Multivariable Risk Modelling and Survival Analysis with Machine Learning in SARS-CoV-2 Infection

Area of Science:

• Medical imaging and artificial intelligence
• Infectious disease research
• Computational pathology

Background:

• The COVID-19 pandemic caused by SARS-CoV-2 has led to significant morbidity and mortality.
• Predicting critical outcomes in SARS-CoV-2 patients is crucial for timely intervention and resource allocation.
• Existing prediction models often lack comprehensive data integration.

Purpose of the Study:

• To develop and evaluate a machine learning model for predicting critical outcomes in SARS-CoV-2 patients.
• To assess the predictive performance of integrating demographic variables, blood tests, comorbidities, and CT-based radiomic features.
• To identify key predictors of severe illness and mortality in SARS-CoV-2 infection.

Main Methods:

• Retrospective analysis of 694 SARS-CoV-2 positive patients.
• Data collection included demographics, blood tests (CRP, LDH, D-dimer), comorbidities, and CT-derived radiomic features.
• A machine learning model was developed using LASSO Cox regression and evaluated on a separate test set.
• Model performance was assessed using C-statistic and Brier scores.

Main Results:

• The study identified D-dimer levels, age, and cardiovascular disease as significant risk factors for critical outcomes.
• The developed machine learning model accurately classified 90% of non-survivors as high-risk in the test dataset.
• High-risk patients identified by the model had a significantly shorter median survival (9 days) compared to low-risk patients (p < 0.001).

Conclusions:

• A machine learning model integrating readily available clinical data (demographics, CT-radiomics, comorbidities, biomarkers) can effectively predict critical illness and mortality in SARS-CoV-2 patients.
• This predictive model can aid clinicians in early risk stratification and management of severe COVID-19 cases.
• The findings highlight the potential of multimodal data integration for improving prognostic accuracy in infectious diseases.