Association of Major Food Sources of Fructose-Containing Sugars With Incident Metabolic Syndrome

This systematic review and meta-analysis examines the association of major food sources of fructose-containing sugars with incident metabolic syndrome.


eAppendix 2. Conversion of OR to RR
For studies with reported hazard ratios, low incidence of MetS (<10%) or odds ratios (OR) between 0.5 and 2.5, values were treated as RRs. OR were converted to RR if OR was less than 0.5 or greater than 2.5 with an incident of MetS greater than 10%. As outlined by Zhang et al. 1 , the following formulae and logic were applied: Thus, P 0 = incidence of the outcome of interest in the non-exposed group; P 1 = incidence of the outcome of interest in the exposed group.

eAppendix 3. Method for Dose-Response Analysis
We modelled dose-response model using RR and 95% CIs from dose categories to understand the shape of the association between the dose of the food source of fructose-containing sugar and the risk of MetS. Data on the dose, distribution of cases and person-years, RRs and 95% CIs were extracted from each study. We defined the assigned dose as the mean consumption in each reported category or quantile. If the assigned were not reported, we approximated the mean dose for each category by using the midpoint of its lower and upper bounds. If the lowest category of a study was open ended, we defined the lowest dose as zero. For open-ended upper categories, we took half of the adjacent category range to estimate the assigned dose. When cohort size or person-year per category were not available, categories were regarded equal in size and follow-up and the case number per category was obtained by the method of Bekkering 2 . We excluded studies from the dose-response meta-analysis that did not report any dose category cut points for the particular food source and studies that provided only RR estimates based on 1-unit increment in dose based on a linear model because these studies were unable to contribute to the assessment of departure from linearity. We fitted a dose-response relationship using restricted cubic splines with 3 knots at 15th, 50th and 85 th percentiles of distribution taking into account the correlation within each category of published RRs and combining the study-specific estimates the one-stage linear mixed-effects meta-analysis 3 . This method estimates the study specific slope lines and combines them to obtain an overall average slope based upon the work of Greenland 4 and Orsini 5 . If restricted cubic splines could not be calculated due to limited number of observations, we fitted a second order fractional polynomial curve to the data 5 and tested for goodness-of-fit of the model using Akaike information criterion (AIC), deviance test (D) and the coefficient of determination (R2) to select the best-fitting model 6 . We reported non-linear associations for a study if Wald test for departure from linearity was significant at p<0.10. RRs below 1 were considered as protective and above 1 as adverse association.
The harmonized criteria classification for MetS takes into account definitions set by the International Diabetes Federation (IDF) and the American Heart Association/ National Heart, Lung, and Blood Institute ATPIII. 7 Both IDF and ATP III definitions include thresholds for waist circumference, elevated triglycerides of ≥150 mg/dL, low HDL of <40 mg/dL in males and <50 mg/ in females, elevated blood pressure of systolic ≥130 and/or diastolic ≥85 mm Hg, and elevated fasting blood glucose of ≥150 mg/d. 8 The ATP III identifies MetS as the presence of any 3 of 5 risk factors. The IDF defines MetS as the presence of abdominal obesity measured through waist circumference, with the addition of any 2 of 4 risk factors. The harmonized criteria defines MetS as the presence of any 3 of 5 risk factors, with specific waist circumference cut-points depending on ethnicity. 8 eTable 1. Search Strategy.   21 3 3 2 8 *Maximum 4 points awarded for cohort representativeness, selection of non-exposed cohort, exposure assessment, and demonstration outcome not present at baseline †Maximum 3 points awarded for follow-up length, adequacy of follow-up, and outcome assessment ‡ Maximum 2 points awarded for controlling for the pre-specified primary confounding variable (age) and additional confounding variables. § A maximum of 9 points could be awarded eTable 4. GRADE Assessment.

*
Although there was evidence of substantial inter-study heterogeneity (I 2 = 68%), the estimates were all in the same direction and there was considerable overlap. Therefore, we did not consider this as serious inconsistency. ** There was substantial heterogeneity (I 2 = 73%, P Q = 0.05) in the pairwise analysis. This was explained by the non-linear dose-response model. Therefore, we did not downgrade for serious inconsistency.   The black diamond represents the pooled risk estimate. Inter-study heterogeneity was tested using the Cochran Q statistic (Chi 2 ) at a significance level of P < 0.10, and quantified by the I 2 statistic. An I 2 value ≥ 50% is considered to indicate substantial heterogeneity. All results are presented as Relative Risks (RR) with 95% Confidence Intervals where estimable.

eFigure 4. Relationship between fruit intake and incident MetS.
The black diamond represents the pooled risk estimate. Inter-study heterogeneity was tested using the Cochran Q statistic (Chi 2 ) at a significance level of P < 0.10, and quantified by the I 2 statistic. An I 2 value ≥ 50% is considered to indicate substantial heterogeneity. All results are presented as Relative Risks (RR) with 95% Confidence Intervals where estimable.

eFigure 5. Relationship between yogurt intake and incident MetS.
The black diamond represents the pooled risk estimate. Inter-study heterogeneity was tested using the Cochran Q statistic (Chi 2 ) at a significance level of P < 0.10, and quantified by the I 2 statistic. An I 2 value ≥ 50% is considered to indicate substantial heterogeneity. All results are presented as Relative Risks (RR) with 95% Confidence Intervals where estimable. eFigure 6. Relationship between honey intake and incident MetS.
The black diamond represents the pooled risk estimate. Inter-study heterogeneity was tested using the Cochran Q statistic (Chi 2 ) at a significance level of P < 0.10, and quantified by the I 2 statistic. An I 2 value ≥ 50% is considered to indicate substantial heterogeneity. All results are presented as Relative Risks (RR) with 95% Confidence Intervals where estimable.