Is Personalized Computational Model of Atrial Fibrillation Really Personalized?

Introduction: Computational modeling-guided, personalized electrophysiology (EP) intervention for atrial fibrillation (AF) is an emerging paradigm of precision medicine. In the published models, advanced imaging and invasive mapping achieve personalization of cardiac anatomy. However, EP cellular personalization is less developed, and parameters are often assumed to be uniform across individual patients Hypothesis: Anatomical personalization alone is not sufficient to recapitulate the individual clinical features of AF in personalized models. Methods: For each of 57 patients (66±10 yr, 30 persistent) referred for catheter ablation of AF, we constructed a personalized 3-D model using pre-ablation CT, invasive mapping, and Courtemanche-Ramirez-Nattel atrial cell model. In the pre-ablation model, we performed virtual ramp pacing with cycle lengths at 200-120ms within the coronary sinus to induce AF. AF was defined as being inducible when it persists for >15s. Simulation was repeated for 5 different published sets of EP parameters of ionic currents. In total, >10k simulations were performed on an NVIDIA V100 GPU cluster, using the Lattice Boltzmann method with spatial resolution of 0.5mm and temporal resolution of 0.05ms, achieving a runtime of 24s per 1s of simulation ( Fig. A ). Results: While 100% AF inducibility was expected in the personalized pre-ablation models, AF inducibility varied between 27% and 96% depending on the set of ionic parameters used ( Fig. B ). The large variability of AF inducibility indicates that those models are highly sensitive to EP parameters, therefore anatomical personalization alone cannot adequately constrain the individual features of AF. Conclusions: The patient-specificity of the current paradigm of modeling-guided, personalized EP intervention is limited. There is an unmet clinical need for incorporation of personalized EP as well as anatomical parameters to achieve true precision medicine.