Mortality Prediction Using SaO₂/FiO₂ Ratio Based on eICU Database Analysis.

Area of Science:

• Critical Care Medicine
• Pulmonary Medicine
• Medical Informatics

Background:

• The PaO2/FiO2 (P/F) ratio is a standard metric for assessing hypoxemia and a criterion in the Berlin definition of Acute Respiratory Distress Syndrome (ARDS).
• P/F ratio measurement is invasive, costly, and impractical in resource-limited environments.
• The noninvasive SaO2/FiO2 (S/F) ratio offers a cost-effective, continuous, and reliable alternative with reduced infection risk.

Purpose of the Study:

• To evaluate the hypothesis that the S/F ratio is noninferior to the P/F ratio in predicting intensive care unit (ICU) mortality.
• To compare the relative predictive abilities of S/F and P/F ratios for ICU mortality using a machine-learning approach.

Main Methods:

• Data from the eICU Collaborative Research Database were analyzed, including demographics, physiological parameters (SaO2, PaO2, FiO2), admission diagnosis, Apache IV score, and mechanical ventilation status.
• An XGBoost machine-learning model was developed and optimized using stratified cross-validation to predict ICU mortality, with AUC as the performance metric.
• Feature importance was assessed using SHAP, ELI5, and built-in XGBoost methods to determine the relative predictive power of S/F and P/F ratios.

Main Results:

• The optimized XGBoost model achieved an AUC of 0.85 on the test set.
• Feature importance analysis revealed the S/F ratio as the strongest physiological predictor of mortality.
• Partial dependence plots indicated a significant increase in mortality probability with S/F ratios above 200.

Conclusions:

• The S/F ratio demonstrated superior predictive power for ICU mortality compared to the P/F ratio in this study.
• The S/F ratio serves as a robust, noninvasive indicator for monitoring hypoxemia and predicting mortality in critically ill patients.
• Further prospective studies are warranted to validate these findings and their clinical utility.