Although the trends of adolescent overweight and obesity have levelled off in some European countries, the prevalence remains high (1). Therefore, nutritional intake and diet quality remains an important focus of scientific community. Furthermore, it is already well recognized that diet, along with environmental and life-style determinants, plays an important role in the aetiopathogenesis and, vice versa, in the prevention of several non-communicable diseases (2), particularly if it starts in early childhood (3). In order to understand the association between diet and different diseases, appropriate tools for dietary assessment need to be developed (4). Standard methods of assessment of dietary intake include 24-hour recalls, food diaries and food frequency questionnaires (FFQs) (5). Currently, all of the available methods for dietary intake assessment have their advantages and disadvantages (5). The choice, therefore, depends on the size of the studied population, price, objectives of the study, among others (6, 7).

In clinical and epidemiological studies, the most commonly used dietary assessment tools are FFQs (5). Quantitative FFQ estimates frequency and quantity of consumption of different foods and beverages. Moreover, FFQ can assess consumption of food during the previous month or previous year. Their advantages over other dietary assessment tools are a lower respondent burden, cost effectiveness and time saving (8). They can also be used to estimate both micro- and macronutrient intake, as well as the intake of foods and/or food groups (9).

There has been limited research into the diet of Croatian adolescents, and therefore, development and validation of FFQ is necessary. FFQs are in most cases validated against other methods that assess dietary intake such as 3-day food records (3DFR) or 24-hour recalls (5). The aim of this study is, therefore, to develop a quantitative FFQ and evaluate its relative validity against two 3-day non-consecutive food records for adolescents living in Croatia.

This study shows that a simple self-administered questionnaire completed by adolescents is a valid tool for measuring energy and nutrient intake among adolescent population in Croatia. The mean nutrient intakes estimated by the FFQ-m were higher than those of the 3D-FR, while the average correlation coefficient for energy and nutrients in our validation study was 0.40. On average, 76.5% adolescents were classified in the same or adjacent quartile of the nutrient intake. In Croatia, very limited data on the diet of adolescents is available, therefore, this FFQ will make the assessment of the diet in a large population of adolescents simpler and less burdensome.

From 84 participants included into the final analysis, the main characteristics of 78 adolescents for whom anthropometric data were estimated are presented in Table 1. In total, 26.9% adolescents were overweight or obese and 14.1% of adolescents were underweight.

The results of our FFQ-m validation study showed a good correlation with the 3DFR for the intake of all macronutrients and some micronutrients (sodium, phosphorus, potassium, calcium, magnesium and iron). To our knowledge, this is the first FFQ developed and validated for Croatian adolescent population. FFQs are low-cost, fast and easy to administer tools, and therefore commonly used in epidemiologic studies (8).

Average intakes of macro- and micronutrients estimated by the FFQ-m and 3DFR are presented in Table 2. A significantly higher intake of energy, most macronutrients (total protein, total fat, SFA, MUFA and PUFA, mono- and disaccharides and dietary fibre) and micronutrients (potassium, phosphorus, iron, retinol equivalents, vitamins B1, B2 and B6, niacin and vitamin C) were estimated by the FFQ-m and compared to the 3DFR. Overestimation of FFQ-m compared to 3DFR was 33.1% on average for macronutrients, ranging from 4.2% for total carbohydrates to 62.7% for PUFA. Overestimation of micronutrients ranged from 1.8% for calcium to 171.4% for vitamin B6. FFQ-m underestimated intakes of cholesterol, sodium, magnesium and zinc compared to 3DFR (Table 2).

The detected overestimation of energy and nutrient intakes has been observed in previous validation studies in children and adolescents as well (21-26). Our FFQ-m largely overestimated the intake of vitamins B1, B2 and B6, and therefore cannot be used for the assessment of these nutrients. Detected overestimation of FFQ in comparison with other dietary methods could be caused by a wide selection of available foods that the participant can choose from (27). Similarly, since 3DFR estimates nutritional intake on random days, consumption of some foods that are part of the FFQ-m might have been missed on days when the 3DFRs were filled. Indeed, frequent underreporting of food intake obtained from food records (28-30) and over-reporting in FFQs (31, 32) have been observed in previous studies.

The correlation coefficients between FFQ-m and two 3DFRs are presented in Table 2. The Spearman’s correlation coefficient showed a significant relationship between the energy intake and all macro- and micronutrients except zinc, retinol equivalent and vitamins B1 and B2. The highest correlation between the two methods was shown for total energy (r=0.701), total fat (r=0.604), total protein (r=0.510), total carbohydrates (r=0.560), SFA (r=0.589) and MUFA (r=0.628).

The average correlation coefficient for energy intake and nutrients in our validation study was 0.40, which is in accordance with other studies (22-33). The highest correlation coefficient was detected for energy intake (r=0.7), similar to that of Morel et al. (26). When separated, in our study mean correlation for macronutrients (r=0.51) was higher than that of micronutrients (r=0.30), mostly due to low correlation coefficients for zinc, retinol equivalent, and vitamins B1 and B2. Similarly, in other studies correlation coefficients for macronutrients were higher than those of some micronutrients (33, 34).

Cross-classification analysis (Table 3) revealed that on average, 76.5% adolescents were classified in the same or adjacent quartile of nutrient intake when comparing the data obtained by the 3DFR. On average, 5.1% adolescents were classified into the opposite quartile and 35.1% adolescents were classified into the same (identical) quartile.

These results are in accordance with the results of previous studies (22-24, 26, 35). The highest percentage of adolescents who were classified into the opposite quartile was observed for retinol equivalent, vitamin B1, vitamin B2 and niacin. The same nutrients have shown the lowest coefficient correlation, possibly for the same reasons that were discussed previously.

Bland-Altman analysis is often used in validation studies in conjunction with correlation coefficients, and is suggested to assess absolute validity of the FFQ (36). This analysis assesses in graphical form the agreement between the methods across the range of intakes by plotting the mean of the two methods against the difference. The mean agreement indicates how well the FFQ and food records agree on average (36). The Bland and Altman plotting (19) showed good agreement among the total energy intake, total protein, total fat, SFA, MUFA, cholesterol, total carbohydrates, calcium, magnesium, potassium, phosphorus and iron. Figs. 1-3 present plots for energy, total protein and calcium intake, showing good agreement between the FFQ-m and 3DFR for all mentioned nutrients, where the Y axis shows the difference between the FFQ-m and 3DFR derived measurements and the X axis represents the average of these measures (37). Plots produced for other nutrients were similar to those shown in Figs. 1-3, except for retinol equivalent, vitamin B1, vitamin B2, niacin and vitamin B6, for which proportional bias was noticed (data not shown).

The main limitation of this study includes fairly low number of participants. Even though we were aiming to recruit the same number of male and female adolescents, male adolescents were more prone to fill out the questionnaires incorrectly and incompletely and were excluded from the study more often than the female adolescents. Additionally, female adolescents are more worried about their appearance and healthy eating habits, and therefore, more willing to participate in similar studies (38). Furthermore, one of the limitations of using 3DFR is that adolescents may change their diets, which creates a distortion of their usual intake (39, 40). However, 3DFR is the best dietary method available for the assessment of dietary intake since it does not depend on the recall (41). Yet another limitation is that we were not able to use the national food database for the assessment of nutrient intake. Although Croatia has its own national food composition database (15), it is outdated and impractical to use. However, by combining different food databases, we were able to analyse all the meals and food items that are typical for the Croatian adolescent population. The last limitation is that FFQ-m was administered on only one occasion. By administering 3DFR and FFQ-m in two different seasons, we would have been able to capture difference in seasonal food intake.

There are also some strengths of this study. Most importantly, the comparison has been made to the 3DFR, which is considered to be the best available dietary method (41). Moreover, FFQ-m and two 3DFRs were applied for each participant within one month. This approach excludes all the possible dietary changes that would have occurred if a longer time period between the two 3DFRs had passed.

In conclusion, the developed food frequency questionnaire (FFQ) and its validation study showed that the modified FFQ (FFQ-m) is a reliable tool to estimate the relative intake of energy, macronutrients and some micronutrients by adolescents in Croatia. It is affordable, quick and, most importantly, has a small respondent burden, which is particularly important for adolescent population. We recommend FFQ-m to be interview-administered, since adolescents are prone to overestimating the frequency and portion sizes of the consumed foods. Nevertheless, FFQ-m could further be adapted and validated for younger children or adults.