Aortic stenosis is characterized by symptoms and adverse events secondary to left ventricular decompensation and valvular obstruction, and it is currently graded based on valve assessment and diagnostic uncertainty.
Echocardiography (ECHO) measurements (ECHO cohort, n = 1,052) were used for:
- Patient analysis
- Stratify high-severity and low-severity phenogroups of AS.
A supervised machine-learning classifier is developed, and the performance is validated with independent markers of disease severity obtained using (CT) computed tomography (CT cohort, n = 752) and cardiovascular magnetic resonance imaging (CMR cohort, n = 160).
The prognostic value was further validated using clinical outcomes (aortic valve replacement [AVR] and death) in the ECHO and CMR cohorts. One-third of the 1,964 patients in the three multi-institutional cohorts (1346 people) had discordant or nonsevere AS severity. According to machine learning, 1,117 patients (57%) have high-severity AS, while 847 patients (43%) have low-severity aortic stenosis. Compared to low severity, patients with high severity in CT and CMR cohorts had left ventricular mass, higher valve calcium scores, and fibrosis.
Among patients who did not receive AVR in the ECHO cohort, progression to AVR and death were quicker in the high-severity group. Machine-learning–based severity classification has improved discrimination (net discrimination improvement: 0.07; 95% confidence interval: 0.02 to 0.12) and reclassification (net discrimination improvement: 0.17; 95% confidence interval: 0.11 to 0.23) for AVR outcome at 5 years. Researchers showed the prognostic value of machine-learning by using ECHO and CMR cohorts and classifications for AS subgroups along with asymptomatic, non-severe, or discordant aortic stenosis.
This study shows us that machine learning can integrate ECHO measurements to enhance the classification of disease severity in most patients with AS, which has the potential to maximize the timing of AVR.