The Comparative Effect of Exposure to Various Risk Factors on the Risk of Hyperuricaemia

Diet Has a Weak Causal Effect

Ruth K. G. Topless; Tanya J. Major; Jose C. Florez; Joel N. Hirschhorn; Murray Cadzow; Nicola Dalbeth; Lisa K. Stamp; Philip L. Wilcox; Richard J. Reynolds; Joanne B. Cole; Tony R. Merriman


Arthritis Res Ther. 2021;23(75) 

In This Article


Population and Average Attributable Fractions

Male sex had the largest PAF and AAF (64.3 to 75.6%, and 29.8 to 30.9%, respectively, in cohorts 1 to 3), being overweight or obese had the second largest measures (PAF = 59.2 to 68.6%, AAF = 23.5 to 26.2%), and inheriting the SLC2A9 rs12498742 A-allele had the third largest effect (PAF = 56.6 to 63.8%, AAF = 22.0 to 23.5%). All other factors had PAFs < 25% and AAFs < 10% (Table 2). Cohorts 2 and 3 had the same ranking of risk factors (sex > BMI > rs12498742 > diet adherence > alcohol consumption > diuretic therapy > age), while cohort 1 ranked the same three risk factors first (sex > BMI > rs12498742), before differing in rankings of the remaining four risk factors (diuretic therapy > diet adherence > age > alcohol consumption) (Figure 1).

In the gout cohort, lack of treatment with urate-lowering therapy had the second largest PAF of 63.2%, after sex (76.7%) (Table 2). The PAFs for BMI and SLC2A9 rs12498742 were lower than for the non-gout cohorts (57 to 69% in non-gout and 48 to 49% in gout), and non-adherence to the Healthy Eating Pyramid guidelines was 21.4% in non-gout (cohort 3) and 11.8% in gout. Average attributable fraction values showed a similar trend (Figure 1).

In sex-stratified analysis in the non-gout cohorts, PAFs for rs12498742 were 77.7 to 96.4% in women compared to 49.2 to 58.9% in men (Table S1). Alcohol was not a risk factor in women across all cohorts, nor age in men in cohorts 2 and 3, and gout (Table S1; 95% CI encompassed 1.0). The lower age limit for recruitment into the UK Biobank, which these two cohorts were derived from, was 40 years, which may have influenced the calculation.

In the non-gout cohorts, of 30 genetically-independent serum urate-associated genetic variants chosen as having the top effects by GWAS[33] evaluated (Table S2), the SLC2A9 rs12498742 variant was the largest, with PAFs ranging from 28.5 to 32.1% and AAFs from 22.0 to 23.5%. For comparative purposes, we summed the PAFs for genetic variants, assuming that the variants act independently of each other to influence the risk of HU, with the individual PAFs summing to > 141% (cohort 1 was 146.2%, cohort 2 was 143.7%, cohort 3 was 141.3%). Summing the AAFs (equivalent to the summing of PAFs, above) resulted in all three cohorts having a summed AAF over 87% (cohort 1 was 97.9%, cohort 2 was 87.1%, cohort 3 was 101.6%). Summed attributable fractions for genetic variants for HU were considerably lower for the gout cohort (PAF was 77.6% and AAF was 44.0%), possibly reflecting selection (collider) bias.

Percent Variance Explained for Serum Urate Levels

In the non-gout cohorts, sex had the most percent variance explained (22 to 27%) (Table 3). The dichotomised BMI exposure was consistently 7 to 9%, with diuretic exposure accounting for 12% variance in the US-based cohort and 4–5% in the UK-derived cohorts. The diet estimate was ≤0.1% and SLC2A9 was 2–3%, similar to our previous report.[10] The gout cohort included urate-lowering therapy exposure in the model, with exposure accounting for the largest proportion (35%) of variance, approximately 10-fold more than any other variable. The use of percent variance explained produced a broadly similar ranking order of risk factors to the PAF and AAF analyses across all four cohorts (Figure 1).

Mendelian Randomisation

Five of the 87 single foods and principal component-derived dietary-associated habits[26] provided evidence of a causal effect (IVW P < 0.05/87 (5.7 × 10− 4)) on urate levels by inverse-variance-weighted MR (Table S3). All five of these dietary habits also had no evidence for an intercept significantly different from zero in the MR Egger analysis (all P > 0.05) indicating no evidence for directional (horizontal) pleiotropy. Four of these dietary habits provided evidence for a causal role (P < 0.05) and yielded similar effect sizes in the weighted median analysis (Table 4). Two of these causal effects were with dairy-related dietary habits (preferentially drinking skim milk and preferentially drinking milk with a higher fat content), and the other two causal effects were for consuming tub margarine and daily dried fruit consumption.

Of the 39 genetic variants that comprised the four dietary-associated habits, 21 are associated with metabolic traits ( [accessed: 2nd June 2020]) and/or traits available in the UKBiobank PheWeb ( [accessed: June 2, 2020]), including 16 specifically associated with BMI or a related body fat trait (Table S4). To test the possibility that the causal association between these four dietary habits and urate levels is due to BMI as a common upstream cause (e.g. change in dietary habits due to weight-loss advice), we applied multivariable MR using the same individual level UK Biobank dataset described in ref..[26] For all four dietary patterns, including BMI in the multivariable analysis resulted in no evidence for a causal effect (P ≥ 0.06), BMI showing a causal relationship with urate levels independent of the dietary habit (Figure 2). Bidirectional MR between BMI and each of the four dietary habits where, by inverse variance-weighted meta-analysis MR BMI was tested for a causal effect on the dietary habits and each of the dietary habits was tested for a causal effect on BMI, conducted to confirm whether BMI is a common upstream cause of dietary habits, provided evidence in both directions (P ≤ 4.4 × 10− 18 for BMI to dietary habit, P ≤ 9.4 × 10− 4 for dietary habit to BMI)—a situation termed "correlated pleiotropy"[43]—except in the margarine analysis for the BMI to dietary habit analysis (P = 0.24) although there was evidence for the dietary habit to BMI analysis (P = 8.4 × 10− 61). This indicates that BMI and the four dietary habits are strongly correlated traits or work through a shared pathway and that the four dietary habits have no effect on urate levels independent of BMI.

Figure 2.

Direct effect of dietary habits on serum urate levels, independent of BMI: multivariable Mendelian randomisation. The solid red arrow and values indicate the causal effect identified in the original inverse variance weighted MR; the dashed red arrow indicates the correlated pleiotropy between BMI and the dietary habit, influencing this original inverse variance weighted MR result; the black arrows and values indicate the causal effect independent of the other exposure variable. Beta values are in mmol/L. Figure 2a relates to results for preferentially drinking skim milk (vs. any other milk type); Figure 2b relates to results for consuming tub margarine (vs. no spread use)—the dashed red arrow is paler in this figure due to the lower confidence surrounding the correlated pleiotropy; Figure 2c relates to results for preferentially drinking milk with a higher fat content; and Figure 2d relates to results for dried fruit consumption (pieces per day)