Implantable medical devices have revolutionized contemporary cardiovascular care, and are used in a wide spectrum of acute and chronic cardiovascular conditions. However, medical device design fault or incorrect use may lead to significant risk of patient injury and represents an important preventable public health risk in the United States. To help identify device-related safety issues, a strategy of active, prospective, post-market safety surveillance has been recommended by the FDA, and evaluated methodologically. This type of surveillance offers significant advantages over traditional adverse event reporting strategies. However, all such approaches are challenged by the need to incorporate learning effects into expectations regarding safety. These learning impacts been repeatedly shown to have dramatic impacts on outcomes during early device experience. Quantifying learning effects on the outcomes associated with high-risk cardiovascular devices will improve our understanding of intrinsic device performance, thereby identifying patient populations best treated with such devices while simultaneously providing necessary feedback to device manufacturers to support iterative improvement in device design. Separately, understanding the impacts of learning may identify opportunities for targeted training as well as help to tease apart institutional and operator characteristics that may accelerate the achievement of optimal outcomes in the use of the specific cardiovascular device. This proposal seeks to extend the previously validated, open-source, active, prospective device safety surveillance tool, by developing and validating robust learning curve (LC) detection and quantification algorithms, designed to simultaneously account for the effects at the operator and institutional levels. We propose a ?blinded? development strategy, in which one team will generate robust synthetic clinical data simulator with LC impacts, and the other team develops and applies LC detection and quantification algorithms, without knowledge of the underlying relationships, determine performance and accuracy through sequential refinement and validation steps. We propose to formally validate the optimized LC tools in real-world data through re-analysis of previously published LC effects on transcatheter valves and vascular closure devices using national cardiovascular registries. In addition, the LC tools will be incorporated into two active, prospective device safety surveillance studies of novel implantable cardiovascular devices using large clinical registries.
This proposal seeks to understand the impact of institutional and physician learning on the safety of newly approved cardiovascular devices, and to use this knowledge to support and improve effective medical device safety surveillance. We propose a ?blinded? strategy of separating simulated dataset generation from the learning effects detection and quantification algorithm development. Incorporating learning effects adjustment into a validated, prospective, near-real-time safety surveillance system, this research will improve public health by identifying poorer performing cardiovascular devices, and provide physicians, device manufacturers and public health officials with better information to optimize the use of medical devices, iteratively improve their design, and identify opportunities for enhanced training that will result in improved patient outcomes.
