Study design
The present cross-sectional analysis combined two samples from the pre-test phase of the German National Cohort Study (GNC) [24]. The first sample was derived from the “P1 project”, a feasibility study from November 2011 until May 2012, aiming to improve study recruitment among Berliners of Turkish descent [25]. The second sample was a control group recruited from April to May 2012 among an unselected sample of Berliners, excluding those with Turkish roots.
Recruitment strategy and participants
Details of the recruitment strategy of the “P1 project” were published by Reiss et al. [25]. Briefly, half of the P1 participants and the whole control sample were recruited as a random sample drawn from the residents’ registration office, whereas the other half of the P1 participants were recruited via Turkish networks in Berlin. General practitioners all over Berlin were informed about the study.
For the first recruitment strategy, randomly chosen persons received an invitation letter including study information and a response form. Up to two reminder letters were sent out in case of non-response. If a phone number was available, potential participants were contacted by phone.
To identify persons with Turkish origin, an onomastic procedure (i.e. selection of individuals based on the origin of their first or family names) was applied for the “P1 project” sample by the Robert Koch-Institute, the national Public Health institute in Germany. Potential participants received all study material in both the German and Turkish language. Bilingual study assistants, mostly medical students, conducted up to three home visits on different days and times. If a potential participant was not at home, an information card with contact details was left at the home address.
For the community-based recruitment approach, key persons in Turkish networks were invited to provide information about the study and to participate in focus groups. Bilingual study staff distributed study materials in migrant settings (shops, doctor’s offices, religious and educational institutions, and societies) and invited people to the study.
Inclusion criteria were 20–69 years of age, the ability to understand the aims of the study, and a signed informed consent (in German or Turkish). All participants received a 15€ expense allowance to cover travelling costs and a letter with results from the study center visit by post.
The study was approved by the ethical review committee of the Charité-Universitätsmedizin Berlin, Germany.
Data collection
Questionnaire
The modified health questionnaire from the German Health Interview and Examination Survey was used to assess socioeconomic status, medical history, health behavior, physical and mental diseases and migration background [26]. A participant was defined as being of Turkish descent if (i) he/she or (ii) at least one parent was born in Turkey [27]. Participants with Turkish roots could choose between self-completed questionnaire versions in the German or Turkish language. In case of difficulties understanding the questions, dyslexia (i.e. reading difficulties) or presbyopia (i.e. age-related reduced ability to focus on near objects), bilingual staff was present in the study center to conduct the questionnaire as a face-to-face interview. Among the control sample the questionnaire was applied as an interview by trained study staff according to standard operating procedures (SOP).
Measures
Physical examinations
In the “P1 project”, weight, height, circumference of hip and waist, blood pressure and heart rate were measured by trained physicians and nurses according to SOPs; in the control sample, weight and height were assessed by self-report.
Blood analyses
A blood sample was taken by trained physicians or nurses according to SOPs. The blood samples were analyzed for 25(OH) D, full blood count, liver enzymes, kidney function, cholesterol levels, and blood glucose. Vitamin D was measured as 25(OH) D in serum via chemiluminescence immunoassay by the “hospital-laborverbund Brandenburg-Berlin”. Participants received their results as postal letter two weeks after their visit in our study center.
Covariates
Physical activity was dichotomized into categories of 150 min or more vs less than 150 min physical activity per week according to WHO recommendations [28]. Body mass index (BMI) was calculated with normal weight being defined as a BMI between greater than 18.5 and less than 25 kg/m2, overweight as BMI from 25 to less than 30 kg/m2 and obesity as a BMI of 30 kg/m2 and higher. Alcohol consumption was categorized into no consumption, moderate consumption (up to 12 grams/day (g/d) for women and up to 24 g/d for men) and hazardous consumption (> 12 g/d for women and > 24 g/d for men). Smoking was dichotomized in never or ex-smoker versus smoker. A person was defined as a smoker, if he or she indicated smoking daily or occasionally, while ex-smokers were defined as former smokers who quit smoking at least one year ago.
Statistical analysis
We computed descriptive statistics for sociodemographic, lifestyle and medical characteristics using mean with standard deviation (SD) and median for continuous variables and absolute and relative frequencies for categorical variables.
Potential influences on the serum vitamin D concentration as the primary outcome were evaluated using univariable and multivariable linear regression models that were stratified for study sample. Means with 95% confidence intervals (95% CI) are presented from univariable regression in the subgroup of women with Turkish background. We used a stepwise backwards approach for the selection of the co-variables in the multivariable linear regression model with a limit for inclusion of p < 0.2. In the final model for the sample with Turkish background the variables BMI, physical activity and sex were included. In the final model for the sample without Turkish background, BMI and sex were included. Resulting Beta-estimates are presented with 95%-CI and p-value.
As a sensitivity analysis we conducted for each study sample stratified univariable and multivariable logistic regression analyses with the binary outcome variable “vitamin D deficiency” based on the recommendations by the Institute of Medicine [29]. Crude odds ratios resulting from univariable logistic regression were estimated for the factors mentioned above. In the multivariable logistic regression model, we used a stepwise backwards approach for the selection of the co-variables with a limit for inclusion of p < 0.2. In the final model for the sample with Turkish background the variables age, physical activity and chronic disease were included. All odds ratios are presented with 95%-CIs. For the sample without Turkish background the variables sex, BMI, physical activity, smoking status and chronic disease were included.
All analyses were explorative rather than strictly hypotheses testing. Missing data were not imputed. All analyses were performed using The SAS System®, Version 9.4 under Windows operating system.