Machine Learning Applied to Clinical Laboratory Data in Spain for COVID-19 Outcome Prediction: Model Development and...

Area of Science:

• Computational biology and bioinformatics
• Machine learning in healthcare
• Epidemiology and public health

Background:

• The COVID-19 pandemic overwhelmed healthcare systems globally, necessitating tools for risk stratification.
• Existing diagnostic and treatment limitations highlighted the need for predictive models for patient outcomes.
• Spain's healthcare system faced significant challenges due to the pandemic's rapid progression.

Purpose of the Study:

• To develop a machine learning model for predicting COVID-19 infection severity and mortality.
• To utilize electronic health records and clinical laboratory parameters for predictive modeling.
• To identify key clinical features associated with poor COVID-19 prognosis to optimize resource allocation.

Main Methods:

• An extreme gradient boosting algorithm was selected for its predictive performance.
• Shapley Additive Explanations (SHAP) were employed to interpret model feature importance.
• A dataset of 1823 COVID-19 patients with 32 predictor variables was analyzed.

Main Results:

• The extreme gradient boosting model achieved high performance metrics, including an Area Under the Receiver Operator Characteristic Curve (AUROC) of 0.97.
• Key predictors for severe outcomes and mortality were identified as lactate dehydrogenase activity, C-reactive protein levels, neutrophil counts, and urea levels.
• The model demonstrated strong accuracy (0.94), sensitivity (0.93), and specificity (0.91) in predicting patient outcomes.

Conclusions:

• The developed machine learning model effectively predicts COVID-19 mortality using laboratory parameters.
• The identified significant features provide insights into disease mechanisms and patient risk factors.
• This predictive tool can aid clinicians in optimizing healthcare resource management for COVID-19 patients.