Determination of Patient-Specific Blood Coagulation Kinetic Parameters via Neural Networks: Toward Thrombosis Prediction in Personalized Medicine.

Purpose: The solution of the system of equations that model the coagulation cascade enables the determination of thrombin production, which is related to blood clot formation and thrombosis. However, traditional models often overlook clinical and hematological variables due to modeling challenges or incomplete understanding. Mathematical models of blood coagulation cascade are typically generalist, presenting limited accuracy. This study aimed to incorporate patient-specific and hematological data into the kinetic parameters of the coagulation cascade to generate individualized thrombin curves and predict the recurrence of venous thromboembolism.

Methods: A sensitivity analysis identified the most influential kinetic parameters for thrombin production. These parameters were adjusted using a model hybrid combining an artificial neural network with a system of ordinary differential equations optimized via a genetic algorithm. The dataset is split into two subsets to prevent data leakage.

Results: Eight kinetic rates were identified as the most sensitive, particularly those related to factor V activation and thrombin-antithrombin III complex formation. Factors such as anticoagulant use, smoking, pulmonary embolism, and factor V Leiden mutation significantly impacted the kinetic parameters. The model presented an AUC of 0.9941 and an accuracy of 0.9872.

Conclusion: The influence of these input variables on the kinetic parameters and thrombin production aligned with their known effects as risk factors reported in the literature. Adjusting the kinetic parameters individualized the model response, providing a clear cutoff point for thrombosis classification based on thrombin production. With further validation, this model could assist in diagnosing and prognosticating thrombosis and identifying new therapeutic targets to regulate thrombin production.