New research shows machine learning is useful in enhancing risk stratification for patients with aortic regurgitation (AR), including the description of 4 distinct phenoclusters within this patient population. Mann Malahfji, MD, of the Houston Methodist DeBakey Heart and Vascular Center, reported these results in a manuscript published online in JACC: Cardiovascular Imaging. “This study represents a significant advancement in the application of precision medicine to chronic aortic regurgitation,” Hector Garcia-Garcia, MD, PhD, of MedStar Washington Hospital Center, Washington, DC, told CRTonline. “The integration of machine learning and cardiac magnetic resonance imaging enables deep phenotyping and yields clinically meaningful clusters with real prognostic value surpassing conventional risk stratification approaches. It is a technically sophisticated method that reveals complex patterns in an objective and reproducible way.” Patients with AR have varying symptoms, rates of symptom development, disease progression timelines and management strategies, but treatment plans often assume a population with similar characteristics. The investigators in this study used unsupervised machine learning to examine phenoclusters in patients with AR, as well as evaluate their impact on prognosis. Characterization of moderate or severe AR was performed using clinical techniques and cardiac magnetic resonance (CMR) across 4 centers in the U.S., where 2 centers were used for derivation of the phenoclusters and the other 2 centers were used for validation. All-cause death was the primary outcome of this study. Partition Around Medoids was the unsupervised clustering pipeline that used 23 clinical and CMR variables to create patient clusters in this group, independent of the outcomes. A total of 972 patients were included in the study (mean age=62 years, 78% male), and 330 patients underwent valve surgery, while 585 patients underwent medical surveillence. The median follow-up was 2.58 years, with a mortality rate of 12%. Within this group of patients, four clusters were derived by machine learning: 1) Younger males with a majority of the bicuspid aortic valve and high extent of left ventricular remodeling (1% mortality); 2) Older males with mostly tricuspid valves and intermediate outcomes (10% mortality); 3) Older males with the highest rates of comorbidities, LV scarring and dysfunction (22% mortality); and 4) Females with higher rates of mortality and burdening symptoms, lower rates of LV remodeling and aortic valve replacement (20% mortality). After adjustment for time-dependent aortic valve replacement and traditional risk factors in AR patients, the algorithm for clustering was independently associated with survival (C statistic 0.77 vs 0.75, p=0.009 in the validation cohort). Overall, 4 unique phenoclusters were discovered within the AR patient population in this cohort using a machine learning approach. "We must be cautious of a potential paradox: in the effort to personalize treatment, we may inadvertently reduce the patient to a data point within a population or cluster, losing sight of their individuality,” said Agostina Sanchez, from MedStar Washington Hospital Center, in an interview with CRTonline. “These models are powerful tools, but they must be complemented by clinical reasoning and a personalized understanding of each patient.” Source: Malahfji, M, Tan X, Kaolawanich Y, et al. Machine learning identification of patient phenoclusters in aortic regurgitation. JACC: Cardiovasc Imaging. 2025 April 16 (Article in press). Image Credit: ArtemisDiana – stock.adobe.com