When Ubiquitous Health Data Stops Helping: Patients' Experience of Breakdowns in the Data-Interpretation-Action Loop at the End of Life

1 Introduction

• At the end of life, a failed interaction may consume time, add burden, or reduce comfort without a later opportunity for repair.The paper frames this as temporal irreversibility: later repair cannot recover time and attention already spent.Source: §1 · p. 2
• The study uses the data–interpretation–action loop as an analytical abstraction and interviews 25 patients as primary informants and 8 clinicians as supplementary informants.Four participant-grounded data forms organize the accounts: state signals, trajectories, algorithmic outputs, and behavioral audits.Source: §1 · p. 2
• Participants described visible data that did not support usable interpretation, suggested actions that no longer fit, and difficulty exiting data exposure embedded in care.Action fit depended on bodily capacities, care priorities, and relationships involved in decision-making.Source: §1 · p. 2
• The paper derives design implications for interpretive stability, care-context alignment, and existential support rather than maximizing data visibility.Its design question is how systems can support interpretation, feasible action, or refusal within patients’ remaining time.Source: §1 · p. 2

2.1 Ubiquitous Computing for End-of-Life Care

• End-of-life conditions include rapidly changing bodies, a shift toward comfort, and shared decisions, giving health data existential weight that neighboring design assumptions may not transfer unchanged.Technologies distribute exposure across hospital rooms, homes, and mobile interfaces.Source: §2.1 · pp. 2–3

2.2 Health Data Exposure to Patients: Promise and Critique

• The data–interpretation–action loop is an analytical abstraction: patients encounter data, construct meaning, and decide whether or how to act.Sensemaking mobilizes signals, prior experience, medical knowledge, and social resources; a breakdown occurs when these do not stabilize a workable interpretation or connect it to feasible action.Source: §2.2 · p. 3
• Existing research describes burdensome or abandoned tracking and context adaptation, but gives less explanation of mixed environments where exposure can continue without patient-initiated inspection.Mortality-focused work foregrounds temporality and relationships, while continuous exposure’s repeated demands to interpret, act, or disengage remain underexamined.Source: §2.2 · pp. 3–4

2.3 Empirical Studies on Ubiquitous Computing for End-of-Life Patients

• Across outcome-oriented, design, and critical studies, patients are rarely examined as interpreters deciding what health data mean and whether to respond.Outcome metrics, workflow descriptions, and interface evaluations alone cannot establish these patient experiences.Source: §2.3 · p. 4
• The paper locates its contribution at the intersection of sensemaking, action, and disengagement from data embedded in care, where later repair cannot be presumed.It treats the loop as an analytical abstraction and the four data forms as participant-grounded organizing categories.Source: §2.3 · p. 4

3 Method

• The qualitative study interviewed 25 terminal-stage patients as primary informants and 8 clinicians as supplementary informants about data exposure in care.Patient first-person experience was the unit of analysis; clinician accounts contextualized clinical interpretation, infrastructure, and workflow.Source: §3 · p. 4, §3.1 · pp. 4–6

3.1 Participants

• The patient sample varied in age, education, illness, care arrangement, and prior experience with intelligent systems, and was recruited primarily through public hospitals.The study recruited through on-site and snowball sampling at public tertiary hospitals.Source: §3.1 · pp. 5–6

3.2 Data Collection

• Semi-structured interviews lasted 45–75 minutes and followed an experiential chain from systems and information through understanding and action to usage boundaries.Patient questions elicited changing strategies such as reduced checking, selective attention, avoidance, and re-engagement; clinician interviews focused on system use, interpretation, decisions, and relationships.Source: §3.2 · pp. 6–7

3.3 Ethics

• The study used ethics approval, consent and capacity checks, withdrawal options, and psychological support measures for participants.Interviewers were trained in palliative communication, participants’ conditions were monitored, and a counselor could provide support or terminate an interview.Source: §3.3 · p. 7

3.4 Data Analysis

• Three authors conducted reflexive thematic analysis, independently reading and coding the dataset before collaboratively clustering and reviewing themes.Patient transcripts were primary evidence for experiential themes, while clinician transcripts contextualized them; themes were traced to transcripts and examined through reflexive discussion.Source: §3.4 · pp. 7–8

4.1 Living with Data: Health Information Contexts in End-of-Life Care

• Patients described an everyday health-data environment organized into four forms: state signals, trajectories, algorithmic outputs, and behavioral audits.Twenty-two patients explicitly reported health data in their lives; these participant-grounded categories capture interpretive and practical work, not a universal taxonomy. Hospital bedside data were especially difficult to avoid, while apps and wearables allowed more partial disengagement; the small home-care group precludes a balanced setting comparison.Source: §4.1 · pp. 8, 10, 11

4.2 When Data Leave Questions Open: Repeated and Unresolved Sensemaking Across Data Types

• Across data types, patients repeatedly tried to infer meanings and causes, but often lacked the context needed for stable interpretations.State signals lacked medical and temporal context; trajectories prompted backward- and forward-looking guesses, sometimes extending to perceived remaining time; algorithmic outputs led patients to reconstruct hidden system logic. Fragmented memory and limited clinical explanation or oversight compounded uncertainty, while behavioral audits were less often associated with this sensemaking pattern.Source: §4.2 · pp. 12–13

4.3 When Action Cannot Follow: Misalignment in Context Management

• Patients described recommendations that assumed bodily capacities they had lost, sometimes making action impossible and reminders a recurring prompt of that loss.Some adaptive tools offered more suitable suggestions only after patients repeatedly restated their changed capacity, shifting context-management work onto them.Source: §4.3 · p. 14
• Patients described long-term health-management goals as misaligned with end-of-life priorities, which often centered on immediate effects and remaining time.Continuous monitoring could feel exhausting, and some data streams remained in place because clinicians considered them useful even when patients no longer did.Source: §4.3 · pp. 14–15
• Patients described data that still mattered but addressed them as decision-makers after decisions had shifted to family members or clinicians.These accounts reflect patients’ perspectives and do not establish how relatives understood or negotiated decision responsibility.Source: §4.3 · p. 15

4.4 The Failure of Disengagement

• Patients reduced engagement by closing apps, ignoring notifications, or avoiding monitors, but described disengagement as partial and unstable because data remained present.State signals were especially difficult to avoid when continuously displayed; avoiding them meant trying not to attend to data already in view, not simply choosing not to open an app.Source: §4.4 · p. 15
• Participants selectively ignored data that felt unactionable, but uncertainty and concern about missing important information drew them back; ambient bedside monitoring also made disengagement harder.Avoidance was easier when data required an extra step to access, but participants described returning to check trajectories or prompts. Repair could require repeating context and was not automatic or cost-free.Source: §4.4 · p. 16

4.5 Contextual Conditions Shaping the Breakdowns

• Breakdowns were shaped by exposure and device placement, social costs of seeking clarification, distributed decision-making, and limited clinical capacity; clinician interviews contextualized, but did not validate, patient experience.Clinician and patient accounts sometimes diverged in how they evaluated limited explanation and oversight, despite little factual disagreement between groups.Source: §4.5 · pp. 16–17

4.6 Expectations for Data Exposure: Less Data, More Support

• Participants preferred data to be less visible and frequent, surfacing selectively when information was serious, urgent, relevant, and actionable in their current condition.They questioned continuous displays and repetitive prompts, and wanted important conditions stated clearly rather than left for them to infer from data.Source: §4.6 · pp. 17, 18, 20
• Participants wanted exposure to recognize their inner lives and offer accompaniment or autonomous environmental adjustments, rather than merely track bodily decline or issue instructions.Examples included cues that make time feel lived, acknowledgment of fear or fatigue, and ambient responses such as gently adjusting room lighting.Source: §4.6 · pp. 18, 20

5.1 Rethinking Data Exposure in End-of-Life Ubiquitous Computing

• At end of life, unresolved interpretation, mismatched action prompts, and difficult disengagement combine within patients’ limited remaining time.The four data forms are comparative lenses, not a universal taxonomy; later repair may improve what happens next but cannot recover time and attention already spent.Source: §5.1 · p. 21

5.2 From Broader Patterns to Situated Manifestations

• The three breakdown patterns may transfer, but their frequency, intensity, and manifestations are situated in technological, relational, and institutional conditions.Because 21 of 25 patients were hospital-based, the hospital–home distinction is descriptive rather than a balanced comparison.Source: §5.2 · pp. 21–22

5.3 Designing for Interpretive Stability

• Interpretive stability means supporting a bounded, provisional understanding of what data indicate and whether they warrant attention, without repeatedly reopening unresolved questions.Proposed support includes explaining source and context, stating uncertainty, and routing unresolved questions through locally defined urgency criteria; algorithmic explanations must not imply unsupported causality or certainty.Source: §5.3 · p. 22
• Behavioral audits should record directly observable actions and remain visually distinct from bodily outcomes, without implying that an action caused a physiological change.Participants used medication, diet, or activity records to look back on whether an understood action occurred.Source: §5.3 · p. 23

5.4 Designing for Care-Context Alignment

• Care-context alignment requires action prompts to fit patients’ current capacities, care priorities, and decision-making roles.Proposed mechanisms include capacity-aware prompts, revisable priority profiles, and role-aware routing; these remain hypotheses, and caregivers’ preferred roles were not studied.Source: §5.4 · pp. 23–24

5.5 Designing for Ubiquitous Existential Support

• Existential support means supporting comfort, dignity, acknowledgment, and personhood when bodily optimization is not the sole or primary goal.Exposure should be negotiable and reversible, and patient-authorized low-demand responses may offer support without requiring another interpretive task; these are design hypotheses, not evaluated outcomes.Source: §5.5 · pp. 24–25

5.6 Limitations and Future Work

• The findings are bounded by a primarily Chinese tertiary-hospital sample, no matched cross-cultural comparison, heterogeneous systems, and retrospective interviews.Caregivers were not interviewed, and 21 of 25 patients were hospital-based; proposed design implications also require participatory, ethically constrained validation.Source: §5.6 · pp. 25–26
• Excluded patients with no relevant prior technology experience or severe hearing or speech impairments may have distinct refusal or accessibility experiences absent from these findings.Source: §5.6 · p. 25

6 Conclusion

• End-of-life health-data systems should support interpretive stability, care-context alignment, and existential support, with evaluation attending to feasible action, unwanted exposure, and patient support.Patients described repeated failed interpretation, mismatched suggested actions, and difficulty sustaining disengagement when data remained embedded in care.Source: §6 · p. 26

• Table 5 is identified as the semi-structured interview guide for patients.Source: Table 5 · p. 31
• The section distinguishes a patient guide from a clinician guide.Source: Table 5 · p. 31, Table 6 · p. 32
• Table 6 is identified as the semi-structured interview guide for clinicians.Source: Table 6 · p. 32

• The approved protocol limited recruitment to patients and clinicians to reduce family-privacy risks and participant burden.Source: B Rationale for Excluding Family… · p. 32
• The study centered patients’ first-person experiences of encountering and living with ubiquitous health data.Source: B Rationale for Excluding Family… · p. 32
• Without direct caregiver interviews, patient accounts cannot be cross-validated or support a comprehensive multi-actor analysis of data interpretation and decision authority.Future work should incorporate caregiver perspectives to examine these collaborative processes.Source: B Rationale for Excluding Family… · pp. 32–33

• Patients needed a physician-confirmed terminal illness with limited life expectancy and sufficient physical and cognitive capacity to participate.Source: C Inclusion and exclusion criteria · p. 33
• Patients also needed to understand the study and provide informed consent; impairments or conditions preventing safe, meaningful participation were exclusion criteria.Source: C Inclusion and exclusion criteria · p. 33
• Strong resistance to relevant technology excluded people without meaningful device or data experience, but not those who later reduced, stopped, avoided, or refused use.Source: C Inclusion and exclusion criteria · p. 33

• Questions moved from concrete, low-burden system-use topics toward interpretation and evaluation after participants became comfortable.Source: D Participant Vulnerability and… · p. 33
• The protocol avoided direct questions about problems or negative experiences, allowing these to emerge through concrete narrative questions.Source: D Participant Vulnerability and… · p. 33
• The protocol balanced eliciting rich experiential data with minimizing emotional costs for terminal-stage patients.Source: D Participant Vulnerability and… · p. 33