Researchers developed and internally validated a machine-learning clinical decision support system designed to predict rapid oxygenation deterioration in mechanically ventilated ICU patients.

Retrospective data from a single university hospital (March 2020–September 2022) involving 3,267 adult patients who received 24 hours of invasive mechanical ventilation.

Because chest imaging wasn’t available to formally diagnose ARDS, the primary endpoint was a strict, rapid loss of oxygenation between two time windows separated by 48 hours.

The gradient-boosted tree model (XGBoost) significantly outperformed baseline logistic regression, achieving a high Area Under the Receiver Operating Characteristic curve.

By tracking dynamic oxygenation trends rather than static, fixed cutoffs, the model successfully identifies patients at risk for severe hypoxemia up to 72 hours in advance.

The study demonstrates that forecasting near-term oxygenation loss in ventilated patients using AI is highly feasible. While these internal results are promising, external validation and prospective clinical trials are required to prove the system’s real-world generalizability and impact on patient outcomes.

Read More