Author + information
- Wendy Tsang,
- Ivan S. Salgo,
- Mark Gajjar,
- Benjamin Freed,
- Lynn Weinert,
- Sandeep Nathan,
- Atman Shah and
- Roberto M. Lang
The Society of Thoracic Surgeons Predicted Risk of Mortality (STS) score was designed to predict operable patient mortality. Recently, it has been used to select inoperable patients for transcatheter aortic valve replacement. Statistical machine learning algorithms (MLA) improve the accuracy of predictive models. We determined if 6 different MLA combining STS and transthoracic echocardiographic (TTE) parameters would improve sensitivity of the STS score alone in predicting mortality in inoperable aortic stenosis (AS) patients.
Baseline demographic, STS scores, TTE parameters (E/E', left ventricular ejection fraction (LV EF), 4–chamber LV global longitudinal strain (GLS–4C)) and clinical outcomes at 2 years were obtained from 96 (36 male, 79+13 yrs) inoperable AS patients undergoing TTE. Six MLA (R, v2.15.1) with leave one out cross validation were used to assess the predictive strength of the STS score with/without the TTE parameters in predicting mortality.
One–year mortality was 46%. STS score alone had good sensitivity (Table). However, STS score with the 3 TTE parameters was superior in predicting mortality using the classification tree and adaptive boosting MLA.
In high–risk AS patients STS score is a strong predictor of mortality. However, MLA combining three easily obtained TTE parameters with the STS score is superior in predicting mortality in inoperable AS patients. Given current resources, this new score may improve risk stratification.
Poster Sessions, Expo North
Monday, March 11, 2013, 9:45 a.m.–10:30 a.m.
Session Title: Structural Heart Disease Intervention
Abstract Category: 49. TCT@ACC–i2: Aortic Valve Disease
Presentation Number: 2114–237
- 2013 American College of Cardiology Foundation