Machine Learning Predicts ICU In-Hospital Mortality in Ards Patients Aged 80 and Above: A Multinational Multicenter...

Area of Science:

• Artificial Intelligence in Medicine
• Clinical Informatics
• Critical Care Medicine

Background:

• Elderly patients with Acute Respiratory Distress Syndrome (ARDS) admitted to the Intensive Care Unit (ICU) have high mortality rates.
• Accurate prediction and risk stratification are crucial for managing this vulnerable population.

Purpose of the Study:

• To develop and validate an interpretable machine learning (ML) model for predicting mortality in ICU patients over 80 with ARDS.
• To enable effective clinical risk stratification for this patient group.

Main Methods:

• Utilized a multicenter cohort including data from Chinese medical institutions and the MIMIC-IV database.
• Eight distinct ML models were trained and evaluated using metrics like AUC.
• The best-performing model (Random Forest) was selected and interpreted using SHAP analysis.

Main Results:

• The Random Forest model achieved the highest performance with an AUC of 0.835.
• The ML model accurately predicted mortality and stratified patients by risk.
• The Random Forest model significantly outperformed the APACHE II score and oxygenation index-based classification.

Conclusions:

• An interpretable ML model was successfully developed for predicting mortality in elderly ARDS ICU patients.
• The model provides accurate risk predictions and stratification, supported by SHAP analysis.
• This offers a scientific basis for improved clinical management of high-risk ARDS patients.