The design of the GCS has been described before 
. Briefly, the GCS is a prospective population-based cohort study, launched in January 2004, which has recruited 50044 adults between 30 and 87 years old from Golestan Province.
Using systematic clustering based on household numbers, a total of 39399 individuals from 326 rural villages and 10645 urban residents were enrolled. Demographics and baseline information including age, sex, education, ethnicity, place of residence, number of owned household appliances, and history of tobacco and opium use were collected using a structured lifestyle questionnaire. Anthropometric data were measured and samples of blood, urine, hair and nails were gathered from the participants by a trained technician after the interview.
Education (highest level attained) and appliance ownership, including bath in the residence, personal car, motorbike, black and white TV, color TV, refrigerator, freezer, vacuum cleaner and washing machine, were used as indicators of SES. Using multiple correspondence analysis, we created a wealth score based on the appliance ownership variables. These scores were calculated and participants were categorized into wealth score quartiles 
Given the lifestyle of this mainly rural population, most of the activities individuals have are at work. As a result, only physical activity at work was looked at in this analysis. Two questions were asked about individuals' work activity: if the person worked every month throughout the year, and if intense physical activity was part of the daily work. Three levels of occupational physical activity were defined based on the answers to these questions: intense physical activity at work, non-intense but regular physical activity and non-intense irregular physical activity.
Individuals were considered tobacco users if they had smoked cigarettes or had used nass, hookah or a pipe at least once a week for a period of 6 months or more. Individuals were categorized into these groups: never smokers, former cigarette smokers, current cigarette smokers, and those who smoked other forms of tobacco (nass, hookah, or a pipe). Current cigarette smokers were further divided into light and heavy smokers if they fell below or above the median pack years for the nondiabetic smokers. Likewise, opium users were defined as those who consumed opium at least once a week for 6 months or more. The self-reported use of opium is a reliable and valid indicator of opium exposure in this population 
Systolic and diastolic blood pressures were measured twice in each arm in the sitting position and averaged. Participants were considered as being hypertensive if they either reported a physician's diagnosis of hypertension, were using anti-hypertensive medication, or fulfilled the criteria of the Joint National Committee on Prevention, Detection, Evaluation, and Treatment of High Blood Pressure (JNC-7) (average systolic blood pressure above ≥140 mmHg, or average diastolic blood pressure above ≥90 mmHg) 
. DM was self-reported based on this question: “Have you ever been diagnosed by a doctor as having diabetes mellitus?”.
Green tea consumption was categorized based on both the frequency and amount of drinking. Non-drinkers did not drink green tea at all, occasional green tea drinkers drank it less than once a week, and frequent drinkers drank it at least once a week, but not every day. Those who drank green tea every day were divided into low and high intake groups based on whether they drank less or more than the median (600 ml), respectively. Black tea consumption was divided into quartiles based on average daily drinking, since there were very few people who didn't drink black tea every day.
Oral health status was summarized using the sum of the number of decayed, missing, or filled teeth (DMFT), and categorized into 3 levels: <20, 21–31, and 32.
Body mass index (BMI), as a measure of overall obesity, was calculated by dividing measured weight (kg) by the square of the measured height (m), and categorized using the World Health Organization (WHO) cutoffs: underweight (BMI<18.5 kg/m2
), normal (18.5≤BMI<25 kg/m2
), overweight (25≤BMI<30 kg/m2
), and obese (BMI≥30 kg/m2
. Waist circumference (WC) was used as a measure of abdominal obesity. Individuals were categorized as either normal or high risk (WC>102 cm in men and >88 cm in women) according to the adult treatment panel (ATP) III criteria 
. Additionally, participants were also categorized into quintiles of WC.
Individuals' body size perceptions at ages 15 and 30 were assessed using a set of drawings (pictograms), ranging from very lean to obese. These pictograms were developed by Stunkard et al 
, and have been shown to have good accuracy for anthropometric assessment in this population 
. The pictograms were scored from 1 to 7 in men, and from 1 to 9 in women (). The highest two categories of pictogram score were combined together due to the relatively small number of observations in these categories. Obesity was defined as a pictogram score of 5 and above 
, Change in pictogram score between ages 15 and 30 was used to assess the association between change in body size during adulthood and DM. Study participants were categorized into four categories: no change, decrease, slight increase (a 1 or 2 category increase) and prominent increase (a more than 2 category increase).
Body size pictograms used in the Golestan Cohort Study.
Five years after recruitment, fasting plasma glucose (FPG) level was measured for a random sample of 3811 cohort participants. The same baseline questionnaire (including DM self-report) was again administered at the time of blood draw. Individuals who had FPG≥126 mg/dl (7.0 mmol/l) (the recommended cutoff of the American Diabetes Association 
) or were under anti-diabetic treatment were categorized as having confirmed DM. This information was used to assess the sensitivity and specificity of self-reported DM in this study.
The GCS was approved by the Institutional Review Boards of the Digestive Disease Research Center of Tehran University of Medical Sciences, the US National Cancer Institute (NCI), and the World Health Organization International Agency for Research on Cancer (IARC). All participants gave written informed consent before enrollment.
The World Standard Population 2000–2005 developed by the WHO 
and the 2009 population provided by the Statistical Center of Iran 
were used for world and national age-standardizations, respectively, using the direct age-standardization method.
We used Poisson regression with robust variance estimator to get unbiased estimates of prevalence ratios (PR). Poisson regressions with robust variance estimator are useful alternatives to log-binomial models; they work equally well when the model is correctly specified, and are not subject to the convergence difficulties 
Multicollinearity was assessed using the variance inflation factor (VIF). Aside from obesity-related covariates, no evidence for serious multicollinearity was observed (all VIFs were below 1.5). Univariate models were first fitted to assess the independent association between each covariate and DM. Potential confounders and mediators were identified using a Directed Acyclic Graph (DAG). Variables included in the DAG were those that have been consistently reported to be associated with DM, have relevant biological mechanisms in the disease process, or have been hypothesized to be associated with DM.
Multivariate models were fitted to assess the direct association between each of the covariates of interest and diabetes. For age, the PR was calculated per 10 years increase in age. Since BMI, WC and pictograms at 15 and 30 years are all measures of obesity, and also showed high VIF (>2.5), separate models were built for each to avoid collinearity. Obesity-related covariates were assessed separately in men and women, but as the effect of BMI on DM was similar in both sexes, only the pooled effect was reported. All these models were further adjusted (according to DAG) for age, ethnicity, place of residence, education and wealth score (2 different indicators of SES), physical activity, tobacco use, opium use, hypertension, green tea consumption, black tea consumption, and DMFT score. Another model was built to assess whether change in body size in early adulthood (between 15 and 30) was associated with DM risk. Since size at a young age is invariably correlated with size later in life, we additionally adjusted this model for body size at 15 years. Finally, the cumulative effect of obesity since 15 years of age was assessed using the combination of obesity at 15 and 30 (pictogram score ≥5) and at the time of recruitment (BMI≥30). Individuals were categorized into 5 groups; never obese, obese at age 15, obese at both ages 15 and 30, obese at age 30 and recruitment, and always obese.
Sensitivity and specificity of self-reported DM were calculated using the data collected 5 years after recruitment. Confirmed DM (defined above) was used as a gold standard for this calculation.
All statistical analyses were performed using STATA statistical software version 11 (Stata Corporation, College Station, TX, USA). We used hotdeck method to impute the missing values for variables with more than 50 missing observations. All tests of hypothesis were conducted at a confidence level of 0.95 under the two-sided alternative.