JMIR Publications

ABSTRACT

Background:

Low patient accrual remains a systemic barrier in cancer clinical trials, with participation rates stagnating between 2% and 8%. Existing research on patient decision-making often lacks the synergy of combining broad digital footprints with in-depth experiential data. This study employs a mixed-methods approach, integrating human qualitative analysis with Large Language Model (LLM) capabilities to evaluate patient attitudes toward trial participation.

Objective:

To understand cancer patients’ attitude toward drug clinical trials, their willingness to participate in drug clinical trials, and the factors influencing these decisions.

Methods:

This mixed-methods study analyzed data from two sources: interview transcripts from 11 cancer patients and online platform comments. A parallel analytical approach was employed, triangulating findings from three distinct analyses: (1) a manual content analysis conducted by two independent researchers with discrepancies resolved through consensus; and (2) two separate computational analyses using distinct large language models (LLMs), Gemini Pro 2.5 and DeepSeek R1. The LLMs processed anonymized data via a consistent, zero-shot prompting strategy to ensure methodological rigor. Thematic outputs were subsequently compared to identify convergent and divergent insights across human and computational interpretations.

Results:

A multi-layered landscape of patient decision-making was revealed through a three-pronged analytical approach. A core set of factors was identified across all analyses, including the dynamic balance between therapeutic hope and risk perception, the foundational role of physician-patient trust, and significant financial pressures. Deeper insights into distinct patient archetypes were generated by the AI-driven analyses. A “Proactive, Goal-Oriented” patient archetype was delineated by the Gemini Pro 2.5 analysis, contrasting with an “Economically-Driven, Reluctantly Optimal” patient group. Furthermore, socio-demographic nuances, such as passive acceptance among the elderly and significant urban-rural disparities in information access, were highlighted by the DeepSeek R1 analysis.

Conclusions:

Enhancing clinical trial recruitment and patients compliance management demands a holistic, multi-pronged strategy that moves beyond simple information provision. It requires simultaneously leveraging trusted physician-patient relationships for education, optimizing trial logistics to alleviate participant burden, and developing culturally-resonant consent frameworks that acknowledge the role of the family. Aligning the procedural demands of science with the humanistic needs of patients is fundamental to building the trust necessary for sustainable research advancement.