Design & setting

This pre-registered (AsPredicted ref: 56358,, three-arm, randomised controlled trial was conducted in an experimental online canteen developed using REDCap (Research Electronic Data Capture), a web application for data collection [7]. The website was designed to simulate an online pre-ordering system for a real-world workplace canteen. An online canteen pre-ordering system is a website which displays the canteen’s menu and allows employees to place their lunch order in the morning for collection later that day. Participants were able to hypothetically order their lunch from 6 menus containing a selection of main hot meals (n = 3), jacket potatoes (n = 10), soup & sandwiches (n = 15), sweet snacks (n = 18), savoury snacks (n = 20), and non-alcoholic drinks (n = 18) based on the menus of a real-world workplace canteen with whom we partnered (see Additional file 1, Appendix D for full menus). In the real-world canteen, main hot meal options (n = 3) change on a daily basis. Participants were randomly assigned to view and choose from the main hot meals for 1 of 5 different days to reflect this. The CONSORT checklist is available in Additional files 2, 3, and 4.


In February 2021, participants were recruited through Prolific Academic, an online participant sourcing platform [8]. To be eligible for the study, participants had to be ≥ 18 years, a UK resident, speak English fluently, and be in full or part-time employment. Those following restricted diets, e.g., vegetarian or dairy-free, were ineligible, as this would affect the acceptability of swaps offered. Prolific Academic pre-screened participants on these criteria and sent invitations to eligible panel members. Potential participants could also access the study via a link published on the Prolific Academic dashboards of all eligible panel members. Participants could follow this link where they were able to confirm their eligibility, read the information sheet, and provide consent.

Randomisation & blinding

Simple randomisation (1:1:1) was performed using Predictiv [9]. Participants were randomised to both a trial arm (1 of 3) and a menu (1 of 5), meaning that participants were evenly allocated to 1 of 15 groups. To do this, the platform allocated eligible participants a random integer between 1 and 15 representing the 15 conditions. To ensure balance, the algorithm ranked the conditions (1–15) based on the number of participants previously allocated to each and allocated the next participant to one of the 7 least used conditions. While investigators were not blinded to condition, they were not able to manipulate any study parameters following the initial study set up, as all study procedures were automated.

Online ordering task

Following randomisation, participants were directed to REDCap where they were asked to indicate their current subjective feeling of hunger. Participants responded to the question “how hungry do you feel” using a slider scale anchored with the extremes of “not at all” (0) on the far left, and “extremely” (100) on the far right [10]. Participants were then asked to imagine they worked for a company that had a pre-ordering website for their canteen and to order their lunch for the day using the website. They were asked to make choices that were in keeping with what they would typically have for lunch during their working day. Lower-energy swaps were automatically offered for originally selected menu items, if a suitable alternative was available. Participants placed one order only and did not pay for this order.

Swaps offered

Swaps offered were pre-determined by the research team using the criteria outlined in this section. The criteria for main hot meals differed to the ones in all other menus. Regardless of the menu, to qualify as a swap, the alternative had to contain at least 50 kcal less than the originally selected item, because a minimum of 50 kcal reduction per-person per-day has been identified as being clinically relevant [11]. Swaps offered were lower-energy items and were almost always from the same menu as the initial selection to ensure that the swap offered was as similar as possible to the initially selected item. Only one swap was offered for each item on a given menu, except for main hot meals where two swaps were offered. The criteria for each individual menu can be found in Sect. 6.2 of Additional file 1. Choices of swaps for main hot meals followed an algorithmic process as per Fig. 1. Main hot dishes served in the real-world canteen are dissimilar to each other to provide variety, therefore offering a swap from within the same menu results in offering a dissimilar meal. Two swaps were offered for each main dish to maximise the potential acceptance of swaps offered. In most cases, the first swap offered was the main with the lowest energy content and the second swap offered was either a jacket potato or an item from the soup and sandwich menu similar to the originally selected item. Similarity was based on the main protein source in the dish. Where possible, the main protein source was matched. For example, a chicken sandwich was offered as a swap for a chicken curry.

Fig. 1
figure 1

Flow diagram outlining the criteria for swaps offered for main dishes

A registered dietitian analysed the energy content of menu items using recipes provided by our partner canteen’s catering company and the nutritional information published on the supplier’s website. A lower-energy swap was available for 71% of options (see Table 1). Swaps were offered immediately after selection. Participants could decline or accept the swap for all menus except the main hot menu by clicking either “No, I will stick with my choice” or “Yes, I would like to swap”. For the main hot menu, participants could decline or accept the swaps offered by clicking: “No, I will stick with my choice”, “Yes, I would like to swap to [name of swap 1]” orYes, I would like to swap to [name of swap 2]”.

Table 1 Menu categories with examples of swaps offered, % swaps available, means (range) of energy content


Participants were randomly allocated to one of the following groups:

  1. 1.

    Control: No swaps offered.

  2. 2.

    Swaps: Swaps offered were accompanied by the message: “How about a swap?”

  3. 3.

    Swaps+PACE: Swaps offered were accompanied by the message: “How about a swap? Save [x] calories = [y] min walk”.

The energy content in kcal was published beside each option on all menus, meaning that all participants, including the control group, could be aware of the initial calories of all items and had the option to calculate the energy difference between menu items. The price was also presented beside each option on all menus, in all conditions. When swaps were offered, the energy content and the price of the swap item was presented. Figure 2 shows how energy content and price information for menu items and swaps offered was displayed in each condition. Prices were based on the 2020 price list provided to us by a real canteen based in the UK.

Fig. 2
figure 2

Intervention example: (A) Main hot meal menu day 1; (B) swaps condition; (C) swaps + PACE condition

Post-trial survey

After placing their lunch order, participants completed a brief exit survey to explore the acceptability of the intervention and record participant information (sex, age, ethnicity, and education along with height and weight for the calculation of body mass index (BMI). The Scottish Physical Activity Screening Questionnaire (Scot-PASQ) is a validated scale and was used to assess whether participants were meeting physical activity guidelines [12]. A shortened 3-item version (Cronbach’s alpha = 0.81) of the Dietary Intent Scale [13, 14], was used to measure dietary restraint. Acceptability was assessed by asking participants how acceptable they felt it would be for their employer to (a) implement a pre-ordering system for their workplace canteen and (b) offer them swaps for their food choices (only those in the intervention groups). Response options were on a scale from 1 (completely unacceptable) to 5 (completely acceptable). Upon completion, participants were debriefed and reimbursed with £0.50. The protocol (Additional file 1) was implemented with minor changes outlined in Sect. 2.12 Statistical Analysis below.

The primary outcome

The primary outcome was the total energy (kcal) of items ordered by each of the three groups, controlling for the energy content of the first item ordered.

Secondary outcomes

  1. (a)

    Secondary outcome (a) was the number of swaps accepted (restricted to the groups being offered swaps), controlling for the number of swaps offered.

  2. (b)

    Secondary outcome (b) was the proportion of participants ordering a lunch that meets the Public Health England (PHE, 2018) guideline of containing 600 kcal or less in each of the three groups.

  3. (c)

    Secondary outcome (c) was Likert scale ratings of intervention acceptability by intervention groups

Exploratory outcomes

  1. (a)

    Exploratory outcome (a) was an interaction analysis examining differences in intervention effects (for the primary outcome) by each of the following variables: sex, age, ethnicity, education, BMI, physical activity level, dietary restraint, and hunger.

  2. (b)

    Exploratory outcome (b) was the effect of price difference between the initially selected item and the swap offered by intervention (restricted only to groups being offered swaps) on swap acceptance.

  3. (c)

    Exploratory outcome (c) was the acceptance of swaps offered (restricted to the groups being offered swaps), separately for each of the 6 categories: (i) main meals; (ii) jacket potatoes; (iii) soup & sandwiches; (iv) sweet snacks; (v) savoury snacks and; (vi) drinks.

  4. (d)

    Exploratory outcome (d) energy ordered controlling for the energy content of the initial item ordered from each menu, separately for each of the 6 categories: (i) main meals; (ii) jacket potatoes; (iii) soup & sandwiches; (iv) sweet snacks; (v) savoury snacks and; (vi) drinks.

Sample size

We aimed to recruit 2,214 participants. With 80% power, this would allow us to detect a 35 kcal difference at an alpha level of 0.05 (an uncorrected analysis) or a 40 kcal difference at an alpha level of 0.016 (Bonferroni). We applied the Benjamini-Hochberg (BH) correction [15] where the alpha level required was between these two bounds. While a 50 kcal reduction would be a clinically relevant energy reduction for adults [16], we powered the minimum detectable effect size to 35-40 kcal, because we expected the relative effect between the experimental groups to be smaller than the effect between the experimental groups and control. Baseline energy estimates (mean = 423 kcal, SD = 236) were taken from a pilot randomised controlled trial conducted in 6 workplace canteens across the UK [17].

Statistical analysis

A pre-specified statistical plan was published (AsPredicted: 56,358) in advance of the analysis and was followed with minor changes outlined below. Participants had to order at least one food item, not order from all menus (because this was deemed as an implausible lunch order), and checkout to be included in the analysis. The 19 participants (1%) in the intervention groups who were not offered any swaps, because they selected the lowest-energy menu items in all the categories they ordered from were included in the analyses.

  • The primary outcome (energy ordered) was analysed using analysis of covariance (ANCOVA), this analysis was pre-registered as ANOVA but ANCOVA was used to control for the energy content of the first item ordered because that was the baseline value of our dependent variable. We controlled for the energy of the first item a participant ordered because we wanted to control for the initial choices participants made but after the first choice their subsequent choices may have been influenced by previous swaps offered.

  • The secondary outcome (a) (swap acceptance) was analysed using ordinal logistic regression controlling for the number of swaps offered. This analysis was pre-registered as swap acceptance on a scale from 0–6, however, in the analysis those who ordered from all 6 menus were excluded as improbable values meaning that the scale ranged from 0–5 swaps accepted.

  • Secondary outcome (b) (lunch ≤ 600 kcal) was analysed using logistic regression.

  • Secondary outcome (c) (intervention acceptability) was analysed using ordinal logistic regression.

  • Exploratory outcome (a) (interaction) was analysed using two-way ANCOVA with post-hoc tests controlling for the energy content of the initial item ordered to examine interaction terms between interventions (swaps and swaps + PACE) and the following variables: sex (male vs female), age (equal to or above vs below the median), ethnicity (white vs non-white), education (none/secondary vs higher), BMI (≥ 30 kg/m2 vs < 30), physical activity level (meeting guidelines vs not meeting guidelines), dietary restraint (at least the median score vs below the median), and hunger (at least the median score vs below the median). A separate model was conducted for each interaction term.

  • Exploratory outcome (b) (overall price difference) was analysed using multilevel logistic regression with an interaction term for intervention and price difference. Analysis was conducted at the ‘swap-level’ and clustering was used to indicate that observations may be correlated within each participant but would be independent between participants. Robust standard errors were used to account for potential heterogeneous effects. For main hot meals where two swaps were offered, price difference was calculated using the price of the swap that was accepted. Where no swap was accepted, the average price difference for the two different swaps offered was used.

  • Exploratory outcome (c) (swap acceptance by menu) was analysed using logistic regression. Separate models were run for each of the six menus. The control group was the reference category in all models.

  • Exploratory outcome (d) energy ordered by menu. This post-hoc analysis was analysed using ANCOVA, with post-hoc tests controlling for the energy content of the initial item ordered from each menu.

Statistical significance was set at P < 0.05, adjusted with Benjamini-Hochberg (BH) correction [15] for the ANCOVA and regression models (Additional file 1). Mean differences or odds ratios (OR) with 95% confidence intervals (CI) were used to report estimates of comparative effectiveness. Statistical analyses were conducted in Stata (version 16) or SPSS (version 25).


Invitations were sent to a random subsample of a pool of 17,773 eligible panel members. Of those invited, 2,477 participants consented and were equally randomised to 1 of the 3 groups. Of those, 2,150 (86.8%) participants followed the instructions, completed the study, and, thus, were included in the analysis (Fig. 3).

Fig. 3
figure 3

CONSORT flow diagram. Note: Participants who did not follow the instructions had implausible menu orders

Participants were on average 36.8 (SD = 11.6) years old. Just over half (54.3%) were female, 80% identified as white, and 51% had completed tertiary-level education (Table 2). Average completion time for the study was 8.2 min (SD = 3.18 and attention checks showed high levels of engagement with the task (Additional file 3, Table 1).

Table 2 Baseline characteristics of participants. Data are presented as mean ± standard deviation or n (%)

Primary Outcome

Participants ordered on average from 3 (SD = 0.91) menus. The average energy content of lunches ordered was 781 kcal (SD: 315 kcal, range: 226 to 2,226 kcal). The average energy content of final lunch orders was significantly lower in both intervention groups when compared with control [control mean = 819 kcal]: swaps -47 kcal [95%CI: -82 to -13, p = 0.003]; swaps + PACE -66 kcal [95%CI: -100 to -31, p < 0.001] (Fig. 4) The difference in the average energy content of final lunches ordered between intervention groups was not statistically significant [-19 kcal, 95%CI: -53 to 16, p = 0.591].

Fig. 4
figure 4

Means and 95% confidence intervals for energy ordered with BH-adjusted p-values: Meets BH threshold*, p < 0.01**, p < 0.001***. C = control; S = swaps; S + P = swaps + PACE

Secondary Outcomes

  1. (a)

    Swap acceptance

    Almost everyone in the intervention groups (99%) was offered at least one lower-energy swap while placing their lunch order. Table 3 shows the percentage acceptance rate of swaps across the intervention groups. Of the 2,936 swaps offered, 413 (14%) were accepted overall. Table 4 shows the percentage acceptance rate of swaps within each menu across the intervention groups. Compared to the swaps only intervention, the swaps + PACE intervention significantly increased the odds of a swap being accepted, when controlling for the total number of swaps offered [OR: 1.63, 95%CI: 1.27 to 2.09, p < 0.001].

    Table 3 Number (%) of swaps offered and accepted by intervention group (n = 1,424)
    Table 4 Number (%) of swaps offered and accepted within each menu by intervention group (n = 1,424)
  2. (b)

    Meeting public health recommendations

    The proportion of participants who ordered a lunch meeting Public Health England’s energy intake recommendation of ≤ 600 kcal was 26%, 31%, and 36% in the control, swaps, and swaps + PACE groups, respectively. Proportions in the swaps group [OR 1.28, 95%CI 1.02 to 1.61, p = 0.0326] and swaps + PACE [OR: 1.57, 95%CI: 1.25 to 1.96, p < 0.001] were significantly higher than control. Those in the swaps + PACE pace group were not significantly more likely to meet guidelines than those in the swaps group [OR: 1.22, 95%CI: 0.98 to 1.52, p = 0.076].

  3. (c)

    Intervention acceptability

    Almost all participants (92%) believed that it would be acceptable for their employer to implement a pre-ordering system for their workplace canteen. Two-thirds (65%) of participants in the intervention groups (i.e., the swaps and swaps + PACE group) believed that being offered swaps for their food choices while pre-ordering would be acceptable (the control group was not asked about the acceptability of swaps). The odds of those in the swaps + PACE group considering the intervention to be acceptable was 1.32 times [95%CI: 1.09 to 1.60, p < 0.004] that of those in the swaps only group.

Exploratory outcomes

  1. (a)

    Moderation analysis

    There was no evidence that the intervention effect depended upon sex, age, ethnicity, education, BMI, physical activity level, dietary restraint, or hunger (all pinteraction > 0.05, Fig. 5; Additional file 3 Tables 3 – 10).

    Fig. 5
    figure 5

    Interaction effect by sex, age, ethnicity, education, BMI, physical activity level, dietary restraint or hunger. Estimated marginal means for final energy ordered controlling for the energy content of the first item ordered with 95% confidence intervals (CIs). C = control; S = swaps; S + P = swaps + PACE. P describes p-values from interaction analysis

  2. (b)

    Swap acceptance and swap price

    Participants in the control, swaps, and swaps + PACE group spent an average of £5.51 (SD = £1.84), £5.29 (SD = £1.82) and £5.31 (SD = £1.86), respectively on their lunch orders. Table 5 shows the average price of initially selected items and swaps offered for each menu for those in the intervention groups. For 50% of the menus (jacket potatoes, sandwiches and drinks), all swaps offered were either cheaper or the same price as initially selected items. For main hot meals, sweet snacks and savoury snacks, swaps offered were up to £0.55 more expensive than initially selected items.

    For every £1 decrease in swap price, the odds of a swap being accepted increased by 1.85 [95% CI: 1.46 to 2.36, p < 0.001]. The interaction analysis demonstrated a greater effect of price difference (between initially selected items and swaps offered) on swap acceptance for those in the swaps group than the swaps + PACE group [pinteraction < 0.002] (Fig. 6).

    Fig. 6
    figure 6

    Interaction analysis for the effect of price difference by intervention. Predicted probabilities and 95% CIs

    Table 5 Prices of initially selected items, swaps offered and price difference for those offered swaps only
  3. (c)

    Swap acceptance and (d) energy ordered by menu

    Analysis for swap acceptance by menu showed some evidence of differences (Table 4), with participants in the swaps + PACE group being significantly more likely to accept swaps offered for jacket potatoes, sandwiches, and drinks compared to those in the swaps only group (Additional file 3, Tables S11-16). Fig. 7 shows the pairwise comparisons for mean energy ordered (and 95% confidence intervals) in each menu by condition. Significant energy reductions were observed for the intervention groups compared with control on all menus with the largest savings for main meals and jacket potatoes.

    Fig. 7
    figure 7

    Means and 95% confidence intervals for energy ordered (adjusted for energy of the initially selected item in each category) with BH-adjusted p-values: *Meets BH threshold. C = control; S = swaps; S + P = swaps + PACE. n = the total number of items ordered from that category across all 3 arms

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit The Creative Commons Public Domain Dedication waiver ( applies to the data made available in this article, unless otherwise stated in a credit line to the data.


This article is autogenerated using RSS feeds and has not been created or edited by OA JF.

Click here for Source link (