English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools annual

Global: Zendy Free Plan

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Funding Acknowledgements EU Horizon 2020 (642676-Cardiofunxion); Spanish Ministry of Economy and Competitiveness (TIN2014-52923-R); Maria de Maeztu Programme (MDM-2015-0502) Background Although clinical guidelines provide valuable help in the management of aortic stenosis (AS), uncertainty remains regarding their strict application in the assessment of stenosis severity, prognosis, and indication for valve intervention, in particular in low-gradient or asymptomatic AS. Evidence regarding the threshold values for aortic valve area (AVA), peak transvalvular velocity (Vmax), and mean transaortic pressure gradient (ΔPm) remains discordant. Interpretable machine learning (ML) approaches have the potential to generate guideline recommendations directly based on (location-specific) data. Purpose To evaluate the use of interpretable ML for risk stratification of AS, in particular to assess the expected improvement with aortic valve intervention (AVI). Methods We conducted a retrospective analysis of 2761 patients, enrolled in a single-centre AS registry between 2000 and 2016. A decision tree (DT) was automatically learned to predict natural-history mortality from cardiovascular causes within 6 years of inclusion, using baseline measurements at the time of inclusion as predictors. The DT was subsequently used to stratify all individuals of the registry. For each subgroup within the DT, the survival of patients that were and were not referred for AVI within 6 months of inclusion (AVI and no-AVI, respectively) was compared. Results The top predictors of 6-year survival were AVA, Vmax, age, posterior wall thickness, and creatinine (see figure). In subgroups 2 and 3, intervention correlated with increased survival. These patients are characterized by AVA &amp;lt; 0.9 cm2 and Vmax ≥ 325.0 cm/s, in agreement with the recommendations for intervention in high-gradient AS. However, there were several subgroups for which performing AVI was not associated with longer survival: • Subgroup 1: AVA &amp;lt; 0.9 cm2 and Vmax &amp;lt; 325.0 cm/s • Subgroup 4: AVA ≥ 0.9 cm2, age &amp;lt; 73 years and creatinine &amp;lt; 1.6 μmol/L • Subgroup 8: AVA ≥ 0.9 cm2, age between 73 and 88 years, creatinine &amp;lt; 1.6 μmol/L, hypertension, and ΔPm ≥ 19 mmHg • Subgroup 9: AVA ≥ 0.9 cm2, age &amp;lt; 88 years and creatinine ≥ 1.6 μmol/L Subgroup 1 corresponds to a discordant grading pattern, characterized by physiological velocities despite small AVA. The results for subgroups 8 and 9 suggest a negative association of AVI with renal dysfunction and hypertension, respectively. Conclusions The AVA cut-off value learned by the DT (0.9 cm2) is consistent with current evidence and identifies patients with high benefit from an intervention. Nonetheless, the data suggest that AVI may not improve the survival of certain AS subgroups, specifically in the presence of a discordant grading pattern, or in association with renal dysfunction or hypertension. The analysis demonstrates the ability of ML approaches to build explainable recommendations grounded on data, and which could be used to easily incorporate local information in addition to the evidence used to formulate guidelines. Abstract 622 Figure. Decision tree aortic stenosis survival

622 Predicting post-intervention survival improvement in aortic stenosis through interpretable machine learning