Servicios personalizados
Inglés (pdf)
Articulo en XML
Referencias del artículo
Traducción automática
Enviar articulo por email
Citado por SciELO 
Citado por Google 
Similares en
SciELO 
Similares en Google 
Permalink
INTRODUCTION
One major limitation when carrying out food surveys is how to measure the size of the portions of the food consumed, mainly because its estimation depends on the respondent's short term memory 1 and the interviewer's experience. The errors that are made lead to a bias in the evaluation of nutrient intake 2, and it is therefore necessary to design tools that achieve greater accuracy in measuring each person's food consumption 2,3.
A number of different visual methods are used to help reduce the error in the estimation of food consumption, for example: the use of household measures, common objects such as dice, tennis balls, desk of cards, etc., to relate their size to the portions consumed 3,4,5. However, the most common errors in using these methods occur when estimating high-volume but low-weight portions of foods 6.
The use of real images of food portions constitutes a useful and highly effective tool to help measure the amount of food consumed 7,8,9,10,11,12,13,14. It also has the advantage of being easily adaptable to local conditions, because it is cheap, easily photocopied and portable 15 and is widely considered as being the tool which most accurately represents the actual food consumed compared to other methods 7. Although, in Ecuador, some manuals and photographic atlases of typical dishes, portions 16 or recipes that can help to measure the amount of food consumed do exist, no visual tool for measuring food consumption has yet been correctly validated and published.
The aim of this study was to check the validity of the content and perception of the images from a photographic atlas of food portions in Ecuador, which was designed as a tool to help adults estimate the weight of their food portions.
METHODS
DESIGN OF THE PHOTOGRAPHIC ATLAS
The selection of foods and prepared dishes was based on information obtained through R24h from several research projects 17,18,19 and the National Health and Nutrition Survey of Ecuador 20. The 68 food items selected were divided into different groups: dairy products, eggs, meat and meat products, fish and seafood, legumes, cereals and tubers, vegetables, fruit, cakes and desserts, oils, fats and sugar, drinks and, finally, a series of traditional Ecuadorian dishes, based on the results of the study by Sánchez-Llaguno et al. 17.
Unfortunately, at the time of designing this photographic food atlas, there were no local or regional data available for measuring the 5th and 95th percentile of the portion size consumed in the country for different foods, which might have helped us to produce a series of images based on these data with fixed increases in the portion sizes. In addition, for many foods, the proportional increase of the portion size of the servings made it impossible to show the actual size of the food consumed, and so an expert committee in Ecuadorian cuisine was called in to establish a range of typical portion sizes for a variety of individuals.
For each food item, the portion was weighed before being photographed, using a Laica Compact Ks1015 electronic scale with 1 g/0.05 oz. precision, and three to four photographs were taken showing a standard portion, as well as larger or smaller amounts, thus making it possible to measure intermediate portions between the images presented in the photographic atlas.
For the technical side of the photography, we used the recommendations set out for the picture books the PANCAKE 21 and PANEU 22 project, which has been used to compile the photographic atlases used in the ENALIA 23,24 surveys, which were validated by the European Food Safety Authority (EFSA) as part of the EU Menu project.
The technical specifications were as follows:
1. Scale and size: all photographs were taken on a scale of 1:1, with a 4-5MB photo size and dimensions of 120 x 80 cm.
2. Camera angle: the photographs were taken at an angle of 35-52 degrees. Glasses were photographed at an angle of approximately 14 degrees.
3. The tableware used was white and of the size normally used in the study area:
Flat, white plates. External diameter: 25 cm; internal diameter: 15.5 cm.
Deep, white dish. External diameter: 19 cm; height: 5 cm.
Glass: Glass. Upper diameter: 7.5 cm; lower diameter: 4 cm. Length: 11.5 cm.
Wide-topped glass: Glass. Upper diameter: 8 cm; lower diameter: 5 cm. Length: 7 cm.
Cutlery: stainless steel. Length of fork: 18 cm; length of knife: 20 cm. Tablespoon: length, diameter: 6 cm; width, diameter: 4 cm. Teaspoon: length, diameter: 4.5 cm; width, diameter: 3 cm. Coffee spoon: length, diameter: 3 cm; width, diameter: 2.5 cm.
The photographic atlas was submitted to a validation process before proceeding to its publication, as detailed in Figure 1 and below.
VALIDATION OF THE PHOTOGRAPHIC ATLAS
Study 1: evaluation by experts
Following the criteria of Hernández et al. 8, who designed a photographic food atlas for Venezuela, a qualitative validation process was carried out by a committee of recognized experts in the field of nutrition. The process of choosing the eight expert judges was based on the following inclusion criteria: a) an outstanding academic or professional career in their specific area (food, nutrition, education); b) proven interest in the health and nutrition fields; c) experience in decision-making; and d) participation in actions, policies and/or programmes linked to food and nutrition issues.
So as not to place limits on the experts' judgment when evaluating the proposed items, the qualitative Delphi technique was used to establish a dynamic process of change, feedback and decision making 25. This technique involves using a series of anonymous rounds to consult the committee of experts on different themes, with the aim of achieving a consensus, while allowing the participants maximum autonomy.
Using an evaluation format consisting of dichotomous responses and open questions for each item, they were asked to evaluate the appropriacy, relevance, design and structure, number of photographs, sequence of information, format and number of food items in the photographic atlas (first round). Next, the research team checked the contributions offered by the judges and incorporated them into the photographic atlas. The edited document was then sent out again for a second round of evaluations, together with a questionnaire evaluating the same issues as before, but using a selection of four answers and an open question for them to give their general assessment of the survey. Thanks to the broad consensus reached in this second round, it was not deemed necessary to carry out further rounds of evaluation.
Study 2: validation of food photographs
In January 2017, 61 adults (aged 18-58 years), belonging to the Pontifical Catholic University of Ecuador, were invited to participate as volunteers in the study. It was decided that the inclusion criterion for participation in the study would be for them to be students, teachers, or university service personnel who were 18 years of age or older and could read and write.
To carry out the analysis of perception, based on the comparison between a portion of pre-weighed food and its estimated one using the photographic food atlas (26), nine different food items were selected (peas, cereals, chaulafán [fried rice], carbonated soft drinks, lentils, mayonnaise, fish, cottage cheese and noodles) represented by the four different portion sizes included in the photographic atlas (except for chaulafán, represented by three images). In total, 35 portions were selected and prepared for the analysis.
The prepared dishes were identified with a code formed by a letter to identify each dish and another letter to define the portion (L, R, X, V), making sure that the code did not provide any clues to the size of the portion presented. Each participant was randomly presented with one portion of each of the different foods (nine dishes per person). None of the participants had consumed the food previously.
The estimation of the amount of food present in the dish was made by two means: a) selecting the image of reference in the photographic atlas; and b) an estimation backed up by a nutritionist, who supplied the participants with household reference measures 10.
The study design was approved by the University of Córdoba (Spain) and all the participants were required to sign an informed consent form.
STATISTICAL ANALYSIS
The first step in assessing the concordance in the estimation of the amount of food using the photographic atlas versus an estimation without it was to employ the intra-class correlation coefficient, with values > 0.40 considered to be acceptable and those > 0.75, as excellent 27, and with the Bland-Altman graphical method for all the food portions analysed 28. This graph includes a horizontal line to mark the mean difference, and two other lines known as limits of agreement, at a distance of ± 1.96 DE; the lower the range between these two limits, the better the agreement. A hypothesis contrast test between the real means and the means obtained by using the atlas or not, using Student's t-test, was also performed.
The results of the estimates made using the photographic atlas and the real measurements were then compared. As proposed by Nelson et al. 26, the percentage of the differences between the estimation using the atlas and the real quantity was calculated using the following formula:
% Differences = [(estimated-real) / real] x 100
A negative difference indicates underestimation, while a positive difference indicates overestimation of that particular portion. To test whether using the atlas significantly overestimated or underestimated portion sizes, the Wilcoxon non-parametric test was used to compare the estimated weights with the actual weights.
Finally, for each of the foods, the percentage of participants who chose the correct photograph (C), the one just above (O) or the one just below (U), was determined when comparing the foods in the photographs of the atlas, calculating the level of agreement between both measurements using the Kappa index 29. To work out the differences in estimation per sex, training and BMI, the Chi-square test was used. The statistical calculations were carried out using SPSS version 15.0 (SPSS Inc., Chicago, United States) and Epidat version 4.1 (Consellería de Sanidad, Xunta de Galicia, Spain). In all the statistical tests, the significance level used was p < 0.05.
RESULTS
EVALUATION BY EXPERTS
In the first evaluation round, the eight experts evaluated all the aspects defined positively. However, they made a series of suggestions and observations that were checked out by the research team and included in the photographic atlas: including changes in the photographs taken for the series of beverages, inclusion of 12 series of photographs (reaching a total of 80 food items) and inclusion of the nutritional value for each food ration. In the second evaluation round, the seven experts who took part considered all the issues analyzed to be either suitable or extremely suitable. After obtaining this unanimous agreement on all the aspects evaluated by the experts participating in this round and its positive general assessment, the survey was not evaluated further, and the content of this Ecuadorian photographic food atlas was considered to be validated.
VALIDATION OF FOOD PHOTOGRAPHS
After reviewing the survey, five participants were ruled out for presenting incomplete or incorrect information in their assessments. Four food ration evaluations were also rejected due to a duplicated selection of images. Finally, 500 food rations were evaluated by 56 participants, 67% of whom were female, with a mean age of 27 ± 8.7 years. The participants were classified according to their BMI as 41% normal, 41% overweight and 18% obese, and their level of education was 4% primary school, 36% secondary school, 35% university students and 25% graduate or postgraduate students.
The intra-class correlation coefficient (Table 1) produced excellent concordance values between the real and estimated portions using the photographic atlas for all foods (> 0.75), except for fish and mayonnaise, which had adequate concordance values (> 0.40). The values obtained from this same index to evaluate the concordance between the real portion and the estimate without the atlas show values of below 0.75 (data not shown), indicating little or no agreement, for lentils (0.276), mayonnaise (0.026) and noodles (0.388). The highest ICC value in the estimation without the atlas was obtained for chaulafán (0.713).
Table I Comparisons between mean actual and estimated portion sizes, mean percentage difference and intra-class correlation coefficients (ICC), using the photographic atlas (56 Ecuadorian adults, January 2017)
*% Difference: [(estimated - actual)/actual] x 100. Difference between actual amounts and amounts estimated assessed by a signed rank test. Difference is significant at †p < 0.05.
The actual food quantities (mean and SD) are shown, together with the quantities estimated with photographs, in table I. The amounts of three foods (ranging from 1.4% for pea to 6.5% for chaulafán) were overestimated, while another six were underestimated (ranging from 0.7% for mayonnaise to 36.5% for cereals).
Figure 2 shows that for the sum of all food portions estimated by each participant, the average difference between the weights of the estimated portion sizes is -20.53 g using the atlas and -201.25 g if done without the atlas, with significant differences between the estimate without the atlas and the amount of actual food present on the plate (p < 0.001). The 95% agreement limits range from -255.20 to 214.13 using the photographic atlas, whereas without the atlas they range from -1,410.37 to 1,007.87.
Figure 2 Bland-Altman plots showing the mean difference (-----------) and 95% limits of agreement (- - -) between the sum of all portions estimated and the actual portions sizes for (a) estimated with the photographic atlas and (b) estimated without the photographic atlas.
For each of the portions of the foods selected in the validation of the photographic atlas, the percentages of correctly estimated, overestimated and underestimated ones were analysed (Table 2). The assessment perception of the images corresponding to the smaller portion (portion A) were correct (80-100%) for peas, cereals, chaulafán, carbonated soft drinks, fish, cottage cheese and noodles. The worst performance was for mayonnaise, with a 6% correctly estimated and an overestimation of 88% for the size of this portion, followed by lentils, with a 50% correctly estimated and an overestimation of 43% of the portions of this size.
Table II Percentage of participants estimating the correct (C), overestimated (O) and underestimated (U) portion size using the photograph atlas with kappa agreement statistics (56 Ecuadorian adults, January 2017)
*Agreement levels: poor, k ≤ 0; slight, 0.01 ≤ k ≤ 0.20; modest, 0.21 ≤ k ≤ 0.40; moderate, 0.41 ≤ k ≤ 0.60; substantial, 0.61 ≤ k ≤ 0.80; almost perfect, 0.81 ≤ k ≤ 1.0.
In the evaluation of the middle portion (portion B), cereals, chaulafán and mayonnaise obtained a correctly estimated over 70%. Cottage cheese obtained the worst performance (18% correctly estimated). In the case of cottage cheese, the deviation was produced exclusively for the biggest portion (76%). In the evaluation of portion C, carbonated soft drinks, chaulafán and mayonnaise obtained a correct estimation of either excellent or good. Portion C of the fish was not successfully chosen by any evaluator, who chose the next portion (portion B) or portion A (54%, data not shown) on 46% of the occasions. As for cereals and lentils, the smaller portion of food was chosen.
Finally, when evaluating the largest portions represented (portion D), high correct estimations for peas, carbonated soft drinks, lentils, fish and cottage cheese were obtained. However, cereals and noodles produced a high level of deviation towards the smaller food portions.
Taking all the serving sizes for each of the foods selected for validation into account, those with a high percentage of hits were chaulafán (n = 53, 89%), carbonated soft drinks (n = 56, 86%) and peas (n = 56, 79%). Mayonnaise (n = 56, 39%) was the food with the lowest hit rate. The agreement between the image as perceived by the evaluators and the corresponding real image, measured by Cohen's kappa coefficient, was excellent for chaulafán and carbonated soft drinks (p < 0.001); good for peas (p < 0.001); moderate for cereals, fish, cottage cheese and noodles (p < 0.001); and poor or weak for lentils and mayonnaise (p < 0.001). There was no statistically significant link between the correct estimation of portion sizes and sex, nutritional status (BMI) and level of studies (p > 0.05).
PHOTOGRAPHIC ATLAS
Once the recommendations had been included and the relevant modifications made, the Photographic Atlas and Tables of Food Composition of Ecuador was published with 80 items of food, represented by 3-4 servings, which generated a total of 298 colour photographs of the different sizes of food portions with information on their composition.
Figure 3 shows an example of the photographs included in the atlas. As can be seen, the food is included with a number of portions in order of size, their net weight in grams and, in some cases, a reference to other possible foods to use their image and estimate their weight.
DISCUSSION
As far as we know, this is the first study made to develop and validate a visual tool to estimate food portions in Ecuador, with the purpose of reducing the error generated when estimating the portions consumed by participants in nutritional studies. With this tool, in addition to measuring the portions of the 80 foods represented, it is possible to choose a portion size between the different images or use one of the images to represent another product of similar proportions and appearance.
As a first step, as commented by Hernández et al. 8, it was considered to be important to obtain an assessment by experts in the field of nutrition. The observations indicated by the different judges were included, and this helped to improve the quantity and quality of the images presented in the photographic atlas. In addition, on the recommendation of the experts, the photographic atlas includes the nutritional composition for each food portion, using Nutriplate 2.0 software 30 developed by the University of Córdoba (Spain) for its calculation.
A validation process was then performed by the instrument, which, as shown by Nelson et al. 10, allows us to evaluate cognitive processes such as perception, conceptualization and memory. In this study, perception (the ability to link a real amount of food to that shown in a photograph) was evaluated, but conceptualization and memory could not be assessed because when the estimates were made, no time had elapsed between seeing the food and using the photographic atlas. This approach is, in fact, less realistic than evaluation in a 24-hour recall situation; nevertheless, it has the advantage of allowing a larger number of estimates of food portions to be made in a one-day session 31) and permits us to understand, correct and prevent the causes of error in estimating portion size when used in clinical and educational contexts, where the aim is to help subjects identify portions of a certain size in relation to a clinical objective (26).
For the validation, nine food items and the 3-4 portions represented in the photographic atlas were selected. Although the number of food portion items validated in this study is not representative of all the foods included in the atlas, we feel that we have chosen those foods which are most commonly consumed or most difficult to estimate 31. In addition, since it was a study in which the participants did not consume the food, it should be noted that it was important not to include an excessive number of food items so as not to tire the participants 26.
The results show that the use of the photographic atlas of food items estimates the amount of food presented more reliably than estimations without it. However, in our study, two of the nine food items analysed produced errors, which suggest problems for their estimation: cereals and fish. For cereals, the mean difference between the size of the serving and the estimated portion sizes was high (-36.5%). This difference is due to an underestimation of the portion, in particular the middle portion (C), where the users underestimated the portion in 75% of the estimates made. This error could be ascribed to the fact that the reduced angle in the photograph did not make the depth of the portion very clear. Nelson et al. 10 comment that evaluating the amount of cereals on a plate would be more successful if the participants had a better idea of the width of the part of the plate not covered by the cereal by reducing the camera angle when the photographs are taken.
In the case of fish, we feel that the errors produced were due to the difference in shape and size of the portion on the plate compared to that shown in the photographs. The images from the photographic atlas showed portions of fish, whereas the plate contained a whole fish fillet. This problem, where the number of food units presented differs from what is shown in the photographs, has already been commented on by other authors 32.
The other foods showed mean differences of -6.2% to +6.5%, and these ranges are similar to those found by other authors. Huybregts et al. 15 reported that the mean differences between the actual size of the serving and the estimated portion size were between -8.4% for couscous and + 6.3% for liquid sauces, while Turconi et al. 13 found mean differences of between -2.7% for bread and +15.9% for vegetables. The range was wider in the study by Robson et al. 33 with differences of between +37.6% (muesli) and -23.3% (crispy rice) and in Venter et al. 5, who found +54.0% for bread and -35.3% for tomato sauce and onions.
The fact that some food items seem to be more difficult to estimate accurately than others is a common finding 31,33. Results from previous studies based on perception found that the estimation varied widely between foods 10,33,34,35,36,37. In the study by Keyzer et al. 38, in which the three cognitive processes of perception, memory and conceptualization were evaluated, errors were found with some food items (e.g., beverages, margarine and bread) while, at the group level, the estimation of food quantities was acceptable.
The different directions and variations in the error associated with different food items highlighted the importance of measuring the extent to which they agreed and not just the mean differences. The proportion of portions estimated correctly was 66%, which was close to the results of previous studies: Ovaskainen 50% 37, Amougou 77% 39, Souza 68% 40, Venter 68% 5, Trolle 50% 23, Lucas 49% 3 and López 60.9% 2. Like other authors 32,37,41, the smallest portion produced the best mean values for correct estimation (80%). It is logical for the smaller portions to obtain higher values of overestimation due to the large amount of images available for selecting a larger portion. However, only in the case of carbonated soft drinks and mayonnaise did we observe the trend found by other authors, where the sizes of the small portions are overestimated and those of the large portions are underestimated 10,11. Other studies showed that the food items served in sauce, without a clearly defined shape, were more difficult to estimate 5,12,15,31. This situation was reflected in our study in the case of mayonnaise, which registered a low percentage for accuracy in estimating the portion (only 39% of estimates were correct).
Some authors suggest that the ability to evaluate the amount of food on a plate can be influenced by certain characteristics of the participants such as age, sex, body mass index or level of studies 10,29,42. However, as in our study, other authors did not find any differences in the evaluation of the portion according to these variables 5,7,13,33,43. Nevertheless, although the subjects chosen represent a fair sample of the adult population as regards literacy 1,10,31,35, we feel it would be better to replicate this study in a population with a low level of education, below adult age or older adults.
After the (qualitative and quantitative) analysis, we can conclude that the Photographic Atlas and Tables of Food Composition of Ecuador 44 developed in this study is a useful and effective tool that allows us to make a more accurate estimation of the amount of food in a portion.
