More than Decision Support: Exploring Patients' Longitudinal Usage of Large Language Models in Real-World Healthcare Settings

Paper Title

More than Decision Support: Exploring Patients' Longitudinal Usage of Large Language Models in Real-World Healthcare-Seeking Journeys

Publication Info

• Topic area: Longitudinal integration of large language models (LLMs) in healthcare-seeking journeys.
• Keywords: Large language models, healthcare-seeking, patient-provider relationship, longitudinal study, emotional support, cognitive scaffolding, decision-making, socio-technical systems, trust calibration, agency.

Background and Problem

• Problem / challenge: Existing studies on LLMs in healthcare focus on specific tasks (e.g., diagnosis, information retrieval) or short-term interactions, neglecting the longitudinal, dynamic nature of real-world healthcare-seeking journeys.
• Significance: Understanding how LLMs can support patients across multi-stage healthcare trajectories is critical for improving patient experiences, addressing resource constraints, and reshaping traditional patient-provider dynamics.
• Motivation and related work: Prior research has explored LLMs for decision support, mental health, and information retrieval but lacks longitudinal, in-situ evidence of their roles across entire healthcare journeys. This study aims to fill this gap by examining how LLMs dynamically integrate into patients' practices and reshape agency, trust, and power.

Solution

• Proposed approach: Conceptualizing LLMs as longitudinal boundary companions that mediate between patients and clinicians across behavioral, informational, emotional, and cognitive dimensions.
• Novelty:
  1. Empirical evidence from a four-week diary study capturing LLMs’ dynamic roles in patients’ healthcare trajectories.
  2. Introduction of the concept of LLMs as longitudinal boundary companions, highlighting their socio-technical impact on patient agency, trust, and power.
  3. Design implications for relational continuity, responsible empowerment, and scaffolding health literacy in LLM-powered healthcare applications.
• Procedure and key techniques:
  • Conducted a four-week diary study with 25 participants managing diverse health conditions.
  • Participants documented daily interactions with LLMs and clinicians, followed by semi-structured exit interviews.
  • Inductive thematic analysis traced evolving roles of LLMs across behavioral, informational, emotional, and cognitive dimensions.

Results

• Concrete findings:
  • LLMs supported patients at four levels: behavioral (decision-making), informational (communication facilitation), emotional (companionship), and cognitive (sensemaking).
  • Patients used LLMs to prepare for clinical encounters, negotiate treatment plans, and sustain health agendas across stages.
  • Emotional support was a prominent role, with LLMs providing reassurance and buffering anxiety during critical moments.
  • LLMs decompressed complex medical advice into actionable guidance, extending sensemaking beyond clinical settings.
• Advantage over baselines:
  • LLMs empowered patients to actively participate in healthcare decisions, reconfiguring traditional power dynamics with clinicians.
  • They provided continuous, personalized support that complemented the episodic nature of clinical care.
• Experiments / evaluation:
  • 587 diary entries and 25 exit interviews analyzed to identify recurring patterns and dynamic shifts in LLM roles.
  • Participants spanned diverse health conditions and AI literacy levels, using both general-purpose and health-specific LLMs.
• Limitations and future work:
  • Study focused on patient perspectives; future work should include clinician viewpoints.
  • Conducted in China; cross-cultural studies are needed to generalize findings.
  • Longer-term studies could explore evolving reliance patterns and emotional connections with LLMs.

Summary

This study investigates how patients integrate LLMs into their healthcare-seeking journeys, revealing their dynamic roles across behavioral, informational, emotional, and cognitive dimensions. LLMs empowered patients to actively negotiate decisions, facilitated communication with clinicians, provided emotional reassurance, and scaffolded sensemaking of medical advice. By conceptualizing LLMs as longitudinal boundary companions, the study highlights their potential to reshape patient-provider relationships and address resource constraints in healthcare systems. Future design should focus on relational continuity, responsible empowerment, and scaffolding health literacy while mitigating risks of misinformation and emotional dependency.