Veggie van program

The VV was a mobile produce market, run by the non-profit organization Community Nutrition Partnership, that offered high-quality produce aggregated from multiple local farms to customers at a reduced price [15]. The socioecological model acts as a guiding framework for addressing the complexities associated with dietary intake in high-need populations. The socioecological model posits that multiple levels of influences (e.g., individual, interpersonal, and community-level factors) intersect to yield outcomes [17]. Social Cognitive Theory provides constructs that support the goal of targeting behavior change on at the individual and environmental levels. Specifically, VV sought to change the food environment (and people’s perceptions of it) while simultaneously improving self-efficacy to purchase, prepare and eat fresh F&V.

Aspects of the food environment addressed by VV included availability, accessibility, acceptability, affordability, and accommodation [18]. VV increased the number of food outlets within target communities by partnering with local community organizations that were already frequented by or in locations near the target population (accessibility) and selling produce at those locations (availability). Locations included health clinics, recreation centers, libraries, housing communities and community centers [15, 19]. The VV mobile market was held weekly during the 6-month intervention period, unless staffing or weather issues prevented the market from operating. Customers could pay week-to-week for shares of produce (i.e., similar to a CSA they would receive a set amount of seasonally available produce items offered at the same price each week) or buy individual produce items at the market. In order to ensure VV offered high quality produce, all the F&V were fresh, locally grown, and often organic (acceptability). Produce was offered at about half the cost of traditional CSA programs (affordability). Unlike CSA programs, no upfront payments or commitments were required though pre-ordering was encouraged and incentivized with monetary discounts. To maximize accommodation, VV accepted multiple forms of payment (cash, credit/debit, check and the Supplemental Nutrition Assistance Program’s (SNAP) electronic benefit transfer (EBT) cards) and visited sites at times which were assessed to be the most convenient for potential customers.

The educational intervention addressed individual skills and behaviors related to F&V consumption with the goal of improving self-efficacy through increased outcome expectations/expectancies, observational learning and behavioral capability. Newsletters and nutrition demonstrations addressed the benefits of healthy eating (expectations). Social marketing campaigns at VV sites encouraged people to come to markets to benefit from VV (expectancies). VV also provided cooking demonstrations, tips for cooking seasonal produce and recipes for items sold at VV (behavioral capability and observational learning). While outside the context of research, customers only received weekly newsletters when they visited the VV, for the purposes of this evaluation, research participants at intervention sites received newsletters by mail or e-mail even if they did not visit VV. Since we could not require study participants to shop at VV, we sent them the newsletters as a way of ensuring that they were aware of the program and at least received some dose of the behavioral intervention.

Recruitment

The Green Cart study and VV teams partnered with organizations that were serving the priority population (lower-income and/or limited access to fresh produce) to facilitate the research. Site and participant recruitment are described in detail elsewhere [19, 20]. We asked partner organizations to collect at least 30 interest forms from community members who were potentially interested in purchasing VV produce and who were willing to participate in a research study. A member of the research team contacted those who completed the forms and asked them to participate in a study. Eligible individuals (Age 18 or older, English speaking, and primary food shopper for their household) were invited to complete a telephone-administered baseline survey and were enrolled in the study. After participant data collection was completed, we randomized sites in pairs to either the intervention or the delayed intervention control group. Institutional Review Board at the university approved all procedures.

Survey and measures

We collected all data over the phone via interviewer-administered surveys at baseline and 6-months. Additional details on data collection and measures can be found elsewhere [20].

Dietary outcomes

The primary outcome, F&V intake (cups/day), was assessed using the 10-item National Cancer Institute F&V screener and calculated according to the screener instructions [21]. A validation study found estimated correlations between the screener and F&V intake from 4 non-consecutive 24-h recalls were 0.67 for men and 0.53 for women [22]. Test-retest reliability for similar food frequency questionnaires (FFQs) is generally good (ICC = 0.65) [23]. Added sugar (servings/day) was calculated from the 7-items selected from the National Health and Nutrition Examination Survey Dietary Screener to capture consumption in the past month including sugar-sweetened beverages, chocolate or candy, pastries, desserts or ice cream [24].

Psychosocial measures

In order to measure the impact of the environmental-level intervention components, we assessed participant’s perceived access to fresh F&V assessed using a 3-item scale [25, 26] adapted to examine access using three different definitions: one assessing perceived neighborhood access, one assessing perceived access near the VV location in their community, and one assessing general perceived access. Possible perceived access scores range from 3 (strongly disagree to all items) to 15 (strongly agree to all items), with a midpoint of 8 indicating a neutral response. The impact of the nutrition education intervention component was assessed by looking at self-efficacy to purchase, prepare and eat F&V. Self-efficacy was assessed using nine questions (shown in Table 4) with response options ranging from 1 to 10 (1 = “very easy” to 10 = “very hard”) [27]. Items were summed to create a total self-efficacy score (range 9 to 90).

VV usage and implementation measures

On the survey, participants were asked to report if they had ever used VV; we also reviewed sales data to see if participants made any purchases at VV during the intervention period. If a participant purchased at least one share of produce, they were recorded as a customer. Participants who only purchased individual produce items were not identified by name and not included in the sales data. In order to understand VV program implementation and fidelity, coordinators were asked to fill out a process measures form after each market. Process data will be reported separately.

Power analysis

Our original power analysis was based on F&V servings/day as calculated by the National Cancer Institute F&V screener. In the sample size estimates, we considered correlated change in F&V intake among participants within a community site (ICC), number of participants within each site and number of sites (clusters) [28, 29]. The VV program was expected to increase the F&V consumption by least 1.25 servings per day or approximately 0.75 cups/day (effect size of approximately 0.35) based on the VV pilot [15]. A sample size of 6 communities per group with 20 participants in each community yielded 0.80 power to detect 0.75 cup difference in mean changes between two groups using two-sided tests of significance at p = 0.05, assuming an ICC of 0.001 and standard deviation of 3.6 based on other cluster randomized trials [30]. We assumed attrition to be no more than 20% based on the pilot study [15]. Thus, a final goal was to recruit at least 25 participants in each community, for a total of 300 participants.

Analysis

A generalized linear mixed model (GLMM) with a random intercept to control for clustering within community sites was used to test the effect of the VV intervention on dietary intake (F&V and added sugar) at 6-months. Additional GLMM variables included: 1) baseline diet as fixed covariate, and 2) baseline diet and education. Including the baseline values as a covariate, in an analysis of covariance (ANCOVA) is known to be a more powerful test than a group comparison of baseline to post-intervention change (26). ANCOVA is not distorted by regression towards the mean bias, whereas a change analysis is subject to that bias [31, 32]. We controlled for education because both income and education were significantly higher among control participants at baseline. These items were highly correlated and education was more consistently reported than income. Additionally, we completed a sensitivity analysis on the final model excluding extreme F&V reporters. Extreme F&V reporters were defined as participants who had a change greater than 9-cups of F&V per day. All secondary outcome analyses were conducted using GLMMs (e.g., usage, perceived access, self-efficacy), removed extreme reporters (when F&V intake was the outcome) and controlled for baseline values, education and clustering within sites.

Demographics

Dietary intake

Perceived access

Self-efficacy

Limitations

In order to maintain partner engagement, we ended recruitment at most sites before we reached our study goals [19]. Consequently our baseline numbers (N = 201) were lower than the anticipated 300. Extended timelines may also have affected drop-out, which was higher than our pilot work (71% actual retention rate vs. 80% anticipated retention rate) [19]. While the research team and the VV staff worked closely together on major decisions that would affect the research, VV staff was ultimately responsible for implementation. Therefore, VV prices, nutrition education programming, and intervention duration were not always consistent across sites. Recruitment and implementation challenges are common in community-based research, especially with minority populations, [33] and the randomized controlled design provides adequate experimental control despite these threats to internal validity.

Another limitation is that this study relied on a food frequency questionnaire (FFQ) to collect diet information. While FFQs are acceptable and have been used in other mobile market research [13, 34], they only reflect limited number of vegetable categories and may not have adequately captured some of the local produce that was commonly available in North Carolina (leafy greens, root vegetables, etc.). As with all self-report dietary assessment measures, results are susceptible to bias. To help improve the accuracy of FFQs, we provided portion size sheets to help participants with their answers, but only about a fifth of participants reported using them when completing the surveys.

As noted above, much of the difference in F&V consumption we saw between the intervention and control groups was due to decreases in the control group. While we don’t know the cause of these decreases, we saw a similar trend in our pilot work which could represent a regression to the mean. We also reviewed the data for a possible effect of seasonality on F&V intake across months and seasons at each time point, but did not find any patterns that suggested a seasonality bias. Future work should consider using 24-h recalls or adapting standard FFQs to account for local or regional produce.

Despite its limitations, these methods represent an improvement to previous mobile market research. Although the study did not reach goal participation numbers, these projections were based on studies that were similar, but that ultimately had different ICCs and standard deviations than the present study. Our estimates also anticipated a smaller (more conservative) difference in the change in F&V intake between intervention and control participants than we saw with our main adjusted analysis so we could more easily detect a significant difference. When extremes removed, the difference was smaller than the anticipated change (0.5/cup vs. a projected 0.75 cups/day difference) which could explain why that comparison did not reach statistical significance.