Prescribing Caution: A Critique of OpenEvidence to Answer Medication-Related Questions

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Pharmacology

Background:

• Large language models (LLMs) like OpenEvidence are increasingly used by healthcare professionals for clinical decision-making and research.
• Existing literature provides a general overview of LLM accuracy in answering clinical questions.
• There is a notable gap in research evaluating OpenEvidence's performance specifically from a pharmaceutical perspective.

Purpose of the Study:

• To review the current literature on the accuracy of OpenEvidence for clinical questions.
• To highlight potential inaccuracies and source summarization errors in OpenEvidence's responses concerning pharmacotherapeutics.
• To provide recommendations for the responsible integration of OpenEvidence in pharmacy practice.

Main Methods:

• Literature review of studies assessing OpenEvidence's accuracy in clinical question answering.
• Analysis of two illustrative case examples demonstrating OpenEvidence's errors in pharmacotherapeutic content.
• Qualitative assessment of source summarization accuracy.

Main Results:

• The current literature on OpenEvidence's accuracy is limited, especially concerning pharmaceutical applications.
• Illustrative examples reveal specific instances of incorrect responses and flawed source summarization related to medication information.
• Identified inaccuracies underscore the need for critical evaluation of OpenEvidence outputs.

Conclusions:

• OpenEvidence, while promising, exhibits inaccuracies in pharmacotherapeutic information, posing risks in clinical settings.
• Pharmacists and healthcare professionals must exercise caution and verify information provided by OpenEvidence.
• Guidelines for responsible use are essential as the adoption of such AI tools grows in healthcare.