Random Survival Forests achieved the best predictive performance for survival in patients with moderate or severe tricuspid regurgitation, with a C-index of 78% and AUC of 82%.
Cohort (n=949)
Do machine learning-based survival models accurately predict mortality in patients with moderate or severe tricuspid regurgitation?
Machine learning models, particularly Random Survival Forests, using clinical and CMR features can effectively predict survival and identify high-risk patients with moderate or severe tricuspid regurgitation.
Absolute Event Rate: 78% vs 66%
Background: Tricuspid regurgitation (TR) is a common valvular heart condition associated with significantly increased mortality. It is often underdiagnosed and undertreated due to limited insight into patient-specific risk prediction and optimal timing of intervention. Machine learning (ML) methods offer the potential to address these gaps by identifying high-risk patients, estimating survival probabilities, and uncovering key risk markers that influence outcomes. Methods: We developed and evaluated models to predict survival curves for a cohort of 949 patients with moderate or severe TR. Three modeling approaches were compared: Cox proportional hazards (Cox PH), Random Survival Forests (RSF), and DeepSurv (a deep learning-based survival model). Models were trained on clinical and imaging features extracted from cardiac magnetic resonance (CMR) studies and patient records. Performance was assessed using the concordance index (C-index) and time-dependent area under the receiver operating characteristic curve (AUC). Kaplan–Meier analysis and multivariable Cox regression were used to identify significant predictors of mortality. Results: RSF achieved the best predictive performance with a C-index of 78% and AUC of 82%, followed by DeepSurv (C-index 72%, AUC 78%) and Cox PH (C-index 66%, AUC 76%). Predicted survival curves for low- and high-risk groups demonstrated clear separation, underscoring the models’ ability to distinguish patient risk. Key predictors of poor survival included older age, tobacco exposure, right ventricular dilation and hypertrophy, right atrial enlargement, and the presence of non-ischemic myocardial fibrosis. These features were independently associated with elevated mortality risk and showed distinct survival differences in Kaplan–Meier analysis. Conclusions: Machine learning-based survival models, particularly RSF and DeepSurv, offer beneficial tools for individualized risk stratification in patients with advanced TR. Structural abnormalities of the right heart and myocardial fibrosis were among the most significant predictors of mortality, highlighting the importance of early detection and timely intervention. Integrating AI-driven survival prediction into clinical workflows could potentially benefit decision-making and enable more personalized management of TR.
Janghorbani et al. (Sun,) conducted a cohort in moderate or severe tricuspid regurgitation (n=949). Machine learning-based survival models (Random Survival Forests and DeepSurv) vs. Cox proportional hazards was evaluated on predictive performance for survival (C-index). Random Survival Forests achieved the best predictive performance for survival in patients with moderate or severe tricuspid regurgitation, with a C-index of 78% and AUC of 82%.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: