The participants, methods, study design and measurements have been described previously . In brief, the InterAct project was initiated to investigate how genetic and lifestyle behavioural factors, particularly diet and physical activity, interact for the risk of developing diabetes and how knowledge about such interactions may be translated into preventive action. As part of the wider InterAct project, consortium partners established a case–cohort study of incident type 2 diabetes (InterAct study) based on cases occurring within 3.99 million person-years accrued between 1991 and 2007 in 340,234 people from eight countries participating in EPIC cohorts. All participants gave written informed consent, and the study was approved by the local ethics committees in the participating countries and the Internal Review Board of the International Agency for Research on Cancer.
We followed a pragmatic high-sensitivity approach for case ascertainment with the aim of: (1) identifying all potential incident diabetes cases; and (2) excluding all individuals with prevalent diabetes.
Prevalent diabetes was identified on the basis of baseline self-report of a history of diabetes, doctor-diagnosed diabetes, diabetes drug use or evidence of diabetes after baseline with a date of diagnosis earlier than the baseline recruitment date. All participants with any evidence of diabetes at baseline were excluded.
Ascertainment of incident type 2 diabetes involved a review of the existing EPIC datasets at each centre using multiple sources of evidence, including self-report, linkage to primary-care registers, secondary-care registers, medication use (drug registers), hospital admissions and mortality data. Information from any follow-up visit or external evidence with a date later than the baseline visit was used. Cases in Denmark and Sweden were not ascertained by self-report, but identified via local and national diabetes and pharmaceutical registers and hence all ascertained cases were considered to be verified. To increase the specificity of the case definition for centres other than those from Denmark and Sweden, we sought further evidence for all cases with information on incident type 2 diabetes from fewer than two independent sources at a minimum, including individual medical records reviews in some centres. Follow-up was censored at the date of diagnosis, 31 December 2007, or the date of death, whichever occurred first. In total, 12,403 verified incident cases were identified during follow-up.
The case–cohort design of the InterAct study differs from the nested case–control design in that a random subcohort is selected instead of a set of matched controls. A subcohort of 16,835 individuals were randomly selected from those with available stored blood and buffy coat, stratified by centre. We oversampled the number of individuals in the subcohort for the proportion of prevalent diabetes cases in each centre to account for later exclusion of individuals with prevalent diabetes from InterAct analyses. After exclusion of 548 individuals with prevalent diabetes, 129 individuals without information on reported diabetes status and four individuals with post-censoring diabetes, 16,154 subcohort individuals were included in the analysis. Because of random selection at baseline, this subcohort also included 778 individuals who developed incident type 2 diabetes during follow-up.
For the present analysis we excluded participants (both cases and participants in the subcohort) within the lowest and highest 1% of the cohort distribution of the ratio of reported total energy intake:energy requirement (n = 619) and those with missing information on diet (n = 117), physical activity (n = 289), level of education (n = 127), smoking status (n = 132) or BMI (n = 167). This resulted in a final sample size of 11,684 type 2 diabetes cases and a subcohort of 15,374 (including 730 of the diabetes cases).
Usual food intake over the previous year (g/day) was estimated using country-specific validated dietary questionnaires  that were administered once at baseline. Estimated individual energy intake (MJ/day) was derived from foods included in the dietary questionnaires through the standardised EPIC Nutrient Database .
Sweet beverages evaluated in the present study were juices and nectars and total soft drinks.
‘Juices and nectars’ combines the information collected on consumption of juices (obtained from either 100% fruit and vegetables or concentrates) and nectars (juices with up to 20% added sugar) across all EPIC countries/centres. The distinction of juices from nectars or of 100% fruit or vegetable juices from juices from concentrates could not be established because of a lack of standardised information across the different EPIC centres. Therefore, these food items were always studied in combination.
‘Total soft drinks’ combines the information collected on consumption of carbonate/soft/isotonic drinks and diluted syrups across all EPIC countries/centres. The different types of soft drinks could be distinguished in all EPIC centres except Italy, Spain and Umeå (in Sweden) and, hence, the variable of total soft drinks could be divided into: (1) sugar-sweetened soft drinks; and (2) artificially sweetened soft drinks. Italy, Spain and Umeå were excluded from analyses of the association between sugar-sweetened and artificially sweetened soft drinks and type 2 diabetes incidence.
Sweet beverages were divided into the following categories of average consumption: <1 glass/month; 1–4 glasses/month; >1–6 glasses/week; ≥1 glass/day, with one glass equivalent to 250 g (~8.8 oz), the standard serving size used in the EPIC dietary questionnaires.
Assessment of other covariates
Standard questionnaires were used to collect information on the participants’ sociodemographic characteristics and lifestyle variables . For the present analysis we used information about: smoking status (never smoker, former smoker and current smoker); alcohol intake (non-drinker, 0.1–4.9, 4.9–15, 15–30, 30–60 and >60 g/day); educational level (no formal education, primary school, technical school, secondary school and university degree); and an ordered four-category index of physical activity (inactive, moderately inactive, moderately active and active) .
Body weight (kg) and height (cm) were measured according to standardised procedures without shoes, except for the centres at Oxford (UK) and France, where self-reported anthropometric values were used. Waist circumference was measured in cm at the narrowest torso circumference or at the midpoint between the lower ribs and iliac crest in all centres but Umeå (Sweden). Weight and waist measurements were corrected to account for protocol differences between centres in clothing worn during measurement, as previously described . BMI was calculated as body weight (kg) divided by height squared (m2). Individuals were categorised into groups of normal weight (BMI <25 kg/m2), overweight (BMI 25–30 kg/m2) and obese (BMI ≥30 kg/m2).
In most participating centres, baseline information was collected on the presence of chronic conditions: hypertension; hyperlipidaemia; and previous cardiovascular disease (angina, stroke and myocardial infarction). Information on family history of type 2 diabetes in a first-degree relative was collected for all participants except for those in Italy, Spain, Germany and Oxford (UK).
Cox proportional hazards regression, modified for the case−cohort design according to the Prentice method , was used to estimate the association between consumption of sweet beverages and incidence of type 2 diabetes. Age was used as the underlying timescale, with entry time defined as the participant’s age at recruitment and exit time as age at diagnosis of diabetes, censoring or death (whichever came first). Centre or country analyses were run separately and HRs were then combined across countries/centres using random-effects meta-analysis. From this analysis I
2, the percentage of variation between countries/centres due to heterogeneity, was calculated.
Exposure variables (consumption of juices and nectars, total soft drinks and, for the centres where the information was available, sugar-sweetened soft drinks and artificially sweetened soft drinks) were assessed as categorical variables (<1 glass/month, 1–4 glasses/month, >1–6 glasses/week, ≥1 glass/day). The tests for linear trend were performed by including median values of consumption within each category in the Cox regression models. For comparison with previous studies, HRs were also calculated per one 12 oz serving size increment in sweet beverage consumption (equivalent to 336 g/day) . In secondary analyses, the study participants were also classified into sex-specific tertiles according to the distribution of intake in consumers within the subcohort or non-consumers. We assessed the association between tertiles of intake and diabetes incidence using models similar to those described earlier, but because the results were very similar in terms of effect size and significance level, these are not presented here.
We ran both crude models and models adjusted for sex, smoking status, alcohol consumption, educational level and physical activity. Juices and nectars and total soft drinks were mutually adjusted. Sugar-sweetened and artificially sweetened soft drinks were also mutually adjusted and also adjusted for juice and nectar consumption.
Sensitivity analyses included adjustment for total energy intake and BMI (as continuous variables). We did this only in a sensitivity analysis because total energy intake and BMI could act as an intermediate of the association between sweet beverages and diabetes risk, in which case they should not be included in the main multiply adjusted model . Also, further adjustments for waist circumference (continuous), presence of hyperlipidaemia (yes, no) and hypercholesterolaemia (yes, no) were performed.
To deal with plausible residual confounding associated with the dietary pattern of those with high consumption of sweet beverages, models were also adjusted for the consumption of vegetables, fruits, nuts, cereals and products, dairy products, red meat, processed meat, sugar and confectionary, cakes and biscuits, and coffee and tea (included in the model as continuous variables expressed in g/day). Also, given that a Mediterranean dietary pattern (MDP) has been shown to be associated with both type 2 diabetes incidence  and consumption of sweet beverages , we adjusted for the relative Mediterranean Diet Score (rMED; score range, 0–18), which assesses adherence to MDP, based on reported consumption of nine dietary components characteristic of an MDP: vegetables; legumes; fruits and nuts; cereals; fish and seafood; olive oil; moderate alcohol consumption; meat and meat products; and dairy products. More information on the construction of rMED can be found elsewhere . Finally, to minimise reverse causality caused by people who may have changed their dietary habits because of an impaired glucose tolerance or chronic disease, we excluded participants with cardiovascular diseases at baseline (stroke, angina and heart disease), those with family history of type 2 diabetes, and participants in the first 2 and 5 years of follow-up.
Effect modification by age group (<55 and ≥55 years old), sex, BMI category (<25, 25–<30 and ≥30 kg/m2) and physical activity level (combining the four-category index into two categories, low and high) was assessed by modelling cross-product terms between these variables and sweet beverages, and conducting stratified analyses.
All statistical analyses were performed with Stata 10.0 (StataCorp, College Station, TX, USA). A level of p < 0.05 was regarded as statistically significant.