Respiratory failure & ventilation
Enhancing predictive modeling for respiratory support with large language model (LLM)-driven guideline adherence
Main finding · IMV, HFNC recommendationResult — go to the resultRelative risk increase 97.33%Authors' conclusion
Integrating LLMs for guideline enforcement improves the interpretability and clinical alignment of counterfactual models in respiratory support decision-making.
Study context
- Population
- Intensive care units (ICU) patients at high risk of invasive mechanical ventilation receiving high-flow nasal cannula (HFNC) or noninvasive ventilation (NIV)
- Participants
- 1,261 ICU encounters
- Exposure
- Discordant vs Concordant
- Outcome
- Invasive mechanical ventilation (IMV), HFNC recommendation
Methodological detailsmodel development and evaluation, structured chart review, LLM (Claude 3.5 Sonnet) in HIPAA-compliant AWS environment, prompted with structured data, clinical notes and guideline criteria
- Model development and evaluation
- Structured chart review
- LLM (Claude 3.5 Sonnet) in HIPAA-compliant AWS environment, prompted with structured data, clinical notes and guideline criteria
Main finding
As reported in the abstract
Relative risk increase
97.33%
Events by group
Observational data: associations, not effects.
Further findings
- Concordance associated with reduced mortality or hospice discharge (OR 0.670)
- p = 0.046
Safety
Errors in LLM recommendations (20-case chart review)
- Cases with errors, most low or moderate risk
- 11/20
- Potentially causing severe harm
- 2/20
Original publication
Enhancing predictive modeling for respiratory support with LLM-driven guideline adherence. Crit Care. Published 14 November 2025. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →