Machine Learning-Based Prediction of In-Hospital Mortality in Severe COVID-19 Patients Using Hematological Markers.

Area of Science:

• Hematology
• Critical Care Medicine
• Machine Learning in Medicine

Background:

• Severe COVID-19 is associated with high mortality rates, necessitating early risk stratification.
• Identifying prognostic factors at hospital admission is crucial for timely intervention and management.
• Hematological parameters offer a readily available resource for predicting patient outcomes.

Purpose of the Study:

• To develop and validate a predictive model for mortality risk in severe COVID-19 patients.
• To utilize hematological parameters obtained at hospital admission for risk assessment.
• To identify key hematological predictors of mortality in severe COVID-19.

Main Methods:

• Retrospective collection of clinical and laboratory data from 396 (Cohort 1) and 112 (Cohort 2) severe COVID-19 patients.
• Feature selection using LASSO method and model development with nine machine learning algorithms, selecting the best performing Logistic Regression (LR) model.
• External validation of the developed model in Cohort 2, including nomogram visualization and subgroup analyses.

Main Results:

• The LR model demonstrated strong predictive performance with an AUC of 0.852 in Cohort 1.
• Key predictors identified include D-dimer, platelets, neutrophil count, lymphocyte count, and activated partial thromboplastin time.
• The model showed good validation in Cohort 2 (AUC = 0.841) and performed well across different age groups.

Conclusions:

• A validated risk prediction model for mortality in severe COVID-19 was successfully developed using only hematological parameters.
• The model, built with the LR algorithm, enables timely and accurate risk stratification and management of critically ill COVID-19 patients.
• This hematology-based model offers a practical tool for improving clinical decision-making in severe COVID-19 cases.