Enhancing stroke risk stratification in atrial fibrillation through non‐Newtonian blood modelling and Gaussian process emulation

Abstract

Atrial fibrillation (AF) is the most common heart arrhythmia, linked to a five-fold increase in stroke risk. The left atrial appendage (LAA), prone to blood stasis, is a common thrombus formation site in AF patients. The LAA can be classified into four morphologies: broccoli, cactus, chicken wing and windsock. Stroke risk prediction in AF typically relies on demographic characteristics and comorbidities, often overlooking blood flow dynamics. We developed patient-specific non-Newtonian models of blood flow, dependent on fibrinogen and haematocrit, to predict changes in LAA viscosity, aiming to predict stroke in AF patients. We conducted 480 computational fluid dynamics (CFD) simulations using the non-Newtonian model across the four LAA morphologies for four virtual patient cohorts: AF + Covid-19, AF + pathological fibrinogen, AF + normal fibrinogen, and healthy controls. Gaussian process emulators (GPEs) were trained on this in silico cohort to predict average LAA viscosity at near-zero computational cost. GPEs demonstrated high accuracy in AF cohorts but lower accuracy when the chicken wing GPE was applied to other morphologies. Global sensitivity analysis showed fibrinogen significantly influenced blood viscosity in all AF cohorts. The chicken wing morphology exhibited the highest viscosity, while the AF + Covid-19 cohort had the highest viscosity. Our non-Newtonian model in CFD simulations confirmed fibrinogen's substantial impact on blood viscosity at low shear rates in the LAA, suggesting that combining blood values and geometric parameters of the LAA into GPE training could enhance stroke risk stratification accuracy. KEY POINTS: Fibrinogen has a significant effect on blood viscosity in the left atrial appendage (LAA) at low shear rates. Gaussian process emulators (GPEs) can predict the viscosity of blood in the LAA at near-zero computational cost. Out of all LAA morphologies, the chicken wing morphology exhibited the highest average blood viscosity. High average blood viscosity in the LAA of atrial fibrilation + Covid-19 patients was observed due to high fibrinogen levels in this cohort. Combining blood values and geometric parameters of the LAA into GPE training could enhance stroke risk stratification accuracy.