To study the validity of the FFQ, the mean of two 24HDR was used as a reference method, and to test the reproducibility of the FFQ, the same questionnaire was administered twice within a one-month period.

From January 2017 till June 2017, recruitment of participants and data collection were carried out in collaboration with the Ministry of Social Affairs (MOSA) through the medico-social centers that serve low to middle class income families. As illustrated in Fig. 1, the first meeting with the participant was thru a face-to-face interview. It lasted around 30-45 minutes and was performed in 2 phases: the first one included collection of general information (including cognitive tests), and dietary interview (FFQ and 24HDR), and the second included anthropometric measurements and other data in relation to health. A 10-minute break separated the 2 phases, to allow the participant to rest. The second interview with the participant, was performed within a one-month period, over the phone, and included both second FFQ and 24HDR. In our study, caretakers were interviewed when subjects were identified to have decline in cognitive functions assessed through the Mini-Mental State Examination (MMSE) [25] and “Test des Neufs Images” (TNI) [26]. In case of significant cognitive decline i.e., when participants scored at or below the population specific cut-off for MMSE or TNI total recall score lower or equal to 9, then the accompanying person, usually a relative living with the participant in the same household, was asked to fill the questionnaire on behalf of the participant. All interviewers were dietitians who received an extensive training before the start of data collection. Each person was interviewed at the MOSA center near his/her home, and for participants unable to attend, the interview was performed by the same research team at home.

Design of the reproducibility and validity study of FFQ among older Lebanese. 24HDR: 24-hour dietary recall, FFQ: food frequency questionnaire

To avoid inter-rater and reporting biases, participants were administered twice the FFQ and 24HDR by the same interviewer, and all filled questionnaires were reviewed by the principal investigator and field interviewers before data entry.

As illustrated by Fig. 1, the study samples were selected from our cross-sectional study on Nutrition and Aging. We included a randomly selected group of 50 and 100 participants respectively, aged 60 years and older, belonging to 2 of the 8 governorates of Lebanon. Individuals who were totally dependent, with known active cancer disease, undergoing dialysis, with advanced liver disease, with major hearing and visual problems, or receiving artificial nutritional support were not included in the study [27].

For both analyses, the size of samples was set for an expected intra-class correlation (ICC) value of 0.75 and a minimum acceptable value of 0.5, with 95% confidence and a desired power of 80%. The number of participants for the reproducibility study was 42 out of 50 initially selected for this analysis (n = 7 declining to do the second interview, and n = 1 for incomplete FFQ). A total of 13 out of 42 had their questionnaire answered by accompanying person, because of low cognitive test results. From an initial sample of 100 individuals included to test for the validity of the FFQ, and after a preliminary analysis, 24 participants were not included in the study: subjects with previously undeclared active cancer (n = 2), unaccompanied individuals with altered mental status and who did not comply with the study protocol (n = 9), those refusing to undergo the second interview (n = 7), those having extreme daily dietary intake (either < 600 Kcal/day (n = 1) or > 4000 Kcal/day (n = 2)), as well as those having extreme restrictions and may not be representing a regular feeding pattern (interview was performed during fasting for more than 14 hours period) (n = 3). A total of 34 out of 76 had their questionnaires answered by accompanying person, because of low cognitive test results. Retention rates within a one-month interval for repeated questionnaires, for FFQ and 24HDR, were respectively 84 and 76%.

Our initial questionnaire, including 113 food items, was developed based on foods commonly consumed by older Lebanese. Particular attention was made to include foods reflecting nutrients with potential link to cognitive decline and frailty [3, 28,29,30,31,32,33,34,35]. This questionnaire was tested on a small group of 10 individuals, followed by a discussion with research dietitians to adjust for misrepresentations of food items and for the final adjustment of the questionnaire. Redundant items and not frequently consumed items were excluded from the list. The final FFQ used for the validity and reproducibility studies included the usual consumption of 90 food items representing all food groups. As illustrated in Table 1, food categories included in the questionnaire are bread and cereals, milk and dairy products, vegetables and fruits, meat, poultry and fish, fats and oils, sweets and desserts, and non-alcoholic beverages, as well as some traditional local foods and dishes. Alcohol consumption was removed from the list because of a very low consumption rate in the studied sample. Foods were grouped based on similarity in their nutritional profiles and frequency of consumption. Consumption of these items were reported as usual portion size used daily, weekly, or monthly. A manual illustrating the usual portions of foods listed in the FFQ, with their respective weights, was developed for the study to help investigators and participants better estimate quantities ingested. Participants were asked to express their answers by describing their habitual food intake the previous year. The final portions consumed were then translated into daily consumption. To account for seasonal variability and based on availability of fruits and vegetables on the Lebanese market, a seasonal coefficient (S) was added to the list of fruits and vegetables. This coefficient is accounted for in the nutrient analysis by dividing consumption with the corresponding coefficient.

Participants were asked to recall and describe in detail and in an open-ended manner the foods and beverages they consumed the previous day, starting from breakfast onwards. The interviewer first asked the participant if the day described was a usual day regarding his/her food consumption and dietary habits. The portion size consumed by the individual was estimated, as with the FFQ, using the portion manual guide, and standard measuring cups and spoons, to help estimate more accurately participant’s consumptions. For composite dishes, participants were always asked to describe the cooking method, the type and quantity of fat used for the whole household. For better analysis, as with the FFQ, brand names for specific foods such as milk, bread, biscuits, chocolates, and processed foods were also reported by the interviewer. As for beverages such as milk (from powder), coffee, tea and other drinks, methods of preparation were also detailed, and specific ingredients were added to the report.

Daily food consumption, reported by the FFQ was analyzed by the Nutrilog software (Nutrilog, version 3.2, France) to extract daily nutrient intake. The United States Department of Agriculture/ Standard reference 28 (USDA/SR28) and the Canadian Nutrient File (CNF 2015) databases were used to analyze nutrient composition of simple foods, and branded foods. For composite traditional dishes and some locally consumed desserts, not found in the database, recipes from known local cookbooks and pastry chefs were entered to the database and used in the nutrient analysis. Data on food consumed were reported simultaneously in (g) and as portions. The nutrient intakes of the two 24HDR were also extracted using the same methods.

To analyze the concordance between FFQ and mean 24HDR in classifying individuals in terms of nutrient adequacy, we calculated The Nutrient Adequacy ratio (NAR) of 17 selected nutrients, including vitamins A, D, E, K, C, B1, B2, B3, B6, B9, B12, and minerals such as calcium, phosphorus, magnesium, iron, selenium and zinc. The NAR was calculated by dividing the estimated nutrient intake of individuals by the age and sex-specific recommended dietary allowance (RDA) for these nutrients, according to the established dietary reference intake (DRI) recommendations [36] For all nutrients, RDA values were used except for vitamin K, where adequate intake (AI) was used as the DRI value for comparison.

IBM-SPSS 20.0 was used for statistical analysis. All macronutrients, 10 vitamins and 5 minerals estimated intakes were reported and compared for the validity and the reproducibility studies respectively. For numerical variables, data was expressed as mean ± standard deviation and median (interquartile ranges), and as numbers and percentages, for categorical variables. Student t-test and chi-square test were used to compare numeric data and categorical data respectively. Normal distribution of calories and nutrients was assessed using Kolmogorov-Smirnov tests, and data was processed accordingly. Characteristics of the reproducibility and the validation study samples were compared using Mann-Whitney U test, for continuous data and Chi-square test for categorical data. Wilcoxon signed-ranks tests were used for comparisons of both related samples of food frequency questionnaires, and for the difference between FFQ and mean 24HDR. For normally distributed data, Pearson’s correlation coefficient was used to test the reproducibility (test-retest) of the FFQ and the relation between estimated macro and micronutrients between the FFQ and 24HRs. For data that was not normally distributed, Spearman’s correlation coefficient was used. Correlation coefficients and their respective 95% confidence intervals (CI) were calculated and p < 0.05 was considered significant. To assess the internal consistency of the FFQ, after standardization of the variables, all items were computed in Cronbach’s alpha statistic. Alpha values between 0.7 and 0.8 indicate a good reliability, and a value higher than 0.9 would imply that all items on the questionnaire were very strongly related to each other and reliable for assessing the construct. To measure within-person variability, intra-class correlations (ICC) were calculated between the first and the second FFQ administration, using one-way random effects model, where people’s effects are random whereas interviewers performed under the same conditions. Values of r < 0.5 indicate a poor reliability, between 0.5 and 0.75 a moderate reliability, between 0.75 - 0.9 a good reliability, and greater than 0.9 excellent reliability [37]. Energy adjusted correlations between nutrients, estimated by FFQ and 24HDR, were calculated by pairwise linear regression models. Percentage agreement was calculated by the ratio of classification in the same or adjacent quartile to distant quartiles between 24HDR and FFQ, and through Kappa statistics. Bland-Altman plots were used to compare and visualize the agreement between 24HRs and FFQ, and both FFQ administrations.

To measure agreement of NAR estimation between the FFQ and the mean 24HDR methods, we calculated the percentage agreement between NAR estimated by FFQ and mean 24HDR by comparing classification in the same or adjacent quartile to distant quartiles, and through Kappa statistics. Concordance between the 2 NAR classifications was also measured using Spearman’s correlation coefficient.