Coagulation parameters predict COVID-19-related thrombosis in a neural network with a positive predictive value of 98

Area of Science:

• * Hematology and Thrombosis
• * Infectious Disease Pathogenesis
• * Artificial Intelligence in Medicine

Background:

• * Thrombosis is a significant complication of COVID-19, linked to alterations in coagulation factors.
• * Predicting COVID-19 thrombosis is challenging due to its multifactorial nature and the inadequacy of single hemostatic variables.
• * Understanding the coagulation system's role is crucial for managing COVID-19 complications.

Purpose of the Study:

• * To develop and validate a predictive model for COVID-19-related thrombosis.
• * To identify key hemostatic and laboratory variables for thrombosis prediction.
• * To improve early detection and management of thrombotic events in COVID-19 patients.

Main Methods:

• * Development of a neural network using hemostatic and laboratory data from 149 COVID-19 patients across two cohorts.
• * Validation of the neural network on patient data collected at hospital and ICU admission.
• * Analysis of variable importance for predicting thrombosis.

Main Results:

• * The neural network identified key predictors including C-reactive protein, sex, and various thrombin generation parameters.
• * The model demonstrated high predictive accuracy, achieving a positive predictive value of 98%-100% for COVID-19-related thrombosis.
• * The model effectively predicts thrombosis risk at the time of hospital admission.

Conclusions:

• * A validated neural network can accurately predict COVID-19-related thrombosis.
• * The model utilizes a combination of hemostatic and general laboratory variables for prediction.
• * This tool offers a promising approach for early identification and prevention of thrombosis in COVID-19 patients.