Respiratory failure & ventilation
Development and validation of VentPilot: an AI-based recommendation system for mechanical ventilation
Authors' conclusion
In multicenter retrospective validation cohorts, fitted Q-evaluation estimated higher returns for VentPilot than for observed clinician behavior under the prespecified reward function, while greater concordance between observed care and VentPilot recommendations was associated with more favorable clinical outcomes. Further clinical evaluation is warranted to establish VentPilot's safety, usability, and clinical impact.
Study context
- Population
- Mechanically ventilated adults at Seoul National University Hospital (SNUH) and Mayo Clinicderivation cohort 3,002 · internal validation cohort 283 · external validation cohort 1,011
- Participants
- 4,296 mechanically ventilated adults
- Comparison
- Estimated return of VentPilot relative to observed clinician behavior under the prespecified reward function; patients with higher versus lower concordance between observed ventilator settings and VentPilot recommendations
- Outcome
- Overall reward (fitted Q-evaluation, prespecified reward function)
Methodological detailsoffline reinforcement learning, multicenter retrospective validation cohorts, fitted Q-evaluation, inverse probability-weighted analysis
- Offline reinforcement learning
- Multicenter retrospective validation cohorts
- Fitted Q-evaluation
- Inverse probability-weighted analysis
Recommendations
As reported in the abstract
Estimate (95% CI)
VentPilot vs observed clinician behavior
difference in overall reward(higher with a), internal validation cohort 1.6 (0.8–2.5)
VentPilot vs observed clinician behavior
difference in overall reward(higher with a), external validation cohort 1.3 (0.6–2.0)
Further findings
| Outcome | Estimate | |
|---|---|---|
| Ventilator-free days · within 28 daysHigher concordance vs lower concordance · inverse probability-weighted analyses | mean difference (more with a), internal validation cohort 5.0 days (95% CI 2.4–7.6) | |
| Ventilator-free days · within 28 daysHigher concordance vs lower concordance · inverse probability-weighted analyses | mean difference (more with a), external validation cohort 3.1 days (95% CI 1.7–4.6) | |
| −50510Difference, days (linear scale) |
28-day mortality and ICU length of stay
- Higher concordance associated with lower 28-day mortality and shorter ICU length of stay
Original publication
Development and validation of VentPilot: an AI-based recommendation system for mechanical ventilation. J Intensive Care. Published 1 October 2026. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →