The study population consisted of participants in the SUNSET study (Surinamese in the Netherlands: Study on health and Ethnicity) [19]. In 1975, almost half the population of the former Dutch colony Surinam migrated to the Netherlands. Approximately 80% of these Surinamese immigrants in the Netherlands are Hindustani ('South Asian', originally from the Indian sub-continent) or African (mixed African, Indian and European, but predominantly of African origin). The SUNSET study is based on a random sample of 2975 individuals, aged 35 to 60 years of age, drawn from the approximately 389000 ethnic Dutch (Dutch) and 72000 Surinamese listed in the Amsterdam population register (figure 1). For the sampling procedure, persons who were born in the Netherlands and whose parents were both born in the Netherlands were presumed to be Dutch. Persons of whom both parents were born in Surinam and persons who were born in Surinam and who had at least one parent who was born in Surinam were presumed to be Surinamese participants.
Between 2001 and 2003, all persons in the sample were approached for face-to-face, structured interviews by trained interviewers who had been matched by sex and presumed ethnicity. The interview included questions on self-identified ethnicity, migration history, demographic variables, lifestyle, and health status. If information on the self-identified ethnicity of the individual was lacking, the origin of the mother, the father and the mother's ancestors were used to classify participants.
The overall response to the interview was 60% (figure 1). Participation rates were higher among women than among men. In addition, participants in the interview were more likely to be married and living with a partner and/or children, and to live in a less urban area (address density of 1500–2500 addresses/km2 vs. ≥ 2500) as compared to non-participants (both non-response and not eligible). However, the absolute and relative differences between participants and non-participants for these characteristics were small and reported trends were similar across ethnic groups (data not shown).
Participants of Hindustani Surinamese, African Surinamese or Dutch origin were also invited for a physical examination at a local health care centre. During the examination, trained physicians recorded the following characteristics: weight in light clothing on a SECA mechanical scale to the nearest 200 grams and height to the nearest 0.01 meter by wall tape measure. Waist circumference midway between the lower rib margin and the iliac crest and hip circumference at the maximum point over the greater trochanters were determined to the nearest 0.01 meter by tape measure. After the subjects had emptied their bladder and had been seated for at least 5 minutes, blood pressure and resting heart rate measurements were obtained from each subject's arm at heart level using an OMRON-M4 semi-automatic sphygmomanometer with an appropriate-sized cuff. All anthropometric measurements were obtained twice and the means (rounded off to the nearest integer) were used for analysis. Fasting glucose (HK/Glucose-6-P dehydrogenase test; Roche Diagnostics, In), high density lipoprotein cholesterol (HDL, Homogenous enzymatic colorimetric test; Roche Diagnostics, In) and triglyceride (GPO-PAP Enzymatic test; Roche Diagnostics, In) levels were determined in serum samples obtained at the time of the physical examination. DM was defined as fasting glucose ≥ 7.0 mmol/l and/or self-reported DM, excluding the self-reported diagnoses of gestational diabetes.
The SUNSET-study was approved by the Institutional Review Board of the Academic Medical Centre of the University of Amsterdam, and carried out in compliance with the Helsinki Declaration. All participants provided a written informed consent.
In the present analysis, we included participants who had participated in both the interview and the physical examination. Of all participants in the interview, 71 were excluded due to missing information on self-identified ethnicity, 182 persons were excluded because they had not undergone the physical exam and 10 persons because fasting glucose measurements were not available (non-response for blood sample). As compared to those who were left in the study, those excluded were similar with regard to gender, self-reported DM and self-rated health (data not shown).
In total, 1434 participants remained in the study, divided into 339 Hindustani Surinamese, 605 African Surinamese, and 490 Dutch (figure 1). Of the Hindustani participants 98.8% were born in Surinam, 99.4 had two parents who were born in Surinam and 92.1% had two parents who were of Hindustani origin. Of the African Surinamese participants, 99.2% were born in Surinam and 99.5% had two parents who were born in Surinam. Moreover, 79.3% of the African Surinamese had two parents who were of African origin.
Characteristics of the ethnic groups were described using means or proportions. In addition, the prevalence of determinants of DM was calculated, directly standardised to the age distribution of the total population. The association of determinants with DM was studied using univariate logistic regression analysis. All variables showing an association of p ≤ 0.25 for the Wald test were selected for the multivariate analyses. Stepwise multiple logistic regression was then performed to construct the optimal risk score for the occurrence of DM among Hindustani Surinamese, African Surinamese and Dutch. Criteria for entry into and exclusion from the model were a p-value for the likelihood ratio test of 0.05 and 0.10, respectively. A new risk score was developed to determine the probability of having DM with a logistic regression model using data that would be routinely available in general practice. Variables considered were ethnicity, biomedical parameters and disease history, e.g. age, BMI, waist circumference, resting heart rate, first-degree relative with DM, hypertension, history of CVD. The risk score is based on the sum of the score of the variables included in the full model (see additional file 1: Risk score for DM SUNSET.pdf).
Subsequently, we evaluated the performance of the risk score and compared it to the sets of screening criteria derived from current guidelines by calculating the area under the Receiver-Operator Characteristic curve (AUC) as a measure of diagnostic accuracy. Before analyzing the data, it was decided to consider an AUC of less than 0.60 to be poor, 0.60–0.75 to be moderate, and higher than 0.75 to be good.
The cut-off for the risk scores at which fasting plasma glucose screening was indicated, was chosen such that the sensitivity was approximately 80%, but not over. Additionally, we determined the specificity, the total population selected for screening, the prevalence in the screened population (the positive predictive value), and the number needed to screen to detect a case of DM (NNS).
Finally, we estimated the diagnostic accuracy of the risk scores for the detection of a new case of DM, to simulate a situation where persons with known DM are excluded from screening. This was done because, ideally, only previously unknown cases would have been used in the derivation of the risk score for screening for unknown DM. However, given issues of power to enable statistical modelling, it was decided to base the score on all cases and to then also estimate the diagnostic accuracy of the risk scores for newly detected DM.
We assessed the ability of a simplified version of our risk score, with points corresponding to the calculated odds ratio, to detect new cases of DM by calculating the number needed to screen to identify a new case of newly detected DM (NNSnew).
To assess the validity of the risk scores, we used bootstrapping techniques to estimate a 'confidence interval' (bCI) around the estimated AUC of all risk scores in our population. We took 1000 random samples, with replacement, from the study population. At each step the parameters in the predictive models were calculated. We subsequently estimated the AUC for each of the models and calculated the 2.5 and 97.5th percentile to indicate the 'confidence interval'[20].
All analyses were performed using the the SAS package, version 9.1 (SAS Institute Inc., Cary, USA.).