De-identified data from Symphony Health, an ICON plc Company, PatientSource®, April 1, 2017 to October 31, 2020, were used. This open claims data source contains patient demographics, medical and procedure claims, and prescription drug claims, including the status of prescription drug claims (i.e., approved, rejected, abandoned). The open claims nature means that a patient’s healthcare activity is captured regardless of maintaining the same healthcare plan if the patient uses providers from the network that supplies data to the database. The database captures more than 75% of all US retail prescription claims, representing over three-quarters of the US population annually across multiple payer channels (i.e., commercial, Medicare, Medicaid); the remaining less than 25% of claims are outside of the covered networks.

The study had a retrospective longitudinal design (Fig. 1). The index date was the date of the first DOAC claim (i.e., apixaban, dabigatran, or rivaroxaban). The study focused on these three DOACs as the most commonly used in the USA [21]. Only patients with the status of the first claim being “approved” (i.e., submitted by a pharmacy and approved for payment by health plans after claims adjudication) were retained.

Study design. DOAC direct oral anticoagulant, SE systemic embolism

The baseline period was defined as the 6-month period of clinical activity before the index date, where clinical activity was based on the first and last patient-level activity flags (i.e., either a pharmacy or a medical claim).

The 90-day landmark period starting on the index date was used to classify patients into the “one-and-done” and “continuer” cohorts. The duration of the landmark period was based on a 60-day or longer gap in DOAC supply to define DOAC discontinuation after 30 or fewer days of the index DOAC claim supply. The gap of 60 days or longer was chosen since it has been commonly used in prior literature to define DOAC discontinuation in NVAF [22, 23], and patients with more than 30 days of supply for the index DOAC claim were excluded since it would not have been possible to classify them as “one-and-done” or “continuer” during the 90-day landmark period (see “Sample Selection”). A sensitivity analysis was conducted using a gap of at least 45 days to define discontinuation, reducing the landmark period length and associated survival bias (see “Sensitivity Analysis”).

The follow-up period used to measure the outcomes started on day 91 post-index and continued until the earliest of the end of clinical activity or data availability.

Patients with the first claim for apixaban, dabigatran, or rivaroxaban were included in the study if they met the following inclusion criteria: (1) at least 6 months of clinical activity before the index date; (2) no claims for other oral anticoagulants (i.e., betrixaban, edoxaban, warfarin) before the index date; (4) at least one claim with a diagnosis for AF (International Classification of Diseases, 10th Revision, Clinical Modification [ICD-10-CM] code I48) during the baseline period or on the index date; and (5) at least 18 years old on the index date.

Patients were excluded from the study if they had (1) at least one claim with any of the following diagnoses or procedures at baseline: mitral stenosis, mechanical heart-valve, venous thromboembolism (VTE), hip or knee replacement surgery, or organ or tissue replaced by transplant; (2) pregnancy during or after the baseline period; (3) at least one claim with a diagnosis of stroke (ischemic or hemorrhagic) or SE during an inpatient admission on the index date or within 30 days prior to the index date (to avoid confounding with the outcomes of ischemic stroke and SE); (4) more than one claim for DOACs (i.e., apixaban, betrixaban, dabigatran, edoxaban, rivaroxaban) on the index date; or (5) more than one final claim status (e.g., approved and abandoned) for the index DOAC on the index date.

Additionally, retaining only patients with the “approved” status of the first claim, the following exclusion criteria for the identification of cohorts during the landmark period were applied: (1) 90 or fewer days of clinical activity post-index (i.e., incomplete landmark period); (2) stroke or SE diagnosis in an inpatient setting during the landmark period; and (3) more than 30 days of supply on the index DOAC claim.

During the landmark period, patients were classified into the “one-and-done” cohort if they discontinued DOAC therapy after the first DOAC claim, i.e., did not have another approved DOAC claim during the landmark period (a 60-day or longer gap in DOAC supply). Patients were classified into the “one-and-done” cohort if they persisted on DOAC therapy beyond the first claim, i.e., had at least two approved DOAC claims during the landmark period.

The first instance of ischemic stroke (ICD-10-CM code I63) and SE (ICD-10-CM code I74) were measured in each cohort during the follow-up period as a composite outcome (i.e., ischemic stroke/SE) and separately. Only diagnoses in claims from an inpatient setting were considered.

To balance baseline characteristics between the one-and-done and continuer cohorts, inverse probability of treatment weighting (IPTW) was applied. The propensity score was computed from a logistic regression model and adjusted for demographic characteristics (i.e., age, sex, region of residence, insurance plan type, index year); comorbidities (Quan–Charlson Comorbidity Index [Quan-CCI] [24], CHA2DS2-VASc score, HAS-BLED score); recent ischemic or hemorrhagic stroke/SE (i.e., presence of event and time from the event to index date); polypharmacy (i.e., use of at least five different medications concurrently); use of antihypertensive agents, antihyperlipidemic agents, antiplatelet agents; the index DOAC medication; and all-cause pharmacy costs. The balance of baseline characteristics was assessed using standardized differences (less than 10% indicated balance) [25].

The probability of an ischemic stroke or SE event during the follow-up period was described between the one-and-done cohort and the continuer cohort using weighted Kaplan–Meier survival analysis. The risk of an ischemic stroke or SE event was compared between the two cohorts using weighted Cox proportional hazard models, with HRs and their 95% CIs and p values reported. Time to the ischemic stroke or SE event was defined as the time from day 91 post-index (i.e., first day following the landmark period) until the first ischemic stroke or SE event during the follow-up period; patients for whom the event was not observed during the follow-up period were censored at the end of the follow-up period.

A sensitivity analysis using a 75-day landmark period was conducted to minimize the loss of patients with early ischemic stroke and SE events (i.e., within the duration of the landmark period).

Data were de-identified and comply with the patient requirements of the Health Insurance Portability and Accountability Act (HIPAA) of 1996; therefore, no review by an institutional review board was required per Title 45 of CFR, Part 46.101(b)(4) [26].