This stepped wedge randomized intervention will apply machine learning algorithms in an electronic health record system to identify primary care patients with non-valvular atrial fibrillation (AF) who are at high risk of stroke and not on anticoagulation therapy. An Anticoagulant Management Service (AMS) will offer support to primary care providers regarding treatment for relevant patients (either warfarin and novel oral anticoagulants).

This study seeks to:

1. increase the proportion of appropriately anticoagulated patients with AF,
2. understand the reasons for lack of anticoagulation, and
3. document the proportion of patients with AF who are appropriately not anticoagulated (e.g. patient refusal, contraindication).

Atrial fibrillation (AF) is the most common type of cardiac arrhythmia and is associated with significant mortality and morbidity from stroke, thromboembolism, and related cardiovascular conditions. While the risk of stroke for AF patients as a whole tends to be greater than the general population; within the AF patient population, the risk of stroke is modified by the presence or absence of additional risk factors such as age, comorbid conditions, and prior stroke history.

There is a wealth of evidence for the effectiveness of anticoagulation therapy to prevent stroke and thromboembolism, but while anticoagulants have been demonstrated to be highly effective at preventing stroke and embolic events among AF patients, they are also known to increase the risk of major bleeding events. Anticoagulation with warfarin and other VKA drugs can be complex to manage. These drugs have narrow therapeutic windows and require close monitoring to stay within the target international normalized ratio (INR). They also have many known food and drug interactions.

In the last few years, several novel oral anticoagulants (NOAC) such as dabigatran, rivaroxaban, and apixaban have entered the market. While each of the NOACs demonstrated non-inferiority to warfarin in a large randomized clinical trial prior to FDA approval, experience with NOACs is limited in practice.

Our intervention will combine the ability of health information technology to filter large volumes of data with human capacity to understand subtleties and barriers for complex clinical decision making. Our intervention will facilitate a connection between patients, treating clinicians, and an established Anticoagulant Management Service (AMS) for coordinated care. We will use information from the EHR to direct additional efforts and resources toward reaching potentially unrecognized or undertreated atrial fibrillation patients with the greatest need for preventive anticoagulation therapy and lowest risk of adverse effects. This electronic safety net will assist with efficient allocation of scarce resources beyond usual care. The proposed clinical decision support/care-coordination process will be designed to address many of these identified barriers to appropriate anticoagulation therapy among AF patients. Here we define "appropriate" anticoagulation as a guideline informed shared decision between individual patients and their care team.

This intervention utilizes a stepped wedge design involving 14 primary care clinics affiliated with the Brigham and Women's Hospital. The timing of clinic entry to the intervention arm will be randomized.

Inclusion Criteria:

• Non-valvular atrial fibrillation as identified by machine learning algorithms
• Primary care provider within Brigham and Women's Hospital
• No evidence of a prescription for an anticoagulant in medical record for at least 1 year

Exclusion Criteria:

-

There is no plan to share individual participant data outside of the staff and participants themselves.