Validation of the German version of the Mediterranean Diet Adherence Screener (MEDAS) questionnaire

Health benefits of the Mediterranean Diet (MD) have been shown in different at-risk populations. A German translation of the Mediterranean Diet Adherence Screener (MEDAS) from the PREvención con DIeta MEDiterránea (PREDIMED) consortium was used in the LIBRE study, investigating effects of lifestyle-intervention on women with BRCA1/2 mutations. The purpose of the present study is to validate the MEDAS German version. LIBRE is a multicentre (three university hospitals during this pilot phase), unblinded, randomized, controlled clinical trial. Women with a BRCA1/2 mutation of age 18 or over who provided written consent were eligible for the trial. As part of the assessment, all were given a full-length Food Frequency Questionnaire (FFQ) and MEDAS at baseline and after 3 months. Data derived from FFQ was compared to MEDAS in order to evaluate agreement or concordance between the two questionnaires. Additionally, the association of dietary intake biomarkers in the blood (β-carotene, omega-3, omega-6 and omega-9 fatty acids and high-sensitivity C-reactive protein (hsCRP)) with some MEDAS items was analyzed using t-Tests and a multivariate regression. The participants of the LIBRE pilot study were 68 in total (33 Intervention, 35 Control). Only participants who completed both questionnaires were included in this analysis (baseline: 66, month three: 54). The concordance between these two questionnaires varied between the items (Intraclass correlation coefficient of 0.91 for pulses at the highest and −0.33 for sugar-sweetened drinks). Mean MEDAS scores (sum of all items) were 9% higher than their FFQ counter-parts at baseline and 15% after 3 months. Higher fish consumption (at least 3 portions) was associated with lower omega-6 fatty acid levels (p = 0.026) and higher omega-3 fatty acid levels (p = 0.037), both results being statistically significant. We conclude that the German MEDAS in its current version could be a useful tool in clinical trials and in practice to assess adherence to MD. ClinicalTrials.gov, registered on March 12, 2014, identifier: NCT02087592. World Health Organization Trial Registration, registered on 3 August 2015, identifier: NCT02087592.


Background
The MD has been tested for its health benefits in different at-risk populations with favourable results. For instance, randomized controlled intervention studies revealed that MD is effective in the primary prevention of cardiovascular diseases [1,2], in lowering hypertension and atherogenic lipoproteins [3,4] and in improving diabetes [5]. More recently, studies in the older population showed an association between MD and improved cognitive function [6]. Epidemiological studies from the European Prospective Investigation into Cancer and Nutrition (EPIC) cohort suggested further that MD might protect against cancer, especially gastric cancer [7], colorectal cancer [8] and bladder cancer [9].
Such trials raised the need for a useful tool to assess MD adherence in study populations. The PREDIMED consortium established a 14-point MEDAS questionnaire, which was validated by administering the established FFQ [10] and MEDAS to 7146 participants from the PRE-DIMED study. The authors found that the average MEDAS score estimate was 105% of the FFQ PREDIMED score estimate, and thus is a valid instrument for rapid estimation of adherence to the MD [11]. Moreover, the PRE-DIMED investigators could show that MEDAS is able to capture a strong monotonic inverse association between adherence to MD and obesity indices in a population of adults with a high cardiovascular risk [12]. In Germany, a multicentre trial, the 'Lifestyle intervention in BRCA 1/2-mutation carriers' (LIBRE) was launched to investigate the effect of a defined lifestyle intervention on breast cancer incidence in women at high genetic risk for this type of cancer [13]. Up to 60% of women with BRCA mutations develop breast or ovarian cancer [14], but not all of them, suggesting that environmental cofactors play a role. Indeed, some studies suggested that physical activity and dietetic intervention help prevent cancer, including breast cancer [15,16]. To test the hypothesis of whether these controllable environmental factors further modulate cancer risk, the LIBRE trial conducts an intervention with clearly defined sport and nutrition components. The nutritional component of the intervention was based on the MD. The German translation of MEDAS was used as an instrument to assess adherence. The purpose of the present study was to validate the German version of MEDAS.

Study population
The LIBRE study is divided into two parts: firstly, a feasibility study to prove the practicability of the lifestyle intervention and consequently, the presently recruiting main trial with the aim of attaining 600 study participants to demonstrate the effects of the lifestyle intervention on the breast cancer incidence in women with BRCA1 or BRCA2 genetic mutations. 68 women, who were all participants of the LIBRE feasibility study, formed the study population for these analyses. The details of the trial have been published elsewhere [13]. The study population (adult women between 18 and 75 years) was recruited from February 2014 to July 2014 in selected consultation centres of the German Consortium of Hereditary Breast and Ovarian Cancer in Kiel, Cologne and Munich. All participants signed an informed written consent. The trial was approved by the responsible ethical committees.
Study participants were randomized into two groups with a ratio of 1:1, stratified by participating centre and previous breast cancer. The intervention group (IG) (n = 35) received a detailed lifestyle intervention over 12 months, and the control group (CG) (n = 33) received no intervention, but standard recommendations for a healthier lifestyle. The lifestyle IG received a supervised physical exercise training program and nutritional education based on the MD. In the first 3 months, the nutritional education took place every fortnight, thereafter at monthly intervals. The CG received minimal nutritional education based on the recommendations of the German Society of Nutrition (DGE -https:// www.dge.de/en/) "Usual Care in Germany"and some general advice for increasing activity in everyday life at the beginning of the study (one session). All participants were asked to fill out both a full-length FFQ and MEDAS at study start (baseline) and 3 months later. We chose both time points to prove whether MEDAS is specific enough by measuring MD-typical changes during the intervention. Only data collected within this period were used for the purposes of the current study. Solely participants who had completed both questionnaires were included in the analysis. These were in total 66 participants at baseline (34 in the IG and 32 in the CG) and 54 at month three (27 in the IG and 27 in the CG).
We used the guidelines from the International Society for Pharmacoeconomics and Outcomes Research (ISPOR) to guide our translation process [17]. MEDAS from the English PREDIMED-publication [11] was translated into German and reviewed by two native speakers in German. It was then translated back into English by a native speaker. Following this, the final version was read and approved in a small group of the study team.

Dietary assessment
MEDAS is a 14-item screener, which consists of 12 questions on food consumption frequency and 2 questions on food intake habits characteristic of the MD (Table 1). Each question was scored with a 0 or 1. One point was given for using olive oil as the principal source of fat for cooking and one for preferring white meat over red meat, and one for consuming each of the following: 4 or more tablespoons (1 tablespoon = 13.5 g) of olive oil/d (including that used in frying, salads, meals eaten away from home, etc.); 2 or more servings of vegetables/d; 3 or more pieces of fruit/d; fewer than 1 serving of red meat or sausages/d; fewer than 1 serving of animal fat/d; fewer than 1 cup (1 cup =100 mL) of sugarsweetened beverages/d; 7 or more servings of red wine/wk.; 3 or more servings of pulses/wk.; 3 or more servings of fish/wk.; fewer than 3 commercial pastries/wk.; 3 or more servings of nuts/wk.; or 2 or more servings/wk. of a dish with a traditional sauce of tomatoes, garlic, onion, or leeks sautéed in olive oil.
If the condition was not met, 0 points were recorded for the category. The MEDAS score (sum of above items) ranged from 0 to 14 points [11].
All participants were also asked to complete a 148item semi-quantitative FFQ. The German version had been validated by the German EPIC investigators [18][19][20][21]. For each item it questions the average serving size, described by photos, and the food frequency during the previous 12 months. Furthermore, it contains questions on cooking oil, frequency of the use of gravy, the fat content of dairy products, the use of sugar and milk in coffee and tea, and the seasonal consumption of fruit and vegetables. The documentation of the questionnaire was done via the study-management-system for Epidemiology and Public Health, which was developed and supervised by the Department of Epidemiology of the German Institute of Human Nutrition Potsdam-Rehbruecke.
Food intake data recorded by FFQ was grouped into the food-based dietary components of MEDAS (Table 1). We validated the dietary assessment data retrieved from MEDAS by comparing it with the data gathered from the validated FFQ and confirmed this by associating with the results from the blood tests. 13. Do you prefer to eat chicken, turkey or rabbit instead of beef, pork, hamburgers, or sausages?

Measurement of dietary intake biomarkers in the blood
To confirm whether MEDAS' tendency towards the MD is consistent, we selected specific biomarkers in the blood which are described in the literature to be associated with consumption of certain MD food components [1,22,23]. Blood samples were taken at the same time points as the completion of both questionnaires (baseline and after 3 month). Following 30 to 60 min of incubation the serum was centrifuged at 3000*g for 10 min in the consultation centres and was overnight delivered chilled together with the EDTA (ethylene diamine tetraacetic acid) blood samples to the central laboratory at the University of Hohenheim.
A part of the serum was passed directly to an external laboratory (Medizinisches Labor Bremen, Bremen, Germany) to measure β-carotene by high performance liquid chromatography (HPLC). The rest of the serum aliquots were stored at −80°C in Hohenheim until the measurement of hsCRP using a sandwich Enzyme Immuno Assay (K 9710S, Immundiagnostik AG, Bensheim, Germany) was done. The erythrocyte membrane was isolated from the EDTA blood and also stored at −80°C in Hohenheim until the fatty acid profile (omega-6-, omega-3-and omega-9-fatty acids) was analyzed after acid esterification using gas chromatography/ mass spectrometry by the Institute of Food Chemistry of the University of Hohenheim [24,25].

Statistical analysis
Patient characteristics were analyzed descriptively, split by study arm. For age and Body Mass Index (BMI), a t-Test was used to determine whether the two groups were statistically different. For all other characteristics, coded as binary items, a Chi 2 test was used.
We then determined the concordance between the answers to the MEDAS questionnaire compared to the answers for corresponding questions in the FFQ questionnaire both at Baseline and at 3 months. First, the absolute agreement in percentage was calculated, which was further investigated using Cohen's kappa (κ) and the intraclass correlation coefficient (ICC). The relative agreement between the corresponding items was examined using the Pearson product-moment correlation.
The agreement between the sum from the MEDAS questionnaire and the equivalent FFQ questions was examined using a Bland-Altman analysis. The mean of the two values was plotted on the x-axis and the difference on the y-axis, in order to determine possible bias. The 95% limits of agreement lines, defined as the mean difference ± 1.96 times the standard deviation of the differences, were also plotted. A linear regression was then carried out with the difference as the dependent value, whose line was added to the plot with its corresponding formula and p-value.
In a further step, we validated whether the MEDAS questionnaire can specifically determine adherence to a MD, the association of blood values for β-carotene, the fatty acids omega-3, omega-6 and omega-9 and hsCRP, and MEDAS items (β-carotene associated with the MEDAS item regarding vegetables and fruits; hsCRP, omega-3 fatty acids, omega-6 fatty acids and omega-9 fatty acids associated with the MEDAS item regarding olive oil; omega-3 fatty acids and omega-6 fatty acids associated with the MEDAS item regarding red meat, fish and nuts). We first applied a t-Test for independent groups (control or intervention) for each of the food items and each of the dietary biomarker values, carried out separately for each of the two time points. Following this, we carried out a multivariate linear regression on the associations described above, where we also adjusted for the study arm.
The statistical analysis was done using R (program for statistical computing) in the R Studio environment Version 0.99.902.

Patient characteristics
Patient characteristics at baseline are outlined in Table 2. The IG comprised 35 women at this point in time, while the CG comprised 32. Considering attributes such as BMI, age and history of breast cancer, these groups did not differ statistically significantly from one another. Both groups included a number of smokers (11% in the IG and 9% in the CG). A vegetarian diet was also followed by a group of the study participants (6% in the IG and 13% in the CG).

Item by item agreement
The absolute and relative agreements between the FFQ and the MEDAS questionnaires were calculated at baseline and at 3 months for the whole sample (Table 3). This concordance at Baseline was highest for questions 1: olive oil as principal source of fat (Pearson's productmoment correlation 0.70, κ 0.70 and an ICC of 0.68), and 12: nuts (Pearson's product-moment correlation 0.72, κ 0.70 and an ICC of 0.68). After 3 months the highest concordance was obtained for questions 9: pulses (Pearson's product-moment correlation 0.86, κ

MEDAS total score agreement
The MEDAS score was analyzed using a Bland-Altman plot (Fig. 1). The mean MEDAS scores were 15% higher than the FFQ score at baseline and 23% higher after 3 months, with the median MEDAS score being higher  Fig. 1 Bland-Altman plots showing the differences between the MEDAS score aggregated from MEDAS and FFQ questionnaires at baseline and after 3 months. The red dashed and dotted lines indicate the mean bias and its 95% confidence interval. The blue dashed and dotted lines depict limits of agreement and their 95% confidence intervals than the FFQ equivalent by 1 point at baseline and 2 points after 3 months. This corresponds to a mean difference of 1.27 at baseline and 2.12 after 3 months, which defines the bias towards higher score sums being obtained by the MEDAS questionnaire. Linear regression analysis revealed a significant increase of the bias with increasing score values (p < 0.001 for baseline, p < 0.001 after 3 months).

Measurement of dietary intake biomarkers in the blood
The possible association between laboratory measurements and intake of particular food groups was first analyzed on a per item basis (Table 4). After the 3-month MD intervention some of these associations showed a statistical significance or near-significance. Consumption of at least 3 portions of fish per week was associated with lower omega-6 fatty acid levels (p = 0.035) and higher omega-3 fatty acid levels (p = 0.053). Consumption of at least 3 portions of fruit per day was associated with higher levels of β-carotene (p = 0.056). Consumption of at least 2 portions of vegetables per day was associated with higher levels of β-carotene (p = 0.004). We have depicted these associations in Fig. 2.
We also examined the same associations in a multivariate model, as reported in Table 5. Consumption of at least three portions of fish per week showed increased levels of omega-3 (p = 0.037) and decreased levels of omega-6 fatty acid (p = 0.026) in the blood, both of which were statistically significant.

Discussion
MEDAS was developed for the Spanish PREDIMED study to expediently determine adherence to the MD and allow an immediate feedback to the patient. This short screener is a validated tool for the rapid assessment of adherence to the MD [11], which is why it was decided to use a German-translation of this questionnaire in the LIBRE study. However, the original MEDAS was not in German. To validate the German translation of this questionnaire, we used the validated German fulllength FFQ as reference.
In general agreement between FFQ and MEDAS questionnaires was of a fair or better level (0.4 and larger values for agreement coefficients) for about half of the MEDAS questions [22]. These differences are likely due to the way FFQ is structured and to the fact that it is differently analyzed. FFQ measures the food frequency of a selected list of German foods with standardized portion sizes for the previous 12 months. These answers are then used to calculate intake for food groups while MEDAS directly asks for the habits and consumption frequency of specific amounts of specific Mediterranean foods during the previous week. In short, the FFQ measures specific details whereas the MEDAS is measuring for components of an overall dietary pattern. For the conversion of food intake data from FFQ into MEDAS food groups only aggregated FFQ data can be used in most cases. For instance, question 3 on vegetables and question 4 on fruits each comprised several questions from FFQ. A further example is that FFQ provides, 3no information on how many times boiled vegetables, pasta, rice, or other dishes with a sauce of tomato, garlic, onion, or leeks sautéed in olive oil are consumed. It asks only about the consumption of tomato sauce, but no further details (e.g. way of cooking or ingredients). Only question 9 (pulses) which has a high concordance is based on a direct answer of FFQ. Additionally, in this study FFQ was completed again within after an interval of just 3 months, which means that the answers for the two time points for FFQ in our analysis overlap for a period of time.
The Bland-Altman analysis showed that the MEDAS score yields higher values for the sum of all items than FFQ with respect to MD. This confirms the results of Schröder et al., who report that the average MEDAS Mediterranean diet score estimate was 105% of the FFQ PREDIMED score estimate [11]. The larger difference between the two score sets after 3 months can be a result of the MD-based intervention in the IG. The MEDAS  questionnaire was specifically developed to detect the adherence of an individual's diet to the principles of the MD. Therefore, MEDAS is more sensitive for MD items than FFQ, resulting in higher scores in individuals abiding by the MD. The FFQ questions, however, assesses the general food intake and does not focus particularly on the consumption of Mediterranean food. Two points can be considered relevant for detecting an adequate implementation of MD in life.
To confirm if the MEDAS' tendency towards the MD is consistent, we analyzed the association between certain MD food items and selected dietary intake biomarkers in the blood thought to be associated with these food items. The traditional MD is characterized by a high intake of olive oil, fruit, nuts, vegetables, and cereals; a moderate intake of fish and poultry; and a low intake of dairy products, red meat, processed meats, and sweets; while wine is consumed moderately and only together with meals [1,23]. In the literature, it is described that a high consumption of fruits and vegetables is associated with higher β-carotene blood levels [26]. Kitamura et al. have previously positively correlated frequency of vegetables and fruit intake with β-carotene, among other things [26]. In our study, subjects who consumed at least two portions of vegetables per day according to the MEDAS questionnaire had higher β-carotene blood levels than those who consumed fewer than two portions at both time points. Those who consumed at least three portions of fruit per day according to the MEDAS questionnaire also had higher β-carotene levels in their blood than those who consumed fewer than three portions. This data supports that the MEDAS results reflect the reported intake of particular nutrients charachterisic of a MD pattern.
The MD is rich in poly-and mono-unsaturated fatty acids due to the high consumption of olive oil, fish and nuts [27]. The amount of saturated fatty acids in the MD is lower than in the Western-style diet, because red meat and processed meat products play a minor part in the Mediterranean nutrition. The Western-style diet is characterized by its highly processed and refined foods and high contents of sugars, salt and fat and protein from red meat [28]. Olive oil is characterized by a high content of mono-unsaturated fatty acids. Oleic acid (C18:1, n-9) is the main component of olive oil [27]. Therefore, we hypothesized that a high consumption of olive oil, fish and nuts and low red meat intake are associated with changes in the fatty acid profile measured in erythrocyte membrane. Barcelo et al. described elevated values of omega-3-fatty acids and low values of omega-6-fatty acids following high olive oil consumption, while the omega-9-fatty acid amount remained unchanged [29]. Our data demonstrated that, more than four tablespoons of olive oil per day were associated with a tendency to higher serum levels of all unsaturated fatty acids (omega-6, −3 and −9) compared with the values measured in subjects who consumed less olive oil. Takkumen et al. described an association between high fish consumption and a change in the omega-6 and −3-fatty acids profile. The amount of omega-6-fatty acids decreased while that of omega-3-fatty acids increased [30]. At least three portions of fish and seafood per week were statistically significantly associated with lower omega-6fatty acids values (24% compared to 26.3%, p = 0.016) and higher omega-3-fatty acids values. While high meat consumption is associated with higher omega-6-fatty acids values [30], such tendencies could also be seen in this study.
Barceló et al. [29] also reported an association between hsCRP values and olive oil consumption. According to their data, a MD enriched with olive oil (1 litre per week) resulted in a reduction of the plasma hsCRP concentration. Such tendencies could also be seen in this study. Individuals who consume more than four tablespoons of olive oil per day had lower values of hsCRP than individuals who consume less olive oil. The described associations between certain food items and blood values indicate that the MEDAS score indeed reflects a MD. Within this context, MEDAS provides reasonable estimates to adequately rank MD adherence.
Study limitations comprise firstly, a small sample size meaning the statistical tests would only have small power. A further limitation of this study is that our findings may not apply to the general population as the participants belonged to a selected population who may have a particular dietary behaviour due to their knowledge about their genetic disposition for breast cancer.
We will be using the German MEDAS in our main trial that aims to recruit 600 study participants. We decided to use the adherence to MD measured by the MEDAS score as one of 3 co-primary endpoints [31].

Conclusions
Despite the study limitations, we conclude that the present version of MEDAS could be a reasonable tool in determining adherence to a MD in German-speaking populations. This short screener is a valid tool for the rapid assessment of adherence to the MD that may also be useful not only for trials but also in clinical practice. The MEDAS score would allow an immediate feedback to study participants or patients regarding their adherence to MD.