Protocol for human evaluation of generative artificial intelligence chatbots in clinical consultations

Area of Science:

• Artificial Intelligence in Medicine
• Clinical Informatics
• Healthcare Technology

Background:

• Generative artificial intelligence (GenAI) offers transformative potential in healthcare.
• Real-world clinical practice necessitates rigorous, evidence-based AI deployment for safety and efficacy.

Purpose of the Study:

• To systematically evaluate large language models (LLMs) as GenAI chatbots in clinical microbiology and infectious diseases.
• To critically assess recommendations from leading GenAI models like Claude 2, Gemini Pro, and GPT-4.0.

Main Methods:

• Development of a standardized, healthcare-specific prompt template for eliciting clinical AI responses.
• Evaluation of AI-generated recommendations by clinician panels using a 5-point Likert scale.
• Assessment across factual consistency, comprehensiveness, coherence, and medical harmfulness.

Main Results:

• The study protocol provides a framework for assessing GenAI chatbots in clinical settings.
• Insights into the feasibility, limitations, and boundaries of GenAI in consultations will be generated.
• Data will inform future research and clinical implementation of AI in healthcare.

Conclusions:

• A standardized evaluation protocol is crucial for assessing GenAI in clinical practice.
• Clinician-led evaluation is essential for ensuring the safety and efficacy of AI tools.
• Development of ethical guidelines and safety guardrails is paramount for patient safety and clinical standards.