Predicting procedural success in patients treated with Cardioband system for severe tricuspid regurgitation by employing a random forest algorithm

Abstract Background Severe tricuspid regurgitation (TR) is associated with high morbidity and mortality despite optimal medical treatment. Transcatheter tricuspid valve intervention (TTVI) is therefore emerging as a novel treatment option, fueling the hope to prolong survival and reduce rehospitalization for heart failure. Obviously, procedural success of TTVI is an important determinant of survival, but predictors for procedural success in patients treated with Cardioband system, which mimics the surgical approach by implanting an annular reduction system and hence targets tricuspid annulus dilatation as the central pathology in most patients, are largely elusive. Purpose This study aims to refine prediction of procedural success in patients with severe TR undergoing TTVI with Cardioband system by employing a random forest algorithm. Methods Procedural success was evaluated in 72 patients enrolled at two tertiary centers in Germany between 2018 and 2020. Key inclusion criterion was TR ≥ III/V° with high symptomatic burden despite optimal medical treatment. Procedural success war defined as patient alive at the end of the procedure, successful Cardioband implantation, and TR reduction ≥ II/V° as assessed on transthoracic echocardiography before discharge. Since 66.7% of patients were classified as “success”, a synthetic minority over-sampling technique was applied in order to train the random forest algorithm on a balanced data set. Results A random forest algorithm reached 85.4% accuracy (AUC: 0.923) in predicting procedural success in a balanced data set using eight parameters from pre-procedural echocardiography as input variables. Partial dependence analysis revealed that enlargement of the tricuspid valve (TV) anteroseptal diameter was most important for model accuracy. Applied to the real-world data set (24 patients classified as “failure” and 48 patients classified as “success”), the now trained random forest algorithm predicted procedural success with high sensitivity (70.8%) and specificity (100.0%), significantly outperforming the no information rate (p-value: 0.0069). Patients with low probability for success were characterized by impaired right ventricular function (TAPSE: 15.5±3.63 mm) and enlarged right sided cardiac diameters (basal right ventricular diameter: 51.6±3.79 mm; TV anteroseptal diameter: 45.0±5.10 mm). Notably, systolic pulmonary artery pressure (sPAP) and TV effective regurgitant orifice area were negatively correlated (R: −0.3004, p-value: 0.0322), and elevation in sPAP was attenuated in patients with low probability for procedural success (sPAP: 34.0±11.7 mmHg). Conclusion A random forest algorithm enables precise prediction of procedural success in patients treated with Cardioband system. TR reduction ≥ II/V° appears less achievable in patients with advanced stages of right heart failure, emphasizing the importance of adequate patient selection and timing of intervention. Funding Acknowledgement Type of funding sources: None.