Characteristics affecting oral anticoagulant therapy choice among patients with non-valvular atrial fibrillation: a retrospective claims analysis

The study was a retrospective analysis of medical and pharmacy claims utilizing administrative claims data from the HealthCore Integrated Research Database (HIRD^SM), a large database of administrative claims for from a large health benefits organization. The HIRD^SM contains longitudinal claims data for approximately 45 million patients belonging to 14 health plans in the Northeastern, Southern, Midwestern, and Western regions of the US; it includes patients covered by health maintenance organizations (HMOs), preferred provider organizations (PPOs), and other health plans (including point of service (POS), consumer-directed health, and indemnity health plans).

All study materials were handled in compliance with the Health Insurance Portability and Accountability Act (HIPAA), and a limited dataset (as defined in the Privacy Regulations issued under HIPAA) was used for this analysis. Patient confidentiality was preserved and the anonymity of all patient data was safeguarded throughout the study. Institutional Review Board approval was not required for this retrospective observational study in which researchers only had access to a limited data set that excluded specified patient identifiers and consisted of anonymized patient information.

Eligible patients were identified from the HIRD^SM during the study period from October 1, 2009 to April 30, 2012. Patient aged 18 years or older were required to have at least two medical claims for AF as primary or secondary diagnosis (ICD-9-CM code 427.31) separated by at least 30 days between October 1, 2009 and October 31, 2011. At least one of these AF medical claims had to be an outpatient claim and both were required to occur prior to or coincide with the study index date (see below for definition). Patients were classified into dabigatran and warfarin cohorts based on their OAC prescription claims filled during the study intake period (from October 1, 2010 to October 31, 2011). During the intake period, dabigatran patients had ≥2 dabigatran prescriptions, warfarin patients had ≥2 warfarin and no dabigatran prescriptions, and the first OAC prescription fill date served as the index date. Continuous health plan enrollment for 12 months preceding (“pre-index period”) and ≥ 6 months following the index date was required. Patients with valvular heart disease/valve replacement or hyperthyroidism (determined via ICD-9-CM codes (Table 1) in the 12-month pre-index period were excluded. Patients without any prescription claims for an oral anticoagulant during the pre-index period were classified as oral anticoagulant naïve (OAC-naïve) patients. Separate analyses were performed for the ‘all-patient’ and ‘OAC-naïve’ cohorts.

The primary outcome variable was use of dabigatran or warfarin on the study index date. Independent variables of interest included patient pre-index demographic and clinical characteristics, healthcare provider specialty, and health plan factors. Demographic characteristics included patient age (at the index date) and gender. Healthcare variables included health plan geographic location (Northeast, Midwest, South, West), and type of health plan [HMO, PPO, other]. A healthcare provider factor (the index OAC prescriber specialty (cardiology, primary care/family/internal medicine,other specialty) was examined. Total patient pre-index out-of-pocket prescription drug costs were calculated.

Comorbid conditions were identified by ICD-9-CM codes associated with medical claims at any position during the pre-index period and included the following: coronary artery disease, myocardial infarction, cardiomyopathy, prior ischemic stroke, prior transient ischemic attack (TIA), prior bleed, heart failure, hypertension, peripheral artery disease, diabetes, and stages II-V chronic kidney disease. In addition, the Elixhauser comorbidity index (ECI) [17, 18] was used to assess patient overall pre-index comorbidity burden, using ICD-9-CM codes for 30 conditions associated with medical claims during the pre-index period. CHADS₂[19] stroke risk factors and total score(calculated based on presence of heart failure (HF), hypertension, diabetes, prior stroke/TIA, and age >75 years) and ATRIA [20] bleeding risk scores (calculated based on presence of anemia, severe renal disease, any prior hemorrhage diagnosis, hypertension, and age ≥75 years) were calculated for study cohorts. AF-related procedures (heart rate and heart rhythm procedures, including insertion of pacemaker or defibrillator, cardioversion, surgical ablation, and left atrial catheter ablation) during the pre-index were assessed.

Additionally, pharmacy prescriptions dispensed during the pre-index period were described for the following medications: warfarin (all-patients cohort only), parenteral anticoagulants (enoxaparin, tinzaparin, dalteparin, fondaparinux), anti-platelet therapy/NSAIDS (clopidogrel, ticagrelor, prasugrel, ticlopidine, cilostazol, dipyridamole, and non-steroidal anti-inflammatory agents [NSAIDS]), heart rate control medications (beta-blockers, calcium channel blockers, digoxin/digitoxin), antiarrhythmic medications (dronedarone, amiodarone, disopyramide, procainamide, flecainide, propafenone, sotalol, deofetilide, quinidine, mexiletine, moricizine), and dyspepsia medications (proton pump inhibitors and histamine receptor antagonists). Finally, the presence of ≥1 prothrombin time (PT) test during the pre-index period was assessed (all-patients cohort only) as a proxy for presence of ≥1 INR tests.

Patient demographic and clinical characteristics, healthcare provider specialty, and health plan factors associated with dabigatran or warfarin use on the index date were analyzed separately for all-patient and OAC-naïve patient cohorts. Two-sample t-tests were used to test differences between continuous variables, chi-square tests were used to test differences between categorical variables, and Wilcoxon rank-sum tests were used to test differences in the cost variables between the dabigatran and warfarin cohorts.

Multivariable logistic regression (LR) with backward elimination was used to identify significant factors associated with dabigatran (reference group = warfarin) use. Separate LR models were conducted for all-patient and for OAC-naïve patient cohorts. The following independent variables were used in the LR models: patient age in years (defined as 18–54, 55–64, 65–74, 75–84, and ≥85), patient gender, ECI score, ATRIA bleeding risk score, separate indicator variables for the presence of relevant pre-index comorbidities (HF, hypertension, diabetes, ischemic stroke, transient ischemic attack, stages II-V chronic kidney disease); indicator variable for pre-index AF-related procedures; pre-index use of warfarin, pre-index PT test (only among all-patient cohort); separate indicator variables for pre-index use of parenteral anticoagulants, anti-platelet/NSAIDS therapy, heart rate control medications or antiarrhythmic medications, and dyspepsia medications; and health plan variables (health plan type and geographic region), the specialty of index OAC prescriber, and total patient pre-index out-of-pocket prescription drug costs. Female and age group variables were forced into the LR models.

Overall, 20,320 patients who received either dabigatran or warfarin therapy on the index date met the study criteria (“all-patient cohort”): 14.9% (3,019) received dabigatran, while 85.1% (17,301) received warfarin. Among patients who received dabigatran, 464 (15.4%) also received warfarin prescriptions during the study follow-up. Among OAC-naïve patients (n = 2,405): 815 (33.9%) patients received dabigatran and 1,590 (66.1%) received warfarin and had no dabigatran prescriptions during the pre-index period.

Patient demographic and clinical characteristics, healthcare provider specialty, and health plan factors between dabigatran and warfarin cohorts were comparable among the all-patient and OAC-naïve cohorts (Table 2). Dabigatran patients were significantly younger than warfarin patients [all-patient cohort: mean (SD) age 67 (11.4) versus 73 (11.4) years, p < 0.0001; OAC-naïve cohort: mean (SD) age 65 (11.9) versus 71 (12.1) years, p < 0.0001] and less frequently females (all-patient: 29.1% versus 39.2%, p < 0.001; OAC-naïve: 27.5% versus 35.4%, p < 0.0001). Among all- and OAC-naïve patients, dabigatran users were more frequently covered by a preferred provider organization (PPO) benefit design than a HMO benefit design (all-patient: 67.8% versus 49.3%, p < 0.0001; OAC- naïve: 68.2% versus 52.9%, p < 0.0001). Compared to warfarin, dabigatran use was less frequent in the Midwest (all-patient: 21.5% versus 36.6%, p < 0.001; OAC-naïve: 22.7% versus 44.9%, p < 0.0001) and more frequent in the West health plans (all-patient: 32.0% versus 22.9%, p < 0.001; OAC-naïve: 32.1% versus 22.4%, p < 0.0001). Patients were more likely to be prescribed dabigatran as their index OAC by a cardiologist (rather than a primary care, or other provider) than warfarin patients (all-patient: 51.0% versus 30.3%, p < 0.001; OAC- naïve: 51.3% versus 32.7%, p < 0.001). Mean patient out-of-pocket pharmacy costs during the pre-index period was higher for dabigatran than warfarin patients [all-patient: mean (SD) $1,012 ($1,585) versus $952 ($1,050), p < 0.0001; OAC-naïve: mean (SD) $870 ($2,135) versus $655 ($735), p < 0.0001].

Compared to warfarin patients, dabigatran patients had lower CHADS₂ risk scores (all patients: mean (SD) 1.9 (1.3) versus 2.2 (1.3), p < 0.0001; OAC- naïve: mean (SD) 1.7 (1.4) versus 2.2 (1.4), p < 0.0001)), lower ATRIA scores (all patient: mean (SD) 2.3 (2.1) versus 3.0 (2.3), p < 0.0001; OAC- naïve: mean (SD) 2.0 (2.0) versus 3.1 (2.6), p < 0.0001)) (Table 3). Dabigatran patients also had fewer pre-index comorbidities than warfarin patients (all patient: mean (SD) ECI 4.0 (2.2) versus 4.3 (2.3), p < 0.0001; OAC-naïve: mean (SD) ECI 3.7 (2.1) versus 4.5 (2.6), p < 0.0001). Dabigatran patients more frequently had a history of ischemic stroke or TIA, but less frequently had hypertension, diabetes, and HF than warfarin users. Dabigatran patients more frequently had pre-index AF-related procedures than warfarin patients among the all-patient cohort (5.1% versus 2.5%, p < 0.0001), but there was no significant difference in the proportion of patients with pre-index stroke/TIA or AF-related procedures between OAC-naïve dabigatran and OAC-naïve warfarin patients.

Among the all-patient cohort, more warfarin patients used warfarin during the pre-index period compared to dabigatran patients (90.8% versus 73.0%, p < 0.0001) (Table 3). Dabigatran patients more frequently used anti-platelet agents/NSAIDS than warfarin patients during the pre-index period among both the all-patient and OAC-naïve cohorts (all-patient: 21.5% versus 16.0%, p < .00001; OAC- naïve: 27.6% versus 21.4%, p < 0.0007). Use of parenteral anticoagulants was more common among dabigatran users in the all-patient cohorts, but less frequent among dabigatran than warfarin patients in the OAC-naïve patients (all patient: 7.7% versus 5.6%, p < 0.0001; OAC-naïve: 1.6% versus 3.7%, p < 0.0040). Use of any heart rate or rhythm control medication was more common among dabigatran than warfarin patients (all-patient: 84.9% versus 82.2%, p < 0.0004; OAC-naïve: 80.2% versus 73.2%, p < 0.0001). Among the all-patient cohort, warfarin users were more likely to have ≥1 PT tests during the pre-index period (90.8% versus 73.0%, p < 0.0001).

The factors that were significantly associated with dabigatran use were generally similar for the all-patient and OAC-naïve patient LR analyses (Table 4); the null hypothesis for lack of model fit was rejected based on a Hosmer and Lemeshow test for the all-patients model (p-value = 0.6416) and for the OAC-naïve patients model (p-value = 0.0754), indicating that both models fit the data well. In both analyses, the adjusted odds of dabigatran use decreased with increasing patient age among patients 65 or older (range of OR = 0.39-0.64; all p ≤ 0.0001 Table 4). Females had slightly lower odds of dabigatran use compared to males (adjusted OR = 0.87, 95% CI 0.79-0.96) in the all-patient analysis but female gender was not a significant predictor in the OAC-naïve analysis. Dabigatran use was also less likely among patients in Midwestern health plans than health plans in the West (all-patient: adjusted OR = 0.55, 95% CI 0.49-0.62; OAC-naïve: adjusted OR = 0.43, 95% CI 0.34-0.56). For both the all-patient and OAC-naïve patient cohorts, pre-index hypertension (all-patient: adjusted OR = 1.16, 95% CI 1.04-1.30; OAC-naïve: adjusted OR = 1.32, 95% CI 1.04-1.69) and ischemic stroke (all-patient: adjusted OR = 1.34, 95% CI 1.18-1.53; OAC-naïve: adjusted OR = 1.42, 95% CI 1.06-1.90) were associated with greater odds of dabigatran use. For both the all-patient and OAC-naïve patient cohorts, pre-index HF and higher risk of bleeding (based on ATRIA score) were associated with lower adjusted odds of dabigatran use. Among the all-patient cohort, prior use of warfarin was strongly associated with lower adjusted odds of dabigatran use (OR = 0.34, 95% CI 0.31-0.38). For the all-patient cohort only, pre-index AF-related procedures, anti-platelet/NSAID medications, dyspepsia medications, rate/rhythm control medications, and pre-index TIA were all positively associated with dabigatran use; odds ratios and associated 95% confidence intervals are shown in Table 4.

Health plan characteristics were significantly associated with the odds of dabigatran use, adjusted for other factors (Table 4). Specifically, patients covered by PPO benefit designs were 68% and 47% more likely to be prescribed dabigatran than patients covered by HMO benefit designs in the all patient and OAC-naïve cohorts, respectively. Compared to patients prescribed their index OAC by a primary care/family/internal medicine physician, patients whose index OAC was prescribed by a cardiologist (all-patient: adjusted OR = 3.12, 95% CI 2.77-3.51; OAC-naïve: adjusted OR = 3.59, 95% CI:2.68-4.81) or other specialist (all-patient: adjusted OR = 1.94, 95% CI 1.72-2.20; OAC-naïve: adjusted OR = 2.22, 95% CI:1.65-2.97) were more likely to receive dabigatran. Finally, for both study cohorts, higher total pre-index period patient out-of-pocket pharmacy costs (per $1,000) were associated with greater odds of dabigatran use (all-patient: adjusted OR = 1.04, 95% CI 1.01-1.08; OAC-naïve: adjusted OR = 1.32, 95% CI:1.17-1.48).

Several limitations should be noted in the interpretation of our study’s findings. Although a number of relevant patient factors were analyzed in this study, other patient factors such as patient socioeconomic status, race, and use of over-the-counter medications such as NSAIDs, are not available in administrative claims and therefore were not included. We were not able to assess patient awareness of dabigatran therapy through direct-to-consumer advertising, nor were we able to control for any effect that treatment outcomes (e.g., bleeding) associated with prior warfarin therapy may have had on dabigatran use in our analyses. Nevertheless, the findings of this study were consistent in the all-patient and OAC-naïve cohorts and support the robustness of this study.