DOI: https://doi.org/10.21203/rs.3.rs-26794/v1
Background: To date, there is no culture-specific and validated Food Frequency Questionnaire (FFQ) available in Ethiopia. We developed a FFQ and evaluated its validity as compared to estimates of a food group and nutrient intakes derived from two 24-Hour Dietary Recalls (24-HRs).
Method: The FFQ has a total of 89 food items. A total of 105 adults aged 20–65 years participated in this study. Wilcoxon- signed rank test, Spearman’s correlation, Cross-classification, kappa, and Bland-Altman analysis were used to validate food group intakes and nutrients derived from FFQ against two 24-HRs.
Results: Mean energy and macronutrient intakes obtained from the FFQ were significantly higher than those obtained from the mean of two 24-HRs. For energy and macronutrients, the correlation between two instruments ranged from 0.05 (total fat) to 0.32 (carbohydrate). Whereas, for micronutrients it ranged from 0.1 (calcium) to 0.49 (vitamin B1). Visual inspection of the Bland-Altman plots for both energy and macronutrients shows no consistent trend across the intake values. For the majority of the food groups, no significant difference was observed in median intake of foods and nutrients between 24-HRs and FFQ. Correlation coefficients for food groups ranged from 0.12 (egg) to 0.78 (legumes). The FFQ showed a fair classification agreement with the 24-HRs for cereals, legumes, and roots and tubers intakes. A systematic trend of overestimation for roots and tubers and under estimation of beverage intakes at higher values was observed when we used FFQ.
Conclusion: The FFQ is valid to assess and rank individuals in terms of intakes of most food groups according to high and low intake categories. However, estimates of roots and tuber and beverages should be interpreted with caution.
Food frequency questionnaire (FFQ) asks respondents about their usual frequency of consumption of each food during a specified time (1). Compared to other dietary assessment methods, such as short-term recall and diet record, FFQ is easier to administer, place less burden on respondents, has a relatively low cost, and provide a rapid estimate. This makes FFQ more feasible and best suited for measuring long term dietary intake for most epidemiological studies and large cohort studies (2).
Interpretation of results from studies of diet-disease that use FFQ is often difficult unless it has been adapted and validated in a population reasonably similar to that being investigated (3). Incorrect information may give rise to false associations between dietary factors and diseases or disease-related markers. The null association could also be attributed to a lack of variation in the dietary exposure in the study population or the inability of the tool to find out existing differences in the diet. Therefore, it is important to assess the degree to which the questionnaire measures the aspect of diet for which it has been designed (3, 4).
Because of the lack of gold standard in dietary assessment, validation studies cannot compare a test method with absolute truth rather they compare one method with another method that is judged to be superior (5). Among the available and feasible comparison methods to validate FFQ, diet records represent an optimal comparison method because of having least correlated error with FFQ. However, when co-operation or literacy of study subjects is limited, 24-hour recall (24-HR) may be more appropriate (3, 6).
To our knowledge, there is no validated standard FFQ in Ethiopia that can help to assess dietary habit among adults. Therefore, this study aimed to develop context-specific FFQ and evaluate the validity against two 24-HRs method.
The FFQ was validated against the average of two 24-HRs. The FFQ was obtained after the second 24-HR. There was a 15 days interval between the first and the second 24-HR. We have used an interactive, multiple passes 24-HR question adapted and validated for use in developing countries (7). The study was conducted among randomly selected 120 Ethiopian adults aged 20 to 65 in Butajira Health and Demographic Surveillance Site (HDSS), from March to April 2019.
We followed five steps to develop the FFQ: choosing appropriate foods, prioritization and categorization of food items, assembling a list of selected foods, frequency and portion size, and expert review and pre-testing. First, information on dietary intakes was obtained from unpublished cross-sectional dietary survey of women living in Butajira from 2018–2019. Information on dietary intake was collected using a single 24-HR. Market and mini-market visits were undertaken on non-consecutive days to identify common brand names and foods that could be relevant and were added accordingly. In addition, we conducted a focus group discussion in Butajira district two weeks before the interview. It was organized by the principal investigator and field supervisors. A group of 6 women were interviewed about foods consumed in the area. We undertook the discussion to identify food items that are typically consumed, including ingredients used, and methods of preparation. The information also included the availability of foods during the different seasons, traditional and ceremonial dishes consumed and foods that are infrequently consumed. Second, we combined similar foods and beverages into a single group of food items. Third, we clustered the related food items together. For closely related foods, we placed more specific items before general items. Forth, the Frequency of intake was evaluated based on the usual intake over 1 month prior to data collection. Seven frequency categories were included ranging from daily to never/one less than per month. Each food item was assigned a portion size. A pre-specified portion size estimation method was employed for the estimation of portion size in FFQ using local house-hold units such as bowl, plate, spoons of different sizes (tablespoon, teaspoon), coffee-cups, tea-cups, water glasses, as well as using photographs. Fifth, experts reviewed the newly developed FFQ (nutritionist from Addis Ababa University) to confirm its content validity. We have discussed on the food list extensively to ensure all relevant food items were included; this was followed by pre-testing of the developed questionnaire in a group of 10 adult women who are comparable to the study participants from a nearby sites. Some minor changes were made based on the finding of the pre-test.
The final developed FFQ consisted of 89 food and drink items. The reference time was determined to be one month prior to data collection point. The food groups include cereals, bread and potatoes, Legumes and pulses, Roots and tubers, vegetables, fruits, egg, milk and dairy, fish and fish-products, meat and poultry, fat and oil, sweets, drinks, and fast foods and pastry.
We have used an interactive, multiple passes 24-HR question adapted and validated for use in developing countries (7). The two 24-HR was conducted on non-consecutive days. We interviewed on weekdays and weekends to capture variance in the intakes across various days of the week. Prior to data collection rigorous training was given and pre-test was conducted. Each interview involved a stepwise series of questions.
First, we asked the participants to report everything that they consumed the previous day, including the night. The opening question was; “After you got up this morning/yesterday morning, when was the first time that you had something to eat or drink?” followed by the questions “What did you eat or drink at that time?” and “Did you eat or drink anything else at that time?” The same three questions were repeatedly been asked until the participants recall all the food and drink items consumed over the specified period. The first pass ended with the questions “Can you remember any other times you had something to eat or drink to?” In the second pass, participants were asked to provide additional detailed information about each item of food and drinks consumed. This includes the name of the food item, where they ate it, brand names, cooking methods, amounts served, and the amount consumed. For homemade dishes, participants were asked for the recipes and ingredients.
On the third pass common household utensils such as bowl, plate, spoons of different sizes (tablespoon, teaspoon), coffee cups, teacups, water glasses were used to improve the memory of the respondents and to assist in completing the recall. A digital food scale (Electronic kitchen scale) was used to measure the weight of the consumed as well as the ingredients used in food preparation to the nearest 1 g. The final pass reviewed all previously recalled information to confirm the accuracy of the record. During the final pass, the data collectors asked the participants to prompt for information about foods and drinks not mentioned that were considered to be easy to forget, such as snacks, fruits, water, and juices (8).
The frequency of intake was evaluated based on the usual intake over the previous month. Nine frequency categories were included ranging from daily to never/one less than per month each food item was assigned a portion size. A pre-specified portion size estimation method based on the average consumption was employed for the estimation of portion size using local house-hold units such as, bowl, plate, spoons of different sizes (tablespoon, teaspoon), coffee cups, teacups, water glasses, as well as using photographs. A visit to local markets and shops were held to purchase and arrange equipment for data collection. To determine the weight of the food items used commonly consumed dishes were made in Ethiopian public Health Institute laboratory by the principal investigator. The measurement was done with an Electronic Seca scale and the average of the 3 measurements was taken. Codes were given for different prepared portions. To help standardize participants understanding the interviewer prepared photographs for each measurement done and showed them to the participants.
The Ethiopian food composition table was used to derive nutrient and energy estimates from the dietary data. The names of foods and drinks, their description, cooking methods, and amounts from both 24-HR and FFQ, were coded and entered into NutriSurvey2007. The FFQ consisted of 89 food and drink items. The food lists were organized into 14 food groups on the basis of prior information. Food estimates from FFQ was calculated using the product sum method. The average frequency of food intake per week and month of the FFQ was converted to a daily intake value (e.g., frequency of 2–3 times per month = 2.5/30.5 times per day. Once the frequency of consumption per day was calculated, daily food intake was computed using product sum method. Daily food intake =∑ (reported consumption frequency of the food item, converted to times per day) *(portion size consumed of that food).
Both the FFQ and 24-HR data were checked for completeness and potential errors. We then entered the data on socio-demographic characteristics using Epi-Data version 3.1 and exported to STATA version 15 for further processing and analysis. Out of 120 study participants, 118 (98.3%) of participants completed 1st 24-HR, 116 (98.6%) completed 2nd 24-HR, 116 (98.6%) completed both 24-HRs and 115 (95.8%) participants completed both the 24-HRs and the FFQ.
The normality of average intake of nutrient and food groups was checked using the Shapiro-Wilk normality test and visualized using Q-Q plots. Parametric tests were used for normally distributed variables, while non-parametric tests were used for most of the variables as the distributions significantly deviated from normality. Those which fulfilled the assumption of normality were described using mean with standard deviation (SD) and those which do not using median with inter-quartile range (IQR).
We evaluated the performance of the FFQ against two 24-HRs using several statistical tests. First, to compare median daily food intakes obtained from the averages of the two 24-HRs and the FFQ, we used the Wilcoxon signed-rank test. To compare mean daily food intakes obtained from the averages of the two 24-HRs and the FFQ, we used paired t-test. Second, to measure the strength and direction of the correlation between the two methods, we computed the Pearson correlation for normally distributed variables, whereas Spearman’s rho for those not normally distributed.
Third, for both the test and reference methods subjects were divided into categories relating to the distribution of dietary intake; quartiles of intake. A comparison of the subjects’ categories showed whether subjects are classified in the same or different categories by the two methods. The result permitted an assessment of the proportion of subjects who are classified correctly. We used a weighted kappa statistic to account for both the correctly classified percentage and the expected participant proportion classified by chance. At last, we used a Bland and Altman plot for assessing limits of agreement between the two methods. The Bland-Altman method is preferable to compare two measurements each of which produced some error in their measures (9).
Table 1 shows the socio-demographic characteristics of the study participants. Of the 120 participants, 115 (95.8%) completed both the 2 day 24-hour dietary recall and the FFQ. A total of 105 study participants were included in the final analysis, of which 43 (41%) were male and 62 (59%) female. The mean age of participants was 31.9 years (SD: 9.2), 33.3% of them had primary education, and 43 (41%) were housewives.
| Characteristics of participants | Frequency | Percent |
|---|---|---|
| Sex | ||
| Male | 43 | 41 |
| Female | 62 | 59 |
| Age category | ||
| 20–29 | 49 | 46.7 |
| 30–39 | 35 | 33.3 |
| 40–49 | 13 | 12.4 |
| 50–65 | 8 | 7.6 |
| Education | ||
| Primary | 35 | 33.3 |
| Secondary | 21 | 20 |
| College/university | 10 | 9.5 |
| No formal education | 39 | 37.2 |
| Occupation | ||
| Farmer and housewife | 5 | 4.8 |
| Housewife | 43 | 41 |
| Employee/private | 3 | 2.9 |
| Merchant | 25 | 23.8 |
| Daily laborer | 8 | 7.6 |
| Unemployed | 8 | 7.6 |
Table 2 shows the mean (SD), median, and 25th, 75th percentiles daily nutrient intakes estimated by the average of two 24-HRs and the FFQ. The mean energy and macronutrient intakes obtained from the FFQ were significantly higher than the average of two 24-HRs. The highest mean difference was for energy 367.6 (CI: 259.0, 476.1), while the lowest was for total fat intake 4.1 (CI: 2.5, 5.7). Similarly, a significant median difference was found in micronutrient intakes between the two measures. The median difference ranged from 0.09 mg/day for vitamin B2 to 391.8ugRAE for vitamin A intake.
| Energy and nutrients | Average of 24-Hour Dietary Recalls | FFQ | Independent sample t-test | ||
|---|---|---|---|---|---|
| Mean | SD | Mean | SD | ||
| Energy (Kcal) | 1449.50 | 421.60 | 1817.10 | 482.30 | 367.60 * |
| Protein (g) | 39.40 | 12.20 | 49.90 | 12.50 | 10.50* |
| Total fat (g) | 17.20 | 5.70 | 21.30 | 6.10 | 4.10* |
| Carbohydrate (g) | 297.60 | 92.30 | 375.20 | 109.30 | 77.60* |
| Median | IQR (25%, 75%) | Median | IQR (25%, 75%) | Wilcoxon signed Rank test | |
| Calcium (mg) | 463.10 | 337.60, 587.90 | 684.40 | 518.00, 796.80 | 0.000** |
| Iron (mg) | 56.00 | 41.90, 83.70 | 67.40 | 54.30, 86.80 | 0.009** |
| Vitamin A (ugRAE) | 259.30 | 51.50, 581.40 | 651.10 | 295.90, 976.60 | 0.000** |
| Vitamin B1 (mg) | 0.70 | 0.49, 0.98 | 0.81 | 0.54, 1.17 | 0.002** |
| Vitamin B2 (mg) | 0.70 | 0.54, 0.84 | 0.79 | 0.64, 0.94 | 0.002** |
| *p-value ≤ 0.05 **p- value < 0.01 | |||||
Table 3 presents the results of correlations between nutrient intakes obtained from the average of two 24-HRs and the FFQ. The Pearson correlation coefficient varied from 0.05 (total fat) to 0.32 (carbohydrate). Except for total fat the correlations were statistically significant. Spearman correlation (rho) obtained for micronutrients ranged from 0.1 (calcium) to 0.49 (vitamin B1). A statistically significant correlations was obtained for vitamin A (p < 0.05) and vitamin B1 (p < 0.05).
| Macro-nutrients | Pearson correlation (95% CI) | Bland-Altman statistics | |
|---|---|---|---|
| Mean Difference (95% CI) | 95% Limit of Agreement | ||
| Energy (Kcal) | 0.24* (0.05, 0.41) | 367.6 (259.0, 476.1) * | -731.9, 1467.1 |
| Protein (g) | 0.22* (0.03, 0.36) | 10.5 (7.6, 13.5) * | -19.7, 40.8 |
| Total fat (g) | 0.05 (-0.14, 0.24) | 4.4 (2.5, 5.7) * | -13.7, 22.5 |
| Carbohydrate (g) | 0.32* (0.14, 0.48) | 77.6 (54.5, 100.6) * | -155.9, 311 |
| Micro-nutrients | Spearman correlation (95% CI) | Bland and Altman statistics for Log transformed data | |
| Mean Difference (95% CI) | 95% Limit of Agreement | ||
| Calcium (mg) | 0.10 (-0.09, 0.29) | 0.15 (0.11, 0.18) * | -0.26, 0.55 |
| Iron (mg) | 0.12 (-0.07, 0.30) | 0.06 (0.007, 0.11) * | -0.45, 0.61 |
| Vitamin A (ugRAE) | 0.45* (0.28, 0.59) | 0.5 (0.34, 0.66) * | -1.1, 2.1 |
| Vitamin B1 (mg) | 0.49* (0.33, 0.62) | 0.1 (0.02, 0.11) * | -0.4, 0.5 |
| Vitamin B2 (mg) | 0.17 (-0.02, 0.35) | 0.05 (0.01, 0.09) * | -0.41, 0.51 |
| *p-value ≤ 0.05 **p- value < 0.01 | |||
| CI: Confidence Interval; | |||
Table 4 shows cross-classification and weighted Kappa statistics of daily intakes of energy, nutrients and food group in quartiles assessed with average of two 24-HRs and the FFQ. The proportion of individuals classified by the FFQ and the average of two 24 hour dietary recalls into the same quartile ranged from 13.4% for total fat to 38.1% for vitamin A. However, the proportion classified into opposite quartiles varied from 3.8% (vitamin B1) to 23.8% (total fat). Weighted kappa values ranged from − 0.04 (total fat) to 0.18 (vitamin A).
| Energy and nutrients | Cross-classification | Kappa statistics | |
|---|---|---|---|
| % in same quartile individuals | % in opposite quartile of individuals | Kappa value | |
| Energy (Kcal) | 34.3 | 8.6 | 0.13 |
| Protein (g) | 33.4 | 8.6 | 0.11 |
| Total fat (g) | 13.4 | 23.8 | -0.04 |
| Carbohydrate (g) | 34.3 | 8.6 | 0.12 |
| Calcium (mg) | 25.7 | 7.6 | 0.09 |
| Iron (mg) | 28.6 | 9.5 | 0.05 |
| VitaminA (ugRAE) | 38.1 | 5.7 | 0.18 |
| Vitamin B1 (mg) | 33.3 | 3.8 | 0.11 |
| Vitamin B2 (mg) | 33.3 | 8.6 | 0.11 |
Table 5 present’s median, and 25th, 75th percentiles of daily food group intakes estimated by the average of two 24-HRs and FFQ. Both methods provide similar median intake estimates for fruits, eggs, meat/poultry/fish, and daily products. For roots and tubers, the 24-HR shows a higher estimate of median intake. FFQ provides a higher estimates of median vegetable intake. A statistically significant median difference was only observed for roots and tubers, eggs and vegetable intake.
| Food group | Average of 24-Hour Dietary Recalls | FFQ | |||
|---|---|---|---|---|---|
| Median | (25%, 75%) | Median | (25%, 75%) | P-value | |
| Cereals | 710 | (548.5, 817) | 648 | (520.9, 852) | 0.997 |
| Legumes | 94.5 | (0, 145.5) | 93 | (21, 134.9) | 0.347 |
| Roots and Tubers | 24.5 | (0, 45) | 11.2 | (0, 31.3) | 0.013* |
| Vegetables | 79.5 | (23.5, 156) | 109 | (45.3, 159) | 0.048* |
| Fruits | 0.0 | 0.0 | 0.0 | (0, 15.4) | 0.367 |
| Eggs | 0.0 | 0.0 | 0.0 | 0.0 | 0.000** |
| Dairy products | 0.0 | 0.0 | 0.0 | (0, 4.9) | 0.087 |
| Meat/Poultry/Fish | 0.0 | 0.0 | 0.0 | 0.0 | 0.068 |
| Beverages | 243 | (152, 334) | 230.4 | (183, 320.6) | 0.971 |
| Wilcoxon signed Rank test *p-value ≤ 0.05 **p- value < 0.01 | |||||
Table 6 shows the correlations between food group intakes obtained from the average of two 24-HRs and the FFQ. Spearman correlation coefficients ranged from 0.12 for egg to 0.78 for legumes. Greater than 0.5 correlation coefficient was observed for legumes (r = 0.78). Correlation (0.2–0.49) were observed for cereals (r = 0.33), Meat/poultry/fish (r = 0.47), fruits (r = 0.46), dairy products (r = 0.45), roots and tubers (r = 0.34), vegetables (r = 0.3) and beverages (r = 0.2). Correlation was low (< 0.2) for egg (r = 0.12).
| Food groups | Spearman correlation (95% CI) | Bland-Altman statistics | |||
|---|---|---|---|---|---|
| Mean Difference (95% CI) | 95% Limit of Agreement | ||||
| Cereals | 0.33* (0.15, 0.49) | 9.9 (-41.9, 61.8) | 535.3, -515.5 | ||
| Legumes | 0.79* (0.71, 0.85) | 2.6 (-8.9, 14.1) | -113.8, 118.9 | ||
| Vegetables | 0.33* (0.15, 0.49) | 7.1 (-15.7, 29.8) | -223.7, 237.8 | ||
| Beverages | 0.20* (0.01,0.38) | 2.9 (-30.9, 36.6) | -339, 344.7 | ||
| Spearman correlation (95% CI) | Bland and Altman statistics | Bland and Altman statistics for Log transformed data | |||
| Mean Difference (95% CI) | 95% Limit of Agreement | Mean Difference (95% CI) | 95% Limit of Agreement | ||
| Roots and Tubers | 0.34* (0.16, 0.45) | 16.2 (6.2,26.3) * | -87.5, 120.1 | -0.04(-0.23,0.15) | -1.96, 1.88 |
| Fruits | 0.46* (0.23, 0.56) | 7.9 (1.4, 14.6) * | -60.2, 76.2 | 0.15 (0.01,0.29) * | -1.25, 1.55 |
| Eggs | 0.12 (-0.07, 0.30) | -2.1 (-3.4,0.8) * | -15.9, 11.7 | 0.19 (0.11,0.27) * | -0.64, 1.03 |
| Dairy products | 0.45* (0.29, 0.59) | 6.2 (1.9, 10.4) * | -37.8, 50.1 | 0.04 (-0.07, 0.14) | -1.03, 1.1 |
| Meat/Poultry/Fish | 0.47* (0.31, 0.61) | 3.8 (-0.8, 8.4) | -43.2, 50.8 | 0.06 (-0.01, 0.13) | -0.68, 0.8 |
| *p-value ≤ 0.05 **p- value < 0.01 | |||||
Table 7 shows cross-classification and weighted Kappa statistics of daily intakes of food groups in quartiles as assessed with average of two 24-HRs and the FFQ. The highest correct classification into the same quartile was observed for “cereals” and “legumes”-i.e., 50.5% and 51.4%, respectively. For the other food groups, the classification into same quartile ranged from 30.5% (beverages) to 40% (roots and tubers). Oppositely classified individuals ranged from 1% (cereals) to 11.4% (beverages). No gross misclassification was observed for intake of legumes. Weighted kappa values ranged from 0.07 (beverages) to 0.35 (legumes).
| Food groups | Cross-classification | Kappa statistics | |
|---|---|---|---|
| % in same quartile individuals | % in opposite quartile of individuals | Kappa value | |
| Cereals | 50.5 | 1 | 0.32 |
| Legumes | 51.4 | 0 | 0.35 |
| Roots and Tubers | 40 | 7.6 | 0.18 |
| Vegetables | 38.1 | 6.7 | 0.17 |
| Beverages | 30.5 | 11.4 | 0.07 |
| (a) (b) | |||
Figure 1 presents the Bland-Altman plots for energy, protein, carbohydrate, total fat, vitamin B1, vitamin A, vitamin B2, calcium and iron. The Bland-Altman plot was used to evaluate the agreement between the FFQ and 24-HR by plotting for each nutrient the difference between the two methods versus the average of the two methods and calculating the limits of agreement and their corresponding 95% CI. Visual inspection of the Bland-Altman plots for both energy and macronutrients shows no consistent trend across the intake values. The FFQ overestimated energy and macronutrient intakes. Except for total fat intake, increased variability of data points was observed for all nutrients both at low, average and high values (wider limits of agreement). Overall, majority of the data points lied between the Limits of Agreements (LOAs). Some outliers were observed for energy and macro-nutrients. Since differences in nutrient intakes were associated with the mean measurement, data related to the micro-nutrient intake were log transformed for Bland and Altman statistics. The result indicates a trend as the FFQ consistently over estimated vitamin A and iron intake at lower value.
Figure 2 shows the Bland-Altman plots for legumes, cereals, vegetables, beverages, roots and tubers, fruits, egg, dairy product and meat/poultry/fish. Data related to roots and tubers was log transformed for Bland and Altman statistics. A systematic trend of overestimation for roots and tubers and under estimation of beverage intakes at higher values was observed when we used FFQ. Majority of the data points are in the 95% of limits of agreement for almost all food groups. A wide limit of agreement was observed for roots and tubers. A wide limit of agreement was observed for roots and tubers.
In this study, we developed and validated a food frequency questionnaire to assess food and nutrient intakes of adults in Ethiopia. We have observed a higher intake of energy and nutrients when the FFQ was used as compared to the average of the two 24-HRs.
Bland-Altman plots show an overestimation of energy and macro-nutrients (carbohydrate, protein and fat) for various data points. We found a low to moderate level of agreement (correlation coefficients) for energy and nutrient intakes between the two methods. The FFQ did not adequately classify subjects with respect to energy, macro-nutrients and most of the micro-nutrients. For the majority of the food groups, median differences in the intake of foods and nutrients between 24-HRs and FFQ were, overall, small and statistically insignificant. For food groups, a moderate correlation was found between the average of the two 24-HRs and the FFQ. The FFQ showed a fair agreement for cereals, legumes, and roots and tubers.
We found that the FFQ overestimated energy and nutrient intakes relative to the average of the two 24 h diet recalls. Overestimation is a common issue reported in various validation studies (10–15). Overestimation in the present study was moderate for intakes of energy (367.9 kcal), protein (10.5 g) and carbohydrate (77.6 g) and slight for intakes total fat (4.4 g) compared to other validation studies conducted using 24-HRs as their reference method (8, 10–13). Overestimation can be attributed to the subject’s tendency to overestimate their actual intake when they are asked to recall the frequency of a large number of foods consumed in an FFQ. Besides, difficulty in conceptualizing the assigned portion sizes and difficulties in reporting the frequencies of usual intake could be an attributing factor(13). It may have also occurred as a result of purposeful over-reporting of food consumption by subjects (7). The use of shorter questionnaires and advances in portion size estimation techniques are suggested to improve overestimation of intakes by FFQs.
This study found moderate correlations (0.2–0.49) between the average of the two 24-HRs and FFQ for energy (r = 0.24), protein (r = 0.22), and carbohydrate (r = 0.32) and low correlation (< 0.2) for fat (r = 0.05). The low to moderate correlation coefficients found between the two methods for energy and macro-nutrient intakes are comparable with other previous validation studies (8, 10, 12, 16). However, our finding was lower than those reported by other studies using 24-HRs as their reference method (8, 10, 12). The observed decrease in correlation coefficients could be interpreted as being the result of using only a two day 24-HRs as a reference method. It was found that nutrient correlations were lower when the reference method of the dietary questionnaire was conducted fewer than eight times (17). We believe that the observed correlation estimates could be improved with additional days of recall as well as with multiple studies over different seasons.
The lower correlation coefficients for iron and vitamin intakes observed in our study is not uncommon in FFQ validation studies (8, 10, 11, 13, 18). A meta-analysis of FFQ validation studies showed that pooled correlation coefficients of nutrient intakes ( total fat, protein, carbohydrate, alcohol, calcium, iron and vitamins) were lower for FFQ validated against 24-HRs rather than food record (17). The possible reason for a low correlation for the vitamin is that vitamin intake tends to vary greatly from day to day (as many vitamins are found in only a small selection of foods)(8).
We observed a moderate to good correlation for almost all food groups. This is in good agreement with previous validation studies assessing food group intakes (8, 10, 19, 20). The good correlation found for vegetable intakes in our study is higher than those reported by other validation studies (8, 10, 12). This may have occurred because of ease of quantifying vegetable intakes as they are often consumed independently in Butajira. The lower correlation of egg intake in our study, in contrast to other studies (10, 11) may have occurred because of not consuming egg on the days where 24-HR was conducted.
The Bland-Altman plot showed a moderate agreement between the two methods for energy and macro-nutrients. Trend was not observed across the intake level in energy and macronutrient intakes. Similarly, another study also showed a moderate level of agreement with no persistent trend across intake levels using a Bland-Altman plot (12, 21). However, ranges for limits of agreement were relatively wide as opposed to another study (13). The observed wide limits of agreements between FFQ and the reference method are common, hence highlighting the limitation of the FFQ in assessing absolute nutrient intake due to wide variability in how the FFQ measured energy and macronutrient intake relative to the average of the two 24-HRs (4).
A tendency towards a poorer agreement in Vitamin A and iron intake between methods was observed with lower levels of intake as shown by the Bland-Altman plot. This poor agreement in iron intake is also reported in another validation study (22). As indicated by a Bland-Altman plots, a systematic mean difference was not observed across the intake levels of Cereals, legumes, vegetables and beverages. Most of the data points are found between the 95% limits of agreement. However, the plot indicated wide limits of agreement which occurs as a result of increased variability.
The present study showed that the FFQ did not adequately classify subjects with respect to energy, macro-nutrients and most of the micro-nutrients as indicted by cross-classification and weighted kappa results (percentage of individuals in same quartile < 50, K values < 0.2) (5). However, the FFQ showed a fair quartile classification agreement for cereals, legumes, and roots and tubers (K values 0.21–0.4). This finding is consistent with previous studies which reported cross-classification and kappa by categorizing intakes into quartiles (10, 11, 22). We found lower values for energy and nutrients with respect to those reported by other studies using similar intake categories (8, 12, 13, 23). The misclassification and low kappa reported in our study may have occurred due to the insensitivity of FFQ to classify individuals into intake categories. The use of a Food diary as a reference method may have also increased classification agreement in the previous studies. FFQ showed a fair quartile classification agreement for cereals, legumes, and roots and tubers (K values 0.21–0.4). Similarly, other studies reported a fair classification agreement for this particular food group(13, 19). For beverage intake, our study indicated a misclassification (30.5%) into opposite quartile supported by low kappa value (k value < 0.21) showing the poor outcome. Other study reported a similar finding for beverage intake (19).
The present study has some limitations that must be acknowledged. First, given that we used a 24-HR dietary assessment method as our reference method, the sources of error between 24-HR and FFQ may tend to be correlated due to conceptualization of portion sizes. However, to lessen this effect we have used a salted replica of actual foods, pictures and calibrated equipment to estimate portion size. Second, we conducted two 24-HR per participant. This might have influenced the result obtained, particularly for estimating usual intakes of foods not consumed on a daily or regular bases such as meat/poultry/fish and the intake of other specific food groups. Third, participants may have purposefully over reported their intake due to social desirability. However, we gave a detailed explanation for the interviewers on how to explain the purpose of the FFQ to the participants using role-playing, small group exercises, and discussions. Fourth, we did not administer FFQ at the onset of the study; therefore, we cannot assess the reproducibility of the instrument.
The strength of the present study are the development of the FFQ based on the latest local dietary survey, focal group discussions, pre-test, and expert reviews, the use of comprehensive statistical analysis, to assess the validity of the FFQ and the use of an interactive, multiple-pass 24-HR adapted and validated for use in developing countries as our reference method..
The study showed that FFQ had good validity to capture intakes of cereals, legumes, vegetables, and beverages both at individual and group levels. However, intakes of root and tuber and beverages are associated with potential systematic bias. Therefore, caution must be exercised when using FFQ for this particular food groups. The FFQ can be used to rank individuals based on cereals, legumes, roots and tubers and vegetable intakes. The supporting individual level validity was acceptable for energy and macro-nutrients as indicated by correlation coefficients and Bland-Altman plots. However, estimates of minerals and vitamins should be interpreted with caution. In summary, FFQ is capable of classifying an individual’s food group intake into quartiles, which is useful in examining the relationships between diet and chronic disease. Food group intake agreement analysis also indicated the usefulness of the tool to assess dietary intakes in large epidemiological studies. However, the use of the tool to assess absolute nutrient intakes at the group level should be exercised with caution.
Ethical clearance was obtained from the institutional review board of School of Public Health, College of Health Sciences, Addis Ababa University. Informed written consent was obtained from the study participants. The study is in compliance with the principles of the declaration of Helsinki.
Not applicable
The datasets generated and/or analyzed during the current study are not publicly available due to the confidentiality regulations of our University, but are available from the corresponding author upon request.
The authors declare that they have no competing interests.
The study did not receive funding. All participants voluntarily agreed to take part.
Conceptualization, I.F., B.S.E., S.H.G., and E.H.; methodology, I.F., B.S.E., S.H.G., and E.H.; software, I.F., B.S.E., S.H.G., and E.H.; validation, I.F., B.S.E., S.H.G., and E.H.; formal analysis, I.F., B.S.E., S.H.G., E.H., and H.Y.H.; investigation, I.F.; writing original draft preparation, I.F., B.S.E., S.H.G., E.H., and H.Y.H.; writing, review and editing, I.F., B.S.E., S.H.G., E.H., and H.Y.H. All authors approved the final version to be submitted.
We are grateful to supervisors, data collectors and study participants
Ilili Feyesa1, Bilal S Endris1, Esete Habtemariam1, Hamid Y Hassen2, Seifu H Gebreyesus1
1Department of Nutrition and Dietetics, School of Public Health, College of Health Sciences Addis Ababa University, Ethiopia; [email protected] (IF); [email protected] (BSE); [email protected] (EH); [email protected] (SHG)
2Department of Primary and interdisciplinary care, Faculty of Medicine and health sciences, University of Antwerp, 2610, Antwerp, Belgium. [email protected] (HYH)
*Correspondence: [email protected]; Tel.: +251919324253