The survey was carried out within the training framework of the Clinical Nutrition class, which is presented in the third year of the medical course in the University of Crete School of Medicine 15. Between the years 1989–2001, all students registered in the third year of their studies (21–49 men and 17–47 women per year) were invited to participate in the study and sign appropriate consent forms. The average annual participation rate was 98.2% (range 95.3–100%). Eighteen students were excluded from the analyses because of missing data on potential confounding variables. The final sample included 989 students (527 males, 462 females), aged 20–40 years (mean age 22, standard deviation [SD] 2) with complete measurements for all variables other than glucose. Blood glucose was available for a sub-sample of 727 students (412 men, 346 women).

Purpose designed questionnaires were administered to ascertain biographical data, lifestyle behaviours on topics including cigarette smoking and alcohol consumption, and medical and family history of cardiovascular diseases. In order to validate the results multiple crosschecked questions on the same topic were addressed to the participants. Smokers were classified as those who stated smoking more than one cigarette per day for at least three consecutive months. Ex-smokers were defined as those who had not been smoking for the last three consecutive months, and non-smokers as those who did not fall in any of the two previous groups. The questionnaires remained unchanged over the years.

A digital scale (Seca) was used to measure body weight (BW) with an accuracy of ± 100 g. Subjects were weighed without shoes, in their underwear. Standing body height (BH) was measured without shoes to the nearest 0.5 cm with the use of a commercial stadiometer with the shoulders in relaxed position and arms hanging freely. Body mass index (BMI) was calculated as BH in kilograms (kg) divided by the square of the BH in meters (m²). Waist circumference (WC) was measured in the middle between 12^th rib and iliac crest at the level of umbilicus and the hips circumference (HC) at the fullest point around the buttocks. WC (cm) was divided by HC (cm) and BH (m) in order to calculate the waist-to-hip (WHpR) and waist-to-height (WHtR) ratio respectively. Means of replicates were used in all anthropometric measurements, which were carried out by the same faculty members over the years.

In order to measure blood pressure, subjects were seated in a chair with their back supported and their arms bared and supported at heart level. Measurement was performed with the student not having ingested coffee or smoked for 30 minutes and after at least five minutes of rest. The appropriate cuff size was used to ensure an accurate measurement. Measurements were taken using a mercury sphygmomanometer applied on the right arm of the participants. First and fifth Korotkoff sounds were recorded for systolic and diastolic readings respectively. Two readings separated by two minutes were averaged. If they differed by more than 5 mmHg, one additional reading was obtained and then averaged.

Early morning ,venous blood samples were drawn for biochemical screening tests, following a 12-hour overnight fast. The blood samples were transferred to the Nutritional Research Laboratory of the University of Crete in tanks containing ice packs so as to maintain a temperature of 3–4°C. Blood was centrifuged and 1.5 mL aliquots were pipetted into plastic Eppendorf tubes. One aliquot was used for blood analysis of triglycerides (TG), total cholesterol (TC) and high-density lipoprotein cholesterol (HDL-C) measurements on the same day of collection, while the other was stored (at -80°C). TG was determined using Fossati's method 16. TC was determined by Allain's method 17. HDL-C was measured by the heparin-manganese precipitation method 18 while low-density lipoprotein cholesterol (LDL-C) was calculated as follows: LDL-C = TC - (HDL-C + TG/5) 19. The ratio of TC to HDL-C (TC:HDL-C) was calculated. Coefficients of variation for measurement error during 1989–2001 ranged between 4.3 and 12.4% for TC and HDL-C. Blood glucose (BG) levels were measured by the exokinase method, using an auto-analyser AU-640 (Olympus Instruments).

Based on the International Obesity Task Force 20, convened by the World Health Organization, a subject with BMI of 25.0 to 29.9 kg/m² was defined as overweight; a BMI ≥ 30.0 kg/m² was defined as obese. The WC, WHpR and WHtR measurements were used to determine the extent of central adiposity. For WC, cut-off points of ≥ 90 cm in men and ≥ 80 cm in women were used 21. A WHpR ≥ 0.9 in men and ≥ 0.8 in women was considered to represent central obesity 21, and WHtR values of ≥ 50.0 in either sex were adopted as cut-offs 22.

Participants were classified as having elevated blood pressure if they reported taking anti-hypertensive medication or had a systolic blood pressure (SBP) ≥ 140 mmHg or diastolic blood pressure (DBP) ≥ 90 mmHg. High-normal blood pressure was defined as blood pressure ≥ 130/85 – < 140/90 mmHg. The cut points correspond to the Sixth Joint National Committee on High Blood Pressure criteria for high blood pressure 23.

Dyslipidaemia was defined as TC ≥ 5.2 mmol/l (200 mg/dl), TG ≥ 1.5 mmol/l (150 mg/dl), LDL-C ≥ 3.4 mmol/l (130 mg/dl), and HDL-C < 0.9 mmol/l (40 mg/dl). These cut-offs correspond to the Adult Panel Treatment III criteria for borderline high TC, TG, LDL-C and low HDL-C levels 24. A TC:HDL-C ≥ 4 was also considered a adverse serum lipid profile 25.

Impaired fasting glucose was defined as BG ≥ 6.1 mmol/l (110 mg/dl) based on the Report of the Expert Committee on the Diagnosis and Classification of Diabetes Mellitus 26.

Statistical analyses were performed using the Statistical Package for Social Sciences (version 8.0, SPSS, Inc) software. Analysis of covariance (ANCOVA) was used to examine for differences in the lipidaemic profiles of men and women, adjusting for the year of examination. The chi-square (χ²) test (or Fisher's exact test for 2 × 2 data tables) was used for categorical data. Although the set of data was collected over a 13-year-period, we did not examine for time trends since the cross-sectional design of the survey and the small number of participants per year would preclude any definite conclusions. However, in order to account for any possible time effects, the year of examination was included in all statistical analyses as an independent variable.

The relationship between obesity and other CVD risk factor variables was first examined with univariate analyses using obesity indices as categorical variables. Average differences in blood pressure, glucose, serum lipids and lipoproteins were calculated using ANCOVA with covariates age, sex, tobacco use and year of examination. Stepwise linear regression techniques were applied to assess the extent to which CVD risk factors depend on the obesity indices considered using the same independent variables. Logistic regression models were applied to calculate odds ratios (OR) for the presence of CVD risk factors after adjustment for age, tobacco use, and year of examination. A level of p < 0.05 was used to indicate statistical significance in all analyses.

The fasting blood glucose levels and lipidaemic profiles of University of Crete medical students are given in Table 1. Men had significantly higher values of BG (p < 0.001), TG (p < 0.001), LDL-C (p = 0.010) and TC:HDL-C (p < 0.001), and lower values of HDL-C (p < 0.001) than their female counterparts. Table 2 presents the distribution of obesity indices, blood pressure, blood glucose, serum lipids and lipoproteins among medical students. The prevalence of overweight or obesity (BMI ≥ 25.0 kg/m²) was 39.5% in male and 23.3% in female students. Central obesity levels were significantly higher in men than women, using the WC (32.6% vs. 21.9%, p < 0.001) or WHtR (37.6% vs. 12.8%, p < 0.001) cut-offs. Forty eight percent of men and 36.1% of women (p < 0.001, difference between the genders) had blood pressure levels ≥ 130/85 mmHg. A significantly higher percentage of males than females had high TG, LDL-C, TC:HDL-C, or low HDL-C (p < 0.001 for all analyses). Only 10 students (8 men, 2 women) had BG levels ≥ 6.1 mmol/l (110 mg/dl).

The relationships between blood pressure, glucose, serum lipids and lipoproteins, and obesity indices treated as categorical data are presented in Table 3. Across the four BMI categories, all CVD risk factor variables except BG demonstrated a significant increase. For WC and WHtR, subjects above the selected cut-off points had significantly higher values of SBP, DBP, TC, TG, LDL-C, TC:HDL-C and lower values of HDL-C. High WHpR values were also related to adverse serum lipid profile.

The linear regression analyses showed that BMI was the best obesity index predicting SBP and DBP in male students (Table 4). The percentage of the variance of these two measurements explained was <5%. WC was also a significant, though not powerful, determinant of SBP. WHtR could better predict most of the serum lipoproteins, explaining 6.2% of TC, 10.6% of TG, 4.2% of LDL-C and 7.3% of TC:HDL-C variance. WHpR was a significant determinant of HDL-C. The same analyses within the female sample showed also that BMI was the most important predictor of blood pressure, explaining 11.9% of the SBP and 8.4% of the DBP variance (Table 5). WHtR could better predict LDL-C levels, and WC was the most important determinant of TG, HDL-C and TC:HDL-C, predicting 7.9, 7.6 and 10% of their variance respectively. In both genders, none of the obesity indices could significantly predict BG levels.

Using the odds ratios for the prediction of the presence of CVD risk factors, subjects above the selected cut-offs had higher risk for high blood pressure and dyslipidaemia (Table 6). Men with WHtR ≥ 50 had the highest risk for elevated TC (OR 2.26) and LDL-C (OR 1.83). Those with WHpR ≥ 0.9 had significantly higher risk for TG ≥ 1.5 mmol/l (OR 2.51), HDL-C < 0.9 mmol/l (OR 2.03), and TC:HDL-C ≥ 4 (2.78). Women with WHtR ≥ 50 had the highest odds for high DBP (OR 2.33) and LDL-C (OR 2.60), and TC:HDL-C ≥ 4 (OR 5.30). The risk for low HDL-C was significantly higher in female students with WHpR ≥ 0.8. Nevertheless, the odds ratios among the obesity indices differed only slightly and their 95% confidence intervals overlapped.