Economic and Clinical Impact of Stroke and Warfarin Use for Patients with Non-valvular Atrial Fibrillation

Background: Atrial fibrillation (AF) is a common clinical problem and potent risk factor for stroke. However, real-world effectiveness and outcomes for AF patients are not well described. Objective: To compare the economic and clinical impact of stroke and warfarin use on patients with nonvalvular atrial fibrillation (NVAF). Methods: This was a retrospective analysis of medical and pharmacy claims of NVAF patients from a large commercial health insurance database (01/01/2005-12/31/2007). Patients were grouped according to stroke or warfarin prescription status. For all groups, demographic, clinical, and pharmaceutical characteristics were analyzed descriptively. Risk-adjusted overall and cardiovascular-related hospital readmission rates in 30 days, length of stay (LOS), clinical outcomes, and health care costs were assessed using propensity score matching. Costs were adjusted to 2007 U.S. dollars using the medical component of the U.S. Consumer Price Index. Results: Out of 18,575 NVAF patients, 3.1% (n=575) experienced a stroke event. Stroke patients were older on average (78.94 vs. 77.28 years, p-value<0.0001) with significantly higher risk-adjusted inpatient mortality (7.14% vs. 2.09%, p-value<0.0001), emergency room visits (79.97% vs. 46.34%, p-value<0.0001), and average LOS measures (overall: 10.20 vs. 6.83 days, p<0.0001; cardiovascular-related: 8.35 vs. 5.90 days, p-value<0.0001). Despite the similarity in Charlson Comorbidity Index scores compared to non-stroke controls, stroke patients significantly higher clinical outcome rates during follow-up for acute coronary syndrome (ACS), ischemic attack, major and minor bleeding patients (p-values<0.0100), and the total cost incurred was nearly three times greater ($33,506 vs. $13,921, p-values<0.0001). NVAF patients were commonly prescribed warfarin (65.60%) and appeared to have a lower prevalence of clinical outcomes, while not incurring significantly higher follow-up costs compared to those not prescribed warfarin ($12,739, standard deviation [SD]=$19,842 vs. $15,358; SD=$45,446; p-value>0.0500). However, a significantly greater proportion of patients with major and minor bleeding were prescribed a combination of warfarin and antiplatelets than those without these events. Conclusions: A stroke after an NVAF diagnosis has a major clinical impact, which translates into a significant economic burden for patients. Warfarin prescriptions did not significantly impact total health care costs, though caution is advised to minimize hemorrhagic events.


INTRODUCTION
Atrial fibrillation (AF) is the most common significant cardiac arrhythmia, affecting an estimated 2.3 million adults in the United States.Its prevalence increases with age, approximately doubling each decade in individuals over age 50 years.In those over age 65 years, AF prevalence is estimated to be 6% 1 and roughly 10% in persons age 80 years or older. 2,3 patients have six times the risk of ischemic stroke, due to cardiogenic embolism, than age-matched controls, resulting in approximately 80,000 strokes per year in North America. 4The risk of stroke in AF patients increases with advancing age and the presence of concomitant risk factors including left ventricular dysfunction, history of hypertension, stroke or transient ischemic attack (TIA), and diabetes mellitus (DM). 5Medical therapy has been shown to ameliorate several complications associated with AF, particularly with regard to reducing stroke risk. 6Among appropriate diagnosed AF patients, anticoagulation can reduce the risk of a first stroke by approximately 68%, 7 and it is the recommended therapy in published guidelines. 8Therefore, an expert panel has recommended that all AF patients over age 75 should be considered for chronic anticoagulation therapy, unless a contraindication exists.
Elderly patients with an ischemic stroke associated with AF are at especially high risk for a recurrent stroke with an annual rate of more than 10% per year. 9Secondary preventative therapy with warfarin is highly effective in reducing this risk.The recurrence rate can be reduced by two thirds with warfarin therapy.Warfarin use was also confirmed to decrease mortality. 10 is a common clinical problem and potent risk factor for stroke, yet there is a lack of studies focusing on the real-world effectiveness and safety of outcomes for AF patients.

Study Sample
This study used data from a large U.S. commercial database from 2003 to 2008.Table 1 represents patient attrition.The prevalence of stroke was analyzed in selected patients aged 65 years or older with claims for at least two primary diagnoses for AF (determined using the International Classification of Diseases 9 th Revision Clinical Modification [ICD-9-CM] codes in Appendix Table A1) occurring within 30 days of one another.Additional criteria for study selection included continuous health plan enrollment for at least 180 days prior to the initial AF diagnosis date (designated as the index date), and the earlier date of at least 180 days following the first AF diagnosis date or until death.
Patients were excluded if they had at least one medical claim for any the following: transient AF caused by a reversible disorder, known presence of atrial myxoma, left ventricular thrombus, active endocarditis, any diagnosis of arterial or venous thromboembolism (VTE), use of selected antiplatelet or anticoagulant agents (low-molecular-weight [LMWH] and unfractionated heparin [UFH], fondaparinux, anisindione, abciximab, cilostazol, anagrelide, pentoxifylline, enoxaparin, dalteparin, or clopidogrel), before the first AF diagnosis.Patients must have had continuous enrollment for at least 180 days before and after the first prescription date.From this final sample (n=19,268), cohorts were assigned; patients were assigned to appropriate cohorts.

COVARIATES AND OUTCOME VARIABLES Patient Characteristics
Many aspects of health care utilization and cost, including treatment selection, therapy patterns, and health outcomes, may be associated with factors not directly measured in administrative claims data.Vast literature exist demonstrating differences in a variety of health-related outcomes for patients of differing educational attainment, income, net worth, race and ethnicity and family structure.Therefore, various patient and clinical factors to estimate risk-adjusted comparisons were captured.
Age was defined as of the index date.In addition, patient age was used to assign them to the following age groups: 65-74, 75-84, 85-94, 95+.Median and maximum values were also provided.Gender and U.S. region variables (Northeast, North Central, South, and West) were included as well.The most commonly used comorbidity index in health outcomes studies is the Charlson Comorbidity Index (CCI), which assigns a weight ranging from 1 to 6 according to disease severity for 19 conditions. 11The index has since adopted several weights, some of which allow outpatient diagnoses to contribute to the score. 12,13morbid Conditions The Appendix Table lists disease codes for comorbid conditions and clinical outcomes measured for the baseline period (≥180 days pre-index date).Patients were flagged for the following conditions: end stage renal disease (ESRD), congestive heart failure (CHF), peripheral arterial disease (PAD), acute coronary syndrome (ACS), hyperthyroidism, obesity, DM, hypertension, ischemic stroke, hemorrhagic stroke, non-central nervous system (CNS) systemic embolism, TIA, catheter ablation, and dyspepsia.Studied medications included warfarin, injectable anticoagulants (enoxaparin, dalteparin, and fondaparinux), antiplatelets (plavix), and combinations thereof.

Outcome Variables
A flag was created for each of the following variables occurring after the index date (Table 2): in-hospital mortality, average number of health care visits per patient per year (pppy), readmission rates 30 days after discharge, length of stay (LOS) (all-cause and cardiovascular-related), ACS, ischemic stroke, hemorrhagic stroke, TIA, major bleeding, non-major clinically relevant bleeding, non-CNS systemic embolism, surgical ablation, cardioversion, intra-cardiac device, myocardial infarction (MI), osteopathic fracture, intracranial hemorrhage (ICH), and gastrointestinal (GI) bleeding.
Average total annual health care costs per patient were captured as well.Since the analysis is from the payer's perspective, reimbursement amounts rather than charges were applied.Expenditures were expressed in 2007 U.S. dollars and were adjusted using the medical component of the U.S. Consumer Price Index.

STATISTICAL ANALYSIS
Baseline and outcome measures were descriptively analyzed using numbers/percentages for dichotomous and means/standard deviation (SD) for continuous variables.Bivariate comparisons were implemented using t-tests or chi-square tests when appropriate with standardized differences included.Non-parametric tests (e.g. the Mann Whitney U-test) were applied if there was a deviation from asymptotical assumptions.
Multivariate analysis was performed using 1:1 propensity score matching (PSM).Propensity scores were estimated via unconditional logistic regression analysis to yield risk-adjusted estimates (or remove overt bias).
Patients were matched if their propensity scores were within ±0.01 units of one another.Covariates in the logistic regression model included age, gender, baseline CCI score, baseline comorbid conditions, baseline health care utilization, and baseline health care costs.To estimate total cost, a generalized linear model (GLM) with gamma distribution and log link function with cost as the dependent variable and demographic/clinical factors used in PSM as independent variables.

DISCUSSION
Stroke in NVAF patients is a clinically relevant event with increased spending implications in Medicare Supplemental patients.Warfarin was commonly prescribed to NVAF patients without a significant increase in total health care expenses.Consideration of the dose intensity of warfarin may help avoid hemorrhagic complications.
An established adverse event associated with warfarin use is hemorrhage, particularly when in combination with antiplatelets. 14,15Observational data from this study generally support this claim; a significantly greater proportion of patients with major and minor bleeding were prescribed a combination of warfarin and antiplatelets than those without these events.Absolute frequencies of warfarin and antiplatelet use were higher in GI bleeding and ICH patients, though their small sample sizes may have contributed to the lack of statistical significance, compared to controls without these conditions.These considerations should be kept in mind when creating an optimal treatment plan for NVAF patients.
A major strength of this study involves the use of a large database of nationwide, physician-designated hospital diagnoses with real-world spending figures from a recent time period.While claims data are extremely valuable for the efficient and effective examination of health care outcomes, treatment patterns, and costs, claims data are collected for the purpose of payment and not research.Therefore, there are certain limitations associated with the use of claims data.The presence of a claim for a filled prescription does not indicate that the medication was consumed or that it was taken as prescribed.Additionally, the presence of a diagnosis code on a medical claim is not positive presence of disease, as the diagnosis code may be incorrectly coded or included as rule-out criteria rather than actual disease.Finally, certain information is not readily available in claims data that could have an effect on study outcomes, such as clinical and disease-specific parameters.
It should be noted that neither regression adjustment addresses problems due to imbalances in unmeasured factors.It is quite possible that outcomes for patients with the same observable characteristics can vary because of unobservable factors such as physician or practice-prescribing patterns.There are several methods that exist to control for unmeasured factors such as the instrumental variable approach, bounding approach and a difference-in-difference estimator.However, these estimators are also confounded by their own limitations.

CONCLUSIONS
A stroke after an NVAF diagnosis has a major clinical and economic impact on patients.Stroke-experienced patients had higher average clinical outcome rates than non-stroke controls as well as more frequent health care resource use (ER visits and longer average hospital LOS) and higher inpatient mortality.These factors contributed to annual total health care costs for stroke-experienced patients nearly three times greater than for those who did not experience a stroke.
Stroke incidence is nearly 5-fold higher in AF patients, so diagnosed patients must be monitored carefully and provided with appropriate pharmacotherapy to avert stroke events. 16,17NVAF patients were commonly prescribed warfarin and appear to have a lower prevalence of clinical outcome conditions, while not incurring significantly higher follow-up costs compared to those not prescribed warfarin.
It should be noted that warfarin patients were also younger than controls on average and had a less complex comorbidity profile than patients without a warfarin prescription.Additional research would be necessary to confirm and explain the real-world outcomes seen in this population in a causative manner.

Table 2 .
Descriptive Statistics for NVAF Patients who did or did not Experience a Stroke

Pharmacotherapy Pattern (AF diagnosis date -June 2007)
Multivariate analysis was performed using 1:1 propensity score matching (PSM).Propensity scores were estimated via unconditional logistic regression analysis to yield risk-adjusted estimates (or remove overt bias).Patients were matched if their propensity scores were within ±0.01 units of one another.Covariates in the logistic regression model included age, gender, baseline CCI score, baseline comorbid conditions, baseline health care utilization, and baseline health care costs.To estimate total cost, a generalized linear model (GLM) with gamma distribution and log link function with cost as the dependent variable and demographic/clinical factors used in PSM as independent variables.

Table 5 .
Risk-adjusted Outcomes for NVAF Patients Prescribed Warfarin or Not

Table 6 .
Pharmacotherapy Patterns for Patients with Bleeding/Hemmorhagic Outcomes after AF Diagnosis