Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Exploring AI-Augmented Sensemaking of Patient-Generated Health Data: A Mixed-Method Study with Healthcare Professionals in Cardiac Risk Reduction

Pavithren V S Pakianathan, Rania Islambouli, Diogo Branco, Albrecht Schmidt, Tiago Guerreiro, Jan David Smeddinck

TL;DR

PGHD are increasingly collected but challenging to interpret in clinical care due to volume and heterogeneity. This paper investigates how LLM-generated summaries and a conversational interface can support HCP sensemaking of PGHD for cardiovascular risk reduction. Using a mixed-methods within-subjects design with 16 HCPs and synthetic one-year PGHD, the authors evaluate a prototype dashboard. Key contributions include empirical insights on benefits and risks, and a set of sociotechnical design implications to improve provenance, personalization, transparency, trust, and data governance, informing the responsible integration of AI into clinical workflows to reduce workload and enhance patient-centered care.

Abstract

Individuals are increasingly generating substantial personal health and lifestyle data, e.g. through wearables and smartphones. While such data could transform preventative care, its integration into clinical practice is hindered by its scale, heterogeneity and the time pressure and data literacy of healthcare professionals (HCPs). We explore how large language models (LLMs) can support sensemaking of patient-generated health data (PGHD) with automated summaries and natural language data exploration. Using cardiovascular disease (CVD) risk reduction as a use case, 16 HCPs reviewed multimodal PGHD in a mixed-methods study with a prototype that integrated common charts, LLM-generated summaries, and a conversational interface. Findings show that AI summaries provided quick overviews that anchored exploration, while conversational interaction supported flexible analysis and bridged data-literacy gaps. However, HCPs raised concerns about transparency, privacy, and overreliance. We contribute empirical insights and sociotechnical design implications for integrating AI-driven summarization and conversation into clinical workflows to support PGHD sensemaking.

Exploring AI-Augmented Sensemaking of Patient-Generated Health Data: A Mixed-Method Study with Healthcare Professionals in Cardiac Risk Reduction

TL;DR

PGHD are increasingly collected but challenging to interpret in clinical care due to volume and heterogeneity. This paper investigates how LLM-generated summaries and a conversational interface can support HCP sensemaking of PGHD for cardiovascular risk reduction. Using a mixed-methods within-subjects design with 16 HCPs and synthetic one-year PGHD, the authors evaluate a prototype dashboard. Key contributions include empirical insights on benefits and risks, and a set of sociotechnical design implications to improve provenance, personalization, transparency, trust, and data governance, informing the responsible integration of AI into clinical workflows to reduce workload and enhance patient-centered care.

Abstract

Individuals are increasingly generating substantial personal health and lifestyle data, e.g. through wearables and smartphones. While such data could transform preventative care, its integration into clinical practice is hindered by its scale, heterogeneity and the time pressure and data literacy of healthcare professionals (HCPs). We explore how large language models (LLMs) can support sensemaking of patient-generated health data (PGHD) with automated summaries and natural language data exploration. Using cardiovascular disease (CVD) risk reduction as a use case, 16 HCPs reviewed multimodal PGHD in a mixed-methods study with a prototype that integrated common charts, LLM-generated summaries, and a conversational interface. Findings show that AI summaries provided quick overviews that anchored exploration, while conversational interaction supported flexible analysis and bridged data-literacy gaps. However, HCPs raised concerns about transparency, privacy, and overreliance. We contribute empirical insights and sociotechnical design implications for integrating AI-driven summarization and conversation into clinical workflows to support PGHD sensemaking.
Paper Structure (38 sections, 5 figures, 5 tables)

This paper contains 38 sections, 5 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Challenges of Multimodal, Longitudinal PGHD in Clinical Care
  4. Sensemaking of Patient Generated Health Data
  5. AI Augmented (Health) Data Sensemaking
  6. Research Gap and Contribution
  7. Method
  8. Study Apparatus
  9. Synthetic Data Generation
  10. Procedure
  11. Participants
  12. Ethics
  13. Measures and Analysis Techniques
  14. Findings
  15. Value of AI Summaries for Sensemaking: Benefits, Evolving Practices, and Emerging Needs
  16. ...and 23 more sections

Figures (5)

  • Figure 1: A snippet of the interface in demo mode showing an overview of the persona, charts of physical activity minutes and corresponding LLM summaries, holistic insights of physical activity minutes, sedentary time, blood pressure and sleep, and charts of sedentary time over a one-week period. Summaries for sedentary time, and other information about blood pressure and sleep and draft physical activity plan are not visible. (Persona name is blinded for review)
  • Figure 2: Expanded view of the conversational interface with response for a query asking to compare sedentary time with physical activity minutes using a graph.
  • Figure 3: Participant interacting with the interface at their workplace.
  • Figure 4: Each session with a HCP was divided into 4 key sections. 1) Informed consent, Demographic Questionnaire, Interface Familiarization 2) Experimental evaluation with AI summary and No AI summary conditions with three randomized persona risk levels and 6 personas in total. 3) Interaction with conversational interface to perform data exploration with natural language. 4) Closing semi-structured interview
  • Figure 5: Charts showing the pairwise comparison of NASA-TLX workload ratings for AI and NO-AI summary conditions (n=16). (A) Barchart of Aggregate NASA-TLX Comparison (B) Spider Chart of Subscale NASA-TLX Comparison