Association of Serum Uric Acid With All-Cause and Cardiovascular Mortality in Diabetes

The prevalence of diabetes is escalating worldwide. There were 537 million people with diabetes globally in 2021, and the number of patients is estimated to reach 783 million by 2045 (1). Cardiovascular disease (CVD) represents the most common cause of death among individuals with diabetes (2). Exploring modifiable risk factors for diabetes favors avoiding premature death and promoting long-term health benefits.

Uric acid is the degradation product of purines. Increased levels of serum uric acid (SUA) is widely recognized as a risk factor for macrovascular and microvascular complications. As a group at high risk of CVD, hyperuricemia is highly prevalent in patients with diabetes (3). However, the optimal levels of SUA in patients with diabetes remain unknown. Previous cohort studies among patients with diabetes showed conflicting findings regarding the association of SUA levels with all-cause and CVD-caused mortality (hereafter, CVD mortality) (4–8). Of note, hyperuricemia is always accompanied by other cardiovascular health risk factors, such as dyslipidemia, obesity, kidney disease, and poor regulation of glucose status. To some extent, cohort studies were partially subjected to limited sample size, short follow-up, and the absence of crucial confounders, which may also alter the relationships between SUA levels and adverse outcomes.

To fill the knowledge gaps, we assessed the associations of SUA levels with all-cause and CVD mortality among U.S. adults with diabetes in a large, nationally representative sample. We also performed a meta-analysis to provide additional evidence for quantifying the associations between SUA levels and mortality.

The National Health and Nutrition Examination Survey (NHANES), conducted by the National Center for Health Statistics (NCHS) of the U.S. Centers for Disease Control and Prevention, is a large-scale, multistage, ongoing, and nationally representative health survey of noninstitutionalized U.S. civilians older than 2 months (9). The survey combines interviews and medical examinations to collect demographic, socioeconomic, dietary, physiological, and laboratory information. NHANES has been approved by the NCHS Ethics Review Board (Protocols 98–12, 2005-06, 2011-17, and 2018-01), and all participants signed informed consent.

Data from 10 cycles of the NHANES survey (1999–2000, 2001–2002, 2003–2004, 2005–2006, 2007–2008, 2009–2010, 2011–2012, 2013–2014, 2015–2016, and 2017–2018) and the National Death Index (NDI) mortality data were combined to construct this cohort study. The participants with diabetes at baseline were defined as having been diagnosed by physicians or as having a glycated hemoglobin A_1c (HbA_1c) value ≥6.5% or a fasting plasma glucose level ≥126 mg/dL (10). The patients with diabetes who had at least one of the following conditions were excluded: 1) individuals aged <20 years; 2) a pregnancy at baseline; 3) having cancer at baseline; 4) missing information on SUA levels, estimated glomerular filtration rate (eGFR) and HbA_1c values; and 5) for whom information on death status was unavailable. Finally, 7,101 patients with diabetes from NHANES 1999–2018 were included (Supplementary Fig. 1).

The SUA measurement was performed with a Hitachi Model 704 multichannel analyzer (Boehringer Mannheim Diagnostics, Indianapolis, IN) in the NHANES 1999–2000 (11), a Beckman Synchron LX20 in the NHANES 2001–2007 (12), a Beckman UniCel DxC800 Synchron in the NHANES 2008–2016 (13), and a Roche Cobas 6000 analyzer in the NHANES 2017–2018 (14). The combination of SUA from multiple cycles of the NHANES for analysis has been reported in previous studies (15,16).

We used the NHANES Public-Use Linked Mortality File through 31 December 2019 (https://www.cdc.gov/nchs/data-linkage/mortality-public.htm), which linked to the NDI data with a probabilistic matching algorithm to determine mortality status. NDI is an NCHS-centralized database (17) of all deaths in the U.S. The cause-specific mortality data in the NDI have been shown to accurately classify deaths, with only a relatively slight possibility of misclassification. The underlying cause of death was identified according to the International Statistical Classification of Diseases, 10th Revision, and CVD mortality was defined as death due to heart diseases (codes I00–I09, I11, I13, I20–I51) and stroke (codes I60–I69) (18). Follow-up time for each person was calculated as the difference between the baseline examination date and the last known date alive or censored from the mortality file.

Information on age, sex, race or ethnicity, education level, family income, smoking status, alcoholic intake, and physical activity was collected using standardized interview questionnaires. Race and ethnicity were classified as non-Hispanic White, non-Hispanic Black, Mexican American or Hispanic, or other. Education level was stratified as less than high school, high school, and college or above. Family income to poverty ratios were categorized as ≤1.30, 1.31 to 3.50, and >3.50; a higher income to poverty ratio represents a better family economic status (19). Participants were classified as nonsmokers, former smokers, and current smokers of cigarettes according to their responses about smoking at least 100 cigarettes during their lifetime and whether they were currently smoking. The amount of alcohol consumption was determined on the basis of the 24-h dietary recall of participants. Current alcohol intake was categorized as none (0 g/day), moderate drinking (0.1–27.9 g/day for men and 0.1–13.9 g/day for women), and heavy drinking (≥28 g/day for men and ≥14 g/d for women).

For physical activity, the inactive group was defined as those with no reported leisure-time physical activity, and the active group was defined as those who had recommended levels of physical activity (i.e., self-reported leisure-time moderate activity [metabolic equivalents ranging from 3 to 6] of 5 or more times per week or leisure-time vigorous activity [metabolic equivalents >6] of 3 or more times per week), and the insufficiently active group was defined as those who were not inactive and did not meet the criteria for recommended levels of physical activity (20). Dietary information was collected through 24-h dietary recall interviews. Total energy intake was calculated using the U.S. Department of Agriculture Automated Multiple-Pass Method (21). We used the Healthy Eating Index-2015 (HEI-2015) to indicate the overall diet quality, with higher scores indicating a better-quality diet (22). Anthropometric and blood biochemical indices were measured following the relevant standardized protocols. BMI was calculated as weight in kilograms divided by height in meters squared and categorized as <25.0, 25.0–29.9, and ≥30.0 kg/m².

Hypertension was defined as taking antihypertensive drugs at baseline, or systolic blood pressure ≥130 mmHg and/or diastolic blood pressure ≥80 mmHg, according to the 2017 American College of Cardiology and American Heart Association hypertension guidelines (23). The HOMA of insulin resistance (HOMA-IR) was calculated according to a previous report (24). Dyslipidemia was defined as having been diagnosed by a physician, or currently taking hypolipidemic medication, or having a triglyceride level ≥150 mg/dL or HDL cholesterol level <40 mg/dL, based on recommendations of the National Cholesterol Education Program Adult Treatment Panel III (25). The eGFR was calculated according to the Chronic Kidney Disease Epidemiology Collaboration equation (26). Urinary albumin was assessed as the albumin to creatinine ratio, using a solid-phase fluorescent immunoassay and modified Jaffe kinetic method with a single, spot sample. Albuminuria was defined as a urinary albumin to creatinine ratio of >30 mg/g (26). Baseline CVD was defined as having been diagnosed with congestive heart failure or stroke by a physician. Information on diabetes duration (years) and medication use (i.e., allopurinol, insulin, and oral hypoglycemic drugs) was obtained through interviews during medical visits.

We used survey modules of SAS software, version 9.4 (SAS Institute, Cary, NC) for statistical analyses, including surveymeans, surveyfreq, surveyreg, and surveyphreg. All statistical analyses accounted for the complex, multistage, stratified, and cluster-sampling design (including oversampling of certain subpopulations) of the NHANES to reflect the nationally representative estimates using sample weights, strata, and primary sampling units embedded in the NHANES data. Baseline characteristics were presented as mean (SE) and percentages (SE). The participants were stratified into quintiles 1–5 (Q1–Q5) according to SUA levels. The linear regression for continuous variables and χ² test for categorical variables were used to detect the difference across the Q1–Q5 groups. The associations of cardiometabolic markers with SUA levels were assessed by generalized least squares estimation, after adjusting for age, sex, race or ethnicity, education level, family income level, smoking status, alcohol intake, physical activity, and HEI-2015.

The associations of SUA level with all-cause and CVD mortality were evaluated by weighted Cox proportional hazards regression models with the following covariates: age, sex, and race or ethnicity (model 1); model 1 plus education level, family income level, smoking status, alcohol intake, physical activity, total energy intake, and HEI-2015 (model 2); model 2 plus BMI, hypertension, dyslipidemia, diabetes duration, history of CVD, HbA_1c, drug use (i.e., allopurinol, hypoglycemic medications, and insulin), eGFR, and albuminuria (model 3). Models were evaluated to determine there was no violation of the proportional hazard’s assumption for all analyses.

Stratified analyses were performed in the strata of age, sex, race or ethnicity (White and non-White), education level (high school or less, or college or above), family income level, smoking status (current smoker or nonsmoker), alcohol intake (drinking or nondrinking), physical activity, BMI (<30, ≥30 kg/m²), and eGFR levels (<60, 60–89, ≥90 mL/min/1.73 m²). The P values for the interaction of SUA levels and stratified variables were also calculated. We further applied the restricted cubic spline (RCS) model without weights to visualize the dose-response association of SUA levels with all-cause and CVD mortality, because there was no available RCS model for complex, multistage sampling survey data. In addition, several sensitivity analyses were conducted to examine the robustness of the main results. First, to minimize the possibility of reverse causality, we excluded patients who died within 2 years of baseline examination. Second, to clarify the possible effect of medicine use on mortality outcomes, we evaluated the associations after excluding the patients with diabetes drug use (i.e., allopurinol, hypoglycemic medications, or insulin). Third, we examined the relationship between hyperuricemia and mortality outcomes using the two definitions of hyperuricemia based on sex (>7.0 mg/dL for men, >5.7 mg/dL for women) and supersaturation point (>7.0 mg/dL) (27). In consideration of the distribution difference in SUA levels between the sexes, we also ran RCS models in different sexes to present dose-response association with mortality outcomes. Two-sided P values <0.05 were considered statistically significant.

We performed a systematic literature search of the Embase and PubMed databases for published cohort studies until 31 December 2021. The search strategies were as follows: (mortality OR cardiovascular mortality) AND (hyperuricemia OR uric acid) AND (diabetes OR type 2 diabetes OR type 2 diabetes mellitus). We included articles if they met the following criteria: 1) cohort design; 2) targeted patients with diabetes or a stratified analysis was performed among patients with diabetes; 3) baseline SUA levels were measured; 4) results were all-cause mortality or CVD mortality; 5) HRs and 95% CIs were provided; and 6) the confounders were adjusted between SUA level and mortality. Data extraction were performed by independent researchers (B.L. and T.T.), and included first author’s name, publication year, sample size, follow-up year, main outcomes, number of cases, HRs and 95% CIs, and confounders. The Newcastle-Ottawa Scale was used to evaluate the quality of identified eligible literature.

SUA levels measured in micromoles per liter were converted to milligrams per deciliter according to the equation 1 mg/dL = 59.48 μmol/L. To examine the association of SUA levels with all-cause and CVD mortality, the HRs (95%CIs) of 1 mg/dL increments in SUA level for each cohort were calculated using the methods of Greenland and Longnecker (28). The median of each SUA level category was calculated as the means of lower and upper bounds. For the open-ended category, the lower bound of the lowest category and upper bound of the highest category were assumed at the same width as the adjacent interval. Finally, the HRs of 1 mg/dL increments for all-cause and CVD mortality were determined by pooling the available results from these cohort studies.

We used I² statistics to evaluate the heterogeneity across studies; a fixed-effect model was used to combine the effect sizes when I² < 50%, otherwise, a random-effect model was adopted. Publication bias was assessed by the funnel plot and Egger test. Sensitivity analysis was performed to examine the role of the single study in the pooled results. In addition, the dose-response meta-analysis required the median, HRs and 95% CIs, number of cases, and sample size per person-year for each SUA level category. If the study did not use the lowest SUA level category as the reference, the Hamling method (29) was used to convert the HRs of other SUA level categories by using the lowest SUA level as the reference. After excluding studies without the aforementioned information or in which there were fewer than 50 cases of death, dose-response meta-analysis was used to evaluate the association between SUA level and mortality. The meta-analysis was conducted using Stata software (version 15.1).

The baseline demographic, lifestyle, and medical characteristics of the cohort are listed in Table 1. This study included 7,101 patients with diabetes (mean age, 57.84 years; 51.7% men) from the NHANES 1999–2018. Participants with higher SUA levels were more likely to be older, male, inactive, obese, and have comorbidities. The least-squares means of cardiometabolic indexes by the Q1–Q5 of SUA levels are presented in Table 2, after adjusting for age, sex, race or ethnicity, education level, income, family income level, smoking status, alcohol intake, physical activity, and HEI-2015. SUA levels were positively related to total cholesterol, triglycerides, insulin levels, and HOMA of insulin resistance, and negatively related to fasting glucose, HbA_1c, HDL cholesterol, and eGFR (for all: P for trend < 0.05).

During 57,926 person-years of follow-up (median, 7.3 years) from the NHANES 1999–2018, we documented 1,900 deaths (n = 674 deaths from CVD). In the fully adjusted model, the multivariable-adjusted HRs for individuals in the highest quintile were 1.28 (95% CI 1.03, 1.58) for all-cause mortality and 1.41 (95% CI 1.03, 1.94) for CVD mortality, as compared with the lowest quintile (Table 3). For a one-unit increment in SUA level, the HRs of all-cause and CVD mortality were 1.07 (95% CI 1.02 1.12) and 1.10 (95% CI 1.02, 1.18), respectively.

The stratified analyses revealed significant interactions between age groups, physical activity groups for all-cause mortality, and BMI categories for CVD mortality, and did not show the interaction between other variables for mortality (Supplementary Tables 1 and 2). The sensitivity analyses excluding patients who died within 2 years of baseline examination (Supplementary Table 3) and patients with diabetes medication use (specifically, allopurinol or insulin therapy) (Supplementary Table 4) yielded similar results for both all-cause mortality and CVD mortality. However, we found no significant association for all-cause and CVD mortality among the patients who did not use oral hypoglycemic drugs (Supplementary Table 4). The significant associations of hyperuricemia with all-cause and CVD mortality were also observed (Supplementary Table 5). The RCS in the total study population (Supplementary Fig. 2) and different sexes (Supplementary Fig. 3) showed a gentle slope of HR when SUA level was ≤7.0 mg/dL, but an increase in HR when SUA level was >7.0 mg/dL, indicating the higher risks among patients with SUA levels >7.0 mg/dL.

As shown in the flowchart of study search and selection (Supplementary Fig. 4), 13 studies were included in this meta-analysis. The baseline characteristics of each cohort are listed in Supplementary Table 6. The total sample size was 51,368. From the sample, 5,242 people died, with 1,682 deaths from CVD, and the mean or median follow-up time ranged from 3.1 to 19.0 years. The included studies were of relatively good quality and the Newcastle-Ottawa Scale scores ranged from 5 to 9.

We included 13 and 8 eligible cohort studies for pooling the HRs of all-cause mortality and CVD mortality, respectively. The pooled HRs per 1 mg/dL increment in SUA level for all-cause and CVD mortality were 1.08 (95% CI 1.05 to 1.11) and 1.05 (95% CI 1.03 to 1.06), respectively (Fig. 1). The subgroup analysis based on the study population with diabetes was performed and showed similar results (Supplementary Fig. 5). Although the funnel plot and Egger test indicated significant publication bias (Supplementary Figs. 6 and 7), the sensitivity analyses showed no potential influence of each study on the pooled effect size (Supplementary Fig. 8). The dose-response meta-analysis of four available cohort studies showed a nonlinear relationship between SUA level and all-cause mortality (Supplementary Fig. 9).

In this large-scale and nationally representative study, we observed consistent findings that high SUA levels were associated with higher risks of all-cause and CVD mortality in patients with diabetes, after adjusting the demographic characteristics, dietary and lifestyle factors, cardiovascular risk factors, diabetes-related parameters, drugs use, baseline diseases, and kidney function. Subsequently, meta-analysis of available cohort studies showed that per 1 mg/dL increment in SUA level, the risks of all-cause and CVD mortality increased by 8% and 5%, respectively, among patients with diabetes.

Among the general population, prospective cohort studies have indicated the U-shaped relationship between SUA levels and mortality outcomes (30,31), hinting at the adverse effect of low and high SUA levels. However, the appropriate target for SUA levels among patients with diabetes is inconclusive. Previous studies as well as our research showed that high SUA level was associated with the augmented risks of all-cause and CVD mortality in patients with type 2 diabetes. One retrospective cohort study of 535 patients with diabetes found that the 1 mg/dL increment in SUA level was associated with 21% higher risk of all-cause mortality (4), though some important confounders, such as use of hypoglycemic drugs and duration of diabetes, were not adjusted. Another cohort study showed that the significant association between SUA concentration and CVD mortality only existed in the prediabetes group and patients with type 2 diabetes, not in residents with normal glucose-tolerance status (5). On the contrary, one cohort study of 1,294 patients with type 2 diabetes reported that SUA level was not an independent predictor of CVD or all-cause mortality (8). Notably, the aforementioned studies among patients with type 2 diabetes did not show the results of HRs of several segments or curves on the association of SUA level with mortality. The data from three cohorts of patients with type 2 diabetes found that both the highest and lowest tertiles of SUA levels were associated with an increased risk of all-cause mortality, as compared with patients with the intermediate tertile of SUA level (7). Another cohort study of 1,540 subjects with type 2 diabetes found that the HRs of the higher quartiles for all-cause mortality were increasing with a significant trend when compared lowest quartile of SUAlevels (6).

The discrepancy in results from these studies may be explained by differences in sample size, duration of the follow-up period, baseline characteristics, and adjusted covariates of the study population. For instance, the durations of diabetes and levels of HbA_1c of participants in these studies were different from the present study. Studies have shown that the higher level of glucose and longer duration of diabetes may alter tubular reabsorption and secretion and lead to a decrease in kidney excretion of uric acid, which may influence the concentration of SUA (32). Moreover, one cohort study found that low SUA levels were associated with a high risk of all-cause and CVD mortality in participants with malnourishment, but not in those without malnourishment, suggesting that nutritional status was important for SUA-associated mortality (30). However, the aforementioned studies did not consider dietary factors. To further reinforce the associations of uric acid with mortality, when the NHANES 1999–2018 data were combined with available cohorts for meta-analysis, the larger sample size and high quality of included studies exhibited more convincing results.

The largest Mendelian randomization study included 110,347 individuals of European ancestry from the Global Urate Genetics Consortium and 343,836 White British individuals from the UK Biobank and found that an increment of 1 SD of genetically predicted SUA levels was associated with increased risks of coronary heart disease and stroke (33). One Mendelian randomization study of 3,316 White individuals also showed that an increase in genetically predicted SUA levels was significant for cardiovascular death and sudden cardiac death (34). In contrast, another Mendelian randomization study of 18,828 Pakistani individuals did not show a causal role of SUA levels in CVDs, such as coronary heart disease, ischemic stroke, and heart failure (35). The discrepancy in results from these Mendelian randomization studies may be due to the differences in race and study outcomes. Interestingly, our stratified analyses indicated higher HRs of Q5 for all-cause and CVD mortality in White people. Additionally, it should be noted that the variants could only explain 3%–7.7% of the variance of SUA levels in large-scale studies. The power to detect the associations might be subject to the explained variance of single nucleotide polymorphisms. The Mendelian randomization study data on SUA levels and adverse outcomes among diabetes are sparse and need to be clarified. Clinical trials on the effect of lowering uric acid treatments indicated conflicting results. Allopurinol, the most common drug for lowering uric acid (16) has been proved to induce a significant improvement in flow-mediated dilation, as the index of endothelial function (36). However, epidemiological and biological studies indicated that allopurinol also can decrease oxidative stress and inhibit free radicals (37). Thus, the effect of allopurinol on cardiovascular health may be explained by the combination of lowering uric acid level and ameliorating oxidative stress (37). Febuxostat, another uric acid–lowering drug, was associated with a higher risk of CVD mortality than allopurinol (38). Thus, evidence from human studies and animal experiments is needed to confirm the health effect of reducing uric acid and underlying mechanisms.

Moreover, the complex interaction between kidney function and SUA levels may be concerning for a potential health effect. Consistent with our results, a study of another cohort of 1,540 patients with diabetes also reported no significant interaction between the SUA level and eGFR for all-cause mortality (6). On the contrary, one longitudinal study of 21,963 individuals indicated a significant interaction between SUA levels and eGFR for all-cause and CVD mortality, and the researchers found that the HRs of a 1 mg/dL increment of SUA level for CVD mortality were 1.21 (95%CI 1.08, 1.36), 1.13 (95% CI 1.07, 1.19), and 1.05 (0.99, 1.12) in participants with eGFR > 90, 60–90, and <60 mL/min/1.73 m², respectively (39). These findings suggest that due to the reduced excretion of uric acid in patients with impaired eGFR, the prognostic role of SUA levels seems to be significant only for higher levels as compared with those of patients with preserved kidney function. More studies are needed to clarify the complex interaction between uric acid and eGFR and its potential effect.

The observed relationship between SUA level and increased mortality is biologically plausible and may involve endothelial dysfunction, cardiometabolic syndromes, and inflammation. Uric acid tends to precipitate as crystals, and the urate deposition on the vessel wall reduces the content of nitric oxide and results in vascular endothelial dysfunction, which may promote the formation of atherosclerosis and thrombosis (40). In addition, the increasing SUA level could promote the oxidative modification of LDLs (41). In the present study, we also show the relationship between higher SUA levels and dyslipidemia, including higher triglyceride levels and lower HDL cholesterol levels, which are notable risk factors of cardiovascular health.

The major strength of this study includes the use of a large amounts of nationally representative data and the combination of that data with those from available published studies for a meta-analysis, which showed a consistent conclusion about a controversial association between SUA level and mortality in patients with diabetes. With the sufficient data collected in the NHANES survey, we could also consider a wide range of confounders. Lastly, a series of sensitivity analyses were performed to ensure the robustness of the results, and the association between SUA level and mortality was still identified.

This study was subject to several limitations. First, although this prospective cohort study has a large sample size with a long follow-up period, we cannot establish causality, due to the nature of the observational study. Second, the NHANES only has a single record of SUA levels, which may not reflect the average levels in the long-term follow-up. However, the single measurement of uric acid levels has been reported in many large-scale cohort studies (6,8). Third, the data from the NHANES could not distinguish different types of diabetes. However, the findings from this study are likely more representative of individuals with type 2 diabetes, because it includes patients aged 20 years or older. Fourth, the relatively small available studies for dose-response meta-analysis limited the power to present the nonlinear association between SUA levels and mortality outcomes. Finally, despite adjustment for many potential confounders, we could not control for all the residual confounding.

In summary, we found that higher SUA level was associated with increased all-cause and CVD mortality among individuals with diabetes. Clinical trials are needed to determine the potential effects of lowering levels of circulating uric acid medications on cardiovascular health.

Acknowledgments. The authors thank the participants and staff of the National Health and Nutrition Examination Survey 1999–2018 for their valuable contributions.

Funding. None.

Duality of Interest. No potential conflicts of interest relevant to this work were reported.

Author Contributions. S.R. proposed and designed the study. B.L., L.C., and J.Y. conducted the statistical analysis. B.L., T.T., and J.Y. completed the literature search and data extraction. B.L., L.C., and T.T. drafted the first version of the manuscript. X.H. and W.B participated in the critical revision of the manuscript. S.R. is the guarantor of this work and, as such, has full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. All authors reviewed and approved the final version of the manuscript.