JMIR Publications

ABSTRACT

Background:

EHealth literacy is essential for men who have sex with men (MSM) to effectively access, evaluate, and use digital health information for health self-management. Understanding the factors associated with eHealth literacy within China’s specific cultural and digital information environment is important for developing effective digital health interventions for MSM. Qualitative research can provide in-depth insights into individuals’ experiences, while capturing the contextual and sensitive dimensions of MSM interview data. Large language models (LLMs) are increasingly being applied to qualitative data analysis. However, limited research has evaluated their ability to interpret qualitative data rich in complex cultural and identity-sensitive contexts.

Objective:

To compare researcher-led and LLM-assisted thematic identification and interpretation when analysing culturally embedded and sensitive qualitative data, and to explore factors associated with eHealth literacy among Chinese MSM through integrated human-LLM analysis.

Methods:

We conducted a multistage qualitative study involving semistructured interviews with 18 Chinese MSM recruited through purposive and maximum-variation sampling. Data analysis was guided by the context-specific LLM workflow. Researchers conducted thematic analysis of the Chinese interview transcripts and established a standardized workflow for LLM-assisted analysis. ChatGPT 5.6, DeepSeek R1, and Gemini 3.5 conducted thematic analyses using the same interview data and analytical tasks. We systematically compared researcher-led and LLM-assisted outputs in terms of theme identification, interpretive depth, and contextual understanding, and revisited the original transcripts to verify the evidentiary basis and contextual appropriateness of the interpretations. A final thematic framework of factors associated with eHealth literacy among Chinese MSM was developed through researcher verification, comparison, and integration.

Results:

The researcher-led, Gemini 3.5-, DeepSeek R1-, and ChatGPT 5.6-assisted analyses respectively identified 6-14, 4-11, 4-18, and 9-42 themes-subthemes. All four analytical approaches converged on three common themes: individual cognitive foundations and experience, the digital health information environment, and social support. LLMs were able to identify culturally embedded expressions and provide contextually appropriate interpretations. However, the LLM-generated outputs differed in their coding emphases and interpretive styles, with Gemini 3.5-generated outputs more frequently exhibiting overinterpretation. A final integrated coding framework comprising six themes and 14 subthemes was developed to characterize factors associated with eHealth literacy among Chinese MSM: individual cognitive foundations and experience; digital technology and health information practices; digital health information environment; perceptions of discrimination against MSM; social support; and medical insurance policy gaps.

Conclusions:

EHealth literacy among Chinese MSM is associated with individual, digital, sociocultural, and health service factors. LLM-assisted coding can provide additional perspectives in analysing culturally embedded and sensitive qualitative data, but researcher verification remains important to ensure that generated interpretations are consistent with the original evidence.