Article Text

Download PDFPDF

Extended report
Allopurinol and the risk of atrial fibrillation in the elderly: a study using Medicare data
  1. Jasvinder A Singh1,2,3,
  2. Shaohua Yu2
  1. 1Medicine Service, Birmingham VA Medical Center, Birmingham, Alabama, USA
  2. 2Department of Medicine at School of Medicine, and Division of Epidemiology at School of Public Health, University of Alabama at Birmingham (UAB), Birmingham, Alabama, USA
  3. 3Department of Orthopedic Surgery, Mayo Clinic College of Medicine, Rochester, Minnesota, USA
  1. Correspondence to Dr Jasvinder A Singh, University of Alabama, Faculty Office Tower 805B, 510 20th Street S, Birmingham, AL 35294, USA;{at}


Objective To assess the effect of allopurinol use on the risk of incident atrial fibrillation (AF) in the elderly.

Methods We used the 5% random Medicare Claims data from 2006 to 2012 to examine the association of allopurinol use and incident AF in a cohort of patients with an absence of AF at baseline (at least 365 days). Multivariable-adjusted Cox regression analyses compared allopurinol exposed and non-exposed periods for the risk of AF, controlling for age, sex, race, Charlson–Romano comorbidity index and use of statins, diuretics, ACE inhibitors and β-blockers. HR with 95% CIs was calculated. Sensitivity analyses considered a longer baseline period (365 days vs 183 days) and individual comorbidities.

Results There were 9244 episodes of incident allopurinol use in 8569 beneficiaries, of which 1366 episodes (14.8%) had incident AF. In multivariable-adjusted analyses, allopurinol use was associated with an HR of 0.83 (95% CI 0.74 to 0.93) for incident AF. In a separate multivariable-adjusted model, compared with no allopurinol use, longer allopurinol use durations were associated with a lower HR of AF: 180 days–2 years, 0.85 (95% CI 0.73 to 0.99) and >2 years, 0.65 (95% CI 0.52 to 0.82). Other factors significantly associated with a higher hazard of AF were: age 75–<85 years and ≥85 years, higher Charlson index score and current β-blocker use. Sensitivity analyses confirmed these findings with minimal/no attenuation of HRs.

Conclusions Allopurinol use was associated with a reduced risk of incident AF in the elderly, especially its use for >6 months duration. Future studies should assess the mechanisms underlying this beneficial effect of allopurinol.

  • Cardiovascular Disease
  • Epidemiology
  • Health services research
View Full Text

Statistics from


Atrial fibrillation (AF) is the most prevalent arrhythmia worldwide1 and the most common arrhythmia leading to hospitalisation.2 Not surprisingly, AF costs $16–26 billion in annual US expenses.3 ,4 AF is associated with increased morbidity, a higher incidence of stroke and heart failure, and increased mortality5–9; for example, in one study of Medicare beneficiaries, a higher proportion of patients with AF experienced stroke (23.1% vs 13.3%) and heart failure (36.7% vs 10.4%) compared with those without AF.3 In another study, mortality was 1.5-fold higher in men with AF and 1.9-fold higher in women with AF.6 In the USA, it is estimated that the number of adults with AF will more than double by the year 2050.10 Thus, AF is an important public health problem.

Allopurinol is the most commonly used urate-lowering therapy for the treatment of hyperuricaemia in patients with gout.11 ,12 Allopurinol is a purine analogue. As a structural isomer of hypoxanthine (a naturally occurring purine in humans), allopurinol competes with the substrate for the enzyme xanthine oxidase, which then leads to reduced uric acid production. In addition to blocking uric acid production, inhibition of xanthine oxidase causes an increase in hypoxanthine and xanthine. Allopurinol reduces oxidative stress,13–15 improves endothelial function,16–20 reduces left ventricular mass21 ,22 and has an anti-ischaemic action.23 Emerging research supports a potential cardioprotective effect of allopurinol including improved outcomes for myocardial infarction (MI), heart failure24–26 and mortality,27–29 although two recent studies failed to find this association.30 ,31 On the other hand, with the exception of one abstract that suggested the association of allopurinol and AF in a single-centre study,32 there are no published studies assessing this association. This is an important area for study, given the prevalence of AF and the availability of allopurinol as a low-cost, generic drug that is well tolerated.

Our study objective was to explore the association of allopurinol with the incidence of AF in the elderly. Specifically, using the Medicare claims data, we assessed whether (1) allopurinol use (vs non-use) and (2) allopurinol use duration, were associated with the risk of incident AF, in the US elderly, 65 years and older.


Study cohort and population of interest

This retrospective cohort study used the 5% random sample of Medicare beneficiaries from 2006 to 2012, obtained from the Centers for Medicare and Medicaid Services (CMS) Chronic Condition Data Warehouse. The 5% random sample file is a standard Medicare analytical file available from CMS data warehouse that is commonly used to address epidemiological and outcomes questions.33–35 The Medicare claims contain information for all insurance claims for each beneficiary including inpatient, outpatient, skilled nursing facility, non-institutional provider, home health, hospice, durable medical equipment services and claims for prescription drugs, and the beneficiary's demographic information. Eligible subjects for the study were Medicare beneficiaries who were 65 years of age or older; lived in the USA; were enrolled continuously in traditional Medicare fee-for-service and pharmacy coverage (parts A, B and D) and not enrolled in Medicare Advantage Plan, who had new treatment with allopurinol (defined in the section below). The study was approved by the Institutional Review Board (IRB) at the University of Alabama at Birmingham. Since this was a database analysis, the IRB waived the need for informed consent.

Exposure definition and covariates

A new allopurinol exposure was defined as the initiation of allopurinol therapy, indicated by a filled allopurinol prescription, after a baseline period of at least 365 days during which no allopurinol prescription was filled. We assigned days of exposure for each allopurinol episode based upon the days' supply for prescription in pharmacy records. Patients were considered exposed for 30 days after the end of days' supply to capture potentially attributable events. If patients filled another allopurinol prescription within 30 days of the end of the current allopurinol prescription, we considered it as a continuous prescription, since patients commonly have some extra days of medication supply due to imperfect medication adherence. If there were more than 30 days between prescription fills, a new continuous allopurinol exposure period started. Allopurinol use duration was categorised as none, 1–180 days, 181 days–2 years, and longer than 2 years. This arbritrary categorisation was performed a priori for clinical relevance with respective categories indicating short-term, intermediate-term and long-term allopurinol use.

Covariates were obtained from the Medicare denominator file and included age at the index date of each allopurinol episode, gender, race/ethnicity, cardiac medications (statins, diuretics, ACE inhibitors and β-blockers) and comorbidity scores in the baseline period for each allopurinol treatment episode. We used the Charlson–Romano comorbidity index score, a validated score developed for claims data.36


The outcome of interest was incident AF. Incident AF was defined as the first occurrence of AF during the study period after initiation of new allopurinol use and its absence during a baseline period of at least 365 days before the allopurinol initiation date, identified by the presence of the International Classification of Diseases, ninth revision, common modification (ICD-9-CM) code, 427.31. This approach of identifying AF using ICD codes has been shown to have a median accuracy of 89% (70–96%) in a systematic review of accuracy of diagnostic codes for AF;37 a recent study showed it to be as high as 96%.38 The follow-up for each treatment episode began on the earliest allopurinol treatment initiation date during the study period and ended on the earliest of the first date of AF, the first date of losing full Medicare coverage, the date of death or the end of the study (31 December 2012). Patients could contribute multiple allopurinol treatment episodes during different time periods.

Statistical analyses

Summary statistics were calculated for the cohort, by occurrence versus non-occurrence of AF and by allopurinol exposure versus not. We performed Cox proportional hazards regression analyses to assess the association between incident allopurinol use and incident AF and between the duration of allopurinol use and incident AF. We performed univariate and multivariable analyses adjusted for age, gender, race, Charlson–Romano comorbidity score and cardiac medications. Since the analyses were episode-level rather than person-level, we accounted for correlated episodes (patients could possibly contribute more than one episode of new use) using the Huber–White ‘Sandwich’ variance estimator39 and calculated robust SEs for all estimates. We also performed a sensitivity analysis using Cox proportional hazards regression models to assess these associations by extending the baseline period to 365 days, instead of at least 365 days. We calculated the HR and 95% CIs.


Characteristics of the patient population

There were 9244 episodes with incident allopurinol use in 8569 beneficiaries during the study period (table 1). Of these, 1366 episodes (14.8%) had associated AF and 7878 did not (figure 1). Compared with episodes without AF, those with AF were younger, more likely to be Caucasian and had a higher Charlson index (table 1). Table 2 shows the incidence rate of AF related to allopurinol exposure and duration of allopurinol use. The mean look-back period was 3.5 years (SD, 1.9 years).

Table 1

Demographic and clinical characteristics of episodes of allopurinol users (baseline with no atrial fibrillation; baseline was longest possible and at least 365 days)

Table 2

Crude incidence rate of atrial fibrillation with allopurinol exposure

Figure 1

Flow chart of study cohort of incident allopurinol users from 2006 to 2012 excluding allopurinol use at baseline (baseline was longest possible and at least 365 days).

Allopurinol use and the risk of AF

Among patients with a new AF, episodes with allopurinol exposure were more likely in those who were Caucasian or had a higher Charlson Comorbidity Index score; a non-significant trend was seen in those with a higher prevalence of cerebrovascular disease (table 3). Univariate HRs are shown in table 4.

Table 3

Demographic characteristics and prevalence of comorbidities of allopurinol episodes with atrial fibrillation

Table 4

Association of risk factors with hazard of atrial fibrillation in patients who received allopurinol with no baseline atrial fibrillation before the index date of allopurinol episode

In multivariable-adjusted analyses, allopurinol use was associated with 17% reduction in the hazard of incident AF, with an HR of 0.83 (95% CI 0.74 to 0.93). Other factors associated with higher hazard of AF were: age 75–<85 years and ≥85 years, higher Charlson–Romano index score and the use of β-blockers (table 4).

In a separate multivariable-adjusted model, compared with no allopurinol use, we found that longer allopurinol use duration was associated significantly with lower HR of AF, but short use <180 days was not: <180 days–2 years, 0.85 (95% CI 0.73 to 0.99), >2 years, 0.65 (95% CI 0.52 to 0.82) and <180 days, 0.92 (95% CI 0.78, 1.09; table 4). The number needed to treat (NNT) to benefit ranged from two for allopurinol use (vs non-use) to two to four for allopurinol use 180 days–2 years versus >2 years (vs non-use), with the assumption that HR could be used as an estimate of the relative risk.

Sensitivity analyses that were performed adjusting for hypertension, diabetes, peripheral vascular disease and coronary artery disease, instead of the Charlson–Romano score, confirmed these findings. Allopurinol exposure was associated with reduction of hazards to 0.83 (95% CI 0.74 to 0.93) (table 5). The respective ORs for allopurinol use durations of <180 days, 180 days–2 years and >2 years (vs non-use), were as follows: 0.92 (95% CI 0.78 to 1.08), 0.85 (95% CI 0.73 to 0.99) and 0.65 (95% CI 0.52 to 0.82), respectively (table 5). Sensitivity analyses that additionally adjusted this model for hypercholesterolaemia (see online supplementary appendix 1), or only adjusted for diabetes, hypertension and hypercholesterolaemia as medical comorbidities (see online supplementary appendix 2), confirmed these findings.

Table 5

Sensitivity analysis additionally adjusted for diabetes, hypertension, PVD and CVD: association of risk factors with hazard of atrial fibrillation in patients who received allopurinol with an absence of atrial fibrillation in the baseline period before the index date of allopurinol episode

Sensitivity analyses that restricted the cohort to an absence of a diagnosis of AF for 365 days (instead of at least 365 days), confirmed the findings above with an HR of 0.90 (95% CI 0.84 to 0.97) for AF with allopurinol use (vs non-use) (see online supplementary appendix 3). The HR was significant at 0.71 (95% CI 0.60 to 0.83) for allopurinol use duration of >2 years, but not for allopurinol use duration 6 months–2 years, 0.95 (95% CI 0.86 to 1.05).

Subgroup analyses for allopurinol use duration by age and gender

In multivariable-adjusted analyses, allopurinol use durations of 181 days–2 years and >2 years were associated with a reduction of HR of AF, most evident for the ≥85 years age groups, female gender and Caucasians (figure 2).

Figure 2

Figures represent multivariable-adjusted hazard ratios of atrial fibrillation with allopurinol exposure of 1–180 days, 181 days to 2 years and >2 years for each patient characteristic. Error bars represent the 95% confidence interval The horizontal hashed line represents a hazard ratio of 1, which represents the reference category of allopurinol use of 0 days (non-use). If the 95% error bar includes 1, the hazard ratio is statistically insignificant.


This study assessed whether incident allopurinol use was associated with a risk of new AF. We found that allopurinol use was independently associated with a lower risk of incident AF in a cohort of elderly patients, using the 5% Medicare claims sample. Compared with patients who did not use allopurinol, allopurinol current users had a 17% lower hazard of MI. Several findings merit further discussion.

First, the reduction in the hazard of AF with incident allopurinol use in the elderly is a novel finding. Specifically, the hazard of AF was reduced by 17% in allopurinol users compared with non-users. This is an important finding, in the absence of any previously published study investigating this relationship. To our knowledge, only one retrospective single-centre study of veterans with heart failure, published as an abstract in 2013 (full report not available), showed that allopurinol use was associated with an HR of 0.5 (95% CI 0.2 to 0.9) of AF.32 The study was limited to patients with heart failure and to US veterans, a significant issue with generalisability of study findings; specifics of study methods are not available. Our observation of potential prevention of AF with allopurinol adds to the emerging literature of its cardiovascular beneficial effect for overall cardiovascular outcomes,24 myocardial infarction,25 heart failure readmission or death,26 and mortality.27–29

AF is associated clinically with hypertension, heart failure, valvular heart disease and ischaemic heart disease, and the presence of these comorbidities increases the risk of AF.40 Heart failure and AF may promote each other; AF can compromise left ventricular function and conversely the left ventricular dysfunction can lead to atrial dilation and pressure overload.40 Chronic atrial stretch, one of the most prominent precursors of AF, can result from left ventricular dysfunction and failure, which is commonly seen with conditions such as heart failure, ischaemic heart disease, etc.41 Allopurinol was associated with suppression of atrial fibrosis and endothelial nitric oxide synthetase (eNOS) in a canine tachycardia-induced cardiomyopathy model, a potential mechanism for lowering the risk of AF.42 Other potential mechanisms include associated antioxidant action,15 anti-ischaemic action,23 reduction of left ventricular mass,21 ,22 improvement of endothelial function16–19 and reduction in blood pressure.43 ,44 Future studies need to characterise the mechanisms for this potentially beneficial effect of allopurinol related to the risk of AF.

Longer duration of allopurinol use was associated with a reduction of the risk of AF, a novel study finding. Allopurinol use for 0.5–2 years and >2 years, were associated with a greater reduction in hazard of AF by 15% and 35%, respectively, suggesting a dose-response gradient. This is similar to a lower risk of cardiovascular events (including MI) due to allopurinol in a longer follow-up of a randomised study documenting an impact of longer duration.45 In addition, higher allopurinol dose was associated with lower mortality in patients with heart failure29 and lower risk of MI in the general population.25 Thus, potential cardioprotective effects of allopurinol may be related to allopurinol duration and dose.

Subgroup analyses indicated that patients who were 65–74 years old and ≥85 years old, women and of Caucasian race, seemed to experience the greatest reduction in the hazard of AF with allopurinol use. These are hypothesis-generating findings that suggest that subsequent studies confirming our findings explore these associations further. This will help to identify patient groups that are likely to benefit the most with allopurinol use. A higher risk of AF with β-blockers is consistent with its recently noted association with a higher risk of stroke,46–49 although mechanisms are still unclear.

Our study has several strengths and limitations. Despite our efforts to control for various confounding factors, our cohort study findings are subject to potential confounding bias. We adjusted the analyses for cardiovascular disease, common cardiovascular conditions and cardiac medications. Second, our database lacked information on other risk factors such as smoking, body mass index and diet components, which can be viewed as risk factors for MI. Third, laboratory test results such as serum urate levels are not available in the Medicare data. Therefore, we could not assess whether the reduction of AF with allopurinol was mediated by reduction in serum urate levels. This is likely, since patients receiving allopurinol have lower levels of serum uric acid than patients not taking allopurinol. Fourth, we were unable to validate the accuracy of the diagnostic code for AF, since charts are not available as part of Medicare claims data. Therefore, there is a risk of some diagnostic misclassification, despite a high median accuracy of 89% previously shown for the ICD-9 code for AF.37 ,38 Such misclassification might have biased the results towards the null, making our estimates conservative; we expect such bias to be small, given the high accuracy of this approach. Fifth, the number of AF events may have been too low to detect statistical significance of findings in certain groups, making our study underpowered (i.e., type II error); availability of the entire data set would likely have overcome this limitation, but limited resources prevented us from doing this. However, the use of a national 5% Medicare sample is a standard approach to many similar studies previously.33 ,34 Sixth, limited resources did not allow us to examine the effect of allopurinol dose on the outcome. Future studies should examine whether a dose-response relationship exists for this effect. Seventh, the NNT assumed that the HR would be similar to relative risk; if this assumption is not correct, the actual NNT may be higher or lower than we estimated. Lastly, these data and analyses are derived from Medicare, and can only be generalised to the US elderly population, not the US general population.

Strengths of the study include the use of a representative US population, the ability to control for the common cardiovascular risk factors, the robustness of findings in multiple sensitivity models, the use of incident user design to avoid the biases of prevalent user design and the examination of both exposure as well as duration of allopurinol use.

In conclusion, this study shows that allopurinol use was associated with a lower risk of AF in the elderly. This potential protective effect is particularly evident after 6 months of allopurinol use and is more pronounced after 2 years of allopurinol use. We identified certain groups of the elderly in whom this effect was most pronounced. Future studies need to examine the mechanism of this potential beneficial cardiovascular effect of allopurinol use.


View Abstract


  • Handling editor Tore K Kvien

  • Contributors JAS: Study design, study protocol, analytical plan, writing the first draft of the paper, critical revision of the paper and decision to submit the paper. SY: data programming, data management, review of the analytical plan, data analyses, critical revision of the paper and decision to submit the paper.

  • Funding This material is the result of work supported by research funds from the UAB Division of Rheumatology and the resources and use of facilities at the Birmingham VA Medical Center. JAS is also supported by a grant from the National Institute of Arthritis, Musculoskeletal, and Skin Diseases (NIAMS) P50 AR060772.

  • Competing interests JAS has received research grants from Takeda and Savient and consultant fees from Savient, Takeda, Regeneron, Iroko, Merz, Bioiberica, Crealta and Allergan pharmaceuticals. JAS serves as the principal investigator for an investigator-initiated study funded by Horizon pharmaceuticals through a grant to DINORA, a 501c3 entity. JAS is a member of the executive of OMERACT, an organisation that receives arms-length funding from 36 companies; a member of the American College of Rheumatology's Guidelines Subcommittee of the Quality of Care Committee; and a member of the Veterans Affairs Rheumatology Field Advisory Committee.

  • Ethics approval The Institutional Review Board at the University of Alabama at Birmingham' approved this study and all investigations were conducted in conformity with ethical principles of research.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Data sharing statement These data are available in the public domain for any author groups to conduct the analyses. We would be happy to share any information with any author groups after approval from relevant committees at our institution (human ethics and patient privacy protection, etc).

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.