NPJ Digit Med

AI tools help draft patient messages but may introduce critical errors

May 15, 2025

Study details: This cross-sectional simulation study included 20 PCPs who reviewed 18 patient portal messages, including four with errors categorized as objective inaccuracies or potentially harmful omissions. The study aimed to assess whether PCPs could identify and correct errors in AI-generated draft responses to patient messages.

Results: While 80% of participants agreed that AI drafts reduced cognitive workload and 75% found them safe, 35% to 45% of erroneous drafts were submitted entirely unedited. Errors were insufficiently addressed by 13-15 participants, highlighting significant patient safety risks.

Clinical impact: AI-generated drafts can alleviate the cognitive burden on PCPs, potentially reducing burnout, but the uncorrected errors in AI drafts pose serious patient safety concerns. Use caution when adopting AI tools for patient messaging and advocate for robust error-detection systems and comprehensive training to ensure safe and effective use.

Source:

Biro JM, et al. (2025, April 24). NPJ Digit Med. Opportunities and risks of artificial intelligence in patient portal messaging in primary care. https://pubmed.ncbi.nlm.nih.gov/40275104/