Abstract Body:

Background: Valvular heart diseases progress heterogeneously, and clinicians rely on serial echocardiography to assess stability versus progression. Transitions in reported severity between examinations are common and can reflect physiologic variability, hemodynamic loading, or timing of follow-up.

Hypothesis: We hypothesize that much of the variability in echocardiographic valve-lesion severity arises from physiological and contextual factors rather than structural progression; regurgitant lesions show greater short-term variability than stenotic lesions; and inpatient examinations exhibit more dynamic grade transitions than outpatient studies.

Methods: We conducted a retrospective, patient-level analysis of the EchoNotes dataset derived from MIMIC-III, encompassing 45,794 echocardiography reports from 2001–2012. For each patient and lesion type (aortic stenosis (AS), aortic regurgitation (AR), mitral stenosis (MS), mitral regurgitation (MR)), successive examinations were used to model transitions among ordinal severity states (normal, mild, moderate, severe). A multinomial transition framework estimated the probability of the next-exam grade conditional on the current grade. Transitions were stratified by follow-up interval (<7, 7–30, 30–90, ≥90 days) and care setting (inpatient vs outpatient). Primary outcomes were grade persistence and transition probabilities by lesion type; secondary analyses assessed the influence of follow-up interval and care setting.

Results: Left-sided lesions were frequently evaluable (AR in 72.7% of studies, AS in 65.2%, MR in 67.8%, and MS in 37.0%). Persistence over short intervals was high (less than 30 days: 89–95%), whereas grade changes increased with longer follow-up (90 days or more: 10–18%). MR showed the greatest variability, with 42% of moderate cases changing grade beyond 90 days. AS was most stable, with more than 80% persisting at the same grade beyond three months. Inpatients had more apparent variability than outpatients (14.2% vs 9.8%, p<0.01); after standardizing for interval, differences largely attenuated except for severe AR, where inpatients were 1.4-fold more likely to improve.

Conclusions: The results revealed distinct progression phenotypes that extend risk stratification beyond a single grade assignment. These insights support precision surveillance intervals and may guide earlier intervention for high-risk phenotypes while informing development of trajectory-aware clinical decision tools.