A large language model improves clinicians’ diagnostic performance in complex critical illness cases
Authors' conclusion
For diagnostically difficult critical illness cases, DeepSeek-R1 generates high-quality information, achieves reasonable diagnostic accuracy, and significantly improves residents' diagnostic accuracy and efficiency.
Study context
- Population
- Challenging critical illness cases from the literature, handled by critical care residents from tertiary teaching hospitals48 cases
- Participants
- 32 participants
- Exposure
- Artificial intelligence (AI)-assisted vs non-AI-assisted critical care residents
- AI-assisted physiciann = 16
- Non-AI-assisted physiciann = 16reference
- Outcome
- Top diagnosis accuracy
Methodological detailsrandomly assigned residents, Reasoning model DeepSeek-R1; each resident handled an average of 3 cases
- Randomly assigned residents
- Reasoning model DeepSeek-R1; each resident handled an average of 3 cases
Main finding
As reported in the abstract
Observational data: associations, not effects.
Further findings
Residents' diagnostic time
95% CI AI-assisted physician 675-1200 · Non-AI-assisted physician 1710-2370
DeepSeek-R1 response quality (Likert grades)
- Completeness, median
- 4.0 (IQR 4.0-5.0; 95% CI 4.0-4.5)
- Clarity, median
- 5.0 (IQR 4.0-5.0; 95% CI 4.5-5.0)
- Usefulness, median
- 5.0 (IQR 4.0-5.0; 95% CI 4.0-5.0)
AI model alone: top diagnosis accuracy and differential diagnosis quality score
- Top diagnosis accuracy
- 60% (29/48; 95% CI 0.456-0.729)
- Median differential diagnosis quality score
- 5.0 (IQR 4.0-5.0; 95% CI 4.5-5.0)
- AI model showed higher diagnostic accuracy than residents
Residents' diagnostic time
- Diagnostic time significantly decreased with AI assistance
Parts of this Brief are withheld: the automatic checks could not confirm them against the abstract.
Original publication
A large language model improves clinicians’ diagnostic performance in complex critical illness cases. Crit Care. Published 6 June 2025. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →