CVD22

CVD22 analyzes relationships between troponin, D-Dimer, CK-MB and other biomarkers and predicts mortality outcomes in COVID-19 patients using explainable machine learning.

Key Features:

• Explainability: Uses SHAP (SHApley Additive exPlanations) to provide transparent, per-feature contributions to model predictions.
• Predictive performance: XGBoost achieved reported metrics on the study dataset with train accuracy 1.0, test accuracy 0.83, precision 0.86, F1-score 0.83, recall 0.80, and AUC 0.91.
• Algorithms evaluated: Five machine learning algorithms were applied and compared on the dataset from Erzurum Training and Research Hospital.
• Feature importance: SHAP-based analysis identified D-Dimer mean, mortality status, CK-MB levels, and glucose as top influential features.

Scientific Applications:

• Mortality risk stratification: Predicts patient mortality risk in COVID-19 cohorts based on biomarker profiles including troponin, D-Dimer, and CK-MB.
• Biomarker prioritization: Ranks biomarkers by influence to identify markers most associated with COVID-19 severity and prognosis.
• Clinical and public health decision support: Provides interpretable model outputs that can inform clinical prioritization and population-level risk assessment.

Methodology:

Five machine learning algorithms were applied to a dataset from Erzurum Training and Research Hospital; XGBoost was selected for best performance and SHAP values were used to determine feature importance, with reported model metrics as stated above.