Machine Learning Accurately Predicts Need for Critical Care Support in Patients Admitted to Hospital for...

Area of Science:

• Medical informatics
• Clinical decision support
• Artificial intelligence in healthcare

Background:

• Hospitalized community-acquired pneumonia (CAP) patients often require critical care interventions such as ventilation, vasopressors, and renal replacement therapy (RRT).
• Accurate prediction of these needs is crucial for timely intervention and resource allocation.
• Existing methods may not fully capture the complexity of predicting these severe outcomes in CAP patients.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for predicting the need for invasive ventilation, vasopressors, and RRT in hospitalized CAP patients.
• To compare the predictive accuracy of ML models against traditional logistic regression (LR).
• To assess the performance of different ML algorithms, including random-forest classifier (RFC), support vector machines (SVMs), Extreme Gradient Boosting (XGBoost), and multilayer perceptron (MLP).

Main Methods:

• A retrospective observational study design was employed.
• Separate ML models were trained using RFC, SVM, XGBoost, and MLP to predict the eventual use of invasive ventilation, vasopressors, and RRT.
• Models were derived and validated in cohorts of COVID-19 and non-COVID-19 CAP patients.

Main Results:

• Random-forest classifier (RFC) models demonstrated the highest accuracy among the tested ML algorithms.
• ML models achieved very high area under the receiver operating characteristic curve (AUROC) values, ranging from 0.74 to 0.95.
• ML models utilized variables like Fio2, Glasgow Coma Scale, mean arterial pressure, creatinine, and potassium for predictions.
• Logistic regression (LR) models were less accurate, with AUROC values ranging from 0.66 to 0.8.

Conclusions:

• Machine learning algorithms, particularly RFC, significantly outperform logistic regression in predicting critical care interventions for hospitalized CAP patients.
• These ML models can enhance clinician judgment for triage and patient management in both COVID-19 and non-COVID-19 CAP cases.
• The developed ML approach offers a promising tool for improving care for severe CAP.