Using a Large Language Model to Support Thematic Analysis of Patient Experiences in Chronic Illness Management:...

Area of Science:

• Qualitative Health Research
• Artificial Intelligence in Healthcare
• Computational Social Science

Background:

• Qualitative health research extensively studies patient experiences with chronic illnesses.
• Large language models (LLMs) offer new avenues for analyzing narrative health data.
• The utility of LLMs compared to human analysis in complex contexts like multimorbidity is underexplored.

Purpose of the Study:

• To evaluate the methodological contribution of LLM-assisted analysis.
• To compare LLM analysis with traditional thematic analysis in replicating and extending qualitative insights.
• To assess LLM capabilities in complex clinical contexts such as multimorbidity.

Main Methods:

• Conducted semistructured interviews with 30 individuals managing multiple chronic illnesses.
• Analyzed interview transcripts using both manual thematic coding and the Claude 3.5 Sonnet LLM.
• Performed a structured comparison of themes, subthemes, and detail levels between manual and LLM approaches.

Main Results:

• Both manual and LLM analyses identified similar core patient experience themes (e.g., healthcare navigation, support systems, emotional challenges).
• Manual analysis yielded more contextually detailed interpretations, while LLMs identified more subthemes.
• Distinct themes emerged: manual analysis highlighted faith and caregiving, while LLMs identified future planning and multiple health conditions.

Conclusions:

• A hybrid approach integrating AI-assisted and human-led thematic analysis enhances analytical depth and scalability.
• LLMs serve as valuable complementary tools in qualitative research.
• Combining automated pattern detection with human interpretation is crucial for robust qualitative analysis.