"Predictability of body mass index for diabetes: Affected by the presence of metabolic syndrome?"

The Tehran Lipid and Glucose Study (TLGS) is a prospective population based study performed on a representative sample of the Tehran population, with the aim of determining the prevalence of non-communicable disease (NCD) risk factors and developing a healthy lifestyle to improve them [11, 12]. The baseline survey was performed from February 1999 to July 2001(phase 1) and 4751 families, which included more than 15,000 residents of district-13 of Tehran aged ≥3 years were selected by cluster random-sampling method. After this cross-sectional phase, subjects entered into a cohort and a prospective interventional study (lifestyle modification education). The current study used the data from 10,368 individuals ≥ 20 years who had baseline examination. After exclusion of participants with prevalent diabetes (n = 1164), and those with missing data regarding fasting and 2 hour post challenge plasma glucose (2 h-PCPG) (n = 884), BMI and MetS definition (n = 199), 8,121 non-diabetic participants remained eligible to be reexamined in two consecutive phases, one from September 2001 to August 2005 (phase 2) and the other from April 2005 to March 2008 (phase3). The same standard approach is followed to collect information across consecutive examinations of the TLGS follow up study. Participants with at least one follow-up examination (5,250) were included for the current study (Figure 1).

Informed written consent was obtained from all participants and the ethical committee of the Research Institute for Endocrine Sciences approved this study.

A trained interviewer collected information using a pretested questionnaire. The information obtained included demographic data, family history of diabetes and past medical history of cardiovascular disorder (CVD), drug use and smoking behavior. Weight was measured, with subjects minimally clothed without shoes, using digital scales (Seca 707: range 0.1-150 kg) and recorded to the nearest 100 g. Height was measured in a standing position without shoes, using tape meter while shoulders were in a normal alignment. Waist circumference (WC) was measured at the umbilical level. Two measurements of systolic blood pressure (SBP) and diastolic blood pressure (DBP) were taken using a standardized mercury sphygmomanometer on the right arm, after a 15 minute rest in a sitting position; mean of the two measurements was considered as the subject's blood pressure [12].

Fasting plasma glucose (FPG), serum HDL-C, and TGs levels were measured by previously reported methods [23, 24]. BMI was calculated as weight (kg) divided by square of the height (m²).

The normal-weight/MetS phenotype was defined as the status of having normal weight (BMI < 25 kg.m^-2) but meeting the MetS criteria. In contrast, obese/without MetS phenotype was defined as the status of being obese or overweight (BMI ≥ 25 kg.m^-2), but not having MetS [18]. We used the updated harmonized definition of MetS [25, 26]. We used WC cutoff points known to be appropriate for Persian men and women [27]. MetS was ascertained in individuals meeting three or more of the following criteria. (1) Waist circumference ≥ 94.5 cm. (2) HDL-C <1.04 mmol.l^-1 (40 mg.dl^-1) in men and <1.30 mmol.l^-1 (50 mg.dl^-1) in women. (3) TGs ≥1.7 mmol.l^-1 (150 mg.dl^-1) or specific treatment for this lipid abnormality. (4) Hypertension defined as SBP ≥130 mmHg or DBP ≥85 mmHg or treatment of previously diagnosed hypertension. (5) fasting glucose ≥5.5 mmol.l^-1 (100 mg.dl^-1) or previously diagnosed diabetes [28]. Smoking status included a record of current, occasional and past smoking history (those who had never smoked were called non-smokers). Positive family history of diabetes was defined as having at least one parent or sibling with diabetes. Participants were classified as having diabetes at the baseline or during follow-up if they met at least one of these criteria: FPG ≥7 mmol.l^-1, or 2 h-PCPG≥11.1 mmol.l^-1 or taking anti-diabetic medication [29]. A previous history of CVD reflected any prior diagnosis of CVD by a physician.

To investigate the sex-specific combined effect of BMI and MetS on the development of diabetes, joint BMI-MetS variable was created. That is, participants were divided into groups based on both their BMI (kg.m^-2) (normal weight, <25, overweight 25-30, and obese ≥ 30) [30] and MetS status [26]. Mean (SDs) were calculated for all continuous variables and the percentage of participants in each BMI-MetS category was determined for all categorical variables. Values for TGs levels were log-transformed since the distribution was highly skewed. Differences among BMI-MetS categories were examined by Chi-square or ANOVA where appropriate. The cumulative incidence of diabetes was calculated as the number of diabetes events divided by the number of subjects at risk in each category of BMI-MetS. The independent and combined effects on diabetes of both BMI and MetS were examined using multivariable logistic regression models separately for each sex. We chose our candidate covariates among the ones that were validated from the literature and new ones that are suspected of playing important roles in the development of diabetes [31, 32]. As such, our covariate selection can be regarded as being guided by scientific as well as numeric evidence. The following variables served as standard candidate risk factors: age, sex, body mass index (BMI), waist circumference, exercise, diabetes, SBP, DBP, smoking, family history of diabetes, TGs, and HDL-C [31]. We followed statistical guidelines with respect to the significance of association of a variable with incident diabetes but also considered scientific and qualitative judgment as well. For example we did not adjusted for waist circumference, TGs, HDL-C, and FPG which are components of the MetS and therefore not appropriate to be adjusted for in prediction models already incorporating MetS. Sex-specific odds ratios (ORs) and their 95% confidence intervals (CIs) were calculated using age- and multivariate-adjusted models, which included age, family history of diabetes, history of cardiovascular disease (CVD), education, smoking and the TLGS interventions. Age- and multivariate-adjusted logistic regression models incorporating the joint variable were also created. The normal weight individuals, without MetS were the referent group when the combined effect of the MetS and BMI were to be evaluated. When BMI was the predictor under evaluation, the reference group was normal weight individuals, regardless of MetS status. When MetS was the predictor under evaluation, the participants without MetS were considered as reference group. Hosmer-Lemeshow test for goodness-of-fit was implemented to assess the calibration of logistic models [33]. Linearity in the regression coefficients of the continuous covariates were analyzed by using multivariate restricted cubic splines [34].

Alternate analyses using Cox proportional hazards models that accounted for interval censoring gave the same results hence, only logistic regression results are presented. Wald tests of the linear hypotheses concerning the logistic regression models coefficients were performed to test the null hypotheses that the ORs (effect size) for one MetS-BMI category was equal to that of another category. All statistical analyses were performed using STATA version 10.0 (StataCorp LP, College Station, Texas) and values of P < 0.05 from 2-sided tests were considered statistically significant.