In-hospital testing of NIVPredict - an AI tool for early prediction of non-invasive ventilation outcome in acute respiratory failure
Authors' conclusion
This study demonstrates superior predictive performance, compared to current clinical indices, of an AI-based tool for NIV outcome prediction on a cohort of patients with overt-acute and acute-on-chronic respiratory failure.
Study context
- Population
- Immunocompromised and immunocompetent patients with hypoxemic acute respiratory failure (ARF) (pneumonia, sepsis or COVID-19) or hypercapnic ARF (acute exacerbation of COPD or obesity hypoventilation syndrome) receiving non-invasive ventilation (NIV)665 training, 422 external, 57 in-hospital
- Comparison
- NIVPredict AI tool vs conventional clinical indices
- Outcome
- Prediction of NIV outcome:
- AUC
- Balanced accuracy
- Follow-up
- Within two hours of initiation
Methodological detailsmachine learning (TabPFN), internal repeated cross-validation, external validation, in-hospital useability testing, Trained on 665 ARF patients from the RENOVATE trial (Brazil); external validation on 422 ARF patients (Italy, Spain, USA); useability tested at University Hospitals of North Midlands NHS Trust, UK, December 2024–November 2025, on 57 eligible ARF patients
- Machine learning (TabPFN)
- Internal repeated cross-validation
- External validation
- In-hospital useability testing
- Trained on 665 ARF patients from the RENOVATE trial (Brazil); external validation on 422 ARF patients (Italy, Spain, USA); useability tested at University Hospitals of North Midlands NHS Trust, UK, December 2024–November 2025, on 57 eligible ARF patients
Performance
As reported in the abstract
AUC and balanced accuracy for NIV outcome prediction
| Test or model | AUC, internal repeated cross-validation | AUC, external validation | AUC, in-hospital testing | Balanced accuracy, internal repeated cross-validation | Balanced accuracy, external validation | Balanced accuracy, in-hospital testing | Balanced accuracy (best of HACOR or Updated HACOR), internal repeated cross-validation | Balanced accuracy (best of HACOR or Updated HACOR), external validation | Balanced accuracy (best of HACOR or Updated HACOR), in-hospital testing |
|---|---|---|---|---|---|---|---|---|---|
| NIVPredict | 0.793 | 0.772 | 0.858 | 78.9% | 74.5% | 85.0% | — not stated in the abstract | — not stated in the abstract | — not stated in the abstract |
| Updated HACOR score | 0.717 | 0.709 | 0.693 | — not stated in the abstract | — not stated in the abstract | — not stated in the abstract | 68.7% | 63.7% | 67.6% |
AUC, internal repeated cross-validation
NIVPredict0.793
Updated HACOR score0.717
AUC, external validation
NIVPredict0.772
Updated HACOR score0.709
AUC, in-hospital testing
NIVPredict0.858
Updated HACOR score0.693
Balanced accuracy, internal repeated cross-validation
NIVPredict78.9%
Updated HACOR score— not stated in the abstract
Balanced accuracy, external validation
NIVPredict74.5%
Updated HACOR score— not stated in the abstract
Balanced accuracy, in-hospital testing
NIVPredict85.0%
Updated HACOR score— not stated in the abstract
Balanced accuracy (best of HACOR or Updated HACOR), internal repeated cross-validation
NIVPredict— not stated in the abstract
Updated HACOR score68.7%
Balanced accuracy (best of HACOR or Updated HACOR), external validation
NIVPredict— not stated in the abstract
Updated HACOR score63.7%
Balanced accuracy (best of HACOR or Updated HACOR), in-hospital testing
NIVPredict— not stated in the abstract
Updated HACOR score67.6%
— not stated in the abstract
Original publication
In-hospital testing of NIVPredict - an AI tool for early prediction of non-invasive ventilation outcome in acute respiratory failure. Crit Care. Published 15 February 2026. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →