A Multimodality Machine Learning Approach to Differentiate Severe and Nonsevere COVID-19: Model Development and...

Area of Science:

• Medical Informatics
• Infectious Disease Research
• Machine Learning in Healthcare

Background:

• Accurate COVID-19 diagnosis and severity classification are crucial for patient outcomes and healthcare system efficiency.
• Current methods for differentiating severe and non-severe COVID-19 rely on limited features that may not fully capture disease complexity.
• Existing diagnostic features may not be readily available at the time of initial patient assessment.

Purpose of the Study:

• To develop and validate a machine learning model for comprehensive and accurate differentiation of severe and non-severe COVID-19 clinical types.
• To identify key medical features that reliably predict COVID-19 clinical severity.
• To provide a robust predictive tool for COVID-19 disease classification.

Main Methods:

• Recruited 214 non-severe and 148 severe COVID-19 patients.
• Utilized 26 clinical characteristics and 26 laboratory test results as input modalities.
• Developed and validated random forest machine learning models using all features and top-ranked features.

Main Results:

• Models achieved over 90% accuracy using clinical data and over 95% using laboratory data independently.
• Top predictors included age, hypertension, cardiovascular disease, diabetes, dimerized plasmin fragment D, troponin I, neutrophil count, IL-6, and LDH.
• A model using the top 10 multimodal features achieved 97% predictive accuracy.

Conclusions:

• Machine learning models can effectively differentiate COVID-19 severity using readily available clinical and laboratory data.
• Identified key features provide insights into the human body's response to SARS-CoV-2 infection.
• Clinical information serves as an initial screening tool, while laboratory results prioritize accuracy for severity assessment.