Calculating In-patient Ventricular Tachycardia Burden and Assessing its Impact on Outcomes

About This Grant

PROJECT SUMMARY/ABSTRACT This is an R21 resubmission that will leverage an existing robust dataset, the University of California San Francisco (UCSF) Ventricular Tachycardia (VT) Database. VT is a potentially lethal arrythmia that occurs in 2-12% of hospitalized patients and is also associated with a 40% increased rate of in-hospital mortality among intensive care unit (ICU) patients. Prior work from our group has shown that there are large variations in VT features which can potentially be used to create a novel metric for “VT burden.” For example, VT can occur only once during hospitalization but can also occur multiple times and number of VT events does not necessarily correlate with a greater risk of mortality. In addition, our prior work has shown that duration and heart rate are important VT features associated with a clinical intervention(s). We hypothesize that, by characterizing these and additional VT features of “VT burden,” we can add precision to current hospital-based ECG monitoring systems and identify patients at greatest risk of poor outcomes. Our transdisciplinary research team included experts in the UCSF Division of Hospital Medicine, School of Nursing, and UCSF Division of Cardiology’s Center for Biosignal Research. The proposed study leverages the existing UCSF VT Database which includes expertly annotated VT (true versus false) from continuously recorded electrocardiographic (ECG) and clinical data for 5,679 consecutively admitted ICU patients with 572,574 hours of physiologic monitoring, which to our knowledge represents the single largest human annotated database to date. This rich dataset was previously used to develop and test a new VT algorithm (UCSF VT algorithm) that improves on poorly performing (90% false positive) existing VT algorithms used in modern-day monitors. The expertly annotated UCSF VT Database includes 14,304 true VTs among 660 ICU patients and detailed electronic health record data that will be used to define VT burden. In Aim 1 of the proposed resubmission, we will use ECG characteristics to develop several features to describe VT burden. In Aim 2, we will use causal inference and machine learning methods to determine the association between these VT burden features and patient outcomes, including incidence of atrial fibrillation, incidence of stroke, heart failure exacerbations, cardiogenic shock, in-hospital cardiac arrest (IHCA), ICU mortality, length of stay (ICU and hospital), and in-hospital mortality. Finally, in Aim 3, we will determine whether there are specific subgroups where VT burden has differential effects on patient outcomes. Our long-term goal is to apply our VT burden features to a new cohort of patients and determine whether they can predict VTs that result in a clinical action(s) (e.g., starting/changing medications, replacing electrolytes, or defibrillation for IHCA), and identify high risk patient who might benefit from pro-active therapies. Ultimately, we will test whether our VT burden features can be displayed on bedside monitors to provide clinicians with greater knowledge regarding their patients’ VT trajectory, significantly impacting in-hospital-based clinical care.