“I think it saved me. I think it saved my heart”: The Complex Journey From Self-Tracking With Wearables To Diagnosis

Despite their nonclinical origins, wearables are emerging as valuable tools in supporting the diagnosis of cardiovascular disease, one of the leading causes of death worldwide. Diagnostic data once only available via a cardiologist is now available to consumers simply by wearing a smartwatch, so understanding how smartwatches currently support diagnosis is important for healthcare providers and for the designers of increasingly sophisticated personal informatics technology. We conducted a qualitative study comprising interviews and analysis of posts on an online community of accounts of smartwatch assisted cardiac diagnosis. Our findings reveal how smartwatches bridge a current gap in clinical diagnostic modalities, facilitating a diagnostic journey instigated and shaped by the interplay of self-collected data, bodily self-awareness, and increasing clinical acceptance. These insights focus attention on the consequences of the democratisation of health data, with ethical and design implications for health providers, consumer electronic companies, and third-party application designers.


INTRODUCTION
Personal informatics technologies support the collection of personally meaningful data, facilitating deeper insights and positive behaviour changes [36].Examples of these technologies include smartwatches, fitness trackers, and smart rings.Technological advances have expanded self-tracking capabilities beyond activity tracking and now facilitate greater insights into specific areas, such as cardiovascular health.Research conducted by Deloitte revealed that 37% of smartwatch owners now track heart health data [38].Through the collection of complex physiological data, including electrocardiograms (ECGs) and continuous heart rate monitoring, consumers can access diagnostic insights once only available in a medical setting.
Cardiovascular disease (CVD) is a leading cause of death worldwide, claiming a life every 1.5 seconds [21].CVD rates are rising [45].This trend is influenced by challenges in accessing care, escalating obesity rates, and growing evidence that COVID-19 infection increases the risk of developing CVD [68,74].Early diagnosis of CVD can be lifesaving and crucial for preserving quality of life, as early diagnosis can slow or prevent further heart damage.However, diagnosis is challenging, with estimates that 1 in 3 heart attacks are initially misdiagnosed [73].Notably, there is a marked gender disparity in the diagnosis and treatment of CVD.Despite being a leading cause of death among women, CVD frequently is underdiagnosed and untreated in female patients [66,73].These challenges in diagnosis and treatment underscore the need for innovative approaches in cardiovascular healthcare.
In this context, PI technology is emerging as a promising tool to support CVD diagnosis.These technologies are capable of identifying episodes indicative of atrial fibrillation (AF) [48].However further applications are emerging.Cardiologists have documented the varied and unexpected uses of self-tracking technologies in peer-reviewed articles illustrating their role in supporting the diagnosis of a range of cardiac conditions, from common conditionssuch as coronary artery disease -to rare cardiac conditions such as cardiac sarcoidosis and cardiomyopathy [26,59,67].Leveraging widely owned PI technologies for heart health monitoring could help address the diagnostic challenges of CVD, potentially improving health outcomes in cardiovascular care.
However, little is known about the lived experiences of individuals who have successfully used self-tracking technology to support their cardiac diagnosis.Gaining insights into these successful experiences is important, as it can reveal common tracking practices that users have found effective.This knowledge has implications for designers of PI technology and could inform health professionals on best practices for utilising this technology.
Therefore our work examines how self-tracking can support the diagnosis of cardiac conditions: • What motivated individuals to engage in self-tracking of cardiac health data?• What self-tracking practices are adopted by these individuals to support a cardiac diagnosis?
To answer these questions, we conducted an analysis of posts to an online community exploring accounts of self-tracking assisted cardiac diagnosis and conducted 15 follow-up interviews.Our research contributes an empirical description of the motivations and self-tracking practices individuals adopt to support a cardiac diagnosis.We found that individuals were motivated to engage in self-tracking of cardiac data to validate body signals, to gather evidence to support a diagnosis when traditional clinical diagnostic methods failed, and to react after an alert from their technology.Understanding of self-collected data was often incomplete, requiring complex collaborative sensemaking and action phases.For many, data shaped engagement with medical professionals, sometimes helping them question a lack of diagnosis.We discuss implications for designers of wearables for cardiac health and their integration into clinical diagnostic pathways.

RELATED WORK
To provide background to this work, we introduce an outline of the interdisciplinary research on self-tracking technologies in cardiovascular diagnosis, and the pathology of cardiac conditions that can be detected by self-tracking technologies.We describe seminal work on personal informatics, outlining the current understanding of self-tracking motivations and practices.We then outline HCI research on cardiovascular disease.

Cardiovascular Disease and Self-Tracking Technologies
First-generation PI devices including smartphone applications and wrist-based devices, utilise photoplethysmography (PPG) sensors for pulse rate measurement [27].This capability facilitates the collection of cardiac health metrics such as heart rate variability (HRV).For some wrist-based devices, the PPG sensor operates automatically during exercise, rest, and sleep, passively measuring pulse rate and rhythm [41].PI devices such as the Apple Watch SE and Withings Scanwatch harness this capability to alert users via notification of abnormally irregular, slow, or high heart rhythms, signalling a possible electrical abnormality with the heart.Each notification encourages a user to seek medical advice.Building upon this, second-generation devices allow consumers to take a medical-grade single-lead electrocardiogram (Apple Watch 4) or a more comprehensive 6-lead electrocardiogram (Kardia Mobile).These devices provide automated rhythm interpretation, allowing users to capture ECGs when symptomatic.
Additionally, the capability of PI devices extends beyond ECGs.Many now utilise PPG measurements to provide users with automated interpretations based on a combination of physiological measurements [5] [71].Bentvelzen et al. [5] coined the term derived metrics to describe this form of system-generated data interpretation.Derived metrics can provide information regarding fitness readiness, stress levels, or overall physical exertion [25].However, there is a limitation to this approach as derived metrics often heavily rely on HRV measurements: high HRV can be indicative of good health.However, in the context of some heart conditions, HRV can spike.HRV spikes can lead to misleading scores suggesting low stress levels, or falsely indicating high readiness for physical activity.
Widespread accessibility of PI devices has instigated a shift from traditional physician-led arrhythmia screening to a consumer-led approach [7].This transformation was catalysed by the regulatory approval of many consumer-grade PI tools for detecting the abnormal heart rhythm atrial fibrillation (AF) [19].AF is an abnormal heart rhythm occurring from the heart's atria, affecting 1 in 3 adults, with prevalence increasing with age [69].AF is not in itself a life-threatening condition, but is associated with a substantially increased risk of stroke (quintupling of risk), cognitive impairment, and dementia [40,53].Strokes associated with AF are often more severe, leading to heightened disability and mortality [69].
AF is also associated with other cardiac conditions like valvular heart disease, particularly mitral valve disease, coronary artery disease, cardiomyopathy, congenital heart disease, and hypertension [42].These conditions can disrupt normal electrical signalling and affect the structural integrity of the heart's atria, leading to an electrical misfiring which can initiate AF.Consequently, whilst PI devices can alert a user to the presence of AF, it could be indicative of a serious and sometimes more life-threatening underlying cardiac condition.
Cases of AF can be hard to detect as the arrhythmia can be paroxysmal (come and go) and one-third of patients can be asymptomatic which can result in delays in diagnosis [31].Twenty percent of patients may only be diagnosed after suffering a stroke [31].Traditional diagnostic tests involve opportunistic manual pulse checks followed by a 12-lead ECG and longer periods of rhythm monitoring via ambulatory monitors in symptomatic patients.However, there is limited access to this technology, which can cause delays in diagnosis [23].The technology is limited in capturing paroxysmal AF as the 12 lead ECG only captures 30 seconds of the heart rhythm; ambulatory monitors record for up to 14 days, however, some cannot continuously record heart rhythm and require user engagement to activate a recording [48].
PI devices are a convenient low-cost alternative to traditional screening methods offering the opportunity to capture asymptomatic cases and potentially avoid AF-related strokes [48].Two key studies have documented the efficacy of wearables in detecting AF.The Apple Heart Study reported a positive predictive value (PPV) of 84% for irregular heart rate notifications and 71% for individual PPG recordings [48].The Huawei Study reported a 99.6% PPV and a negative predictive value of 96.2% [18].
However, the utility of consumer-led AF screening remains uncertain.There is a lack of evidence available to determine whether widespread AF screening in this format will be beneficial or harmful [7].PI devices are often marketed towards wealthy health-conscious individuals, who have a different risk profile for AF and stroke [23].The Apple Heart and Huawei Heart studies recruited a small percentage of high-risk participants (those aged over 65 years old) accounting for only 5.9% and 1.8% of the study populations [27,48].As a result, low rates of AF were identified for example 0.52% of all participants in the Apple Heart Study [48].
Whilst high accuracy in detecting AF was reported in controlled environments [48].Independent research suggests significantly lower accuracy rates in nonclinical settings, amongst the elderly population, who are at higher risk of AF [20].Accurate results in this demographic were only achieved when automated rhythm interpretation was supplemented with clinician assessment [20].There is speculation that this technology may assist in AF detection in patients with unexplained stroke.Interestingly, a survey revealed that 92% of cardiologists would recommend wearables for AF screening in post-stroke patients [39].Despite this high rate of clinical acceptance, understanding how such technology is used in a real-world setting by those with varying digital literacy and reduced cognitive functions is not well understood.
Rapid developments in PI technology have outpaced independent research validation, and their impacts on clinical outcomes are unknown [20].A significant concern is the psychological impacts of wearables, particularly health anxiety.False positive results and inconclusive AI-generated reports from these devices have been found to exacerbate health anxiety and lead to unnecessary use of medical resources [57].Rosman et al. [55] highlight this issue, noting that wearables may amplify anxiety in patients, especially those with paroxysmal AF [57].They report a case where a patient received heart rate notifications and ambiguous AI-generated ECG results, this triggered the development of new-onset health anxiety [55].This concern is compounded as anxiety and depression are common in AF patients, often leading to a higher burden of symptoms and reduced quality of life [65].Rosman et al. [55] caution that wearables, by facilitating excessive symptom monitoring, could intensify health awareness, exacerbating anxiety and depression in those already vulnerable to such conditions [57].Despite these critical implications, the potential negative impacts on users' psychological health have been underexplored.

Self-Tracking in Personal Informatics
Personal informatics tools allow the collection of data to support personal reflection, which can generate insights to facilitate positive behaviour change.People are increasingly using this technology to track health data [61].In a review of the HCI literature, Epstein et al. noted that 83% of personal informatics papers covered aspects of health [15], in particular physical activity or mental health [15].
A pioneering model for exploring how people use personal informatics tools is Li et al. 's Stage Based Model [36].The stage-based approach proposed tracking as an iterative, organised process involving: preparation, collection, integration, reflection, and action [36].Li et al. noted that these stages may be user-driven, system driven or a combination of both [36].Choe et al. expanded this approach by describing three data collection methods: manual, automated, and semi-automated (a combination of both) [9].Rooksby et al. [54] extended the discussion by characterising self-tracking as "lived informatics".This approach considered PI technologies in a users' everyday life, emphasising the messiness of engagement and arguing that they shouldn't be characterised as rational data scientists [54].Rooksby et al, identified five overlapping styles of self-tracking: directive tracking, which is when individuals track with a specific goal in mind; documentary tracking, whereby individuals are motivated by curiosity; collecting rewards involves tracking data for achievements and rewards; fetishized tracking is for interest in technology per se; and diagnostic tracking [54].Diagnostic tracking refers to individuals looking for links between data and bodily experiences such as stomach pain and tiredness [54].It closely aligns with the original vision of self-tracking held by the quantified self-movement.
Epstein et al. [17] combined the stage-based model and lived informatics and proposed the 'lived informatics' model which characterises tracking as a dynamic process with integrated approaches, such as collection and reflection occurring simultaneously [17].The model expands on Li et al's work by adding processes such as deciding, selecting, lapsing, and resuming tracking [17].Whilst Rooksby et al. acknowledge the tracking style: diagnostic tracking.It is still unclear what motivations and practices diagnostic trackers adopt and how these experiences unfold in everyday life [54].Recent work in personal informatics has focused on the lived experience of tracking technologies and interactive elements that aim to support it such as flexibility [4], personalisation [32] and preferences in how to log and engage with data [1].

Self Care Technologies & Cardiovascular Disease
The HCI community has explored the design and use of self-care technologies for chronic condition management [44].In cardiovascular disease, HCI researchers have explored the role of technology in both prevention and management in the post diagnosis phase [43]: for example, tablet-based tools to allow hypertensive patients to monitor and transmit blood pressure data to their medical records [35].Others considered the socio-technical challenges of self-monitoring at home in preventing hypertension [35].
Research on the use of technology in cardiac rehabilitation (CR), which is prescribed to patients post coronary event, has been extensive.A systematic review by Tadas et al. [63] reported that research has mostly involved the design of technology to promote positive behaviour changes such as increasing activity and medication adherence [66].Li et al. [36] have started designing a digital platform for patients living with coronary artery disease [52].By incorporating wearable cardiac data and a conversational agent they hope to deliver personalised lifestyle modification.Physiological data from wearables will be used to continuously risk stratify patients, seeking to reduce and prevent further coronary events [62].Supporting patients in the transition from hospital to self-management at home after a coronary event has also been explored [64].
To support heart failure patients Derboven et al. [12] co-designed a smartphone application to work in combination with cognitive behavioural therapy.By measuring vital signs via a wearable, predictive mathematical models were utilised to create tailored lifestyle modification advice to patients.If deterioration in health or wellbeing was detected the patient would receive tailored advice [12].This work has progressed to proof-of-concept trials evidencing positive outcomes for patients [11].
In summary, the HCI community has focused on how and why people use personal informatics technologies [44].Others have explored self-care technologies with a particular focus on patients living with coronary artery disease [64].Little research has focused on the user experience of using self-tracked data in the diagnostic phase of a cardiac condition involving arrhythmia.This is the focus of the current study.

METHOD
To explore our research questions, we conducted a mixed methods approach consisting of a qualitative analysis of online wearable community posts (n=253) and an interview study (n=15).This approach aligns with past PI and self-care technology studies that utilised mixed datasets for analysis (e.g., [16,43]).The online community analysis explored a breadth of participant experiences to further understand specific practices and methods adopted to support a cardiac diagnosis.Interviews allowed us to capture a detailed and more nuanced understanding of these experiences.As identifiers, data extracts drawn from the online community are referred to as OCXX, and interview responses PXX.University of Bristol research ethical approval was obtained.

Online Community Data Collection
We conducted an analysis of several forums of an online community.Following suggested practice in the BPS ethics guidelines for internet mediated research we do not reveal the forum in order to reduce the risk of compromising the anonymity of individuals [6].The online community's extensive user base allowed us to access a broad range of perspectives.Users within these technology forums discussed myriad self-tracking technologies, allowing us to capture a spectrum of experiences across various personal informatics technologies.
Forum posts used in this analysis are publicly available.To protect user privacy, any quotations used in our findings have been modified without changing their inherent meaning, and we took additional steps to ensure these modified quotations could not be traced back to their original source via an online search [28].We were unable to collect demographic information for the authors of these posts.Publicly available demographic information for the online community indicates that most members are under the age of 30, with fewer than 5% of users aged over 65 (at higher risk of cardiovascular disease) [13].The online community that was selected had moderators for each forum to ensure forum rules were adhered to.Users were able to preserve anonymity through the use of pseudonyms when registering.
We collected posts relating to the use of self-tracking technologies in the context of supporting a cardiac diagnosis.To do this we initially searched for PI technology-focused forums via a search engine.The following search terms were used alongside the name of the online community "apple watch", "whoop", "fitbit", "garmin".This allowed us to identify forums dedicated to PI technology.Selected forums were then individually searched using the online community search engine.Initial reviews of relevant posts across the forums allowed the compilation of a key list of search terms relating to successful experiences.Search terms combined commercial technology names, heart conditions, technology features, and outcomes.Example search terms include "apple watch", "ventricular tachycardia" and "high heart rate notifications", as well as "diagnosis".A further example is "withings", "irregular heart beat alert" and "AF diagnosis".The key search terms were then used to search within each forum.A second search was conducted using a search engine in a "incognito window", this ensured previous search history did not impact data collection.This allowed us to generate a list of relevant URLs.We then utilised the online community's API in conjunction with the associated URLs to extract a total of 7664 posts, belonging to 73 threads from 2018 (the year cardiac data was introduced to the Apple Watch) to 2023 (the time of analysis).After filtering out duplicates, we then undertook a second-pass analysis to identify first-person narratives.All other posts were excluded.This approach yielded 253 posts, where community members discussed their personal experience with PI technologies supporting a cardiac diagnosis.

Interview Data Collection
To gain deeper insights into the lived experiences and daily practices of our participants, we undertook a series of interviews.Drawing from the online community study, we formulated a semi-structured interview topic guide to deeply explore the human experience of using PI technology to support a cardiac diagnosis.The interview was piloted with a person who used a wearable to aid diagnosis of a heart condition, and wording and order of the questions were modified slightly to aid comprehension and flow of the interview.To develop rapport and open conversation, the first author discussed her husband's experience of using PI to support cardiac diagnosis.Questions were then asked to explore the cardiac diagnosis journey and technology experience, such as "Can you tell me about your experience of using your Apple Watch and how it led you to discover you had a heart condition?"and "Can you describe your experience of interpreting the data collected on your device?" and "How do you currently determine whether or not to trust the data collected by your device?".Interviews lasted between 60-90 minutes, resulting in approximately 18 hours of audio recordings.Given the geographical spread of participants, interviews were conducted online and audio recorded.Participants were reimbursed for their time with a £20 Amazon gift voucher or charitable donation of their choice.

Participants & Recruitment
Recruitment was conducted directly by contacting the authors of posts on the online community, through a sign-up survey on cardiac diagnosis with wearables posted on technology forums, and through word of mouth.Interview participants had to be aged over 18 years old and have had first-hand experience using selftracking technologies to support a cardiac diagnosis.All interview participants were self-declared to have received a clinical diagnosis.Interview participants were diverse in age, technology use, and cardiac conditions diagnosed; they were based in either Europe or the US (see Table 1).Although many participants reported using multiple PI tools, the participant table lists the primary wearable used in the diagnosis of their cardiac condition.

Data Analysis
The online community and interview data was analysed in an inductive "bottom up" approach [47], following Braun & Clarke's thematic analysis method [8].A preliminary analysis of the online community data was made.This approach helped identify areas that we needed to collect more data on and aided the formation of the interview questions accordingly.Once interviews began data was analysed in parallel.
The first author initially coded the data, with codes being developed and iterated after each interview.The probes used with the interview topic guide evolved in response to the data being collected.Upon completion of all interviews, the first author re-read the online community posts and interview transcripts in full and open-coded the data in full.All authors met on a weekly basis to collaboratively review and discuss the coding.
The first, second and third authors then collectively and systematically compared and contrasted experiences across the interview participants by mapping out each experience by creating a flow diagram for each participant.The initial motivation was to be able to visualise the general diagnosis journey participants went through to present as a contribution, but the complexity of each journey would not allow this approach.Instead, these mappings allowed the identification of trends, similarities, and differences.Given the diverse geographical location of participants, further distinctions were identified such as interactions with medical professionals and experiences with healthcare systems.This cross-comparison led to the formulation of distinct themes, which were iterated on by all authors through co-analysis.The derived themes were compared against theoretical models in personal informatics and resulted in higher-level overarching themes.The online community posts underwent the same thematic analysis approach as part of the entire data corpus; however, the cross-comparison aspect was not applicable to the online community posts due to user anonymity.

Positionality Statement
The inspiration for this research originated from a personal connection to the subject matter.The first author was closely involved in a situation where personal informatics technology played a pivotal role in supporting the diagnosis of a rare cardiac condition called myocarditis in a family member.The first author's background in clinical cardiology became instrumental in collaboratively interpreting and acting upon the data from the PI technology.Witnessing this experience, the first author found herself informally adopting an ethnographic lens, observing how the family member continued to interact, rely upon, and interpret the data from his PI technology.It is through this unique experience that the first author came to reflect on the complexity of using PI technologies in the context of supporting a cardiac diagnosis.This personal involvement has undeniably shaped the approach to this research.It not only provided her with greater empathy and understanding for the participants she interviewed but also influenced how she interpreted their experiences in the interviews and the online community posts.

FINDINGS
We present the findings in two parts, drawing on the experiences of the online community (OC) and interview participants (P).First, we present three vignettes that capture the depth and nuances of the temporal lived experience of using PI as part of diagnosis.We chose these narratives because they encapsulate the complexity of participant experiences, whilst also representing common themes echoed by others in the study.These real-life vignettes serve as a precursor to the thematic analysis that follows [56].We then characterise the motivations that led participants to self-track (to validate bodily signals; to gather evidence; and as a reaction to alerts) and also their self-tracking practices (sense-making, datadriven interactions with healthcare professionals, and self-tracking during the diagnostic process).James was feeling well and preparing to go on holiday when unexpectedly he lost consciousness.It took his wife, Julie, some time to wake James up.After a few minutes, they decided to use a pulse oximeter, which they had purchased during the pandemic.Placing it on his finger, the oxygen levels read 95% and heartrate read 33 bpm (normal readings 96-99% 50-90 bpm).Unsure if this was an invalid reading, Julie subsequently placed the device on multiple fingers only to get the same readings.She decided to check her own finger and all the readings were within normal range.

Vignettes
Julie then realised that James could use the ECG function on his Apple watch.With the help of Julie, James took his ECG but it read "heart rate too low".Julie called his personal physician, who advised her to call an ambulance.Ambulance technicians and the emergency staff dismissed the reading from the pulse oximeter, which James thinks is because it doesn't produce a record or perhaps because it is a cheap device.In the emergency department, James told the doctor that the watch ECG didn't work immediately after passing out.The doctor looked at Apple Health on James' phone and upon reviewing this ECG data, said "you've just admitted yourself to the hospital".
James believes the data provided by his watch was an essential piece of information that prompted the doctors to admit him: "Apple watch and Oura ring saved my life".That evening in the hospital James's heart stopped completely, requiring an emergency pacemaker implantation procedure.When reflecting on how different things could have been he says "had I not had that information they would have sent me home from the hospital".and "there's a good chance during my time at home I would, well it would have been the end".James has since activated the heart rate alerts on his watch and has bought a watch for his daughter who also has a heart condition.always felt a skipped heartbeat; it was subtle yet persistent and she found it deeply unsettling.This weighed on Mia's emotional wellbeing and she grappled with disturbing thoughts: was she a "ticking time bomb" about to "keel over and die"?Despite many visits to cardiologists, they found no abnormalities so she started to doubt her intuition.Mia used Fitbit for many years as an avid fitness fanatic, but she found the individual metrics overwhelming and transitioned to Whoop.Mia was intrigued by the simplicity of Whoop's recovery scores, which translate various data points into a daily measure of bodily preparedness (low scores can indicate overtraining or sickness whilst high scores suggest the body is well-rested and ready for a day of strenuous activity).Mia would check her recovery score each morning and started to realize that her highest recovery scores always aligned with the days her heart felt out of rhythm.The score meant to signal that her body was ready for a day of vigorous activity became an indicator of concern for Mia.She started to use her phone to measure her pulse which presented a visual representation of her heartbeat and realized that the visual always looked odd on days she had her highest recovery scores.Mia started to believe her data confirmed her suspicions that something was wrong, and this gave her the confidence to see a cardiologist one more time.The cardiologist encouraged Mia to buy the Apple watch as this would allow her to capture her ECG at times when her heart felt out of rhythm.
With the Apple Watch ECGs and further clinical tests, Mia finally received a diagnosis: a benign (non-harmful) arrhythmia which is common in athletes and does not require treatment.The association of the high recovery score on Whoop with arrhythmia is probably due to the use of high heart rate variability in this derived metric.Her diagnosis has brought about a sense of relief and understanding, dispelling uncertainty and as Mia puts it made it "easier to continue living life".She has gone back to using the Whoop as she prefers the simplicity of the recovery score to the more medical data provided by the Apple Watch and feels able to appropriate the metric to interpret her own body signals.
4.1.3Timekeeping and the Surprise Diagnosis.Oliver (P11) is in his early 30s and based in the UK.He started wearing the Withings smartwatch with a traditional watch aesthetic 3 years ago, which he used predominantly to tell the time.After starting to wear the watch, Oliver received a notification from his watch saying "Signs of Afib detected, carry out ECG".Oliver had "only had the watch for a week" and had never heard of AFib before, so he thought his power tools were affecting the watch or maybe it was caused by a faulty software update.He ignored the alert and the advice to take an ECG.
The alerts continued and after a few months, Oliver started to realize that he was feeling quite unwell.He decided to mute the alerts, but still keep a record of them.He checked the muted AFib alerts to see there was a significant number stored.Keeping the alerts muted Oliver continued "to make sure it wasn't, self-induced", but he continued to feel unwell and started to notice his heart racing during the day.This started to happen more often.Oliver finally decided to take his ECG with his watch.The Withings interpretation indicated "signs of atrial fibrillation".Oliver decided to share his ECG online.After feedback from peers familiar with his watch, he subsequently booked an appointment with his GP.
The GP did not review Oliver's watch data despite Oliver offering it for review, and he reflected that "they told me not to wear the watch at all, they told me to turn everything off and don't wear it!".The GP arranged several tests at the local hospital and it took a few months for these tests to be done.Oliver was prescribed medication to treat his abnormal heart rhythm by his GP.However, Oliver never received a referral to see a cardiologist.
Oliver still gets alerts, but only when he is sleeping and he no longer feels his heart race during the day.Oliver continues to look at his "Signs of Afib" alerts and at his heart rate graph, and reflects on how he has been feeling.He is not concerned by the alerts as he feels "all right" in himself.Oliver says he will only go back to the GP if he feels ill again, and the experience "impacted my trust in them [GP]".Oliver suggested that "they want to diagnose you, they don't want to be told because the device has told you." When discussing the potential for wearables and heart health, Oliver reflects that there is a risk that people won't seek help until they are experiencing symptoms, he acknowledges that maybe "that's probably when it's too late".

Motivations to Engage in Self-Tracking for Heart Health
Many participants were motivated to self-track to validate their bodily experiences to themselves and to others, fuelled by challenges in trusting their perceptions of their bodily signals.Others noted challenges in having their bodily experiences recognized and acted upon by medical professionals, often when symptoms did not fit with more established diagnostic criteria.In some cases, selftracking device alerts served as a nudge to specifically self-track cardiac health data.Others engaged in self-tracking as a form of evidence gathering to support diagnosis.

Validating Bodily Signals.
Before medical engagement and diagnosis, the first step participants took in becoming aware of an illness was to recognize symptoms, which were often combined with a bodily intuition that something 'feels' wrong.Participant motivations to engage in self-tracking of cardiac data were instigated by these felt bodily experiences: "I know how I felt. . .something was going on" (P2).Often challenges in understanding bodily experiences instigated self-tracking of heart data.P5 for instance had never used the ECG function on his watch until he started to recognise unusual symptoms: "Suddenly my heart rate shot up. . .I became flush and warm and hot and felt a lot of discomfort in my chest.
And I remembered ohh right, this thing can take an ECG.So I did, And just watching the patterns was a wild experience for me, because I thought 'ohh yeah, OK, something is happening.I'm not making this up'.And the watch said, 'seems like atrial fibrillation for sure.'"(P5) P5's experience illustrates a level of uncertainty about his physical experiences; they happened suddenly and are ambiguous in nature.The participant expressed relief when data supported his bodily experiences, and this was further confirmed by an AI-driven ECG interpretation.Adding a layer of tracked data to confirm his subjective experiences, gave reassurance that symptoms were physiological and not a product of anxiety, stress, or imagination.P4 shared similar motivations to validate bodily experiences explaining that when symptoms occurred, he would take an ECG to gain data to verify his felt experience: "having something which can verify that was something which I wanted" (P4).
Participants expressed that having validation made their bodily experiences feel more tangible and objective, and therefore they felt it was worth taking action over: "I felt my heart fluttering so I took my ECG on my watch, it confirmed I was in atrial fibrillation so I went to the emergency room." (OC189).For these participants, data validation had positive implications -validation prompted them to seek medical attention: "I was feeling pretty well, but when out running I noticed my heart would race all of a sudden, the racing wouldn't stop.My watch just launched an ECG feature, so I used it just to see what was happening.The ECG flagged an irregular heartbeat warning. . .I started to feel concerned and went to A&E [the Emergency Room] that evening" (OC61).The ECG feedback backed up the participant's feelings and influenced taking action.In this instance, the smartwatch ECG results led to the early diagnosis of valvular heart disease (OC61).
Wearables provide a valuable mechanism for validating body signals.However, validation can sometimes be misleading and dangerous as it might encourage misinterpretation of symptoms.P4 found comfort in using his wearable when experiencing chest discomfort explaining, "it might just be. . .indigestion or something like that. . . to be able to just have a quick ECG trace and you know.. it might put my mind to rest a little bit".In this case, the technology may have given a false sense of security and may have led to a misinterpretation of symptoms.Wearables are designed to identify a specific electrical abnormality of the heart (AF) and cannot identify conditions such as coronary artery disease, which may present with similar symptoms.

Evidence
Gathering.The practice of evidence gathering took place when initial medical investigations failed to detect an abnormal heart rhythm.Going beyond initial responses to body signals, some participants were motivated to self-track by the need to capture heart rhythm abnormalities that traditional diagnostic tests didn't capture.For instance, one participant shared "I bought my watch because they could never capture what was going on with my heart with the hospital-issued heart monitors.I sent my watch recordings to the cardiologist who was pleased I managed to capture what was going on" (OC101).As heart rhythm abnormalities can be sporadic and unpredictable, participants emphasised the convenience and ease of capturing an ECG through wearables as a vital element in supporting their cardiac diagnosis.P12 developed a perceived racing heart rate after a COVID-19 infection, however clinical tests failed to capture an arrhythmia.He decided to purchase a smartwatch which allowed him to capture an ECG whilst his heart was racing, allowing him to engage with diagnostic technology at the right time outside of clinical settings: "I wouldn't have known I had AF if I didn't have the watch. . .without the watch, I would still be searching" (P12).
In contrast, other participants took up self-tracking because doctors could not establish the root cause of perceived symptoms: "Since I was a teenager I would get these dizzy spells and feel like I couldn't control my breathing, The doctors could never find anything wrong, so I bought my Fitbit.So I can say hey, look, I'm really not bonkers. . . the graph shows it" (OC206).Self-collected data for this participant was a critical component in receiving a diagnosis.For a prolonged period, they were left searching for an explanation for the cause of their symptoms.Data brought confidence to seek a diagnosis one more time, causing a shift in the traditional patientdoctor relationship.
For participants who had already been using self-tracking technology, their tracking goals suddenly shifted when traditional diagnostic methods failed to capture their sporadic heart rhythm abnormalities: "We can't have heart monitors and Zio Patches [hospital issued heart monitor] attached to us at all hours of the day at all days of the year."(P5).P6 compared wearing his watch to always having a camera on hand: "They say the best camera is the one you have on you at the time, and the same thing with this.I mean the only way I was getting that recording [ECG] was through a watch".This sentiment was echoed by many who found traditional methods were unable to capture rhythm abnormalities: "The cardiologist had me wear their heart monitor for 2 weeks, but it didn't catch anything.They just put it down to anxiety and stress.I started using the ECG feature on my watch, I was able to show these to my doctor. . .this concerned him so he made a referral to see a specialist cardiologist.I'm now waiting on my heart operation" (OC207) This illustrates the transient nature of symptoms, and how the watch provided useful and actionable data not only to the user but also to the medical professionals involved in the participant's care.OC200 had a similar experience "The doctors really appreciated my watch ECGs, as every time I made it to the hospital my episodes had passed" (OC200).For these participants, their medical professionals had a positive response to their self-collected data.
For others, heart rate data was a form of evidence that shaped responsiveness from healthcare professionals: "My watch is the primary reason why my general practitioner got a cardiologist involved, I showed her all my heart rate data and that's when she started to be alarmed, and made a cardiology referral straight away" (OC187).Some participants described how self-collected data shaped medical care by expediting access to specialized care: "I'm glad I had this [watch] because it accelerated the timeline of getting diagnosed" (P5).
The participants described the utility of self-collected data as evidence, which shaped engagement with medical professionals.There is currently no non-invasive medical device available that can collect ECG data for prolonged periods of time.The sporadic and unpredictable nature of arrhythmias often leads to doctors organising heart monitors which can be worn for as long as 14 days.Often these diagnostic tests are ordered repeatedly to capture evidence of an arrhythmia.We illustrate how participants sought an alternative route to diagnosis through self-tracking.We must highlight however that whilst many experiences of using data as evidence were positive, this was not the case for all participants.Oliver (see section 4.1.3)did not have a positive response to his self-collected data from his general practitioner.P1 shared a similar experience in the emergency room when her self-collected ECG which indicated AF whilst feeling very unwell: "they don't care. . .they are very busy. . .they don't like you to be so assertive I suppose".

Alerts as Nudges to
Self-Track Health Data.Automated alerts often served as the initial cue to recognise potential health issues, nudging participants to consider their cardiac health status: "One morning I woke up to a low heart rate alert on my watch.I took an ECG on my watch and that captured what was going on." (OC31).The alert provoked this individual to focus on data collection and quickly reflect on a snapshot of data presented to them.The smartwatch alert facilitated a call to action, whilst the notification did not provide guidance on what data to collect in real-time, this participant recognised the importance of taking their ECG urgently.
Participants also showed how alerts caused a shift from passive tracking to actively engaging in tracking.P8 explained that he bought his watch primarily because "it was useful and it told the time. . . .I wasn't wearing it, purposely to check any particular health issue".He explained that he only engaged in self-tracking after receiving an alert, which eventually led to clinical engagement.He was diagnosed with valvular heart disease, and credits the smartwatch for starting that journey: "the alert was the key thing" (P8).P8 was asymptomatic to his arrhythmia; the alert was a critical factor in supporting this participant in recognising ill health.We also see in Oliver's vignette that he started to engage actively in self-tracking, after reflecting on the number of alerts he received.Whilst in his case he did not trust the alerts initially, they did spark a transition from passive tracking to more engagement with tracked data.In Oliver's case, self-tracking was instigated by a need to make sense of the alerts he had received.For P7 the high volume and continuous stream of alerts prompted him to shift from passive to active self-tracking.Notably, the majority of participants did not turn off health notifications.
These participant experiences illustrate how alerts often instigated a shift in tracking goals, as in P7's case where he went from counting steps he explained he used his watch "predominantly to keep an eye on my walking" and transitioned to tracking his heart data after receiving an alert.Alerts for many were unexpected as participants in some cases had no symptoms related to heart health: "I wasn't aware of any symptoms. . .before the watch warned me" (P8).The alerts led some participants to re-evaluate and sometimes expand their self-tracking practices.Upon receiving alerts on his current device, P8 upgraded his smartwatch to one with additional heart health functionality to capture more detailed data, like ECGs, explaining his "natural inclination was to buy a new watch" (P8).Many participants upgraded to be able to access the ECG collection capability, combined with improved battery longevity.The improved battery longevity allowed for longer periods of passive monitoring of heart rate, which is required to fully utilise system driven alerts.This illustrates how the alert served as a nudge to participants and resulted in a shift in self-tracking goals, setting them on the journey toward diagnosis.Participants also expanded their self-tracking toolkits with the adoption of blood pressure monitors and smart scales (P10) and continuous glucose monitoring (P2) (this participant did not have diabetes) to understand their heart health and overall health.
Alerts facilitated further data collection for many participants, and for our participants heart rate and rhythm alerts often prompted a user to seek medical advice that eventually led to diagnosis.Some smartwatches also offered guided prompts to users to self-collect an ECG in real-time.However, we found an unintended consequence of alerts was that they often sparked participants to enter a cycle of continuous data collection over longer periods of time.Some participants felt the need to collect extensive data over time to verify and trust the alert's accuracy before consulting a medical professional.

Self-Tracking Practices to Support Diagnosis
Participants adopted a variety of self-tracking practices to make sense of cardiac data, to engage in clinical encounters.As the diagnostic process is not immediate, we also show how participants continued to use their devices throughout the sometimes long journey toward diagnosis.

Making
Sense of Personal Cardiac Data.Participants reported struggles in understanding their self-collected health data, and some participants developed their own strategies to interpret cardiac data: "what I understand. . . .if its something that's repeatable and not normal. . .definitely get it checked out" (P2).This participant took a systematic approach, taking several measurements to compare data if it deviated from the norm for him to feel like it was actionable and viable to show a clinician.P6 also adopted his own strategy by using visual pattern-based cues to understand his ECG data: "My eye for it is basically, you know, it's all symmetrical.It's rhythmical, you know.Then I think, OK, that's not so bad.But if it's, you know, crazy in all over the place, then it's like, OK, this probably requires a closer look" (P6).Participants interpreted data through lenses that they had personally constructed, and set their own tipping point for starting the diagnosis process with a clinician.However, some participants did not make that interpretation leap with their cardiac data, and instead went straight to the professionals: "I'm not a doctor. . .I can't read an ECG" (P2).
For many participants, symptoms played a key role in shaping how they understood self-collected data.For some their subjective experiences unlocked understanding of their objective data, such as when Mia's (see 4.1.2)high recovery scores on her wearable always aligned with the days she experienced her symptoms.For Mia, her bodily self-awareness influenced how she interpreted and made sense of the recovery score, so much so that her interpretation of the data contradicted the AI-generated interpretation from her device and she appropriated the metric as a sign of concern.When probed she explained "It needs to correspond with how I feel".She preferred to collect and reflect on data that aligned with her bodily awareness, and this influenced the self-tracking technology she chose to continue to use through the diagnosis process.Mia's experience showed the simplicity of the derived metrics facilitated understanding, which Mia personally constructed.
In contrast P7 did not recognize any symptoms.When receiving an alert he struggled to make sense of the data as it did not align with his bodily self-awareness and came unexpectedly.He ignored alerts for several months, sense-making became a prolonged and evolving process.As P7 had been counting steps, but had never actively engaged with his health data before receiving an alert, he was unfamiliar with his baseline "normal" measurements.He was sceptical when he first received an alert, unsure if his watch was "playing up" or if it was caused by a "software update".He purchased a third-party app to visualise his heart rates.Still unsure, he bought a blood pressure monitor which allowed him to see his "heart rate plummets at times" via a live heart rate visualization.After some time, he started to collect his ECG on his watch "but it's just a pretty graph that goes zigzag up and down and I haven't got a clue about what it means really, realistically. . . .But it is nice to have. . ..it does show it [the heart] playing up" (P7).For this participant, sensemaking was an iterative process in which he used multiple self-tracking tools over several months to form his own interpretation of the data, and only established understanding after a prompt from his wife instigated collaborative sense-making with a healthcare professional.However, sensemaking could not always be an iterative process, as in James' vignette where we see the urgency with which he and his wife had to make decisions.Whilst James and his wife did not fully understand the data they were collecting, they were able to understand it enough to call his personal doctor who advised them to call an ambulance.While James was fortunate to be able to gain access to a professional to facilitate collaborative sense-making in an emergency, this is not always the case in different health systems.Similarly, P6 explained his access to a specialist medical professional meant they did not need to understand complex cardiac date: "There is nothing to watch. . ..I view it the way I kind of view accounting. . ..it doesn't really interest me. . .it's not something I want to expend brain power on. . ..I pay a guy for that. . .that's why I have a cardiologist." (P6).
When participants struggled to access specialist care from a cardiologist they leaned on AI-generated interpretations to elicit an understanding of ECG data: "I do rely on the interpretation it provides me" (P4).P1 was more explicit suggesting she leaves the interpretation to the smartwatch: "no, no it just says it" (P1) when referring to AI generated reports.Some participants expanded their self-tracking tool kit to gain more sophisticated AI interpretation tools during their diagnosis journey.P12 supplemented his watch with a dedicated ECG recording device so that it could interpret a greater range of heart rhythms and provide him with more detailed reports.
All interview participants eventually engaged in collaborative sensemaking with a medical professional, leading to a diagnosis.Whilst many participants had positive experiences with medical professionals and their self-tracked cardiac data, a minority of participants did not have a positive experience as the health professional did not support them in understanding their self-collected data.The failure in collaborative sense-making was seen in Oliver's vignette (4.1.3)resulted in Oliver deciding to base his understanding of objective data purely on his subjective experiences, revealing that he would not go back to the doctor unless he started to feel unwell.

Using
Personal Cardiac Data in Clinical Engagements.After self-tracking cardiac data, it was used to help participants to engage with clinicians.P2 explained what happened when he shared his self-collected ECG in the emergency room "I was able to show them what I had captured. . .and they actually took it relatively seriously, I guess cause they were younger doctors. . .they had a cardiologist in the hospital and the ER look at it." Data in some cases acted as a prompt to engage with doctors:"I don't know how long it would have been.I would have gone without knowing I had A-fib if it wasn't for my watch and you know, throwing up that red flag to which got me to talk to my cardiologist." (P12).
Data also supported participants in foregrounding their own lived experiences, and they reported a perceived or real need to provide evidence to counteract clinician assumptions:"Thanks to the watch, the doctors started taking me seriously, they just always thought I was suffering from stress and anxiety" (OC105).For many participants, notably, many female interview participants, tracking practices were shaped by a need to collect data before engaging with medical professionals.P14 explained that her mother, aunt and cousin had palpitations but didn't get a diagnosis until it was too late and they developed heart failure.P14 elaborated that she also "had palpitations since the age of 17", but at the time her doctor believed her symptoms were due to hormones and stress.The participant's interactions with her doctor at the age of 17 deeply affected her ongoing confidence and trust in medical professionals.P14 believed that symptom evidence alone was not enough, and she felt strongly that she needed to have a role in her treatment approach.Following advice from her cousin, P14 collected her ECGs and "printed them all, I went in with a stack of papers like this high [participant holds hand up indicating a substantial amount of paper]".When the cardiologist suggested he would arrange further tests, she told him "well go ahead, but that's a roll of the dice if you're gonna catch anything...I don't want any more of these events."She encouraged the doctor to look at her self-collected data and gave him a list of times and durations of her abnormal heart rhythms.The doctor wrote a prescription and arranged further tests.P14 finally received her diagnosis at age 65.Upon reflection she said "if I had not had my watch, and not done the ECG. . .I may have gone down the same path they [mother, aunt, and cousin] did without treatment. . .I'm 100% convinced" (P14).
Other participants also leveraged data collection to challenge established perspectives from healthcare professionals: "My doctors completely dismissed my fainting spells. . .always blaming them on anxiety. . ..my watch tracked that my heart rate was dropping all of a sudden when I was collapsing. . ... I had the proof I needed to be taken seriously. . .I finally have my diagnosis" (OC130).This participant used data as evidence to challenge the conception that they were suffering from anxiety, and they believed objective evidence was essential to challenge the initial assessment of their healthcare provider.
We illustrate how cardiac health data empowered participants to engage with clinicians and importantly challenge the care they received.Many participants felt that presenting to a doctor with symptoms was not enough.In cardiovascular medicine, women are known to present with atypical (unusual) symptoms and combined with a deep cultural belief that heart disease is a man's disease [34], women are often underdiagnosed and undertreated [24].Female participants used technology to validate and counteract preconceptions.

Continued Self-Tracking
During the Diagnostic Process.Some participants in this study faced prolonged diagnostic periods as they were dealing with complex heart conditions like valvular heart disease and heart failure.These conditions often require myriad tests to confirm diagnosis.Participants found themselves caught in no man's land: knowing they had a cardiac condition, but uncertain about the cause, treatment or how serious the condition was.P7 was awaiting MRI results, aware his heart was failing but unsure why.He did not know how serious the condition was or to what extent the condition was going to impact his quality of life.He explained that he continued to use alerts as they helped him "take more notice" and helped him "take note".He observed that he was starting to "feel more short of breath, palpitations" and "a few more dizzy spells than normal."Whilst the alerts helped the participant become more in tune with his newly diagnosed condition, he was unable to act on the data as he could not easily access care from a cardiologist: "I'm not really bothering that much with them [the alerts]" as "I know it's something they [the doctors] are trying to sort out. . .what they will end up doing I don't know".Whilst alerts held value for this participant, they also evoked helplessness as he alone could not act on his self-collected data.
Diagnosis, particularly for conditions like AF is not a static oneoff event, but a dynamic, ongoing process.This is due to the fluctuating nature of the condition with the abnormal rhythm coming and going.This was reflected by participants continuing to self-track whilst waiting for test results and clarification.They continued to use the data to determine if more medical engagement should be sought.P6 explained how his smartwatch became a useful preliminary gauge, that enabled him to monitor his symptoms in real-time before deciding whether medical engagement was required "I knew it was a toy. . .but I had enough faith in it that it would tell me if there is enough of a problem...I don't wanna go running to my doctor every time I have a flutter" (P6).Other participants continued to self-track to assist in the interpretation of bodily signals, whilst treatment approaches were optimised.P3 was in the process of having her medication titrated, the medication was required to reduce extra heartbeats caused by myocardial scarring.Unsure if the medication was working P3 explained her approach "There have been times where I feel like my heart pounding, but the data says my heart rate is 80".For this participant, data became an integral part of how she interpreted her body signals, and she became highly dependent on her device ECG functionality: "I rely on that 100% like if it shows that everything's fine and I, but I feel my heart pounding out of my chest and I know it's probably just fluke.Just let it go.We'll check it again in 5 minutes."The participant used data to override her interpretation of bodily experiences, choosing to trust the automated data interpretation her device provided.Later she admitted that losing this functionality post-diagnosis would be devastating: "Oh I'd be really sad, that is more important to me than I think I realise" (P3).Self-collected data became an integral part of the participants everyday health management.With self collected data influencing how she engaged and reflected on her heart condition.
Other participants explained how their experiences influenced their decision to upgrade: "I was wearing the watch and it alerted me to a high heart rate whilst I was in hospital with COVID, I was able to tell my nurse who called the cardiac team.As soon as I got out of hospital I upgraded my watch" (OC190).Participants upgraded self-tracking tools if they wanted functionality that allowed passive tracking of heart rhythm with integrated alerts.P4 noted that he didn't trust himself to be aware of all his arrhythmia episodes and his watch was helpful in keeping "an eye on things".P4 explained, "That's why I use this device.I want to make sure that I've got some sort of eye on. . .I wanted to be able to catch episodes when I might not be aware myself".
The emotional toll that participants experienced was significant and many participants noted that their device brought reassurance throughout the diagnosis process.As OC95 noted, "Once you've been through something like that you get a lot of anxiety around your health, the watch gives me a lot of reassurance".This sense of emotional security was echoed by many "the watch is a comfort to me, to know I can monitor my heart and share that information with the doctor, its valuable to me" (OC91).
As diagnosis rarely came immediately upon engaging medical advice, our participants continued to use the wearables to track their cardiac health, to determine when they needed to go back to their doctor, and for emotional support in a time of uncertainty.

DISCUSSION
This study provides an empirical account of how wearables can support cardiac diagnosis.This has previously only been reported in the popular press and case studies in medical journals.Based on this study we present design implications for how wearables may be designed to support appropriate medical engagement and cardiovascular diagnostic pathways.

Design Implications for Wearables in Cardiac Diagnosis
5.1.1Critical health data and user experience.Our findings show that self-tracking cardiac health does not follow a linear [36] or simple path [17] from data collection, to reflection and action.User interpretation of cardiac data is varied and sometimes inaccurate, leading to fragmented understanding, affecting reflection and action.Unlike basic health metrics like counting steps, cardiac data interpretation involves potentially safety-critical decisions.Participants viewed their data through a unique, personal lens.This approach could lead to negative consequences such as health anxiety from false positives or complacency from false negatives.
To mitigate these challenges design features could include easier data sharing capability with healthcare professionals allowing clinician interpretation and feedback.To facilitate greater understanding and appropriate action, clear explanations of data, potential risks and guidance on when to seek medical advice should be provided to users.
We observed varied user preferences in interpreting data, suggesting the need for versatile visualisations that cater to diverse levels of data and health literacy.This is crucial as misinterpretation could contribute to negative consequences and carries the risk of inappropriate or delayed medical engagement, negatively impacting health outcomes [37].Psychological factors such as illness denial also contributed to delayed medical engagement in our participants [22].To address this, designers could draw on approaches such as soma design or micro-phenomenology, to enhance users' awareness of body signals alongside automatically tracked data [30,51].There have been moves towards this in HCI, particularly with regards to taking self-tracking beyond data, towards a felt experience, from biodata to soma data [3,30].Designers could encourage users to acknowledge and log symptoms alongside self-tracked data.Ayobi et al. demonstrated the potential of customisable visualisations to track felt experiences when managing multiple sclerosis [4].Translating felt experiences into personalised visualisations may foster deeper reflection and understanding [4].
5.1.2Trust and Validation in Self-Tracking.Throughout the study, trust was multifaceted and shaped by self-validation and verification in a clinical setting.Some users often found their self-collected data overlooked by general practitioners, yet taken more seriously by cardiologists.In these cases, we saw examples of participants collecting data that went beyond self-reflection towards that which could be used to challenge clinical assessments.Chung et al. [10] demonstrated how patient generated data (PGD) supported conversation in clinical consultation by acting as a boundary artefact.Our findings point to a necessity to make data visualisations that are clinically credible and cater for specific clinical contexts.
Levels of trust evolved as participants moved from passive to active tracking.Some participants collected extensive data to validate their felt experiences, especially in cases where symptoms were subtle or absent.System alerts intended to prompt medical interactions for many participants led to further self-tracking initially which was followed by seeking medical advice.This behaviour may reflect varying levels of trust combined with preconceived beliefs about when it is appropriate to engage with a health professional.Many participants rationalised a health concern before seeking medical advice [2].Our findings also suggest a belief that medical advice should only be sought when experiencing overt symptoms.PI interfaces could offer educational content about early-stage symptoms and guidance on conditions which may not have overt symptoms.
5.1.3Designing for Multiple Stakeholders.Elsden et al. [14] and Potapov et al [50] have discussed the social aspects of personal informatics, and this is no different for wearables used for heart health.In our data, we saw the crucial role of a spouse in the emergency response to cardiac wearable data.As with much of health and care, sensemaking is not experienced in a vacuum.There are many stakeholders involved, such as family, friends, neighbours, and professionals [58].Social factors influenced the process of sensemaking for participants.Often a loved one was involved in sensemaking and encouraging a participant to seek medical advice, which has parallels to the family informatics literature in HCI [49].Additionally, our study shows participants seeking assistance from strangers online to aid understanding.This behaviour has also been seen in other chronic condition research where people are not getting enough clinical support, which has implications for the design of future wearables [46].They are often considered 'personal' informatics, but the roles of others is key in diagnosis and overall cardiac health.Chung et al. support this notion, calling for the development of PI models to consider the roles that experts play across self-tracking stages [10].

5.1.4
Informing Future PI Models: Alerts & AI.. Lived informatics models from Epstein et al. [17] and particularly Rooksby et al. [54] consider the chaotic and often non-rational nature of self-tracking.However, they do not consider the complexities introduced by alerts, AI-driven interpretation, and derived metrics.Our study found that participants relied heavily on alerts and AI interpretations for health monitoring.Current PI models presume active engagement in data collection and user interpretation [17,36].In contrast, our findings illustrate how system-generated alerts were critical triggers for participants to move between phases of passive, proactive, and reactive engagement in self-tracking.Alerts impacted multiple stages of selftracking.They particularly impacted reflection phases, prompting users to reflect on specific data, guided at a system level which was often associated with automated interpretation.Over-reliance on alerts may mean users no longer have a deep engagement with data and may not engage in personal reflection.There is also a risk that alerts and derived metrics may impact an individual's agency to set tracking goals, reflect and act upon data.These are important considerations for future PI models Our findings indicate that whilst health alerts aim to prompt actions such as seeking medical advice, they often result in unpredictable responses, including initial disregard and delayed action until alerts become frequent.This can lead to cycles of data collection and re-examination of previously gathered data to enhance understanding.Ignoring legitimate health alerts poses health risks, raising ethical considerations for the design of PI devices.For instance, should health alerts be shared with experts such as a user's physician or should users retain complete autonomy and privacy?There has been work on notifications in HCI [60], and further research is needed to examine the psychological impact of receiving a health warning via a wearable.
Whilst we recognise the importance of alerts in supporting an initial diagnosis, designers must also consider how they can trigger health anxiety [55].Customisable alerts for users with a confirmed diagnosis may mitigate this risk.Collaboration between a user and health professional in customising these alerts could help address anxiety and prevent misuse of medical resources.

Integrating Wearables into the Clinical Pathway
Participants discussed the tension between out-of-clinic data and in-clinic data.This cannot be easily generalized, as different participants trusted one rather than the other depending on context.Interestingly, the ongoing data collection afforded by wearable devices elevated participants' perceptions of this data compared to the clinical diagnostic technologies.This was exacerbated by the limitations of clinical technologies.Illness episodes are sporadic for some participants, so the ability to capture everyday cardiac data whilst experiencing symptoms was invaluable for participants.Despite the success of many participants in navigating the clinical care pathway towards diagnosis, there were instances where this did not occur based on the medical role that their point of contact played.Access to specialist care (e.g.cardiologists) is not necessarily guaranteed.For many in the United Kingdom for example, and for those on certain health plans in the United States, general practitioners act as gatekeepers to specialised care.Research indicates that GPs can safely rule out atrial fibrillation from self-collected ECGs but when an abnormality is present, they are incorrect 50% of the time at reading the self-collected ECG [33].This suggests that if a GP suspects any rhythm abnormality they should seek confirmation from a cardiologist, who can interpret abnormal self-collected ECGs with greater accuracy [72].This highlights that the design of data outputs and data visualisation should be designed with the diagnostic pathways in mind, supporting both user and clinician.One approach could be to include educational content specifically aimed at supporting clinician interpretation of self-collected ECGs.
Designers should also be mindful that accessing specialist care may not be prompt, resulting in delays in collaborative sensemaking and action phases of self-tracking.Designers could consider supportive measures to enhance user understanding during this time.For example, automated data interpretation features and adaptive outputs could prompt a user to act if health metrics deteriorate.Data visualisations that help users recognise data stability could support wellbeing during the waiting period to see a specialist or receive treatment.It is important to recognise that data outputs between phases of reflection and action could negatively impact a user's wellbeing leading to feelings of helplessness or fear.
There are many barriers to using patient-generated data (PGD) in clinical settings such as unfamiliarity, lack of standardisation and limited clinical acceptance due to a lack of scientific validation [70].Prior work by West et al. [70] note that the barriers to PGD being used in clinical settings cannot always be overcome through design but rather require changes in practice.Whilst many barriers to PGD being utilised still exist, participants in this study may have faced fewer obstacles due to changing practices which are specific to the clinical setting.Wearables produce standardised single lead ECGs which are increasingly recognised in cardiac clinical pathways.For instance the European Society of Cardiology (ESC) guidelines for the management of AF now state that wearable ECG recordings can be accepted as diagnostic for AF [29].However, a recent survey by the ESC revealed that 17% of cardiologists did not feel confident in diagnosing AF from a wearable-based ECG recording [39].This suggests a need to educate clinicians on clinical guidelines and the emerging role of PI tools [39].The survey also revealed that 29% of cardiologists would not prescribe anti-stroke medication (the first line of treatment for AF) based on wearable ECG recordings evidencing AF [39].This discrepancy illustrates that there is hesitation by some clinicians in prescribing therapeutics based on PI tools [39].Whilst clinical acceptance is high amongst cardiologists and highest amongst electrophysiologists, further research is needed to understand acceptance rates in other medical specialities [39].Initial work could focus on understanding acceptance rates amongst the gatekeepers to specialist care: such as emergency medicine and general practitioners.

Future Work and Limitations
This study was designed to document successful experiences of individuals using PI tools to support a cardiac diagnosis.This focus inherently biases our work towards positive outcomes, risking overstating the capabilities of PI technology.The prevalence of these experiences should not be interpreted as representative.The lead author was the sole coder for the thematic analysis.We acknowledge that this approach might introduce analytical bias.Subsequent themes were discussed and reflected upon with other authors to mitigate this risk.Additionally, as participants in the online ethnography component self-reported clinical diagnoses, there is a risk of self-selection and survivorship bias.Our interview demographic reflected western educated, industrialized, rich, and democratic populations, with the majority being male.Only five interview participants identified as female.For these reasons, our results cannot be generalised.Although we captured some indications of differences between cardiac healthcare experiences for men and women.Future research is required to understand gender-specific differences in challenges, motivations and tracking practices.
Increasing adoption of wearable technology alongside rising clinical acceptance means there is a pressing need to explore a full breadth of user experiences in the context of cardiac health data.Exploring both positive and negative experiences is essential to fully inform the future design of PI technology.Rosman et al. note the risk wearables pose in causing negative health behaviours such as obsessive symptom surveillance, which is an important avenue for future research [55].Future PI models should consider the impact of system-driven alerts, AI's role in data interpretation, and the use of derived metrics to interpret self-collected data.Further work is needed to understand how these factors influence the various stages of self-tracking.

CONCLUSION
"Without the watch, I would never have connected the dots that something dangerous was going on. . . it saved my life. . .shame it cant fix the NHS too." (OC209) We have provided a qualitative account of the emerging role of PI technology in successfully supporting a cardiac diagnosis.We have documented the motivations participants initially had to self-track cardiac data, such as bridging current gaps in clinical diagnostic modalities.We found that during the diagnostic journey, self-tracking practices were instigated and shaped by the complex interplay of self-collected data, bodily self-awareness, and increasing clinical acceptance.These insights reveal the consequences of the democratisation of complex health data.We highlight design challenges future health providers, consumer electronic companies, and third-party application designers face in further supporting the diagnosis of cardiovascular disease.
4.1.1Overlooked Data and The Lifesaving Electrocardiogram.James (P13) is in his early 60s and is based in the US.He wears the Oura ring and an Apple watch.He wears the ring at night to track his sleep and wears his watch during the day to keep track of his activities.He often reviews his sleep data each morning and activity scores at the end of each day.

Table 1 :
4.1.2Misleading Metrics and the Search for Answers.Mia (P9) is in her mid-30s and lives in the US.She uses a Whoop fitness tracker, predominantly to track her sleep.From her teenage years, Mia Interview Participant Information