Respiratory failure & ventilation
Large language models for optimizing clinical trial recruitment in intensive-care unit (ICUs): application to ventilator-induced diaphragm dysfunction
Authors' conclusion
LLM-based prescreening offers a promising approach for identifying trial candidates in critical care, prioritizing candidates for clinician review.
Study context
- Population
- Consecutive ICU stays, Montpellier University Hospital, 2024; prescreening for a phase 2 trial targeting ventilator-induced diaphragm dysfunction
- Participants
- 1,342 ICU stays
- Outcome
- Criterion-level F1-score vs a reference standard of 50 ICU stays annotated by six clinicians
Methodological detailsretrospective screening of consecutive ICU stays, expert adjudication, single-centre
- Retrospective screening of consecutive ICU stays
- Expert adjudication
- Single-centre
Performance
As reported in the abstract
Criterion-level F1-score
GPT-OSS:120B
0.82
- F1 scores across LLM-based pipelines
- 0.73 to 0.82
Further findings
Real-world prescreening of 2024 ICU stays (LLM-assisted, expert-confirmed eligibility)
- ICU stays screened
- 1,342
- Patients with ≥ 3 days of mechanical ventilation identified
- 532
- Potentially eligible after exclusion criteria
- 185
- Confirmed eligible by expert review
- 133
- PPV
- 72% (95% CI 65-78)
- Selected model applied to consecutive ICU stays, Montpellier University Hospital
Clinician review time
- Estimated reduction with LLM-assisted workflow
- 86%
Original publication
Large language models for optimizing clinical trial recruitment in ICUs: application to ventilator-induced diaphragm dysfunction. Crit Care. Published 11 June 2026. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →