Matthew Sussman, MA

: To assess comparative resource use for patients given a new diagnosis of nonvalvular atrial fibrillation (NVAF) and newly treated with dabigatran or warfarin in a US real-world setting.

: A retrospective cohort analysis was conducted using de-identified electronic health record data from a large, nationwide database of integrated delivery networks (IDNs).

: We identified newly diagnosed NVAF patients, newly treated with dabigatran or warfarin ≤90 days of diagnosis between October 1, 2010, and September 30, 2012. Patients had ≥1 year of data prior to initiating therapy (index date). Patients in the warfarin cohort were propensity score matched 1:1 to patients in the dabigatran cohort and were followed up to 1 year after index to assess all-cause, stroke-related, and bleed-related healthcare resource use.

: A total of 3890 patients (1945, dabigatran; 1945, warfarin) were included and balanced through the matching process. The probability of an all-cause hospitalization was significantly lower in the dabigatran cohort compared with the warfarin cohort (42.8% vs 47.5%; P = .007). The dabigatran cohort had significantly lower per-patient per-year (PPPY) hospital (1.07 vs 1.20), emergency department (ED) (0.36 vs 0.51), and physician office visits (10.64 vs 18.13), as well as mean length of hospital stay (4.0 vs 4.6 days), compared with the warfarin cohort (all P <.001). Total PPPY stroke-related hospitalizations and physician office visits, and total PPPY bleed-related ED visits, were also significantly lower for the dabigatran cohort compared with the warfarin cohort (all P <.05).

: Newly diagnosed NVAF patients, newly treated with dabigatran and receiving care in US IDNs, experienced fewer all-cause healthcare resources than patients initiating warfarin.

Nonvalvular atrial fibrillation (NVAF), which refers to atrial fibrillation (AF) without valvular disease, accounts for more than 95% of all cases of AF

 in the United States. Clinical trial data have shown that oral anticoagulation therapy substantially reduces the risk of stroke in patients with AF.

Warfarin is the most prevalent anticoagulant prescribed in the United States today.

 The drug has been shown to reduce the risk of stroke by about two-thirds compared with placebo.

 However, warfarin is associated with a number of therapeutic challenges. For instance, warfarin requires frequent monitoring to maintain international normalized ratio (INR) within a narrow therapeutic range (2.0-3.0).

For INR values that fall outside of the optimal therapeutic range, the risk of hemorrhage increases with the level of INR,

 with yearly rates of major bleeding ranging from 0.5% to 6.5% and fatal bleeding ranging from 0.1% to 1.0% among warfarin-treated patients.

 Additional challenges of warfarin use include the ability of dietary vitamin K intake and concomitant medications to interact with warfarin, which may influence INR maintenance and the time spent within therapeutic range.

Dabigatran, approved by the FDA in October 2010, is the first novel oral anticoagulant (NOAC) to become available for clinical use in more than 50 years.

 It has been shown to be associated with significantly lower rates of stroke, systemic embolism, and intracranial hemorrhages compared with warfarin.

Due to the recent emergence of pharmaceutical alternatives to warfarin, payers in the United States have been faced with the option of placing NOACs on their formularies, thereby intensifying the need for evaluations of the overall economic impact—particularly healthcare resource utilization and direct medical costs&mdash;across treatment types.

Limited real-world data exist on healthcare utilization associated with NOAC therapies; as such, the primary objective of this analysis was to assess comparative resource use for patients receiving a new diagnosis of NVAF, and newly treated with dabigatran or warfarin, using de-identified electronic health record (EHR) data from integrated delivery networks (IDNs). With an emphasis on the continuum of healthcare services, this unique data source provides additional insight on real-world estimates of resource utilization for this patient population.

A retrospective cohort analysis was conducted using de-identified EHR data from October 1, 2010, through September 30, 2013. During the study period, the Humedica EHR database contained de-identified data on more than 30 million individuals across 38 states. Humedica integrates claims, prescription, and practice management data by partnering with medical groups, IDNs, and hospital chains to directly obtain data from EHRs in real time.

Approximately 40% of patients in the Humedica database have linked inpatient and outpatient records, allowing for a complete analysis of a patient’s full experience in the IDN system. This subset of patients in the EHR who are from an IDN setting formed the basis for the patient population in this study.

The population of interest for this analysis included adult patients ≥18 years, newly diagnosed with NVAF and newly treated with an oral anticoagulant (OAC), either dabigatran or warfarin. The date of the first OAC prescription was defined as the index date, which must have occurred between October 1, 2010, and September 30, 2012, to allow for 1-year baseline and follow-up periods.

To ensure cohorts of newly treated patients, patients were only included if they had no prescriptions for any OAC during the 12-month pre-index period. Additionally, to ensure cohorts of newly diagnosed NVAF patients, the first observed NVAF diagnosis (identified by International Classification of Diseases, Ninth Revision, Clinical Modification code 427.31 in any position) during the 12-month pre-index period was required to appear on the index date or within 3 months prior to the index date.

Patients were required to have ≥1 EHR encounter in the 12 months prior to the index date (defined as the baseline period) and ≥1 EHR encounter following the index date (defined as the follow-up period). Furthermore, patients must have had integrated inpatient and outpatient data, which allow for a complete picture of resource utilization in the IDN setting. Patients in the warfarin cohort were propensity score matched 1:1 to patients in the dabigatran cohort, and only matched pairs were included in the final data set.

Patients were required to have ≥2 prescriptions of the index OAC (including the prescription on the index date) for study inclusion. Additionally, on the index date or within 3 months prior to the index date, patients were required to have ≥1 inpatient, ≥2 physician office visits, ≥2 emergency department (ED) visits, or ≥1 physician office visit and ≥1 ED visit on distinct service dates with an NVAF diagnosis.

Patients with the following were excluded from the study sample: diagnosis of hyperthyroidism (potentially reversible underlying cause of AF) or valvular heart disease during the 12 months pre-index or an indication of cardiac surgery or diagnosis of pericarditis, myocarditis, or pulmonary embolism within 3 months prior to the first NVAF diagnosis.

Patients were followed for a maximum of 12 months. The follow-up period commenced on the index date and ended on the earlier of the following: (a) last EHR encounter date in the observed data or (b) the last day of the 12-month postindex period.

Additional Details on the Matching Process

Patients in the warfarin cohort were matched 1:1 to patients in the dabigatran cohort using the propensity score matching (PSM) method. The propensity score was defined as the probability of being treated with dabigatran (dependent variable) based on a set of baseline characteristics, including sex, age, geographic region, month of index date, Deyo-Charlson comorbidity index, CHADS2 stroke risk score, HEMORR2HAGES bleeding risk score, and selected comorbidities (ie, acute myocardial infarction, cardiomyopathy, congestive heart failure, chronic obstructive pulmonary disease/emphysema, diabetes, hyperlipidemia, hypertension, peptic ulcer/gastroesophageal reflux disease, ischemic stroke/transient ischemic attack, venous thromboembolism). PSM by logistic regression, with the Nearest Neighbor method within a caliper of 0.20 of the standard deviation of the estimated logit, was used to match dabigatran and warfarin cohorts.

Comparative healthcare resource utilization during the follow-up period was the primary outcome of interest for this analysis. Incidence rates for the first healthcare resource use were reported by type of service (ie, hospitalization, ED visit, physician office visit) for all-cause, stroke-related, and bleed-related events.

The total number of all-cause, stroke-related, and bleed-related resource utilization events were assessed and reported by type of service (ie, hospitalization, ED visit, physician office visit). Among patients with at least 1 hospitalization, the average length of stay per all-cause hospitalization was estimated.

Kaplan-Meier survival analyses were conducted to assess the time to first all-cause hospitalization, time to first stroke-related hospitalization, and time to first bleed-related hospitalization following index for each cohort. A log-rank sum test was used to test the significant differences between survival curves.

In addition to using a PSM method for identifying matched patients, a sensitivity analysis was conducted using a doubly robust approach in the form of a multivariate Cox proportional hazards model; this was to assess the association between OAC treatment and risk of first all-cause hospitalization, first all-cause ED visit, and first all-cause physician office visit.

 The primary covariate used in the model included the presence of dabigatran as the index therapy. Incidence rates were presented as a rate per 100 patient-years to allow for variable follow-up. Incidence rates were calculated using the number of patients identified in each cohort having an event during the follow-up period as the numerator. The denominator consisted of all patients with NVAF in the cohort (the denominator was censored based on the date of first all-cause or disease-specific event, last EHR encounter date in the observed data, or the last day of the 12-month postindex period). For each incidence rate, a 95% confidence interval was derived assuming a binomial distribution.

Outcomes tested for statistical significance (defined as P <.05) between the dabigatran and warfarin cohorts, including baseline demographic and clinical characteristics and number of healthcare resource utilization events. Chi-squared contingency tests were used for comparison of proportions, while nonparametric Wilcoxon rank sum tests were used to evaluate the statistical significance of differences in continuous variables.

Baseline Demographic and Clinical Characteristics

Prior to matching, 1945 patients with NVAF who met the inclusion criteria initiated treatment on dabigatran, while 11,473 initiated treatment on warfarin. The matched analysis included 1945 patients in the dabigatran and warfarin cohorts, nearly 58% of them male and with a mean age of approximately 70 years (Table 1).

Following the PSM process, mean CHADS2 and CHA2DS2-VASc scores were approximately 1.3 and 2.5, indicating moderate to high stroke risk, respectively.

 Additionally, the HEMORR2HAGES score was approximately 1.50, indicating low to moderate bleeding risk.

Coronary artery disease, diabetes, hyperlipidemia, and hypertension were the most common comorbid conditions identified among the matched study population. Dabigatran and warfarin patients were well matched on demographic and clinical characteristics, with no significant differences observed in age, sex, region, selected comorbidities, Deyo-Charlson comorbidity index, Elixhauser comorbidity index, stroke risk (CHADS2), and bleeding risk (HEMORR2HAGES). The average duration of follow-up was 342 days (±64) for the dabigatran index cohort and 332 days

Results from the Kaplan-Meier analyses demonstrated the time to first all-cause hospitalization to be significantly (P = .007) longer for the dabigatran cohort compared with the warfarin cohort (Figure 1). In fact, only 42.8% of the dabigatran cohort had an all-cause hospitalization at the end of 1 year compared with 47.5% of the warfarin cohort.

Similarly, the time to first stroke-related hospitalization was significantly (P = .024) longer for the dabigatran cohort (Figure 2). Moreover, just 3.5% of the dabigatran cohort had a stroke-related hospitalization at the end of 1 year compared with 4.9% of the warfarin cohort. Results from the analysis of time to first bleed-related hospitalization were not significant.

First all-cause hospitalization and ED visits per 100 patient-years were considerably lower in the dabigatran cohort compared with the warfarin cohort (Table 2). Incidence rates for first stroke-related and bleed-related events (hospitalizations, ED visits, physician office visits) were also lower in the dabigatran cohort.

The per-patient per-year (PPPY) total number of all-cause hospitalizations, ED visits, and physician office visits were all significantly (P <.001) lower for the dabigatran cohort compared with the warfarin cohort (mean all-cause PPPY hospitalizations, 1.07 vs 1.20; mean all-cause PPPY ED visits, 0.36 vs 0.51; mean all-cause PPPY physician office visits, 10.64 vs 18.13) (Table 2).

Additionally, the dabigatran cohort had a significantly lower mean length of stay for all-cause inpatient admissions (4.0 vs 4.6 days; P <.001) compared with the warfarin cohort. Moreover, the PPPY total of stroke-related hospitalizations and physician office visits was significantly lower for the dabigatran cohort compared with the warfarin cohort (mean stroke-related PPPY hospitalizations, 0.06 vs 0.10, P = .03; mean stroke-related PPPY ED visits, 0.00 vs 0.01, P = .65; mean stroke-related PPPY physician office visits, 0.16 vs 0.29, P = .02) (Table 2).

The PPPY total number of bleed-related ED visits were significantly (P = .02) lower for the dabigatran cohort, while the bleed-related hospitalizations and physician office visits were not statistically significantly different between cohorts (mean bleed-related PPPY hospitalizations, 0.05 vs 0.03, P = .49; mean bleed-related PPPY ED visits, 0.01 vs 0.03, P = .02; mean bleed-related PPPY physician office visits, 0.05 vs 0.15, P = .57) (Table 3).

Results of the sensitivity analysis using adjusted Cox regression models to predict risk of first all-cause hospitalization, ED visit, or physician office visit were consistent with the unadjusted matched results. In particular, treatment with dabigatran was associated with a significantly (P <.01) lower likelihood of a first all-cause event.

A total of 1945 matched patients, newly diagnosed with NVAF and newly treated with dabigatran or warfarin, were analyzed from the de-identified EHR IDN dataset. The time to first all-cause hospitalization and time to first stroke-related hospitalization were significantly longer for the dabigatran cohort.

Moreover, total PPPY all-cause resource use (hospitalizations, ED visits, and physician office visits), total PPPY stroke-related hospitalizations and physician office visits, and total PPPY bleed-related ED visits were significantly lower for the dabigatran cohort compared with the warfarin cohort during the 1-year follow-up period.

To the best of our knowledge, no other published articles of retrospective database design have compared resource utilization using IDN data in this population of newly diagnosed NVAF patients who have been newly treated with dabigatran or warfarin. Other recently published retrospective database studies of dabigatran users have focused on characteristics affecting OAC treatment selection among patients with NVAF,

 thus demonstrating an important need to measure economic outcomes in this population.

Using prospectively collected data from the RE-LY (Randomized Evaluation of Long-term anticoagulant therapY: dabigatran vs. warfarin) clinical trial, Connolly et al estimated the risk of stroke or systemic embolism among patients with AF randomized to 1 of 2 fixed dose arms of dabigatran (110 or 150 mg twice daily) or adjusted-dose warfarin.

Compared with the warfarin arm, the authors found a statistically significant reduction in the primary outcome measure associated with the 150-mg dose of dabigatran. Despite differences in sample selection criteria between the RE-LY trial and our EHR-based study, we, too, found a statistically significant reduction in stroke events (ie, stroke-related hospitalizations) among patients newly treated with dabigatran.

Our study is subject to certain limitations that are inherent in EHR data in general, primarily the lack of continuous eligibility/enrollment data, which potentially underestimates the level of observed utilization events. As a proxy for ensuring continuous eligibility, patients were required to have at least 1 medical encounter in the baseline period and at least 1 follow-up encounter, which assumes continuous enrollment in the health plan before and after index.

Additionally, National Drug Code (NDC) information was not reliably provided by data sources contributing to the EHR database, making it difficult to identify prescriptions. To avoid misclassifying prescriptions, this study used the drug name field, rather than the NDC field, for dabigatran or warfarin identification.

Articles

This study assesses comparative resource use for patients given a new diagnosis of nonvalvular atrial fibrillation treated with dabigatran or warfarin in a real-world setting.

Resource Use Among Nonvalvular Atrial Fibrillation Patients

Atrial fibrillation (AF), a common risk factor for stroke, is characterized by a rapid, irregular heartbeat that can cause blood to pool in the atria and increase the risk of blood clot formation.

 AF is the most common form of sustained arrhythmia, affecting more than 2 million people in the United States. The prevalence of AF has been increasing, and as the overall US population ages, it is predicted that more than 5.5 million people will be affected by 2050.

 Prevalence of AF among the general population is approximately 1%, but increases with age to approximately 9% for persons 80 years or older.

	AF has been shown to increase the risk of ischemic stroke (IS) 5-fold due to the high risk of clotting associated with the condition,

 and furthermore, patients with AF are at higher risk of increased stroke severity.

 Consequently, this may lead to higher costs given that the cost of severe stroke has been shown to be 11% to 71% higher when compared with minor stroke.

 Additionally, anticoagulation therapy to prevent IS can increase the risk for hemorrhagic stroke (HS) in patients with AF.

 Total direct costs attributable to AF have been estimated at $6.65 billion annually within the United States, including approximately 350,000 hospitalizations and 5 million office visits per year.

 Moreover, the total societal burden of stroke in the United States (ie, direct and indirect costs for all stroke types) has been estimated at $34.3 billion annually.1 Stroke costs have also been shown to vary across stroke types, with a recent analysis estimating the 4-year medical expenditures for IS and HS to be $39,396 and $48,327, respectively, among Medicare beneficiaries.

	Analyses of long-term observational data have suggested that age-adjusted stroke incidence has declined considerably over the past few decades

 However, limited recent data on stroke incidence among patients with AF exist, particularly from a US perspective.

 Similarly, despite the increased risk of stroke among patients with AF, limited data exist on the economic burden of stroke in this patient population. A 2013 literature review on the economics of anticoagulant therapy for stroke prevention in AF noted this gap in stroke-related cost data in the United States.15 Moreover, the literature search and review resulted in no published data comparing the economic burden of IS and HS among patients with AF.

	In recent years, the effectiveness of multiple new oral anticoagulants (OACs) for stroke prevention in AF patients has been evaluated in comparison to warfarin.

 Health plans would be required to invest considerable time and effort to undertake analyses to understand the epidemiology and costs of stroke in their enrolled populations. Addressing the current epidemiologic and economic data gaps outlined above will provide necessary estimates for conducting risk-benefit analyses for new stroke prevention strategies among an AF population.

	To facilitate the understanding of total cost of stroke in AF patients, we created a model based on key estimates derived from a large database that allows for the estimation of both IS and HS among patients with AF from the perspective of a representative health plan.

This study was a model simulation to estimate the cost of IS and HS in AF patients of a hypothetical health plan of 1 million members. The study involved 2 phases of data analysis. In the first analysis, annual stroke incidence rates among patients with AF, with stratification by stroke type (ie, IS and HS), were estimated by a retrospective claims data analysis. In the second analysis, age- and gender-specific annual stroke incidence estimates from the first analysis, along with 1-year resource use and cost estimates associated with IS and HS from a previously published study,

 were applied to estimate the annual cost of stroke care in the hypothetical health plan. Patients who experienced and survived a stroke hospitalization were assigned the cost associated with a nonfatal index event plus 1 year of stroke-related follow-up cost, while patients who did not survive a stroke event were assigned the cost of a fatal ischemic or hemorrhagic index event only. Annual healthcare costs for both stroke types were estimated for the hypothetical health plan by summing costs for all age- and sex-specific cohorts (

Both the annual stroke incidence analysis and the previously published resource use and cost of stroke study were performed using the combined administrative medical and pharmacy claims data of 2 Truven Health MarketScan databases: 1) the Commercial Claims and Encounters database, and 2) the Medicare Supplemental and Coordination of Benefits database. Data from January 1, 2005, to December 31, 2011, were used for the analyses. The MarketScan Commercial Claims and Encounters database has claims for approximately 100 self-insured employers and 12 health plans representing approximately 40 million covered lives in 2008. The population covers employees and dependents 64 years and younger. The MarketScan Medicare Supplemental and Coordination of Benefits database focuses on patients with Medicare coverage plus employer-paid commercial plans, and includes 2.5 million lives from 2008. The database includes both employer-paid and Medicare-paid components of healthcare.

Incidence of stroke in AF analysis: patient selection. 

To be eligible for the annual stroke incidence analysis, patients were required to meet the following selection criteria: adult patient (aged ≥18 years); at least 1 inpatient claim or 2 outpatient claims for AF (

International Classification of Diseases, Ninth Revision, Clinical Modification

] code 427.31) during the identification period (ie, January 1, 2006, to November 30, 2011)19; at least 12 months continuous enrollment prior to the initial AF claim (ie, baseline period); and no stroke diagnoses in the 4-to-12-month period prior to the initial AF claim (a stroke diagnosis may have occurred in the 3 months prior to the index AF diagnosis to allow for undiagnosed AF prior to the stroke event). The date associated with the first AF diagnosis was termed the index date.

Incidence of stroke in AF analysis: study measures & data analysis. 

Stroke events were identified by the presence of at least 1 inpatient hospitalization claim with a primary diagnosis for stroke (

 codes 433.01, 433.11, 433.31, 434.xx for IS; 430.xx, 431.xx, 432.xx for HS) either in the 3 months prior to the index AF claim or in the 12 months following the index AF claim. Annual incidences were evaluated for each calendar year from 2006 to 2011 as rates per 100 person-years to allow for the fact that some patients might not have enrolled for a full year in the MarketScan database. Incidence rates were stratified based on age, sex, and stroke type. Incidence rates were calculated using the number of patients identified with having a stroke event during a given calendar year (either in the 3 months prior to or during the 12 months following the index AF claim) as the numerator. The denominator consisted of all AF patients identified in a given calendar year (the denominator was censored based on the date of a stroke event or disenrollment from the plan). For each incidence rate, a 95% confidence interval (CI) was derived assuming a binomial distribution. All analyses were performed using SAS version 9.3 (SAS Institute, Cary, North Carolina).

Demographic characteristics observed in AF patients eligible for the incidence analysis were applied to a hypothetical health plan population of 1 million members to determine the plan’s AF patient age and gender distribution. Age- and gender-specific incidence rates for IS and HS were then applied to the estimated AF population to establish the expected total number of ISs and HSs experienced by the health plan’s AF population during a given year. An average incidence rate calculated from all years in the analysis (2006-2011) was used in the modeling application.

	The number of fatal IS and HS cases among AF patients in the health plan were estimated based on IS and HS mortality rates observed in a previous analysis of the MarketScan databases (2005-2011) aimed at assessing the resource use and cost of stroke in AF patients.18 In that analysis, estimates of index stroke hospitalization mortality rate, index stroke hospitalization costs (ie, all costs occurring throughout the time frame of the index hospitalization, stratified by survival outcome), and stroke-related costs 12 months post index for all components of medical care were available (all 2011 US$). Cost estimates were adjusted, controlling for demographics, comorbidities, baseline anticoagulant use, and baseline resource use. 

presents a summary of the cost inputs that were used in the current study.

	In the cost simulation, AF health plan members who did not survive the index stroke hospitalization were assigned the cost associated with a fatal IS or HS event, while AF members who survived a stroke were assigned the cost associated with a nonfatal index event plus 12 months of stroke-related medical care costs. Costs for all age- and sex-specific cohorts were then summed to capture the total stroke-related costs experienced by the health plan for a given year.

	The impact of variations in stroke incidence rates and estimated stroke-related costs were evaluated in 1-way sensitivity analyses. High and low 95% CI estimates for incidence and cost data inputs were entered into the model. Corresponding results demonstrate the potential range of annual stroke-related costs experienced by a typical US health plan.

Based on analyses of the 2 Truven Health MarketScan databases, 20,318 patients with AF experienced a HS or IS from 2006 to 2011. The average (± SD) age of the stroke population was 75.8 (± 11.4) years, with 50.2% of patients being female. Forty-one percent of patients received treatment with an OAC during the baseline period; 65.8% of patients had hypertension, 36.8% had coronary artery disease, 26.7% had diabetes, and 24.3% had congestive heart failure.

Annual IS rates per 100 person-years were considerably higher than HS rates among the cohort of patients with AF. Furthermore, the overall annual rate of IS and HS per 100 person-years tended to decline with time. For instance, rates of IS events per 100 person-years were 1.84 in 2006 and 1.37 in 2011. Similarly, rates of HS events per 100 person-years were 0.37 in 2006 and 0.25 in 2011 (

	For IS, females experienced a higher incidence rate during each year of analysis (range = 1.62-2.31 per 100 person-years) compared with males (range = 1.17-1.50 per 100 person-years). HS rates were similar between males and females with the incidence of either stroke type tending to increase with age, with the highest rates observed among those 85 years or older (

Based on previously published data on the prevalence of diagnosed AF among an adult population (ie, age ≥18),

 we estimated that there would be approximately 9500 AF patients per 1 million adult enrollees in a typical health plan. Demographic characteristics (ie, baseline age and sex data) observed in AF patients eligible for the stroke incidence analysis were applied to the hypothetical health plan population, as were age- and sex-specific annual stroke incidence rates for both IS and HS. This resulted in an estimate of 142 and 29 new cases projected of IS and HS for the health plan during 1 year, respectively.

	Total annual stroke-related healthcare costs for the health plan were estimated at $4,930,787, with 74% of costs attributable to IS (ie, $3,672,304 annual stroke-related costs attributable to IS vs $1,258,482 attributable to HS) (

When the estimated rates representing the lower bound of the 95% CI for annual incidence of both IS and HS were entered into the model, there were 124 and 21 new cases of IS and HS among AF patients in the health plan during 1 year, respectively. This resulted in total annual stroke-related healthcare costs of $4,109,449. When the estimated rates representing the upper bound of the 95% CI for annual incidence were used, there were 159 and 37 new cases of IS and HS, respectively, resulting in total annual stroke-related healthcare costs of $5,727,283.

Applying the estimates representing the lower bound of the 95% CI for all stroke-related costs to base case annual incidence stroke rates resulted in total annual stroke-related costs of $4,775,668 ($3,585,039 attributable to IS and $1,190,629 attributable to HS) for the health plan. Using estimates representing the upper bound of the 95% CI for all stroke-related costs in the model yielded a total annual stroke-related cost of $5,091,968 ($3,761,707 attributable to IS and $1,330,261 attributable to HS).

This study used multi-year data from 2 large administrative claims databases to assess the incidence of stroke, as well as stroke-related medical costs, among patients with AF. Our findings suggest that rates of incidence of IS and HS among patients with AF have declined over the 6-year period from 2006 to 2011, with rates of IS being higher among female patients versus males during each year, and incidence rates of both stroke types increasing with age. Applying observed average incidence rates and costs to a typical health plan of 1 million enrollees resulted in an annual stroke-related healthcare cost of $4,930,787, with 74% of costs being attributable to IS due to higher prevalence rates and lower mortality compared with HS. Resulting total annual costs to the health plan were more sensitive to variations in incidence rates than in stroke-related cost inputs.

An incidence-based model was developed to estimate the annual ischemic and hemorrhagic strokeâ€“related healthcare costs to a hypothetical US health plan with 1 million enrollees.