A territory-wide study of arrhythmogenic right ventricular cardiomyopathy patients from Hong Kong

Abstract

Abstract Background Arrhythmogenic right ventricular cardiomyopathy/dysplasia (ARVC/D) is a hereditary disease characterised by fibrofatty infiltration of the right ventricular myocardium that predisposes affected patients to malignant ventricular arrhythmias, dual-chamber cardiac failure and sudden cardiac death (SCD). Methods This was a territory-wide retrospective cohort study of patients diagnosed with ARVC/D between 1997 and 2019. The primary outcome was incident ventricular tachycardia/ventricular fibrillation (VT/VF). The secondary outcomes were new-onset heart failure with reduced ejection fraction (HFrEF) and all-cause mortality. Results This study consisted of 115 ARVC/D patients (median age: 60 [44.1–70.2] years; 58% male). Of these, 51 and 24 patients developed incident VT/VF and new-onset HFrEF, respectively. Five patients underwent cardiac transplantation, and 14 died during follow-up. Multivariate Cox regression identified prolonged QRS duration as a predictor of VT/VF (P<0.05). Female gender, prolonged QTc duration, the presence of epsilon waves and T-wave inversion (TWI) in any lead except aVR/V1 predicted new-onset HFrEF (P<0.05). Female gender, prolonged QTc duration and the presence of epsilon waves, in addition to the parameters of older age at diagnosis of ARVC/D, prolonged QRS duration and worsening ejection fraction predicted all-cause mortality (p<0.05). Clinical scores were also developed to predict new-onset HFrEF (Table 1a-c) and all-cause mortality (Table 2a-c). This was followed by the application of a non-parametric machine learning survival analysis models for outcome prediction. These machine learning algorithms better capture nonlinear and interactive patterns within survival data compared to traditionally used Cox regression models, which assume the existence of a hazard function between survival data and censored outcomes. The present study introduced weighted random survival forests models for the prediction of incident VT/VF, HFrEF and all-cause mortality. Findings indicate that these machine learning wRSF models performed the best in the prediction of all three aforementioned outcomes compared to other analytical methods. Conclusion Clinical and electrocardiographic parameters are important for assessing prognosis in ARVC/D patients. Machine learning algorithms appear to be the most optimal tools for event prediction, and as such should potentially be used to aid risk stratification and decision-making in the clinical setting. Funding Acknowledgement Type of funding sources: None.