Machine learning-based cross-sectional exploration of depression-chronic pain associations in chronic respiratory disease patients.

History-taking is a fundamental clinical skill, and its quality directly affects diagnostic accuracy [,] and subsequent clinical reasoning []. For medical students, mastering a complete history-taking process and developing fluent clinical interview skills are core tasks in the transition from classroom learning to clinical practice []. However, achieving this transition has become increasingly difficult, as the erosion of bedside teaching and the decline of core clinical skills have been recognized as growing challenges in medical education []. Simulation-based instruction and standardized patient (SP) encounters have been proposed as effective alternatives to address this gap, with evidence supporting their value in improving student confidence and clinical competence [,]; however, their systematic implementation remains constrained by time and resource demands, limiting opportunities for repeated practice and structured feedback []. Virtual patient (VP) systems have emerged as a promising approach to address these challenges, offering interactive and repeatable practice environments suited to clinical reasoning training []. Existing evidence supports their educational value: a systematic review of conversational VP interventions found consistent improvements in history-taking and clinical reasoning competencies across multiple studies [], and a single-cohort intervention study demonstrated statistically significant gains in both self-reported competence and confidence, corroborated by improved objective structured clinical examination performance in medical and surgical history-taking []. Notably, that same systematic review found that student satisfaction tended to be higher when VP systems incorporated AI and natural language processing to enable more realistic conversational interaction [], suggesting that AI-enhanced dialogue may be associated with more favorable learner experiences compared with scripted or menu-driven formats. Building on this trajectory, recent advances in large language models have enabled a new generation of AI-powered virtual patient (AI-VP) systems capable of open-ended natural language interaction and automated feedback [-]. However, empirical evidence specifically evaluating AI-VP systems in history-taking education remains limited and preliminary, and a recent scoping review of AI applications in medical education underscored both the promise of these technologies and the need for rigorous evaluation []. Sustainable adoption therefore depends not only on technical performance and educational effectiveness, but also on user acceptance [,]. From a training-evaluation perspective, Kirkpatrick’s 4-level model provides a useful framework for considering how learners respond to educational interventions. Its first level, “Reaction,” focuses on learners’ immediate responses to training, including satisfaction and perceived value, and is sometimes considered an early indicator of subsequent learning engagement, though the assumption that Level 1 reactions predict higher-level outcomes has been critically examined in medical education [,]. Nevertheless, Level 1 evaluation remains a recognized starting point for understanding learner responses to novel educational technologies, particularly in early-stage implementations where downstream outcome data are not yet available. The Technology Acceptance Model (TAM) and related frameworks such as the Unified Theory of Acceptance and Use of Technology (UTAUT) emphasize that perceived usefulness, ease of use or effort expectancy, and related performance beliefs are central determinants of technology acceptance [,]. A recent systematic review of TAM in medical education suggested considering broader acceptance frameworks, including TAM2 extensions, for educational technologies in context []. Although mixed methods work on trust and acceptance of AI are emerging in medicine [], research that combines quantitative associations with qualitative exploration of underlying experiential patterns in medical education remains limited. In this study, Kirkpatrick Level 1 provides the curricular-evaluation positioning, whereas TAM/UTAUT-style acceptance logic informs the operational acceptance outcome (use intention, recommendation intention, and overall satisfaction). We treated acceptance as an exploratory, theory-informed construct rather than as a formal test or validated adaptation of TAM or UTAUT, and the experience composites correspond only partially to established acceptance constructs, as detailed in the Methods.