Robert B. Parker, PharmD

: Warfarin anticoagulation significantly decreases ischemic stroke risk among patients with atrial fibrillation (AF) over long-term use, but is often associated with discontinuations due to various reasons. The aim of this study was to determine economic effects of warfarin discontinuation in patients with nonvalvular AF (NVAF).

: Adult patients (aged ≥18 years) with NVAF to whom warfarin was prescribed in MarketScan Databases (January 1, 2008-June 30, 2012) were analyzed. Warfarin discontinuation was defined as a gap of  ≥45 days in warfarin prescriptions within 1 year after warfarin initiation. Patients with discontinuation were matched 1:1 to persistent patients based on a propensity score method. Matched patients were followed for 1 year after warfarin discontinuation. Negative binomial regression and generalized linear model were conducted to compare healthcare utilization and costs between discontinuation and persistent groups.

: A total of 27,000 patients were included. The discontinuation group was more likely than the persistent group to be hospitalized (mean annualized number of hospitalizations = 0.51 [SD = 1.34] vs 0.37 [0.78]) and to visit emergency departments (EDs) (mean annualized number of visits = 1.14 [2.44] vs 1.01 [1.95]). The discontinuation group also had greater annualized hospitalization costs ($10,216.95 [56,102.23] vs $6834.79 [38,222.68]), ED costs ($296.40 [1149.75] vs $259.96 [883.59]), and total healthcare costs ($26,990.48 [69,585.79] vs $21,472.50 [46,512.08]) (P <.0001 for all comparisons). These comparisons were similar after adjusting for patient characteristics.

: Warfarin discontinuation is associated with increased healthcare utilization and costs in patients with NVAF. Due to limited healthcare resources, decision makers may need to understand the reasons behind nonpersistence and consider alternative/new anticoagulants.

Atrial fibrillation (AF) is one of the most common cardiac rhythm disorders, affecting an estimated 5.2 million adults in the United States.

 The number of patients with AF is expected to increase to greater than 12 million by 2030.

 AF is a serious risk factor for thromboembolism and mortality.

To determine the economic impact of AF, Coyne et al analyzed three federal US databases and calculated total annual healthcare costs for treating AF.

 They found annual direct costs for treating AF were $6.65 billion in 2005 dollars, 73% of which was hospitalization costs, 23% outpatient treatment costs, and 4% prescription drug costs.

 A more recent study estimated that patients with AF incurred total incremental medical costs of $8705 per patient on an annual basis and that the overall incremental medical costs of AF were $26 billion annually.

Previous studies have further examined the disease burden of AF and its related clinical events such as stroke and bleeding.

 For example, Wu et al assessed the direct and indirect costs associated with AF in a privately insured population and found that excess annual direct costs of AF were $12,349 per capita in 2002 dollars.

Likewise, Lee et al estimated the treatment costs of AF in a Medicare population using a 5% random sample of Medicare beneficiaries, and they found that incremental annual treatment costs were $14,199 per capita in 2004 dollars.

 Given the high treatment costs associated with AF, as demonstrated by the aforementioned studies, disruption of treatment efforts and the resulting negative consequences for patient health threaten to further increase the disease burden imposed by AF.

Anticoagulants such as warfarin are associated with marked reductions in the risk of thromboembolic events, particularly ischemic stroke, in patients with AF.

 Yet, despite its proven clinical efficacy, warfarin treatment may be discontinued for a variety of reasons, including lack of compliance, bleeding, and achieving stable cardiac rhythm.

 The rate of warfarin interruption/discontinuation was estimated to be more than 45% within 1 year after warfarin initiation, and it is associated with an increased risk of stroke.

 Casciano et al assessed the economic burden of warfarin non-adherence/underutilization among patients with nonvalvular AF (NVAF) in a commercially insured population.

 Healthcare resource utilization and costs during the 18 months after AF diagnosis were compared between patients with the proportion of days covered (PDC) by warfarin >0.8 (high) and 

0.8 (low) versus patients with no warfarin exposure.

They found that patients receiving warfarin were 27% less likely to incur hospitalizations and 16% less likely to incur emergency department (ED) visits than patients who did not receive warfarin.

 Both low and high PDC were associated with lower all-cause inpatient costs compared with no warfarin exposure.

 However, this study did not examine the economic impact of warfarin discontinuation, a critical gap in the existing knowledge base.

The current study is among the first evaluations of the economic outcomes of warfarin discontinuation in patients with NVAF. The objective of the study was to determine the effects of warfarin discontinuation on utilization and costs of physician visits, hospitalizations, and ED visits, as well as total healthcare costs in patients with NVAF. This may assist clinicians and formulary/decision makers in raising awareness of the importance of warfarin persistence and in developing effective educational strategies to improve outcomes of anticoagulant therapy in this patient population.

This study used Truven Health MarketScan Commercial Claims and Encounters, and Medicare Supplemental and Coordination of Benefits Databases (MarketScan, licensed by Truven Health Analytics) between January 1, 2008, and June 30, 2012. MarketScan is compliant with the Health Insurance Portability and Accountability Act and provides access to medical and prescription drug claims for individuals who have an employer-sponsored health insurance plan in the United States.

This study was approved by the Institutional Review Board of the University of Tennessee Health Science Center.

A retrospective, cohort-matched design was used to compare NVAF enrollees who discontinued warfarin therapy with those who were persistent with this medication. The study cohorts were selected using the following criteria (Figure):

Patients were required to have at least two claims with primary or secondary diagnosis of AF, defined using 

International Classification of Diseases, Ninth Revision, Clinical Modification 

) code 427.31, separated by ≥30 days and ≤12 months and at least 1 outpatient claim from January 1, 2008, to June 30, 2011.

In the same period, patients filled at least 1 prescription for warfarin. Warfarin prescription was determined by either generic or brand name (generic name: warfarin sodium, warfarin potassium; brand names: Coumadin, Jantoven, Marevan, Lawarin, Waran, Athrombin-K, Warfant). The names of medications that patients filled were identified by linking outpatient pharmaceutical claims to the Red Book using National Drug Codes.

Patients had no record of use of warfarin or International Normalized Ratio (INR) monitoring (defined using Current Procedural Terminology codes 99363, 99364, 3555F, 85610, G0248, G0249, G0250) in the 6 months prior to warfarin initiation.

 AF diagnosis had to occur within 30 days of warfarin initiation to cover delayed entry of diagnosis codes, maximize the probability that patients were newly diagnosed AF patients, and maximize the likelihood that study subjects were receiving warfarin for stroke prevention.

Patients had continuous enrollment in prescription drug plans from 6 months prior to and at least 2 years after warfarin initiation.

Patients were aged 18 years or older as of warfarin initiation date.

Patients with mitral or aortic valvular repair or replacement (

 codes: 394.0, 394.2, 396.0, 396.1, 396.8, v43.3, v42.2, 35.10-35.14, 35.20-35.28), transient perioperative AF (

 codes: 36.10-36.19, 37.10-37.12, 37.31-37.33, 37.40, 35.00-35.04, 35.31-35.39, 35.41-35.42, 35.50-35.56, 35.60-35.63, 35.70-35.73), or hyperthyroidism (

Patients were followed for up to 1 year after warfarin initiation for determination of persistence and discontinuation status. Warfarin persistence (WP) was defined as warfarin therapy without a gap of ≥45 days between the end date of the former prescription and the start date of the current prescription, or with INR monitoring at least every 42 days.

 Warfarin discontinuation (WD) was defined as a gap in warfarin therapy of 

45 days between the end date of the former prescription and the start date of the current prescription, and without INR monitoring at least every 42 days.

A propensity score method was used to match WD patients 1:1 to WP patients to create cohorts with balanced patient characteristics. In this propensity score method, a logistic regression was used to predict the probability of discontinuation for each patient as their propensity score.

The variables used in the logistic regression included: age, gender, warfarin initiation year, CHADS

 (which is an acronym and clinical prediction score for stroke risk representing such predictors as Congestive heart failure, Hypertension, Age, Diabetes mellitus, and Stroke) score, geographic region, type of health insurance plan, bleeding in the 6 months prior to warfarin initiation, hospital or ED visits in the 6 months prior to warfarin initiation, and Charlson Comorbidity Index (CCI) score calculated in the 6 months prior to warfarin initiation.

Following the propensity score calculations, matched patients were subsequently followed for up to 1 year to determine the association between warfarin discontinuation and study outcomes. Patient follow-up started from the date of discontinuation for WD patients and after the same duration of warfarin therapy for the matched WP patients. The matched date was defined as the index date of the study. Patients were followed for up to 1 year starting from the index date. Follow-up ended when study outcomes were detected, a WD patient restarted warfarin, or a WP patient discontinued warfarin.

Study outcomes included healthcare resource utilization (physician visits, hospitalizations, and ED visits) and costs. Each type of event was defined as a count variable for the utilizations or a quantitative variable for costs. Patients’ utilization and costs of physician visits were obtained and summarized from outpatient claims by specifying codes on place of service, procedure group, and first procedure code.

Utilization and costs of hospitalizations were obtained and summarized from inpatient admission records. Utilization and costs of ED visits were obtained and summarized from outpatient claims by specifying codes on place of service, provider type, service type, procedure group, and revenue code. All cost categories were reported from the total gross cost perspective. Utilization and costs were annualized and costs were inflation adjusted to reflect 2013 US dollar market value by using Consumer Price Index of medical care.

To determine the association between warfarin discontinuation and healthcare utilization and costs, independent samples 

 tests and a test of equality of variances were conducted to compare the WD group with the WP group. Negative binomial regression was conducted to compare healthcare utilization between WD and WP patients. Generalized linear model with log link and Gamma distribution was conducted to compare healthcare costs across groups. Ratios (for example, incidence rate ratios [IRRs]) greater than one estimated from these models would suggest positive association between warfarin discontinuation and healthcare utilization and costs.

To determine the association between warfarin discontinuation and healthcare utilization and costs, prior utilization of services or healthcare costs incurred during the baseline period was added to the following baseline variables as a marker for baseline aggregate disease burden experienced by patients: age (<65 or ≥65 years), gender, US geographic region (Northeast, North Central, South, and West), type of health insurance plan (comprehensive plans, preferred provider organizations [PPOs], health maintenance organizations, and other plans).

For example, when analyzing the relationship between warfarin discontinuation and costs of hospitalization, the costs of hospitalization within the 6 months prior to warfarin initiation was added as an independent variable in the regression model. Additionally, some patient characteristics were included and determined for the 6 months prior to warfarin initiation: a diagnosis of congestive heart failure, hypertension, diabetes mellitus, stroke or transient ischemic attack (TIA), bleeding or anemia, and CCI score. All analyses were conducted using SAS 9.3 (SAS Institute Inc, Cary, North Carolina) and Stata 12 (Stata Corporation, College Station, Texas). Statistical significance was set 

A total of 27,000 patients were included in the final analysis after matching (Figure). Average age was 71.87 years (SD = 11.09), mean CHADS

 score was 1.64 (1.18), and mean CCI score was 1.61 (1.80) (

). As detailed in Table 1, 58.45% of patients were male, 42.17% had PPO insurance, and the rates of patients diagnosed with congestive heart failure, hypertension, diabetes mellitus, history of stroke or TIA, hemorrhagic stroke, anemia, and bleeding within the 6 months prior to warfarin initiation were 22.54%, 52.91%, 22.94%, 9.50%, 0.32%, 13.02%, and 10.53%, respectively. Almost 60% of patients were either hospitalized or had ED visits (Table 1).

Differences in average annualized healthcare utilization and costs per patient between the two groups are summarized in 

. WD patients compared with WP patients were more likely to be hospitalized (annualized mean number of hospitalizations = 0.51 [1.34] vs 0.37 [0.78], respectively; 

 <.0001) and to have ED visits (1.14 [2.44] vs 1.01 [1.95], respectively; 

 <.0001). The WD group compared with the WP group had fewer physician visits (12.94 [9.50] vs 15.27 [9.61], respectively; 

The WD group compared with the WP group had significantly higher mean costs of hospitalization ($10,216.95 [56,102.23] vs $6834.79 [38,222.68], respectively; 

 <.0001), costs of ED visits ($296.40 [1149.75] vs $259.96 [883.59], respectively; 

 <.0001), and total healthcare costs ($26,990.48 [69,585.79] vs $21,472.50 [46,512.08], respectively; 

 <.0001). Additionally, the average total healthcare costs of the WD group were 25.7% higher than the WP group. The WD group compared with the WP group had significantly lower mean cost of physician visits ($1232.17 [1547.09] vs $1302.23 [1379.14], respectively; 

 <.0001). Hospitalization costs accounted for 35.19% of total healthcare costs.

After adjusting for baseline characteristics, the comparison between the two groups on adjusted healthcare utilization and costs remained the same as the unadjusted results (Table 2 and Table 3). For example, the average number of hospitalizations was 0.48 for the WD group and 0.35 for the WP group (Table 2), so the WD group was 35% more likely to have an increase of 1 hospitalization compared with the WP group (IRR: 1.35; 95% CI, 1.28-1.43; Table 3).

Adjusted mean annualized hospitalization costs were $8826.62 for the WD group and $6077.59 for the WP group (Table 2), so the WD group incurred 45% higher costs than the WP group (ratio : 1.45; 95% CI, 1.30-1.62; Table 3). The WD group was 10% more likely to have an increase of 1 ED visit compared with the WP group (IRR: 1.10; 95% CI, 1.04-1.16; Table 3) than the WP group, and incurred 12% higher costs in ED visits (ratio: 1.12; 95% CI, 1.02-1.22; Table 3).

Additionally, the WD group had 21% higher average total healthcare costs (adjusted mean costs: $24,279.46; ratio: 1.21; 95% CI, 1.16-1.28) compared with the WP group (adjusted mean costs: $19,989.33; 

 <.0001). However, the WD group had 18% lower rates of physician visits (IRR: 0.82; 95% CI, 0.81-0.83) and 8% lower costs of physician visits (cost ratio: 0.92; 95% CI, 0.90-0.94) compared with the WP group (

Although interruptions in warfarin treatment are consistently associated with adverse outcomes, limited information was available in the literature on the economic impact of warfarin discontinuation prior to this study.

 Thus, the current study represents one of the first published works to examine the economic outcomes of warfarin discontinuation in patients with NVAF, including utilization and costs of physician visits, hospitalizations, and ED visits, as well as total healthcare costs.

Our results indicate warfarin discontinuation was associated with higher total healthcare costs and higher hospital and ED utilizations/costs compared with warfarin persistence. These findings are unsurprising given the effects of warfarin discontinuation on health outcomes among patients with NVAF.

Previous studies have found that warfarin discontinuation is associated with increased risk of ischemic stroke. For example, a study by Ewen et al found that patients with 1, 2, or more warfarin interruptions (defined as prescription gaps over 45 days) had higher stroke incidence than those without warfarin interruption (

 Two or more warfarin interruptions were associated with an increased risk of stroke (relative risk: 2.29; 95% CI, 1.29-4.07) after adjusting for patient characteristics.

Deitelzweig et al also found that stroke risk was higher with warfarin discontinuation than during continuous warfarin therapy (hazard ratio: 1.60; 95% CI, 1.35-1.90; 

 Therefore, it can be logically inferred that due to warfarin discontinuation, patients experience increased negative health outcomes such as stroke, which results in a greater number of hospitalizations and ED visits, and, in turn, increases healthcare costs.

Our results also show that the WD group had fewer physician visits and lower costs related to physician visits compared with the WP group. We speculate that this is directly related to warfarin discontinuation. For example, patients who have discontinued warfarin require fewer follow-up visits with physicians for monitoring (eg, INR) than those who remain on warfarin, and hence have lower costs associated with physician visits.

Although seemingly counterintuitive, higher number of physician visits and higher physician visit costs are likely positive indicators that a patient is remaining persistent with warfarin. Such persistence, in turn, may result in lower hospital, ED, and total healthcare utilizations and costs.

This study’s findings can be compared with previously mentioned studies that examined the disease burden of AF. In a study by Wu et al among the privately insured population, the annual direct healthcare costs of AF were $12,349 per capita in 2002 dollars ($18,297 in 2013 dollars).

 Lee et al found that treatment costs of AF among the Medicare population was $14,199 in 2004 dollars ($19,376 in 2013 dollars).

The present study found that costs of treating NVAF for WD patients versus WP patients were $26,990 and $21,473, respectively. As the estimates in this study were similar to those found by Wu et al and Lee et al (although neither of these studies considered the effects of warfarin discontinuation), these comparisons testify to the reliability of the findings from the current study.

While having made interesting and important findings utilizing a large, geographically diverse sample with substantive generalizability, this study has limitations mainly due to the use of a claims database. First, claims databases are not constructed for research purposes, so the inherent limitations include the unavailability of important clinical factors including INR measures, other test results, use of over-the-counter medications such as aspirin, or other important clinical measures such as liver and renal function.

Second, this study may misclassify patient persistence status when categorizing patients based on warfarin refill history. Patients may take different dosing than what was indicated in the database (eg, they may split pills) and may discontinue warfarin therapy before exhausting prescription supply. To address this potential issue, although warfarin days of supply were typically 30 days, a prescription gap of 45 days was used to define discontinuation.

However, without patient self-recorded/reported confirmation, this study’s warfarin persistence categories may potentially involve minor misclassification. Another limitation is that the cost analyses were limited to direct medical costs because indirect costs such as absenteeism and presenteeism due to illness were not included in the claims database. Further, the reasons for differential utilization and costs between the two study groups cannot be pinpointed based on claims database analysis.

Lastly, this study may have had a selection bias issue: Individuals included in the final cohort may have been systematically different from the typical cohort of warfarin users due to the continuous enrollment requirements, propensity score matching, and general requirement to have commercial insurance. However, the consistency of this study’s findings with previous studies confirmed the reliability of the study findings from current study. Despite inherent limitations, well-designed observational studies can and have played an important role in representing real-life scenarios in clinical practice with high generalizability.

Articles

This study seeks to determine the economic effects of warfarin discontinuation in patients with nonvalvular atrial fibrillation.