The research data were generated as part of a study investigating the utility of large language models (LLMs) in solving clinical cases. Seven clinical cases, previously generated using different LLMs and subsequently reviewed and refined based on assessments by clinical pharmacists, were each solved by three LLMs. Each model answered two general and five specific questions related to each clinical case. Standardized instructions were provided to all models, and all interactions were conducted in Slovenian.
The dataset contains textual responses generated by the LLMs to questions concerning the identification of drug-related problems, the proposal of pharmaceutical interventions, and the assessment of various aspects of treatment effectiveness and safety. The data are organized according to the clinical case, question, and model that generated the response. The dataset consists of qualitative, unstructured or semi-structured textual data, enabling comparison of the professional accuracy, comprehensiveness, clinical relevance, and specificity of responses generated by different models. The responses served as the basis for subsequent evaluation by clinical pharmacists and other LLMs, using a five-point Likert scale and qualitative comments.
|
