Critical Appraisal Tools for Evaluating Artificial Intelligence in Clinical Studies: Scoping Review

Area of Science:

• * Health research
• * Artificial intelligence (AI)
• * Clinical decision-making

Background:

• * Growing use of predictive and generative AI in health research.
• * Potential for systematic errors in AI studies, similar to traditional clinical studies.
• * Need for critical appraisal tools to translate AI evidence into clinical practice.

Purpose of the Study:

• * Identify existing critical appraisal tools for AI clinical studies.
• * Examine the concepts and domains covered by these tools.
• * Frame the research question using the Population-Concept-Context (PCC) framework.

Main Methods:

• * Searched medical and engineering databases (MEDLINE, Embase, CINAHL, PsycINFO, IEEE) up to April 2024.
• * Included clinical primary research with critical appraisal tools; excluded non-human research.
• * Adhered to PRISMA-ScR and PRISMA-S guidelines; used Rayyan for screening.

Main Results:

• * Included 70 records, identifying 46 tools: 26 reporting guides, 16 critical appraisal tools, 2 for study quality, and 2 for risk of bias.
• * Nine papers focused on bias classification or mitigation.
• * 15 chatbot assessment studies/reviews were identified, showing heterogeneity.

Conclusions:

• * Reporting tools are predominant, followed by critical appraisal tools; risk of bias tools are fewer.
• * Critical appraisal must consider AI-specific biases, fairness, and mitigation strategies.
• * Chatbot assessment is a rapidly evolving field requiring urgent progress in design, reporting, and appraisal.