The main characteristics of the five population groups (N = 4110) are given in Table . The ethnic Norwegians were oldest and subjects from Sri Lanka youngest. Socioeconomic factors differed markedly between the ethnic groups, more for women than for men. Tamils from Sri Lanka had the lowest proportion with low education and subjects from Turkey the highest, and the latter had the lowest proportion with income-generating work. Ethnic minority men had longer duration of residence in Norway than women. Subjects from Pakistan and Turkey had longer residence than those from Vietnam and Sri Lanka.
Known diabetes was reported by 238 subjects (111 women/127 men), of these 18 (6.8%) were diagnosed before the age of 25 years. The total number with diabetes was 406 (176 women/230 men) when including the survey-detected cases constituting 37% of cases in women and 45% in men. The age-standardized diabetes prevalence (self-reported and total in Figure ) and adiposity variables (Table ) differed markedly between the ethnic groups. The highest prevalence of diabetes was found in subjects from Sri Lanka and Pakistan and the lowest in Norwegians. Subjects from Turkey had the highest BMI and the highest proportion with BMI ≥ 30 kg/m2, but subjects from Pakistan and Sri Lanka had the highest WHR and a substantially higher proportion with BMI ≥ 25 kg/m2 and BMI ≥ 30 kg/m2 compared to the Vietnamese. Nearly all women from Sri Lanka/Pakistan (85-92%) and 49% of women from Vietnam were overweight using the ethnic specific criteria (BMI ≥ 23 kg/m2), and the majority of female subjects from Sri Lanka/Pakistan (66-82%) but only 25% of female subjects from Vietnam were obese (BMI ≥ 25 kg/m2). Mean WSR was highest in subjects from Pakistan and Turkey and lowest in subjects from Vietnam. Applying the standard BMI definition for obesity (BMI ≥ 30 kg/m2) in the other female groups, 56% from Turkey and 19% from Norway were obese. All ethnic minority groups performed less heavy physical activity in leisure time, compared with Norwegians. No differences in age, years of education, self-reported diabetes or BMI were found between those included and the 515 subjects excluded from The Immigrant Health Study.
The age-standardized prevalence of diabetes was higher in the ethnic minority groups than in Norwegians for any level of BMI, WC and WHR when used as continuous variables except for men from Turkey in the lowest range of the adiposity variables (data not shown). When the adiposity measures BMI, WC and WHR were categorized (Figure ), the age-standardized prevalence of diabetes increased more with increasing levels of these variables in ethnic minorities than in Norwegians. The age-standardized prevalences within each category of the adiposity measures were significantly different between the groups in both genders (all p-values < 0.001, likelihood ratio tests). Subjects from Norway with BMI ≥ 30 kg/m2 had lower prevalence of diabetes than subjects from Sri Lanka and Pakistan with BMI 25-30 kg/m2. Lower prevalence in Norwegians was also found when applying higher cut-off values for WC for Norwegians (women: ≥ 88 cm, men: ≥ 94 cm) than for subjects from Sri Lanka and Pakistan (women: WC ≥ 80 cm, men: ≥ 90 cm). For the other ethnic groups the diabetes prevalence rates were higher than for the Norwegians in each adiposity category, but as there were few cases in some groups (Turkey: low adiposity groups, Vietnam: high adiposity groups), the confidence intervals were wide. However, the prevalence of diabetes in Vietnamese women was higher than in Norwegian women for all categories of WC and for those with WHR ≤ 0.85.
Before performing logistic regression analyses (Table , Figure ), all continuous variables were standardized to allow for comparison. Overall, we found significant interactions between ethnicity and gender (p = 0.0003), and ethnicity and BMI (p = 0.017). In the gender specific logistic regression analyses adjusted for age, all adiposity measures were significantly associated with diabetes except in women from Turkey (Table ). Ethnic differences in the OR for diabetes were only found for body height and income-generating work in women. As the age-adjusted OR for diabetes was highest for WHR in all ethnic subgroups except the Vietnamese, this anthropometric measure was used in the subsequent multivariate models.
We assessed whether the ethnic differences were significant after adjustments in Model 1: age, Model 2: age and adiposity (WHR) and Model 3: age, adiposity and SEP (body height, education and income-generating work), with Norwegians as the reference group (Figure ). OR for diabetes was significantly increased for the four minority groups for both genders after adjusting for WHR and age. In women, the increased OR for diabetes persisted after further adjustment for all the SEP factors (Model 3) (Turkey: 2.9 (95% CI 1.30-6.36), Vietnam: 2.7 (1.29-5.76), Sri Lanka: 8.0 (4.19-15.14), Pakistan: 8.3 (4.37-15.58). No SEP variables were independently related to diabetes in women in Model 3.
In men, having no income-generating work was independently related to diabetes (OR 1.6 (1.10-2.27)) along with age (2.0 (1.59-2.40)), WHR (1.9 (1.49-2.29)) and ethnicity (see Figure ). For men from Sri Lanka and Pakistan the increased OR compared with Norwegians persisted after adjusting for all SEP variables, with identical OR (3.0 (1.80-5.12)). For men from Turkey, however, when adding any of the SEP indicators into the model, the increased OR for diabetes compared with Norwegians was no longer significant. For men from Vietnam the increased OR for diabetes persisted when adding either income-generating work (1.8 (1.04-3.20)) or education (2.5 (1.47-4.23)), but when adding body height no ethnic difference was found (1.8 (0.96-3.40)).
Applying WC or WSR instead of WHR did not explain more of the ethnic differences, nor did adding any of the two physical activity variables, smoking, duration of residence in Norway and parity (for women) to the model. No significant interactions were found in the full multivariate model.