Pre-interventional transesophageal echocardiography as a reliable predictor of residual shunt following patent foramen ovale closure

Figures

• 

  Fig. 1. Morphological characterization of PFO through transesophageal echocardiography in patients with paradoxical embolism undergoing percutaneous PFO closure. A Sankey plot illustrating the association of residual shunt after 6 months and PFO shunt size at baseline. Persistence of inter-atrial shunt was significantly (p < 0.05) higher in case of large or moderate PFO size and did not occur in patients with small shunt size. B Proportion of patients with residual shunt (RS) after 6 months was significantly increased (p < 0.05) in case of PFO with atrial septum aneurysm (ASA). C Box plots demonstrate significantly (p < 0.05) increased length of septum primum at baseline in patients with RS. D Septum primum length was significantly (p < 0.05) increased in patients with RS. E Likewise, tunnel length of PFO was significantly (p < 0.05) increased in patients with RS when compared to those with successful PFO closure. F Representative echocardiographic image depicting the anatomical complexity of PFO associated with RS following transcatheter closure

• 

  Fig. 2. A mismatch between device size and anatomical characteristics is linked to a significant increase in residual shunt (RS) leak following PFO closure. A In patients with PFO receiving a closure device with right atrial (RA) disk size diameter greater than 25 mm, the occurrence of peri-device leak was significantly (p < 0.05) higher compared to those who received a smaller device. B The ratio between RA disk size and excursion of septum primum was significantly (p < 0.05) lowered in patients with RS in contrast to those with a successfully closed shunt. C Similarly, the ratio between device size and degree of shunt size at baseline was significantly (p < 0.05) decreased in patients with RS post PFO closure at six months follow-up. D Multivariable regression analysis identified occluder disk size, PFO shunt size, and excursion of septum primum as independent predictors of ineffective PFO closure. Odds ratio of the significant (p < 0.05) features is presented as median as well as the lower and upper quartile

• 

  Fig. 3. Machine learning analysis of echocardiographic data enhances diagnostic accuracy for predicting residual shunt following percutaneous PFO closure. A Unsupervised Uniform Manifold Approximation and Projection (UMAP) of all TEE data of PFO anatomy at baseline unveiled a minor separation between patients with RS and those with successful closure after six months. B Comparison of model performance between various machine learning algorithms to predict RS. Extreme gradient boosting (XGBoost) unveiled a superior (p < 0.0001) mean absolute error (MAE). C Receiver operating characteristic (ROC) analysis of XGBoost algorithm to estimate the likelihood of RS. The supervised cross-validated machine learning model demonstrated a high diagnostic accuracy based on area under the curve (AUC). The model was trained (yellow curve) in a random set of patients and validated (purple curve) using a tenfold cross-validation loop. D The transformed likelihood for predicted peri-device leak based on the XGBoost model of baseline echocardiographic data demonstrated a high sensitivity (p < 0.0001) in distinguishing between patients with RS (purple) and those with successful closure (yellow) after 6 months

• 

  Fig. 4. Morphological features of PFO are significantly interrelated and can predict outcomes after percutaneous PFO closure. A Most important variables predicting adverse RS based on echocardiographic data at baseline prior to intervention. Included parameters are sorted according to the relative importance of gain parameters and include device- and patient-related features. B SHapley Additive exPlanations (SHAP) values for individual features to predict RS. Feature importance is listed in descending order of mean values and colored accordingly, with dots representing patients with PFO. Positive SHAP values indicate an increased likelihood for RS, while negative SHAP values indicate a lower likelihood of the outcome. C Comprehensive correlation matrix synopsizing the interrelationship of baseline demographic characteristics, morphological features of PFO, occluder device, and outcome of patients undergoing transcatheter PFO closure. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. D Representative TEE image of RS following percutaneous intervention