Machine learning modelling for cavotricuspid isthmus length unveils impaired atrial relaxation in cardioverted individuals
A Mircea, J Solana-Munoz, K Bode, C Jahnke, P Antiochos, A Masi, A Luca, J Scwitter, E Pruvot, I PaetschAbstract
Background
Patients with typical right atrial flutter in whom electrical cardioversion (EC) was performed prior to an ablation carried out in an interventional MRI (iMRI) environment show an altered cavotriscuspid isthmus (CTI) contraction pattern compared to that of patients without EC. Our previous study identified a clinical profile linked to reduced CTI contractility, particularly among patients with a history of EC, dyslipidemia and smoking.
Purpose
We aimed to compare and characterize the biomechanical properties of the CTI in EC and non-EC patients using machine learning regression models, while integrating the significant clinical variables associated with reduced CTI contractility.
Methods
32 patients (66±10 years, 17 smokers, 17 dyslipidemic, 15 cardioverted) treated within an iMRI for atrial flutter from a German center formed our analysis cohort. The following CTI metrics were collected: (1) CTI systolic and diastolic curvilinear lengths, (2) pouch depth and tricuspid valve (TV) - Eustachian valve (EV) distance in diastole, (3) normalized elongation (NE) defined as the diastolic minus the systolic CTI curvilinear length divided by the systolic curvilinear length. The CTI metrics and clinical variables (EC, smoking status, dyslipidemia) significantly associated with NE made up the features for predicting the CTI systolic and diastolic curvilinear length using the Lasso regression model. The association between EC and CTI metrics was assessed using point biserial tests.
Results
Lasso regression showed that the straight-line distance in diastole between TV - EV and pouch depth were sufficient for predicting CTI curvilinear diastolic length. The coefficients for CTI systolic length, EC, smoking status and dyslipidemia had no predictive value. In contrast, adding EC and smoking status improved model performance when predicting CTI curvilinear systolic length. The mean absolute error (MAE) decreased from 5.63 mm to 4.86 mm, and R² increased from 0.53 to 0.621. The point biserial tests run between EC status and the four CTI metrics were significant for pouch depth and CTI systolic length (r = -0.37, p = 0.03; r = 0.42, p= 0.01, respectively). EC individuals are more likely to display longer CTI systolic length and shallower CTI during diastole (Figure 1). Pathophysiologically, the EC-induced stunned myocardium might exhibit impaired metabolism. The interplay between the metabolically active phase (diastole) and a negative Bowditch effect, might result in higher CTI systolic length and shorter pouch depths for EC individuals (Figure 2).
Conclusion
Electrically cardioverted patients are more likely to display impaired right atrial diastole as reflected by a longer CTI systolic length and shallower CTI in diastole.