Background
Stroke is the second leading cause of cardiovascular mortality worldwide and among the top five causes of death in the U.S. [
1,
2]. In the U.S., 6.4 million people, 2.7 % of the adult population, suffered from stroke, according to the 2012 National Health Interview Survey [
3]. Stroke is associated with significant disability, and a negative impact on function and quality of life [
4‐
7]. The treatment for stroke has rapidly evolved over time, and outcomes continue to improve [
8]. While some of the disease risk factors for stroke are well known, there is a relative lack of studies assessing whether certain commonly used medications can modify the risk of stroke.
Allopurinol is a urate-lowering therapy (ULT) that is commonly used for the treatment of hyperuricemia [
9,
10]. In addition to its urate-lowering effect related to of the inhibition of xanthine oxidase by its active metabolite, oxypurinol, recent studies have suggested other mechanisms of actions of allopurinol, some dependent and some independent of xanthine oxidase inhibition [
11‐
18]. Many of these effects including improvement of endothelial function and reduction of oxidative stress [
14,
19‐
23], reduction of glycosylated hemoglobin [
14], and attenuation of intercellular adhesion molecule-1 levels [
24], may potentially lower the risk of stroke. To our knowledge, it is not known whether allopurinol use is associated with a lower risk of stroke.
Previous studies of allopurinol examined a composite cardiovascular outcome (that included stroke) and provided contradictory findings, ranging from a protective effect [
25,
26] to an increased risk [
27,
28]. These contradictory findings were related at least partially to differences in patient populations (renal failure, hypertensive nephropathy, heart failure vs. gout), study design (randomized vs. observational) and covariates/confounders adjusted in the analyses. In a 7-year follow-up of a 2-year randomized trial in patients with chronic kidney disease, adjusted for race, sex and renal function, allopurinol use was associated with lower risk of 0.43 (95 % CI, 0.21 to 0.88) of cardiovascular events (including CAD, cerebrovascular disease, heart failure and peripheral vascular disease) [
25]. In an observational study of 187 patients with hypertensive nephropathy, allopurinol use was associated with an adjusted hazard of 0.34 (
p = 0.04) for cardiovascular disease (CAD, heart failure and stroke) [
26]. In a study of heart failure patients using Scottish database, Wei et al. reported that with a minimum follow-up of 5 years, compared with non-users, allopurinol users had a no significant difference in cardiovascular event (nonfatal myocardial infarction (MI), nonfatal stroke and cardiovascular mortality), 0.88 (95 % CI, 0.73 to 1.05) [
27]. Kok et al. [
28] studied >4,000 patients with newly diagnosed gout and no pre-existing severe form of CVD using a prevalent user design and reported that allopurinol use was associated with an increased adjusted hazard of 1.25 for cardiovascular event (hospitalization for MI, stroke, hypertension etc.).
None of the previous studies examined the association of allopurinol with stroke as an outcome in the general population. Therefore, findings can only be generalized to patient sub-populations with hypertensive nephropathy, renal failure, gout or heart failure, and then only to a cardiovascular composite outcome, that includes peripheral vascular disease, heart failure, MI, angina etc. in addition to stroke. Thus, well-designed studies examining the effect of allopurinol use on the risk of stroke are needed, especially for adults 65 years and older. This is important since 2/3
rd of patients with stroke in the U.S. are adults 65 years and older [
29]. Therefore, our objective was to assess whether incident allopurinol use was independently associated with a reduction in the risk of incident stroke in the elderly, using a population-based approach. We hypothesized that allopurinol use and its duration of use will each be independently associated with a reduction in the risk of incident stroke in adults 65 years and older. In exploratory analyses, we examined whether stroke risk reduction with allopurinol varied by age, gender and race.
Methods
Study cohort and population of interest
We conducted a retrospective cohort study using the 5 % random sample of Medicare beneficiary claims from 2006 to 2012. These data were obtained from the Centers for Medicare and Medicaid Services (CMS) Chronic Condition Data Warehouse and included beneficiaries’ demographic information and insurance claims (inpatient, outpatient, skilled nursing facility, non-institutional provider, home health, hospice, durable medical equipment and prescription drugs).
Study eligibility criteria were: (1) Medicare beneficiaries who were 65 years of age or older and lived in the U.S.; (2) continuous enrollment in traditional Medicare fee-for-service and pharmacy coverage (Parts A, B and D) and non-enrollment in Medicare Advantage Plan; and (3) new treatment with allopurinol (defined below). The Institutional Review Board at the University of Alabama at Birmingham approved the study, and waived the need for patient informed consent for this database study. All investigations were conducted in conformity with ethical principles of research.
Exposure definition and covariates
Incident allopurinol use, i.e. new allopurinol use, was defined as a new filled allopurinol prescription, with no allopurinol prescription filled during a look-back baseline period of 365 days. Each day of observation within each episode was labeled as allopurinol exposed or non-exposed based upon the days’ supply for allopurinol prescription in pharmacy records after the beginning of allopurinol treatment episode. We allowed up to 30 days stock carry over. Patients were considered exposed for 30 days after the end of days’ supply to capture events attributable to allopurinol, after which a new continuous allopurinol exposure period started. Allopurinol use duration was categorized as none, 1 to 180 days, 181 days to 2 years, and longer than 2 years, based on clinical relevance and current use patterns of allopurinol [
9,
10]. A patient could contribute multiple allopurinol treatment episodes during different time periods.
We obtained the following covariates in the baseline period for each allopurinol treatment episode from the Medicare denominator file: age, gender, race/ethnicity, residence (Northeast, South, Midwest, West) and comorbidity scores, derived using Charlson-Romano comorbidity index score, a validated score for comorbidity assessment consisting of common medical comorbidities [
30], adapted for claims data by Romano et al.. Comorbidities included in Charlson-Romano comorbidity index were: myocardial infarction, congestive heart failure, cerebrovascular disease, dementia, chronic pulmonary disease, connective tissue disease, peptic ulcer disease, mild liver disease, diabetes without complications, diabetes with complications, paraplegia and hemiplegia, renal disease, cancer, moderate to severe liver disease, metastatic cancer and AIDS/HIV [
31]. Charlson-Romano index score is a valid comorbidity measure [
32,
33].
Study outcome
Incident stroke was our study outcome of interest, defined as the first occurence of stroke during the study period after the initiation of a new allopurinol prescription. Incident stroke was identified by the presence of International Classification of Diseases, ninth revision, common modification (ICD-9-CM) codes for stroke (430.xx, 431.xx, 433.x1, 436.xx, 434.xx, except 434.x0); subgroup analyses were for hemorrhagic stroke (430.xx, 431.xx) and ischemic stroke (rest of the codes). This approach using ICD-9-CM codes is accurate for identification of cases and risk factors, and has high positive predictive value, usually exceeding >90 % [
34‐
36]. The earliest allopurinol treatment initiation date during the study period marked the beginning of the follow-up for each treatment episode that ended on the earliest of first date of stroke, losing full Medicare coverage, the date of death, or the end of the study (12/31/12).
Statistical analyses
We calculated descriptive statistics by the occurrence of stroke among allopurinol users and by allopurinol exposure among episodes with stroke. The main analysis focused on stroke (including all subtypes). We performed Cox proportional hazard regression models to assess the association between allopurinol exposure or duration of allopurinol use and incident stroke. Multivariable analysis were adjusted for age, gender, race and Charlson-Romano comorbidity score. We accounted for correlated episodes using the Huber-White “Sandwich” variance estimator and calculated robust standard errors for all estimates, since patients could possibly contribute more than one episode of new allopurinol use. We performed sensitivity analyses using Cox proportional hazard regression models to assess these associations by replacing Charlson-Romano index score with selected comorbidities, including risk factors for coronary artery disease (CAD) in one model, and replacing with peripheral vascular disease (PVD) and CAD in another separate model. We calculated hazard ratio (HR) and 95 % confidence intervals (CI). We also performed subgroup analyses by assessing the relationship of allopurinol use and allopurinol use duration to each type of stroke, ischemic vs. hemorrhagic stroke.
Discussion
We used an incident/new user design (new users of allopurinol indicated by a new allopurinol filled prescriptions) and a population-based approach to examine the effect of allopurinol use on the risk of stroke in the U.S. elderly. We controlled for patient demographics, comorbidity and use of cardio-protective medications to understand whether the association of allopurinol was independent of other CAD risk factors and treatments. This new knowledge is important, since the majority of strokes occur in adult Americans 65 years and older [
29]. Subgroup analyses examined whether association of allopurinol was evident for separate analyses for ischemic and hemorrhagic stroke. Several study findings merit further discussion.
Our finding of independent association of incident allopurinol use with a reduction of hazard of incident stroke in the older adults compared to no allopurinol use, is novel. We are unaware of other studies that have examined this specific question. Most previous studies in this area of research had several limitations in that: (1) all studied a composite cardiovascular outcome that included heart failure, angina, MI etc. in addition to stroke [
25‐
28]; (2) most used prevalent user design [
26,
28]; and (3) all studies were in patient subpopulations with limited generalizability [
25‐
28]. Composite outcome assessment did not allow the interpretation regarding the risk of stroke, due to inclusion of related outcomes, which might have different mechanisms than stroke, i.e., heart failure, angina etc. It is therefore not surprising that findings ranged from a reduced risk of with allopurinol [
25,
26] to an increased risk of composite cardiovascular outcome that included stroke [
27,
28].
The stroke risk reduction associated with allopurinol use noted in our study and others may be related to several potential mechanisms. Allopurinol use was associated with attenuation of intercellular adhesion molecule-1 levels in patients after a recent stroke [
24]. Allopurinol use improved endothelial function and reduction of oxidative stress [
14,
19‐
23] and reduced glycosylated hemoglobin [
14]. Allopurinol reduced serum urate, and an elevated serum urate was associated with higher levels of chronic inflammatory markers in cardiac disease [
37].
Regardless of the mechanism of action, the significant reduction of stroke risk associated with allopurinol has practical implications. The hazards of stroke were reduced by 9 % for allopurinol use vs. non-use and reduced by 12–21 % with longer allopurinol use durations. Thus, the magnitudes of hazard reduction were moderate, not large. Allopurinol is commonly used for the treatment of gout and reduced morbidity related to joint inflammation and pain, including disability, functional limitation and joint destruction. The potential added benefit of stroke reduction with allopurinol should be shared with patients with gout at the time of allopurinol treatment initiation. This discussion should impact the overall risk/benefit ratio of allopurinol, as well as serve as a motivating factor for improving allopurinol adherence and persistence, which is low for allopurinol in the first year after initiation [
38]. More evidence is needed before an informed benefit/risk analysis of allopurinol use can be shared with patients with asymptomatic hyperuricemia without gout, where allopurinol use is currently not recommended [
39].
Maximum hazard reduction in incident stroke with allopurinol use occurred in the >6 to 24 months and >2 years after new allopurinol use, with 12 % and 21 % hazard reduction, respectively. We noted an incremental stroke risk reduction associated with longer duration of allopurinol use. This important finding indicates that allopurinol use is needed for >6 months to experience this potential benefit.
In exploratory multivariable-adjusted subgroup analyses, allopurinol use durations of 181 days to 2 years and >2 years were associated with a reduction of hazard of stroke, most evident for the age group 75–84, female gender and people who were white. Our study identified the sub- groups of patients likely to benefit the most with longer durations of allopurinol use. These findings need to be confirmed in future studies.
Our exploratory analyses added further clarity to this protective effect of allopurinol. Risk reduction associated with allopurinol use was noted for ischemic stroke, but not hemorrhagic stroke. These two types of stroke differ quite a bit in pathophysiology and outcomes [
40]. It was reassuring that the protection for overall stroke risk is primarily due to protection against ischemic stroke, which is consistent with allopurinol’s effect on oxidative stress, endothelial function and chronic inflammation [
14,
19‐
23,
37].
Our study has several strengths and limitations. The diagnoses of CAD risk factors and Charlson-Romano index were based on the presence of ICD-9-CM codes, which may have led to misclassification bias. However, the codes for outcomes and covariates have reasonable accuracy [
34‐
36], and the use of ICD-9-CM codes for Charlson-Romano index is a standard valid approach that is used commonly in analyses of large claims databases, such as Medicare. The observational study design does not allow us to eliminate confounder bias, due to unmeasured confounding; however, we included important demographic, clinical, risk factors and cardiac medications to account for important potential confounders. Serum urate levels were available for a small proportion of patients and therefore we could not examine whether the associations we noted were mediated through the reduction of serum urate or not. The database also did not allow us to include other variables such body mass index, diet and exercise, which might be potential risk factors for stroke.
The contribution of healthy adherer effect can not be ruled out, since patients adherent to/using allopurinol may have other healthy behaviors, such as high adherence to preventive care, other medications, healthier lifestyle etc., which may all reduce the risk of stroke etc. A filled prescription is a surrogate for, but can not be equated to the actual use of allopurinol. Since study’s eligibility criteria included that patients to have continued coverage, these findings are likely not applicable to elderly who loose Medicare coverage. Whether these findings can be generalized to populations other than adults 65 years and older is not known, and further study is needed in this area.
Study strengths include a large sample size, the use of a sample that allows generalizable findings to Medicare enrollees, inclusion of MI/stroke risk factors in analyses to avoid confounding bias, and the robustness of estimates in sensitivity analyses. Implementation of new user design (i.e., examining only new users of allopurinol) in our study avoids over-adjustment for factors on the causal pathway, unlike the prevalent user design, used in previous studies.