Unsupervised trajectory-based clustering identifies a rapid progression phenotype in patients with aortic stenosis

Background Aortic stenosis (AS) is the most prevalent valvular heart disease in developed countries and demonstrates substantial inter-individual variability in its rate of progression. Current follow-up strategies rely on uniform echocardiographic follow-up intervals, reflecting an incomplete understanding of underlying patient-specific disease trajectories. Identification of distinct progression phenotypes may enable more precise risk stratification and inform optimal timing of intervention. Methods Patients with moderate or less AS who underwent multiple serial echocardiographic examinations were consecutively included. Growth mixture modeling was applied as an unsupervised trajectory-based approach using baseline and longitudinal transaortic peak velocity measurements. The optimal number of clusters was determined based on Bayesian information criteria and classification entropy. Results A total of 210 patients (age 74 ± 13 years, 53% female) with 520 echocardiographic examinations were analyzed. Three distinct phenotypes of AS progression were identified: a mild AS with slow progression (Group 1, n = 121, progression rate 0.15 ± 0.02 m/s/year), a moderate AS with slow progression (Group 2, n = 72, 0.17 ± 0.02 m/s/year), and a rapid progression phenotype (Group 3, n = 17, 0.84 ± 0.08 m/s/year; Figure A). Groups 1 and 2 demonstrated similar progression slopes but differed in baseline severity. In contrast, Group 3 exhibited a markedly steeper slope despite a wide distribution of baseline transaortic peak velocities. Baseline aortic valve area and mean pressure gradient were comparable between Groups 2 and 3. Although age and sex did not differ significantly among the three groups, Group 3 demonstrated a more concentric LV geometry with significantly higher relative wall thickness (Groups 1 vs 2 vs 3; 0.40±0.08 vs 0.44±0.09 vs 0.38±0.09, p = 0.037) and tended to have smaller left ventricular dimensions and lower stroke volume index. Importantly, Group 3 exhibited a trend toward higher valvulo-arterial impedance (ZVa, Group 1 vs 2 vs 3; 3.2±0.9 vs 3.3±1.1 vs 3.8±1.3, p=0.139), and demonstrated higher aortic valve resistance (1.47±0.54 vs 2.06±0.70 vs 2.09±0.70 dyne·s·cm⁻⁵, p<0.001), indicating an increased functional valvular load. Scatter plot analysis confirmed that Group 3 was driven primarily by slope rather than baseline severity. Consistently, only a weak correlation was observed between baseline velocity and progression rate (r = 0.19, p = 0.006, Figure B). Conclusions Unsupervised trajectory-based clustering identified a rapid progression phenotype not predictable from baseline severity and characterized by increased functional valve impedance. Given the substantial heterogeneity in AS progression, our findings support a shift from uniform follow-up strategies toward more individualized surveillance.Figure A.trajectory-based phenotypingFigure B.trajectory-based phenotyping