The TeenFADS is a cross-sectional study that investigated the associations between dietary intake and health and pubertal development of adolescents. Data were collected using an online questionnaire that consisted of several sections including a dietary assessment section (webFFQ) and school visits for the anthropometric measurements. This current study was designed to validate the webFFQ using multiple 1-day FR as reference method. Participants completed the webFFQ, underwent a one-on-one training to do FR, and then provided 6 1-day FR over 2–3 months. They utilized their personal mobile phones for their FR. The Institutional Review Boards of Loma Linda University (LLU) and Andrews University approved the study protocol and the informed consent process for both parental permission and assent of the adolescent participant.
The TeenFADS involved adolescents aged 12–18 years who attended middle and high schools near Adventist universities in southern California and Michigan. Convenience sampling was done to recruit volunteers for the validation sub-study from among those who completed the webFFQ and attended the anthropometry clinic (n = 601). A question at the completion of the web-based questionnaire asked respondents if they were willing to participate in a sub-study. Those who responded positively were contacted and participants were selected based on the following criteria: (1) owned a mobile phone with good camera resolution; (2) had unlimited texting/calling; (3) after their one-on-one training, demonstrated through a return demonstration test the ability to follow video and/or printed instructions on how to take digital photographs correctly; and (4) could text messages and digital photographs to an email account set up for the study. Initially, 108 adolescents volunteered but of these, 23 did not meet the first two requirements. Of those who passed the training (n = 85), eight did not proceed with the study due to a busy schedule, five either lost or did not receive their fiducial markers and could not proceed with the return demonstration, and two lost interest in the study before it started. Thus, a total of 70 participants entered the study.
Web-based food frequency questionnaire
The webFFQ is a 151-item self-administered, semi-quantitative dietary assessment questionnaire that was designed to assess the soy intake and habitual diet of adolescents. The food list was initially composed of foods from a pilot questionnaire constructed at LLU that was administered among adolescents to determine their soy foods intake. Foods deemed commonly eaten based on interviews of multiethnic adolescents were added to this list. Registered Dietitians (RDs) who work with adolescents reviewed the list to ensure that the food names used were appropriate for this population. A comprehensive list composed of 151 food items, 36 of which were soy-containing, were included in the final version (see Additional file 1: Table S1). Food items were grouped into eight categories that constituted eight screens in the webFFQ. The webFFQ was then pre-tested on 25 adolescents of the same demographic characteristics but were not participants in the study. This was done to determine clarity, interpretation, length of completion time, and if respondents would use the pop-up feature for additional information about an item if they needed further clarification. No conceptual difficulties in filling out the webFFQ were reported and the majority of participants did not find it necessary to seek additional information about the food items. Technical issues with use of specific browsers were resolved before administering the webFFQ.
Foods in the webFFQ are categorized as convenience foods (32 items), protein-rich foods (29 items), starches/cereals (17 items), vegetables/fruits (21 items), dairy products (10 items), beverages (24 items), snacks/sweets (11 items), and soups/legumes (7 items). These food groups are divided over eight screens (see Additional file 2: Figure S1). Food items are arranged in such a way that various types of a specific food are grouped (e.g., dairy milk: regular, low-fat or non-fat). Protein-rich foods such as meats also have their meat alternative counterparts (e.g., chicken, vegechicken) (see Additional file 3: Figure S2). More information about the food and examples pop up when the pointer is hovered over the food name (see Additional file 4: Figure S3). Fixed portion sizes are based on familiar measuring devices, e.g., cup, tablespoon, 12-fluid ounce can, and others.
On the webFFQ, respondents are asked to self-report the frequency of their intake during the past month. Frequency of intake is selected from a drop-down list of frequency categories which are: never/rarely, 1–3 times per month, once per week, 2–4 times per week, 5–6 times per week, once per day, 2–3 times per day, and 4 or more times per day (see Additional file 4: Figure S3). For seasonal foods, e.g., certain fruits, “when in season” is included as frequency of intake category. To ensure that no item is skipped, respondents are not allowed to move on to the next food category until all items on the screen are completely filled out. However, they can go back to make changes to their responses in any food category if desired.
The current variety of meat analog brands has various formulations. More specifically, some contain more gluten than soy protein or contain no soy protein at all. To differentiate between these formulations and more accurately determine soy and isoflavones intake, a separate window of different brand names for meat analogs appears for respondents to choose from when a frequency of “once per week” or greater is chosen. Frequency of “once per week” or greater for meat intake also shows a separate window where respondents choose the type of meat eaten (beef, chicken, turkey, lamb or pork). It takes approximately 25 min to complete the webFFQ.
Food records with digital photographs
Multiple day food records provide a better measure of food and/or nutrient intake and, thus, is commonly used as a criterion reference measure in evaluating or validating FFQs [26, 27]. This dietary assessment method is often paper-based and recording is done on consecutive days. Weighing or taking a photograph of the foods before and after they are eaten are some of the approaches used to improve the accuracy of food records. As the criterion measure for this study, we determined the feasibility of using personal mobile phones for photographing foods to be consumed and recording intake as part of the food recording process among our young study participants. This method is described in a forthcoming article . Instead of consecutive days of recording, the protocol was designed to reduce the burden associated with continuous food recording which may result in deviation from usual intake  or boredom, and to make the procedure more attainable among adolescents.
Before the data collection, participants were trained and then tested through a return demonstration of a one-day FR that was then evaluated to determine readiness, or additional training, for the study. During the trial runs, a few of the participants were excluded for losing their fiducial marker. Laminating the fiducial marker, adding a hole on one side and inserting a 2-foot long 1-inch wide ribbon through the hole resolved the problem of misplacing or losing the marker. Participants were randomly scheduled for six 1-day food recording days over a 2- to 3-month period. All five weekdays (Monday to Friday) and one weekend day (Sunday) were covered in scheduling the recording days.
The FR had two components: (1) digital photographs before- and after-intake of a meal that captures a fiducial marker for size approximation of objects in the digital photograph; and (2) text message that captures the time of intake, meal type (breakfast, snack, lunch, or dinner), and the foods eaten with corresponding amounts. Participants used their personal mobile phones to take digital photographs and record their intake. They sent their text report and photographs of their intake to a central location (i.e., the email address for the study). Reports from each participant (all digital photographs and texts sent during their reporting day) were reviewed and collated into food records the following day by trained research assistants. The amount of intake reported in the text was compared with the before- and after-eating digital photographs (after-eating digital photographs were only sent if there were leftovers) to determine if amount reported was estimated accurately or needed to be adjusted. If reports were incomplete, unclear, have discrepant information or other issues, the participant was contacted for clarification. Otherwise, the FR was cleared and logged on as completed.
Isoflavones and nutrient intake determination
Dietary intake reports on both the webFFQ and FR were coded using the Nutrition Data System for Research (NDS-R) software version 2012 developed by the Nutrition Coordinating Center, University of Minnesota, Minneapolis, MN, for nutrient composition and analysis. The NDS-R has a comprehensive database of about 18,000 foods—which reflects variations according to preparation methods and component ingredients—and 163 nutrients/nutrient ratios/other food components derived from the USDA database and several other sources . The comprehensive meat analogs database of the Adventist Health Study-2  was utilized in determining the nutrient profile of meat analogs. Frequency values of average consumption during the past month were converted into frequency of intake per day, e.g., frequency of 1–3 per month = 2/30 = 0.067, and so on. Thus, factors used were 0 for “never/rarely”, 0.067 for “1–3 times per month”, 0.143 for “once per week”, 0.429 for “2–4 times per week”, 0.786 for “5–6 times per week”, 1.0 for “once per day”, 2.5 for “2-3 times per day”, and 5.0 for “≥4 times per day”. Nutrient amounts were relative to the fixed portion/serving sizes in the webFFQ. For vegetables that are eaten either raw or cooked, the average of raw and cooked nutrient components was used. For mixed foods, recipes were created by trained RDs and nutrient composition was based on the proportion of ingredients in the recipe. For single food items where several types exist, the NDS-R software computes nutrient composition for the most commonly eaten type (the “default” food), a determination made using nationally representative market research data . Nutrient and isoflavones intake per day was computed afterwards using the product-sum method . Daily FRs were coded on NDS-R separately for each participant.
All 70 participants selected for this validation sub-study completed six recording days. However, 15 participants had to be excluded from the analyses: four due to improbable intake on the webFFQ which we predefined as >4500 kcal or <1000 kcal for boys and >3500 kcal or <900 kcal for girls, and 11 due to poor or non-compliance with the protocol in food recording. Thus, the analyzable data was based on information from 55 participants.
Most of the nutrients and isoflavones had skewed distributions. Absolute median intake estimates from the webFFQ and the FR were compared using Wilcoxon signed-ranks test. Intake values were either log- or square root-transformed to achieve normality before Pearson’s bivariate correlation analyses were performed. Correction for within-person variation in multiple FR measurements was performed using the formula r
[1 + (Sw2/nSb2)]½  where r
= corrected correlation coefficient, r
= crude correlation coefficient between FFQ and mean of the multiple food records, Sw2 = within-person variance of the multiple food records, Sb2 = estimate of the between person variance in the reference method (food records), and n = number of repeated measures of the food records. Using this method for deattenuation or correction of the correlation coefficients creates conditions in which normally distributed errors for r
cannot be assumed. Thus, instead of the traditional asymptotic methods to determine confidence intervals about r
, we computed ‘distribution-free’, non-parametric 95 % confidence intervals using the bias-corrected and accelerated (BCa) bootstrap re-sampling method , with each confidence interval determined from the distribution of r
’s from 2000 samples.
Analytical tests were performed without energy adjustment and with energy adjustment using the residual method. The ability of the webFFQ to rank respondents according to intake was tested by determining classification agreement between quartiles of consumption levels according to webFFQ responses and the mean of the multiple FR. Proportions of exact agreement, within 1 quartile agreement, and gross misclassification were also estimated. Additional graphical determination of agreement with the reference method using Bland-Altman plots was done for the soy isoflavones (total, daidzein, genistein, and glycitein). Total isoflavones was computed as the sum of daidzein, genistein, and glycitein. Analyses were performed using the Statistical Analysis System (SAS) statistical software package version 9.4 (SAS Institute Inc., Cary, NC, USA) and R version 3.1.2 .