Early Prediction of ICU Mortality in Patients with Acute Hypoxemic Respiratory Failure Using Machine Learning: The...

Area of Science:

• Critical Care Medicine
• Respiratory Medicine
• Medical Informatics

Background:

• Early prediction of Intensive Care Unit (ICU) death in acute hypoxemic respiratory failure (AHRF) is vital for timely therapeutic interventions.
• Identifying modifiable and non-modifiable risk factors within the initial 24 hours of AHRF can improve patient outcomes and survival rates.

Purpose of the Study:

• To develop and validate a predictive model for ICU mortality in patients with AHRF.
• To identify key clinical features within the first 24 hours of AHRF that are associated with ICU death.

Main Methods:

• A prospective, multicenter, observational cohort study involving 1241 patients with AHRF on mechanical ventilation.
• Development of a logistic regression model using genetic algorithm and machine learning (ML) for variable selection.
• Model development and testing on 75% of the cohort, with final validation on the remaining 25%.

Main Results:

• The final model included six predictors: patient age, and 24-hour values of PEEP, FiO2, plateau pressure, tidal volume, and number of extrapulmonary organ failures.
• The ML model achieved a high Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.88 (95% CI 0.86-0.90) during development.
• Validation confirmed adequate model performance with an AUROC of 0.83 (95% CI 0.78-0.88).

Conclusions:

• Machine learning and traditional statistical methods can create effective models for predicting ICU death in ventilated AHRF patients.
• The developed model demonstrates encouraging performance in predicting ICU mortality as early as 24 hours post-diagnosis.
• Further research is necessary to identify modifiable factors that can be targeted to prevent ICU deaths in this patient population.