In rural and underserved communities, small food retailers (e.g. corner and convenience stores) may be one of the only food retailers available(Reference Liese, Weis and Pluto1,Reference McGuirt, Jilcott Pitts and Ammerman2) . Small food retailers typically sell foods and beverages that are less healthy when compared to large food retailers, which perpetuates the cycle of low access and poor dietary quality among rural and disadvantaged communities(Reference Laska, Borradaile and Tester3,Reference Lucan, Karpyn and Sherman4) . To improve access to healthy foods, public health interventions have been implemented to stock and promote healthier foods in small food retailers(5–Reference Langellier, Garza and Prelip8). More research is needed to better understand how food and beverage purchases at these small food retailers are associated with dietary intake(5,Reference Jilcott Pitts, Wu and Truesdale9) .
One key component of a healthy diet is fruit and vegetable (FV) intake. Traditional methods to assess FV intake include self-reported measures. Unfortunately, these measures are often fraught with social desirability bias(Reference Kirkpatrick, Collins and Keogh10–Reference Archer and Blair12). Carotenoids are the colourful phytonutrients found in vegetables and fruits, while plasma carotenoids are correlated with FV intake and serve to approximate overall FV intake(Reference Campbell, Gross and Martini13,Reference Lampe, Huang and Neuhouser14) . Measurement of skin carotenoid score (SCS) is emerging as a promising method to assess FV intake in field-based nutrition studies, as it is relatively quick and portable(Reference Ng, Delgado and Borlaug15,Reference Bakırcı-Taylor, Reed and McCool16) . To date, no studies examine healthfulness of purchases at small food retailers and associations with skin carotenoids. Therefore, the purpose of this study was to determine if customer purchases at small food stores are associated with the overall healthfulness of dietary intake as approximated by SCS.
Methods
Research design and setting
This cross-sectional study is a secondary analysis of data from a larger evaluation study of the North Carolina Healthy Food Small Retailer Program (NC HFSRP), which is detailed in other manuscripts(5,Reference Jilcott Pitts, Wu and Truesdale9,Reference Jilcott Pitts, Wu and Truesdale17–Reference Haynes-Maslow, Jilcott Pitts and Boys20) . A total of twenty-two small food retailers were evaluated, including customer intercept surveys and bag checks among 1261 customers between 2017 and 2020. While the NC HFSRP was first designed to promote healthy food access in rural areas, it was eventually expanded so that seven of the stores were in rural areas and fifteen were in more urban areas. To classify stores as rural or urban, the stores were spatially joined to census tract-level data from the USDA Food Access Research Atlas dataset which defines rural and urban based on the geographic centroid of the census tract, with census tracts containing more than 2500 people being urban, and all others being rural(21).
Study population and data collection
Participants in this study were a convenience sample of adults (18 years or older) who purchased food or beverage items at small food retailers participating in the larger evaluation study(18). Verbal consent was given after research staff provided individuals with information about the study. If interested and eligible, participants completed a survey with demographic information and researchers conducted a bag check by recording detailed information on each item purchased, consistent with procedures in prior studies(Reference Caspi, Pelletier and Harnack22,Reference Kiszko, Cantor and Abrams23) .
Skin carotenoid scores
SCS were measured using the Veggie Meter®, a non-invasive, portable device that uses pressure-mediated reflection spectroscopy to quantify skin carotenoids(Reference Ermakov and Gellermann24). The Veggie Meter® has been validated against plasma carotenoids and self-reported FV intake and has been successfully used in community settings among racially and ethnically diverse participants(18,Reference May, Jilcott Pitts and Stage25–Reference Bayles, Peterson and Jilcott Pitts28) . SCS are associated with plasma carotenoids and FV intake(Reference Jilcott Pitts, Moran and Wu29,Reference Mayne, Cartmel and Scarmo30) , key components of the federal Dietary Guidelines for Americans(31–Reference Chiavaroli, Viguiliouk and Nishi34).
Each participant’s right index finger was inserted into the instrument’s finger cradle to bring the fingertip in contact with the collecting contact lens, while simultaneously applying light pressure so that blood was temporarily squeezed out of the measured skin region(Reference Ermakov, Ermakova and Sharifzadeh27). Participants’ fingers were scanned three times in the single-scan mode to obtain a mean SCS. SCS range from 0 to 800, with a higher score indicating more skin carotenoids present. The device was calibrated with a light and dark reference before each measurement session.
Bag check purchase data
Bag check data were analysed to generate a total of four scores detailing individuals’ purchases: (1) Healthy Purchases; (2) Healthy Purchases-Proportion; (3) Carotenoid-Containing Purchases; and (4) Carotenoid Containing Purchases-Proportion. To classify foods, one Registered Dietitian/Nutritionist (RDN (PhD) and one Nutritional Epidemiologist (PhD) were given a list of all recorded food and beverage items and independently classified items as healthy or unhealthy and carotenoid-containing or non-carotenoid-containing. Coding results were compared, and a second RDN, PhD, served as the tiebreaker on items when consensus was not reached. The criteria used to classify healthy items followed the 2020 Dietary Guidelines for Americans and included fruits, vegetables, eggs, beans, bread, low-fat dairy products, lean protein and 100 % juice(31). Common non-healthy items included processed meats, sweetened drinks, salty and sweet snacks, and ready-to-eat meals(Reference Mayne, Cartmel and Scarmo30,Reference Chiavaroli, Viguiliouk and Nishi34) . The criteria used to classify carotenoid-containing were that items must have had ≥ 0·1 mg per serving of either α-carotene, β-carotene, β-cryptoxanthin, lycopene, lutein and zeaxanthin (the most common forms of carotenoids in the human diet)(Reference Eggersdorfer and Wyss35). Examples of carotenoid-containing foods and beverages included orange juice, tangerine fruit cups, avocadoes, string beans, tomatoes, green peppers, yams, spinach, eggs, sweet potato, salad and collards. Further details on food and beverage classifications can be found in Table 1.
Table 1 Decision criteria for coding bag check food and beverage items as healthy or unhealthy, and example food and beverage items*
* All items not to be consumed by an adult (e.g. infant formula) or non-edible (e.g. gasoline and laundry detergent) were not coded or included in the denominator of proportion scores.
To calculate the Healthy Purchases score, participants received a score of 1 if any healthy items were purchased and a score of 0 if no healthy items were purchased. To calculate the Healthy Purchases-Proportion score, all items in a purchase were scored as healthy (1) or not healthy (0) and summed, and the number of healthy purchases was divided by the total number of items purchased (healthy and unhealthy). Multi-component items (e.g. BLT sandwich) were coded as their individual components (e.g. bacon, lettuce, tomato and bread).
To calculate the Carotenoid-Containing Purchases score, participants received a score of 1 if any carotenoid-containing items were purchased and a score of 0 if no carotenoid-containing items were purchased. To calculate the Carotenoid-Containing Purchases-Proportion score, all items in a purchase were scored as a carotenoid-containing food (1) or non-carotenoid-containing foods (0), and carotenoid-containing foods were summed and then divided by the total number of items purchased (carotenoid and non-carotenoid containing foods). Multi-component items were coded as their individual components, similar to coding for the healthy proportion purchases.
Analysis
Data analyses were conducted using SAS version 9.4 (SAS Institutes). Univariate statistics, including means and frequencies, were used to describe demographic variables. Chi-square tests and two-sample t tests were used to compare the demographics between those who purchased healthy/carotenoid items and those who did not. To examine associations between purchase of healthy items and carotenoid-containing items and VM®-assessed skin carotenoids, t tests and multiple linear regression models were computed, including covariates of sex, age, race and smoking status. These covariates were included because they could all be associated with FV intake and with SCS(Reference Madore, Hwang and Park36–Reference Parks, Blaser and Smith39). The four main independent variables considered were Healthy/Carotenoid-Containing Purchasing scores and Healthy/Carotenoid-Containing Purchases-Proportion scores. Data were collected for 1086 participants; however, due to missing data, sample sizes ranged from 966 to 967 in linear regression models. In an analysis of missing data, there were no significant differences between those missing and not missing data on the variables of interest (healthy/carotenoid-containing purchases and skin carotenoids). For bivariate analyses between proportion of healthy items and proportion of carotenoid-containing items, Pearson’s correlation coefficients were used, with Cohen’s criteria for the strength of association (0 < 0·3 = small; 0·3–0·5 = moderate; > 0·5 = large)(Reference Cohen40).
Results
Healthy and carotenoid-containing foods and beverages purchased
Table 1 shows examples of healthy foods purchased, including any produce purchased, lean meats, 100 % juices, plain popcorn and plain nuts, milk, and yogurt. Unhealthy items (the majority of items purchased) included fatty meats, non-100 % juices, crackers, chips, candy, cakes, and flavoured popcorn and peanuts with excessive added salt, sugar or fat.
The carotenoid-containing foods purchased included orange juice, cranberry apple juice, cranberry apple raspberry juice, pineapple juice, tangerine fruit cup, avocadoes, string beans, tomatoes, green peppers, yams, spinach, tomatoes and sweet potatoes.
Participant characteristics
The characteristics of study participants are found in Table 2, for the overall sample, and by those purchasing healthy and carotenoid-containing food and beverage items. Most participants were male (55·1 %) and either Black/African American (60·0 %) or White (27·6 %). The mean participants’ age was 43·4 years (sd = 15·3). The mean VM®-assessed skin carotenoids was 236·4 (sd = 81·9). (Data not shown.)
Table 2 Characteristics of 1086* small food retailer store participants, overall and by number and percentage with healthy and carotenoid-containing food and beverage purchases
* Individuals did not report demographics for the following characteristics: age (n 15), sex (n 3), race/ethnicity (n 14) and smoking (n 89).
** Other category was excluded.
Associations between healthy items purchased, proportion of healthy items purchased and skin carotenoid scores
Table 3 shows that among the 395 participants who purchased a healthy item, mean VM®-assessed SCS was 246·6 (sd = 84·3), and among the 691 participants who did not purchase a healthy item, mean VM®-assessed SCS was 230·6 (sd = 80·0). This difference was statistically significant, P = 0·002 (Cohen’s D = 0·20). There was also a small, positive statistically significant correlation between proportion of healthy items purchased and VM®-assessed SCS (r = 0·077, P = 0·011). (Data not shown.)
Table 3 Mean Veggie Meter®-assessed skin carotenoid scores for those with and without healthy and carotenoid-containing purchases
Table 4 shows that, when controlling for sex, age, race and smoking status, purchase of a healthy item was positively and statistically significantly associated with VM®-assessed SCS, such that individuals who purchased a healthy item had significantly higher VM®-assessed SCS compared to those who did not purchase a healthy item (estimate = 12·96; se = 5·31; P = 0·015).
Table 4 Adjusted linear regression analyses* for the association between healthy purchases, carotenoid-containing purchases and Veggie Meter®-assessed skin carotenoid scores among 1086 study participants. The outcome SCS had a mean 236 (82)**
SCS, skin carotenoid status.
* Data were collected for 1086 participants; however, due to missing data, sample sizes ranged from 966 to 967 in linear regression models.
** The models are adjusted for sex, age, race and smoking.
In adjusted linear regression analyses controlling for age, sex, race and smoking, % healthy items purchased was positively associated with VM®-assessed SCS, such that individuals who purchased 1 % more of healthy items had a higher VM®-assessed SCS (estimate = 0·13; se = 0·07; P = 0·057).
In the four models in Table 4, age was associated with SCS such that older individuals had lower VM®-assessed SCS. Sex was associated such that males had higher VM®-assessed SCS than females. Race was associated such that Black and other races had higher VM®-assessed SCS than White individuals. Smoking was associated with VM®-assessed SCS such that those who did not smoke had higher VM®-assessed SCS compared to those who did.
Table 3 also shows that purchase of healthy non-carotenoid containing items was also associated with higher VM®-assessed SCS such that those who purchased any healthy non-carotenoid containing items had a significantly (P = 0·020) higher SCS (243·2 (81·7) compared to those who did not (230·5 (80·0)).
Associations between carotenoid-containing items purchased, proportion of carotenoid-containing items purchased and skin carotenoid scores
Table 3 shows that among the eighty-three participants who purchased a carotenoid-containing item, mean VM®-assessed SCS was 260·2 (sd = 92·2). Among the 1003 participants who did not purchase a carotenoid-containing item, mean VM® assessed SCS was 234·5 (sd = 80·8). The difference was statistically significant, P = 0·006 (Cohen’s D = 0·32). There was also a small, positive statistically significant correlation between proportion of carotenoid-containing items and VM®-assessed SCS (r = 0·096, P = 0·002). (Data not shown.)
Table 4 shows that, when controlling for sex, age, race and smoking status, purchase of a carotenoid-containing item was positively and statistically significantly associated with VM®-assessed SCS, such that individuals who purchased any carotenoid-containing items had significantly higher VM®-assessed SCS compared to those who did not purchase any carotenoid-containing items (estimate = 23·77; se = 9·58; P = 0·013).
In the adjusted linear regression model, % carotenoid-containing items purchased was positively and statistically significantly associated with VM®-assessed SCS, such that individuals who purchased 1 % more of carotenoid-containing items had a higher VM®-assessed SCS (estimate = 0·41, se = 0·13, P = 0·002).
Discussion
In the current study, participants who purchased healthy and/or carotenoid-containing foods had significantly higher skin carotenoids than those who did not purchase any healthy or carotenoid-containing foods. A previous study suggests that FV purchases have the highest concordance between purchases and consumption(Reference Appelhans, French and Tangney41). Our findings suggest that healthy and carotenoid-containing purchases at small food stores are possibly indicative of carotenoid-rich FV intake due to the association with SCS.
The strengths of this study include large sample size and the use of an objective dietary measurement. Additionally, the study was conducted in an underserved area, where communities could benefit greatly from this research. Limitations include the use of a convenience sample from one state, which reduces generalisability. Participants were also not asked whether they were purchasing items for themselves or for others. For this reason, we do not know if and what proportion of food/beverages purchased were consumed by the participant. However, many items purchased from small stores are single-serve snacks or ready-to-eat foods(Reference Pinard, Shanks and Harden42), increasing the likelihood that the participant was also the consumer of the food/beverage. A further limitation is that the determination of what constituted an ‘item’ was challenging given the varied range of foods purchased. Yet a more precise measure based on weight was not feasible in this study.
Finally, there are many ways to define healthy and unhealthy foods/beverages and the definitions chosen may be debated. Nevertheless, the use of three health and nutrition professionals to code healthy and unhealthy foods based on established nutrition-related guidelines adds increased credibility to the results.
Conclusions
Customers who purchased healthier food and beverages from small food retailers had higher skin carotenoids. This suggests that customer purchases at small stores are indicative of dietary healthfulness as approximated using skin carotenoids. Further, these customers who purchased foods higher in carotenoids, and had higher skin carotenoid levels, did so without a specific intervention aimed at increasing skin carotenoid scores. More research is needed to understand additional factors that could improve the number of carotenoid-rich foods offered and the percentage of individuals who purchase carotenoid-rich foods at small food retailers. Lastly, the current study indicates that measuring both purchases of healthy and carotenoid-containing foods, as well as skin carotenoids, are both potentially feasible and effective tools in evaluations of similar public health nutrition interventions.
Acknowledgements
The authors are grateful to the store owners and customers who volunteered to participate in this study.
Financial support
Support for this article was provided by Policies for Action, a national programme of the Robert Wood Johnson Foundation, grant number: 76101. The views expressed here do not necessarily reflect the views of the Foundation.
Conflicts of interest
The authors have no conflicts of interest to report.
Authorship
S.J.P, V.C.S. and J.D. conceptualised the paper. J.D. wrote initial drafts of the paper and conducted preliminary analyses. S.J.P. edited the final version of the paper. V.C.S., K.P.T. and K.K. categorised food items into appropriate categories. J.D. and S.BJ.P. conducted data analyses, and K.P.T. and Q.W. contributed to data analyses. J.T.Mc. and L.H.M. assisted in conceptualising the study and interpreting results. All authors reviewed initial drafts and provided substantive comments on final drafts. All authors reviewed and approved the version of the paper as submitted.
Ethics of human subject participation
This study was conducted according to the guidelines laid down in the Declaration of Helsinki, and the East Carolina University Institutional Review Board reviewed and approved the study (UMCRB 16-002420). Verbal informed consent was obtained from all participants. Verbal consent was witnessed and signified by completion of the bag check and questionnaire.
Open access
