Empirical dietary index for lower urate concentrations and risk of gout: evidence from cohort studies

Abstract

Background

High blood urate concentrations are a causal risk factor for the development of gout. There is no dietary pattern that specifically targets on lowering plasma urate concentrations or gout risk.

Objectives

This study aimed to derive a dietary pattern that predicts lower plasma urate concentrations and to examine this diet in relation to the risk of gout and related cardiometabolic conditions, including hypertension, coronary artery disease (CAD), stroke, and type 2 diabetes (T2D).

Methods

An Empirical Dietary Index for Normo-Uricemia (EDINU) was developed using 7-d diet records and plasma urate concentrations in the Lifestyle Validation Study (LVS) and prospective associations between the EDINU and disease risks were assessed using multivariable Cox regression in the Nurses' Health Study (NHS) and Health Professionals Follow-up Study (HPFS), using prospective cohort data. Replications were conducted in National Health and Nutrition Examination Survey (NHANES) and UK Biobank.

Results

The EDINU positively ranks low-fat milk, blueberries, grapes, and cheese as negative predictors of urate and negatively ranks mixed vegetables, liquor, red meat, liver, artificially sweetened beverages, tomato products, wine, and salad dressing as positive predictors. The EDINU showed significant correlations with plasma urate concentrations in both discovery and replication studies (Spearman correlation of –0.23 in LVS or –0.33 in NHANES). Higher EDINU scores were associated with lower gout risk in 3 independent cohort studies with a hazard ratio, comparing extreme quintiles, of 0.48 (95% confidence interval: 0.42, 0.55) in the NHS/HPFS or 0.65 (0.48, 0.88) in UK Biobank. The EDINU was inversely associated with a lower risk of hypertension, stroke, and T2D, but not CAD, in the NHS/HPFS.

Conclusions

A replicated empirical index predicting lower plasma urate is associated with significantly lower risks of gout and related cardiometabolic conditions. Consuming such a diet with lower uricemic potentials could be a novel, promising approach to preventing gout.

Introduction

Gout is a rheumatologic condition caused by the formation of monosodium urate crystals in the joints, leading to pain, swelling, and inflammation, which can result in chronic joint damage, disability, reduced quality of life, and is often associated with cardiovascular–kidney–metabolic comorbidities [1]. High plasma urate concentrations are the primary causal risk factor for the development of gout [2,3]. Diet is among the key modifiable risk factors for modulating urate concentrations, as well as the clinical endpoint of gout, as demonstrated in several large population studies [[4], [5], [6]]. However, dietary recommendations for gout prevention continue to emphasize the purine content of individual foods [7,8], which fails to account for the complex interactions between individual food items or nutrients. It remains ambiguous to the public (as well as individuals living with gout) who wish to choose optimal diets for maintaining urate concentrations in the normal range and lowering gout risk, while simultaneously maintaining a high diet quality for the overall health. Few studies have been dedicated to developing a dietary pattern that specifically targets plasma urate concentrations and to examine such a dietary pattern in relation to gout risk. In this regard, it has been a fruitful strategy to use causal risk markers, such as inflammatory cytokines and insulin, as the instrument to identify dietary factors and their combinations that maximize the predictability of such factors in an agnostic fashion, which are linked to health outcomes such as cardiovascular disease (CVD) and type 2 diabetes (T2D) [[9], [10], [11]]. The dietary patterns derived from these causal biomarkers are often strongly associated with the risk of diseases influenced by these factors [10,12], including gout [13], but no studies have been conducted to construct a dietary pattern that specifically targets urate concentrations.

To fill this critical knowledge gap in plasma urate metabolism and gout research, the current study aims to use the same agnostic approach to develop an Empirical Dietary Index for Normo-Uricemia (EDINU) and to evaluate the prospective association between EDINU and incident gout and other related chronic diseases, including hypertension, coronary artery disease (CAD), stroke, and T2D in large prospective cohort studies. To rule out the role of chance in this effort of developing empirical dietary indices, we also sought to replicate the performance of this index and its association with gout risk in independent populations.

Methods

Study population

The current study design comprises 4 components (Figure 1):

• 1)The development of EDINU in the Lifestyle Validation Study (LVS): the LVS is originally designed to validate self-reported data, including diet, among participants from the Nurses’ Health Study (NHS) and Health Professionals Follow-up Study (HPFS) who were free of a history of CVD, cancer, or major neurologic diseases. In the LVS, diet was assessed using 2 sets of 7-d diet records (7DDRs). Of note, we restricted the analyses to 400 LVS males and females with low estimated glomerular filtration rate (eGFR). This strategy was based on some critical considerations. It is known that impaired kidney function leads to significantly reduced clearance of urate, often resulting in hyperuricemia. Plasma urate metabolism equilibrium in the human body depends on the balance between urate production from the diet or endogenous metabolism and urate excretion through the kidney primarily (and intestine to a much less extent) [[14], [15], [16]]. As such, it is more feasible and likely to observe a stronger dietary impact on plasma urate concentrations among individuals with lower kidney function than in the general population [[14], [15], [16]], because urate accumulation due to diet will manifest at lower kidney function but otherwise be resolved through an efficient urinary excretion. In a sensitivity analysis, we nevertheless included all participants in the development of EDINU.
• 2)The replication of EDINU in the Mind Body Study (MBS) and NHANES: the MBS is a substudy among 233 NHSII participants, aiming to investigate the interplay between mental and physical health, with a focus on assessing lifestyle and dietary factors [17]. A total of 68 MBS females with low eGFR were included, who had existing dietary data assessed using validated semiquantitative food frequency questionnaires (SFFQs) [18] and plasma urate concentrations. The NHANES is a nationally representative survey designed to evaluate the health and nutritional status of United States population. NHANES participants aged >20 y with 24-h diet recall data and plasma urate data from the latest cycle (2017–March 2020) were considered. We focused on males and females with low eGFR and further excluded those without reporting any of the food items in the EDINU to improve data quality.
• 3)Prospective cohort analyses in the NHS and HPFS: the NHS cohort was launched in 1976 with 121,700 female United States registered nurses, aged 30–55 y, who completed a questionnaire regarding their demographics, medical history, and lifestyles. The HPFS was initiated in 1986, involving 51,529 male health professionals, aged 40–75 y, also completed a similar baseline questionnaire. In both cohorts, follow-up questionnaires were collected biennially to update participant information and identify new cases of gout, CVD, T2D, and other chronic diseases.
• 4)Replication of associations between EDINU and gout risk in the UK Biobank: the UK Biobank is a prospective cohort study that enrolled ∼500,000 participants, aged between 40 and 69 y, from 2006 to 2010 at 22 centers across the United Kingdom. Between 2009 and 2012, nearly 42% of the participants repeatedly completed 24-h dietary recalls. The population selection was described in Supplemental Method 1.

FIGURE 1.

Overview of study design. (A) Development of the Empirical Dietary Index for Normo-Uricemia (EDINU): analyses were conducted among 400 Lifestyle Validation Study (LVS) participants with low estimated glomerular filtration rate (eGFR) who had 2 sets of 7-d diet records and plasma urate data. The EDINU was developed using the least absolute shrinkage and selection operator (LASSO) model within a leave-one-out cross-validation framework among 58 predefined food groups. (B) Replication of the EDINU: analyses were conducted to replicate correlations between EDINU and urate concentrations. Replication 1 was conducted in the Mind Body Study (MBS); analyses were done among 65 participants with low eGFR who completed a semiquantitative food frequency questionnaires (SFFQs) and had plasma urate data. Replication 2 was from the NHANES, including 1951 participants with low eGFR who had 24-h dietary recall data, and serum urate measurements. (C) Application in cohort study: the EDINU was calculated using both parametric and nonparametric methods in the overall Nurses’ Health Study and Health Professionals Follow-up Study cohorts. The EDINU scores were cumulatively averaged based on every 4-y food frequency questionnaire. Prospective cohort analyses were conducted to evaluate the association between EDINU and the risk of gout and related cardiometabolic diseases including hypertension, type 2 diabetes, coronary artery disease, and stroke over >30 y of follow-up. (D) Replication of the EDINU and gout: participants were enrolled from the UK Biobank who are free of gout and have 24-h dietary recalls. Prospective cohort analyses were conducted to replicate the association between EDINU and the risk of gout.

The NHS, NHS II, and HPFS were approved by the institutional review boards of the Brigham and Women’s Hospital and the Harvard T.H. Chan School of Public Health. NHANES has approval from the Institutional Review Board of the National Center for Health Statistics. The UK Biobank study was approved by the North West Multi-Center Research Ethics Committee. All participants provided written informed consent.

Measurements of plasma urate

In LVS and MBS, plasma metabolomics profiling was conducted using high-throughput liquid chromatography-mass spectrometry techniques at the Broad Institute of MIT and Harvard [19,20]. Hydrophilic interaction liquid chromatography with positive ionization mode detection was used to measure urate and creatinine. These analyses were performed using an LC-MS system comprising a Shimadzu Nexera X2 U-HPLC (Shimadzu Corp.) coupled to a Q Exactive mass spectrometer (Thermo Fisher Scientific). The metabolites demonstrated acceptable assay reproducibility with coefficients of variation (CVs) <20% [19]. The levels of metabolites were log-transformed and standardized to z-scores. Missing data were imputed using the half-minimum method.

In the NHANES, serum urate concentrations were measured using the Beckman Unicel DxC 800 Synchron Clinical System. The process involves the oxidation of uric acid by uricase, producing peroxide, which reacts with peroxidase and 4-aminophenazone to form a colored product measured at 546 nm. Serum creatinine concentrations were measured using the Roche Cobas 6000 Chemistry Analyzer. The method involves enzymatic reactions where creatinine is converted to creatine, then to sarcosine, and finally to a detectable colored product. For both urate and creatinine, the intra-assay CVs < 3% based on replicate measurements within the same day [21]. These analyses were conducted by the Advanced Research and Diagnostic Laboratory at the University of Minnesota.

Calculations of eGFR

The Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) Creatinine Equation (2021) was used to estimate the eGFR. In the LVS and MBS, because the unitless creatinine concentrations were quantified using the metabolomics profiling, we first converted the creatinine z-scores into all positive values by adding the absolute value of the lowest z-score to each score, and then applied to this formula. Of note, creatinine concentrations measured via metabolomics showed strong correlations with plasma creatinine concentrations and eGFR [22]. In NHANES, the eGFR was calculated by using serum creatinine concentrations. In these 3 populations, we estimated population-specific tertile cutoff points, respectively, and individuals in the lowest tertile were considered as having a lower eGFR. We also applied the Modification of Diet in Renal Disease formula to calculate eGFR as a sensitivity analysis and found that the selected individuals were largely identical.

Of note, any measurement errors associated with metabolomics measurements of creatinine and the calculation of eGFR are unrelated to the 7DDR assessments and thus non-differential in nature. We observed a strong correlation of 0.73 between creatinine concentrations measured via metabolomics compared with those measured using the enzymatic method in a small subset of 100 HPFS participants. Furthermore, previous studies have reported that creatinine measured using the metabolomics platform showed a strong inverse correlation with eGFR calculated based on serum creatinine (r = –0.81) [22].

Assessments of diet and covariates

In the LVS, diet was assessed using 2 sets of 7DDRs that were administered 6 months apart and included all foods and beverages consumed during the 2 wk, with portion sizes, meal times, preparation methods, and other details [23]. Nutrient and food intake was calculated from 7DDRs based on the Nutrition Data System for Research at the University of Minnesota Nutrition Coordinating Center [23]. In the MBS, NHS, and HPFS cohorts, diets were assessed using validated SFFQs. Participants reported their habitual intake (never to ≥6 times per day) of a standard portion size of food on each SFFQ. Frequencies and portions of each individual food item were converted to average daily intake for each participant (serving/d). The reproducibility and validity of these SFFQs in measuring food intake have been previously reported in detail [24,25].

In the NHANES, diet was assessed by utilizing the 24-h dietary recall method to collect participants’ intake over the previous day (g/d). The 24-h recalls have been collected on 2 d using the USDA Automated Multiple-Pass Method [26]. In the UK Biobank, dietary assessment was conducted using the Oxford WebQ, which is a validated web-based questionnaire for 24-h dietary recalls [27].

In terms of the covariates in NHS and HPFS cohorts, the baseline and biennial follow-up questionnaires collected information on demographics, lifestyles, medical history, and medication use, including sex, age, height, body weight, physical activity, smoking status, family history of diabetes and CVD, multivitamins use, aspirin use, baseline hypertension, baseline high cholesterol, menopausal status (only in NHS), postmenopausal hormones use (only in NHS), history of kidney failure (only in HPFS), and use of diuretics. Physical activity was measured by using a metabolic equivalent of task (MET). Baseline BMI was calculated by dividing weight in kilograms by height in meters squared. Total energy intake and alcohol consumption were calculated from the cumulative average of dietary assessments conducted every 4 y through SFFQs. The covariate assessments in the UK Biobank were described in Supplemental Method 1.

Ascertainment of gout and other disease outcomes

In the NHS and HPFS, the incidence of gout, hypertension, CVD (including myocardial infarction, coronary artery bypass graft surgery, and stroke), and T2D was ascertained using various methods. Participants who self-reported a physician-diagnosed case of gout received a supplementary questionnaire based on the American College of Rheumatology survey criteria to confirm the diagnosis [28]. Incident gout was defined as meeting ≥6 of 11 of these criteria. To ensure accuracy, 2 board-certified rheumatologists reviewed the cases of 50 males from the HPFS who reported a gout diagnosis [29]. The concordance rate between the survey criteria and their medical records was 94%. In each biennial questionnaire, participants were also asked to report any new diagnoses of high blood pressure made by their physicians. This self-reporting method has been previously validated within the NHS and HPFS cohorts [30,31] The ascertainment of other disease outcomes is provided in Supplemental Method 2.

Statistical analysis

Development of EDINU

In the discovery phase, food items in 7DDRs from LVS were grouped to 58 predefined categories, such as poultry, fish and seafood, red meat, dairy-based desserts, diet beverages, juice, legumes, refined grain foods, coffee, tea, etc. (Supplemental Table 1). Participants were randomly assigned to either the training set or the testing set in a 7 to 3 fashion. We conducted least absolute shrinkage and selection operator (LASSO) regression in the training set within a leave-one-out cross-validation framework and then applied the score to the testing set [[32], [33], [34]] The EDINU was calculated as the weighted sum of the selected food groups with weights equal to the flipped coefficients from the optimal feature selection approach, with higher scores indicating lower potential for hyperuricemia:

$$\text{EDINU}=–({\beta }_1{M}_1+{\beta }_2{M}_2+{\beta }_3{M}_3+\dots +{\beta }_n{M}_n)$$

where M_n represents the intake of the ith food group (grams/d with Z-score standardization); β_n represents the coefficient associated with the ith food group.

Spearman correlation coefficients were calculated to evaluate the strength of associations between the EDINU and plasma urate concentrations. Furthermore, to enhance the generalizability to other populations and to further minimize the likelihood of overfitting, we also used a nonparametric method to calculate the EDINU. We ranked participants by quintiles and gave positive or negative scores for each food group identified by the LASSO analysis. For food items predicting lower urate concentrations, participants in the highest quintile received a score of 5, the second highest quintile a score of 4, and so forth. Conversely, for food items predicting higher urate concentrations, reversed score of 1–5 was assigned to the quintiles (Supplemental Table 2). We also applied a modified weighting approach to derive the EDINU, multiplying all coefficients by 100 to obtain integer values (Supplemental Table 2). Test–retest reliability of the dietary index was assessed using the intraclass correlation coefficient (ICC) based on 2 sets of 7DDRs.

Validation of EDINU in the MBS and NHANES

At the replication phase, the EDINU was calculated as previously described in the discovery phase in MBS and NHANES. In the MBS, blueberries, grapes, and mixed vegetables were not included in the dietary score calculation because these food items were not included in the MBS food frequency questionnaire (FFQ). Spearman correlation analyses were calculated to examine the relationships between the EDINU and plasma urate concentrations among participants with low eGFR. In addition, in the NHANES, a random error-corrected correlation coefficient was further calculated based on data of 2 separate 24-h recall interviews using the method developed by Rosner et al. [35] (Supplemental Method 3).

Cohort analyses

The baseline was set at 1984 for the NHS and 1986 for the HPFS. We excluded participants who did not return a SFFQ; those who reported an unusual total energy intake at baseline (<500 or >3500 kcal/d for the NHS, and <800 or >4200 kcal/d for the HPFS); and individuals who completed only the baseline questionnaire. For the analysis where gout was the outcome, we further excluded individuals with prevalent gout at baseline. When analyzing hypertension as the outcome, individuals with baseline hypertension were excluded. For analyses on CAD, stroke, and T2D, we excluded those diagnosed with T2D, CVD (including nonfatal myocardial infarction, fatal CAD, and fatal and nonfatal stroke), or cancer at baseline (Supplemental Figure 1).

Person-time was calculated from the date of return of baseline FFQ until the date of outcome diagnosis, death, or the end of follow-up (for gout, 2010–2012; for other disease outcomes, 2016–2018), whichever occurred first. Cox proportional hazards models with time-varying covariates were used to evaluate hazard ratios (HRs) and 95% confidence intervals (CIs) for the association between EDINU and the risk of chronic diseases. The EDINU was cumulatively averaged, but updating stopped upon the development of intermediate outcomes. To alleviate the potential reverse causality that participants with existing diseases might change their usual diet intake, we stopped updating diet once participants developed cancer or CVD during follow-up. Because the proportion of missing values of covariates was low, the missing values were replaced from the previous follow-up cycle, and otherwise missing indicators were used. The proportional hazards assumption was evaluated using a likelihood ratio test comparing the model with and without an interaction term between age and dietary indices. Analyses were stratified by age (in month), calendar year, and cohort. Multivariable models were adjusted for ancestry (White, Asian, African American or other), physical activity (<3, 3 to < 9, 9 to < 18, 18 to < 27, ≥27 METs/wk), smoking status [never, former, or current (1–14, 15–24, or ≥25 cigarettes/d)], menopausal status [premenopausal or postmenopausal (never, or postmenopausal hormone use), NHS only], baseline hypertension (yes or no), baseline high cholesterol (yes or no), total energy intake (quintiles), and baseline BMI (<21, 21–24.9, 25–29.9, 30–34.9, or ≥35). For gout, history of kidney failure (HPFS only), and diuretic use were further adjusted. For hypertension, CAD, and stroke, family history of CVD (yes or no), multivitamin use (yes or no), and aspirin use (yes or no) were further adjusted. For T2D, family history of T2D (yes or no), multivitamin use (yes or no), and aspirin use (yes or no) was further adjusted. The P values for trend were calculated by modeling the median value of each quintile as a continuous variable in the regression analysis. The dose–response relationships of EDINU with the risk of chronic diseases were assessed using restricted cubic spline regression. Subgroup analyses by sex, baseline hypertension, physical activity, smoking status, BMI, diuretic agents use, and polygenic risk score (PRS) were performed. The PRS for urate was built using 114 single-nucleotide polymorphisms (SNPs) based on a trans-ancestry genome-wide association study of plasma urate [36]. Each SNP was assigned a weight based on its relative effect size on serum urate concentrations, and the weighted values were summed to construct the PRS. The P values for interaction in all subgroup analyses were calculated using the Wald test.

In sensitivity analyses, we developed a series of models to calculate the HRs (95% CIs) for gout risk associated with the EDINU. In each model, we sequentially excluded one food group considered in the EDINU to assess the impact of each item on the overall dietary effect. Furthermore, we conducted partial least squares (PLS) analyses to recognize patterns of intake of other food items that are associated with EDINU in LVS [37], and subsequently created an expanded EDINU dietary pattern score in the NHS and HPFS. This expanded EDINU dietary pattern score incorporates food groups with factor loadings ≥|0.17| in PLS. We ranked participants by quintiles and given positive or reverse scores for each food group according to their correlations with the EDINU. The association between this expanded EDINU dietary pattern and gout risk was further examined in the NHS and HPFS cohorts, using the same approach as in the analysis of EDINU in relation to gout. We also conducted sensitivity analyses by including the Dietary Approaches to Stop Hypertension (DASH) score as an additional covariate in the multivariable-adjusted models to examine whether the observed associations were independent of DASH pattern. The cohort analyses in UK Biobank were described in Supplemental Method 1. All analyses were performed using SAS 9.4 statistical software and R version 4.1.0. Statistical tests were 2-sided, and P values of <0.05 were considered as statistically significant.

Results

The LVS included 400 participants with low eGFR (62.0% females, mean age 67.6 y) (Supplemental Table 3). LASSO analysis identified mixed vegetables, liquor, red meat, liver, artificially sweetened beverages, tomato products, wine, and salad dressing as positive predictors of plasma urate, and cheese, grapes, blueberries, and low-fat milk as negative predictors (Figure 2A). The model explained 20.1% of the deviance (%Dev), with EDINU showing Spearman correlations of –0.33 (training dataset) and –0.14 (testing dataset) with urate concentrations (Figure 2B). The MBS study included 65 females with low eGFR (mean age 60.0 y) (Supplemental Table 3), showing a Spearman correlation of –0.14 between EDINU and urate (Supplemental Figure 2). In NHANES, a random error-corrected correlation was –0.33 (95% CI: –0.24, –0.42) (Supplemental Figure 3). For the empirical index derived based on all LVS participants regardless of eGFR concentrations yielded lower correlations (–0.21 in training, –0.10 in testing) (Supplemental Figure 4) and 8.4% %Dev. The EDINU showed modest test–retest reliability, with an ICC of 0.34 (95% CI: 0.27, 0.41) based on the average of the 2 dietary assessments.

FIGURE 2.

(A) Food groups with uricemia potential: coefficients for food groups selected by the least absolute shrinkage and selection operator (LASSO) model for the plasma urate concentration among participants with low estimated glomerular filtration rate (eGFR) in the Lifestyle Validation Study (LVS). Positive coefficients are displayed on the right, whereas negative coefficients are on the left. (B) Correlation coefficients between EDINU and plasma urate concentrations, as determined by LASSO in the LVS. The EDINU positively weighted negative plasma urate predictors and vice versa for positive predictors.

Table 1 shows baseline characteristics of participants by EDINU quintiles in NHS and HPFS. Compared with the lowest quintile, those in the highest EDINU quintile were more likely nonsmokers, physically active, less obese, consumed fewer calories, and had higher Alternative Healthy Eating Index (AHEI) and DASH scores. EDINU correlated weakly with AHEI (r = 0.09) and moderately with DASH (r = 0.22).

TABLE 1.

Baseline characteristics of participants free of gout from Nurses’ Health Study (1984) and Health Professionals Follow-up Study (1986).

Variable	Quintile of EDINU
	1 (N = 23,869)	2 (N = 23,870)	3 (N = 23,869)	4 (N = 23,870)	5 (N = 23,869)
Age1	51.2 (8.3)	51.2 (8.2)	51.3 (8.3)	51.76 (8.4)	52.53 (8.7)
Ancestry
White, %	96.9	96.9	96.6	96.5	97.3
Others, %	1.8	1.0	1.0	0.9	1.0
Asian, %	0.9	1.0	1.1	1.2	0.8
African American, %	1.0	1.1	1.3	1.3	1.0
Never smoker, %	36.4	42.6	46.4	49.4	54.8
Physically active, %	51.2	50.9	50.9	50.7	53.6
BMI (kg m2)
Underweight, %	25.4	29.6	32.4	33.9	34.3
Healthy weight, %	22.0	22.3	22.7	23.4	23.5
Overweight, %	39.3	40.0	35.5	33.7	33.4
Obesity, %	13.3	11.2	9.4	9.0	8.7
Total energy intake (kcal/d)	2029.9 (618.7)	1884.6 (556.2)	1771.3 (533.9)	1668.0 (529.7)	1807.2 (571.3)
AHEI	48.2 (11.0)	49.2 (11.2)	49.4 (11.5)	50.5 (11.4)	50.8 (11.4)
DASH	22.2 (4.7)	22.3 (4.9)	22.6 (5.0)	23.5 (5.0)	25.7 (4.7)
Baseline hypertension, %	24.9	21.2	20.1	19.9	19.0
Baseline hypercholesterolemia, %	10.3	9.3	9.2	9.6	10.3
Diuretic drug use, %	14.0	11.5	10.5	10.6	9.3

Values are means (SD) or medians (Q25, Q75) for continuous variables; percentages for categorical variables, and are standardized to the age distribution of the study population. Values of polytomous variables may not sum to 100% due to rounding.

Abbreviations: AHEI, Alternative Healthy Eating Index; DASH, Dash Style Diet Score; EDINU, Empirical Dietary Index for Normo-Uricemia.

¹Value is not age adjusted.

During follow-up, 2686 gout, 58,005 hypertension, 9094 CAD, 5527 stroke, and 12,528 T2D cases were documented. Table 2 presents HRs for gout and cardiometabolic diseases by EDINU quintiles. In pooled multivariable analysis, the highest EDINU quintile was associated with a 52% lower gout risk (HR: 0.48; 95% CI: 0.42, 0.55). This association was consistent across sex, genetic susceptibility, and lifestyle factors (Supplemental Table 4, Supplemental Figure 5). Sensitivity analyses excluding individual food groups showed HRs for gout ranging from 0.45 to 0.58 (Supplemental Figure 6). Adjusting for the DASH score did not substantially change the results: the HR comparing extreme EDINU quintiles was 0.52 (95% CI: 0.45, 0.59) for gout.

TABLE 2.

Hazard ratios (95% CIs)¹ for gout and related chronic diseases according to quintiles of the Empirical Dietary Index for Normo-Uricemia (EDINU) by weights (parametric method).

Variable	Quintile of EDINU					HR (95% CI) per SD	P trend
	1	2	3	4	5
Gout2
Cases/person-years	310/517,034	447/530,485	544/541,452	656/538,878	729/510,887
Model 1 HR (95% CI)	1 (reference)	0.82 (0.74, 0.92)	0.66 (0.59, 0.74)	0.55 (0.49, 0.62)	0.39 (0.34, 0.45)	0.77 (0.74, 0.79)	<0.001
Model 2 HR (95% CI)	1 (reference)	0.91 (0.82, 1.01)	0.76 (0.68, 0.85)	0.65 (0.58, 0.74)	0.48 (0.42, 0.55)	0.82 (0.79, 0.85)	<0.001
Hypertension3
Cases/person-years	11,830/1,016,494	12,159/1,329,566	11,922/1,511,397	11,344/1,537,565	10,750/1,325,355
Model 1 HR (95% CI)	1 (reference)	0.88 (0.86, 0.90)	0.83 (0.81, 0.86)	0.80 (0.78, 0.82)	0.77 (0.75, 0.79)	0.92 (0.91, 0.92)	<0.001
Model 2 HR (95% CI)	1 (reference)	0.90 (0.88, 0.93)	0.88 (0.85, 0.90)	0.85 (0.83, 0.87)	0.84 (0.81, 0.86)	0.94 (0.93, 0.95)	<0.001
CAD3
Cases/person-years	1759/578,746	1823/617,442	1832/619,535	1851/602,778	1829/569,785
Model 1 HR (95% CI)	1 (reference)	0.94 (0.89, 0.98)	0.92 (0.88, 0.97)	0.90 (0.86, 0.94)	0.86 (0.82, 0.90)	0.95 (0.93, 0.96)	<0.001
Model 2 HR (95% CI)	1 (reference)	0.99 (0.95, 1.04)	1.02 (0.97, 1.07)	1.02 (0.97, 1.07)	1.00 (0.95, 1.05)	1.00 (0.98, 1.01)	0.83
Stroke3
Cases/person-years	1139/578,853	1115/617,627	1072/619,790	1104/603,012	1097/570,003
Model 1 HR (95% CI)	1 (reference)	0.87 (0.80, 0.94)	0.80 (0.74, 0.87)	0.79 (0.73, 0.86)	0.77 (0.71, 0.84)	0.91 (0.88, 0.93)	<0.001
Model 2 HR (95% CI)	1 (reference)	0.91 (0.84, 0.99)	0.86 (0.79, 0.93)	0.85 (0.78, 0.93)	0.85 (0.78, 0.92)	0.94 (0.91, 0.96)	0.004
T2D4
Cases/person-years	2875/530,242	2810/575,305	2550/583,840	2307/568,761	1986/537,725
Model 1 HR (95% CI)	1 (reference)	0.92 (0.87, 0.97)	0.81 (0.76, 0.85)	0.76 (0.72, 0.81)	0.69 (0.65, 0.73)	0.89 (0.87, 0.90)	<0.001
Model 2 HR (95% CI)	1 (reference)	1.02 (0.97, 1.08)	0.97 (0.92, 1.03)	0.96 (0.91, 1.01)	0.89 (0.84, 0.94)	0.97 (0.96, 0.99)	0.03

Abbreviations: CAD, coronary artery disease; CI, confidence interval; CVD, cardiovascular disease; HPFS, Health Professionals Follow-up Study; NHS, Nurses’ Health Study; T2D, type 2 diabetes.

¹Analyses were stratified by age (in month), calendar year, and cohort. Model 1 was unadjusted. Model 2 was adjusted for ancestry (White, Asian, African American or other), physical activity (<3, 3 to < 9, 9 to < 18, 18 to < 27, ≥27 METs/wk), smoking status [never, former, or current (1–14, 15–24, or ≥25 cigarettes/d)], menopausal status [premenopausal or postmenopausal (never, or postmenopausal hormone use), NHS only], baseline hypertension (yes or no), baseline high cholesterol (yes or no), total energy intake (quintiles), and baseline BMI (<21, 21–24.9, 25–29.9, 30–34.9, or ≥35).

²For gout, history of kidney failure (HPFS only), and diuretic use were further adjusted.

³For hypertension, CAD or stroke, family history of CVD (yes or no), multivitamin use (yes or no), and aspirin use (yes or no) were further adjusted.

⁴For T2D, family history of T2D (yes or no), multivitamin use (yes or no), and aspirin use (yes or no) were further adjusted.

Participants in the highest EDINU quintile, compared with the lowest, had a 16% lower hypertension risk (HR: 0.84; 95% CI: 0.81, 0.86), a 15% lower stroke risk (HR: 0.85; 95% CI: 0.78, 0.92), particularly for ischemic stroke (Supplemental Table 5), and an 11% lower T2D risk (HR: 0.89; 95% CI: 0.84, 0.94), but the association with CAD was null (HR: 1.00; 95% CI: 0.95, 1.05) (Table 2). These findings were consistent when we applied modified weights or a nonparametric EDINU that emphasizes the same food items (Supplemental Table 6).

In the UK Biobank, participants in the highest EDINU quintile consumed fewer calories, were more likely nonsmokers, had lower hypertension rates, and had better socioeconomic conditions (Supplemental Table 7). With 2599 gout cases identified over 2,393,292 person-years, the highest EDINU quintile was associated with a 25% lower gout risk (HR: 0.75; 95% CI: 0.66, 0.86) (Table 3), with a random error-corrected HR of 0.65 (0.48, 0.88).

TABLE 3.

Hazard ratios (HRs; 95% CIs) for gout according to quintiles of the Empirical Dietary Index for Normo-Uricemia (EDINU) in the UK Biobank.

Variable	Quintile of EDINU					P trend
	1	2	3	4	5
Gout
Cases/person-years	747/470,510	556/479,097	483/482,049	399/480,672	414/480,964
Model 1 HR (95% CI)1	1 (reference)	0.83 (0.75, 0.93)	0.76 (0.68, 0.85)	0.65 (0.57, 0.73)	0.65 (0.58, 0.74)	<0.001
Model 2 HR (95% CI)2	1 (reference)	0.88 (0.78, 1.00)	0.84 (0.73, 0.95)	0.75 (0.65, 0.86)	0.75 (0.66, 0.86)	<0.001

Abbreviation: CI, confidence interval.

¹Model 1 was unadjusted.

²Model 2 was adjusted for sex, age, ancestry (White, Asian or Asian British, Black or Black British, and others), total energy intake (quintiles), Townsend Deprivation Index (quintiles), physical activity (low, moderate, high defined by International Physical Activity Questionnaire activity group), smoking status (current, never, previous), baseline BMI (<18.5, 18.5–24.9, 25–29.9, or ≥30), baseline hypertension (yes or no), baseline high cholesterol (yes or no), and history of kidney failure (yes or no).

Figure 3 shows dose–response relationships between EDINU and disease outcomes. Associations with gout, hypertension, and T2D were nonlinear (all P for test of nonlinearity <0.05), whereas stroke showed a linear association (P for linear association <0.001).

FIGURE 3.

Dose–response relationship of the Empirical Dietary Index for Normo-Uricemia (EDINU) with risk of gout (A) and related cardiometabolic diseases (B–C).

Analyses were stratified by age (in months), calendar year, and cohort. Models were adjusted for ancestry (White, Asian, African American or other), physical activity (<3, 3 to <9, 9 to <18, 18 to <27, ≥27 METs/wk), smoking status [never, former, or current (1–14, 15–24, or ≥25 cigarettes/d)], menopausal status [premenopausal or postmenopausal (never, or postmenopausal hormone use), Nurses’ Health Study (NHS) only], baseline hypertension (yes or no), baseline high cholesterol (yes or no), total energy intake (quintiles), and baseline BMI (<21, 21–24.9, 25–29.9, 30–34.9, or ≥35). For gout, history of kidney failure (HPFS only), and diuretic use were further adjusted. For hypertension, family history of CVD (yes or no), multivitamin use (yes or no), and aspirin use (yes or no) were further adjusted. For T2D, family history of T2D (yes or no), multivitamin use (yes or no), and aspirin use (yes or no) were further adjusted. Solid curves indicate HRs, and dashed curves depict 95% CIs. CI, confidence interval; CVD, cardiovascular disease; HPFS, Health Professionals Follow-up Study; HR, hazard ratio; T2D, type 2 diabetes.

Sensitivity analyses using PLS analyses to recognize patterns of intake of other food items that are associated with EDINU identified other fruits (banana and cantaloupe), whole grains, cold breakfast cereal, juice, umbellifers (carrots and celery), apple, orange juice, soy products, cruciferous vegetables, condiments, and citrus as positive correlates of the EDINU; or processed red meat, pizza, poultry, and beer as negative correlates (Supplemental Figure 7). We further considered these food items and derived the expanded EDINU dietary pattern (Supplemental Table 8). This expanded EDINU dietary pattern, incorporating these foods, was associated with a 53% lower gout risk (HR: 0.47; 95% CI: 0.41, 0.54) (Supplemental Table 9).

Discussion

In this study, we developed an empirical dietary index predicting lower plasma urate, which identified low-fat milk, blueberries, grapes, and cheese as negative predictors of urate concentrations, and mixed vegetables, liquor, red meat, liver, artificially sweetened beverages, tomato products, wine, and salad dressing as positive predictors. The validity of EDINU was further verified in MBS females and NHANES population. In males and females of the United States, we found substantially lower gout risk among participants who better adhered to such a diet, and this inverse association was replicated in an independent United Kingdom population. In addition, this index was associated with a lower risk of related cardiometabolic diseases, especially hypertension, stroke, and T2D. These associations persisted by sex, genetic predisposition, and lifestyle and were robust to multivariable adjustments and in multiple sensitivity analyses.

To our knowledge, our study is among the first efforts that systematically identified dietary factors that predict plasma urate concentrations and constructed a dietary index accordingly. Prior research has identified some specific dietary components that may influence the risk of gout. For example, high-purine foods, such as red meats, organ meats, and certain seafood, have been consistently associated with an increased risk of gout [29]. Conversely, dairy protein and vitamin C intake may lower the incidence of gout [38,39]. In addition, several studies have highlighted the protective effects of healthy dietary patterns. For instance, among females, adherence to the DASH diet was associated with a 32% lower risk, whereas the Mediterranean diet may reduce the risk by 12% [5]. Similarly, adherence to the AHEI, the Prudent diet (characterized by higher intake of fruits, vegetables, whole grains, poultry, and low-fat dairy), or plant-based diets may also lower gout risk [5,40]. In contrast, an adherence to a typical Western diet or unhealthful plant-based diet was linked to a higher risk of gout [5,38]. The 2020 American College of Rheumatology Guideline for the Management of Gout emphasizes managing lifestyle factors, such as adopting a low-purine diet to mitigate risk and limiting excessive intake of alcohol and high-fructose corn syrup [7,41]. Although these prior findings collectively underscored the importance of diet quality in gout prevention, the dietary patterns identified so far did not specifically target plasma urate concentrations or gout risk.

The EDINU developed and replicated in this study consists of some food items that are known to potentially modulate plasma urate concentrations. For example, dairy products, especially low-fat dairy intake, are known for their urate-lowering potentials [42]. Blueberries and grapes are rich in polyphenols and vitamin C [43], which may reduce urate concentrations by regulating urate excretion transporters [44]. A clinical trial has shown decreased plasma urate concentrations by blueberry intake [45]. In contrast, red meat, liver, and alcohol have been recognized for their urate-increasing effects [29]. There are also some food items included in the EDINU for which there is no or less direct evidence previously suggesting their link with urate concentrations, including mixed vegetables, salad dressings, and certain tomato-based products. The mixed vegetables is a combination of types of starchy vegetables, typically including peas, carrots, corn, and green beans. Green peas and beans are known purine-rich vegetables [29] and starchy vegetables may not confer the same health benefits as nonstarchy vegetables [46]. Prior studies have found a positive association between tomato consumption and urate concentrations, although whether this relationship is causal remains to be further elucidated [47]. Moreover, dressings that contain high-fructose corn syrup or added sugar may potentially lead to elevated urate concentrations or gout incidence [48,49]. Further research is needed to fully understand the biological underpinnings that explain the overall EDINU and its components in relation to urate metabolism and gout risk.

We observed a stronger association between the EDINU and the risk of gout than with other dietary patterns reported in previous studies [5,38,40]. Moreover, this association between the EDINU and gout was replicated in the UK Biobank, which assessed diet using very different instruments (a limited number of 24-h recalls), strongly supporting the robustness of the associations of interest. The associations are also persistent in subgroups defined by sex, genetic susceptibility of uricemia, and lifestyle factors. These observations may help rule out the role of chance in our findings. Apart from gout, the association between EDINU and cardiometabolic diseases, specifically for hypertension, ischemic stroke, and T2D was observed as well, which indicates that dietary patterns influencing plasma urate concentrations may have a broader public health impact beyond gout. In addition, these observations also demonstrated the “safety” or “healthfulness” of consuming such a diet.

The strengths of this study included the use of 7DDR data to characterize diet, prospective cohort design, standardized ascertainment of gout cases, large sample size, long-term follow-up, standardized diagnosis, and control of multiple confounders. We adopted the agnostic approach to developing the EDINU, the effectiveness of which has been demonstrated in previous studies. Furthermore, the EDINU was validated and replicated using both internal and external datasets, including MBS, NHANES, and UK Biobank, ensuring the robustness and generalizability of the findings across varied populations. Nevertheless, the limitations of this study shall be acknowledged. First, measurement errors in dietary assessment might arise from the use of FFQs in large cohorts, yet employing cumulatively averaged method mitigates random errors when assessing long-term dietary patterns [50]. Second, although adjustments were made for potential confounders, the possibility of residual or unmeasured confounding cannot be entirely ruled out in observational studies, although the consistent associations across United States and United Kingdom populations may make it less a concern. Last, the background dietary context poses another limitation. For instance, in populations with very low dairy consumption, the relevance of our EDINU is lower.

An empirical index modestly predicting lower plasma urate was associated with a substantially lower risk of gout in United States individuals. Consuming such a diet with lower uricemia potentials could be a promising approach to preventing gout. Clinical trials and follow-up observational research are warranted to further substantiate the findings from the current study.

Consent to participate

Return of the completed questionnaire was considered to imply consent, as approved by the institutional review boards.

Consent to publish

We confirm that consent to publish has been received from all participants.

Author contributions

The authors’ responsibilities were as follows – QS, XW: conceived and designed the study; XW: performed the statistical analysis and data interpretation, and drafted the manuscript; WZ, MW, BL, YH, SW, HH, YH, HKC: conducted the technique review and edited the manuscript; QS, XW, SKR: responsible for the later revision of the code and manuscript; and XW, QS: guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Ethics approval

The Nurses’ Health Study, Nurses’ Health Study II, and Health Professionals Follow-up Study were approved by the institutional review boards of the Brigham and Women’s Hospital and the Harvard T.H. Chan School of Public Health. NHANES has approval from the Institutional Review Board of the National Center for Health Statistics. The UK Biobank study was approved by the North West Multi-Center Research Ethics Committee. All participants provided written informed consent.

Data availability

Data described in the manuscript, codebook, and analytic code will be made available on request.

Funding

This study was funded by the NIH (grant nos. UM1 CA186107, U01 CA176726, U01 CA167552, P01 CA87969, R01 HL034594, R01 HL035464, R01 HL60712, R01 DK120870, R01 DK126698, R01 DK119268, U2C DK129670, DK119268, R01 ES022981, R01 ES036206, and R21 AG070375). SKR was supported by a Postdoctoral Fellowship Award from the Canadian Institutes of Health Research.

The funders had no role in the study design; in the collection, analysis, and interpretation of data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication. All authors confirm the independence of researchers from funders.

Conflict of interest

The authors declare no conflict of interest.

Appendix A. Supplementary data

The following is the Supplementary data to this article: