2.1. Study design and setting

The WRAPPED study was a natural experiment with a prospective matched controlled cluster design, with participants clustered within 36 study supermarkets to account for the store-based intervention. The flow diagram, Fig 1, illustrates the sampling frame for all outcome data points: store sales, customer purchasing, dietary data and household waste. The study took place between March 2018 and May 2022 and was registered with ClinicalTrials.gov (prospectively registered NCT03573973). It was approved by the University of Southampton, Faculty of Medicine Ethics Committee (ID 20986.A9) and conducted in accordance with the Declaration of Helsinki and Data Protection regulations. Modifications from the original protocol [21] were approved by both the University of Southampton, Faculty of Medicine Ethics Committee and the independent Study Steering Committee. The updated protocol is available on the project funder’s website [22]. This study is reported as per Consolidated Standards of Reporting Trials (CONSORT and CONSERVE) guidelines (S1 CONSORT Checklist and S1 CONSERVE Checklist).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. WRAPPED study flow chart and sample size.

https://doi.org/10.1371/journal.pmed.1004575.g001

The study setting was discount supermarket stores in England. The collaborating supermarket has over 1,000 stores nationwide and holds ~2% of the UK grocery market share [23].

This study sampled 36 stores, 18 intervention and 18 control stores [21]. Intervention stores were selected by the research team from the full list of stores due for refurbishment and structural changes to their in-store environment during the study period by the company. Retail store refurbishment occurs for all stores roughly every 5–7 years to keep stores’ decor fresh and equipment energy efficient. Intervention stores were selected by the research team from company planograms. Any store set for refurbishment which did not have the fruit and vegetable section in the first aisle previously were included. Control stores were matched to an intervention store on: i) sales profile, ii) customer profile, and iii) neighbourhood deprivation (Index of Multiple Deprivation (IMD)) [13,24]. Control stores were geographically distant from intervention stores to reduce contamination effects.

2.2. Intervention and control conditions

The intervention was implemented continuously for 6 months with two components executed simultaneously: i) more prominent positioning of fresh fruit and vegetables near store entrances, and ii) expanding the numbers of fresh fruit and vegetables available. The positioning or availability of frozen and canned fruit and vegetables were not altered.

The control condition was the previous layout of stores with a limited range of fresh fruit and vegetables, placed at the back of the store and other products in their usual locations.

2.3. Participant eligibility and recruitment

Women aged 18–60 years were the target population because improving the diets of women in this age group will improve their own health, and the short and long-term health of their children [25,26]. Women within this age group are also primarily responsible for domestic food-related tasks that influence their partners and families’ diets [27,28]. Eligible participants were women aged 18–60 years, holding a study supermarket loyalty card and had shopped in a study store 12 weeks before recruitment. Women under 18 or over 60 years of age, without a loyalty card or who shopped online only were not eligible.

Recruitment occurred in five waves between May 2018 and October 2021 with each pair of stores recruited over the same pre-implementation period. Eligible women, identified from loyalty card data, were sent a letter inviting them to participate. Participation involved completing four telephone interviews with a researcher at baseline and 1, 3 and 6 months after intervention implementation, and providing oral consent to be interviewed and written consent to share their loyalty card data. The letter did not contain details about the intervention and was sent by the supermarket to comply with data protection laws. Interested women contacted the research team directly via Freephone, text or email, were screened for eligibility and provided informed consent. To boost participant numbers, in-store recruitment was employed. The research team approached women customers while shopping in study stores and offered an information sheet. Interested women were subsequently phoned and consented. This method proved effective at enhancing sample diversity and was used for all 36 stores. Participants were offered up to 3× £10 Love2Shop or Amazon vouchers for their time. Each £10 voucher was sent to participants after completion of the baseline, 3- and 6-month questionnaires.

Due to COVID-19 restrictions, in-store recruitment was not possible and fewer participants responded to invitation letters than had done in the study’s pilot phase [21,24]. Consequently, the age criteria for participants were expanded from 18–45 years to 18–60 years and a sixth wave of recruitment was undertaken between September 2022 and February 2023 to boost purchasing data numbers (primary outcome) according to post-hoc sample size calculations (S1 Box). A letter was sent by the collaborating supermarket chain. Participants were offered a £15 Love2Shop or Amazon voucher for completing one telephone interview and consenting to share their loyalty card data.

All participants completed a demographic questionnaire about their age, ethnicity, marital status, highest educational qualification, employment status, number of children, weekly money spent on groceries and whether they shopped for most of their household’s food from the study supermarket chain. Participants provided home postcodes to identify neighbourhood deprivation status (IMD).

2.4. Outcome measures

Data collected included 9 months continuous store sale transaction data (secondary outcome) and participant loyalty card purchasing data (primary outcome) from the collaborating supermarket covering three time periods: i) baseline (3 months prior to intervention implementation); ii) short-term intervention effects (0–3 months post-intervention commencement) to represent habit formation [29]; and iii) longer-term intervention effects (3–6 months post-intervention commencement) to assess sustained changes. To understand intervention effects on household members’ diets (secondary outcomes), interview-administered telephone questionnaires obtained information about participants’ and their child’s diet aged 2–6 years (where applicable) at three time points: baseline, and 3 and 6 months following intervention implementation. If a participant had more than one child aged 2–6 years, she was asked to answer questions about her eldest child in that age group. Frequency of household fruit and vegetable waste (secondary outcome) were also collected at these time points.

Store sales of fresh fruit and vegetables were provided as numbers of items for each product sold in each study week. Participant purchasing data covering the same categories were provided as the number of items for each product purchased at each store visit during the study period. Purchasing data were aggregated to present data as items per household per week. A week of sales and purchasing data were removed from the analysis if the store was closed for the full or part of the week. The time period of store closure for structural changes to the instore environment varied across intervention stores according to the size of store and extent of in-store changes. Across all intervention stores, closures affected one to four weeks of sales and purchasing data; these data were removed from the analysis for both intervention and control groups. Store sales and individual purchasing datasets consisted of 11–14 weeks before the intervention and 24–28 weeks afterwards.

Measures of women’s and children’s dietary quality were assessed using published, validated tools [30,31]. Participants were asked how often in the previous month they (or their child) consumed each of the 20 Food Frequency Questionnaire (FFQ) foods. A dietary quality score for each woman or child was calculated by multiplying their standardised reported frequency of consumption of each FFQ food by corresponding weightings derived from published tools and summing the results. Dietary quality scores were standardised (mean = 0, SD = 1). Higher scores represent better dietary quality characterised by higher intakes of vegetables, fruit, water and wholegrain bread and lower intakes of white bread, processed meats, chips, crisps and sugar. Household waste data were collected by asking participants ‘over the past month how often has your household thrown away (as a result of food spoiling or serving/cooking too much) fruit’. The same question was asked separately for vegetables. Response categories were the same as those used for the FFQs.

2.5. Statistical analysis

Descriptive variables are given as percentage (frequency) for categorical variables and median (interquartile range) for non-normally distributed continuous variables. Differences between participants that shopped at intervention and control stores were tested using chi-squared tests for categorical variables and Mann-Whitney rank sum tests for non-normally distributed continuous variables.

Store sales data were analysed using an Interrupted Time Series [32] (full details S2 Box). Weekly sales per store were transformed to normal distributions using Fisher–Yates transformations [33] to protect commercially sensitive sales figures (standard deviations (SDs)). The time series models were fitted separately to account for store pairing followed by random effects meta-analysis [34,35] to synthesise differences between pairs of stores at the time of intervention, and 3 and 6 months post-intervention. This method enabled: i) retention of the store pairing design, ii) comparisons between pairs and iii) overall statements of study effect size and precision [36]. Results were interpreted on the original items sold per week scale by calculating the equivalent change on the original scale to the change from the median on the Fisher–Yates transformed scale. The collaborating supermarket chain sells only packaged fruit/vegetables (not singly), with each item averaging 5 portions (~400 g). This information informed conversions from items to portions. Full evaluation of the implementation of the intervention and differences between intervention and control stores in the positioning and availability of fruit and vegetables has been published elsewhere [37]. In brief, the intervention was implemented with close adherence to the study protocol but noted some variation in the distance the produce section moved forward and the overall increase in fruit and vegetable availability in intervention stores. Hence, process evaluation results identified the need for additional analyses including stratifying by: i) distance fresh fruit and vegetables moved forwards in the store (>14 m versus <14 m), ii) post-intervention position of fresh fruit and vegetables (first/last half of first aisle), and iii) post-intervention availability of fresh fruit and vegetables (≥73/<73 Stock Keeping Units (SKUs)) [37]. These analyses involved running random effects meta-analyses on the results of the Interrupted Time Series models, described above, but with intervention stores grouped by each position/availability stratification variable, and their paired control stores.

For household purchasing data time series analysis was not possible because data were right-hand skewed (i.e., 67% of women’s weekly purchases had no fresh fruit or vegetables). Three-level multilevel models were also tested but did not fit the data but returned a singular fit, likely due to smaller than anticipated participant numbers in each store. Singular fit arose as the store variance component was estimated as effectively zero, suggesting minimal clustering at the store level. Outcome data were dichotomised to indicate whether each week resulted in any study food purchases. Two-level multilevel models, with weekly purchasing nested within women and pooling data across stores, were implemented using a difference-in-difference approach [38] (full details S2 Box). Women’s data were analysed according to their recruitment store to conform with intention-to-treat analysis. Intervention effects according to woman’s educational level (up to 16 years of age or beyond 16 years of age) were analysed according to the study protocol (secondary outcome). Process evaluation results identified the need for the same additional analyses as the store sales data, plus study supermarket as main food shop [37]. These analyses involved running models in separate stratification groups defined by education/position/availability. To determine the consistency of the effect of the intervention on household purchasing with analyses accounting for clustering at store level, two-level multilevel models were fitted in 17 store pairs individually (the model would not run in all 18 pairs because of data sparsity in one store pair), followed by random effects meta-analysis [30,31] to synthesise differences between pairs of stores. This approach could not be used for stratified analyses because a singular fit was returned.

Intervention effects on changes in dietary quality from baseline to 3 and 6 months post-intervention were assessed using linear regression models. Diet was the outcome and intervention group and diet at baseline were predictors. A second set of regression models included confounding variables (age, money spent on groceries, number of children in the household and woman’s education) determined before analyses using directed acyclic graphs (DAG) (S1 and S2 Figs) [39]. Planned multilevel linear regression models (with women clustered within stores) returned a singular fit, likely for the same reason the household purchasing data did not fit. Regression models were fitted in each pair of stores were therefore fitted separately and random effects meta-analysis [34,35] used to synthesise results. Intervention effects on household fruit and vegetable waste per week was assessed using the same approach (S3 Fig).

Analyses were performed in Stata 14 [40], except the time series models which were fitted in R using packages nlme [41,42] and msm [43]. Consistent with current statistical thinking [44] our interpretations focus on effect sizes, and their precision, rather than emphasising statistical significance relative to any boundary.

3.1. Participant characteristics

A total of 865 women were assessed for eligibility, of which 580 were recruited; 300 were from control stores and 280 from intervention stores (Fig 1). Most women provided purchasing data (n = 475, 248 from control stores and 227 from intervention stores which provides 85% power at 5% significance level (2-sided) (S1 Box)) and 360 women provided information about their dietary and household fruit and vegetable waste patterns (190 from control stores and 170 from intervention stores). Of these 360 women, 250 reported living with children (aged <18 years) and 127 provided data about their child aged 2–6 years. Attrition rates for diet and waste data were 13% at 3-month and 19% at 6-month follow-ups. There were slight differences in participant characteristics at baseline between participants that shopped at intervention and control stores, with intervention women less likely to identify as being of white ethnicity and more likely to live in more deprived neighbourhoods (Table 1). More than half of the sample were aged 31–45 years, 72% identified as being white British, 40% had low educational attainment (no qualifications beyond age 16 years) and 63% were in paid employment. Almost a third reported that the study supermarket chain was where they purchased most of their groceries (31%).

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Baseline characteristics of all participants (n = 580).

https://doi.org/10.1371/journal.pmed.1004575.t001

3.2. Store sales (secondary outcome)

The synthesised results using meta-analysis are shown in Figs 2–4, illustrating results at baseline, and 3 and 6 months post-intervention implementation, respectively. Sales of fresh fruit and vegetables were 0.32 SDs (95% CI 0.11, 0.5; p = 0.002) greater in the intervention stores than would be predicted by the model counterfactuals at the time of intervention; the equivalent differences were 0.23 SDs (95% CI −0.05, 0.52; p = 0.10) at 3 months post-intervention and 0.18 SDs (95% CI −0.16, 0.52; p = 0.29) at 6 months post-intervention (secondary outcome). These changes are approximately equivalent to 2,525 (95% CI 775, 4,115), 1940 (95% CI 380, 3,950) and 1,450 (95% CI −945, 3,950) extra fruit and vegetable portions per store, per week at intervention, and 3 and 6 months post-intervention implementation, respectively.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Meta-analysis of increase in store sales of fresh fruit and vegetables (SDs) in intervention stores compared to that predicted by model counterfactuals at intervention implementation.

https://doi.org/10.1371/journal.pmed.1004575.g002

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Meta-analysis of increase in store sales of fresh fruit and vegetables (SDs) in intervention stores compared to that predicted by model counterfactuals at 3 months post-intervention implementation.

https://doi.org/10.1371/journal.pmed.1004575.g003

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Meta-analysis of increase in store sales of fresh fruit and vegetables (SDs) in intervention stores compared to that predicted by model counterfactuals at 6 months post-intervention implementation.

https://doi.org/10.1371/journal.pmed.1004575.g004

There were somewhat greater increases in sales of fresh fruit and vegetables in the intervention stores where the produce section moved forward more than 14 m (Table 2; difference = 0.48 SDs (95% CI 0.30, 0.67; p < 0.001) at intervention; difference = 0.40 SDs (95% CI 0.12, 0.68; p = 0.005) at 3 months post-intervention; difference = 0.30 SDs (95% CI −0.13, 0.74; p = 0.17) at 6 months post-intervention) and lower to no increase in stores where the fruit and vegetables moved fewer than 14 m further forward (difference = 0.15 SDs (95% CI −0.16, 0.45; p = 0.35) at intervention; difference = 0.01 SDs (95% CI −0.38, 0.40; p = 0.97) at 3 months post-intervention; difference = −0.01 SDs (95% CI −0.57, 0.56; p = 0.98) at 6 months post-intervention); however confidence intervals were wide and overlapped for stores where produce moved forward more than 14 m compared to less than 14 m. The changes in stores where fruit and vegetables moved forward more than 14 m are approximately equivalent to 3,645 (95% CI 2,350, 5,305), 3,115 (95% CI 960, 5,350) and 2,350 (95%CI 720, 6,045) greater fruit and vegetable portions per store, per week at intervention and 3 and 6 months, respectively. These findings were not replicated for intervention dose identified as being in the first, compared to the last half of the first aisle (Table 2; first half of aisle difference = 0.41 SDs (95% CI 0.25, 0.57; p < 0.001) at intervention; difference = 0.27 SDs (95% CI 0.02, 0.53; p = 0.04) at 3 months post-intervention; difference = 0.12 SDs (95% CI −0.27, 0.52; p = 0.53) at 6 months post-intervention compared to last half of first aisle difference = 0.09 SDs (95% CI −0.37, 0.56; p = 0.69) at intervention; difference = 0.15 SDs (95% CI −0.52, 0.81; p = 0.67) at 3 months post-intervention; difference = 0.61 SDs (95% CI −0.24, 1.47; p = 0.16) at 6 months post-intervention). Similarly, no notable differences in effect size were observed between stores that sold ≥73 fresh fruit and vegetables SKUs post-intervention or those which sold <73 SKUs (S1 Table; ≥ 73 SKU difference = 0.38 SDs (95% CI 0.19, 0.57; p < 0.001) at intervention; difference = 0.23 SDs (95% CI −0.01, 0.48; p = 0.06) at 3 months post-intervention; difference = 0.02 SDs (95% CI −0.39, 0.42; p = 0.94) at 6 months post-intervention compared to <73 SKU difference = 0.30 SDs (95% CI −0.01, 0.62; p = 0.06) at intervention; difference = 0.24 SDs (95% CI −0.20, 0.68; p = 0.28) at 3 months post-intervention; difference = 0.31 SDs (95% CI −0.24, 0.86; p = 0.27) at 6 months post-intervention).

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Increase in store sales of fresh fruit and vegetables (SDs) in intervention stores compared to that predicted by model counterfactuals at intervention, and 3- and 6-month follow-up post-intervention by dose (positioning).

https://doi.org/10.1371/journal.pmed.1004575.t002

3.3. Participant loyalty card purchasing data (primary outcome)

Of the 475 participants with purchasing data, there were 5,077 store visits according to their loyalty cards over the study period, 2,704 from control store participants and 2,373 from intervention store participants. Of the 5,077 visits, 1,724 (955 control women visits, 769 intervention women visits) were not at stores where women had been recruited and, while 72 of these visits were to another study store, only three visits were to stores in the opposite study arm.

Modelled proportions of women’s fruit and vegetable purchases show an overall decline in purchasing over time (Fig 5). The difference in the proportion of participants who purchased fruit and vegetables in intervention compared to control stores was very similar at baseline (−0.1% (95%CI −6.1%, 6.0%)) but was 0.6% (95%CI −5.6%, 6.7%) at 3 months and 3.3% (95%CI −2.5%, 9.2%) at 6 months after exposure to the intervention (Table 3) (primary outcome). In analyses accounting for clustering at store level in 17 store pairs (S4 Fig and S3 Table) consistent findings were observed to the two-level model findings in all 18 store pairs, increasing confidence in the main results.

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Effect of intervention on proportion of households purchasing fresh fruit and vegetables at baseline, 3- and 6-month follow-up post-intervention.

https://doi.org/10.1371/journal.pmed.1004575.t003

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. Modelled proportion of women purchasing food items in intervention and control stores.

https://doi.org/10.1371/journal.pmed.1004575.g005

Results for stratification by highest educational qualification, as a marker of socioeconomic status, showed a beneficial increase in the proportion purchasing fruit and vegetables amongst participants who shopped at intervention stores only after 6 months when analyses were restricted to those with fewer educational qualifications, although tests for interaction were not statistically significant by level of education (Table 3; educated up to age 16 years difference = −6.2% (95% CI −15.8%, 3.4%) at intervention; difference = −6.9% (95% CI −17.0%, 3.2%; p = 0.94) at 3 months post-intervention; difference = 1.5% (95% CI −8.2%, 11.3%; p = 0.09) at 6 months post-intervention compared to educated over 16 years of age difference = 3.9% (95% CI −3.9%, 11.7%) at intervention; difference = 4.8% (95% CI −3.1%, 12.7%; p = 0.79) at 3 months post-intervention; difference = 5.2% (95% CI −2.3%, 12.8%; p = 0.66) at 6 months post-intervention) (secondary outcome). Among participants whose main supermarket was the study store, the proportion purchasing fruit and vegetables was greater amongst those shopping at intervention stores than control stores only after 6 months of intervention exposure, although again tests for interaction did not show robust differences (Table 3; study store is main store interaction p = 0.49 and difference = −1.7% (95% CI −13.6%, 10.2%; p = 0.49) at 3 months; interaction p = 0.18 and difference = 4.0% (95% CI −7.4%, 15.4%; p = 0.07) at 6 months compared to study store is not main store difference = 0.9% (95% CI −5.9%, 7.6%; p = 0.82) at 3 months; difference = 2.1% (95% CI −4.5%, 8.7%; p = 0.91) at 6 months). When stratifying by intervention dose using distance produce section moved, the difference in proportion of fruit and vegetables purchasing between participants shopping at intervention verses control stores showed some increases from baseline to 3 (difference = 2.5% (95% CI −5.5%, 10.6%; p = 0.61) and 6 months (difference = 5.5% (95% CI −2.3%, 13.3%; p = 0.20) among those who shopped at stores where the produce section moved ≥14 m. The effect sizes were greater than those observed among intervention participants who shopped at intervention stores where the produce section moved <14 m (Table 3; difference = −2.1% (95% CI −11.4%, 7.1%; p = 0.80) at 3 months; difference = 0.7% (95% CI −8.1%, 9.5%; p = 0.69) at 6 months), although tests for interaction were not statistically significant. There were similarly no consistent differences in effect size according to stratification by number of SKUs (≥73 or <73) in intervention stores or for positioning of produce section in the first half or last half of the aisle in intervention stores (S2 Table).

3.4. Dietary variables (secondary outcomes)

Women’s diet findings (Table 4) indicate changes in dietary quality of 0.07 SDs (95% CI −0.09, 0.23) at 1 month post-intervention, −0.15 SDs (95% CI −0.33, 0.03) at 3 months post-intervention and 0.25 SDs (95% CI 0.10, 0.40) at 6 months post-intervention (secondary outcome). Equivalent changes for children’s dietary quality were 0.34 SDs (95% CI −0.30, 0.98), 0.43 SDs (95% CI −0.39, 1.25) and 0.06 SDs (95% CI −0.13, 0.24) at 1-, 3- and 6-month follow-up, respectively, with uncertainty indicated by wide confidence intervals (secondary outcome).

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Effect of store intervention on women’s and children’s dietary quality from baseline to 1-, 3- and 6-month follow-up post-intervention.

https://doi.org/10.1371/journal.pmed.1004575.t004

Download:

• PNG
  larger image
• TIFF
  original image

Table 5. Effect of store intervention on change in frequency of waste from baseline to 1-, 3- and 6-month follow-up post-intervention.

https://doi.org/10.1371/journal.pmed.1004575.t005

3.5. Household waste (secondary outcome)

Results for reported household waste (Table 5) revealed little change in frequency of fresh fruit being thrown away at 1- (−0.01 times per week (95% CI −0.24, 0.22)) and 3-month follow-up (0.07 times per week (95% CI −0.18, 0.32)) (secondary outcome). Similarly, there was little effect on change in frequency of fresh vegetables being thrown away at 1- (−0.02 times per week (95% CI −0.20, 0.17) and 3-month follow-up ((−0.02 times per week (95% CI −0.25, 0.20)) (secondary outcome). At 6 months post-intervention, differences in reported frequency of waste in intervention compared to control stores were 0.20 times per week (95% CI 0.07, 0.32) for vegetables and 0.12 times per week (95% CI −0.12, 0.35) for fruit.