Validation of a Brief Vegetable Variety Questionnaire to Assess Diet Quality

Abstract

Background

Obtaining estimates of food intake via the 24-h dietary recall is laborious and expensive. Simpler methods for estimating dietary quality are needed in clinical settings and for evaluating nutrition education interventions.

Objectives

The objective of this study was to validate a simple, pictorial tool for assessing diet quality using vegetable variety as the indicator.

Methods

The My Veggies tool, containing 28 pictures of vegetables and items about food preparation, eating at home, and eating fast food, was administered to 258 healthy adults aged 18–65 y. To assess convergent validity, 3 dietary recalls were used to calculate the Healthy Eating Index (HEI) 2015 and dietary energy density (DED), both of which serve as measures of diet quality. To determine criterion validity, height, weight, and body composition (as measured by dual-energy X-ray absorptiometry) were assessed. Using cluster analysis, responses to My Veggies generated 3 vegetable variety groups: high variety (HV; n = 84), medium variety (MV; n = 107), and low variety (LV; n = 67). Values for HEI, DED, body mass index (BMI) (in kg/m²), and body composition were compared across the variety groups to determine whether differences existed.

Results

Demographic characteristics did not differ between the variety groups. The LV group had lower total HEI score (median = 54.4; IQR = 15.6) than HV (62.1; 17.3) and MV (68.0; 18.6) groups (P < 0.0001 for both). The LV variety group had a higher DED score (median = 2.6; IQR = 0.4) than the HV (2.2; 0.6) and MV (2.3; 0.5) groups (P < 0.0001 for both). Both HV and MV groups reported cooking from scratch more often than LV group (P < 0.0001), and HV and MV reported eating out and fast food less frequently than LV (analysis of variance, P < 0.0001; post hoc mean comparisons of HV and MV compared with LV, P < 0.0001). The LV group had higher BMI (P < 0.003) and body fat percentage (P < 0.005) than MV and HV groups.

Conclusions

This study provides evidence of the convergent and criterion validity of a simple, pictorial assessment tool for evaluating diet quality based on vegetable variety. Overall, adults with lower vegetable variety scores had lower diet quality and higher body mass index and fat mass. My Veggies can be used by health professionals as an evaluation tool for community nutrition education interventions or as a screener for counseling in a clinical setting.

This trial was registered at clinicaltrials.gov as NCT02367287.

Introduction

Vegetables of all types are an important core element that contributes to a healthy diet pattern. The Dietary Guidelines for Americans state that a healthy eating pattern requires not only an increase in vegetable consumption across all vegetable groups, but also that these nutrient-dense vegetables replace more energy-dense foods. Vegetables provide unique health benefits, including nutrients and bioactive compounds such as antioxidants, carotenoids, and flavonoids [[1], [2], [3], [4]]. Specific bioactive compounds in vegetables can vary greatly [[5], [6], [7]]. Hence, the recommendation to choose a variety of vegetables is critical to supporting adequate nutritional status and health. For this reason, the current MyPlate consumer guide encourages consumers to “vary your vegetables.”

Our laboratory published a study on low-income women, demonstrating that vegetable variety may be a useful focal point for assessing diet quality [8]. In that study, women who typically consumed ≥10 different vegetables per week ate a larger quantity of vegetables per day and had higher diet quality scores, as measured by 2 variables: total Healthy Eating Index (HEI) score and dietary energy density (DED). To move this research forward, we postulated that a simple instrument featuring a variety of vegetables could be a useful tool for estimating diet quality rather than the more arduous 24-h dietary recalls. Further, for application to a population with limited resources and literacy, and often with English as a second language, pictorial representation is preferable to reduce cognitive burden [9,10]. This instrument could be used by food and nutrition education programs for screening and assessing intervention impact, and by health clinics to guide counseling efforts.

With this in mind, we developed a tool that queried vegetable use with high-resolution photos of different vegetables. Respondents were asked to indicate the vegetables they included in their diet during the past week. The prototype included 2 or 3 different photographs of each vegetable to make them easily identifiable.

In the study described herein, our goal was to determine whether a simple, pictorial assessment tool, My Veggies, can measure vegetable variety as a surrogate for diet quality. We explored whether the My Veggies tool can identify groups of people with diets of higher or lower quality, as indicated by HEI scores, thus testing convergent validity. To establish criterion validity, we tested whether groups with higher-quality diets had better anthropometric values. The sample population used was part of a larger phenotyping study [11] that measured several metabolic and physiological variables (clinicaltrials.gov NCT02367287). Use of My Veggies was not part of the original trial protocol, but was added shortly after the phenotyping study commenced, once institutional review board approval was obtained.

Methods

Participants

People living near Davis, CA, United States were invited to participate in this exploratory study if they were aged between 18 and 65 y. A total of 258 healthy adults signed informed consent forms, completed dietary recalls, the My Veggies tool, and anthropometric and body composition measurements. A consort diagram of recruitment, screening, enrollment, and study completion is shown in Supplemental Figure 1. The protocol received ethical approval from the Institutional Review Board of the University of California, Davis (approval number 691654).

Description of the vegetable variety instrument “My Veggies”

Photographs (n = 28) of 25 single vegetables and 3 vegetable groupings, along with questions (n = 3) about food preparation and eating out, were included. The front page of the tool is shown in Figure 1 [12]. The tool was developed with a convenience sample of 16 parents who were primary food purchasers and preparers in their households, recruited from 4 federally funded preschool sites (n = 4). Using an iterative process, parents answered questions about vegetables their families consume and how they buy, prepare, and serve them. They reviewed and selected color photographs of each vegetable type. Eight versions were created until the final version, with 28 vegetable types, was reached. The vegetable photographs were obtained from 3 USDA-funded websites with permission and the researchers’ food photograph library. Refer to Supplemental Table 1 for content of selected nutrients in the pictured vegetables.

Figure 1.

Front page of the My Veggies tool. A link to the complete tool can be found elsewhere [12]. A list of the pictured vegetables is provided in Supplemental Table 1.

Convergent validity using dietary assessment

Recent dietary intake was assessed using the Automated Self-Administered 24-h (ASA-24) recall from the National Cancer Institute of the NIH [13]. A training session was conducted on-site during the first study visit to explain the process for entering foods and beverages, and then a “practice” recall was conducted by the participant to ensure that they understood how to use the website. Following this training, participants were sent unscheduled email prompts to complete Self-Administered 24-h dietary recalls on 2 weekdays and 1 weekend day to capture individual variability, which is known to be high from day to day [14]. The participants logged in to the website at home to complete these recalls between the 2 scheduled study visits, which were ∼10 to 14 d apart. The recall period was from midnight to midnight the day prior to receipt of the prompt. Study staff checked entries before uploading them to a secure data capture system (REDCap; University of California) [15,16]. The 3 ASA-24 recalls recorded at home were averaged to calculate the HEI-2015 scores and the energy density of the diet. HEI scores were calculated using the SAS code provided by the NCI. DED, expressed as kcal/g, is a measure of available energy per unit weight and was calculated from the reported diets. In this study, DED was calculated based on foods without beverages, as this approach has been used in a previous study [17].

Criterion validity using anthropometry

Excess clothing (shoes, hats, jackets, scarves, etc.) and equipment (mobile phones, keys, backpacks, etc.) were removed, and height and weight were measured in duplicate to the nearest 0.1 cm and 0.1 kg, respectively. Measurements were repeated a third time if the difference between measurements was >0.2 kg or 0.2 cm, respectively. Permanently mounted stadiometers were used to measure height, and electronic scales (Scale-tronix 6002, Welch Allyn or Tanita BWB-627A, Toledo Scale) were used to measure weight. Whole-body dual-energy X-ray absorptiometry was performed by CA-licensed limited-use X-ray technicians using Hologic Discovery QDR Series 84994 instrumentation. The whole-body scan captured total lean mass, total fat mass, and percentages of lean and fat mass. Further details are available elsewhere [11]. Demographic descriptors of participants are displayed by sex in Supplemental Table 2.

Statistical analysis

Descriptive statistics for demographic variables included means, SDs, SEMs, and percentages. Vegetable pictures were organized by MyPlate categories as follows: red/orange, dark green, beans, starchy, and other vegetables. To test convergent validity, K-means clustering [18] was used to identify clusters based on variables from My Veggies, including the total number of vegetables chosen, percentage of pictures chosen, question about cooking from scratch, question about eating out, question about eating fast food, total count of red/orange vegetables, total count of dark green vegetables, total count of beans, total count of starchy vegetables, and total count of other vegetables. This statistical procedure allowed detection of patterns in these responses. Scree plots were used to indicate the best number of clusters, and 3 groups, representing high variety (HV), medium variety (MV), and low variety (LV), best fit the data (Supplemental Figure 2). Clusters were compared with respect to demographic, anthropometric, and questionnaire-related variables. We used chi-squared tests of association for categorical variables and analysis of variance or Kruskal-Wallis test when residual analysis indicated nonnormality and/or nonhomogeneity in within-cluster variances. Variables such as HEI total vegetable subscore, with a ceiling value of 5, had a large number of subjects at or near 5. They were dichotomized into 2 values and analyzed as binary variables (5 compared with <5). Other subscores from the HEI-2015 were analyzed using either chi-squared or Kruskal-Wallis test, and, if significant, pairwise comparisons were performed.

Criterion validity was assessed using analysis of variance (ANOVA) to compare BMI, weight, and body composition measures across subjects in the LV, MV, and HV clusters. Contrasts, including pairwise comparisons and combining HV + MV compared with LV alone, as well as combining LV + MV compared with HV alone, were carried out using Bonferroni adjustments or rank-based Bonferroni adjustments for non-normal variables. For pairwise comparisons of proportions, we used Holm-adjusted P values. Finally, we inspected the contributions of each cell in the contingency table when using chi-squared association test to investigate each cluster’s contribution to the chi-squared statistic.

Results

Commonly consumed vegetables

Of the 28 vegetable pictures, tomatoes, potatoes, corn, garlic and/or onion, and carrots were chosen with the highest frequency. Rankings of frequency for all participants (males and females) and the vegetable variety clusters are available in Supplemental Table 3.

Applying K-means clustering to the dataset yielded 3 groups: HV (n = 84), MV (n = 107), and LV (n = 67). The physical characteristics of the participants in the groups are shown in Table 1. The mean ± SD energy intake of the vegetable variety groups was 1650 ± 317, 1678 ± 294, and 1796 ± 332 kcal/d for the HV, MV, and LV, respectively (ANOVA, P < 0.012), with higher intake in the LV group. We found significantly higher energy intake in the LV group than in the MV and HV groups (P < 0.05).

Table 1.

Anthropometric characteristics and demographic variables of vegetable variety groups formed by K-means cluster analysis¹

	High variety n = 84	Medium variety n = 107	Low variety n = 67	P value or χ2
Males:females	42:42	50:57	37:30	0.552 2
Age, y	43.0 ± 13.5	42.1 ± 13.3	38.1 ± 13.5	0.068
BMI, kg/m2	27.3 ± 4.7	27.0 ± 4.9	29.6 ± 5.6	0.003
Lean body mass, kg	51.9 ± 11.5	52.5 ± 11.9	54.3 ± 12.3	0.461
Body fat mass, kg	23.9 ± 10.1	23.0 ± 10.8	28.6 ± 12.8	0.005
Body fat, %	30.0 ± 9.7	29.1 ± 10.5	32.5 ± 9.6	0.089
Racial ethnicity, %				0.109 2
White/Caucasian	64	69	45
Asian	7	10	16
Black/African American	4	4	6
Hispanic/Latino	16	9	21
Middle Eastern	1	2	0
Native American	0	0	2
Mixed	8	6	9
Education, %				0.117 2
Doctoral degree	17	11	12
Master’s degree	25	17	24
Bachelor’s degree	39	41	28
Some college	18	24	31
High school or less	1	7	6
Household income, %				0.693 2
>$100,000	41	29	28
$80,000–$99,000	11	12	5
$60,000–$79,000	13	14	10
$40,000–$59,999	11	13	18
$20,000–$39,999	13	13	16
$0–$19,999	7	11	16
Declined to respond	4	7	6

¹Values for the anthropometric variables are mean ± SD. P values are based on analysis of variance for 3 groups.

²The proportion of males to females, racial-ethnic groups, education level, and household income by vegetable variety groups was tested using the chi-squared test of association.

The total HEI score differed between clusters (P < 0.0001). The LV group had lower total HEI score (median = 54.4; IQR = 15.6) than HV (62.1; 17.3) and MV (68.0; 18.6) groups (P < 0.0001 for both) (Figure 2). The HEI subscores for total vegetables and greens and beans followed a similar pattern to the total HEI scores (P < 0.0001; Figure 2). In addition to total vegetables and beans and greens, the HEI-2015 includes 11 other subscores. Of these subscores, pairwise comparisons revealed that the MV group consumed more total fruit than the LV group (P < 0.011), and both the HV and MV groups consumed more whole fruit than the LV group (P < 0.025). The whole grains subscore was higher for the HV and MV groups than for the LV group (P < 0.05), as was the saturated fat subscore (P < 0.05). Finally, the seafood/plant protein subscore was higher for the MV group than for the LV group (P < 0.008). No other subscores differed significantly between the variety groups. All subscore values for each variety group are shown in Supplemental Table 4.

FIGURE 2.

Healthy Eating Index (HEI) scores for the vegetable variety groups. Top panel: the total HEI scores, which range from 0 to 100, are depicted using box-and-whisker plots, with red box representing the high variety group, green box representing the medium-variety group, and blue box representing the low-variety group. Dots represent outliers. Analysis of variance (ANOVA) indicated that the group means differed (P = 0.0001). Post hoc comparisons with Bonferroni correction indicated that the low-variety group differed from the high- and medium-variety groups (P < 0.0001 for both). Matching lowercase letters next to the boxes indicate that values did not differ, whereas different letters indicate differences. Bottom panel: the HEI vegetable subscores, which range from 0 to 5, are depicted using box-and-whisker plots, as in the top panel. Subscores differed between groups, and the high- and medium-variety groups had higher scores than the low-variety group (ANOVA, P = 0.0001; mean comparisons, P < 0.0001 for both), as indicated by lowercase letters.

DED differed between groups (P < 0.0001). A lower value represents a healthier diet than a higher value. The LV variety group had a higher DED score (median = 2.6; IQR = 0.4) than the HV (2.2; 0.6) and MV (2.3; 0.5) groups (P < 0.0001 for both) (Figure 3). Thus, the higher energy density of the LV group indicates that this group had a lower diet quality than the HV and MV groups.

FIGURE 3.

Dietary energy density values of vegetable variety groups. The high-variety group is represented by the red box, the medium-variety group by the green box, and the low-variety group by the blue box. Analysis of variance indicated that the group means differed (P = 0.0001), and post hoc comparisons using Bonferroni correction indicated that the low-variety group differed from the high- and medium-variety groups (P < 0.0001 for both). Matching lowercase letters next to the boxes indicate that values did not differ, whereas different letters indicate differences.

The HV and MV groups cooked from scratch more often than the LV group (P < 0.0001). The HV and MV clusters ate out and consumed fast food less frequently than the LV cluster (ANOVA, P < 0.0001; post hoc mean comparisons of HV compared with LV, P < 0.0001, and MV compared with LV, P < 0.0001). Results by group are illustrated in Figure 4.

FIGURE 4.

Eating behaviors captured by the My Veggies tool for the vegetable variety groups are depicted as red box for high variety (HV), green box for medium variety (MV), and blue box for low variety (LV). The panel on the left depicts responses to “I cook from scratch,” and the vertical axis values are from a Likert scale ranging from no (score = 0) to some days (score = 1), most days (score = 2), almost every day (score = 3), or every day (score = 4). The middle panel depicts responses to “I eat out ___ times a week” and the right panel depicts responses to “I eat fast food ___ times a week.” Responses to both questions ranged from 0 to 7 times per week. Each behavior was tested for normality and found to be non-normal; thus, the data were analyzed using the chi-squared or Kruskal-Wallis test, and Bonferroni-adjusted values indicated that HV + MV > LV. Matching lowercase letters next to the boxes indicate that values did not differ, whereas different letters indicate differences.

Criterion validity

Body fat mass and BMI differed between groups (P < 0.005 for both variables). The LV group had higher weight, BMI, and body fat mass than the other 2 groups (P < 0.05) in post hoc pairwise comparisons (Supplemental Figure 3).

Discussion

The results of this study provide evidence of the convergent and criterion validity of My Veggies, a simple, pictorial tool for assessing diet quality. The My Veggies tool identified participants who clustered together by the number of vegetable types they ate per week. These identified clusters, or vegetable variety groups, had distinct levels of diet quality, as measured by the HEI and DED. A pairwise comparison supported our hypothesis that those in the LV group had overall worse diet quality. Also, we found that the HV and MV groups had better BMI and body fat mass, providing evidence of criterion validity with objective anthropometric measures. These findings indicate that My Veggies can be used instead of 24-h dietary recalls in settings where a quick assessment of diet quality is desirable. The My Veggies tool is easy to administer and requires only that the educator/counselor follow the simple instructions and explain to their clients to check off pictures of the foods they ate in the last week. The use of pictures helps clients complete the tool, which takes only ∼10 min. Practitioners do not need to explain portion sizes or food amounts, spending less time on assessment and more on counseling or education in intervention settings.

In this study, HEI represents how well a diet conforms to the recommendations in the 2015 Dietary Guidelines for Americans. The total score can range from 0 to 100, with scores between 90 and 100 indicating a diet of excellent quality and scores below 60 indicating a poor-quality diet [19]. Based on trends in diet quality from the NHANES dietary survey “What we eat in America,” the total HEI score for adults averaged 59, indicating that the diets of American adults need improvement [20]. In the present study, the LV group had a median total HEI score of 55, suggesting a poor-quality diet. The HEI subscores, including total vegetables, greens and beans, total fruit, whole fruit, whole grains, saturated fat, and seafood/plant proteins, were also inferior to those of the LV group, suggesting that other dimensions of the diet were mediocre compared with the HV group.

DED has been used in research to designate diets that will prevent obesity [21]. DED can be calculated using only solid foods, excluding beverages, or using all foods and beverages consumed [17]. However, there is no consensus on this approach [22].

This study had many strengths, including the use of a combination of standard dietary data collection protocols and robust objective measures, including direct measurement of height, weight, and body composition, implemented by experienced technicians.

The concept of developing a simple pictorial assessment tool to estimate diet quality is novel. However, there are challenges associated with its validation. First, the use of the ASA-24 to estimate diet quality may have resulted in underreporting or misreporting [23]. Although obtaining multiple 24-h recalls can alleviate potential related bias, the magnitude of misreporting has been found to be consistent across serial recalls [24]. Nevertheless, the recalls we collected were subjected to “data cleaning” to identify misclassifications and better describe actual foods individuals ate [25]. In contrast, using My Veggies eliminates the need for multiple recalls and for a dietary technician or professional, such as a registered dietitian, to administer the tool with fidelity and analyze data accurately. My Veggies, on the other hand, is self-administered as a paper-and-pencil instrument and can be self-scored. Second, because My Veggies is not a quantitative assessment of vegetables, it eliminates the need to estimate portion sizes, which are known to be a significant source of error in reporting dietary intake [26,27]. My Veggies was designed to reduce respondent burden and to accommodate a wide range of literacy levels, making it convenient to use in community and clinical settings.

We used HEI and DED as standards for diet quality because they were used in our previous work; however, they are not necessarily gold standards, and numerous alternatives are available, as listed in Arimond and Deitchler [28]. IIn general, the literature defines diet quality as comprising 2 elements: a healthy diet pattern and estimation of how this pattern affects chronic disease risk. Although there is an abundance of studies demonstrating one or both attributes, the consensus is that the best standard for assessing diet quality depends on the research question.

This study used a convenience sample, which limits generalizability and external validity. First, selection bias is possible due to the nonprobabilistic nature of our sample. Second, because My Veggies did not include an exhaustive list of vegetables, the validity of the tool may not hold in situations where the included vegetables differ from those readily available in geographic areas or consumed by demographic groups not represented in our sample. For instance, other types of vegetables may play a more prominent role in providing variety in different settings, such as the popularity of okra in the Southern United States and Kohlrabi in the Northern United States. We recommend further testing My Veggies with more geographically and demographically representative groups.

In conclusion, we provide evidence supporting the validity of My Veggies for assessing diet quality. Participants who reported consuming an LV of vegetables using My Veggies had a lower diet quality and higher energy intake, BMI, and body fat. My Veggies is the only diet quality assessment tool for adults validated with objective measures, such as anthropometry. Given the resource-intensiveness of current methods, a simple tool such as My Veggies is long overdue. The next step is to develop a scoring system for its application and interpretation. This will allow the identification of those on low-quality diets, enabling targeted intervention in community and clinical settings.

Author contributions

The authors’ responsibilities were as follows – NLK, MST, MKS, KDR: designed the research; NLK: conducted the research; CMD, JS, XZ: analyzed data; NLK, MST, MKS, CMD, KDR: wrote the paper; NLK: had primary responsibility for the final content; and all authors: read and approved the final manuscript.

Data availability

The data described in the manuscript may be made available on request pending approval from all authors.

Declaration of Generative AI/ and AI-assisted technologies in the writing process

The authors declare that no generative AI or AI-assisted technologies were used in the writing of this manuscript.

Funding

Sources of funds used for this project included in-house USDA CRIS Projects 2032-51530-025-00D and 2032-51530-026-00D, as well as the National Institute of Food and Agriculture USDA award number 2015-68001-23280.

Conflict of interest

The authors report no conflicts of interest.

Appendix A. Supplementary data

The following is the Supplementary data to this article: