Abstract
Abstract 12895: Machine Learning Discovery of the Patterns and Impact of Recurrent Bleeding in Patients With Left Ventricular Assist Devices
Circulation (New York, N.Y.), v 140(Suppl_1 Suppl 1), pp A12895-A12895
19 Nov 2019
Abstract
BackgroundBleeding abnormalities after left ventricular assist device (LVAD) implantation are a common adverse event (AE) but typically do not occur as events. Our goal was to identify the sequences of associated AEs and predictors of outcome in LVAD patients with recurrent bleeding.MethodsA combination of machine learning (ML) and traditional statistical approaches were used to define a patient cluster with predominant bleeding AEs recorded in INTERMACS (2006 to 2015). Hierarchical clustering with ward linkages identified the cluster of patients with recurrent bleeding AEs (RBC). Markov modeling was used to order and illustrate the AE patterns associated with repeated bleeding. Multivariable Cox regression determined risk-adjusted predictors of death and transplantation in patients within the RBC.Results863 patients with 5,912 AE’s defined the RBC. There were 2,908 other AEs associated with recurrent bleeding, the most common being infection (48.7%, n=1,415) and cardiac arrhythmias (13.5%, n=394). The most significant pattern of AEs associated with recurrent bleeding were chronic bleeding (defined > 5 bleeding AEs) and infection (Figure). Bleeding was commonly of mucosal origin (93.9%, n=3,612) while mediastinal bleeding was rare (6.1%, n=236) (p<0.0001). The median number of bleeding events was 9 [IQR 6, 12]. Device explant occurred in 14% (n=121) and transplantation in 7.0% (n=60). Median survival in patients with recurrent bleeding was 2.4 years [IQR 1.3, 3.7]. Chronic bleeding within the RBC was an independent predictor of death (HR 2.2, 95% CI 1.6, 3.1, p<0.0001), but not transplantation (p=0.157).ConclusionsML techniques helped to identify and evaluate patients at highest risk of recurrent bleeding following LVAD implantation. In LVAD patients with recurrent bleeding, more than 5 bleeding events was an independent risk factor for death.
Metrics
3 Record Views
Details
- Title
- Abstract 12895: Machine Learning Discovery of the Patterns and Impact of Recurrent Bleeding in Patients With Left Ventricular Assist Devices
- Creators
- Laura Seese - University of PittsburghFaezeh Movahedi - University of PittsburghSarah Burki - University of PittsburghJames Antaki - Cornell UniversityArman Kilic - University of PittsburghManreet Kanwar - Cardiology, Allegheny General Hosp, Pittsburgh, PAChristopher Sciortino - University of PittsburghRobert Kormos - University of Pittsburgh
- Publication Details
- Circulation (New York, N.Y.), v 140(Suppl_1 Suppl 1), pp A12895-A12895
- Publisher
- by the American College of Cardiology Foundation and the American Heart Association, Inc
- Resource Type
- Abstract
- Language
- English
- Academic Unit
- Cardiology
- Other Identifier
- 991021932199204721