Assessing GPT-4's accuracy in answering clinical pharmacological questions on pain therapy.

Abstract

Aims: This study aimed to evaluate the accuracy and completeness of GPT-4, a large language model, in answering clinical pharmacological questions related to pain therapy, with a focus on its potential as a tool for delivering patient-facing medical information. The objective was to assess its reliability in delivering medical information in the context of pain management.

Methods: A cross-sectional survey-based study was conducted with healthcare professionals, including physicians and pharmacists. Participants submitted up to 8 clinical pharmacology questions on pain management, focusing on drug interactions, dosages and contraindications. GPT-4's responses were evaluated based on comprehensibility, detail, satisfaction, medical-pharmacological accuracy and completeness. Additionally, responses were compared to the German Drug Directory to assess their accuracy.

Results: The majority of participants (99%) found GPT-4's responses comprehensible, while 84% considered the information detailed enough. Overall satisfaction was high, with 93% expressing satisfaction, and 96% deemed the responses medically accurate. However, only 63% rated the information as complete, with some identifying gaps in pharmacokinetics and drug interaction data. Usability was evaluated as good to excellent, with a System Usability Scale score of 83.38 (± 10.26).

Conclusion: GPT-4 demonstrates potential as a tool for delivering medical information, particularly in pain management. However, limitations such as incomplete pharmacological data and the potential for contextual carryover in follow-up questions suggest that further refinement is necessary. Developing specialized artificial intelligence tools that integrate real-time pharmacological databases could improve accuracy and reliability for clinical decision-making.