The Costs of Warfarin Underuse and Nonadherence in Patients with Atrial Fibrillation: A Commercial Insurer Perspective

BACKGROUND: Atrial fibrillation (AF) imposes a substantial clinical and economic burden on the U.S. health care system. Despite national guidelines that recommend oral anticoagulation for stroke prevention, the literature consistently reports its underuse in AF patients with moderate to high stroke risk. OBJECTIVES: To assess the economic burden of underuse and nonadherence of warfarin therapy among patients with nonvalvular AF in a commercially insured population. METHODS: Claims data between January 2003 and December 2007 from the Thomson Reuters MarketScan Research Database were used. Patients diagnosed with nonvalvular AF who were continuously enrolled for at least 12 months prior to and 2 months following their diagnosis, who had a CHADS2 score ≥2, and were not at high risk of bleeding (ATRIA score less than  5, HEMORR2HAGE score less than  4, and HAS-BLED score less than  3) at baseline were included. Patients were followed for up to 18 months after the AF diagnosis date to assess the level of warfarin utilization. Health care resource utilization and cost during follow-up among patients with the proportion of days covered (PDC) by warfarin greater than 0.8 (high) and ≤0.8 (low) versus patients with no warfarin exposure were assessed. Multivariate negative binomial regressions and generalized linear models were used to estimate differences in resource utilization and cost, respectively. RESULTS: Of the 13,289 subjects included in this analysis, 47% had no warfarin exposure; 31.5% had low PDC; and 21.5% had high PDC. The rates of ischemic stroke and transient ischemic attack (per 100 patient-years) were significantly lower for the groups that had high and low PDCs as compared with the group with no warfarin exposure (P  less than  0.001). Multivariate analysis showed that patients with high PDC were 27% less likely (P  less than  0.001) to incur hospitalizations, and 16% were less likely (P = 0.019) to incur emergency room visits than patients who did not receive warfarin, but the differences between low PDC patients and no warfarin exposure were not significant. Although both low and high PDC were associated with lower all-cause inpatient cost (P  less than  0.001), only high PDC was associated with a lower post-index all-cause total cost (P  less than  0.001) compared with no warfarin exposure. CONCLUSIONS: Our results confirm that underutilization and nonadherence of warfarin among nonvalvular AF patients is both prevalent and costly. Warfarin use among patients with moderate to high stroke risk and low to moderate bleed risk demonstrated a stroke benefit without a significant increase in intracranial hemorrhage. Adherence to oral anticoagulant therapy was associated with a significant reduction in inpatient service use and total health care cost. Improving adherence to oral anticoagulation is important to attaining the clinical and economic benefits of therapy.

A trial fibrillation (AF) is the most common type of cardiac arrhythmia, and it affects more than 3 million U.S. adults. 1 AF is responsible for 15% to 20% of all strokes 2 and confers almost 5 times the risk of stroke and 1.9 times the risk of death versus patients without AF. 2,3 Total direct cost related to AF treatment in the United States is reported to be more than $6 billion annually. 4 The overall economic burden of stroke from a societal perspective, including both direct and indirect costs, was estimated at $34.3 billion in 2008. 5 Direct cost of AF-related strokes was estimated to be approximately $8 billion. 6 Clinical guidelines recommend the use of oral anticoagulants for long-term stroke prevention in AF patients who are at intermediate to high risk of stroke (i.e., CHADS 2 score > 1), assuming patients are not at high bleeding risk. 7,8 However, studies have consistently reported that many patients who were eligible for oral anticoagulation did not receive warfarin. Zimetbaum et al. (2010) 9 report that only 42.1% of patients with a CHADS 2 score ≥ 3 received warfarin therapy, and according to a systematic literature review, less than 60% of patients with high stroke risk are on oral anticoagulant (OAC) • Atrial fibrillation (AF) is estimated to affect more than 3 million U.S. adults. It is associated with significant increase in the likelihood of stroke, resulting in a substantial clinical and economic burden. • The efficacy of warfarin for stroke prevention in AF patients is well established, but the literature consistently shows that it is underused even in patients at high risk of stroke. • Maintaining adherence to warfarin therapy is challenging.
Previous data have shown that 1 in 4 AF patients discontinue warfarin therapy within the first year of receiving their initial prescriptions.

What is already known about this subject
• High warfarin adherence (proportion of days covered > 80%) was associated with fewer inpatient admissions and emergency room visits and lower total health care cost, compared with no use of warfarin at all. These data underscore the economic and clinical benefits of improving oral anticoagulant use in AF patients.
been continuously enrolled in the health plan for at least 12 months prior to the AF index date and at least 2 months after the AF index date to be eligible for inclusion. Eligible patients were required to have moderate to high stroke risk as assessed by a CHADS 2 score ≥ 2 17 and did not have high bleed risk as assessed by an ATRIA bleeding risk score < 5, 18 the HEMORR 2 HAGES score < 4, 19 or the HAS-BLED score <3. 20 ICD-9-CM codes associated with inpatient and outpatient medical claims in the 12-month pre-index period were used to determine baseline stroke and bleeding risk scores. Patients with ICD-9-CM codes corresponding to contraindications to warfarin, according to the package insert for Coumadin, as well as those presenting with valvular and transient AF were excluded (see Appendices). Patients with warfarin prescription claims or prothrombin/INR (PT/INR) claims in the 12 months prior to AF index date were also excluded.

Key Variables of Interest
Warfarin exposure during the 18-month follow-up period after the AF index date was identified by the presence of outpatient prescription claims as well as the timing of PT/INR claims. Consistent with the algorithm validated by Go et al. (2003), 21 patients were considered to be continually on warfarin when the gaps between 2 prescription claims were ≤ 60 days apart. When the gaps between 2 prescription claims were > 60 days apart, patients were considered to be continually on warfarin if there was a PT/INR claim every 42 days. If there was no PT/ INR claim every 42 days, then patients were considered off warfarin from the thirty-first day after the end of the first prescription days of supply until the start of the next prescription fill date. The level of warfarin exposure for patients who had at least 1 warfarin prescription claim was assessed using proportion of days covered (PDC), calculated as the total days of supply associated with warfarin prescription claims during the follow-up period (less any overlapping days of supply) divided by the length of follow-up after the AF index date. The level of warfarin exposure was classified as high (PDC > 0.8) and low (PDC ≤ 0.8), and patients who did not fill a warfarin prescription after the AF index date were considered to have a PDC = 0 (no warfarin exposure). The use of the 0.8 cut point was based on conventions used in other conditions 22 and based on the distribution of PDC values of warfarin users in our sample to ensure there was an adequate number of subjects in each PDC group.
Health care resource utilization and costs during the followup period for the no warfarin exposure group (PDC = 0) were compared with those with high and low PDC groups. Inpatient hospitalizations, length of stay (LOS), emergency room (ER) visits, and outpatient office visits during the 18-month period after the AF index date was assessed. All cost variables were evaluated from the payer perspective and were based on therapy. 10 Contributing further to AF morbidity and mortality is poor adherence with warfarin therapy and high discontinuation rates among those who received prescriptions. 11,12 According to a recent study of 4,188 Kaiser Permanente (U.S.) patients with AF, more than 1 in 4 patients (26%) starting on warfarin discontinued therapy in the first year despite a low overall hemorrhage rate (2.3% of patients). 13 Suboptimal utilization of OACs can lead to higher health care cost associated with strokes that would be prevented by effective anticoagulation therapy. As oral anticoagulation is also associated with an increased risk of bleeding, 14 risk-benefit assessment is crucial in the anticoagulation treatment decision. Data on the economic impact of OAC underuse are limited. Based on decision-analytic modeling, Caro (2004) 6 estimated that optimizing INR (international normalized ratio) control among AF patients that received warfarin and increasing warfarin use in AF patients could result in a reduction of health care costs by $2.4 billion per year. 6 The objective of this study was to estimate the economic burden from the payers perspective associated with suboptimal utilization of warfarin therapy in a commercially insured AF population with moderate to high stroke risk without contraindications for warfarin treatment. This study utilized medical claims data to identify clinical events and costs and prescription claims data to assess warfarin use, which provides a more precise estimate of the economic implications of warfarin underuse and nonadherence than previous economic modeling.

■■ Methods Study Design and Data Source
This study was a retrospective, observational, quasi-experimental study using de-identified health care claims data from the Thomson Reuters' MarketScan Commercial Claims and Encounters and Medicare Supplemental and Coordination of Benefits databases from January 2003 to December 2007. These databases include individual-level enrollment and cost data across inpatient, outpatient, and prescription drug services from 45 large employers, health plans, and government and public organizations. The MarketScan database has been used in published research conducted in AF patients previously. 9,15,16

Study Population
The study sample consisted of adult incident AF patients who would be candidates for OAC. Specifically, AF was identified based on the presence of at least 2 medical claims with primary or secondary AF diagnoses (using the International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9-CM] diagnosis code 427.31) separated by at least 30 days and not more than 12 months, of which at least 1 claim was in the outpatient setting. The date of the first AF medical claim was defined as the AF index date. Patients were required to have amounts paid by the insurer to the provider (i.e., paid claims). Inpatient cost included services rendered during a hospital stay. Outpatient cost included physician office visits, hospital outpatient visits, ER visits, outpatient laboratory tests, medical procedures, and radiological exams, excluding outpatient prescription drug cost. Medical cost was defined as the total cost of all inpatient and outpatient medical services. Prescription drug cost included all outpatient pharmacy claims. Due to the variable length of follow-up for eligible study patients, cost measures were expressed on a per-patient per-month basis.
Ischemic stroke or bleeding events during the post-index period were assessed in each PDC group based on the presence of medical or inpatient claims with either a primary or secondary diagnosis (Table 1). Since the occurrence of strokes or bleeds may have resulted in the initiation or termination of warfarin therapy, which would influence warfarin exposure PDC levels, we censored the warfarin PDC calculation to either the day prior to the occurrence of such an event or at the end of the 18-month follow-up period, whichever occurred first.

Statistical Analyses
Descriptive analyses were used to compare baseline characteristics among patients at different PDC levels. A series of mul-tivariate regressions were performed to examine the effect of warfarin exposure level on health care resource use and costs. Negative binomial regression was used to assess health care resource use variables and a generalized linear model (GLM) with gamma distribution and log link was used to assess cost outcomes based on the skewed distribution encountered in our data and previous modeling studies of cost. 23 When the proportion of patients with zero costs was > 10%, two-part modeling using a preliminary logistic regression to obtain probabilities of incurring cost given the independent variables in the model, followed by a GLM on the subset of the population that incurred cost, was performed.
Covariates in the regression models included patient demographics, insurance type, comorbidities, stroke (CHADS 2 ) and bleed (ATRIA) risk factors, and pre-index measures of the same economic outcome variable under evaluation. Pre-index health care resource use and costs were identified by medical claims in the 12-month period before the AF index date. Comorbidities from the Elixhauser Comorbidity Index 24 were identified based on ICD-9-CM codes associated with inpatient and outpatient claims during the 12-month pre-index period. Due to a high degree of multicollinearity between ATRIA, HEMORR 2 HAGES, and HAS-BLED bleeding risk scores (Pearson correlation coefficient = 0.6), we controlled for ATRIA bleeding risk score 13 in the models because the ATRIA risk factors can be more reliably defined using medical claims compared with the other indices. To account for variable follow-up time and to control its effect on cost variance, weights were determined based on the months of enrollment in the study divided by the study period (18 months) for each observation. All data management and analyses were conducted using SAS version 9.3 (SAS Institute Inc., Cary, NC).

■■ Results Population Characteristics and Warfarin Utilization
The study sample consisted of 13,289 patients with a CHADS 2 score ≥ 2 and no obvious contraindication to warfarin ( Figure  1). Nearly half of the patients (n = 6,253; 47%) did not receive warfarin, and only 21.5% (n = 2,852) of patients had a PDC > 0.80. Patients who did not receive warfarin were more likely to be ≥ 75 years of age, and they had a slightly higher CHADS 2 score (mean score = 2.37 vs. 2.33; P < 0.001) partly due to the higher prevalence of congestive heart failure than those who received warfarin. Patients who did not receive warfarin also had slightly higher bleeding risk on average as assessed by all 3 bleeding risk indices. Of the patients who did receive warfarin, 65% did so within 30 days of the index diagnosis. The mean follow-up time for patients not exposed to warfarin was 14.2 months, significantly lower compared with 15.6 months (P < 0.001) for patients who received warfarin (    After controlling for covariates and baseline utilization, the high warfarin PDC group was associated with 27% fewer inpatient encounters (P < 0.001), 16% fewer ER visits (P = 0.019), and nearly a 40% reduction in the number of hospital days (P < 0.001) compared with patients who did not receive warfarin ( Figure 2). In contrast, inpatient and ER services utilization were not significantly different between patients with low warfarin PDC and patients who did not receive warfarin. Patients who received warfarin, regardless of PDC level, had all types of bleeding events (major, minor, and other), patients who received warfarin had a significantly higher bleeding rate than patients who did not receive warfarin (P < 0.05). There was no significant difference in the rate of intracranial hemorrhage between patients who received warfarin and patients who did not receive warfarin. However, the major gastrointestinal (GI) hemorrhage rate was significantly higher among patients with warfarin PDC ≤ 0.80 than patients who did not receive warfarin. The risk of major GI bleeding was similar between patients with high warfarin PDC and patients who did not receive warfarin. Table 4 summarizes unadjusted health care service utilization differences between the pre-index and post-index follow-up period. Differences in post-and pre-index hospitalizations (-0.008 vs. 0.024; P < 0.001) and LOS (-0.011 vs. 0.095; P < 0.001) were significantly lower for the high warfarin PDC

Unadjusted Mean (SD) Utilization Across Warfarin Exposure Groups During the Pre-and Post-index Periods, Per Patient Per Month
significantly more outpatient visits in the post-index period than patients who did not receive warfarin. Patients with low and high PDCs had 21% and 32% greater outpatient visits, respectively, during follow-up compared with those who did not receive warfarin (P < 0.001 for both).

Health Care Costs
Increases in pre-versus post-index period inpatient, outpatient, and total cost were significantly lower for the high PDC group, compared with the group with no warfarin exposure (15.45 vs. 133.46, P = 0.002; 121.94 vs. 187.31, P < 0.001; 187.35 vs. 339.71, P < 0.001, respectively). Patients with high PDC showed a decrease in stroke-related inpatient cost in the postindex period (-$11.14), significantly lower than patients with no warfarin use who showed an increase in these costs in the post-index period ($20.20; P = 0.0006). Additionally, outpatient stroke-related cost increases from the pre-index period were significantly lower for the high warfarin PDC group compared with patients with no warfarin exposure (1.36 vs. 8.00, P < 0.0001; Table 5). Adjusted mean pharmacy cost was significantly higher for both patients with low (adjusted cost ratio = 1.08; P < 0.001) and high PDCs (adjusted cost ratio = 1.16; P < 0.001) than patients who did not receive warfarin (Table 6). After controlling for covariates and pre-index health care cost, adjusted mean allcause inpatient (adjusted cost ratio = 0.88; P < 0.001), all-cause outpatient (adjusted cost ratio = 0.73; P < 0.001), and all-cause   medical (adjusted cost ratio = 0.72; P < 0.001) costs were significantly lower in patients with high warfarin PDC than patients who did not receive warfarin. Stroke-related medical cost was 60% lower (adjusted cost ratio = 0.40; P < 0.001) and bleed-related medical cost was 22% lower (adjusted cost ratio = 0.78; P < 0.05) for patients with high warfarin PDC compared with those with no warfarin exposure. Patients with low warfarin PDC had significantly lower mean adjusted all-cause inpatient cost (adjusted cost ratio = 0.90; P < 0.001) but similar all-cause outpatient cost (adjusted cost ratio = 0.96; P = 0.09) and all-cause medical cost (adjusted cost ratio = 0.97; P = 0.27) compared with patients who did not receive warfarin. Stroke-related medical cost was 48% lower for patients with low warfarin PDC compared with patients with no warfarin exposure. Overall, total health care costs were 13% lower for patients with high PDC (P < 0.001) but similar for patients with low PDC compared with patients who did not receive warfarin.

■■ Discussion
Our study found that only 53% of patients with moderate to high risk of stroke and not at high bleeding risk received warfarin therapy. Even among patients who received warfarin, only 40% were adherent to warfarin with a PDC > 0.80. Previous studies using administrative claims and patient registries have found warfarin underuse to be a prevalent problem in AF patients 9,10 and that maintaining optimal INR control 25 and minimizing unintended discontinuation of warfarin therapy to be a significant challenge among patients who had received warfarin therapy. 13,26 Our study represents one of the first attempts to estimate the economic burden associated with warfarin underuse and nonadherence. Although AF patients who received warfarin had higher pharmacy cost and incurred more outpatient visits, adherence to OAC therapy resulted in fewer inpatient admissions and ER visits, leading to significant reductions in total health care cost. Greater adherence as measured by PDC, compared with no use, clearly shows beneficial economic results in terms of lower total, inpatient, and outpatient costs during follow-up. Ischemic stroke rates for the low and high warfarin exposure groups were significantly lower than the no warfarin exposure group. However, there was not a significant difference in the ischemic strokes rates between the high and low warfarin PDC groups. While there was a greater rate of GI bleeds among the low PDC group compared with those not taking warfarin (P = 0.001), we did not find a significant difference between the high PDC and no warfarin group. This observation may be explained through a widely established association between poor adherence to warfarin and unstable INR, which has been shown to increase bleeding risk. 13,20 This relationship between poor adherence and bleeding risk may also explain the significantly greater number of observed ER encounters for the low PDC group.

Generalized Linear Model Estimation of Adjusted Mean Monthly Cost During Study Follow-up Period a
The findings of our study should be viewed in light of several limitations. First, although the study used a large nationally representative database of patients recently diagnosed with AF, it may be subject to selection bias. The MarketScan database under-represents the U.S. Medicare population, a dominant patient segment in AF. Additionally, in working with large sample sizes, observed differences may be statistically significant but clinically irrelevant. For example, our observed baseline difference in the CHADS 2 score (mean score = 2.33 vs. 2.37; P < 0.001) between patients that used and did not use warfarin may not translate to notable clinical difference in stroke risk. Patients were required to have at least 2 months of continuous eligibility after the AF index date. Patients who died within 2 months of the index AF diagnosis were not included in the study. By excluding patients who may have died from stroke or intracranial bleeding, the study can potentially over-or underestimate the cost of warfarin underutilization. Second, as with other retrospective claims data analysis, adherence in this study was measured by prescription claim records. In the absence of additional clinical information, we were unable to determine how well patients follow the instructions given by their health care providers and if discontinuation of warfarin therapy is medically necessary. Third, we defined eligibility of OAC therapy using CHADS 2 and published bleeding risk algorithms. Although the CHADS 2 score has been available and validated for stroke risk stratification, it remains unclear to what extent it has been routinely used in clinical practice to guide anticoagulation decisions. Current treatment guidelines for anticoagulation do not specify a standard instrument that should be used for bleeding risk assessment. 8 Limited data exist on how well these bleeding risk scores can predict bleeding in clinical practice and to facilitate risk-benefit assessment of OAC treatment decisions. More data on the validity of these scores in clinical practice are needed.
This study identifies a significant economic and clinical burden driven by previously established warfarin underutilization among patients with AF. These findings reveal a missed opportunity to reduce costs by avoiding strokes that can be effectively prevented by OAC therapy. The results underscore the need for actions by health insurers and policy makers to develop programs to overcome these challenges in prescribing OAC therapy.

■■ Conclusion
The economic implications of underutilizing warfarin in nearly half of the commercially insured nonvalvular AF population are significant. Adherence in terms of greater warfarin exposure was shown to reduce health care resource use as well as overall health care cost. Given the chronic nature of AF, improving adherence is critical to attaining the clinical and economic benefits of OAC therapy.

ICD-9-CM Diagnoses for HEMORR 2 HAGES Bleeding Risk Factors
Hypertension Same codes as used in the CHADS 2 risk scheme (Appendix C) Ethanol abuse Same codes as used in HEMORR 2 HAGE risk scheme (Appendix E) Stroke Same codes as used in the CHADS 2 risk scheme (Appendix C) Prior bleed Same codes as used in the ATRIA risk scheme (Appendix D), also including anemia codes LABILE INRs Not available in the data; will be assumed to be 0 for all patients Abnormal renal or liver function Same codes as used in HEMORR 2 HAGE risk scheme (Appendix E) ICD-9-CM = International Classification of Diseases, Ninth Revision, Clinical Modification; INR = international normalized rate.