Introduction

Monitoring existing public policies and programs in nutrition requires knowledge of the foods and amounts consumed. From birth to adulthood, recall of dietary intakes gives useful information for the prevention or treatment of chronic health diseases and maternal and infant health promotion, among others. The accuracy of intake recalls is limited by factors such as the respondent’s memory, ability to estimate food amounts, food types, age, education level, and other variables, resulting in a gap between the actual intake and the reported information in the survey¹. Visual aids, such as household measures, photographs or food replicas are helpful in the conceptualization of portion sizes and quantities^2-7. Photo atlases have been recognized as reliable and easy-to-use tools for surveys, providing the opportunity to show foods in different serving sizes according to the types and amounts commonly consumed in a region or country. Additionally, photos are simpler to transport than three-dimensional models, which is crucial for national face-to-face surveys⁸. Thus, a validated food atlas including commonly consumed foods (ideally by age and gender) is particularly valuable for any country^1,9.

The First National Health and Nutrition Survey (ENNyS) conducted in Argentina between 2004 and 2005 collected dietary data by using 24-hour dietary recalls. A paper-based atlas of photographs for Argentinian foods was used to help in estimating portion sizes¹⁰. The Second National Health and Nutrition Survey (ENNyS2) conducted in Argentina between 2018 and 2019¹¹, also collected dietary data by using 24-hour dietary recalls, but the development of a validated digital atlas was considered due to several vantages. It would enable the inclusion of a vast collection of high-quality colour photographs at low cost, facilitate portability with reduced burden on field researchers and allow periodic updates. In addition, it may enable free web downloads. The objective of this article is to describe the development of the Digital Photographic Atlas of Argentinean Foods (AFDAA) used for ENNyS2 and to assess its accuracy in visually estimating the amounts of foods commonly consumed in Argentina.

Methodology

Development and design

Photograph series were planned for commonly consumed foods in Argentina following the recommendations of Nelson et al.¹². For most of them, information about usual food intake was obtained from the first ENNyS, which was conducted in 2005 and was the only national and representative data at the time^13,14. All the foods consumed by at least 1% of the population were considered for inclusion in the AFDAA. That initial list was completed with other foods and dishes of nutritional importance at a population level.

The AFDAA contained both foods and dishes prioritizing foods difficult to quantify by a description or other means (i.e.: whole fruits)¹². Whenever possible, servings were presented in 4 pictures representing different weights (size 8x5 cm). In most cases, the first photograph represented the 50^th percentile of the intake of children between 6 and 24 months; and the other 3 pictures represented the 25^th, 50^th and 75^th percentiles of adult intakes (Figure 1). When 2 percentiles were too similar, intermediate points were selected taking into consideration an adequate visual perception and the plausibility of the intake. In the case a percentile represented a portion size too small or too big for visual perception, its weight was also modified to make the pictures more meaningful. For some foods (i.e.: cookies) or preparations (i.e.: jello), it did not make sense to take more than one picture, therefore several serving sizes were placed into the same image and the different serving sizes were identified with letters to avoid selection bias¹².

Figure 1. Example of series of 4 dishes which represent different amounts of rice stew. 

The atlas contained a final count of 292 photographs classified according to the Argentinean Food Guide (vegetables and fruits: 15 series; legumes, cereals, potatoes, bread and pasta: 34 series; milk, yoghurt and cheese: 5 series; meats and eggs: 15 series; oils, nuts and seeds: 8 series; optional foods, sweets and fats: 20 series, and water and beverages: 6 series)¹⁵. In supplementary (https://www.renhyd.org/renhyd/article/view/1925/1163) material we present the full set of pictures included in the AFDAA used during ENNyS2.

Photographs were taken in standardized and controlled conditions following the recommendations available on this matter^12,16. Several working days were needed to photograph all the foods and dishes prepared by a professional team right before the shooting session. A professional photographer was in charge of the photograph setting following precise instructions to ensure standardized lighting, angle, and distance. All the pictures from the same series were taken with a fixed camera maintaining the shooting angle (45 or 90 angle degree depending on the type of food), the distance from the plate and the general layout. Plates and surfaces for the photographs were carefully selected to guarantee good color contrast. Knives, forks, or spoons were placed on the sides of the plates in most series to improve the real perception of sizes and shapes (see figure). Each food to be photographed was weighed using a digital electronic scale model Sistel Clipse 5 V2,5 kg (precision 1 g).

Results

After three validation sessions, a total of 277 participants were interviewed. Sociodemographic information is presented in Table 1. Participants` mean age was 26.8 years. A total of 2,761 observations were made across the three validation sessions for 30 different foods or dishes.

The real mean estimated weights and average weight differences of each evaluated food/dish are shown in Table 2. Fruit rings and gnocchi presented the lowest and highest average percentage difference, -1.2% and 58.5% respectively. Over half (17/30) of the evaluated foods and dishes had an average percentage difference equal to or lower than 20%. Mean differences were ≤10% for 8 foods/dishes (grapes, lettuce and tomato salad, peas, cereal fruit rings, french fries, farfalle with and without sauce and breaded meat), between 10 and 20% for 9 foods/dishes (grated carrot, tomato fresh cubes, beans, mashed potatoes, potato boiled in cubes, oil, quince paste, rice and rice stew); over 20 and up to 30% for 3 foods/dishes (fruit salad, cacao powder and jam) and over >30% for the rest of the foods/dishes.

The range (minimum and maximum) of percentage differences of estimated weight differed by food or dish. The largest range was for cereal fruit rings (-58.3 to 650.0) and the smallest for French fries (-28.4 to 20.9).

As per the average percentage difference, 14 foods/dishes quantities were underestimated and 8 were overestimated. For underestimation, the largest mean difference was -56.5% (grated cheese); for overestimation the largest mean difference was 58.5% (gnocchi).

Among the 30 foods/dishes evaluated, 19 had 50% or more observations with differences within 30% of the real weight; out of those 30, gnocchi had the lowest proportion and french fries had the highest (2.73% and 100% respectively). Those 11 series that had more than 50% of the observations outside the +/- 30% range were excluded from the atlas and are shown in (Table 3).

Discussion

We developed a photographic atlas of Argentinean foods using dietary data from the first ENNyS. A set of 30 photographs of foods and dishes was evaluated to visually estimate amounts of foods. The average percentage difference observed was less than 20% for 57% of the evaluated sets (17/30) and at least 50% of the observations were within ±30% of the real weight range for 63% (19/30) of the evaluated foods and dishes. Finally, pictures outside that range (11 sets) were eliminated from the tool used for ENNyS2.

Comparing results among studies is challenging because validation procedures, sample sizes, types and numbers of foods validated, quality and quantity of pictures displayed, skills tested, and other characteristics widely vary across studies; therefore, any generalization has to be made with caution^5,19. For example, some studies compared pictures with real intake²⁰, while others used experimental designs¹⁸. Some studies tested a limited number of pictures and others a large number (from 6 to 45)^16,21. Some studies allowed participants to choose one particular picture¹⁸ while others offered a continuum of weights to select from¹⁶. Our study focused on depicting a realistic scenario anticipating the use of this tool in a 24-hour dietary survey during the ENNyS2 the participants were given the freedom to choose any amount of food to describe the plates shown. Moreover, none of the real plates had similar characteristics or weights to any of the pictures to reduce selection bias as described by other authors¹⁸.

The proportion of sets with differences within the range of ±30% stipulated for this study was generally in agreement with the proportion of differences considered adequate or acceptable in several studies^3,4,18,22,23. Not many studies reported differences as percentages. Nevertheless, Frobisher in his study with a similar population found a range between -11% and 73%²⁴; in our study the percentage of difference was between 1.2% and 58.5%.

By observing the range of the differences between the real and the estimated weight, it seems evident that there is a large variability in individual capability of photograph perception; however, the AFDAA proved useful to estimate population mean intakes, as pointed by other authors as well^18,25.

Tendency for over or underestimation was different among reports. Vereecken in a study with a similar design but conducted on a different population, showed a comparable percentage of estimations within 10% difference with the real weight to our study (29% and 26% respectively)²². Robson did not show a clear tendency to over or underestimation²⁶, Frobisher reported a tendency of overestimation data²⁴, lastly, Lazarte reported a tendency of underestimation across subjects as our findings²⁷.

The over or under-estimation may differ according to the type of food. Some studies reported similar foods to the ones tested for the AFDAA that make comparisons possible; Faggiano and coworkers found that rice and mixed salad were underestimated by participants and our data were similar to those findings; however, carrots and potatoes performed in opposite directions in both studies²⁸.

Two previous studies assessed the performance of food atlases in Argentinean populations^3,4. Overall, our study found a similar percentage of acceptable sets to López´s (63% and 56% respectively). However, in their study, breaded meat was one of the worst estimated foods, whereas in our study its average percentage difference was less than 10% and more than 50% of observations were between a range of differences equal to or less than 30% of the real weight. On the other hand, in both studies, rice had a similar percentage of correct answers. As for tomato, our results were worse than López; however, this author did not specify the way the tomato was served. This difference could be due to the different presentations of the tomato in each study³. In Navarro´s, 8 out of the 118 items tested (potato, french fries, mashed potato, tomato, grated carrots, peanuts, rice, and gnocchi) are comparable with our study. Their general agreement percentage (51%) was in concordance with ours and the foods with a better agreement were similar to ours except for tomato, peanuts and gnocchi⁴.

The validation process permitted to identify of pictures to be included in the tool used in the 24HR during ENNyS2. There is agreement that the use of tools, such as photographs reduces misinformation during dietary recall²⁹; however, caution should be given to the fact that error can be introduced if a biased tool is used³⁰. Therefore, our decision to eliminate pictures that did not meet the minimum criteria seemed appropriate for the intended use of the tool.

Some limitations of our work include the small number of foods evaluated (30 out of 103 series). Even though it may be desirable to evaluate more pictures, the whole process is not only expensive, but also time consuming as it requires a large amount of equipment and personnel. Also, we cannot rule out the potential bias of central tendency for some of the sets of photographs for which an uneven number of pictures was shown. Even when we made all efforts to have four pictures of each meal to avoid that central tendency, in some cases it was not possible given that the grams estimated from percentiles of intake from the first ENNyS were identical⁶. Another limitation of this study is that participants were only adults, mostly educated. Therefore, it remains to know how useful this tool is in helping children and teenagers to report their intakes, as well as other populations. Also, since the participants had a real amount of food in front of them to make the comparison, skills like conceptualization and memory were not tested in this study. Given that some authors also recommend testing atlases in the same conditions that they will be used, it is still necessary to validate the AFDAA in a context like a 24-hour recall¹². Finally, given that weight and height data were self-reported and voluntary, some were missing limiting the possibility to evaluate the performance of the pictures according to BMI.

To our understanding, the study design has some strengths as well. First of all, the AFDAA was built with intakes from national data, providing the tool with a real range of amounts of foods consumed. Another strength is that during the evaluation sessions, participants were allowed to select a particular picture or any amount in between pictures as it would happen in a real-life situation. Finally, since this atlas is digital, it is possible to eliminate some series based on their performance during the evaluation sessions.

Conclusions

The information used to construct the AFDAA, the depuration process for the pictures that did not perform as expected, and the general results of the series during the evaluation sessions make this atlas a valuable tool for the estimation of group intakes. This tool developed and validated in this study was available for the Argentinean National Nutrition and Health Survey. Also, given that this is the first digital and free-to-use atlas in our country, it is expected to be a useful tool in other research settings in the future. Further research should focus on testing this tool in a more diverse population sample, as well as the inclusion of new photographs, to increase its validity and scope. Finally, it would be advisable to evaluate the perception of the usefulness of the tool among users like nutrition-related professionals.