Presentation (CTI)

OBJECTIVES: There is an increasing recognition of the importance of including the patient voice within health technology assessment (HTA) processes. Thematic analysis is used widely to interpret qualitative patient-centred research but is often resource-intensive and subject to variability between analysts. Advances in artificial intelligence (AI) have the potential to streamline analysis, yet comparative evaluations in clinical contexts remain limited. This study aimed to compare AI-assisted and human-led thematic analysis of patient interview data, using treatment preferences among individuals with non muscle-invasive bladder cancer (NMIBC) as a case study.
METHODS: Three semi-structured focus groups were conducted with patients diagnosed with NMIBC to explore their experiences, values, and preferences regarding treatment options. Transcripts were analysed independently using two approaches: (1) human-led inductive thematic analysis, and (2) AI-assisted thematic analysis using Regulaido by Pharos Labs.
RESULTS: Human-led analysis identified eight primary themes, while AI-assisted analysis identified five and employed a more complex coding approach. There was substantial overlap in the content of the themes generated between the two analyses, although differences were observed in thematic granularity and depth. Human-led analysis produced more nuanced themes relating to emotional and contextual factors (e.g. how well-informed patients were by HCPs and life following radical cystectomy), whereas AI generated outputs tended to consolidate these into broader categories. Determinants of treatment preference also differed, with AI-analyses identifying fewer, broader, categories and missing considerations including living with a urostomy and travel to treatment centres.
CONCLUSIONS: AI improved consistency and efficiency but lacked some emotional and contextual factors identified by the human-led analysis, including some determinants of treatment preference. The omission of certain practical, emotional, and quality-of-life considerations could implicate how patient preferences are represented within HTA. A hybrid approach combining AI efficiency with human interpretive insight may optimise qualitative analysis and encourage the inclusion of patient-centred research in HTA.