A cohort study of the association between serum urate and cardiovascular disease in a New Zealand population.

1. Rheumatology Advance Access published October 11, 2012
RHEUMATOLOGY
Original article doi:10.1093/rheumatology/kes269
Is serum urate causally associated with incident
cardiovascular disease?
Simon Thornley1, Roger J. Marshall1, Rod Jackson1, Dudley Gentles1,
Nicola Dalbeth2, Sue Crengle3, Andrew Kerr1,4 and Sue Wells1
Abstract
Downloaded from http://rheumatology.oxfordjournals.org/ at University of Auckland on October 14, 2012
Objective. With studies reporting both positive and negative associations, the influence of serum urate on
incident cardiovascular disease (CVD) is uncertain. We sought to determine whether serum urate is caus-
ally associated with incident CVD.
Methods. Participants were aged 30–80 years and were screened for CVD risk in primary care between
2006 and 2009. Participants had blood pressure, lipids, age and ethnic group recorded at assessment,
with record linkage providing drug dispensing, hospital diagnoses and laboratory test results. Outcomes
were derived from hospital diagnoses and mortality records until December 2009. Cox models were used
to assess the influence of exposures on outcomes.
Results. A total of 78 707 people, free of CVD, were enrolled, and 1328 CVD events occurred during
follow-up. Serum urate was recorded before baseline assessment in 43% (34 008/78 707) of participants.
After adjustment for confounding factors, a 2 S.D. difference in serum urate (0.45 vs 0.27 mmol/l) was
associated with a hazard ratio (HR) of 1.56 (95% CI 1.32, 1.84). This was more than double that of the
equivalent distributional change in high-density lipoprotein cholesterol (adjusted HR 1.22) and one-third
greater than that for HbA1c (adjusted HR 1.41).
Conclusion. Serum urate is likely to be causally associated with CVD. This supports public health action
to reduce urate levels in populations with significant burdens of the disease.
Key words: cardiovascular diseases, risk factors, epidemiology.
CLINICAL
SCIENCE
Introduction
characterizes research on the effect of serum urate on
‘Measurement of serum uric acid levels is unlikely to en- the incidence of cardiovascular disease (CVD), reflected
hance usefully the prediction of [coronary heart disease] in these two contrasting quotes from European and
CHD, and this factor is unlikely to be a major determinant Japanese researchers, respectively. Meta-analyses of ob-
of the disease in general populations’ [1]. ‘These results servational studies [3, 4] have reported some association
showed that hyperuricaemia has a strong association between hyperuricaemia (serum urate >400 mmol/l) and
with . . . death in all causes, coronary heart disease, . . . and incident CHD. One study [3], for example, reported an
indicated that serum uric acid seems to be a considerable increase in risk of CVD (pooled relative risk 1.46; 95%
risk factor for reduced life expectancy’ [2]. Uncertainty CI 1.20, 1.73), but the association diminished with
adjustment for established risk factors (pooled relative
1
Section of Epidemiology and Biostatistics, School of Population risk 1.09; 95% CI 1.03, 1.16). Such factors included age,
Health, 2Department of Medicine, Faculty of Medicine and Health systolic blood pressure, cholesterol, smoking and either
Sciences, 3Te Kupenga Hauora Maori, School of Population Health,
University of Auckland and 4Counties Manukau District Health Board,
a diagnosis or other laboratory indices of diabetes.
Hospital Road, Otahuhu, Auckland, New Zealand. However, generally it is not clear which factors to adjust
Submitted 19 June 2012; revised version accepted 29 August 2012. for. Some, such as systolic blood pressure, may be on the
Correspondence to: Simon Thornley, Section of Epidemiology and causal pathway between serum urate and CVD, thus
Biostatistics, Level 4, School of Population Health, Tamaki Innovation mediating the relationship rather than acting as a confoun-
Campus, Corner of Merton and Morrin Roads, Glen Innes, The
University of Auckland, Private Bag 92019, Auckland 1142, New
der, since there is evidence that high levels of serum
Zealand. Email: s.thornley@auckland.ac.nz urate causally influence the onset of hypertension [4, 5].
! The Author 2012. Published by Oxford University Press on behalf of the British Society for Rheumatology. All rights reserved. For Permissions, please email: journals.permissions@oup.com 1

2. Simon Thornley et al.
Thus whether serum urate is causally associated with test in the 5-year period before their PREDICT assess-
CVD remains uncertain. ment; otherwise data for these variables were missing.
The prevalence of gout is extremely high, by interna-
tional standards, among Maori ($6%) and Pacific people
¯ Outcomes
($8%) living in New Zealand [6, 7]. Together these ethnic
CVD events (acute coronary syndromes, coronary pro-
groups comprise between 20% and 25% of the total popu-
cedures, stroke, transient ischaemic attack, haemorrhage
lation, so serum urate is measured frequently in New
stroke, peripheral vascular disease, peripheral arterial
Zealand. We have been conducting a large cohort study
procedures and congestive heart failure) were identified
of CVD risk factors that involves linking risk factor data
from hospital discharge and mortality records using
derived from primary care-based screening to laboratory
International Classification of Disease codes, as previ-
results, drug dispensing, hospital admission diagnoses,
ously described [11]. They covered the period from
and mortality records. Here we explore the relationship
January 2006 to December 2009. The analysis was of
between urate levels and the incidence of CVD.
the outcome of any CVD event, including immediate
death that was coded as caused by such conditions.
Methods
Downloaded from http://rheumatology.oxfordjournals.org/ at University of Auckland on October 14, 2012
Study design Other data
The PREDICT cohort has been described elsewhere [8, 9]. Patient medications, specifically anti-hypertensive and
Briefly, PREDICT is a web-based program for assessing gout medication, were obtained from the New Zealand
and managing CVD risk that is integrated with commonly Pharmaceutical Information Database [12], a register of
used general practice electronic medical records in New community dispensing, and were merged with PREDICT
Zealand. PREDICT generates a standardized cardiovas- records. Anti-hypertensive drugs included b-blockers,
cular risk profile on each patient and calculates a 5-year calcium channel blockers, angiotensin converting
percentage CVD risk based on the patient’s age, gender, enzyme (ACE) inhibitors, a-blockers and thiazide diuretics.
ethnic group, blood pressure, total to high-density lipo- Gout medication included allopurinol, colchicine and
protein (HDL) cholesterol ratio, family history of CVD, probenecid.
smoking and diabetes status, using Framingham Demographic information was taken from the PREDICT
study-derived equations [10]. PREDICT clinical informa- database and, in the case of ethnicity, the variable was
tion is either entered by the general practitioner at the recorded in both PREDICT and national data sources.
time of risk assessment (blood pressure, diabetes Ethnicity data were classified according to a standardized
status, family history of CVD, smoking status and family data protocol [13], in which Maori and then Pacific ethnic
¯
history of CVD) or derived from practice management groups are prioritized if more than one ethnic group was
software from information stored about the patient (age, recorded. Individuals were coded as either Maori, Pacific,
¯
gender, ethnic group and total to HDL cholesterol ratio). Indian or Other. The Other group were mainly of New
A risk profile is stored for each patient and is linked to an Zealand European descent.
encrypted, unique, National Health Index. The index
enables linkage with national health records, such as hos- Analysis
pital discharge codes, mortality, community drug dis- Statistical analysis centred on developing a survival model
pensing and laboratory test results. Enrollees between for incidence of any CVD event, taking serum urate as a
January 2006 and 15 October 2009 were selected for potential predictor. We used a Cox regression model, with
this analysis since full dispensing and coded mortality re- age at event as the time variable rather than time to event
cords were available for this period. Those aged <30 or from baseline enrolment. This approach is increasingly
>80 years at the time of screening were excluded. We recommended for observational epidemiological analysis
also excluded people with a history of prior CVD or of CVD [14, 15]. Time, therefore, was considered as left
heart failure, identified by a general practitioner diagnosis truncated (patients are observed conditional on survival
of CVD or hospitalization with CVD in the past 5 years, or up until that point) and right censored. Proportional
those dispensed a loop diuretic in the 6 months before hazards assumptions were checked using scaled
assessment who were assumed to have heart failure. Schoenfeld residuals [16]. Restricted cubic splines were
used to investigate the relationship between continuous
Serum urate variables and time to event as a check on the modelling
The urate concentration of enrollees in PREDICT were ob- assumption of linearity. HRs for continuous variables were
tained by anonymized linkage to community laboratory reported by comparing the relative hazard between the
data in the Diagnostic Medlab database. The database 16th and 84th centiles of the variable (1 S.D. either side
was the sole provider of laboratory tests, outside hospital, of the mean for a normal distribution) [17]. This allows
to the greater Auckland region until late 2009. If a person direct comparison with relative hazards of binary vari-
had had many serum urate measurements, we used the ables, since the 16th and 84th centiles are roughly equiva-
one prior to their first PREDICT assessment. Serum urate lent to the one-unit difference between 0 and 1 of a binary
and other laboratory variables (HbA1c and HDL choles- variable. That is, 2 S.D.s on a binary scale with a mean of
terol) were only recorded if a patient had had a laboratory $0.5 is 1.
2 www.rheumatology.oxfordjournals.org

3. Serum urate and incident CVD
FIG. 1 A DAG, characterizing the influences on risk of recordings and other unobserved influences (colliders),
developing CVD. we believed that including such variables was more
likely to result in biased effect estimates than excluding
them from the analysis [19].
We also tested for interactions, in particular, between
serum urate and gender. From other studies of CVD using
similar risk factor profiles, effect estimates have been
observed to change with age. These were tested using
the likelihood ratio (P < 0.05). When using age at event as
the time variable, such interactions may manifest as viola-
tions of the proportional hazards assumption. We tested for
such violations using the cox.zph function, which calculates
tests of the proportional hazards assumption for each cov-
ariate by correlating the corresponding set of scaled
Schoenfeld residuals with a suitable transformation of
Downloaded from http://rheumatology.oxfordjournals.org/ at University of Auckland on October 14, 2012
time, based on the Kaplan–Meier estimate of the survival
function [20]. Stratification allowed different baseline haz-
Dark font variables are those that are observed, whereas
grey font variables represent those that are unobserved. ards for exposure groups, where the proportional hazards
t: variable recorded at baseline; tÀ1: variable recorded assumption was likely to be contravened.
before baseline; LDL: low-density lipoprotein cholesterol; Because many serum urate and other laboratory values
Trigs: serum triglycerides; meds: medications. were missing, we used multiple imputation, so that the
modelling results reported were based on averaging
model parameters over 10 random imputations [16].
To help decide model structure and which variables to Variables that were used in the multiple imputation mod-
include in a model, we sketched a diagram of likely causal elling were gender, serum metabolic markers (creatinine,
mechanisms and pathways [a directed acyclic graph HDL cholesterol, triglycerides, HbA1c and urate), age at
(DAG)] [18], and used some of the ideas proposed by enrolment, systolic blood pressure, smoking status, dia-
Pearl to examine causality. The DAG we constructed betes, ethnic group, death and CVD event, along with an
included measured and unmeasured influences on CVD interaction between serum urate and sex.
risk that we felt were important (Fig. 1). Lines with an end All analyses were done using R software (version 2.14.1)
arrow represent a causal link acting in the direction of the [21]. The package rms [22] was used for the Cox regres-
arrow. Those lines that are solid represent a link that can sion analysis and the functions aregImpute and
potentially be observed from our data, dashed lines are fit.mult.impute for multiple imputation. The aregImpute
causal mechanisms we believe to hold but cannot be function finds transformations that optimize how each
observed in our data. variable may be predicted from every other variable,
From Fig. 1, we considered a diagnosis of gout (derived using additive semiparametric models. The fit.mult.impute
from dispensing data) as likely to be a mediator of the function was then used to average sets of regression
influence of urate on CVD risk. We assumed that lipopro- coefficients and compute variance and covariance, ad-
tein concentrations are causally associated with CVD, and justed for the error derived from the uncertainty from im-
so they are included in the model. Blood pressure (BP), putation of missing data. This project was approved by
measured before baseline (BPtÀ1 in Fig. 1) and measured the national Multi Region Ethics Committee in 2007 (MEC/
at baseline (BPt), were considered mediators of the influ- 07/19/EXP).
ence of urate on CVD risk, and were therefore not con-
trolled for in the model. Blood pressure lowering and statin Results
therapies were considered to represent colliders [18],
which may introduce bias from an unobserved variable Table 1 shows the baseline characteristics of the
that would otherwise not influence our analysis, so these PREDICT cohort, that is, when first enrolled in the
variables were excluded. Although certain blood pressure PREDICT database. Men (56%) outnumbered women,
medications, such as thiazide diuretics, are known to raise and the mean age was 55 years, with women rather
serum urate, we believe that this is a relatively small influ- older. Pacific people accounted for 23% of the cohort,
ence on the causal paths compared with those otherwise Maori 16% and Indian 7%. The remainder of the cohort
¯
identified in the DAG. Furthermore, blood pressure and and majority ethnic group, Other, was 86% European, 6%
treatment of the condition were regarded as mediators Chinese and 8% composed of a large variety of ethnic
(caused by high urate levels), as was a diagnosis of groups. More women than men were diagnosed with dia-
gout. In view of these considerations, to block backdoor betes; however, HbA1c levels were equivalent. Men had
paths, using Pearl’s terminology (adjust for confounding higher levels of serum urate, higher lipid ratios and lower
variables), the regression analysis adjusted for gender, levels of HDL. Women were less likely than men to smoke
HbA1c, ethnic group, smoking status and lipoprotein con- cigarettes.
centrations (HDL cholesterol). Because post-treatment Urate values were approximately normally distributed,
variables are likely to be strongly influenced by baseline with a lower mean in women compared with men.
www.rheumatology.oxfordjournals.org 3

4. Simon Thornley et al.
TABLE 1 Baseline characteristics, by sex
Characteristic Men Women Total
Total 43 815 34 892 78 707
Age at baseline, mean (S.D.), years 52.9 (10.5) 57.1 (9.8) 54.8 (10.4)
Ethnic group, n (%)
Other 23 971 (54.7) 19 311 (55.4) 43 282 (55.0)
Maori
¯ 6578 (15.0) 5696 (16.3) 12 274 (15.6)
Pacific 9967 (22.8) 7766 (22.3) 17 733 (22.5)
Indian 3299 (7.5) 2119 (6.1) 5418 (6.9)
Serum urate, mean (S.D.), mmol/l; n=34 008, (43%) 0.39 (0.09) 0.32 (0.09) 0.36 (0.09)
HbA1c (%); n = 33 075 (42%)
Mean (S.D.) 6.66 (1.71) 6.68 (1.67) 6.67 (1.69)
Diagnosis of diabetes, n (%) 7709 (17.6) 7170 (20.6) 14 879 (18.9)
Current smoker, n (%) 8630 (19.7) 5370 (15.4) 14 000 (17.8)
Systolic BP, mean (S.D.), mmHg 130.3 (17.0) 131.3 (18.4) 130.8 (17.7)
Downloaded from http://rheumatology.oxfordjournals.org/ at University of Auckland on October 14, 2012
Total cholesterol/HDL ratio, mean (S.D.) 4.31 (1.35) 3.72 (1.17) 4.04 (1.31)
HDL cholesterol, mean (S.D.), mmol/l; n = 68 224 (87%) 1.30 (0.35) 1.55 (0.43) 1.41 (0.41)
TABLE 2 Serum urate, by sex, ethnic group, diabetes and FIG. 2 Density plot comparing the distributions of
smoking status observed (dashed line) and imputed urate values
(solid line).
Serum urate,
mean (S.D.),
mmol/l
Characteristic
Men Women
Ethnic group
Other 0.37 (0.08) 0.30 (0.08)
Maori
¯ 0.41 (0.09) 0.34 (0.09)
Pacific 0.42 (0.09) 0.36 (0.09)
Indian 0.35 (0.08) 0.29 (0.07)
Diabetes
Yes 0.37 (0.09) 0.34 (0.09)
No 0.39 (0.09) 0.32 (0.08)
Current smoker
Yes 0.39 (0.09) 0.33 (0.09)
No 0.39 (0.09) 0.32 (0.09)
Initial model checking revealed that the proportional
hazard assumption was not supported for gender and eth-
There were, however, 57% missing serum urate levels, nicity. For this reason the model was stratified on these
58% missing HbA1c and 13% missing HDL cholesterol. factors, meaning that separate baseline hazards were
No other data items had missing values. Mean assumed for all permutations of gender and ethnic
observed urate values varied substantially by ethnic group. A gender by urate and ethnic group by urate inter-
group, but varied little by diabetes or smoking status action was tested for, but neither was found to be statis-
(Table 2). Maori and Pacific ethnic groups were highest,
¯ tically significant.
whereas mean levels among the Other and Indian Table 3 shows the results of the Cox regression analysis.
ethnic groups were about half to two-thirds of an S.D. In the model, comparing measures at the 16th and 84th
lower. centiles showed an HR of 1.56 (95% CI 1.32, 1.84) for
A total of 1328 CVD events occurred during follow-up, serum urate. The relative hazard was linear (on the loga-
with 167 patients dying during this period. Median follow- rithmic scale), which indicated that an increase in serum
up time was 538 days, with a maximum of 1424 and a urate of 0.29 mmol/l [=ln(2)/(

5. urate))] doubled the relative
minimum of 1. The distributions of observed and imputed hazard of CVD across the distribution of observed urate
urate values (Fig. 2) show that the variance is reduced values. In the study population, the association between
among imputed compared with observed values, because a 2 S.D. difference in urate level (67% relative increase) on
imputed values are regressed to the mean. incident CVD risk was higher than adjusted effects of the
4 www.rheumatology.oxfordjournals.org

6. Serum urate and incident CVD
TABLE 3 Measures of association with CVD from a Cox proportional hazards model
Comparisona
Crude HRb (95% CI) Adjusted HRc (95% CI)
Factor High Low
Serum uratea, mmol/l 0.45 0.27 1.76 (1.56, 1.99) 1.56 (1.32, 1.84)
Non-smoker 1.00 0.00 1.71 (1.50, 1.95) 1.63 (1.43, 1.86)
HbA1ca, % 8.00 5.50 1.41 (1.29, 1.54) 1.41 (1.29, 1.55)
Serum HDLa, mmol/l 1.03 1.79 1.75 (1.49, 2.04) 1.22 (1.03, 1.45)
n = 78 707. aComparisons for the continuous variables are taken at the 84th and 16th centiles (z = 1) for comparability with
binary variables. The binary variables are, by definition, set at 0 and 1. bStratified by sex and ethnic group. cAdjusted for all
variables in the table and stratified by gender and ethnic group.
Downloaded from http://rheumatology.oxfordjournals.org/ at University of Auckland on October 14, 2012
For a 60-year-old male smoker in the Other ethnic
FIG. 3 Measures of association with CVD from a Cox
group with a serum urate level of 0.27 mmol/l, HbA1c of
model.
6% and serum HDL cholesterol of 1.34 mmol/l, the 5-year
risk of CVD equates to 1 À (0.813/0.871) = 6.7%. If his
serum urate is about 2 S.D. greater, that is 0.45 mmol/l,
his 5-year cumulative incidence of CVD will increase to
10.1% (about a one-third increase). This calculation is
based on an estimate of the baseline survival (S0) of
88.9% at 60 years and 83.8% at 65 years.
Discussion
We found that, given plausible assumptions, raised serum
urate is likely to have a substantial causal effect on the
incidence of CVD, of a similar or greater magnitude to the
effects of high HbA1c or low HDL cholesterol levels. A
Cox-modelled survival estimates (with 95% confidence strength of the study was that it was based on a large
intervals) by time for serum urate at the 16th centile cohort, n = 79 707, in which the information had been col-
(0.27 mmol/l) compared with the 84th centile (0.45 mmol/l) lected in a systematic and standardized way. An asso-
for a non-smoking male in the Other ethnic group, with ciated weakness, however, is the incompleteness of the
HDL cholesterol = 1.34 mmol/l and HbA1c = 6%. laboratory data, and in particular of the serum urate
values. To deal with these missing values, we used mul-
tiple imputation, which is recommended to reduce bias in
equivalent change in the distribution of HbA1c (37% rela- effect estimates when compared with complete-case
tive increase) and HDL cholesterol (155% relative analysis [16]. In a sensitivity analysis, using complete-
increase). case analysis, our observed effect estimates were similar
An illustration of the extent to which a change in serum to those derived from multiple imputation [the adjusted HR
urate affects survival is shown in Fig. 3. It gives the of the effect of urate was 1.59 (95% CI 1.32, 1.90), com-
Cox-modelled survival plot for a man in the Other ethnic pared with 1.56 in Table 2]. We can only speculate that if
group, with HDL cholesterol of 1.34 mmol/l, HbA1c 6% serum urate had been measured in everyone, the effect
and a non-smoker, showing the average change asso- estimates would not have differed substantially.
ciated with a 2 S.D. difference in serum urate on survival We have also explored sensitivity analyses of multivari-
probability. This association (change in urate on survival ate effect estimates (see supplementary data, available at
probability) increases with age. Rheumatology Online). These included carrying out sep-
From the model, the cumulative incidence over a spe- arate analyses by gender, with and without imputation of
cified period of time, e.g. the 5-year risk, can be derived missing values, varying the time scale chosen, including
using the formula: those who had used loop diuretics at baseline assess-
ment, and dividing urate values by quintile. Generally
Sðt þ Þ
1À these analyses were congruent with the effect estimates
SðtÞ
reported in Table 3.
where t is age of an individual, is the interval over which A positive aspect of the study was that we explicitly
the cumulative incidence is calculated (typically 5 or 10 considered causal paths to determine variables to adjust
years) and S is the survival probability, estimated by the in the model [18]. This process is conceptually appealing
Cox model. and less likely to underestimate the effect of serum urate
www.rheumatology.oxfordjournals.org 5

7. Simon Thornley et al.
or to introduce inappropriate confounder adjustments. the exposure. Some studies have reported effects of
Whether our DAG is an accurate reflection of the causal serum urate on CVD as a continuous variable. For ex-
interaction between the variables in our study may be ample, the Framingham study [23] (n = 6763) found no sig-
debated; for example, whether systolic blood pressure is nificant association between serum urate and CHD
a mediator of the effect of urate or whether it should be incidence in their adjusted Cox model among men, but
included as a confounder in regression equations. To ex- a significant increase in risk with women. We found a stat-
plore the effects of variable selection based on the DAG, istically significant association between urate and CVD,
we carried out a sensitivity analysis, altering the model by without a gender  urate interaction, and so our measures
(i) adding a diagnosis of diabetes in the model; (ii) inter- of association were aggregated over the whole popula-
changing HDL cholesterol with the total cholesterol/HDL tion. Our study reached similar conclusions to the
ratio and (iii) adding blood pressure as a potential con- National Health and Nutrition Examination Survey
founder. The first two changes to the model resulted in (NHANES)-I [24] cohort study (n = 5926), which found
a 2% change in the adjusted HR for serum urate. After that serum uric acid levels were significantly associated
inclusion of diabetes, the effect of HbA1c was reduced with CVD mortality after adjustment for known confoun-
from 1.41 to 1.36 (comparing individuals with baseline ders. The NHANES study used CVD mortality as its out-
Downloaded from http://rheumatology.oxfordjournals.org/ at University of Auckland on October 14, 2012
measures of 8.00 and 5.50%). If systolic blood pressure come event while our study used hospital admission or
is included, our adjusted effect of urate diminished death from CVD. Its advantages were that it had long
slightly [for 0.45 compared with 0.27 mmol/l; the HR follow-up time (16.4 years) and included both white and
reduced from 1.56 (95% CI 1.32, 1.84) to HR = 1.48; black participants (not present in the Framingham cohort).
(95% CI 1.25, 1.75)]. The authors reported age-adjusted HRs for cardiovascu-
Family history of ischaemic heart disease may also be lar mortality in participants aged 45–54 years that were
considered a confounder of the relationship between similar to ours: 1.28 (95% CI 1.08, 1.52) in men and 1.43
urate and CVD. For this study, family history of CVD was (95% CI 1.16, 1.37) in women for each 0.06 mmol/l rise in
recorded as a history of CHD or ischaemic stroke in a serum urate level. Our study, using age as the
first-degree relative (father or brother 55 years, mother time-to-event variable, is likely to adjust for the effect of
or sister 65 years). When this variable was added, how- age more accurately than the more traditional methods
ever, the point and interval estimate for serum urate re- used in the NHANES study, which aggregated age into
mained unchanged (see supplementary data, available at 15-year intervals. Despite these differences in methods,
Rheumatology Online). this study was concordant with our findings.
The nature of the cohort, composed of a generally high Our study provides statistical evidence that elevated
CVD risk population rather than a designed sample, is also serum urate is likely to cause CVD. Biological evidence
a potential weakness. The high prevalence of diabetes also supports our findings, highlighting the role of serum
($20%) reported among the cohort along with the urate in the pathogenesis of endothelial injury [25]. If this
gender distribution favouring men was expected. causal relationship is true, attention turns to what deter-
Nevertheless, there is substantial heterogeneity in the mines serum urate, and how serum urate may be lowered,
risk profile of the study population, so there is no reason either by lifestyle modification, or through drug treatment.
to believe that the lack of representativeness was an im- In addition to the role of purine-rich foods, fructose intake
portant problem for these analyses. may also contribute to hyperuricaemia through increased
Using age as the outcome variable in the Cox model hepatic synthesis of purines [26, 27]. This gives impetus
imposes an assumption that entry into the study is inde- for public health action to limit intake of sugar, the dom-
pendent of age. This is unlikely to be true here, given that inant source of fructose in the diet, as well as purine-rich
national CVD screening guidelines recommend initiation foods. In addition, this finding raises the question of
of screening at specified ages, which differ for varying whether pharmaceutical reduction of urate levels, with
ethnic groups. However, use of age in survival modelling urate-lowering therapy such as allopurinol, can reduce
is now the recommended approach for cohort studies that CVD risk. Such an idea is strengthened by the results of
estimate CVD incidence [15]. This choice of time scale is a small (n = 65) cross-over trial of allopurinol therapy in
justified because the use of age at event results in less patients with symptomatic, effort-dependent angina.
bias of effect estimates compared with time to event [16]. Treatment with 600 mg of allopurinol per day for
This method is increasingly recommended if the period of 6 weeks increased the time to S-T segment depression
observation is the time from birth to event, as it is when by 43 s (95% CI 31, 58) compared with otherwise equiva-
modelling CVD risk, rather than time from intervention to lent treatment with placebo [28].
event, as it is in a randomized trial [15]. Maori and Pacific people in this study had mean levels
¯
There are some differences and similarities between our of serum urate $15% higher than Others, Indians and
results and other studies. Some studies report the asso- other Asians. We speculate that dietary rather than gen-
ciation between raised serum urate in a binary fashion etic differences are more likely to account for these differ-
(presence or absence of hyperuricaemia) [3]. In our ences, because New Zealand Maori almost all share
¯
study we found evidence of a log-linear relationship be- significant ancestry with New Zealand Europeans due to
tween urate concentration and CVD risk, so studies that widespread intermarriage, yet their mean urate values
employed a cut-off are likely to underestimate the effect of vary substantially. In our regression analysis, to obtain a
6 www.rheumatology.oxfordjournals.org

8. Serum urate and incident CVD
model with good fit, we stratified by ethnicity rather than 9,458 incident cases and 155,084 controls: prospective
evaluating the effects of this variable. However, because study and meta-analysis. PLoS Med 2005;2:e76.
CVD incidence and prevalence are significantly higher in 2 Tomita M, Mizuno S, Yamanaka H et al. Does
Maori and Pacific peoples compared with Europeans
¯ hyperuricemia affect mortality? A prospective cohort
[29–31], future work will evaluate whether variations in study of Japanese male workers. J Epidemiol 2000;10:
serum urate levels account for ethnic variations in the 403–9.
prevalence of CVD. In conclusion, serum urate is asso- 3 Kim SY, Guevara JP, Kim KM, Choi HK, Heitjan DF,
ciated with incident CVD and is likely to be a significant Albert DA. Hyperuricemia and coronary heart disease: a
factor in the disease’s causal pathway. systematic review and meta-analysis. Arthritis Care Res
2010;62:170–80.
¨
4 Sundstrom J, Sullivan L, D’Agostino RB, Levy D,
Rheumatology key messages
Kannel WB, Vasan RS. Relations of serum uric acid to
. Substantial uncertainty exists about the effect of longitudinal blood pressure tracking and hypertension
serum urate on CVD. incidence. Hypertension 2005;45:28–33.
. After making causal assumptions, we found a 5 Feig DI, Soletsky B, Johnson RJ. Effect of allopurinol on
strong association between urate and CVD.
Downloaded from http://rheumatology.oxfordjournals.org/ at University of Auckland on October 14, 2012
blood pressure of adolescents with newly diagnosed es-
. Urate differs substantially between ethnic groups, sential hypertension. JAMA 2008;300:924–32.
and may explain differences in CVD burden.
6 Winnard D, Kake T, Gow P et al. Debunking the myths to
provide 21st century management of gout. N Z Med J
2008;121:79–85.
Acknowledgements 7 Winnard D, Wright C, Taylor WJ et al. National prevalence
of gout derived from administrative health data in
The authors would like to thank the National Health Board Aotearoa New Zealand. Rheumatology 2012;51:901–9.
Analytic Services and Diagnostic Medlab Ltd for supplying 8 Bannink L, Wells S, Broad J, Riddell T, Jackson R. Web-
anonymized data. They would also like to thank affiliated based assessment of cardiovascular disease risk in rou-
general practitioners and practice nurses and patients be- tine primary care practice in New Zealand: the first 18,000
longing to the ProCare Network, Auckland PHO Ltd, patients (PREDICT CVD-1). N Z Med J 2006;119:U2313.
HealthWest, East Health Services, National Maori ¯ 9 Wells S, Kerr A, Broad J, Riddell T, Kenealy T, Jackson R.
Hauora Coalition, Alliance, Te Tai Tokerau and Manaia The impact of New Zealand CVD risk chart adjustments
PHOs. for family history and ethnicity on eligibility for treatment
(PREDICT CVD-5). N Z Med J 2007;120:U2712.
PREDICT was developed by a collaboration of clinical epi-
demiologists at the University of Auckland, IT specialists 10 Anderson KM, Odell PM, Wilson PW, Kannel WB.
at Enigma Publishing Ltd (a private provider of online Cardiovascular disease risk profiles. Am Heart J 1991;
121(1 Pt 2):293–8.
health knowledge systems), primary health care organiza-
tions, non-governmental organizations (New Zealand 11 Riddell T, Wells S, Jackson R et al. Performance of
Guidelines Group, National Heart Foundation, Diabetes Framingham cardiovascular risk scores by ethnic groups
New Zealand, Diabetes Auckland), several district health in New Zealand: PREDICT CVD-10. N Z Med J 2010;123:
50–61.
boards and the Ministry of Health. The PREDICT software
platform is owned by Enigma Publishing Ltd (PREDICT is 12 New Zealand Health Information Service. Pharmaceutical
a trademark of Enigma Publishing Ltd). Information Database (PHARMS). Wellington, New
Zealand: New Zealand Health Information Service, 2011.
Funding: This research was supported by a grant from the 13 Ministry of Health. Ethnicity data protocols for the health
National Heart Foundation of New Zealand and a Health and disability sector. Wellington, New Zealand: Ministry of
Research Council Clinical Research Training Fellowship Health, 2004.
(grant number 11/145).
14 Conroy RM, Pyorala K, Fitzgerald AP et al. Estimation of
Disclosure statement: N.D. has received consultancy fees ten-year risk of fatal cardiovascular disease in Europe: the
from Takeda, Ardea Biosciences, Metabolex and SCORE project. Eur Heart J 2003;24:987–1003.
Novartis. All other authors have declared no conflicts of ´ ´
15 Thiebaut ACM, Benichou J. Choice of time-scale in Cox’s
interest. model analysis of epidemiologic cohort data: a simulation
study. Stat Med 2004;23:3803–20.
16 Harrell FE. Regression modeling strategies. New York:
Supplementary data Springer, 2001.
Supplementary data are available at Rheumatology 17 Thornley S, Marshall R. Measures of association in
epidemiological studies: how best to compare discrete
Online.
and continuous variables? J Biomet Biostat 2012;3:e111.
18 Pearl J. Causality: models, reasoning, and inference. New
References York: Cambridge University Press, 2000.
1 Wheeler JG, Juzwishin KDM, Eiriksdottir G, Gudnason V, 19 Gelman A, Hill J. Causal inference using regression on the
Danesh J. Serum uric acid and coronary heart disease in treatment variable. In: Data analysis using regression and
www.rheumatology.oxfordjournals.org 7

9. Simon Thornley et al.
multilevel/hierachical models. New York: Cambridge 27 Nakagawa T, Hu H, Zharikov S et al. A causal role for uric
University Press, 2007. acid in fructose-induced metabolic syndrome. Am J
20 Therneau TM, Grambsch PM. Modeling survival data: Physiol 2006;290:F625–F631.
extending the Cox model. New York: Springer, 2000. 28 Noman A, Ang DSC, Ogston S, Lang CC, Struthers AD.
21 R Development Core Team. R: a language and environ- Effect of high-dose allopurinol on exercise in patients with
ment for statistical computing. Vienna: R Foundation for chronic stable angina: a randomised, placebo controlled
Statistical Computing, 2007. crossover trial. Lancet 2010;375:2161–7.
22 Harrell FE Jr. rms: regression modeling strategies. 29 Thornley S, Chan WC, Crengle S et al.
R package version. 3.3-2, http://cran.r-project.org/web/ Sociodemographic differences in prevalence of
packages/rms/index.html, 2011. diagnosed coronary heart disease in New Zealand
estimated from linked national health records. N Z Med J
23 Culleton BF, Larson MG, Kannel WB, Levy D. Serum uric
2011;124:21–34.
acid and risk for cardiovascular disease and death: the
Framingham Heart Study. Ann Intern Med 1999;131:7–13. 30 Tobias M, Chan W, Wright C, Jackson R, Mann S,
Yeh L. Can the incidence and prevalence of coronary
24 Fang J, Alderman MH. Serum uric acid and cardiovascular
heart disease be determined from routinely collected
mortality. JAMA 2000;283:2404–10. national data? Population-based estimates for New
Downloaded from http://rheumatology.oxfordjournals.org/ at University of Auckland on October 14, 2012
25 Hayden MR, Tyagi SC. Uric acid: a new look at an old risk Zealand in 2001–03. Aust N Z J Public Health 2008;32:
marker for cardiovascular disease, metabolic syndrome, 24–7.
and type 2 diabetes mellitus: the urate redox shuttle. 31 Chan WC, Wright C, Riddell T et al. Ethnic and
Nutr Metab 2004;1:10. socioeconomic disparities in the prevalence of
26 Emmerson BT. Effect of oral fructose on urate production. cardiovascular disease in New Zealand. N Z Med J 2008;
Ann Rheum Dis 1974;33:276–80. 121:11–20.
8 www.rheumatology.oxfordjournals.org