ICU admission and mortality classifiers for COVID-19 patients based on subgroups of dynamically associated profiles...

Area of Science:

• Medical Informatics
• Computational Biology
• Epidemiology

Background:

• The COVID-19 pandemic, caused by SARS-CoV-2, has overwhelmed global healthcare systems due to high transmissibility.
• There is a critical need to understand dynamic patient associations and identify subgroups for better clinical outcome prediction.
• Robust classifiers for Intensive Care Unit (ICU) admission and mortality are essential for managing COVID-19 patients.

Purpose of the Study:

• To develop and validate a computational pipeline for analyzing time-series clinical data of COVID-19 patients.
• To identify distinct patient subgroups with common clinical trajectories using clustering techniques.
• To build accurate predictive models for ICU admission and mortality by integrating dynamic association analysis and patient stratification.

Main Methods:

• An automated data curation workflow was employed to enhance the quality of multi-time-series clinical data.
• Dynamic Bayesian Networks (DBNs) were utilized for dynamic association analysis to detect feature connectivity.
• Self-Organizing Maps (SOMs) and trajectory analysis were used for patient subgroup identification, followed by Multiple Additive Regression Trees (MART) for classification.

Main Results:

• The pipeline achieved improved classification performance: 0.83 sensitivity and 0.83 specificity for ICU admission, and 0.74 sensitivity and 0.76 specificity for mortality.
• Inclusion of additional data enhanced mortality prediction classifiers by 4% in sensitivity and specificity.
• Identified risk factors for ICU admission included lymphocyte count, SatO2, and PO2/FiO2 ratio; for mortality, they included neutrophil and lymphocyte percentages, PO2/FiO2, LDH, and ALP.

Conclusions:

• This study presents a novel approach combining dynamic modeling and clustering for COVID-19 patient stratification.
• The developed method successfully identified homogeneous patient groups and improved the accuracy of ICU admission and mortality classifiers.
• The findings highlight key clinical features and biomarkers associated with severe COVID-19 progression, aiding in risk stratification and clinical decision-making.