Projecting COVID-19 disease severity in cancer patients using purposefully-designed machine learning

Area of Science:

• Oncology
• Infectious Diseases
• Computational Biology

Background:

• Predicting COVID-19 outcomes in cancer patients is challenging.
• Clinical variables' predictive value and interactions for COVID-19 severity in cancer are unclear.

Purpose of the Study:

• To develop a machine learning model for predicting COVID-19 severity in cancer patients.
• To classify patients into severe-early, severe-late, or non-severe outcomes based on early clinical data.

Main Methods:

• Utilized machine learning algorithms on data from 348 cancer patients.
• Classified patients based on clinical variables available at or before COVID-19 diagnosis.
• Defined outcomes as severe-early (within 3 days), severe-late (after 3 days), or non-severe oxygen support requirements.

Main Results:

• The algorithm achieved an Area Under the Receiver Operating Characteristic Curve (AUROC) of 70-85%.
• This significantly outperformed previous methods and univariate analyses.
• High accuracy was achieved using a combination of patient information, diagnoses, lab work, and cancer type, indicating combinatorial risk factors.

Conclusions:

• A computational tool was developed to identify high-risk cancer patients with COVID-19 early.
• This tool can assist in clinical decision-making and treatment selection for these patients.