A Mendelian Randomization Study of Circulating Uric Acid and Type 2 Diabetes

Elevated serum uric acid concentrations have been associated with higher diabetes risk in observational studies (1,2). Meta-analyses reported 6–17% higher diabetes risk with every 59.48 µmol/L (1 mg/dL) higher uric acid concentration (1,2). If this observed association were found to be causal, uric acid–lowering therapies could be used in diabetes prevention. However, whether uric acid causes diabetes is still a matter of debate (3,4). Uric acid concentrations are closely linked to other diabetes risk factors, such as obesity, which makes it difficult to determine the independent effects of uric acid when limited to observational analysis alone (3,4). Evidence from human intervention studies on the effect of uric acid–lowering therapy on glucose metabolism is very limited and inconsistent (5–7).

The concept of Mendelian randomization (i.e., using genetic variants as an instrumental variable), can be applied to test and estimate the causal effects of risk factors on disease outcomes (8). Because alleles are randomly allocated during gamete formation, the association of a genetic variant with risk of a disease outcome is unlikely to be confounded by other factors. Also, reverse causality is abrogated.

Three meta-analyses together identified 31 loci associated with uric acid (9–11). Variants at such loci can be used as genetic instruments to estimate the unconfounded effect of uric acid on diabetes risk. Only one Mendelian randomization study on uric acid and diabetes risk has been performed (12) and reported no evidence for a causal effect. That study used a small number of single nucleotide polymorphisms (SNPs) (8 identified in the first meta-analyses [9]) and used different studies to estimate the association between the genetic score and diabetes, the association between the genetic score and uric acid, and the association between uric acid and diabetes (i.e., the three sides of the Mendelian randomization triangle [13]).

In the current study, we used a multilocus Mendelian randomization approach to estimate the unconfounded effect of uric acid on diabetes risk. We performed instrumental variable estimation within the same study, using data on genetic variants in 24 uric acid–associated loci, and measured uric acid concentrations among 24,265 individuals, including 10,576 case subjects with incident type 2. We then bolstered the sample size by including summary-level data from the Diabetes Genetics Replication And Meta-analysis (DIAGRAM) consortium, bringing to 41,508 our total number of case subjects with diabetes.

European Prospective Investigation into Cancer and Nutrition (EPIC)-InterAct study is a large, prospective case-cohort study involving individuals in 26 study centers from 8 European countries (Denmark, France, Germany, Italy, the Netherlands, Spain, Sweden, and the U.K.) that is nested within EPIC (14). Most participants were aged 35 to 70 years and were recruited between 1991 and 2000, mainly from the general population. The EPIC-InterAct study, drawn from a total cohort of 340,234 individuals comprising 3.99 million person-years of follow-up, was designed to investigate the interplay between genetic and lifestyle factors and type 2 diabetes risk (15). A total of 12,403 verified incident cases of type 2 diabetes were identified. A center-stratified, random subcohort of 16,154 individuals was selected for analysis. Because of the random selection, this subcohort also included a random set of 778 individuals who had developed incident type 2 diabetes during follow-up. All participants gave written informed consent, and the study was approved by the local ethics committees and the International Agency for Research on Cancer Internal Review Board.

For the observational part of this analysis, we excluded participants with missing uric acid (n = 1,873) or covariable (n = 1,641) data, leaving 24,265 (10,576 case subjects, 14,364 subcohort participants, including 675 case subjects in the subcohort) participants for analyses. For the instrumental variable analysis, we excluded participants with missing uric acid (n = 1,875), genetic (n = 8,634; including 4,063 from Denmark, because at the time of analysis, genetic data were not yet available from the Danish cohort), BMI (n = 141), or biomarker (n = 11) data, leaving 17,118 (7,319 case subjects, 10,235 subcohort participants, including 436 case subjects in the subcohort) participants for analyses.

Ascertainment and verification of incident diabetes has been described in detail elsewhere (15). In short, case subjects with incident diabetes were identified through self-report, linkage to primary care registers, secondary care registers, medication use, and hospital admissions and mortality data. Information from any follow-up visit or external evidence with a date later than the baseline visit was used. To increase the specificity of the case definition, we sought further evidence for all case subjects with information on incident type 2 diabetes from less than two independent sources at a minimum, including individual review of medical records. Participants were followed up for occurrence of diabetes until 31 December 2007.

Nonfasting blood samples were taken at baseline. Laboratory measures were performed by the Stichting Huisartsen Laboratorium Groep (Etten-Leur, the Netherlands) on serum (except for participants in the Umeå center in Sweden, where only plasma samples were available) or erythrocyte samples that had been previously frozen in ultra–low-temperature freezers at −80°C or in liquid nitrogen. Serum uric acid, triglycerides, glucose, and HDL were measured using a Cobas enzymatic assay (Roche Diagnostics, Mannheim, Germany) on a Roche Hitachi Modular P analyzer. Erythrocyte HbA_1c was measured using Tosoh (HLC-723G8) ion exchange high-performance liquid chromatography on a Tosoh G8.

DNA was extracted from buffy coat from a citrated blood sample using standard procedures on an automated Autopure LS DNA extraction system (Qiagen, Hilden, Germany) with PUREGENE chemistry (Qiagen). In total, 8,536 (3,942 case subjects, 4,859 subcohort participants, including 265 case subjects in the subcohort) participants were genotyped with CardioMetabochip+, a customized version of the CardioMetabochip (Illumina, San Diego, CA), using a Sequenom iPLEX array (Sequenom, San Diego CA). The remaining 8,582 participants (2,941 case subjects and 5,812 subcohort participants, including 171 case subjects in the subcohort) were genotyped with the Illumina 660W quad chip (Illumina), using TaqMan (Applied Biosystems, Carlsbad, CA). Missing genotypes for participants genotyped with the Illumina 660W quad chip were imputed by assigning the mean genotype at each locus for case subjects and noncase subjects separately, for individuals successfully genotyped. Genotypes for 15 of 24 SNPs were imputed. We selected SNPs that passed the significance threshold of P < 5 × 10^–8 in three large-scale genome-wide association study (GWAS) meta-analyses of uric acid (9–11) that were identified from searching PubMed with key words “GWAS” and “uric acid” or “urate.” No SNPs were in linkage disequilibrium with each other. The alleles were coded 0, 1, or 2 according to the number of uric acid–raising alleles. We then calculated a genetic score by summing the number of risk alleles. To take into account that effect sizes of individual SNPs differ, we calculated a weighted genetic score by weighing the individual SNPs by their effect on uric acid by using estimates from the previously published GWAS meta-analyses (9–11). Supplementary Table 1 provides an overview of the SNPs included in the genetic score and the weights assigned to each SNP.

Baseline information on lifestyle, diet, and medical history were obtained from self-administered questionnaires. Weight and height were recorded by trained health professionals during a visit to a study center. Presence of hypertension was defined based on self-reported diagnosis and/or use of medication. Physical activity was assessed by questionnaire and classified as inactive, moderately inactive, moderately active, and active, according to the Cambridge Physical Activity Index (16). The estimated glomerular filtration rate was estimated using the Chronic Kidney Disease Epidemiology Collaboration equation, with creatinine standardized to the Roche enzymatic method (17).

Associations of individual SNPs with uric acid were assessed with linear regression among the participants in the subcohort. Uric acid was modeled per 59.48 µmol/L (1 mg/dL), SNPs were modeled per uric acid–increasing allele (additive model), and associations were adjusted for study center. Associations of individual SNPs and the uric acid–related genetic score (per SD increase) with incident diabetes were examined with modified Cox regression that accounted for the case-cohort design (Prentice-weighted model [18]), adjusted for study center. We calculated country-specific hazard ratios (HRs), and used random-effects meta-analysis to calculate a pooled HR. We investigated associations of the uric acid–related genetic score (per SD increase) with potential confounders by using linear regression for continuous and logistic regression for dichotomous confounders.

For the observational association of uric acid and incident diabetes, we estimated country-specific HRs and pooled them through meta-analysis. We used I² to quantify heterogeneity between countries. Interactions with sex, age, and BMI were tested within each country by including interaction terms in the multivariable models. Country-specific estimates were pooled, as described above.

For the instrumental variable estimate of uric acid on diabetes risk, we used the weighted genetic score to estimate the unconfounded effect of an increase in uric acid of 59.48 µmol/L (1 mg/dL) on diabetes risk. We applied the two-stage control function estimator approach (19) for this instrumental variable estimate. Instrumental variable estimates were adjusted for study center, and in a second model, sex and BMI were added. Country-specific estimates were pooled as described above. The analyses were repeated in strata of sex, age, BMI, and duration of follow-up. Furthermore, we generated instrumental variable estimates of uric acid on glycemic traits (nonfasting glucose and HbA_1c), as described above.

Proportional hazard assumptions were inspected visually using log-minus-log plots, with no deviations detected.

Analyses were repeated after excluding participants with HbA_1c >6.5% (n = 22,146 for observational analysis and n = 15,380 for instrumental variable analysis). Furthermore, the observational association of uric acid and diabetes was estimated in the population used for the instrumental variable analysis (n = 17,118 instead of 24,265). We reanalyzed the instrumental variable estimate of uric acid on diabetes risk using the nonweighted genetic score, excluding SNPs that were not statistically significantly associated with uric acid in our study, excluding proxy SNPs with r² < 0.80, and excluding SNPs (rs734553; rs2231142) with the strongest effects on uric acid (Supplementary Table 1).

We estimated the power for the Mendelian randomization analysis at a two-sided α of 0.05 based on the sample size and proportion of cases, strength of the genetic instrument, and the expected causal HR using the online tool mRnd (http://glimmer.rstudio.com/kn3in/mRnd/) (20).

To maximize power, we additionally incorporated data made publicly available by GWAS consortia. For fasting glucose (n = 58,074) and HOMA-insulin resistance (n = 37,073), we used data from the Meta-Analyses of Glucose and Insulin-related traits Consortium (MAGIC), which is a collaborative effort that combined data from multiple GWAS to identify genetic determinants that affect glycemic and metabolic traits. Participants were of European ancestry and genotyped with the Metabochip (21). Data are publicly available at http://www.magicinvestigators.org/. For diabetes, we used data from DIAGRAM consortium, which meta-analyzed genetic variants on Metabochip in 34,840 case subjects with diabetes and in 114,981 control subjects from 37 studies (22). All studies participating in DIAGRAM included men and women, participants were mainly of European ancestry, the mean age varied from 43 to 72 years, and the mean study-level BMI varied from 25.9 to 33.4 kg/m² among case subjects with diabetes and from 22.3 to 28.3 kg/m² among control subjects. Data are publicly available at http://diagram-consortium.org/downloads.html.

For DIAGRAM, we selected the same 24 SNPs (either directly or in linkage disequilibrium >0.85) and extracted the odds ratios (ORs) and accompanying SEs. Diabetes estimates were meta-analyzed with ORs from InterAct (after excluding EPIC-Norfolk, which contributes to DIAGRAM), using fixed-effects meta-analysis on the log-scale, to generate a summary estimate for each SNP and diabetes risk. We then used pooled SNP-diabetes effect estimates (including up to 41,508 case subjects with diabetes) and external weights from uric acid GWAS (Supplementary Table 1) for instrumental variable analysis. Exactly the same process was repeated in MAGIC but without meta-analyzing MAGIC and InterAct because fasting glucose and HOMA-insulin resistance are not quantified in InterAct. We generated instrumental variable estimates for each SNP by dividing each SNP-trait effect estimate by the corresponding SNP-uric acid estimate. The analysis took into account the uncertainty in the SNP-trait and SNP-uric acid estimates by using the Δ method to estimate SEs of instrumental variable ratio estimates (23). We then pooled instrumental variable estimates across SNPs using fixed-effects meta-analysis to generate the summary causal effects. All analyses were performed using Stata 13.1 software (StataCorp LP, College Station, TX).

The mean (SD) age in the subcohort was 52 (10) years, and 65% were men. The mean (SD) uric acid concentration was 280 (77) µmol/L among the subcohort and 333 (83) µmol/L among case subjects with diabetes (Table 1). Mean uric acid ranged from 327 µmol/L in Italy and Sweden to 351 µmol/L in Spain among men, and from 241 µmol/L in Germany to 261 µmol/L in the Netherlands among women.

In the observational analysis, uric acid was associated with higher diabetes risk, with an HR of 1.51 (95% CI 1.42, 1.62) per 59.48 µmol/L (1 mg/dL) uric acid. After adjustment for confounders, the observed association attenuated but remained present, with a corresponding HR of 1.20 (95% CI 1.11, 1.30) in the multivariable model. BMI was the largest contributor to this attenuation (Table 2). Additional adjustment for red meat and vitamin C did not alter the findings (HR 1.22 [95% CI 1.11, 1.34]). The association remained consistent when we explored the association using the population selected for the instrumental variable analysis (HR multivariable model 1.25 [95% CI 1.13, 1.38]). Excluding participants with HbA_1c >6.5% yielded a multivariable HR of 1.26 (95% CI 1.17, 1.36).

Although all country-specific HRs directed toward a higher diabetes risk with higher uric acid concentrations, there was substantial heterogeneity among the countries (I² = 70%, P = 0.001; Supplementary Fig. 1). Heterogeneity remained present when the analyses were stratified by age, sex, and BMI, with no significant interactions for age and sex (P = 0.16 and P = 0.77, respectively, for interaction) and borderline significant (P = 0.06) for BMI, with no substantially different results in BMI strata (data not shown). After Sweden was excluded from the analysis, heterogeneity attenuated substantially, with I² of 48% (P = 0.07), and the association remained present (HR 1.17 [95% CI 1.09, 1.25]).

Individual uric acid–associated SNPs were all directly associated with uric acid, with the strongest association for rs734553 on locus SLC2A9 (Table 3). The individual SNPs were generally not associated with diabetes risk (Table 3).

The mean (SD) uric acid–associated genetic score was 1.55 (0.25) in the subcohort and case subjects with diabetes and was normally distributed among the study participants. Each SD higher genetic score was associated with a 17 µmol/L (95% CI 15, 18) higher uric acid concentration (Supplementary Table 2). The genetic score explained 4% of the proportion of variance of uric acid (F statistic = 462). The genetic score did not associate with diabetes risk (HR 1.01 [95% CI 0.97, 1.05] per SD higher genetic score; Supplementary Fig. 2).

The uric acid–associated genetic score was associated with higher triglyceride concentrations (β = 0.01 mmol/L [95% CI 0.001, 0.02] per SD higher genetic score), and a borderline association was identified with vitamin C intake and physical activity. Remaining potential confounders or mediators were not associated with the genetic score (Supplementary Table 3).

Using the uric acid–associated genetic score to estimate the unconfounded effect of uric acid (per 59.48 µmol/L [1 mg/dL]) on diabetes showed no evidence for an effect (HR 1.01 [95% CI 0.87, 1.16]). There was no substantial heterogeneity among countries (I² = 16%, P = 0.31; Supplementary Fig. 3). This did not materially change after further adjustment for sex and BMI (Table 2). No differential effects were found in subgroups based on sex, age, BMI, and duration of follow-up (Supplementary Table 4). Furthermore, there was no evidence for an effect of uric acid on glycemic traits (Supplementary Table 5).

Excluding participants with HbA_1c >6.5% yielded an HR of 1.02 (95% CI 0.89, 1.17). Using the nonweighted genetic score as the instrumental variable instead of the weighted genetic score yielded an HR of 0.96 (95% CI 0.71, 1.30). Excluding SNPs from the weighted genetic score that were not associated with uric acid in our study did not change our findings (HR 1.02 [95% CI 0.89, 1.17]) and neither did excluding proxy SNPs with r² < 0.80 (HR 0.99 [95% CI 0.85, 1.16]). Adjustment for triglycerides, vitamin C, and physical activity did not materially alter the estimate (HR 0.97 [95% CI 0.82, 1.15]).

Inclusion of DIAGRAM increased our dataset to 41,508 case subjects with diabetes and yielded a summary causal estimate of OR 0.99 (95% CI 0.92, 1.06) (Table 2 and Supplementary Fig. 4). Exclusion from this combined dataset of the two SNPs that most strongly associated with circulating uric acid (rs734553 in SLC2A9 and/or rs2231142 in ABCG2) did not alter the summary estimate (Supplementary Table 6).

Power calculations for our Mendelian randomization analysis are reported in Supplementary Table 7. In InterAct, we had 100% power to detect an HR of 1.51, 68% power to detect an HR of 1.20, and 31% power to detect the same effect estimate when we excluded rs734553. Inclusion of DIAGRAM increased the power to detect an HR of 1.2 for all sensitivity analyses to more than 90% (Supplementary Table 7), meaning that the estimates derived from the combined analysis (InterAct and DIAGRAM) were well powered for all scenarios.

In this large European case-cohort study, we found a 20% higher diabetes risk per 59.48 µmol/L (1 mg/dL) higher circulating uric acid concentration in multivariable observational analysis. Instrumental variable analysis did not confirm this association and suggested no evidence of a causal effect of circulating uric acid on diabetes risk.

The results of the observational analysis are in line with previous reports (1,2). Two previous meta-analyses showed a 6–17% higher diabetes risk per 59.48 µmol/L (1 mg/dL) uric acid. We found a 20% higher risk per 59.48 µmol/L (1 mg/dL), which is comparable to the previous studies. However, residual confounding and/or reverse causality may explain these associations because we did not find evidence for such an association in instrumental variable analysis. The results of our instrumental variable analysis generally agree with previous studies. Our findings are in agreement with the previous Mendelian randomization study of uric acid and diabetes that included fewer uric acid–associated loci and used different studies to estimate the three sides of the Mendelian randomization triangle (12). Moreover, a study of Yang et al. (11) showed no association of a genetic score for uric acid with plasma glucose concentrations, in line with our results. Studies that used a genetic uric acid score or SLC2A9 as an instrumental variable also suggested a bystander role for uric acid in other metabolic and cardiovascular traits, namely, metabolic syndrome (24,25), ischemic heart disease (26), markers of subclinical atherosclerosis (27), markers of adiposity (28), and triglycerides (29). The results for blood pressure are mixed, with reports of no effect (26), reducing effects (30,31), and increasing effects (32) (Supplementary Table 8).

There are observations that support a potential causal role of uric acid, whereas others suggest a bystander role. Hyperinsulinemia decreases renal excretion of uric acid, leading to increased blood concentrations of uric acid (3), supporting a bystander role. Furthermore, subclinical chronic inflammation may precede the development of diabetes (33), and uric acid generation may be increased as a result of oxidative stress. Support for a causal role comes from a recent study showing that intestinal knockdown of uric acid resulted in hyperuricemia and development of metabolic syndrome in mice (34). Moreover, there are reports that xanthine oxidase inhibitors (pharmacological agents used to lower uric acid) may improve endothelial function, which may reduce insulin resistance (3). However, it has been suggested that this may represent an additional effect of enzyme inhibition that is unrelated to uric acid, because therapies other than xanthine oxidase inhibitors that reduce uric acid concentrations did not show the same benefits to endothelial function (7,35). Inhibition of xanthine oxidase may improve endothelial function by reduction of oxidative stress instead of lowering of uric acid (7).

Strengths of our study are its large sample size (especially including data from DIAGRAM, which provided a cumulative total of more than 40,000 case subjects with diabetes and bolstered our power for sensitivity analyses), heterogeneous European population, and availability of a comprehensive range of potential confounders. Moreover, uric acid concentrations were available for all participants and were measured centrally to optimize comparability of uric acid concentrations among participants. Furthermore, our findings showed robustness in sensitivity analysis.

A potential limitation of our study is that the genetic score explained only 4% of the variation in uric acid. The percentage of explained variation is very comparable to previous Mendelian randomization studies (36), and the corresponding F statistic was high, indicating the study was unlikely to suffer from weak instrument bias (13). Second, our study investigated the effect of circulating uric acid in blood and does not necessarily also reflect effects of intracellular uric acid. Individual SNPs in the gene score may have differential effects on uric acid concentration by body compartment (34,37). Despite this, it is not plausible there will be common pleiotropy among the individual SNPs included in the score, and any pleiotropic roles of SNPs should be balanced out by use of a polygenic score (38). Third, our study population was of European ancestry, which limits generalizability to populations of other ancestries.

Mendelian randomization studies are a valid way to explore evidence for causality, given that certain assumptions are met. First, there has to be a strong association between the instrumental variable and risk factor of interest. All SNPs used in this study have been shown to be strongly associated with uric acid concentrations in large meta-analyses of GWAS (9–11). Some SNPs did not associate with uric acid in our study; however, the null-association remained present when we excluded those SNPs from the genetic score. Moreover, we strengthened our instrumental variable by using a genetic score of multiple uric acid–associated SNPs. No SNPs were in linkage disequilibrium with each other, which justifies combining those SNPs.

Second, the instrumental variable must be independent of potential confounders (confounders in the association between uric acid and diabetes). To test this, we examined the associations of the genetic score with potential confounders and found an association with triglycerides. However, whether this is a true confounder or a downstream consequence of uric acid pathways can be debated. Moreover, because we did not find an association of uric acid and diabetes in the instrumental variable analysis, it is not likely that this is explained by the higher risk of hypertriglyceridemia in individuals with a high genetic score. Indeed, when we additionally adjusted the instrumental variable estimate of uric acid on diabetes risk for triglycerides, the null-effect remained. The observed higher triglyceride concentrations suggest that, although uric acid may not be causally involved in development of diabetes, there may be a separate causal role for uric acid in this metabolic disorder.

Third, the instrumental variable affects the outcome only through the risk factor of interest. This assumption cannot be tested and should be considered using information on the underlying biology. None of the SNPs used in this study were in linkage disequilibrium with loci known to influence diabetes risk (22,39,40), which strengthens this assumption. Moreover, the vast majority of SNPs identified in the meta-analysis of Kolz et al. (9) were involved in regulating urate transport across cell membranes, which suggests that these SNPs directly influence uric acid levels. However, SLC2A9, the strongest uric acid–associated locus, transports not only uric acid but also glucose and fructose (41) and exchanges uric acid for glucose (42), leaving room for possible pleiotropy. Moreover, SLC2A9 has recently been shown to have differential effects on urinary and intestinal secretion of uric acid in a mouse model, suggesting a rise in serum uric acid due to reduced urinary secretion could be counterbalanced by increased intestinal secretion and decreased portal vein levels (34). Similar contrasting roles have been reported for ABCG2 (37). A sensitivity analysis excluding the SNPs in these loci did not alter the result (Supplementary Table 6).

In conclusion, our study does not support the hypothesis that circulating uric acid has a causal effect on diabetes risk. Our findings therefore suggest that increased uric acid concentrations are a consequence of an adverse metabolic profile rather than a cause of diabetes and that uric acid has limited value as a therapeutic target in preventing diabetes.

Acknowledgments. The authors thank staff from the Technical, Field Epidemiology, and Data Functional Group Teams of the MRC Epidemiology Unit in Cambridge, U.K., for carrying out sample preparation, DNA provision and quality control, genotyping, and data-handling work. The authors specifically thank S. Dawson for coordinating the sample provision for biomarker measurements, A. Britten for coordinating DNA sample provision and genotyping of candidate markers, N. Kerrison, C. Gillson, and A. Britten for data provision and genotyping quality control, and M. Sims for writing the technical laboratory specification for the intermediate pathway biomarker measurements and for overseeing the laboratory work (all MRC Epidemiology Unit, Cambridge, U.K.). The authors thank all EPIC participants and staff for their contribution to the study. The authors thank N. Kerrison (MRC Epidemiology Unit, Cambridge, U.K.) for managing the data for the EPIC-InterAct project.

Funding. Funding for the EPIC-InterAct project was provided by the European Union Sixth Framework Programme (grant number LSHM_CT_2006_037197). In addition, EPIC-InterAct investigators acknowledge funding from the following agencies: I.S., Y.T.v.d.S., J.W.J.B.: Verification of case subjects with diabetes was additionally funded by NL Agency grant IGE05012 and an Incentive Grant from the Board of the UMC Utrecht; M.V.H.: Medical Research Council Population Health Scientist Fellowship (G0802432); J.M.H.: Health Research Fund of the Spanish Ministry of Health, Murcia Regional Government (No. 6236); P.W.F., P.M.N.: Swedish Research Council; P.W.F.: Swedish Diabetes Association, Swedish Heart-Lung Foundation; R.K.: German Cancer Aid, German Federal Ministry of Education and Research (BMBF); K.T.K.: Medical Research Council U.K., Cancer Research U.K.; T.K.: German Cancer Aid, German Cancer Research Center (DKFZ), German Federal Ministry of Education and Research (BMBF); K.O., A.T.: Danish Cancer Society; S.P.: Compagnia di San Paolo; J.R.Q.: Asturias Regional Government; O.R.: The Västerboten County Council; N.S.: Spanish Ministry of Health Network, Red Temática Investigación Cooperativa en Cáncer (ISCIII RD06/0020/0091); A.M.W.S., D.L.v.d.A: Dutch Ministry of Public Health, Welfare and Sports, Netherlands Cancer Registry, LK Research Funds, Dutch Prevention Funds, Dutch ZON (Zorg Onderzoek Nederland), World Cancer Research Fund, Statistics Netherlands; R.T.: AIRE-ONLUS Ragusa, AVIS-Ragusa, Sicilian Regional Government.

Duality of Interest. P.W.F. received funding from Novo Nordisk. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. I.S. had access to all data for this study, analyzed the data, drafted the manuscript, and takes responsibility for the manuscript contents. M.V.H. helped with analyses and drafting of the manuscript. T.M.P. provided analytical tools. All authors qualify for authorship according to American Diabetes Association criteria. They all contributed to conception and design, interpretation of data, revising the article critically for important intellectual content, and final approval of the version to be published. I.S. is the guarantor of this work and, as such, had 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.