대규모 언어모델을 활용한 질적 데이터의 양적 지표화 : 물리치료 융합을 위한 새로운 방법론 = (Quantification of Qualitative Data Using Large Language Models: A Novel Methodology for Physical Therapy Integration)

다국어 초록 (Multilingual Abstract)

Qualitative analysis constitutes an essential methodological approach in physical therapy research, offering comprehensive insights into patients' subjective experiences, perceptions of therapeutic interventions, and recovery trajectories, thereby facilitating evidence-based intervention development and optimizing clinical outcomes. Nevertheless, conventional qualitative analytical approaches applied to physical therapy patient interviews encounter substantial limitations regarding analytical efficiency and inter-rater reliability. The present study examines the integration of Large Language Models (LLMs) into qualitative research methodologies within the physical therapy domain to address these methodological constraints. Semi-structured interviews were conducted with one stroke patient as a pilot case, and three LLMs (ChatGPT, Claude, and Gemini) were employed with 160 iterative measurements each to implement the Large Language Model Quotient (LLMq) methodology—a quantitative framework for systematic qualitative coding—to evaluate its validity and applicability within the physical therapy research context. Intraclass correlation coefficient (ICC) analysis demonstrated ICC(3,1) of 0.802 and ICC(3,k) of 0.924, meeting Cicchetti's (1994) criteria for excellent reliability. Findings demonstrate that this analytical approach significantly enhances the efficiency of qualitative data analysis (reducing analysis time from weeks-months to hours-days) while mitigating subjective interpretive bias (ICC 0.802), thereby substantiating the considerable methodological value that LLM-based approaches confer upon physical therapy research endeavors.