Early Prediction of COVID-19 Ventilation Requirement and Mortality from Routinely Collected Baseline Chest...

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Epidemiology

Background:

• The COVID-19 pandemic caused by SARS-CoV-2 has overwhelmed global healthcare systems.
• A significant number of patients initially diagnosed with mild COVID-19 later developed severe disease.
• Accurate early prediction of severe outcomes is crucial for patient management and resource allocation.

Purpose of the Study:

• To develop predictive models for COVID-19 mechanical ventilation and mortality.
• Utilize routinely collected patient data, including chest X-rays (CXR), complete blood counts (CBC), demographics, and patient history.
• Enable early risk stratification at the time of diagnosis.

Main Methods:

• Retrospective collection of data from 5739 COVID-19 patients.
• Application of four machine learning algorithms with feature selection and data balancing techniques.
• Validation of models for ventilatory support (1508 patients) and mortality (1513 patients) endpoints.

Main Results:

• A predictive model for ventilation requirement achieved an AUC of 0.87 and balanced accuracy of 0.81.
• A mortality prediction model yielded an AUC of 0.83 and balanced accuracy of 0.80.
• Combined data features consistently outperformed individual data sets for predicting intubation and mortality.

Conclusions:

• The developed models demonstrate practical utility for hospital resource planning.
• These predictive tools can assist in prioritizing patients during the COVID-19 pandemic.
• Routinely collected clinical and radiological data are effective for predicting severe COVID-19 outcomes.