A retrospective cohort study was conducted with deidentified medical and pharmacy claims data from January 1, 2010, to December 31, 2012, using a nationwide commercial U.S. managed care health plan affiliated with Optum Clinformatics Data Mart (Optum Insight, Eden Prairie, MN). Adult patients were included who were aged ≥ 18 years with at least 1 AF or atrial-flutter diagnosis claim identified using the medical file (inpatient or outpatient) with an International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) code of 427.31/427.32 during the pre-index period. Patients with at least 2 pharmacy claims of dabigatran or rivaroxaban (NOACs) were identified based on National Drug Code numbers. Patients with at least 6 months of pre-index and 12 months of post-index continuous eligibility with a permissible gap of 45 days were included in the cohort. Patients with hyperthyroidism (ICD-9-CM code 242.9) were excluded from the cohort, since hyperthyroidism may be the probable cause of AF. The CHA₂DS₂VASc score was used to quantify the stroke risk and was scored from 0-9 points based on risk factors for stroke that included congestive heart failure, hypertension, age 65-75 years, diabetes, peripheral vascular disease, female gender (constituting 1 point each), and previous occurrence of transient ischemic attack or stroke and age ≥ 75 years scored 2 points.^12-14 Patients with CHA₂DS₂VASc scores ≥ 1 in the pre-index period were included in the cohort.^13,14

The index date was defined as the date of the first prescription fill of NOAC medication during the measurement period. Patients with concomitant use of warfarin and NOACs during the post-index period were excluded. To avoid exclusion of the few patients who used warfarin in the pre-index period before treatment with NOACs, NOAC users with pre-index warfarin use were included based on the definition of “warfarin-naive” that was applied in the Randomized Evaluation of Long-Term Anticoagulation Therapy (RELY) trials. Patients were considered to be warfarin-naive if they had not used warfarin during the 2 months before initiation of a NOAC.¹⁵ Data from the RELY trials showed no heterogeneity between patients who had previous warfarin use and those with no previous warfarin therapy. We used this criterion to ensure that we captured all NOAC users and to avoid potential channeling bias regarding previous warfarin therapy for assessment of outcomes. Thus, the index date of warfarin-naive users was based on the first prescription fill of NOACs. Patient outcomes were assessed over a post-index follow-up period of 12 months. The assessment period included index date to 3 months, 6 months, and 12 months.

Proportion of days covered (PDC) was the preferred method to measure adherence, where the numerator was defined as number of days covered by NOAC drugs as a class (using first fill date and days of supply) and the denominator was the days between first fill and end of the study, disenrollment, or death, whichever occurred first. PDC also helped to adjust early refills.¹⁶ Overall adherence to NOAC use was calculated using PDC from the index date to the end of the assessment period (3, 6, and 12 months). Patients with a PDC ≥ 80% were classified as adherent.^6,17 Adherence was also calculated for an assessment period of 3, 6, and 12 months using the following 3 severity groups:

The per-patient post-index cost for a follow-up of 12 months was calculated. The medical cost was defined as the total amount paid for inpatient and outpatient medical services plus inpatient and outpatient professional fees and copays. The total health care cost was the sum of all medical and drug costs using the amount paid by the patient through prescription fills (pharmacy claims and copays). The cost measures were expressed as average per-patient annual all-cause costs. The cost incurred for the years 2010-2012 was adjusted for inflation to 2016 U.S. dollars based on the Consumer Price Index.

A logistic regression model was used to examine the predictors of medication adherence based on PDC ≥ 80% over 12 months post-index (dependent variable, 1/0). Age, region, insurance plan type, gender at index date of NOAC use, pre-index CCI, stroke risk based on CHA₂DS₂VASc score, bleeding risk based on HAS-BLED score, and cardiac medication use that included statin use, angiotensin-converting enzyme inhibitor or angiotensin receptor blockers (ACE-ARB) use, and beta blocker use were used as covariates and included in the full model.

Generalized linear models with gamma distribution and log link were also used to obtain and compare the adjusted annual costs between adherent and nonadherent patients, defined based on PDC ≥ 80%. The gamma distribution was preferred, since we expected a skewed distribution of the cost, and the gamma model is an acceptable approach to handle cost data.²⁰ Age, CCI scores, stroke risk based on CHA₂DS₂VASc scores, region, insurance type, gender, bleeding risk based on HAS-BLED scores, and cardiac medication use including statin, ACE-ARB, and beta blocker use were used as covariates. The total annual cost, along with its subcomponents (drug cost and medical cost) were also compared between the patients stratified based on CCI score and stroke and bleeding risk at a significance level of P ≤ 0.05. Data were analyzed using SAS Enterprise Guide version 7.1 (SAS Institute, Cary, NC).

Of the 25,120 NOAC users identified during the study period, 14,618 had 2 or more prescription fills. Of these patients, 5,364 patients had used warfarin in the pre-index period, and, 3,726 patients who were not warfarin naive had been excluded, leaving 1,638 patients had been included as NOAC users because they fit the definition of warfarin naive. Because of concomitant use of warfarin and NOACs in the post-index period, 1,010 patients had also been excluded. Based on the other inclusion criteria (diagnosis of AF in the pre-index period, aged ≥ 18 years, continuous enrollment for 6-month pre-index and 12-month post-index, and a CHA₂DS₂VASc score ≥ 1), 2,891 NOAC patients formed the final cohort (Figure 1).

The mean age of the cohort was 64 years, with more men (70%) than women. Most of the patients were either from the South or the Midwest (68%). The proportion of patients with moderate and high risk of stroke was higher (high 36% vs. moderate 45% vs. low 19%) compared with low-risk patients. Based on CCI scores, most (80%) of the patients had at least 1 comorbid condition. Patients were well distributed across bleeding risk based on the HAS-BLED scores. More than 45% of the cohort used statins and ACE-ARB class of cardiovascular drugs (Table 1).

Based on baseline characteristics, age, stroke risk, type of insurance, statin use, bleeding risk, CCI, and region were significantly (P ≤ 0.05) different across the adherent and nonadherent patient groups. The mean age of patients classified as adherent was slightly higher (approximately 67 years [standard deviation {SD} = 10]) compared with the nonadherent group (approximately 62 years [SD = 11]). Gender and beta blocker use was consistent between the adherence-based cohorts. The proportion of patients with moderate to high risk of stroke (CHA₂DS₂VASc score ≥ 4) was substantially higher among adherent versus nonadherent patients (43% vs. 28%; Table 1).

Adherence was measured during follow-up periods of 3, 6, and 12 months. In the overall sample, the proportion of adherent patients (PDC ≥ 80%) declined during the period from 3, 6, and 12 months (72% for 3-month follow-up, 65% for 6-month follow-up, and 54% for 12-month follow-up). This trend was evident across all subgroups based on CCI, CHA₂DS₂VASc, and HAS-BLED scores (Table 2).

Adherence increased with a higher risk of stroke. Based on the 12-month follow-up, the proportion of patients adherent to NOACs in the low-risk and moderate-risk groups were 41% and 52%, respectively, and were significantly (P < 0.001) lower compared with the proportion of adherent patients (65%) in the high-risk group. Similar results were observed during the follow-up periods of 3 and 6 months. Higher adherence among higher risk groups was evident in the subgroups based on bleeding risk and CCI scores over the 3 follow-up periods (3, 6, and 12 months; Table 2).

Based on the logistic regression model, age, region, CCI scores, stroke risk, statin and ACE-ARB use, and bleeding risk were significant predictors of medication adherence to NOACs. Adherence rate was higher with increasing age (odds ratio [OR] = 1.04, 95% confidence interval [CI] = 1.02-1.05), and patients in the Midwest had higher adherence compared with the other regions. Patients with higher stroke risk were 1.56 times (95% CI = 1.14-2.12) more likely to be adherent compared with low-risk patients. Similarly, patients with CCI scores of 1-2 had higher odds (OR = 1.25, 95% CI = 1.01-1.54) of being adherent than those with CCI scores of 0. Patients using statins and ACE-ARBs tended to be more adherent (OR for statins = 1.31, 95% CI = 1.11-1.53 and OR for ACE-ARBs = 1.19, 95% CI = 1.02-1.39) to NOAC therapy compared with patients who were not taking these cardiovascular medications. Finally, patients with a moderate risk of bleeding were 1.07 times (95% CI = 0.88-1.30) more likely to be adherent compared with low-risk patients (Table 3).

As expected, the annual adjusted per-patient drug cost for adherent users was significantly higher compared with nonadherent users ($5,595 vs. $2,233; P ≤ 0.001). However, the annual adjusted per-patient medical cost that included all inpatient and outpatient costs was significantly lower for adherent users compared with nonadherent users ($23,544 vs. $30,485; P ≤ 0.001). The total annual adjusted per-patient cost for adherent users was also significantly lower than nonadherent users ($29,742 vs. $33,609; P = 0.005). Overall, the high drug cost among adherent users was offset by lower inpatient and outpatient costs (medical costs) compared with nonadherent users (Table 4).

Most of the patients had moderate to high risk of stroke based on the CHA₂DS₂VASc score (low risk 19.29%, moderate risk 44.82%, and high risk 35.89%). The annual per-patient drug cost was slightly lower in patients with low risk of stroke compared with patients with moderate and high risk (low risk $3,347 vs. moderate risk $3,664 vs. high risk $3,602). The annual per-patient medical cost for NOAC users with high risk of stroke was lower than the cost of patients with low risk of stroke. A similar trend was evident for the total cost in which the cost incurred by high-risk patients was lower than the cost incurred by low-risk patients (Table 4).

Patients were well distributed by bleeding risk based on the HAS-BLED score (low risk 36.36%, moderate risk 34.49%, and high risk 29.15%). The annual adjusted per-patient drug cost did not significantly differ across groups based on the risk of bleeding (low risk $3,596 vs. moderate risk $3,508 vs. high risk $3,501). However, the annual adjusted per-patient medical cost increased significantly with an increased bleeding risk (low risk $22,516 vs. moderate risk $24,916 vs. high risk $34,276; P ≤ 0.001). High medical cost was the driver for the total annual cost, in which the same trend of higher total cost with increased risk of bleeding was consistent (Table 4).

Most of the patients had at least 1 comorbidity (79.8%) based on CCI scores. There were 32.74% of patients with 3 or more comorbid conditions. Similar to the subgroups based on bleeding risk, the annual adjusted per-patient drug cost did not significantly differ across groups based on CCI scores. However, the total adjusted per-patient cost increased significantly with increase in CCI scores (low risk $27,189 vs. moderate risk $29,641 vs. high risk $39,216; P ≤ 0.001), which was mostly driven by the high medical costs across highly comorbid patients (Table 4).

A few limitations need to be considered when interpreting our results. Although PDC has been widely used and accepted, it is an indirect method of assessing adherence because it is based on insurance claims (prescription fills). It should also be acknowledged that Optum Clinformatics Data Mart is a nationwide database, but the majority of the study patients were from the southern and midwestern regions. This database also underrepresents the older Medicare population.

As with all studies using claims databases, our data lacked information on race and ethnicity, long-term disease history, reasons for therapy discontinuation, and clinical variables such as INR values and body mass index. However, clinical markers of severity, such as CHA₂DS₂VASc, HAS-BLED, and CCI, reflected risk due to hypertension, previous cardiovascular disease, diabetes, and other comorbidities. Aspirin use was not comprehensively captured in the claims database because of its availability as an over-the-counter drug; however, since all patients were prescribed NOACs, the differential use of aspirin might be unlikely among patients with a CHA₂DS₂VASc score > 1, who are recommended anticoagulant therapy based on guidelines from the American College of Cardiology and the American Heart Association.

It is also important to acknowledge the differences in mechanism of action and costs between dabigatran and rivaroxaban. A higher proportion of our sample was prescribed dabigatran, since it was approved earlier. The newer drugs apixaban and edoxaban, which came into the market after 2012, were not included in our analysis. Since the measurement of PDC and cost was in the same period, inferences suggest a cross-sectional association instead of a sequential (longitudinal) causal relationship. This is a question that can be explored in future studies.

Despite these limitations, this study provides new evidence that emphasizes the importance of considering overall cost and adherence information in conjunction with the assessment of stroke and bleeding risk to inform the optimal choice of anticoagulant therapy and ascertain the overall risk-benefit of NOAC therapy.