Respiratory failure & ventilation
Development and validation of a machine learning model for real-time prediction of invasive mechanical ventilation weaning readiness
Authors' conclusion
This study presents three ML models for real-time prediction of IMV weaning readiness, offering a promising approach to enhance clinical decision-making and optimize patient care.
Study context
- Population
- Adults admitted after 2010 with invasive mechanical ventilation >24 h, MIMIC-IV (development, internal validation) and AmsterdamUMCdb (external validation)8,565 MIMIC-IV
- Comparison
- 24-h model vs 48-h model vs 72-h model
- Outcome
- Discrimination for IMV weaning readiness (auROC, auPR, performance metrics)
Methodological detailsretrospective database, internal and external validation, XGBoost
- Retrospective database
- Internal and external validation
- XGBoost
Performance
As reported in the abstract
Model performance, external validation cohort
| Test or model | Areas under Receiver Operating Characteristic (auROC) | auPR | Specificity | Sensitivity |
|---|---|---|---|---|
| 24-h model | 0.847 | 54.17 | 0.79 | >0.75 |
| 48-h model | 0.795 | 54.56 | >0.75 | |
| 72-h model | 0.789 | 59.4 | 0.63 | >0.75 |
Areas under Receiver Operating Characteristic (auROC)
24-h model0.847
48-h model0.795
72-h model0.789
auPR
24-h model54.17
48-h model54.56
72-h model59.4
Specificity
24-h model0.79
48-h model— not stated in the abstract
72-h model0.63
Sensitivity
24-h model>0.75
48-h model>0.75
72-h model>0.75
Further findings
Sensitivity analyses by age and admission reason (subgroup)
- Lower performance in older (> 60 years) and neurosurgical patients
Original publication
Development and validation of a machine learning model for real-time prediction of invasive mechanical ventilation weaning readiness. J Crit Care. Published 27 May 2025. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →