Improving Safety of Cardiovascular Implantable Electronic Devices in Veterans

Background: This proposal is intended to support the career development of Sanket Dhruva, MD, MHS, a Staff Cardiologist at the San Francisco VA and Assistant Professor of Medicine at the University of California, San Francisco into an independent VA health services researcher with the training and experience necessary

to conduct innovative research and develop interventions that improve safety of Veterans with cardiovascular implantable electronic devices (CIEDs: pacemakers and implantable cardioverter defibrillators [ICDs]). Even though more than 10% of the 55,000 Veterans followed by VA have suffered CIED-related complications, there

has not been any systematic evaluation to identify failed CIED leads using VA’s data systems. Significance/Impact: This research will close Dr. Dhruva’s knowledge gaps in biostatistics, data science, and qualitative methods, enabling him to generate actionable, high-quality evidence to inform VA cardiac

electrophysiologists to implant the safest devices in Veterans. This research will also enable him to identify CIED leads that have already been implanted in Veterans but are at risk for failure, thereby informing strategies to avoid clinical sequelae of failure (such as inappropriate shocks and death) for individual Veterans.

This proposal is directly aligned with operational priorities set forth in VHA Directive 1189 (published in January 2020) to “monitor the safety of CIEDs,” HSR&D Priorities of a Learning Healthcare System and improving Veteran Quality of Care and Safety, and supports VHA’s priority of becoming a High-Reliability Organization.

Innovation: This research is innovative through its application of advanced statistical methods to leverage a comprehensive, longitudinal database of Veterans with CIEDs, the VA National Cardiac Device Surveillance Program (NCDSP), including temporally dense CIED-generated data, to address the large-scale, complex

problem of identifying CIED lead failure. Additionally, this research provides information about the unexplored question of physician selection of manufacturer and model of device to implant and the role of safety data. Specific Aims: Aim 1: To compare risk-adjusted failure rates of different cardiovascular implantable

electronic device (CIED) lead models among Veterans. H1: We will detect one or more CIED lead models with statistically and clinically significantly higher failure rates when compared to other leads of the same type (e.g. ICD lead when compared to all other ICD leads). Aim 2: To develop risk prediction models of all-cause CIED lead failure among Veterans by applying

supervised machine learning methods to repeated measures from CIED remote monitoring data. H2: Risk prediction models will detect lead failure with high discrimination (area under the curve [AUC] ≥0.85) and adequate calibration at 3 months and 12 months post-assessment. Aim 3: To conduct a pilot study to determine the effect of an academic detailing and audit and

feedback intervention on the specific CIED lead models implanted in Veterans. H3: Post-intervention, Veterans will more often be implanted with lead models associated with the lowest failure rates. Methodology: Aim 1 will use sequential propensity score-adjusted simulated prospective survival analyses

applied to a dataset of the NCDSP linked to VA’s Corporate Data Warehouse and Medicare data. Aim 2 will apply two supervised machine learning techniques, elastic net and random forests, to quarterly patient- generated data from CIEDs to create prediction models. Aim 3 will include qualitative interviews of cardiac

electrophysiologists about device selection and the development, implementation, and evaluation of an academic detailing and audit and feedback intervention for cardiac electrophysiologists in 3 VISNs. Implementation: This research will enable Dr. Dhruva to become an independent VA HSR&D investigator who

conducts research to improve outcomes for Veterans with CIEDs and those who will receive one in the future.