CoChat

Talk2Care: An LLM-based Voice Assistant for Communication between Healthcare Providers and Older Adults

Open access
Yang, Ziqi, Xu, Xuhai, Yao, Bingsheng, Rogers, Ethan, Zhang, Shao, Intille, Stephen, Shara, Nawar, Gao, Guodong Gordon, Wang, Dakuo · 2024 · Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies

The DOI matches its Crossref record; 135 of the 135 references it cites were found in a registry or an index; 33 of the 64 sources its claims cite were checked for support.

Abstract

Despite the plethora of telehealth applications to assist home-based older adults and healthcare providers, basic messaging and phone calls are still the most common communication methods, which suffer from limited availability, information loss, and process inefficiencies. One promising solution to facilitate patient-provider communication is to leverage large language models (LLMs) with their powerful natural conversation and summarization capability. However, there is a limited understanding of LLMs' role during the communication. We first conducted two interview studies with both older adults (N=10) and healthcare providers (N=9) to understand their needs and opportunities for LLMs in patient-provider asynchronous communication. Based on the insights, we built an LLM-powered communication system, Talk2Care, and designed interactive components for both groups: (1) For older adults, we leveraged the convenience and accessibility of voice assistants (VAs) and built an LLM-powered conversational interface for effective information collection. (2) For health providers, we built an LLM-based dashboard to summarize and present important health information based on older adults' conversations with the VA. We further conducted two user studies with older adults and providers to evaluate the usability of the system. The results showed that Talk2Care could facilitate the communication process, enrich the health information collected from older adults, and considerably save providers' efforts and time. We envision our work as an initial exploration of LLMs' capability in the intersection of healthcare and interpersonal communication.

Overview

CoChat Summary

## TL;DR

Talk2Care is a pilot LLM-powered voice assistant and provider dashboard designed to improve asynchronous communication between older adults at home and healthcare providers. Across interviews and user studies with 10 older adults and 9 providers, the system was reported to improve usability, enrich shared health information, and reduce providers’ effort and time.

## Key findings

  • In the need-finding interviews, older adults and providers both described communication breakdowns in existing methods: providers were hard to reach, messages were often unclear or incomplete, and older adults had accessibility and literacy challenges.
  • In the older-adult user study, the VA system had an average SUS score of 75.5±17.1 and participants rated it as simple to understand (4.6±0.5), accurate in understanding their words (4.3±0.8), reliable (4.2±1.0), and supportive for home-based healthcare needs (4.2±1.0).
  • In the provider user study, the dashboard had an average SUS score of 85.8±9.8; providers also reported low task load on NASA-TLX (mental demand 2.2±1.0, physical demand 1.4±0.5, temporal demand 1.7±0.9, effort 2.0±1.1, frustration 1.1±0.3, performance 6.1±0.6).
  • Providers rated the system helpful for communication with older adults (4.6±0.5 out of 5) and for offloading work (4.4±0.9), and 5 out of 9 said the dashboard could assist triage and prioritization.

## Methods

  • The paper used a mixed design: two semi-structured interview studies for need-finding, followed by two scenario-based user studies evaluating a pilot system.
  • Sample: 10 older adults (65+ years old, living in single-family homes in the northeastern U.S.) and 9 healthcare providers (including physicians, nurse practitioners, and related professionals).
  • Analysis and evaluation: interview transcripts were analyzed with thematic analysis; user studies used questionnaires including SUS and NASA-TLX, plus exit interviews and comparison against traditional asynchronous messaging/phone-call workflows.

## Limitations

  • The paper states that the user-study sample sizes were limited: 10 older adults and 9 healthcare providers.
  • The paper states that participants were recruited from a university pool and may have had higher education, digital literacy, or better understanding of LLMs than other groups, so results may not generalize to other populations.
  • The paper states that providers were mainly physicians rather than a wider range of roles; it suggests future work include broader patient and provider populations.
  • Evident from the text: the system is a proof-of-concept prototype and the evaluation focused mainly on usability and perceived benefits, not on module-level performance such as question quality, summary quality, or risk-prediction accuracy.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1145/3659625

Cites

135 of 135 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Image 1
CC-BY · From the PDF
Fig. 1 Overview of Talk2Care System. The system consists of two modules. 1) The patient module: An LLM-powered VA interface (in purple) that generates natural conversation with home-based older adults to collect health information and forward it to healthcare providers; 2) The provider module: A dashboard interface (in green) that summarizes the key information from the older patient conversation to assist providers who are responsible for communication ( e.g. , nurses and patient navigators). Note that Talk2Care does not provide specific healthcare advice. Our current implementation does not involve an actual electronic health record (EHR) system, which can be a promising future direction.
Fig. 1 Overview of Talk2Care System. The system consists of two modules. 1) The patient module: An LLM-powered VA interface (in purple) that generates natural conversation with home-based older adults to collect health information and forward it to healthcare providers; 2) The provider module: A dashboard interface (in green) that summarizes the key information from the older patient conversation to assist providers who are responsible for communication ( e.g. , nurses and patient navigators). Note that Talk2Care does not provide specific healthcare advice. Our current implementation does not involve an actual electronic health record (EHR) system, which can be a promising future direction.CC-BY · From the PDF
Fig. 2 The Component of Talk2Care System for Home-based Older Adults. The VA interface has multi-turn personalized conversations with the older adult to collect related health information. The LLM-powered Question Generation Module is responsible for taking the older adult's words and generating questions for effective information collection. The prompt design of this module is detailed in Figure 3. Another LLM-powered Content Loopback Module is to make sure that key information from the older adult (e.g., pain level) is accurate by double-checking the content, a common healthcare communication practice. The older adult's information, conversation protocol, and conversation log are stored in the information database.
Fig. 2 The Component of Talk2Care System for Home-based Older Adults. The VA interface has multi-turn personalized conversations with the older adult to collect related health information. The LLM-powered Question Generation Module is responsible for taking the older adult's words and generating questions for effective information collection. The prompt design of this module is detailed in Figure 3. Another LLM-powered Content Loopback Module is to make sure that key information from the older adult (e.g., pain level) is accurate by double-checking the content, a common healthcare communication practice. The older adult's information, conversation protocol, and conversation log are stored in the information database.CC-BY · From the PDF
Image 4
CC-BY · From the PDF
Fig. 4 The Component of Talk2Care System for Healthcare Providers. The information dashboard summarizes and highlights key older adults' information. The main content on the dashboard is generated by three LLM-powered modules: (1) The Content Summary Module formats the conversation log and user information into a clinical note structure. (2) The Information Highlight Module color-codes the parts in the conversation log that require attention. (3) The Risk Prediction Module suggests the health risk (low, moderate, and high) based on the current conversation log. Providers can take notes or further actions on the dashboard, which are then stored in the information database.
Fig. 4 The Component of Talk2Care System for Healthcare Providers. The information dashboard summarizes and highlights key older adults' information. The main content on the dashboard is generated by three LLM-powered modules: (1) The Content Summary Module formats the conversation log and user information into a clinical note structure. (2) The Information Highlight Module color-codes the parts in the conversation log that require attention. (3) The Risk Prediction Module suggests the health risk (low, moderate, and high) based on the current conversation log. Providers can take notes or further actions on the dashboard, which are then stored in the information database.CC-BY · From the PDF
Image 6
CC-BY · From the PDF
Fig. 6 LLM-powered Dashboard Interface Layout for Healthcare Providers. The dashboard presents the conversation summary (center), raw log, and its highlights (right). The colored dot represents the risk prediction (left and center). Providers can take specific follow-up actions or write down notes under each conversation summary, after which they can mark one session as done.
Fig. 6 LLM-powered Dashboard Interface Layout for Healthcare Providers. The dashboard presents the conversation summary (center), raw log, and its highlights (right). The colored dot represents the risk prediction (left and center). Providers can take specific follow-up actions or write down notes under each conversation summary, after which they can mark one session as done.CC-BY · From the PDF
Fig. 7 User Study 3 Scenario with an Older Adult Participant.
Fig. 7 User Study 3 Scenario with an Older Adult Participant.CC-BY · From the PDF
Fig. 8 Subjective Ratings on Different Perspectives of The Usability of The VA System. Error bars show the standard deviation. The average SUS score is 75.5 ± 17.1, indicating good usability.
Fig. 8 Subjective Ratings on Different Perspectives of The Usability of The VA System. Error bars show the standard deviation. The average SUS score is 75.5 ± 17.1, indicating good usability.CC-BY · From the PDF
Fig. 9 Dashboard Interface Example for Scenario B: Post-Surgery Follow-up. The example for Scenario A can be found in Figure 6, with the same interface and different scenario-specific content.
Fig. 9 Dashboard Interface Example for Scenario B: Post-Surgery Follow-up. The example for Scenario A can be found in Figure 6, with the same interface and different scenario-specific content.CC-BY · From the PDF
Fig. 10 Subjective Ratings on Different Perspectives of The Usability of The Dashboard System Design. The average SUS score is 85.8 ± 9.8, indicating very good usability.
Fig. 10 Subjective Ratings on Different Perspectives of The Usability of The Dashboard System Design. The average SUS score is 85.8 ± 9.8, indicating very good usability.CC-BY · From the PDF

Verification

Claims the paper makes

6 Valid · 16 Not valid · 7 Not checked · 163 Own results and interpretations

72 more claims · 1 with their cited sources checked

Create a free account to see all 192 claims and every source verdict.

See all claims — free account
  • It may state two separate claims as one. 55
  • Probably not a substantial claim. 79

Create a free account to see these claims.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Citations

Cites

Show all 135 references

Research works printed but not linked: 2

  1. Farah Magrabi, Elske Ammenwerth, Jytte Brender McNair, Nicolet F De Keizer, Hannele Hyppönen, Pirkko Nykänen, Michael Rigby, Philip J Scott, Tuulikki Vehko, Zoie Shui-Yee Wong, et al. 2019. Artificial intelligence in clinical decision support: challenges for evaluating AI and practical implications. Yearbook of medical informatics 28, 01 (2019), 128--134.
  2. Brendan Spillane, Emer Gilmartin, Christian Saam, Benjamin R Cowan, and Vincent Wade. 2018. ADELE: Care and Companionship for Independent Aging.. In ICAHGCA@ AAMAS. 18--24.

Cited by

Items 101–102 of 102

Metadata

Authors
Yang, Z., Xu, X., Yao, B., Rogers, E., Zhang, S., Intille, S., Shara, N., Gao, G. G., & Wang, D.
Affiliations
  • University of Michigan (US)
  • Massachusetts Institute of Technology (US)
  • Rensselaer Polytechnic Institute (US)
  • Northeastern University (US)
  • MedStar Health Research Institute (US)
  • Johns Hopkins University (US)
Venue
Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
Published
2024-05-13
Volume
8
Issue
2
Pages
1-35
ISSN
2474-9567 (ISSN-L)
OpenAlex
W4396919238
Cited by
120
References
89
Citations by year
2026: 50 · 2025: 64 · 2024: 6
FWCI
102.62
Type
journal article
Language
en
Open access
Locations
  • Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies (publishedVersion, cc-by)
  • DSpace@MIT (Massachusetts Institute of Technology) (submittedVersion, cc-by-nc)
Topics
  • Health Sciences › Medicine › Geriatrics and Gerontology › Technology Use by Older Adults
  • Health Sciences › Medicine › Public Health, Environmental and Occupational Health › Telemedicine and Telehealth Implementation
  • Health Sciences › Health Professions › Health Information Management › Innovation in Digital Healthcare Systems
Keywords
large language models, voice assistants, patient-provider communication, older adults, asynchronous communication, telemedicine, conversational interfaces, usability evaluation, user needs

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-09-29, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-09-29 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.