CONCLUSION
The HCM-AF-Risk Model effectively identifies HCM patients with AF, in addition to predictors of AF in HCM. Our approach attains good performance (0.74 sensitivity, 0.72 specificity, C-index = 0.80) improving upon current AF risk scores from community cohorts, while addressing the imbalance between high-risk and low-risk cases that is inherent in most clinical data. The set of clinical attributes identified by our method as indicative of AF – and serving to justify the severity level assigned by the classifier, includes several hitherto un-identified markers of AF in HCM, and suggests that HCM patients with AF have a more severe cardiac HCM phenotype.
Funding: This work was funded in part by NSF IIS EAGER grant#1650851 (to HS), an award from the John Taylor Babbitt (JTB) foundation (Chatham, New Jersey) and startup funds from the UCSF Division of Cardiology (to MRA).