Impact of gout on the risk of atrial fibrillation

Abstract

Objectives. To examine the risk of atrial fibrillation (AF) at the time of first diagnosis of gout compared with matched controls and to follow incident gout patients and their matched controls after diagnosis to compare their subsequent risk of AF.

Methods. From the UK Clinical Practice Research Data-link, 45 378 incident gout patients and 45 378 age-, sex-, practice-, registration year- and index year-matched controls were identified. Index dates were initial diagnosis date for gout patients and their matched controls. The risk of AF at diagnosis [odds ratios (ORs), using conditional logistic regression] and after the diagnosis of gout [hazard ratios (HRs), using Cox proportional models] were estimated, adjusted for BMI, smoking, alcohol consumption, ischaemic heart disease, heart failure, heart valve disease, hyperthyroidism and other comorbidities and medications.

Results. The prevalence of AF at index date in gout patients (male, 72.3%; mean age, 62.4 ± 15.1 years) was 7.42% (95% CI 7.18, 7.66%) and in matched controls 2.83% (95% CI 2.67, 2.98%). The adjusted OR (95% CI) was 1.45 (1.29, 1.62). The cumulative probability of AF at 1, 2, 5 and 10 years after index date was 1.08, 2.03, 4.77 and 9.68% in gout patients and 0.43, 1.08, 2.95 and 6.33% in controls, respectively (log-rank test, P < 0.001). The adjusted HR (95% CIs) was 1.09 (1.03, 1.16).

Conclusion. This population-based study indicates that gout is independently associated with a higher risk of AF at diagnosis and the risk is also higher after the diagnosis.

gout, atrial fibrillation, relative risks, population study, CPRD

Rheumatology key messages

• Gout patients often have multiple comorbidities.

• Gout is independently associated with a higher risk of atrial fibrillation.

  Electrocardiogram is warranted for gout patients at diagnosis.

Introduction

Gout is the most common inflammatory arthritis worldwide and affects 1 in 40 individuals in the UK [1]. In addition to episodic acute arthritis, gout also results in chronic joint damage, subcutaneous tophi and peri-articular inflammation [2]. It is generally accepted that gout is highly associated with the features of metabolic syndrome [3] and chronic renal impairment [4]. In addition, gout patients are also at higher risk for developing many different conditions, such as cardiovascular [5–12], metabolic [3, 11, 13], renal [4, 11] and many other comorbidities [14–16], which collectively lead to an increased mortality [17, 18]. A recent study of co-morbidities that associate with gout also provided evidence for an association with cardiac arrhythmias [11], an observation not reported previously.

Atrial fibrillation (AF) is the most prevalent cardiac arrhythmia present in clinical practice [19]. In the UK, the incidence of chronic AF was reported in 2002 as 1.7/1000 person-years, and this increased markedly with age [20]. Affected patients are at increased risk of heart failure, thromboembolic events such as stroke, and increased mortality [21, 22]. The Framingham study reported a 4- to 5-fold risk for stroke in patients with AF [23]. In addition, AF is also associated with a 2-fold risk of silent stroke [24]. The major risk factors recognized for AF include heart failure, ischaemic heart disease, hypertension, thyroid disease, heavy drinking and obesity [25]. Interestingly, stroke and many predictors for AF are also common comorbidities in gout patients [11], suggesting that there may be a potential link between gout and AF. For example, a recent study has indicated that hyperuricaemia is associated with a larger left atrial dimension [26, 27] and a greater risk of AF [26–28]. However, despite some evidence that hyperuricaemia, the prerequisite for gout, may be a risk factor for AF, the association between gout and AF has not been formally studied.

Therefore, using data representative of the general population of the UK from the Clinical Practice Research Data-link (CPRD), this study aimed to examine the risk of AF at the time of first diagnosis of gout compared with matched controls. We further followed incident gout patients and their matched controls after diagnosis to compare their subsequent risk of AF.

Methods

We hypothesized that gout patients have a higher risk of AF at the time of first diagnosis and that the risk would continue to be higher after the diagnosis of the disease. We addressed this hypothesis using both retrospective (prior to diagnosis) and prospective case control study (after diagnosis) within the CPRD. The CPRD complies with the Declaration of Helsinki. Our analysis was approved by the Independent Scientific Advisory Committee for MHRA Database Research (protocol 11-021R). All data in this study were totally anonymized and therefore patient consents are exempted.

Data source

In the UK, general practitioners are the gatekeepers for all resources of the health care system. CPRD is a database containing primary care data prospectively collected by general practitioners in 684 practices who were trained to record medical information to ensure the accuracy of the data but were unaware of the research that used the information from the database. Currently the database contains data on >15 million people across the UK and is broadly representative of the UK general population in age and sex structure [29]. The database has been described previously [29] and has been validated for many diagnoses, including gout [30] and many other medical conditions [30–34]. A recent systematic review reported that the median percentage of cases confirmed by validation was 89% (33–100%) [33]. Importantly, 87% of diagnoses from secondary care specialists were captured and recorded [33]. The database contains comprehensive information on patient demographics, lifestyle factors, medical diagnoses, results of investigations and examinations and prescribed medications. The CPRD is also linked to additional data sources including secondary care, the Office for National Statistics cause of death data and information from specific disease registries.

Study design

In this study, our source population was based on the August 2014 version of CPRD. The study population consisted of all incident gout patients diagnosed between 1997 and 2005 who had at least 3 years of continuous registration. Each incident gout patient was matched at random to one control patient who was registered in the same practice and did not have gout and any prescription of urate-lowering treatment. Control patients were frequency-matched in a 1:1 ratio to incident gout patients by year of birth (±2 years), gender, general practice and year of first continuous registration (±2 years) at this CPRD practice. The same index date was assigned to each of the matched controls. As with incident gout patients the matched controls had at least 3 years of registration prior to the index date. For prospective analysis, follow-up time started from index date and ended when they had AF, died, transferred out of the practice or at the last data collection date of the practice at which they were registered, whichever came first.

Study groups

We classified patients according to Read codes, which are diagnostic codes used by general practitioners to define diseases, procedure and other patient characteristics in UK primary care. Patients with gout were identified using a code list for gout, which our group has been using consistently [1, 11, 35] and which was validated previously with an overall ascertainment rate of 90% [30]. The case definition was based on physician diagnosis using 18 Read codes indicative of incident gout [1]. The gout diagnosis in the CPRD has been validated previously by a review of medical records of 10 confirmed and 28 probable gout patients showing that 10 out of 10 confirmed cases and 24 out of 28 probable cases were true gout patients (overall ascertainment rate 90%) [30]. In the UK, urate lowering therapy (ULT) is indicated only for gout or uric acid nephrolithiasis. However, some patients did receive ULT prior to their first recording of gout diagnosis or without any diagnosis. These patients were excluded from this study.

AF outcome assessment

Our primary outcome of interest was AF-based physician diagnosis. This case definition has been validated previously [20, 36]. These validation studies contacted practitioners for a questionnaire asking them to classify 1714 AF patients according to a set of criteria. Among 1606 valid questionnaires, only 66 patients were confirmed not to have AF [20, 36].

Assessment of covariates

We collected life style characteristics (alcohol use, smoking and BMI), comorbidities (hypertension, myocardial infarction, congestive heart failure, peripheral vascular disease, cerebrovascular disease, heart valve disease, hyperthyroidism, dementia, chronic pulmonary disease, rheumatologic disease, peptic ulcer disease, mild liver disease, liver disease, diabetes mellitus, renal diseases, cancer and anaemia), medication use (antihypertensive, glucocorticoid, statin, lipid-lowering agents, hypoglycaemic agents, insulin, aspirin, anticoagulants and hormone replacement treatment) and number of consultations in the 1 year prior to the index date. Life style characteristics were ascertained using the nearest possible measurement prior to the index date. Comorbidities were ascertained within 10 years and medications and the number of visits to practitioners was ascertained within 1 year prior to the index date.

Statistical analysis

Patient characteristics at index date were compared between gout patients and matched controls. The prevalence of AF was calculated using the number of people ever diagnosed with AF during the past 3-year period before the index date as the numerator and the number of incident gout patients or matched controls as the denominator. Odds ratios (ORs) and 95% confidence intervals (CI) were used to estimate the association between gout and AF at diagnosis. Conditional logistic regression was used to adjust for the aforementioned covariates. Missing data for BMI, smoking and alcohol status were coded as unknown. Both gout patients and matched controls were followed up from the index date to the earliest date of occurrence of AF, death, transfer out or end of study (31 December 2013), whichever came first. Kaplan–Meier product-limit analysis was used to estimate the cumulative probability of AF in people with incident gout and matched controls. Hazard ratios (HRs) with 95% CIs were calculated for AF using a Cox proportional hazards model. Only people without AF at index date were considered in the Cox model. We checked the proportional hazards assumptions by examining the log–log survival curves in our models. The HRs were adjusted by the aforementioned covariates. Data were missing for alcohol use, smoking and BMI in 24.76, 17.62 and 17.62% of patients, respectively. These lifestyle characteristics are generally obtained when patients visit GPs. Therefore we assumed that data were missing at random (the missing pattern does not depend on the unobserved data). We used multiple imputation to handle these missing data using SAS MI procedure. The imputation model used logistic regression for alcohol and smoking and regression for BMI. We included all lifestyle, comorbidity and consultation frequency variables in the imputation model. The procedure generated five imputed datasets. The datasets were pooled for logistic regression and Cox proportional hazards models using Rubin’s strategy [37]. We performed a sensitivity analysis using a different method handling missing values (multiple imputation and setting a new category for missing value). All statistical analyses were performed using SAS statistical software, version 9.3 (SAS Institute, Cary, NC, USA).

Results

Study population

We identified 45 378 incident gout patients during the period 1997–2005 [mean age 62.4 (15.1) years; 72.3% being men]. Female patients were older than male patients and had more comorbidities and medications use. They were 1:1 matched to 45 378 controls with the same age and sex structure (Table 1). The median observation periods (interquartile range) before and after index date were 15 (9–27) and 9 (5–12) years with no statistical difference between gout patients and controls. Gout patients consumed more cigarettes and alcohol and had a higher BMI, Charlson score and more general practice than controls. The prevalence of hypertension, ischaemic heart disease, cerebrovascular disease, heart failure and heart valve disease was significantly higher in gout patients than controls in both genders. More gout patients were prescribed medications (Table 1).

Table 1

Characteristics of incident gout patients and controls

Characteristic	Men			Women
	Gout (n = 33 012)	Controls (n = 33 012)	P	Gout (n = 12 366)	Controls (n = 12 366)	P
Age, mean (s.d.), years	59.9 (14.7)	59.8 (14.8)	0.49	69.4 (13.8)	69.3 (13.9)	0.66
BMI, n (%), kg/m2			<0.001			<0.001
<18.5	176 (0.55)	676 (2.18)		214 (1.78)	543 (4.64)
18.5–24.9	6858 (21.33)	11 831 (38.10)		3117 (25.94)	5021 (42.94)
25.0–29.9	14 585 (45.36)	13 091 (42.16)		4111 (34.22)	3863 (33.03)
≥30	10 533 (32.76)	5453 (17.56)		4572 (38.06)	2267 (19.39)
Smoking, n (%)			<0.001			<0.001
Non-smoker	8867 (26.86)	8040 (24.35)		4336 (35.06)	4389 (35.49)
Current smoker	5737 (17.38)	4915 (14.89)		1925 (15.57)	1546 (12.50)
Ex-smoker	18408 (55.76)	20057 (60.76)		6105 (49.37)	6431 (52.01)
Alcohol consumption, n (%), units/week			<0.001			<0.001
Never/Ex-drinker	2361 (7.15)	3359 (10.18)		3185 (25.76)	3333 (26.95)
Current 1–9	16526 (50.06)	19526 (59.15)		7863 (63.59)	8068 (65.24)
Current ≥10	14125 (42.79)	10127 (30.68)		1318 (10.66)	965 (7.80)
Charlson score, mean (s.d.)	2.65 (1.22)	1.99 (1.39)	<0.001	2.31 (1.26)	2.98 (1.39)	<0.001
No. of visits to GP, median (IQR)	14 (5–25)	6 (1–17)	<0.001	24 (15–35)	14 (5–25)	<0.001
Comorbiditiesa, n (%)
Hypertension	10 401 (31.51)	5266 (15.95)	<0.001	5591 (45.21)	3312 (26.78)	<0.001
Ischaemic heart disease	4582 (13.88)	2606 (7.89)	<0.001	1926 (15.57)	919 (7.43)	<0.001
Cerebrovascular disease	1759 (5.33)	1263 (3.83)	<0.001	997 (8.06)	639 (5.17)	<0.001
Heart failure	2456 (7.44)	660 (2.00)	<0.001	1520 (12.29)	481 (3.89)	<0.001
Vulvular heart disease	644 (1.95)	283 (0.86)	<0.001	436 (3.53)	158 (1.28)	<0.001
Hyperthyroidism	116 (0.35)	71 (0.22)	0.001	151 (1.22)	130 (1.03)	0.21
Medicationsb, n (%)
Aspirin	6380 (19.33)	4029 (12.20)	<0.001	3125 (25.27)	1820 (14.72)	<0.001
Anticoagulant	1896 (5.74)	6.47 (1.96)	<0.001	930 (7.52)	259 (2.09)	<0.001
Statin	5049 (15.29)	2776 (8.41)	<0.001	2189 (17.70)	1066 (8.62)	<0.001
Diuretics	9761 (29.57)	3706 (11.23)	<0.001	7443 (60.19)	3307 (26.74)	<0.001
Beta antagonist	3877 (11.74)	1851 (5.61)	<0.001	2062 (16.67)	1019 (8.24)	<0.001
Calcium channel antagonist	4518 (13.69)	2634 (7.98)	<0.001	2374 (19.20)	1443 (11.67)	<0.001
ACEi/ARB	7088 (21.47)	3055 (9.25)	<0.001	3774 (30.52)	10605 (12.98)	<0.001
Glucocorticoid	1404 (4.25)	937 (2.84)	<0.001	993 (8.03)	559 (4.52)	<0.001

^aSelected comorbidities.

^bSelected medications.

ACEi, angiotensin converting enzyme antagonist; ARB, Angiotensin receptor antagonist; IQR, interquartile range.

Retrospective observation

At time of index date, the proportion of people having AF was significantly higher in incident gout patients (7.42%; 95% CI 7.18, 7.66%) than in controls (2.83%; 95% CI 2.67, 2.98%; P < 0.001). The prevalence of AF at index date was higher in both men and women with gout than in their respective controls (Table 2). Using a conditional logistic regression models, the unadjusted OR was estimated to be 2.89 (95% CI 2.70, 3.09) for AF. After consideration of all covariates including traditional risk factors for AF such as cardiovascular diseases and hyperthyroidism, gout was still associated with an increased odds for AF, with an adjusted OR of 1.45 (95% CI 1.29, 1.62). Separate estimates for men and women were shown in Table 2. Both men and women with gout had a higher AF prevalence than that in the controls.

Table 2

Prevalence and odds ratios for atrial fibrillation by gout status, stratified by sex

Gout	AF prevalence (%; 95% CI)	Matched OR (95% CI)a	Adjusted OR (95% CI)b
Overall
Yes	7.42 (7.18, 7.66)	2.82 (2.63, 3.04)	1.44 (1.28, 1.63)
No	2.83 (2.67, 2.98)	Reference	Reference
Men
Yes	6.69 (6.42, 6.70)	2.85 (2.61, 3.11)	1.43 (1.23, 1.67)
No	2.49 (2.33, 2.67)	Reference	Reference
Women
Yes	9.36 (8.85, 9.88)	2.79 (2.47, 3.16)	1.41 (1.13, 1.78)
No	3.71 (3.39, 4.06)	Reference	Reference

^aMatched for age, sex, general practice, index year and registration year.

^bAdditionally adjusted for BMI, smoking, alcohol consumption, ischaemic heart disease, heart failure, heart valve disease, hyperthyroidism and other comorbidities and medications.

Follow-up data after index date

As shown in Fig. 1, the cumulative probability for incident AF was significantly higher in gout patients than in matched controls at all times since index date (log-rank test P < 0.001). The cumulative probability of AF in gout patients who were free of AF at first presentation was 1.08% at 1 year, 2.03% at 2 years, 4.77% at 5 years and 9.68% at 10 years after index date. In comparison, the cumulative probability of AF in controls was 0.43, 1.08, 2.95 and 6.33% at 1, 2, 5 and 10 years after index date (log-rank test P < 0.001). After a median follow-up of 9 years, 3534 gout patients and 2322 matched controls who were free of AF at index date developed AF. The matched HR was 1.66 (95% CI 1.59, 1.74) and adjusted HR was 1.09 (95% CI 1.03, 1.16). Separate estimates for men and women are shown in Table 3. The unadjusted and adjusted HR estimates were similar in magnitude for both men and women.

Fig. 1

Cumulative probability for AF in gout patients (red line) and matched controls (blue line)

Table 3

Incidence of and hazard ratios for atrial fibrillation by gout status, stratified by sex

Gout	AF incidence (per 1000 person years; 95% CI)	Matched hazard ratio (95% CI)a	Adjusted hazard ratio (95% CI)b
Overall
Yes	9.68 (9.36, 10.00)	1.61 (1.53, 1.70)	1.09 (1.03, 1.16)
No	6.11 (5.86, 6.36)	Reference	Reference
Men
Yes	8.85 (8.50, 9.20)	1.60 (1.51, 1.70)	1.09 (1.01, 1.16)
No	5.60 (5.33, 5.89)	Reference	Reference
Women
Yes	12.19 (11.48, 12.92)	1.61 (1.53, 1.70)	1.12 (1.02, 1.24)
No	7.53 (7.00, 8.08)	Reference	Reference

^aMatched for age, sex, general practice, index year and registration year.

^bAdditionally adjusted for BMI, smoking, alcohol consumption, ischaemic heart disease, heart failure, heart valve disease, hyperthyroidism and other comorbidities and medications.

Sensitivity analysis

Next, we undertook a sensitivity analysis by replacing missing values (alcohol consumption, smoking and BMI categories) by a new category and performing logistic regression and Cox proportional models again. As shown in Table 4, all estimates were very similar to estimates in the primary analysis.

Table 4

Sensitivity analysis with missing value replaced by a new category

Gout	Matched OR (95% CI)a	Adjusted OR (95% CI)b	Matched hazard ratio (95% CI)a	Adjusted hazard ratio (95% CI)b
Overall
Yes	2.89 (2.70, 3.09)	1.45 (1.29, 1.62)	1.66 (1.59, 1.74)	1.09 (1.03, 1.16)
No	Reference	Reference	Reference	Reference
Men
Yes	2.94 (2.70, 3.20)	1.47 (1.27, 1.69)	1.60 (1.51, 1.70)	1.09 (1.01, 1.16)
No	Reference	Reference	Reference	Reference
Women
Yes	2.79 (2.49, 3.14)	1.34 (1.08, 1.65)	1.66 (1.52, 1.81)	1.12 (1.01, 1.23)
No	Reference	Reference	Reference	Reference

^aMatched for age, sex, general practice, index year and registration year.

^bAdditionally adjusted for BMI, smoking, alcohol consumption, ischaemic heart disease, heart failure, heart valve disease, hyperthyroidism and other comorbidities and medications.

Discussion

This study used a large primary care database in the UK to compare the risk of AF in gout patients with matched controls at diagnosis (index date) and also to estimate the risk of future AF diagnosis. The prevalence of AF was 2-fold higher in gout patients than controls at diagnosis. Furthermore, ∼12% of gout patients developed AF within 5 years from the diagnosis whereas only 6% of matched controls developed AF. The incidence of AF was 60% higher in AF-free gout patients than matched controls. After controlling for known predictors for AF (such as ischaemic heart disease, heart failure, valvular heart disease and hyperthyroidism) [25], other comorbidities and medication use, gout was still independently associated with AF although the magnitude of association was diminished. Gout is probably an independent risk factor for AF, despite it also being associated with many comorbidities that also contribute to development of AF. Overall, the burden of AF is very high at diagnosis of gout and the risk of developing new comorbidity is also elevated in incident gout patients compared with the general population.

Currently, there is no explicit explanation for the link between gout and AF. The potential mechanism underlying the increased risk of AF in gout patients is hyperuricaemia. Increasing evidence suggests that uric acid participates in the atrial remodelling process that enhances the risk of AF [38]. Previous studies suggesting a higher risk of AF in individuals with hyperuricaemia are scarce. Cross-sectional studies generally found a higher prevalence of AF in hyperuricaemic patients with or without heart diseases [39–43]. However, these studies were retrospective in nature and many were based on hospital records, which may introduce selection bias. The prospective studies supporting the link between hyperuricaemia and AF include the Atherosclerosis Risk in Communities study, which observed 15 382 AF-free patients aged 45–64 years for a median of 16.8 years [44]. The study found that 1 s.d. increase in serum uric acid levels is associated with a HR of 1.56 for AF in black Americans but no significant association was identified for white Americans. A recent study following 400 patients with type 2 diabetes found that a 1 s.d. increment serum uric acid level was associated with ∼2.5-fold increase in the risk of incident AF [45]. Our finding with a higher risk of AF in gout patients at diagnosis supports the assumption that hyperuricaemia is associated with an increased risk of AF because invariably gout patients are exposed to long-term hyperuricaemia.

After diagnosis, the incidence of AF in gout patients is still 60% higher than their matched controls. However, the adjusted HRs for AF diminished, despite it still being statistically significant. There are two potential explanations. Firstly, some patients may be treated by urate-lowering treatment that may then reduce their risk of developing AF. However, against this is the finding in our recent studies that the majority of gout patients did not receive urate-lowering treatment [1] even long after the initial diagnosis [35]. Secondly, hyperuricaemia not only increased the risk of AF but also increased the risk of other cardiovascular diseases that hasten the development of AF. For example, gout is consistently associated with heart failure [11, 12] and ischaemic heart disease [5–9], both of which are established risk factors for AF [25]. Since many gout patients already had such risk factors for AF at or shortly after diagnosis [11], the effect of gout on AF may be diluted. Nevertheless, the absolute risk of AF in gout patients is ∼60% more than age- and sex-matched controls. Therefore, vigilance for AF should be entertained for all gout patients at diagnosis or thereafter.

AF is one of the most important risk factor for stroke [46]. Our findings linking gout and AF support previous studies documenting that gout patients have a higher risk of stroke [10, 11]. We previously compared the risks of comorbidities in 39 111 patients with incident gout and 39 111 matched controls using the CPRD and found that gout was associated with an OR of 1.50 for stroke at diagnosis and with a HR for 1.29 for stroke after diagnosis [11]. This finding was supported by another study linking CPRD and data from secondary care [10]. Collectively these findings indicate that gout patients are at risk for both AF and stroke. Our study cannot differentiate the impact of gout and AF on stroke, but gout is known to have a high burden of cardiovascular comorbidities in addition to AF [11], so cardiovascular vigilance seems warranted and current guidelines and recommendations [47, 48] support this practice. However, none of these practice recommendations and guidelines specifically advise undertaking an electrocardiogram on gout patients, which is an inexpensive screening for AF. Therefore, this study supports the case for a clinical cardiovascular assessment and inclusion of an electrocardiogram as a part of the initial assessment of gout patients at diagnosis and close observation, for example, by annual assessment, for the occurrence of incident AF, especially for the elderly and those having other AF risk factors.

There are several limitations to this study. Firstly, misclassification bias could exist since the identification of gout patients was based on physician diagnosis, rather than according to classification criteria [49, 50] or to the gold standard of urate crystal identification. However, the validity of gout diagnosis in the CPRD has been investigated and found to be high [30]. Similarly there may have been some misclassification of AF, though again this diagnosis has been validated previously and the recordings of a diagnosis of AF are generally reliable [20, 36]. In addition, there is no reason to suspect a differential misclassification of AF between gout and non-gout patients. Secondly, differential ascertainment bias between incident gout patients and controls cannot be excluded entirely. However, we have adjusted for consultation frequency in our models and the observation periods before and after the index date were comparable between cases and controls. Thirdly, potential confounders may exist and bias our results toward null. For example, the Framingham Heart Study incorporates age, sex, systolic blood pressure, treatment for hypertension, PR interval, clinically significant cardiac murmur, BMI and heart failure into a risk prediction model for AF [51]. Among these factors, the PR interval on an electrocardiogram and clinically significant cardiac murmurs are not available in the CPRD.

In conclusion, gout patients have a higher risk of AF at diagnosis and the risk is also higher afterwards. This study suggests using an electrocardiogram as a part of the initial assessment of gout patients at diagnosis and close observation for the occurrence of incident AF after initial diagnosis of gout.

Acknowledgements

The study was supported by the University of Nottingham with methodological assistance. The sponsors of the study had no role in design and conduct of the study; collection, management, analysis and interpretation of the data; and preparation, review or approval of the manuscript. C.D.M. is funded by the National Institute for Health Research (NIHR) Collaborations for Leadership in Applied Health Research and Care West Midlands, the NIHR School for Primary Care Research and a NIHR Research Professorship in General Practice (NIHR-RP-2014-04-026). The views expressed are those of the authors and not necessarily those of the National Health Service (NHS), the NIHR or the Department of Health. Author contributions: C.F.K. had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. C.F.K., W.Z. and M.D. conceived and designed the study. C.F.K. and W.Z. obtained the funding and acquired the data. C.F.K., M.J.G. and W.Z. performed and supervised the statistical analysis. C.F.K., M.J.G., C.M., W.Z. and M.D. analysed and interpreted the data. C.F.K. and W.Z. drafted the manuscript. All authors contributed to the critical revision of the manuscript for important intellectual content. W.Z., M.J.G. and M.D. supervised the study. M.D. is the guarantor.

Funding: This work was funded by the National Science Council of Taiwan (project 103-2314-B-182A-070-MY2) and Chang Gung Memorial Hospital (project CMRPG3A0624).

Disclosure statement: M.D. is on ad hoc advisory boards concerning osteoarthritis and gout for AstraZeneca, Nordic Biosciences and Novartis; and has received a grant from AstraZeneca for a Sons of Gout Investigator-led research project. W.Z. was a member of the Advisory Board and received consultation honorarium from Savient Plc for pegloticase between 2012 and 2014. All other authors have declared no conflicts of interest.