘hard topic’, despite materials. increasingly well-established clinical approaches to risk For our interviewees – personally, scientifically or finan- communication (e.g. Harkins et al., 2015). Indeed, some cially invested in the futures of digital health – digital phe- respondents suggested that this potential for harm may be notyping approaches draw from the wider promise of data intensified in the context of cognitive and behavioural the possibility of a different way of understanding and data that are seen as ‘a bit of your identity’ in a way that bio- representing the ageing brain. For these researchers, as for logical markers are not (Digital health researcher). Risk and advocates of digital psychiatry such as Insel, the potential its communication serves as an engine for scientific and of the data shadows of Alzheimer’s disease is that of an 10 Big Data & Society individualised, longitudinal and ‘ecological’ mode of repre- early detection do not necessarily indicate the current sentation. The construction of this way of seeing illustrates state of the brain, instead creating data shadows that are how the representational apparatus of Alzheimer’s disease cast forward in time. In the context of research that is is being assembled and which aspects of the disease are moving away from a focus on symptomatic diagnosis of being exposed to, or concealed from, view. It also, disease towards early detection and the identification of however, shows how this process reflects and reproduces the risk of future dementia, analyses of digital traces are a wider discourses around the possibilities and perils asso- means of knowing both the current and (possible) future ciated with inferences made through our digital traces. self. Thus, our respondents highlight how digital phenotyping These considerations point to the emerging complexities and the elaboration of digital data shadows intersects with of digital phenotyping as developers attempt to integrate the existing practices of data collection and analysis, as well collection of digital ‘biomarkers’ by active or passive as the ontological and epistemological commitments of means into existing clinical domains. Our use of the data contemporary Alzheimer’s disease research. Participants shadow to explore these themes, and their representational draw attention to the potential for complementarity but and material articulations, points to the ways in which also tensions between digital data and those that emphasise digital phenotyping requires continued attention to inter- biology, biomarkers and the therapeutic models that target twined ethical and epistemological considerations. The them. These relations further gesture towards the commer- empowering, enabling and threatening aspects of the cial arrangements and health system architectures asso- shadow highlight the challenges associated with controlling ciated with different ways of seeing, and emphasise the or containing the Alzheimer’s disease data self. In addition, persistent forms of strategic relationality through which they suggest the importance of continuing to revisit what is, data shadows are cast, bodies made visible and visualisable and what is not, made visible in the process of representing or left in obscurity (cf Leonelli et al., 2017). the ageing brain, what is gained or lost with the commit- In addition, however, in our respondents’ discussions of ment to longitudinal, situated representation, and what fea- digital phenotyping the ‘data shadow’ is more than repre- tures of ageing continue to remain unseen. sentational, playing a material role in the emergence and promise of new technologies and tools. The Alzheimer’s Acknowledgements disease data shadow has the potential to be materially force- The authors would like to thank the editors of the special issue and ful both in the lives of data subjects and to empower and to the other panellists at the 2020 4S/EASST session for their com- enable – but also to threaten. For many of our respondents, ments on this paper, and particularly the reviewers for their helpful this is captured in both the opportunities the data shadow and considered feedback. We would also like to thank the partici- pants in our interviews and focus groups for their time and affords for ‘self-care’ and the facilitative role that the data contribution. shadow can play as it stands in for the data subject in their interactions with healthcare and insurance. Declaration of conflicting interests Conversely, our respondents highlight the threat this shadow poses to the individuals living ‘in the shadow’ of The author(s) declared no potential conflicts of interest with future dementia, connecting emerging discussions around respect to the research, authorship, and/or publication of this article. digital phenotyping with longstanding bioethical debates (Karlawish, 2011; Post, 1996). This threat looms large in Funding a field in which the question of ‘would you want to know’ is a recurrent feature of popular and policy discus- The author(s) disclosed receipt of the following financial support sions around the early detection of disease. For our respon- for the research, authorship, and/or publication of this article: dents, this threat is to both those living with such shadows, RM and AC’s work was supported by Welcome Trust grants 213579 and 206194. NB’s work was supported by an ESRC and to those looking to establish a space for diagnostic Postdoctoral Fellowship and grant (MR/N029941/1) from the innovation and negotiate the challenges of ‘innovating National Institute for Health Research (NIHR) and the Medical with care’ in Alzheimer’s disease (Boenink et al., 2016). Research Council (MRC). Further, the potential to enable – within a particular political and economic configuration of preventative medicine – also ORCID iDs suggests the potential challenges associated with the Richard Milne https://orcid.org/0000-0002-8770-2384 absence of data, or living without a data shadow as digital Alessia Costa https://orcid.org/0000-0002-3761-9080 phenotyping practices become embedded in healthcare or Natassia Brenman https://orcid.org/0000-0002-6567-2129 insurance markets. The ability of data shadows to empower, enable and References threaten, and the consequences of their absence, draws par- ticular attention to the way shadows are situated in time. As Beer D (2021) A history of the data present. History of the Human our respondents described, digital phenotyping tools for Sciences 34(3–4): 385–398. Milne et al. 11 Bergson H (1913) Time and Free Will: An Essay on the Immediate Hedgecoe A (2008) From resistance to usefulness: Sociology Data of Consciousness. London: George Allen and Company. and the clinical use of genetic tests. BioSocieties 3(2): 183– Birk R and Samuel G (2020) Can digital data diagnose mental 194. health problems? A sociological exploration of ‘digital pheno- Hirschtritt ME and Insel TR (2018) Digital technologies in psy- typing’. Sociology of Health & Illness 42(8): 1873–1887. chiatry: Present and future. Focus 16(3): 251–258. Boenink M (2018) Gatekeeping and trailblazing: The role of bio- Insel TR (2017) Digital phenotyping. JAMA 318(13): 1215. markers in novel guidelines for diagnosing Alzheimer’s Karlawish J (2011) Addressing the ethical, policy, and social chal- disease. Biosocieties 13(1): 213–231. lenges of preclinical Alzheimer disease. Neurology 77(15): Boenink M, van Lente H and Moors E (2016) Diagnosing 1487–1493. Alzheimer’s disease: How to innovate with care. In: Boenink Kourtis LC, Regele OB, Wright JM, et al. (2019) Digital biomar- M, van Lente H and Moors E (eds) Emerging Technologies kers for Alzheimer’s disease: The mobile/wearable devices for Diagnosing Alzheimer’s Disease. Dordrecht: Springer, opportunity. npj Digital Medicine 2(1): 9. pp.263–275. Kvavilashvili L and Ellis J (2004) Ecological validity and twenty Brayne C and Kelly S (2019) Against the stream: Early diagnosis years of real-life/laboratory controversy in memory research: A of dementia, is it so desirable? BJPsych Bulletin 43(3): 123– critical (and historical) review. History and Philosophy of 125. Psychology 6: 59–80. Chaytor N and Schmitter-Edgecombe M (2003) The ecological Largent EA, Harkins K, van Dyck CH, et al. (2020) Cognitively validity of neuropsychological tests: A review of the literature unimpaired adults’ reactions to disclosure of amyloid PET on everyday cognitive skills. Neuropsychology Review 13(4): scan results. PLOS ONE 15(2): e0229137. 181–197. Leibing A (2014) The earlier the better: Alzheimer’s prevention, Cicourel AV (1982) Interviews, surveys, and the problem of eco- early detection, and the quest for pharmacological interven- logical validity. The American Sociologist 17(1): 11–20. tions. Culture, Medicine and Psychiatry 38(2): 217–236. Cicourel AV (1996) Ecological validity and ‘white room effects’: Leibing A and Schicktanz S (eds) (2020) Preventing Dementia?: The interaction of cognitive and cultural models in the prag- Critical Perspectives on a New Paradigm of Preparing for matic analysis of elicited narratives from children. Old Age. New York: Berghahn Books. Pragmatics & Cognition 4(2): 221–264. Leonelli S, Rappert B and Davies G (2017) Data shadows: Dagum P (2018) Digital biomarkers of cognitive function. npj Knowledge, openness, and absence. Science, Technology, & Digital Med 1(1): 1–3. Human Values 42(2): 191–202. Douglas-Jones R (2021) Bodies of data: Doubles, composites, and Lock M (2013) The Alzheimer Conundrum. Princeton: Princeton aggregates. Journal of the Royal Anthropological Institute University Press. 27(S1): 159–170. Lupton D (2012) M-health and health promotion: The digital Dumit J (2012) Drugs for Life: How Pharmaceutical Companies cyborg and surveillance society. Social Theory & Health Define Our Health. Durham: Duke University Press. 10(3): 229–244. Ebner-Priemer U and Santangelo P (2020) Digital phenotyping: Lupton D (2018) How do data come to matter? Living and becom- Hype or hope? The Lancet Psychiatry 7(4): 297–299. ing with personal data. Big Data & Society 5(2): Engelmann L (2020) Into the deep – AI and total pathology. 2053951718786314. Science as Culture 29(4): 625–629. Lupton D (2019) Data Selves: More-than-Human Perspectives. Eubanks V (2018) Automating Inequality: How High-Tech Tools Cambridge: Polity. Profile, Police, and Punish the Poor. New York: St Martin’s Lyon D (2014) Surveillance, Snowden, and big data: Capacities, Press. consequences, critique. Big Data & Society 1(2): Fadell AM, Matsuoka Y and Sloo D, Veron M, and Google LLC. 2053951714541861. (2018) Monitoring and reporting household activities in the McGoey L (2017) The elusive Rentier Rich: Piketty’s data battles smart home according to a household policy. US9872088B2. and the power of absent evidence. Science, Technology, & Available at: https://patents.google.com/patent/ Human Values 42(2): 257–279. US9872088B2/, accessed 10 June 2021. Milne R (2018) From people with dementia to people with data: Goriunova O (2019) The digital subject: People as data as persons. Participation and value in Alzheimer’s disease research. Theory, Culture & Society 36(6): 125–145. BioSocieties 13(3): 623–639. Graham M and Shelton T (2013) Geography and the future of big Milne R, Diaz A, Badger S, et al. (2018) At, with and beyond risk: data, big data and the future of geography. Dialogues in Expectations of living with the possibility of future dementia. Human Geography 3(3): 255–261. Sociology of Health & Illness 40(6): 969–987. Haggerty KD and Ericson RV (2000) The surveillant assemblage. Nettleton S (2004) The emergence of E-scaped medicine? The British Journal of Sociology 51(4): 605–622. Sociology 38(4): 661–679. SAGE Publications Ltd. Harkins K, Sankar P, Sperling R, et al. (2015) Development of a Orne MT (1962) On the social psychology of the psychological process to disclose amyloid imaging results to cognitively experiment: With particular reference to demand characteris- normal older adult research participants. Alzheimer’s tics and their implications. American Psychologist 17(11): 776. Research & Therapy 7(1): 1–9. DOI: 10.1186/ Pickersgill M (2019) Digitising psychiatry? Sociotechnical expec- s13195-015-0112-7 tations, performative nominalism and biomedical virtue in Hawkes N (2016) Sixty seconds on … solanezumab. BMJ 355: (digital) psychiatric praxis. Sociology of Health & Illness i6389–i6389. 41(S1): 16–30. 12 Big Data & Society Post SG (1996) On not jumping the gun: Ethical aspects of APOE and Human Augmentics, Artificial Intelligence, Sentience. gene testing for Alzheimer’s disease. Annals of the New York Abingdon: Routledge, pp.25–38. Academy of Sciences 802: 111–119. Stoichita VI (1997) Short History of the Shadow. London: Raballo A (2018) Digital phenotyping: An overarching framework Reaktion Books. to capture our extended mental states. The Lancet Psychiatry Trister AD, Dorsey ER and Friend SH (2016) Smartphones as new 5(3): 194–195. tools in the management and understanding of Parkinson’s Reardon S (2017) Former US mental-health chief leaves google disease. npj Parkinson’s Disease 2(1): 1–2. for start-up. Nature. DOI: 10.1038/nature.2017.21976. White RW, Doraiswamy PM and Horvitz E (2018) Detecting neu- Ruppert E (2012) The governmental topologies of database rodegenerative disorders from web search signals. npj Digital devices. Theory, Culture & Society 29(4–5): 116–136. Medicine 1(1): 1–4. Schicktanz S, Schweda M, Ballenger JF, et al. (2014) Before it is too Zook M, Dodge M, Aoyama Y, et al. (2004) New digital geog- late: Professional responsibilities in late-onset Alzheimer’s raphies: Information, communication, and place. In: Brunn S, research and pre-symptomatic prediction. Frontiers in Human Cutter S and Harrington JJr. (eds) Geography and Technology. Neuroscience 8: 21. London: Kluwer Academic Publishers, pp.155–176. Schüll ND (2018) Self in the loop: Bits, patterns, and pathways in Zuboff PS (2019) The Age of Surveillance Capitalism: The Fight for a the quantified self. In: Papacharissi Z (ed) A Networked Self Human Future at the New Frontier of Power. London: Profile Books.

Journal

Big Data & SocietySAGE

Published: Jan 11, 2022

Keywords: Data shadow; digital phenotype; data double; digital health; ageing; Alzheimer’s disease

There are no references for this article.