Healthcare Resource Utilization among Non-Valvular Atrial Fibrillation Patients Who Switched from Warfarin to a Novel Oral Anti-Coagulant

Among patients with non-valvular atrial fibrillation (NVAF), switching from warfarin to novel oral anticoagulants (NOACs) is common, yet clarifying the differences in the effect of NOACs on all-cause healthcare resource utilization (HCRU) are unknown. Adult NVAF patients who switched from warfarin to dabigatran, apixaban, or rivaroxaban were identified in MarketScan databases between 10/2010-12/2015. Patients had 12 months pre-period (index date was 1st NOAC claim) and were followed up to 12 months until medication discontinuation, end of enrollment, inpatient death, or 12/2016. Overall, 8,679 and 5,761 dabigatran switchers were matched (1:1) to rivaroxaban and apixaban switchers (mean age 73-74 years). Compared with rivaroxaban switchers, a lower proportion of dabigatran switchers had an inpatient (IP) visit (20.0% vs. 21.6%, p=0.008). Dabigatran switchers had lower per-patient-per-month (PPPM) total outpatient (3.87 vs. 4.06, p=0.002), emergency department (ED; 0.48 vs. 0.52, p=0.026), outpatient office (1.17 vs. 1.22, p<0.001), and other outpatient (2.71 vs. 2.83, p=0.043) visits compared with rivaroxaban switchers. A similar proportion of dabigatran and apixaban switchers had an IP visit (20.7% vs. 21.2%); compared with apixaban switchers, dabigatran switchers had significantly more PPPM IP visits (0.23 vs. 0.21, p=0.031) but significantly lower ED visits (0.47 vs. 0.52, p=0.016). Post-discharge 30-day readmission rates were comparable among warfarin-to-NOAC switching groups. Time to readmission was longer for dabigatran versus rivaroxaban switchers (8.2 vs. 7.8 days, p<0.001), but comparable with apixaban patients (8.1 vs. 8.4 days). Switching to dabigatran after warfarin discontinuation may lower HCRU among NVAF patients compared with switching to rivaroxaban or apixaban.


Introduction
Atrial fibrillation (AF) is estimated to affect 3 to 6 million people in the United States (US), with prevalence expected to double by 2030, accounting for approximately 15% of all strokes [1][2][3][4][5][6] . Patients with AF are four to five times more likely to have a stroke than patients without AF 2,3,7 . Non-valvular atrial fibrillation (NVAF), defined as AF in the absence of mitral stenosis or valvular prostheses, accounts for 95% of all AF cases in the US 2,8 .
Anticoagulation therapy has been the standard of care for reducing risks of thromboembolic and ischemic stroke events in patients with AF going back many years 9 . Antithrombotic treatment guidelines in the US, Canada, and Europe recommend chronic oral anticoagulation (OAC) for patients with AF, rather than no therapy, aspirin, or a combination therapy of aspirin with clopidogrel, especially for patients with an intermediate or high risk of stroke [9][10][11][12][13][14] . Warfarin had historically been the most commonly prescribed OAC since its approval in 1954; however, recent US Food and Drug Administration approvals have made available four mechanistically novel oral anticoagulants (NOACs), including the direct thrombin inhibitor dabigatran (approved October 2010), and factor Xa inhibitors rivaroxaban, apixaban, and edoxaban (approved in July 2011, October 2012, and January 2015, respectively) for prophylaxis of ischemic stroke and other complications 10,13 . To this end, the recently revised American Heart Association (AHA)/American College of Cardiology (ACC)/ Heart Rhythm Society (HRS) guidelines advocate the use of NOACs over warfarin 15 .
Since dabigatran's approval, several real-world clinical practice studies have evaluated the associated clinical outcomes, healthcare utilization, costs, and treatment patterns, mostly with warfarin as the comparator [16][17][18] . A study of the RE-LY trial comparing the effectiveness of dabigatran versus warfarin found dabigatran superior in reducing the risk of both stroke and systemic embolism 19 . While a large proportion of newly diagnosed AF patients initiate OAC therapy with warfarin, many of them may discontinue warfarin due to regular monitoring of International Normalized Ratio, interactions of warfarin with food, alcohol, and other drugs, complications, new comorbidities, and certain genetic variations that may predispose patients to adverse events 15,20,21 . To date, there is limited real-world data available assessing differences between NOACs for those who switched from warfarin. The primary objective of this study was to compare allcause healthcare resource utilization of warfarin-treated patients diagnosed with NVAF who switched to dabigatran, rivaroxaban, or apixaban.

Objectives
The objective of this study was to compare allcause healthcare resource utilization among patients diagnosed with NVAF who were treated with warfarin and subsequently switched to dabigatran, rivaroxaban, or apixaban in a head-to-head comparison using a large realworld data source.

Study design and data source
This retrospective analysis of patients diagnosed with AF compared healthcare resource utilization using real-world healthcare claims data for patients in the US with commercial and/or Medicare supplemental insurance coverage treated initially with warfarin and who subsequently switched to dabigatran, rivaroxaban, or apixaban. All study data was obtained from de-identified health plan enrollment records, inpatient and outpatient medical claims and outpatient prescription claims using International Classification of Diseases, 9th and 10th Revision, Clinical Modification (ICD-9-CM and ICD-10-CM) codes, Current Procedural Terminology 4th edition (CPT-4) codes, Healthcare Common Procedure Coding System (HCPCS) codes, and National Drug Codes (NDCs).
The data source was administrative healthcare claims data from the 2009-2016 MarketScan® Commercial Claims and Encounters (commercial) Database and Medicare Supplemental and Coordination of Benefit (Medicare) Database (IBM Watson Health, Cambridge, MA). These databases contain the complete longitudinal records of inpatient services, outpatient services, and prescription drug claims for commercially-insured and Medicareeligible patients covered under a variety of health plans, including dates of service, places of service, and all payments. All database records are de-identified and fully compliant with US patient confidentiality requirements set forth in Sections 164.514 (a)-(b)1ii of the Health Insurance Portability and Accountability Act (HIPAA) regarding the determination and documentation of statistically deidentified data. Because this study used only de-identified patient records and did not involve the collection, use, or transmittal of individually identifiable data, Institutional Review Board approval to conduct this study was not necessary.

Subject selection
Patients were selected for the study if they had an inpatient or outpatient medical claim with at least one diagnosis code for AF (ICD-9-CM diagnosis code 427.31 or ICD-10-CM diagnosis codes I480, I481, I482, or I4891) in any position between October 1, 2010 and December 31, 2015 (patient selection period) and at least one outpatient pharmacy claim during that period for warfarin on or after the first observed AF diagnosis and prior to initiation of dabigatran, rivaroxaban, or apixaban. Patients were further required to have at least one outpatient pharmacy claim for dabigatran, rivaroxaban, or apixaban after a pharmacy claim for warfarin. The inclusion dates for patients receiving these three NOACs depended on the drugs' launch dates, as follows: dabigatran 10/01/2010-12/31/2015, rivaroxaban 11/01/2011-12/31/2015, and apixaban 12/01/2012-12/31/2015. The date of the first pharmacy claim for dabigatran, rivaroxaban, or apixaban was the patient's index date.
All patients were required to be continuously enrolled for 12 months prior to the index date (pre-index period) and for up to 12 months following the index date (followup period). Patients were required to be at least 18 years of age on their index date.
Patients were excluded if they had an inpatient or outpatient medical claim that included any of the following procedures or diagnoses indicating possible valvular disease within 6 months prior to the first observed AF diagnosis: cardiac surgery, hyperthyroidism, myocarditis, technique without replacement and allowing a caliper of 1/4 of the standard deviation of estimated propensity scores. To examine the quality of the match, standardized differences were calculated before and after the matching, with an absolute value less than 10 indicating an acceptable match.

Outcome measures
All-cause healthcare utilization was reported overall and by type of service: inpatient, emergency department (ED), outpatient, physician office visits, other outpatient services, and pharmacy. The rate for all outcomes was calculated across all patients, regardless if they had a specific service, and among patients with at least one service. Healthcare resource utilization was reported on a PPPM basis to account for variable length of follow-up by dividing the total utilization over the observation period by the number of months of the observation period, and then averaged across patients to obtain PPPM.
Hospital length of stay (LOS) was reported in the follow-up period as the total number of inpatient days in the follow-up period divided by the number of inpatient admissions. Cumulative LOS (bed days) was reported as the total number of inpatient days in the follow-up period.
Readmissions within 30 days following discharge were evaluated among patients with at least 30 days of followup post discharge. The payment mechanism for Medicare Part A causes the majority of those patients to be missed using standard readmission methodology; therefore, a published algorithm was utilized to help identify Medicare readmissions 25 .

Key explanatory variables
Demographic variables measured on the index date included age in years, sex (male or female), payer type (commercial or Medicare), health plan type, United States Census Bureau geographic region of residence (Northeast, North Central, South, West), physician specialty on the index claim, duration of post-index period, and days between the index drug's launch date and index. Total costs were measured in the 12-month pre-index period.
Clinical characteristics measured during the 12-month pre-index period included the following: General measure of overall health status included the Deyo-Charlson Comorbidity Index (DCI), the number of unique medications, number of inpatient visits observed, and number of physician visits observed 26 . Specific comorbidities included chronic kidney disease, chronic obstructive pulmonary disease, pneumonia, cirrhosis, hepatitis, coronary artery disease, diabetes mellitus, gastrointestinal (GI) bleeding, heart failure, ischemic stroke, transient ischemic attacks (TIAs), hemorrhagic stroke, intracranial bleeding, extracranial bleeding, myocardial infarction (MI -acute pericarditis, pulmonary embolism, valve replacement, chronic rheumatic heart disease, or valvular disease, and for evidence of pregnancy (codes available upon request). Patients were also excluded if they had pharmacy claims for an anticoagulant other than warfarin (apixaban, dabigatran, rivaroxaban, argatroban, dalteparin, edoxaban, enoxaparin, fondaparinux, heparin, or tinzaparin) in the 6-month pre-index period, if they did not discontinue warfarin during the follow-up period, or if they had been indexed on a 10mg dose of rivaroxaban.
The follow-up period was a variable-length up to 12 months from the index date until the earliest evidence of switching to a different NVAF medication, discontinuation, inpatient death, end of continuous enrollment, or study end (December 31, 2016). Healthcare resource utilization was reported per patient per month (PPPM) to account for the variable length of follow-up periods [22][23][24] .
Switching was defined as the presence of a different anticoagulant (apixaban, dabigatran, edoxaban, rivaroxaban, warfarin, argatroban, dalteparin, enoxaparin, fondaparinux, heparin, or tinzaparin) within 30 days of the end of the days' supply of the index medication and evidence of discontinuing the index medication (definition below).
Discontinuation was defined as the lack of subsequent claims for the index medication beyond 90 days following the exhaustion of the previous claim's days' supply. The discontinuation date was the last day of supply immediately preceding the >90-day gap.

Matching
To control for demographic and clinical characteristics that could potentially affect the interpretation of study outcomes in head-to-head comparisons, propensity score matching was conducted to match apixaban and rivaroxaban patients to dabigatran patients. Matching factors included age, payer (commercial vs. Medicare), sex, geographic region of residence, health plan type, baseline comorbidities, stroke risk scores, bleeding risk scores, baseline total costs, and baseline medication use (beta blocker, calcium channel blocker, diuretics, other antihypertensives, antihyperlipidemics, steroids, anti-diabetic medications, antiarrhythmics, antiplatelets, non-steroidal anti-inflammatory drugs, selective serotonin reuptake inhibitors, and serotonin-norepinephrine reuptake inhibitors). Propensity scores were generated using a series of logistic regression models to predict the probability that a patient who discontinued warfarin switched to dabigatran vs. rivaroxaban and dabigatran vs. apixaban based on observed characteristics. Once each patient was assigned a propensity score, dabigatran switchers were matched 1:1 against the available pools of rivaroxaban and apixaban patients using the nearest neighbor matching and/or old), paraplegia/hemiplegia, psychiatric disorders, and venous thromboembolism. Stroke risk was assessed using the CHADS 2 and CHA 2 DS 2 -VASc scores, based on the presence of specified diagnoses from inpatient and outpatient claims in the 12-month pre-index period 27 . CHADS 2 score is calculated based on Congestive heart failure, Hypertension, Age, Diabetes, and Stroke/transient ischemic stroke. The CHA 2 DS 2 -VASc score adds vascular disease and female, has different points for two age groups, and includes all the other components of CHADS 2 . Bleeding risk was assessed using the HAS-BLED and ATRIA scores, also based on specified diagnoses in inpatient and outpatient claims during the 12-month pre-index period 28 . HAS-BLED is based on evidence of hypertension, abnormal renal or liver function, stroke, major bleeding, labile international normalized ratio (INR), elderly, and drug or alcohol use. Components of ATRIA are anemia, severe renal disease, age, hemorrhagic diagnosis, and hypertension.
International Normalized Ratio (INR) testing was determined via claims with a CPT or HCPCS procedure code. In addition, the following timings were measured: days from the first AF diagnosis to warfarin exposure, days from the first AF diagnosis to index drug initiation, days from the exhaustion of the days' supply of the last warfarin claim to index drug initiation, and the length of warfarin therapy.

Statistical analysis
Categorical variables are presented as counts and percentages. Continuous variables are summarized by providing means and standard deviations. Statistical tests of significance are conducted for evaluating differences between rivaroxaban or apixaban cohorts with respective matched dabigatran cohorts using Chi-square tests for categorical variables, and t-tests for continuous variables. A critical value of p<0.05 was specified a priori as the threshold for statistical significance. Data management, analytic file building, and statistical analyses were conducted using SAS software, version 9.4 (SAS Institute Inc, Cary, North Carolina).
Mean days from AF diagnosis to index medication initiation were significantly shorter for dabigatran than for either rivaroxaban (339 vs. 790 days, p<0.001) or apixaban (358 vs. 1065 days, p<0.001), owing to a combination of shorter days from AF diagnosis to warfarin initiation, shorter length of warfarin therapy, and shorter time from the end of warfarin therapy to index drug initiation ( Table 2).
The proportion of patients with at least one inpatient admission was lower for dabigatran compared with rivaroxaban (20.0% vs. 21.6%, p=0.008), but similar between dabigatran and apixaban patients (20.7% vs 21.2%, p=0.522) (Figure 2). Hospital LOS and cumulative days in hospitals were similar among all three cohorts (0.9-1.1 and 1.2-1.3) ( Table 3). The percentage of patients with at least one ED visit was lower for dabigatran compared with rivaroxaban (27.4% vs. 29.7%, p<0.001), but similar between dabigatran and apixaban patients (28.3% vs 29.8%, p=0.074). Significantly fewer dabigatran patients had claims for other outpatient services versus either

Discussion
This study is the first real-world data assessment comparing healthcare resource utilization among NVAF patients initially treated with warfarin who eventually switched to either dabigatran, rivaroxaban, or apixaban. This comparison is timely and important because many NVAF patients are expected to switch from warfarin to NOACs following the 2019 AHA/ACC/HRS guidelines 15 that recommend NOACs over warfarin. In this retrospective database analysis of healthcare resource utilization using head-to-head comparisons of warfarin switchers initiating dabigatran vs. rivaroxaban and dabigatran vs. apixaban, we found significantly higher proportions of rivaroxaban patients with hospital and ED visits compared with dabigatran-treated patients. Use of other outpatient services was lower for dabigatran compared with both rivaroxaban and apixaban users. Overall, there were fewer outpatient visits and ED visits PPPM for dabigatran patients compared with the other two NOAC groups. In addition, the number of office visits and other outpatient visits PPPM were significantly lower for dabigatran patients compared with rivaroxaban patients, though similar when compared with apixaban.
An earlier study by Gilligan et al. 32 found dabigatran patients had significantly fewer inpatient admissions and ED visits, and higher numbers of outpatient visits and pharmacy claims than apixaban patients. In the same study, compared with rivaroxaban, dabigatran patients had fewer admissions, outpatient visits, and pharmacy claims, and similar ED visits compared with apixaban and rivaroxaban, as well as lower healthcare resource utilization across all service categories (with exception to LOS) when compared with warfarin patients.
This study adds to the existing head-to-head comparison literature by focusing on NVAF patients who switched to NOACs from warfarin using real-world data that cover a large range of health plans, providers, and both commercial and Medicare payers. Consistent with previous comparisons, the current results display significant evidence of lower healthcare resource utilization for dabigatran compared with rivaroxaban for hospitalizations, ED visits, and outpatient visits. Dabigatran patients also displayed lower numbers of ED and outpatient visits compared with matched apixaban patients. Patients, providers, and payers may find significant benefits from these lower healthcare resource intensities associated with the use of dabigatran after warfarin discontinuation. Indeed, forthcoming studies will enable finer discernment of additional resource savings that can potentially be associated with a monetary value for our healthcare systems.

Limitations
This study was subject to limitations, including those inherent with retrospective administrative healthcare claims analyses. Claims are intended to support reimbursement and therefore variables found on claims have coding limitations and possible data entry errors. NVAF was defined based on diagnosis and procedure codes, thus misclassification error is possible for NVAF, covariates, and study outcomes due to missing or inaccurate codes. Medication usage was based on filled prescriptions that patients were assumed to have taken as prescribed; however, it was unknown whether medications were actually taken. Specific information for the reasons why patients switched from warfarin to a NOAC, such as improved efficacy or diminished side effects, are not available in claims data and remain unknown. The variance around the mean time from warfarin discontinuation to NOAC initiation extended from one to several months, and it is possible that these patients may have been treated with something other than an anticoagulant prior to initiating any of the NOACs. Patients' medical and prescription history was limited to administrative claims during the reporting months in this study. Due to the large sample sizes, small differences computed as statistically significant between comparators may not have significant clinical or cost importance. There may be systematic differences between the treatment groups that account for differences found in healthcare resource utilization. While propensity score matching provided adjustment for differences between treatment groups, there is the potential for unmeasured confounders given information was limited to characteristics that could be measured using administrative claims. This study was based on commercially-insured and Medicare covered individuals in the MarketScan databases, which are convenience samples of contributing US commercial and Medicare payers. Consequently, results of this analysis may not be generalizable to other US or international populations of NVAF patients, with other insurances or without health insurance coverage. Despite these limitations, this study provides a valuable insight among patients who switched from warfarin therapy, and the resulting effects on their healthcare utilization after switching to different NOACs.

Conclusions
Warfarin treated NVAF patients switch to NOACs due to various reasons, and this switching is expected to increase per the new AHA/ACC/HRS guidelines. Although the current literature provides head-to-head comparisons on healthcare resource utilization among NOACs, there are little data on comparisons in patients who switched to NOACs from warfarin. The current study indicates that the use of dabigatran following warfarin therapy may enable significant healthcare resource utilization savings compared with those who switched to rivaroxaban or apixaban following warfarin therapy. These findings provide valuable information that may facilitate physicians' treatment decisions in guiding patients who may be candidates for switching to a NOAC from warfarin.

Authors' Full Disclosures
Jessica Franchino-Elder, Briain O Hartaigh, and Cheng Wang, are employees of Boehringer Ingelheim Pharmaceuticals. Xue Song, Caroline Henriques, and Amy Sainski-Nguyen are employees of IBM Watson Health, which received compensation from Boehringer Ingelheim Pharmaceuticals for the overall conduct of the study and preparation of the manuscript. Adrienne Gilligan was an employee of IBM Watson Health during the time of the study and writing of the manuscript.