Abstract 11784: Machine Learning Analysis of Progression From Moderate to Severe Tricuspid Regurgitation

Abstract

Background: Tricuspid regurgitation (TR) is linked to poor outcomes and its progression is associated with higher mortality. Left heart disease, age and sex are factors related to TR progression. We assessed echocardiographic (Echo) and demographic predictors of moderate TR progression using Machine Learning Natural Language Processing (ML NLP). Methods: 54,315 patients were retrospectively assessed for moderate TR with an Echo between May 2017 to October 2020 and a follow-up Echo 30 days to 113 months. Exclusion criteria included prior pacemaker/ICD, tricuspid prosthesis or band/ring, history of Ebstein, tetralogy of Fallot and tricuspid endocarditis. Cardiac Intelligence TM (Mpirik, Milwaukee, WI) identified moderate TR, demographic and Echo indices using ML NLP of the Echo report. Our endpoint was progression from moderate to severe TR within 1 or 2 years. Univariate Cox proportional hazard model was used with univariate analysis. Results: Of 3,489 moderate TR patients, 1,564 had a subsequent Echo and 771 met study criteria. Table 1 shows baseline demographic and Echo data. 15% and 24% of moderate TR patients progressed to severe TR within 1 and 2 years, respectively. Black race, left ventricular size, moderate/severe right ventricular (RV) dysfunction and Doppler-based RV systolic pressure were found to be univariate predictors of progression (Figure 1). Conclusion: ML NLP based on input from Echo-derived parameters using a real-world sample can offer insights into risk stratification of moderate TR regarding risk of progression to severe disease. Including further clinical and demographic variables will allow to improve performance of a disease prediction ML model.