Respiratory failure & ventilation
Early prognosis prediction in mechanically ventilated patients using machine learning for tertiary care hospital settings
Authors' conclusion
The automated machine-learning method for prognosis prediction may assist clinicians in the early triage of patients.
Study context
- Population
- Mechanically ventilated ICU patients, three intensive care units (ICUs) at AIIMS, Delhi98 clinical parameters
- Participants
- 660 participants
- Comparison
- K-nearest-neighbour (KNN) model with SHAP-selected features vs Sequential Organ Failure Assessment (SOFA)-only model
- Outcome
- Survival/death during ICU stay (prediction performance)
Methodological detailsmachine learning, 10-fold cross-validation, 70/30 train-test split, three ICUs
- Machine learning
- 10-fold cross-validation
- 70/30 train-test split
- Three ICUs
Performance
As reported in the abstract
Test performance for ICU mortality prediction
| Test or model | AUROC | Sensitivity | Specificity | PPV | NPV | Accuracy |
|---|---|---|---|---|---|---|
| KNN | 0.84 | 0.82 | 0.77 | 0.78 | 0.82 | 0.80 |
| SOFA-only | 0.73 | 0.82 | 0.63 | 0.69 | 0.78 | 0.73 |
AUROC
KNN0.84
SOFA-only0.73
Sensitivity
KNN0.82
SOFA-only0.82
Specificity
KNN0.77
SOFA-only0.63
PPV
KNN0.78
SOFA-only0.69
NPV
KNN0.82
SOFA-only0.78
Accuracy
KNN0.80
SOFA-only0.73
Further findings
Feature selection (SHAP and Random Forest)
- Features selected
- 39
- SOFA score feature importance
- 7.3%
- International normalized ratio (INR) feature importance
- 4.5%
- Respiratory rate feature importance
- 3.4%
Original publication
Early prognosis prediction in mechanically ventilated patients using machine learning for tertiary care hospital settings. J Clin Monit Comput. Published 27 November 2025. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →