PROJECT SUMMARY/ABSTRACT Atrial fibrillation (AF) affects about 10% of older adults and accounts for a growing proportion of strokes. Lifelong oral anticoagulation, with warfarin or a non-vitamin K antagonist oral anticoagulant (NOAC), is recommended for stroke prevention in most AF patients. However, the drugs increase the risk of bleeding and the adherence to the lifelong drug therapy is poor, leaving many patients under-treated. The recently-approved Watchman device offers an attractive alternative to lifelong drug therapy for AF stroke prevention. However, the device has been studied in only two clinical trials, both of which compared it to warfarin. Little is known about how Watchman compares with the current mainstay therapy, NOACs, or how Watchman compares with no treatment in patients who have difficulties taking anticoagulation drugs. Furthermore, for preventive treatment with a high upfront cost and some procedural risks, a personalized approach is needed to target Watchman to patients who are most likely to benefit and avoid it in those who have little to gain. Therefore, the overall objective of this project is to address these evidence gaps and develop new prediction tools to optimize the use of Watchman for AF stroke prevention. In Aim 1, we will conduct comparative effectiveness studies using a large national administrative database (OptumLabs) that contains insurance claims for over two million patients with AF of all ages and races from all 50 states with linked EHRs in a subset. The findings will provide timely evidence to address the key unanswered questions highlighted in current practice guidelines and will facilitate the design, analysis, and interpretation of future clinical trials. In Aim 2, we will develop and validate machine-learning models to predict how Watchman compares with non-invasive therapies. We will develop the models using the OptumLabs data, and validate the models in two RCTs and two large health systems? EHRs, thereby validating models in both clinical trial and routine practice settings. The new prediction models will provide personalized estimates for the benefits and harms, and thus, engage patients in making informed choices consistent with their preferences and ease clinicians? cognitive burden. In Aim 3, we will assess how the Watchman decisions made in contemporary practice agree with those suggested by the new prediction models. We will use machine-learning methods to identify patient and provider characteristics associated with incongruent decisions. Such findings will highlight patient and provider groups who may particularly benefit from the decision support, thereby informing future implementation and translation efforts. We have assembled a team with complementary clinical and research expertise, a solid record of successful collaboration, and extensive experience in outcomes research and prediction modeling. We also have developed a web-based decision aid that is ready to translate the prediction models to reduce unwarranted variation in care delivery, patient outcomes, and medical costs.
Explore the research topics touched on by this project. These labels are generated based on the underlying awards/grants. Together they form a unique fingerprint.