PaperlayerView exact revision on arXiv

Source-linked AI summary

CareMirror: Bringing Caregiver Wellbeing into the Dementia Care Ecosystem

Jiayue Melissa Shi, Ethan Nguyen, Drishti Goel, Upasana Natarajan, Shashwat Srivatsa, Daniel S. Brown, Violeta J. Rodríguez, Dong Whi Yoo, Ravi Karkar, Koustuv Saha

arXiv:2609.17434v1cs.HCcs.AIcs.CLcs.CY

TL;DR

Family caregivers provide much of the day-to-day care for people living with dementia, and CareMirror explores how their wellbeing support might connect with dementia care. Interviews with 14 caregivers found value in recognizing wellbeing changes and accessing timely, personalized support, alongside tensions around clinical connection. The study offers design considerations within the scope of its small, nonrepresentative sample.

  • Problem

    Family caregivers provide much of the day-to-day care for people living with dementia.

    Source: §1 · p. 2

  • Method

    The authors used CareMirror as a design probe and conducted semi-structured interviews with 14 caregivers.

    Source: §7 · p. 21, §8 · p. 25

  • Results

    Caregivers valued recognizing wellbeing changes and accessing timely, personalized support, while clinical connection raised tensions around privacy, disclosure, emotional burden, and agency.

    Source: §8 · p. 25

  • Takeaways & Limitations

    The findings suggest that making caregiver wellbeing visible should not be an endpoint.

    Source: §8 · p. 25

    The caregiver participants were not representative of the broader dementia caregiver population, particularly given the relatively small sample.

    Source: §7.4 · p. 24

Abstract

from arXiv · show

Family caregivers of people living with dementia shoulder emotional and practical responsibilities, yet their own wellbeing often remains peripheral to dementia care. We built CareMirror, an envisioned caregiver wellbeing ecosystem with interconnected caregiver- and clinician-facing interfaces for longitudinal reflection, personalized support, and caregiver-controlled sharing with clinical care. We conducted semi-structured interviews with 14 caregivers, using CareMirror as a design probe to examine how they perceived this ecosystem and what expectations, concerns, and boundaries emerged around clinical connection. Caregivers valued attention to their wellbeing, longitudinal awareness, context-sensitive support, and clinical visibility when it could lead to meaningful follow-up. However, repeated reflection could become burdensome or emotionally difficult, automatic clinical sharing could inhibit candid disclosure, and participants wanted control over what information entered clinical care. They also expected AI to support reflection and communication without replacing caregiver voice or clinician judgment. We contribute design considerations for proactive, clinically connected caregiver wellbeing support.

1 Introduction

Caregiver wellbeing can remain peripheral to dementia care, while changing needs and limited clinician insight between visits leave caregivers to notice changes and find support. CareMirror probes a connected approach, and interviews surfaced both valued support and tensions around burden, sharing, and AI.

Motivation

  • Caregivers’ wellbeing may remain peripheral to dementia care, while discrete support places the work of noticing changes and finding help on caregivers whose time is constrained.Clinicians may have limited insight into caregivers’ wellbeing between encounters.Source: §1 · p. 2

CareMirror and study

  • CareMirror links caregiver and clinician interfaces for wellbeing assessment, reflection, personalized support, and potential communication with care teams.The system was used as a design probe rather than presented as a finalized model for caregiver support.Source: §1 · p. 2
  • In a semi-structured interview study, 14 caregivers discussed caregiving experiences and wellbeing needs, explored CareMirror’s proposed wellbeing features, and considered how caregiver wellbeing support might connect with clinical care.Source: §1 · p. 3

Findings

  • Caregivers valued a dedicated space for wellbeing, longitudinal awareness, and support tailored to their current caregiving circumstances.They also valued clinical visibility when it could lead to meaningful follow-up, while rejecting automatic sharing and seeking control over what entered clinical care.Source: §1 · p. 3
  • Caregivers said repeated wellbeing activities could add demands, and revisiting difficult experiences was not always beneficial.They also expressed concern about AI interpreting or communicating their experiences without sufficient human oversight.Source: §1 · p. 3

Contributions

  • The paper contributes design considerations for longitudinal caregiver wellbeing support that preserves sharing control and positions technology as supporting, not replacing, caregiver expression and clinician judgment.The findings highlight tensions among caregiver reflection, clinical visibility, caregiver agency, and professional involvement.Source: §1 · p. 3

2 Related Work

The related work frames dementia caregiving as a changing experience that can strain caregivers’ wellbeing and capacity to seek support. It reviews tracking, digital support, and collaborative care while identifying a gap in integrating caregivers’ own wellbeing into connected care.

2.1 Dementia: Condition and Caregiving

  • Dementia caregiving brings changing emotional and practical demands, while caregivers may have limited time and capacity for self-care or formal support.Caregiver needs and their intensity change alongside dementia progression.Source: §2.1 · pp. 3–4

2.2 Digital Technologies for Caregiver Wellbeing Tracking and Reflection

  • Tracking and digital support can help caregivers attend to wellbeing, but existing approaches include self-monitoring, resources, social support, and AI with contextual-fit and safety concerns.Reviews report varied intervention effects and sustained-engagement challenges; this work motivates AI as one component of a broader ecosystem rather than a replacement for human support.Source: §2.2 · pp. 4–5

2.3 Caregiver–Clinician Communication and Collaborative Care

  • Collaborative care technologies support information exchange, yet connected care remains organized mainly around the care recipient, with less attention to caregivers’ own wellbeing.The cited work motivates a caregiver wellbeing ecosystem spanning everyday and clinical contexts.Source: §2.3 · p. 5

3 Design Goals

CareMirror’s design goals draw on prior work and clinician coauthor discussions to frame caregiver wellbeing as part of a dementia care ecosystem. The first goal opens a brief, voluntary caregiver wellbeing pathway during care encounters, with attention to follow-up responsibility.

3 Design Goals

  • CareMirror’s design goals were informed by prior work and refined through clinician coauthor discussions about assessment, useful information, and feasible follow-up.The goals frame an infrastructure for caregiver wellbeing within the dementia care ecosystem.Source: §3 · p. 6

3.1 DG1: Opening a Caregiver Wellbeing Pathway within the Dementia Care Ecosystem

  • DG1 proposes a brief, voluntary way for caregivers to reflect on their wellbeing during a dementia care encounter.Validated measures can surface caregiver needs, but entering clinical care creates an expectation that concerning results receive review and, when appropriate, support.Source: §3.1 · p. 6

3.2 DG2: Following Caregiver Wellbeing Over Time

CareMirror’s goals span longitudinal awareness, timely personalized support, and clinician awareness with caregiver-controlled sharing. The proposed ecosystem aims to reflect changing needs while leaving interpretation and sharing decisions with caregivers and clinicians.

3.2 DG2: Following Caregiver Wellbeing Over Time

  • DG2 proposes repeated check-ins and review of prior responses so caregivers can notice wellbeing changes between clinical encounters.Check-in timing and frequency can be adjusted to caregiver preferences and recent responses, and the system surfaces patterns from caregiver-provided information rather than automatically inferring mental health.Source: §3.2 · pp. 6–7

3.3 DG3: Connecting Caregivers with Timely and Personalized Support

  • DG3 proposes support matched to caregivers’ current needs and circumstances, with relevant resources available when needs arise.Options include self-care strategies, dementia-care information, local groups, respite, and professional support; the chatbot is not intended to provide clinical care or replace qualified professionals.Source: §3.3 · p. 7

3.4 DG4: Supporting Clinician Awareness and Caregiver Follow-Up

  • DG4 proposes helping clinicians understand caregiver-reported needs and determine when follow-up may be appropriate, while caregiver control over sharing remains central.The dashboard informs rather than automates clinical judgment; caregivers may share wellbeing information without sharing every response or conversation.Source: §3.4 · pp. 7–8

4 Designing CareMirror

CareMirror links caregiver wellbeing experiences to clinical attention through connected caregiver- and clinician-facing interfaces, while retaining boundaries around privacy and data access. Its components support longitudinal reflection, clinician review, and asynchronous awareness, but are not designed for real-time response.

4.1 Design Process and Clinical Context

  • CareMirror connects caregivers’ everyday wellbeing with clinical attention while preserving boundaries around privacy and data access.The ecosystem pairs caregiver- and clinician-facing portals to make wellbeing experiences more visible and actionable.Source: §4 · p. 8, §4.1 · p. 8

4.2 Caregiver-Facing Portal

  • The caregiver portal combines wellbeing assessments, mood check-ins, longitudinal tracking, AI-supported reflection, and access to clinical, community, and local resources.Assessments can be saved, skipped, or revisited; shorter recurring measures are used to support tracking while reducing burden.Source: §4.2 · p. 9, Figure 2 · p. 9

4.3 Clinician-Facing Portal

  • The clinician portal offers population- and individual-level views of caregiver wellbeing, potential concerns, AI-generated insights, and follow-up options.Indicators are intended to direct attention to patterns for further review, rather than independently characterize clinical or safety risk.Source: Figure 3 · p. 10, §4.3 · p. 10

4.4 System Architecture and Data Flow

  • Caregiver-generated data pass through a backend for storage and LLM-based analysis, with summaries and potential-concern indicators surfaced for asynchronous clinician review.These indicators are not real-time alerts and do not replace clinician review of caregiver-reported information.Source: §4.4 · p. 10, Figure 4 · p. 11
  • Deployment would require protocols for reviewing concerning information, follow-up and escalation, and communicating expectations to caregivers; these workflows were not implemented or evaluated.CareMirror is designed for asynchronous awareness, not real-time monitoring or an implied immediate clinician response.Source: §4.4 · p. 11

5 User Study

The study used CareMirror as a design probe in semi-structured interviews to examine caregivers’ experiences, support needs, and expectations for clinical connection. Surveys and iterative thematic analysis complemented the interviews.

5 User Study

  • Semi-structured interviews used CareMirror as a design probe to elicit caregivers’ needs, expectations, concerns, and boundaries around proactive wellbeing support and clinical connection.The study explored perceptions of the proposed ecosystem rather than evaluating task completion.Source: §5 · p. 11, §5.2 · p. 13

5.1 Participant Recruitment

  • The team interviewed 14 caregivers: 11 family caregivers and 3 professional caregivers, recruited primarily through dementia- and caregiving-related online communities.Participants were required to be adults, current or former caregivers, and U.S. residents.Source: §5.1 · pp. 11–12

5.2 Interview Procedure

  • Remote interviews included guided interaction with CareMirror and discussion of caregiving experiences, wellbeing needs, clinical sharing, privacy, and AI interpretations.Participants were encouraged to think aloud and describe how features might fit their routines.Source: §5.2 · pp. 12–13

5.3 Data Analysis

  • The analysis combined descriptive statistics from entry and exit surveys with qualitative analysis of interview transcripts and notes.All 14 participants completed the surveys; anonymized transcripts were analyzed alongside interview notes.Source: §5.3.1 · p. 13, §5.3.2 · p. 13

5.3.1 Descriptive Analysis.

  • Entry surveys covered the R-CWBS and mental wellbeing concerns, while exit surveys assessed usability and intervention appropriateness.The reported entry responses represented a range of caregiving-related wellbeing challenges and responsibilities.Source: §5.3.1 · p. 13

5.3.2 Qualitative Analysis.

  • Reflexive thematic analysis iteratively refined 467 open codes into four high-level themes about caregivers’ expectations, concerns, and boundaries.Two authors led initial coding, with broader team review and feedback during theme development.Source: §5.3.2 · pp. 13–14

5.4 Privacy, Ethics, and Reflexivity

  • Participants could decline questions, pause, or stop interviews, and the team treated sharing boundaries as central to the inquiry.Identifying information was removed from transcripts and study data were stored in secure, access-controlled environments.Source: §5.4 · p. 14

6 Findings

Participants welcomed CareMirror’s potential to extend caregiver wellbeing support through longitudinal tracking, AI-supported reflection, and clinical connection. They also raised concerns about privacy, sharing, and access, emphasizing the need for caregiver agency and control.

6 Findings

  • Participants valued longitudinal tracking, AI-supported reflection, and clinical connection when these could help recognize wellbeing changes and reduce the burden of communicating them to clinicians.They welcomed extending support beyond one-time wellbeing assessments.Source: §6 · p. 14
  • Participants raised concerns about privacy, data sharing, and AI and clinician access, making caregiver agency and control central to clinical integration.Technical integration alone was not presented as sufficient.Source: §6 · p. 14
  • The exit-survey figure displays SUS and IAM score distributions, with dotted lines marking the median of each distribution.The supplied figure description does not report score values or comparisons.Source: Figure 5 · p. 15

6.1 RQ1: Caregivers’ Perceptions about CareMirror

This section examines caregivers’ perceptions of CareMirror’s wellbeing support, longitudinal reflection, and clinical connection. Caregivers valued contextualized support and clinical visibility that enables meaningful follow-up, while noting potential burdens of repeated reflection and tensions between clinician awareness and candid disclosure when information may be shared automatically.

6.1.1 Perceived Usability and Appropriateness.

  • 81.61 (mean SUS score; median=80.0) indicated above-average, acceptable usability, while 17.43 (mean IAM score; median=18.0) indicated high appropriateness within this limited participant pool.CareMirror was explored as a prototype, and limited interface fidelity may have influenced usability ratings.Source: §6.1.1 · p. 15

6.1.2 Clinical Encounters Could Provide an Entry Point for Caregiver Support.

  • Caregivers valued explicit attention to their emotional and mental wellbeing, which could bring needs often overlooked in dementia care into view.Participants described the assessment’s range of wellbeing domains as helpful and saw caregiver support as extending beyond training in dementia care.Source: §6.1.2 · pp. 15–16
  • Repeated wellbeing activities could feel like another task, so participants favored brief, voluntary reflection and configurable check-in timing rather than assumed frequency.Participants also cautioned that repeated monitoring might focus attention on distress; the study did not evaluate this effect through longitudinal use.Source: §6.1.2 · p. 16
  • Repeatedly directing attention toward distress could feel burdensome or heighten symptom-focused attention, although the study did not test this through longitudinal use.Participants’ concerns suggest that check-in frequency and timing should not be assumed to benefit everyone.Source: §6.1.2 · p. 16

6.1.3 Longitudinal Awareness Was Valuable but Could Also Reinforce Distress.

  • Because caregiver wellbeing changes with caregiving demands, participants valued longitudinal views that contextualized check-ins and let them revisit earlier experiences.Returning to past entries, including notes from a better day, could provide reassurance.Source: §6.1.3 · p. 16
  • Resurfacing difficult caregiving records could itself be distressing, depending on the record’s content and the caregiver’s current emotional state.Participants saw a need to preserve reflection’s value while remaining sensitive to when and how distressing past experiences return to view.Source: §6.1.3 · pp. 16–17

6.1.4 Timely Support Needed to Reflect Each Caregiver’s Context.

  • Participants valued immediately available AI guidance, but its usefulness depended on reflecting individual caregiving circumstances and approaching the quality of human support.Some found chatbot responses helpful yet general or less good than a human response, while conversational interaction could feel more personal.Source: §6.1.4 · p. 17
  • Participants saw resource recommendations as useful when they connected reported wellbeing or emotional experiences with support relevant to what caregivers were experiencing.They likewise valued AI resources as a way to connect reported wellbeing with relevant support.Source: §6.1.4 · p. 17
  • Participants’ responses highlighted contextual relevance as an important dimension of timely, personalized support.This relates to the personalized support envisioned in DG3.Source: §6.1.4 · p. 17

6.1.5 Clinical Visibility Was Valuable When Paired with Meaningful Follow-Up.

  • Caregivers valued clinician visibility when it provided context for their experiences and helped clinicians interpret broad wellbeing labels.Notes accompanying mood reports could clarify what labels such as “good” or “low” meant to an individual caregiver.Source: §6.1.5 · pp. 17–18
  • Automatic clinical reporting could make caregivers less willing to disclose honestly, as greater clinician awareness could make the reflection space feel less comfortable.Participants described a tension between making wellbeing visible in clinical care and preserving a space for candid reflection.Source: §6.1.5 · p. 18
  • Caregivers valued clinical visibility when it could prompt appropriate follow-up, while emphasizing that clinician attention should fit their immediate needs.Participants sometimes welcomed a clinician check-in but also wanted to be left alone at times.Source: §6.1.5 · p. 18

6.2 RQ2: Envisioning a Clinically-connected Caregiver Wellbeing Ecosystem

Caregivers envisioned wellbeing as part of dementia care, while expecting clinical connection to preserve their agency and support meaningful follow-up. Their expectations centered on caregiver-focused care, responsive visibility, negotiated sharing, and AI that supports rather than mediates human care.

6.2.1 Caregiver Wellbeing Belongs within Dementia Care, but Should Remain Caregiver-Centered.

  • Caregiver wellbeing warrants attention for caregivers’ own sake, not only as a means of improving care for the person living with dementia.Participants described caregivers as both partners in dementia care and individuals whose wellbeing is itself an object of care.Source: §6.2.1 · pp. 18–19

6.2.2 Making Caregiver Wellbeing Visible Creates a Responsibility to Respond.

  • Wellbeing visibility needed interpretable indicators and longitudinal patterns, with concerning changes linked to resources or clinical follow-up.Participants wanted clearer communication of wellbeing states and ways to compare how they were doing across days.Source: §6.2.2 · p. 19

6.2.3 Connecting Caregivers and Clinicians Requires Negotiated Data Boundaries.

  • Clinical sharing should be negotiated, since automatic reporting could inhibit candid disclosure and caregivers wanted control over what information clinicians received.Participants envisioned permission-based sharing and boundaries around information, rather than clinical access as the default consequence of reflection.Source: §6.2.3 · pp. 19–20

6.2.4 Technology Should Support Rather than Mediate the Caregiver–Clinician Relationship.

  • AI could help communicate caregiver experiences, but participants expected caregiver expression and clinician judgment to remain central.Concerns about hallucinated or inaccurate information entering clinical records reinforced the importance of human involvement.Source: §6.2.4 · p. 20

7 Discussion

Across interviews with 14 caregivers, participants valued a dedicated space for wellbeing, longitudinal awareness, and support tailored to caregiving circumstances. The discussion frames clinically connected support as requiring caregiver-centered recognition, careful disclosure boundaries, and realistic pathways to human response.

7.1.1 From Care Partners to Recipients of Care.

  • Caregivers should be recognized not only as care partners but also as people whose wellbeing warrants support in its own right.The discussion extends caregiver wellbeing beyond its possible effects on care recipients.Source: §7.1.1 · p. 21

7.1.2 From Tracking to Care: Making Wellbeing Visible Creates a Responsibility to Respond.

  • Longitudinal wellbeing data can help recognize and interpret changes, but revisiting difficult periods carries emotional weight and should connect to appropriate support.The proposed progression is from visibility, to interpretation, to response.Source: §7.1.2 · p. 22

7.2.1 Clinical Connection Should Be Negotiated, Not Assumed.

  • Clinical connection should be negotiated because anticipated clinical visibility may change what caregivers disclose, and sharing preferences can evolve over time.Participants envisioned permission-based sharing and selective disclosure rather than exposing the entirety of a private reflective space.Source: §7.2.1 · p. 22

7.2.2 AI Should Support, Rather than Mediate, the Care Relationship.

  • AI summaries may ease synthesis and communication, but they represent caregivers’ experiences and should not displace caregiver expression or clinician judgment.Caregivers should be able to inspect and contextualize such representations, which should remain subject to clinician judgment.Source: §7.2.2 · pp. 22–23

7.2.3 Clinical Visibility Does Not Create Clinical Capacity.

  • Greater clinical visibility does not itself provide capacity for timely human support, and new information streams may create responsibilities without resources.The study leaves operational questions about accountability and follow-up unresolved.Source: §7.2.3 · p. 23
  • Implementing clinically connected support requires stakeholder input and further work on workflows, costs, funding, and equitable access.The discussion notes that pay-per-use support could disadvantage caregivers with fewer financial resources, while targeting only substantial distress could depart from a proactive model.Source: §7.2.3 · p. 23

7.3 Design Implications for Caregiver Wellbeing Ecosystems

The design implications emphasize reducing the work of receiving support, treating longitudinal reflection as emotionally consequential, preserving caregiver control over sharing, and keeping AI subordinate to caregiver and clinician judgment.

7.3 Design Implications for Caregiver Wellbeing Ecosystems

  • Wellbeing support should reduce rather than add caregiver work through brief, skippable interactions, adaptable check-in frequency, and support responsive to current needs.Proactive systems should help reduce the work of seeking support rather than turn self-care into another responsibility.Source: §7.3.1 · p. 23
  • Treat wellbeing histories as emotionally consequential: give caregivers control over when they revisit them and connect concerning patterns with resources or follow-up.The intended progression is from reflection, to interpretation, to response.Source: §7.3.2 · pp. 23–24
  • Keep reflection private by default and let caregivers choose what to share, with whom, and under what circumstances, revising preferences as needs change.Selected summaries, trends, or concerns can inform clinical care without making the full reflective space visible.Source: §7.3.3 · p. 24
  • Use AI to synthesize and communicate wellbeing information while allowing caregivers to inspect and correct representations and keeping clinician judgment central.Clinicians should be able to distinguish caregiver-reported information from AI-generated summaries or inferences.Source: §7.3.4 · p. 24

7.4 Limitations and Future Directions

The study reports limits in sample representativeness and clinical perspective, and identifies longitudinal, diverse, and workflow-focused research as future directions.

7.4 Limitations and Future Directions

  • The relatively small sample recruited through online communities is not representative of the broader dementia caregiver population, and the findings are not intended to be generalizable.The study instead sought to understand caregiver expectations and design preferences.Source: §7.4 · p. 24
  • Because clinicians were not study participants, the findings capture caregivers’ anticipated perspectives rather than clinicians’ needs, concerns, or workflow constraints.The envisioned ecosystem depends on clinicians interpreting caregiver wellbeing information and deciding when and how to respond.Source: §7.4 · p. 24
  • Future work should examine evolving caregiver engagement and preferences through longitudinal deployments and recruit more diverse backgrounds and care contexts.Further research can investigate how caregiving stage, healthcare access, and digital literacy shape expectations for proactive, clinically connected support.Source: §7.4 · p. 25

8 Conclusion

CareMirror explores how caregiver wellbeing support might fit into the broader dementia care ecosystem, based on semi-structured interviews with 14 caregivers. Caregivers valued longitudinal, personalized support, but wanted clinical visibility paired with meaningful response and control over what is shared.

8 Conclusion

  • CareMirror is an envisioned proactive wellbeing ecosystem exploring integration of caregiver support into the broader dementia care ecosystem.Source: §8 · p. 25
  • Caregivers valued technologies that help them recognize and make sense of wellbeing changes over time and access timely, personalized support.Source: §8 · p. 25
  • Connecting caregiver wellbeing with clinical care raised tensions around privacy, disclosure, emotional burden, and caregiver agency.Source: §8 · p. 25
  • Wellbeing visibility should lead to meaningful interpretation and pathways to response, while caregivers control what enters clinical contexts and when.Source: §8 · p. 25
  • The work highlights an opportunity for caregiver–clinician ecosystems to recognize caregivers as individuals with wellbeing needs of their own.Source: §8 · p. 25
Loading 2609.17434v1…