Study design and recruitment
Reproducibility was assessed by comparing a test and retest of the FFQ. Relative validity was assessed by comparing intake measured by the FFQ (test) with 7-day weighed food records. Recruitment of participants took place from November 1st until November 24th in 2017. Data were collected in the time period from November 2017 until January 2018. To be included in the study, participants had to be 20-30 years old, without children and give their consent to participate. The lower range of the age group was based on targeting young adults as they move away from home and start their independent life, thereby deciding their own diet. The upper range of the age group was based on the Norwegian age of first child birth, which were 29 for women and 32 for men in 2018, respectively [13]. The study included both preconception and periconception young adults but did not distinguish between the two. The participants needed to have access to internet, possess the necessary skills to complete an internet-based questionnaire and be willing to weigh and record their intake of food for seven consecutive days. Finally, they needed to be able to meet in person at the University of Agder, Kristiansand, Norway, at least once to attend an instructional meeting.
The study was advertised on social media, among students at the Faculty of Health and Sport Sciences at University of Agder, and through word of mouth. The advertisement led potential participants to the study website, which contained a general outline of the study, an invitation to participate, contact information, and a button for enrolment in the study. When signing up for the study, the participants gave their consent to participate and selected one of 11 possible instructional meetings to attend. As participants signed up, an e-mail containing a link to the online FFQ was sent to their e-mail address. The e-mail also contained information about the FFQ, a deadline for completing the FFQ, and information on where to meet for the instructional meeting. If a participant had not completed the FFQ 1 week before their scheduled meeting, a reminder was sent by e-mail.
For the test-retest reproducibility investigation, a link to the FFQ retest version was sent to participants via e-mail on December 13th, 2017, resulting in at least 19 days between test and retest of the FFQ.
Study population
During the recruitment period of approximately 4 weeks, 32 participants signed up. Of these, 29 completed the first FFQ (of which three did not complete the entire FFQ). In the course of 11 instructional meetings 25 participants attended and started their 7-day WFR. Of these 25 participants, 22 completed the WFR (one participant recorded 6 days) and returned their recording booklet.
The FFQ retest was completed by 21 of the 22 participants that completed the WFR. A total of 21 participants completed all the components of the study (Fig. 1).
The food-frequency questionnaire
The FFQ was developed from an existing FFQ targeting adolescents [14]. In the first stage of development, the original FFQ was sent to five volunteers using a purposive sampling, at which point they completed the FFQ and then answered questions according to an interview guide. Feedback from the volunteers were considered by the authors. The inputs deemed relevant based on our understanding of the age group and development of an FFQ were included in the revision of the questionnaire. In addition, the original questionnaire was revised in accordance to input from the authors and expert colleagues at the faculty. In the second stage of development, the revised version of the FFQ was sent to two new volunteers, using purposive sampling. The participants were interviewed in accordance with the interview guide upon completion of the FFQ. Based on their feedback the final version of the FFQ was created, using the online survey tool SurveyXact [15]. The food-frequency questionnaire can be found in Additional file 1. The majority of the food items in the original FFQ were included in the revised version. Some food items were excluded (food items not found relevant by the participants during the development, food items targeting children, i.e. specific sweetened breakfast cereals), and some food items were added (i.e. low- and full fat options, a wider range of condiments). Changes to frequency of intake were based on some participants requesting a greater degree of variation in order to reflect their dietary intake (i.e. adding ‘more than 3 cups per day’ for coffee).
The FFQ consisted of 146 questions aiming to reflect diet in a four-week retrospect by asking respondent to report their average intake of the specified food items during the last 4 weeks. Completing the FFQ took about 25 min. The FFQ started with the respondents’ sex, age, self-reported height and weight, and level of education, followed by 121 questions related to average consumption of foods and beverages. These were divided into different sections (beverages and dairy products, bread and grain products, lunch meats, dinner meals and side dishes, fruits and vegetables, desserts, cakes and snacks). The FFQ ended with 13 questions regarding food habits and 7 questions related to physical activity, screen time, sleep and tobacco use.
The response alternatives regarding frequency of intake varied according to food and beverage questions. For beverages and dairy products (not including yoghurt), the interval range was ‘never’, ‘1-3 per month’, ‘1-3 per week’, ‘4-6 per week’, ‘1 per day’, ‘2-3 per day’ and ‘more than 3 per day’. Dinner meals and side dishes used the interval ranges ‘never’, ‘1-3 per month’, ‘1 per week’, ‘2-4 per week’ and ‘more than 4 times per week’. Fruit and vegetables, desserts, cakes and snacks all used the interval range ‘never’, ‘1-3 per month’, ‘1 per week’, ‘2-3 per week’, ‘4-6 per week’ and ‘1 or more times per day’. The respondents reported their food intake in ‘units per month’, ‘units per week’ or ‘units per day’. The unit measurements differed between sections and foods, whereas most questions were related to a standard portion size (e.g. cup of coffee, a piece of bread, an apple). For some questions, extra information was provided (e.g. maize = 2 tablespoons or soda = 0.5 l).
Calculating food and nutrient intake was overseen by the corresponding author. All food and beverage related questions were linked to a corresponding food-code in the Norwegian food composition table [16]. The Norwegian Food Safety Authority’s “Weights, measures and portion sizes for food” [17] and the web-page “Food-Recipes” [18] were used when assigning portion sizes. Amount in grams/millilitres was calculated using portion sizes and reported frequency of intake per day (based on 1 month being 28 days, as the FFQ reported intake in a four-week retrospect). FoodCalc [19] was used to process the FFQ, based on nutritional values from the Norwegian food composition table [16]. Food and beverage items assessed by the FFQ were organized into 28 non-overlapping food groups according to nutrient profile, biological classification or culinary usage.
The 7-day weighed food record
At the instructional meetings, the participants received general information on how and why the study was conducted, and instructions on how to weigh and record their food. They also received the equipment necessary to implement the WFR. The participants were encouraged to maintain their normal diet during the recording period, as any change in diet could influence the validation of the FFQ. Every participant received a weighing scale (Swordfish SFKSW14E) and was told to use this rather than a personal weighing scale. The recording booklet contained two pages of information on how to weigh and record food. This information was reviewed, followed by a review of the booklet itself. To accurately record the weight of the food, participants were given boxes to weigh the remains after a meal. A practical example was conducted with weighing and recording of a test meal. When eating out, participants were instructed to weigh and record their food as usual if possible. If the weighing scale was not accessible, participants were instructed to take note of what they ate, take a picture if possible and estimate portion size. The participants were offered a pre-paid envelope if they did not have the opportunity to deliver the recording booklet in person after completing the WFR. The WFR started the following day and continued for seven consecutive days. The participants were encouraged to make contact by e-mail or telephone if they had any questions.
Data entry of the food records was conducted by the corresponding author in collaboration with two research assistants. Calculating food and nutrient intake was overseen by one of the co-authors. Foods and beverages recorded by the participants in the WFR were linked to a corresponding food-code in the Norwegian food composition table [16]. The Norwegian Food Safety Authority’s “Weights, measures and portion sizes for food” was used when converting from volume measurements to grams and for calculating the weight yield factor from cooking [17]. When participants provided recipes, these were replaced with the closest approximate food-code. We used FoodCalc [19] and the Norwegian food composition table [16] to calculate food and nutrient intakes from the WFR. Food and beverage items assessed by the WFR were aggregated into the same 28 non-overlapping food groups used for the FFQ.
Statistical analysis
Descriptive analyses were used to evaluate the characteristics of the participants (age, height, weight, body mass index (BMI), level of education). Most of the nutrients used to assess reproducibility and relative validity were not normally distributed and presented as median with 25th and 75th percentile, although some were considered to be normally distributed and therefore presented as mean with standard deviation (SD). The food groups used to organize the FFQ test-retest and WFR were not normally distributed, and therefore presented as median with 25th and 75th percentile. We used Spearman’s rank correlation coefficient to examine the correlations for the test-retest reproducibility and the relative validity. Correlation coefficients between 0.5 and 0.7 has shown to be common when testing the reproducibility between two administrations of an FFQ [10], and a Spearman correlation coefficient above 0.5 is recommended for nutrients in dietary validation studies [20].
We also examined energy intake and intake of vegetables and fruits assessed by the FFQ and the WFR by Bland-Altman plots, i.e. by plotting the mean energy intake and intake of vegetables and fruits (x-axis) against their mean difference for each participant [21]. Further, the total intake per day for the food and beverages included in each food group were ranked into tertiles of intake for the FFQ and WFR. The FFQ’s ability to categorize participants into the correct tertile of intake was assessed by calculating the percentage of agreement. Participants were presented as percent correctly classified or grossly misclassified. Participants correctly classified were categorized in the same tertile of intake for both measurements, whereas participants that were grossly misclassified were categorized in non-adjacent tertiles. Unweighted Cohen’s Kappa statistics was analysed for food intake in each food group ranked into tertiles to identify the strength of agreement. Values of Kappa, according to Masson et al., are categorized as follows: < 0.20: poor agreement, 0.21-0.40: fair agreement, 0.41-0.60: moderate agreement, 0.61-0.80: good agreement, and > 0.80: very good agreement [20]. Self-reported height and weight were used to calculate BMI (kg/m2). The significance level was set to 5%, and all statistical analysis were carried out using the computer program IBM SPSS Statistics for Windows, version 24 (IBM Corp., Armonk, N.Y., USA).