The Role of Various Forms of Training on Improved Accuracy of Food-Portion Estimation Skills: A Systematic Review of the Literature

ABSTRACT

This study describes the types of food-portion tools used and changes in accuracy for food-portion size estimation by adult populations after an intervention of food-portion education and training. This systematic review searched 7 scientific databases. Only internally comparable study designs were included. Studies were tabulated for nutrition- and non–nutrition-trained university students and the general population. Included studies were assessed for level of evidence and quality, including risk of bias. Thirteen studies were reviewed, with 8 targeting university students. Food type, length of training, number of tools, and the impact of repeated use on food-portion estimation were summarized. Estimation accuracy calculations across studies were not consistent, and training was found to improve portion-size estimation accuracy in the short term (4 wk). Computer-based training tools only identified for the general population were equally or less effective and shifted estimation from under- to overestimation. This review suggests that education with food-portion tools may be effective in improving estimation skills in university-recruited participants and the general population. Computerized tools for university students are required, likely combined with other tools for improved estimation accuracy. The use of food models or multiple tools is more effective until a tailored computerized solution is developed. Repeated training is needed to maintain skills over time. This systematic review was registered with PROSPERO at http://bit.ly/2mZK3u3 as CRD42016038110.

Introduction

Effective education about food-portion sizes and how these compare to recommendations is a fundamental skill of nutrition practitioners. Portion-size estimation can be completed with the use of a range of tools, including 2-dimensional food images, line drawings, food models, measuring cups and spoons, and reference objects such as a packet of playing cards. However, in practice, portion estimation is commonly completed in the absence of any aids and relies on an individual's perception, conceptualization, and memory (1–3). Dietitians may use only their hands (or their client's hands) to represent portion sizes (4). Although this is convenient, there are many opportunities for error, including the need to mentally translate the nonrelated objects to the correct portion.

Studies have found that a variety of food-portion tools can be used to improve portion-size estimation by adults (2, 5–8). Flat images usually require >1 scene to determine volume (9). Food shape remains a factor in the success of portion-size estimation, with amorphous foods, those lacking a defined shape such as pasta or rice, proving the most difficult in both a traditional and computer-based formats (2, 6, 10, 11). Estimation errors are also affected by the serving amount and how close the food is to the portion tool in both size and shape (8).

Past work that reviewed the effectiveness of portion-size tools identified a lack of detailed descriptions and an incomplete research base (12). Several studies focused on the evaluation of one tool, or a comparison of tools (2, 10, 13). A food-portion book was compared with actual food servings with no training provided. Participants were asked to serve a standard portion for a meal, which was weighed, and the following day they were asked to recall the portion using the portion book. In total, 77% of adults and 74% of children estimated the portion accurately (2, 5–8). Adding an aspect of a 3-h training session was provided to participants for 57 commonly consumed foods presented in 125 different forms. The training improved portion estimates for many foods, except for amorphous ones, which varied substantially both before and after training (2, 10, 13). Finally, the form of the training can vary whereby the concepts of food portion rather than the estimates using computer-based training can also be implemented. This was stressed as being of importance for nursing professionals, suggesting that portion training has application beyond the field of nutrition science (2, 10, 13).

Although a range of studies have explored different approaches to portion training and tested portion tools in practice, to the our knowledge there is still a need to assess the effectiveness of food-portion tools in improving adults’ portion-size estimation ability and accuracy. This systematic review addressed the PICO (Population, Intervention, Comparator, and Outcomes) question: “Do food portion tools improve food portion size estimation by adults?” The review aimed to describe the types of food-portion tools used and changes in accuracy for food-portion size estimation by adult populations after an intervention of food-portion education or training or both.

Methods

Combinations of the following search terms were used: portion size, accuracy, estimation, tools, aid, and teaching, with the refining Boolean operators of NOT children OR adolescents where required. The systematic review was registered with PROSPERO on 27 April 2016 as CRD42016038110 (available at http://bit.ly/2mZK3u3). The search was conducted with the use of the following databases: CINAHL (https://www.ebscohost.com/nursing/products/cinahl-databases/cinahl-complete), Cochrane library (https://www.cochranelibrary.com/), MEDLINE (http://www.ovid.com/site/catalog/databases/901.jsp), PubMed (https://www.ncbi.nlm.nih.gov/pubmed/), Scopus (https://www.scopus.com/home.uri), Web of Science (https://clarivate.com/products/web-of-science/), and Wiley Online Library (https://onlinelibrary.wiley.com/). An example search is as follows for the Scopus database: search terms [TITLE-ABS-KEY (“portion size”) AND TITLE-ABS-KEY (accura* OR estimat*) AND TITLE-ABS-KEY (tool* OR aid* OR teaching) AND NOT TITLE-ABS-KEY (child* OR adolesc*)]. No date restrictions were applied in line with Cochrane Review guidelines (14). The date last searched was September 2016.

Publication selection was guided by the Academy of Nutrition and Dietetics Evidence Analysis Manual and the National Health and Medical Research Manual for Systematic Review (15). Reporting followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines and statement. After the searches, studies were extracted into EndNote software (EndNote x7.5.1.1, 2016; Thomson Reuters) for data management. Duplicates were removed. Publications were screened for inclusion initially on the basis of title and abstract. Peer-reviewed, English-language studies addressing adults (aged 18–65 y) were included. Studies investigating children, adolescents, and adults aged >65 y were excluded to reduce factors affecting portion estimation, such as impaired memory (16) or lack of experience with food selection. Portion-size estimation error rates have also been higher for younger children (17) and older adults (16). Where studies extended past either side of the age criteria, they were assessed individually and included only where means and SDs indicated that 95% of the participants were within the age criteria.

Inclusion and exclusion criteria

Included studies were required to have a control group or to show pre- and postestimation results. Studies evaluating the effect of portion training on the estimation of portion sizes and studies that had a valid comparator for evaluation, such as a weighed food record, were also included. Hand-searching of reference lists was conducted for all included studies.

The key outcome for inclusion in the systematic review was testing of food-portion estimation accuracy at baseline and shortly after a training intervention. A secondary outcome of interest was a demonstrated understanding of food-portion concepts. Where studies measured estimation at numerous time points, these were also reviewed. Where studies were published in >1 publication, the earliest publication was used with all other published studies excluded. Studies were excluded if they reported on portion-size estimation error without an educational component, were not focused on adults, or addressed an education program for parents of children. Furthermore, studies specific to diseased populations were also excluded due to the contextual differences and differing motivations for learning.

Screening and extraction

Data were collated independently by one researcher (AH) and collated into a tabular form separated by those involving nutrition- and non–nutrition-trained university students (Table 1) and those involving the general population (Table 2) and were reviewed by a second researcher (YP). Data were summarized by publication year, study design, quality rating, population, setting, intervention, intervention period, follow-up, training outcome, and results.

TABLE 1.

Study characteristics evaluating training effects on portion-size estimation accuracy in university students¹

Study (reference)	Study design	Tool	Intervention	Follow-up	Training impact	Estimation measure	Post-training outcomes2
Nutrition-trained university students
Arroyo-Izaga et al. (10)	Pre- post	Real foods	3 h (visual)	1 wk	Improved	Absolute	Amorphous: 60.5% ± 21.5%; solid: 50.0% ± 17.4%; liquid: 45.5% ± 20.6% (all, P < 0.001)
Studies involving university students (nonnutrition-related majors)
Byrd‐Bredbenner and Schwartz (6)	RCT		30-min 2D/3D PSMAs	Immediate	Improved	Mean estimation	Scores: 7.1 ± 2.7 (2D) and 7.0 ± 2.6 (3D); 60% accuracy
Hausman et al. (7)	Pre-post	Foods, measuring cups, and PSMAs	Training program using stimulus equivalence	Participant dependent; average of 5 training sessions, 1- wk extension3	Improved	Mean error	16.1%; 7 of 9 maintained skills (19.84%)
Yuhas et al. (18)	RCT	Food models	10 min	Immediately	Improved; solids: 45.1%; better liquids	Mean error	92.4% (no training), 65.7% (training) (P = 0.0001); solids: 45.1%; liquids: 78.6%; amorphous: 111.6% (P = 0.05)
Bolland et al. (19)	Non-RCT	Labeled food models	10 min	Immediately; 1 wk, 4 wk later	Improved; after 4 wk, disappeared for 3 of 6 foods	NS	Trained (58.7% ± 71.9%); untrained (94.0% ± 116.1%) (P < 0.05); 1 wk (61.5% ± 60.0%); 4 wk (76.9% ± 104.5%) (all, P < 0.05)
Trucil et al. (20)	Pre-post	Real foods, measuring cups and PSMAs	Equivalence-based instruction; 1 to many training sessions	Until 90% correct for training foods, maintenance phase 1 and 2 wk after	Improved; maintained 2 wk	Absolute	9.4–19.3%; generalization foods4: 9.2–22.2%
Bolland et al. (21)	RCT	Household measures and real foods ± food models	10 min	Immediately	Improved (14 of 19 foods)	Mean error	116% ± 72.4% (untrained); 57.8% ± 58.8%(trained)
Slawson and Eck (22)	Pre-post	Food models, real foods	Standard and intense program with verbal instructions	Unspecified	NS	NS	Standard: solids (111% ± 33%, P < 0.001); liquids (100% ± 31%, P < 0.01); intense: amorphous (124% ± 35%, P < 0.08)

¹NS, not specified; PSMA, portion-size measurement aid; RCT, randomized controlled trial; 2D, 2-dimensional; 3D, 3-dimensional.

²Values shown are means ± SDs unless otherwise indicated.

³Foods estimated after the maintenance period are referred to as extension.

⁴Generalization foods were used to evaluate the extent of generalized responding from participants.

TABLE 2.

Study characteristics evaluating training effects on portion-size estimation accuracy in the general population¹

Study (reference)	Study design	Tool	Intervention	Follow-up	Training impact	Estimation measure	Post-training outcomes2
Ayala (5)	RCT	Computerized	1 h	Immediately	Improved estimation, greater self-efficacy	Mean error	Group 1: 10.2 of 14 (P < 0.01); group 2: 9.4 of 14
Howat et al. (23)	RCT	Food models ± life-sized food photographs	22.25 h over 5 wk	3 and 11 d	Improved; majority overestimated; remained for 11 d (24% ± 25%)	All measures	Photos lowered errors (12% ± 22%) compared with models (35% ± 38%)
Weber et al. (24)	RCT	Real foods and food models	1 h	Immediately	Improved	Absolute and difference	Solids: 17.8% ± 10.1% (trained, P < 0.01); amorphous: 28.0% ± 21.5% (trained, P < 0.001); improved when in cups
Martin et al. (25)	Pre-post	Photographs	HMR Calorie System3 (a 7-page system using anchor points) and photographic assessment of calorie estimation	6-mo study; baseline (weeks 1–4); training (weeks 5–8); review ongoing, end (weeks 22–24)	Improved	Mean error	19% ± 16% (P < 0.05) with foods; 19% ± 16% (P = 0.33) with photographs
Riley et al. (26)	Pre-post	Computer food-portion reference	Individualized	30 min	No more accurate	NS	Shift from under- to overestimation

¹NS, not specified; RCT, randomized controlled trial.

²Values are shown as means ± SDs unless otherwise indicated.

³HMR Weight Management Services Corp.

Assessment of bias

Each included study was also assessed for relevance, validity, and potential bias by using the Academy of Nutrition and Dietetics Evidence Analysis Quality Criteria Checklist (27). Included studies were also classified according to the National Health and Medical Research Council’s levels of evidence (15). This process was undertaken by the same researcher who extracted the data (AH), and issues were discussed with the research team (YP and AM) until consensus was reached. Methods of determining portion estimation accuracy were summarized. The principal outcome measure used for this systematic review was the difference in mean food-portion estimation accuracy between the control group (or preintervention) and the postintervention group.

Results

The database search identified 147 publications, and hand-searching resulted in 7 additional studies. A total of 13 studies were included in the final systematic review (Supplemental Figure 1). Seven studies received a positive quality rating and 6 received a neutral quality rating. Eight studies focused on university students (Table 1). The following sections outline patterns identified from the included studies.

Method of measuring and reporting portion-size estimation accuracy

The most common method used by studies of portion-size estimation accuracy was the percentage error in estimation against actual portion size by weighing of the food items with the use of either an absolute method (assuming a positive error whether an over- or underestimation is made) (5, 7, 18–20, 23) or a difference method (distinguishing overestimation as positive errors and underestimation as negative errors) (23) Weighting of the food items was achieved by using the following equation:

(1)

Two studies compared both methods (10, 24). One study (25) reported portion-size accuracy calculated by using . This approach was similar to the difference method, except that it reported the result as a number rather than a percentage. Other methods addressed the percentage accuracy (22) or accuracy ratio (26) using the median estimate. Percentage accuracy was calculated by the equation , whereas the accuracy ratio was determined by using . The mean numbers of correctly estimated food models were reported (5), although another study (6) standardized portion-size estimate scores by awarding 1 point for estimates ≤33% of actual portion size.

Training for portion estimation accuracy

Overall, 10 studies found that training improved portion-size estimation accuracy in adults. Of these, 7 studies (5, 6, 10, 18, 19, 24, 25) reported significant results for training, whereas 2 studies reported mixed results (21, 22) for some but not all food items tested in support of training.

Studies that found significant effects from training interventions reported improved mean absolute estimation errors from 92.4% (untrained) to 65.7% (trained) (P = 0.0001) (18) and 94.0% (untrained) to 58.7% (trained) (19) (Table 1). One study reported a mean ± SD reduction in estimation error from 66.40% ± 12.81% to 53.37% ± 10.15% (P < 0.05); however, the post-test was carried out 1 wk after the initial estimation (10). Interestingly, 3 studies (7, 23, 24) reported lower starting estimation errors in untrained or pretrained groups, although improved accuracy was seen for all (Table 1).

Only 1 publication involved participants who were nutrition-trained university students (Table 1) (10). The remaining studies either involved a general population or university students, with the majority nonnutrition majors (Table 2). Four studies (5, 6, 21, 23) tested the estimation accuracy of different training methods and 3 studies (5, 6, 23) found improved estimation with the use of either tool. No significant difference was found between 2-dimensional or 3-dimensional portion tools (6) (Table 1). Group-training methods using food models resulted in more correct estimations than a computer-trained group, with both outperforming the control group (5) (Table 2). The use of food models improved the accuracy of estimation, although life-sized photographs additionally enhanced accuracy (23). Groups trained with food models or household measures and real foods did not differ significantly in their estimation accuracy; however, combined results were significant for 14 of 19 foods tested compared with the untrained group (21). Computer-based training, which showed food images with superimposed reference objects, improved estimation accuracy in some, but not all, foods, whereas a shift from under- to overestimation was observed (26) (Table 2).

Training over time for portion estimation accuracy

Whether the effects of training were sustained over time was reported by only 4 studies (7, 19, 20, 23). Estimation group errors were consistently lower than untrained group estimation errors for a period of ≤4 wk after training (19). However, at an individual food level, the training effects disappeared for some foods, such as fish, meatloaf, and applesauce (19). Other studies had shorter follow-up periods and found that training benefits could remain for periods of between 1 and 2 wk (7, 20, 23). Mixed results were seen for foods used outside of the training food set, referred to as extension foods in one study, with estimation error increasing to 60.5% (from 43.7% pretraining) with one extension food and decreasing to 30.3% with another extension food. Foods estimated after the maintenance period were referred to as extension foods.

Food types and portion estimation accuracy

Three of the studies reported estimation accuracy by food type (solid, liquid, and amorphous) (10, 18, 22) and found that solid and liquid food forms showed the most significant improvements after pre-/post-training [290% ± 168% to 111% ± 33% (P < 0.001) and 67% ± 22% to 100% ± 31% (P < 0.01), respectively]; however, estimation of amorphous foods required intense training to produce nonsignificant results, although with less variability (195% ± 102% to 124% ± 35%; P < 0.08) (22). The studies used the absolute method (10) and difference in mean (18) accuracy measures, with one study not comparing estimation error for the study overall (22) (Table 1). Two studies (10, 24) reported significant post-training improvements for amorphous foods; however, this remained the least accurately estimated food overall, ranging from a mean ± SD of 28.0% ± 21.5% to 60.5% ± 21.5%.

Training of food-portion understanding

Only one study assessed food-portion understanding in addition to estimation accuracy (5). Self-efficacy was tested on the basis of confidence in judging portion sizes, which showed a significant improvement over time for the trained groups [Wilks’ λ = 0.85, F_(2,64) = 5.55, P < 0.01: effect size = 0.15]. The secondary outcome, understanding of food-portion concepts, showed a trend toward significance [Wilks’ λ = 0.916, F_(2,63) = 2.872, P < 0.064: effect size = 0.084] (Table 2).

Discussion

The aim of this systematic review was to investigate whether particular food-portion tools are more effective for food portion-size estimation accuracy in adults. This systematic review found that a range of tools and training methods are effective for portion estimation. Portion-size estimation accuracy was more likely affected by the amount of time participants were exposed to education with the tool, the complexity of food type used in the training, or the method of measurement used to calculate the estimation error.

The effectiveness of training is driven by the ability to impart portion-size knowledge to participants that can, in turn, be applied to practice. The use of measurement aids, such as golf and tennis balls, may have been effective because they are easily relatable even when used in pictorial form (6). It should be noted that the accuracy rate when using these aids was 60%, indicating substantial error still remained.

Food models or real foods were commonly used for training purposes. Real foods may be thought to be more effective, given that they mirror the target outcome. This could be due to the fact that food models may be more conducive to handling and reflects the type of education that clients are more likely to receive in practice. Interestingly, estimation accuracy was found to improve when food models were used in combination with life-sized photographs (23).

There did appear to be more benefit from physically viewing and manipulating food models rather than seeing images on a computer screen, although the level of computer interactivity was variable between studies, with one showing 2-dimensional food images only (5), whereas another allowed greater interactivity and feedback (26). The computer training studies were in a general population and either showed no training benefit (26) or a benefit that was less than that in the group trained with real foods and food models (5). None of the university-based publications included computerized components.

There are technical considerations when displaying objects to participants on computers, such as an object may appear to be different in size on screen than its actual size, which affects perception and estimation skills (26). However, the inclusion of a reference object of known size assists with this. Another way to give perspective is to include several food images on the same page (25, 28). This may assist participants to gain perspective and relativity and create relationships between the foods. Interestingly, the same study (25) reported significant results for participants; however, it did not find this effective when tested with photographs of meals. This suggests that having a 2-dimensional photograph with an entire meal may not provide enough information to translate to an accurate estimation. Technology can provide many benefits for users for efficiency and learning (29). However, the correct use of the technology needs to be employed to ensure that the user has the tools required to improve their learning. The equivalence-based instruction method (7, 20) improved estimation ability; however, no comparative statistics were provided. Equivalence-based instruction is grounded in the idea that different stimuli are considered to be equivalent for teaching the same concept—for example, a picture and a word for the same food.

The method of measurement used to test portion-size accuracy in these studies played a role in the results. Although the majority of studies chose to use an absolute method, the difference method was applied by some. Individual negative and positive estimation results cancel each other out, lowering the overall effect (10). This may also affect nonsignificant estimation findings if used as the sole calculation method (21).

The length of training is a key consideration. A balance needs to be found between length and participant burden. Two of the studies classified as short in length (minutes or hours) and one as medium length (days) reported mixed results or results suggesting that training length may be a factor (21, 22, 26). A short training session with food models or food photos would be practical in a dietetic education setting; however, it is also worth considering dedicated sessions for individuals who require more intensive training, particularly because some food types, such as amorphous foods, require more training to achieve improved estimation results (22). Two of the studies classified as having a long training length (weeks) produced promising results with small participant groups (7, 20). However, the true length of intervention was unclear in both because they were based on participant progress and achievement of target success rates. Although this, in theory, is effective, it is not efficient and it is difficult to evaluate the feasibility of implementation. The other 2 studies classified as of long duration were components of larger investigations (23, 25). In these studies, participants were exposed to other health-related exercises such as keeping food records, weekly group meetings, and/or cognitive-behavioral therapy strategies, which may have made them more conscious of foods and portion sizes.

Training accuracy in food-portion estimation is a vital skill, particularly if the skill can be maintained. When retested, generally 1 wk later, results supporting estimation accuracy remained better than untrained or baseline estimations (7, 10, 19, 20, 23). When retested at 4 wk (19) results remained significantly better than in untrained subjects for all foods combined. Individual variation did, however, indicate that the training effect had been lost for 3 of the 6 foods. Training accuracy over time was affected in one study by food choice, with an “extension food” (noodles) estimated better than baseline measures by only 3 of the 9 participants. A second “extension test” was completed by 7 of the participants (using crackers) and 6 of the estimates improved. The small number of foods tested, choice of food, and sample size may affect these results.

Food type is an important consideration when estimating portion size. The studies reviewed found that solids and liquids showed the most significant improvement after training. Amorphous foods were more problematic, and it was found that even when studies saw improved estimation, the estimates were still less accurate when compared with solid and liquid foods (10, 18, 22, 24). The implication of this is that, in dietetic practice, any amorphous foods should be paid close attention to. It may be wise to spend more time on these foods rather than others to minimize potential errors.

Only one study in this systematic review included a measure of participant understanding and found that training improved confidence (5). Although estimation accuracy is the ultimate goal, there are many other factors involved. A participant's confidence in judging portion sizes relates to self-efficacy, which is useful feedback for the effectiveness of portion education. Food-portion knowledge provides insight into an individual's broader nutrition understanding and may indicate where further education may be required.

A strength of this systematic review lies in the positive and neutral quality ratings of the individual studies. Although the overall rating cannot be clearly defined as positive or neutral, it can be said that no negative bias was identified. This systematic review was limited by the variability in populations studied, reducing the ability to make comparisons between population groups and the ability to generalize the results beyond the participant groups that were studied. Two studies compared training groups in which the groups received different types of training rather than one group that received no training (6, 24). There was limited consistency between the studies in terms of methods of calculation and reporting of outcome. As a result, comparisons of estimation accuracy could only be made for the overall training outcome. Finally, the data extraction and review of bias of the studies were also limited to the review of studies by one reviewer, with consensus about studies reviewed by a second reviewer rather than both reviewers reviewing all studies.

Six of the 13 studies were classified as level 4 evidence, given the before-after design, which affected the overall evidence base rating. The majority of the participants were female given the selective sampling of some of the studies. In addition, the majority of studies were university based and many of these used introductory nutrition courses to access participants. This may produce a sample who is more interested in general health and nutrition than the general community. The rate of participants majoring in nutrition was, on average, 10%; however, given the introductory nature of the courses it might be assumed that portion-size estimation was not an area that had been covered yet (10). University students, even if studying nutrition, have been found to have low estimation accuracy at baseline (30). Interestingly, one study (26) found that 22% of their participants had received some form of past food-portion estimation instruction; however, it was not evident from their reporting method how these estimation errors compared with others. Also of note, the same study did not find a training method that produced more accurate estimation.

This systematic review suggests that education with food-portion tools may be effective in improving estimation skills in university-recruited participants and the general population. The use of food photos has been found to increase the accuracy of food-portion estimation compared with unaided estimates (31–33), suggesting the use of image as a useful recall method. Improvement was seen for a range of tools and training methods. It appears that computerized tools were limited and food models may be a suitable tool until tailored computerized tools are developed. When using the tools, the length of intervention and food types were key factors affecting estimation accuracy. Consideration needs to be given to the length of education, given that shorter durations may hinder estimation ability. More intensive training may be needed for amorphous foods, which were least accurately estimated, and improved estimation skills were maintained over a period of time, to a maximum length of 4 wk. It is unknown whether certain tools require longer education sessions than others for similar estimation improvement or if all amorphous foods are equally difficult to estimate.

It would be beneficial for future studies to investigate the benefits of repeated exposure to training, as in a refresher course, because food-portion estimation is a skill necessary to maintain for the long term. It would also be interesting to explore whether estimation skills are aligned with self-efficacy, which was not explored in detail in this systematic review.

Notes

The authors reported no funding received for this project.

Author disclosures: AH, AM, and YP, no conflicts of interest.

Supplemental Figure 1 is available from the “Supplementary data” link in the online posting of the article and from the same link in the online table of contents at https://academic.oup.com/advances/.