Predictive Machine Learning Models and Survival Analysis for COVID-19 Prognosis Based on Hematochemical Parameters

Area of Science:

• Clinical Medicine
• Biochemistry
• Data Science

Background:

• The COVID-19 pandemic caused millions of deaths globally, necessitating accurate patient management strategies.
• Hematochemical alterations, including inflammatory markers, are observed in COVID-19 patients.
• Predicting clinical outcomes like mortality and intensive care unit (ICU) admission is vital for patient care.

Purpose of the Study:

• To identify unfavorable laboratory predictors for mortality and ICU admission in COVID-19 patients.
• To develop robust machine learning models for predicting severe COVID-19 outcomes.
• To establish a prognostic signature using hematochemical parameters for disease severity.

Main Methods:

• Retrospective analysis of 303 COVID-19 patients' data from the Polyclinic Hospital of Bari.
• Survival analysis using Kaplan-Meier curves and Cox Regression to identify significant predictors.
• Comparison of machine learning models (Decision Tree, Support Vector Machine) for outcome prediction.

Main Results:

• C-reactive protein (min value) was the most significant predictor for both mortality (HR=17.963) and ICU admission (HR=1.789).
• Erythrocytes (max value) also showed significance for mortality prediction (HR=1.765).
• Decision Tree model achieved 89.66% ROC-AUC for predicting death, while Support Vector Machine achieved 95.07% for ICU admission.

Conclusions:

• Hematochemical parameters, particularly C-reactive protein and erythrocyte levels, serve as strong prognostic indicators for COVID-19 severity.
• Machine learning models can effectively predict mortality and ICU admission risk using these identified factors.
• These findings can aid in characterizing disease severity and guiding clinical management of COVID-19 patients.