Does the Royal Horticultural Society Campaign for School Gardening increase intake of fruit and vegetables in children? Results from two randomised controlled trials.

The Child And Diet Evaluation Tool

The primary aim of the two linked trials is to determine whether or not children who participate in the RHS advisor-led gardening intervention increase their fruit and vegetable consumption more than those who receive the teacher-led gardening intervention; or whether the teacher-led intervention increases their fruit and vegetable consumption more than no intervention at all. The effectiveness of either intervention (RHS advisor-led or teacher-led) will be determined by an increase in mean intake of one of the following: mean intake of fruit, mean intake of vegetables, or mean intake of fruit and vegetables at follow-up, after adjusting for baseline intake. Dietary intake, with a focus on fruit and vegetable intake, was measured using a modified version of the Child And Diet Evaluation Tool (CADET) questionnaire. 62 The main aim of the CADET diary is to collect accurate information on children’s fruit and vegetable intake, whilst also collecting information on all foods that the children consumed in a 24-hour period.

Part one of the CADET diary comprises a list of 115 separate food and drink types, divided into 15 categories. The categories of foods are cereal (five items); sandwich/bread/cake/biscuit (10 items); spreads/sauces/soup (seven items); cheese/egg (six items); chicken/turkey (three items); other meat (nine items); fish (five items); vegetarian (three items); pizza/pasta/rice (eight items); desserts/puddings (three items); sweets/crisps (four items); vegetables and beans (18 items); potato (two items); fruit (13 items); and drinks (nine items). Part two consists of food-related questions to identify daily consumption of milk, bread, sugar, spreads and fruit juice. It also includes general demographic questions about the family household, questions about the children’s and parents’ attitudes towards fruit and vegetables, and the availability of fruit and vegetables at home.

Portion sizes for children aged 8–11 years

The dietary information from the CADET diary was transferred to a Microsoft Access spreadsheet (Microsoft Corporation, Redmond, WA, USA) using our established in-house software, named Diet And Nutrition Tool for Evaluation (DANTE) (Nutritional Epidemiology Group, Leeds). This used standard predefined algorithms to convert food items into total daily nutrient values for each child, based on the composition of foods. 67 Although the CADET diary upon which the school and home food diaries were based has previously been validated in children aged 3–7 years, it has not been used to collect dietary information in 8- to 11-year-olds. As this study includes children aged 7–10 years, it was necessary to change the standard portion sizes in DANTE to reflect the children’s intake for each year of age (i.e. 8, 9, 10 and 11 years), and to account for differences in intake for boys and girls.

Protocol for determining portion sizes for children aged 8, 9, 10 and 11 years

The portion sizes for ages 8, 9, 10 and 11 years were obtained from the National Diet and Nutrition Survey (NDNS) of young people aged 4–18 years. 68 The NDNS was conducted to explore food consumption and nutrient intake in the general population living in privately owned houses across Britain. The NDNS data are based on an interview and a 4-day food diary as well as blood and urine samples. The NDNS is the most detailed nutrition survey conducted across Britain. A recent update of the report (2008/09–2009/10) confirmed that the overall diet intake was similar to the previous findings. Owing to the validity of these data, it was decided that they would be used to update the CADET portion sizes for older children. 18

From the NDNS data, the mean portion size, number of participants, standard deviation (SD), and maximum and minimum values were extracted. Nearly all the food items used in the CADET were available from the NDNS data and were then further broken down into each age category by gender. Whereas commonly consumed items such as apples and bananas were consumed by, on average, a higher number of participants in each age group (32 and 24 children on average per age group, respectively), several items were consumed by, on average, fewer than five participants per age group. For these foods, the portions had notably higher variation compared with those foods with a higher number of observations. The likelihood that these portion sizes reflected those of the general population that consumed them was questionable. Furthermore, some food items, once broken down into age/gender categories, were found to have missing data. To improve the validity of those foods with low or missing numbers of participants, the rules in the following sections were applied.

Missing data

If any foods included in the CADET dairy were not available as specific codes/items on the NDNS database, then a similar food item was substituted.

Food items with, on average, fewer than five participants per age/gender category

If the item was consumed by fewer than five participants, on average, per age/gender category, an appropriate nutritionally similar food with an average of 10 or more participants per age/gender category would be obtained. The average of the two means would then be calculated in an attempt to reflect actual intake for each category.

An example of this is kiwi, which had, on average, only one person per age/gender category. It also had no value for girls aged 8 years. For kiwi, an average of kiwi and peach, nectarine, plum, apricot and mango was used to ensure a better representation of the average portion sizes for the different age groups consuming them, based on gender. The aim was to smooth out the data where there were extreme values based on one person, and to gain a more valid estimation of intake. For each food that was changed, a line graph was produced containing both the pre-existing food, for example ‘kiwi’, and the modified food, for example ‘average of kiwi and peach, nectarine, plum, apricot and mango’, to visually confirm that the portion sizes looked appropriate. The reason for doing so was to confirm the direction of change in consumption, as at different ages and for different foods, children can not only increase, but also decrease their consumption. Tables 1 and 2 along with Figures 2 and 3 show the portion sizes for children aged 3–11 years, to demonstrate the overall change in portion sizes by age. The calculations of portion sizes as consumed were made only for children aged 8–11 years.

FIGURE 2.

Girls’ portion sizes by age, for different fruits in CADET.

FIGURE 3.

Boys’ portion sizes by age, for different fruits in CADET.

Modifications

Of the 115 foods in the school and home food diaries, 21 had no dietary examples from the NDNS data and 16 had an average sample size of fewer than five participants. Table 3 lists the food items from CADET that did not have a NDNS portion size, and the food groups used as a substitute to create an appropriate portion size for consumers. Table 4 lists the foods with an average of five or fewer participants per age group in the NDNS data, and the food groups that were used as a substitute.

Table 5 shows the final portion sizes for all vegetables, as used in DANTE for the CADET diaries, for boys and girls across the different age groups. Overall, there is a general trend towards a small increase in vegetable consumption for both boys and girls. However, there is more variability between the different ages for both boys and girls in fruit intake (Table 6). Melon and watermelon portion sizes vary greatly between year groups; this could be a consequence of the infrequent consumption of both of these fruits. It was decided not to overmanipulate the data and to leave these portion sizes as they are, as the NDNS data are based on weighed intakes from a nationally representative sample.

Development of the questionnaire on knowledge and attitudes towards fruit and vegetables

One of the secondary outcome measures was ‘Can participating in a school gardening intervention improve children’s ability to identify specific fruit and vegetables and their attitudes towards fruit and vegetables?’

Since the RHS gardening intervention is an educational resource that teaches children about fruit and vegetables through gardening, it has the potential to have an impact on children’s general knowledge of fruit and vegetables. Therefore, one of the other main aims of this study was to explore change in children’s knowledge and attitudes towards fruit and vegetables, to see if there was a difference from baseline to follow-up. A short questionnaire was developed to identify children’s knowledge and attitudes towards fruit and vegetable consumption, and to assess gardening activity levels (see Appendix 1, Child knowledge and attitudes questionnaire). The knowledge questions assessed children’s ability to recognise different fruit and vegetables. Children were presented with a list of 13 fruits, 17 vegetables and one herb, with a colour picture for each, and were asked to draw a line connecting each name with the right picture. The attitude questions were based on previously validated research. 73 Children were asked if they agreed or disagreed with ideas about fruit and vegetables, and were presented with a list of 10 statements, five of these about fruit and five about vegetables. An example is ‘I enjoy eating fruit’. The children had to circle one of four options: ‘agree a lot’, ‘agree a little’, ‘disagree a lot’ or ‘disagree a little’. Images of smiley, neutral or sad faces were presented above each statement to help the children work out their response.

The gardening questions assessed the children’s gardening experience, in terms of what they have grown and what they have tasted. The children were asked to confirm if they had done any gardening (‘yes’ or ‘no’) and then write in the space provided if they had grown any fruit or vegetables. They were then asked to confirm if they had tasted any of the fruit or vegetables they had grown (‘yes’ or ‘no’) and to write down what they had tasted.

To assist with the varying levels of reading ability, this questionnaire was read out to the children as a class, to help them with any difficult words. Furthermore, the teachers and teaching assistants were encouraged to help those children who might struggle with this task, and children were encouraged to put their hands up if they had any questions.

Process measures questionnaires

There were two process measures components for this study; the first was a gardening telephone interview, to identify current level of gardening activities within the school (see Appendix 1); the second was a gardening activity process measures questionnaire to identify the gardening activities that had taken place in each academic year in each school.

School gardening questionnaire

The school gardening questionnaire was a telephone interview. It was designed to identify the school’s baseline gardening level. This questionnaire was based on the RHS benchmarking scheme, which ranks schools in the following categories: (1) planning, (2) getting started, (3) growing and diversifying, (4) sharing best practice and (5) celebrating with the wider community. The schools were asked a series of questions to identify different aspects of gardening currently occurring in their school garden. The questions were focused on the following aspects of gardening in schools: school culture and ethos, the school garden, teaching and learning, and community. Within each of these areas there were several questions that reflected different levels of development within school gardening relating to the five stages of developing a school garden. These questions were adapted as simple ‘yes or no’ questions to be used in a telephone interview. The interviewee was the school staff member who was most involved in the school garden within each school. The questions were structured according to the five categories.

Gardening activity process measures questionnaire

The aim of the gardening activity questionnaire was to identify the level of adherence to the intervention by the schools involved, and to identify any gardening activities that are being undertaken by the control schools. The main aim of the process evaluations was to capture which fruit and vegetables each school grows and harvests. They also aimed to identify which year groups had been involved in the garden each year, whether or not they had started a growing or environmental club, and to find out if the schools had any success or failure stories around the school garden. This information was captured via e-mail in September 2010 for trial years 1 and 2, and again in September 2011 for both trials.

For the schools involved in the RHS intervention, more in-depth information about their intervention activities was captured by the regional advisor and was used to outline changes in school gardening. From this, the level of involvement in the intervention by each school and their adherence to the intervention was identified, as well as success and failure stories reported by the regional advisor himself.

For trial 2 intervention schools, another process measure captured was the level of involvement in the twilight sessions, whereby the regional advisor kept a record of the teacher’s attendance. With this type of intervention, schools were expected to tailor the intervention to their individual needs. By monitoring what activities are undertaken in the school garden, aspects of the intervention that could be associated with dietary change were identified.

Study population

A total of 74 Year 3 and 4 children from two local primary schools in Leeds (mean age 8.4 years) participated in the pilot study. This involved three different class groups: one Year 3, one Year 4 and a mixed Year 3 and 4 class. To evaluate whether the DVD should be sent home or viewed in school, one class was allocated to receive the DVD to watch in class, another was allocated to be given the DVD to take home and the third class was allocated not to receive the DVD at all.

Masters students in nutrition were recruited and trained to administer the CADET diaries and the child knowledge and attitudes questionnaire. The students were asked to record everything the children ate at school by completing the school food diary, and then to go through the child knowledge and attitudes questionnaire as a class. At the end of the school day, one class of children was given the home food diary, one class was asked to watch the DVD before they were given the home food diary, and the final class was given the DVD and the home food diary and asked to watch the DVD with their parents.

Home food diary and instruction DVD

One of the aims of the pilot study was to evaluate whether the DVD should be shown in the classroom at school, or sent home with the children for them and their parents to watch together; there were concerns about children forgetting to return the DVD to the school the next day, and losing the DVD. The results indicated that children who received the DVD to take home and watch with their parents had a higher home food diary return rate (83%) compared with those who watched the DVD in class (73%) or did not receive the DVD (52%). Of those parents who confirmed that they had watched the DVD, all completed the home food diary correctly. Therefore, it was decided that all children should receive the DVD to take home and watch with their parents to improve the quality of the data collected.

School food diary

The fieldworkers were also required to complete the school food diary for all the children in the pilot study. It was brought to our attention that Yorkshire pudding was not included in the school food diary, as one school had it as part of its school dinners; it was then added to both the school and home food diaries. There was also a comment from one of the parents about the home food diary; they stated that they would prefer their ethnicity to be classified as ‘British Asian’ rather than ‘Asian British’. This was rectified.

Data collection protocol

On the second day of data collection, the fieldworkers had two tasks: (1) to check that the home food diary was completed properly, and (2) to complete a diet recall for those children who did not return the home food diary. These results reveal that 25% of the total sample did not return the home food diary. Of the children who were allocated to watch the DVD with their parents, only 17% needed a diet recall to be taken, compared with 27% of those who watched the DVD at school and 42% of those who did not receive the DVD.

Knowledge and attitudes questionnaire results

To assist with the psychological questions and the variability in children’s reading ability, the knowledge and attitudes questionnaire was read aloud to the children and completed together as a class. Teachers were encouraged to assist any children who they thought might struggle with completing the questionnaire.

Administration of the questionnaires was successfully completed. There were six different sections in the child questionnaire. There was only one section which children struggled to complete; this was section 4, containing psychological questions about gardening and fruit and vegetable self-efficacy. Children were asked to respond ‘agree a lot’, ‘agree a little’, ‘disagree a little’ or ‘disagree a lot’ to each of these questions (presented in Table 8). In view of the feedback from fieldworkers, five of the questions were removed. Furthermore, a smiley face or sad face was added under the different options (‘agree a little’, etc.) to help children choose how to respond to each of these questions.

The results also revealed that, on average, when asked how many fruit and vegetables one should eat every day to stay healthy, 52% of the children were not aware that they should consume at least five portions of fruit and vegetables a day.

Trial 1: Royal Horticultural Society-led intervention versus teacher-led intervention

The RHS introduced its Campaign for School Gardening to schools in the London region in the autumn of 2009. The RHS campaign provided intensive support to 10 schools in each region through support from an RHS School Gardening Regional Advisor (the RHS-led intervention). The remaining schools had access to support through twilight training sessions for staff and other activities (the teacher-led intervention). A sample size of 10 schools received the RHS-led intervention, as this was the maximum number of schools that could be supported by one regional advisor. Further details of the intervention components are discussed later in this chapter.

Twenty-six schools from four boroughs in London (Wandsworth, Tower Hamlets, Greenwich and Sutton) were recruited for trial 1. Of the 26 schools, 10 were randomly allocated to receive the RHS-led and 16 to receive the teacher-led intervention. The allocation sequence was generated using Stata Version 11 (StataCorp LP, College Station, TX, USA). All schools were allocated at the same time. It was not possible to randomise schools in trial 1 to receive no intervention at all owing to RHS policy.

Rationale for trial 2

In trial 1 it was not possible to randomise schools to receive no intervention at all (control/comparison group) as it is RHS policy to provide support to all schools who register an interest in the campaign. As a consequence of this, the second set of schools was recruited into a linked trial, trial 2, to provide a ‘no intervention’ arm, i.e. a comparison group.

Trial 2: teacher-led versus delayed intervention

Thirty-two schools from four boroughs in London (Lewisham, Lambeth, Merton and Newham) were recruited for trial 2. These boroughs are adjacent to the trial 1 boroughs. Of these schools, 16 were randomly allocated to receive the teacher-led intervention and 16 were used as comparison schools. The comparison schools received no active intervention during the trial. However, they were informed that once the study ended follow-up collection in February 2012, they would be able to attend the twilight sessions offered to the teacher-led schools.

It was not possible to blind the schools to their intervention group because of the nature of the intervention. The fieldworkers were blinded to the allocation of schools to the intervention (RHS-led or teacher-led) and comparison arms of the study.

Trial 1 inclusion criteria

All non-fee-paying primary schools within four London boroughs (Wandsworth, Tower Hamlets, Greenwich and Sutton) with classes in Key Stage 2 (Years 3–6; children aged 8–11 years) were invited to take part in the study.

Trial 2 inclusion criteria

All non-fee-paying primary schools within four London boroughs (Lewisham, Lambeth, Merton and Newham) with classes in Key Stage 2 (Years 3–6; children aged 8–11 years) were invited to take part in the study.

Exclusion criteria for trials 1 and 2

Independent schools, special schools, schools without all four year groups in Key Stage 2 at primary school (Years 3–6) and small schools with fewer than 15 pupils per year group were excluded.

Proposed sample size

Based on our previous school-based trial, Project Tomato,65 the SD for daily consumption in this age group was estimated to be 85 g for vegetables and 143 g for fruit, with an associated intraclass correlation coefficient of 0.125 for vegetables and 0.114 for fruit. With the proposed sample of one Year 3 class and one Year 4 class from each school, the sample size needed to detect a half-portion (40 g) difference in vegetable intake with 90% power would be 627 children per group, approximately 13 schools. 82 To have 90% power to detect a one-portion difference in fruit intake (one portion = 80 g), 482 children per group would be required, i.e. about 10 schools.

The Project Tomato research identified that approximately 75% of participants completed the dietary questionnaire at baseline and follow-up; therefore, to allow for possible withdrawals and children changing schools, it was decided that 16 schools would be randomly allocated to each group, except for the RHS-led intervention, where the sample size requirements were determined by the staffing levels at the RHS. As a consequence, the RHS-led group had a sample size of 10 randomised schools only, which was carried out by the trial research team.

Discontinuation criteria

Analysis followed the principle of intention to treat. Therefore all schools and children are included in the analysis according to the group to which they were initially randomised. All reasonable and ethical steps were taken to ensure completeness of follow-up of outcome measures.

School withdrawal

If a school wished to withdraw from the trial, the study team would post a data collection form to the head/class teacher along with a freepost envelope. The data collection form would record the following: reasons for withdrawal; whether or not anything could have been done to make taking part in the study easier; confirmation that they no longer wanted to take part in the intervention and receive information/training/materials; and whether or not they still allowed us to use data collected to date and to collect data at round two (i.e. follow-up collection) in October 2011.

Child withdrawal

This occurred if a parent requested to remove their child from the trial. It was anticipated that this request would go to the school, the RHS or the study team at the University of Leeds. Whoever was the first point of contact with the parent was required to inform the other relevant groups (school/RHS/University of Leeds) by telephone, letter or e-mail. A record of any child withdrawals was kept in the database. On receipt of this information, the study team would send a letter to inform the class teacher that the child was to be withdrawn from the study. A data collection form and freepost envelope would be sent via the class teacher to the parent. A covering letter would make it clear to the parent that although the child would not receive any self-study or home-based materials, he or she would not be left out of whole-class activities, as to do so would involve taking the child out of the class while these activities were occurring. The parent would be asked to complete the data collection form and post it back to the Nutritional Epidemiology Group at the University of Leeds in the freepost envelope.

Assessment of harm

On rare occasions, children or schools may need to discontinue the randomised intervention. This may, in most cases, be only a temporary withdrawal; for example, if a child injures him or herself with a spade. Minor adverse reactions were not considered grounds for discontinuing. However, these events were captured either by the RHS regional advisor for the RHS-led schools, or by the Nutritional Epidemiology Group team, through the process measures e-mail, for the teacher-led schools. All adverse events were reported to the Trial Steering Committee. The same notification procedures applied for school or individual withdrawal.

Interim analysis and stopping rules

No interim analyses of trial outcomes were planned.

Randomisation

Cluster randomisation, with school location and borough to identify each ‘cluster’, was used to randomise the schools. The schools were randomised by the study team by geographic location of their London borough and using Stata. From each primary school, one Year 3 class and one Year 4 class was asked to consent to be part of the trial. These classes were randomly selected if there was more than one class in that particular year group.

General considerations

All data collected from these two trials have been reported and presented according to the revised CONSORT statement in Chapter 6. 83

Ethical approval

Ethical approval was obtained through the University of Leeds Research Ethics Committee in 2009. Written informed consent was obtained first from all schools and then from all parents whose children were in the classes chosen to participate in the trial data collection. Schools and parents were informed about the potential risks and benefits of participating in the trial through the information sheet. Participants’ parents gave informed consent, with the opportunity to ‘opt out’ of the study if they did not wish their child to take part. If the parents did not wish their child to participate in the study, the child was still able to take part in the growing activities in the class; however, his or her food intake and child knowledge and attitudes questionnaire were not recorded.

Intervention definitions

• RHS-led intervention These schools received an intervention delivered by the RHS regional advisor.

• Teacher-led intervention Staff from these schools attended twilight sessions of the garden programme at a nearby participating school. The twilight sessions were run by the RHS regional advisor.

The Campaign for School Gardening aims to:

• inspire and empower schools to get growing and to give children the chance to grow and create gardens

• demonstrate the value of gardening in enriching the curriculum, teaching life skills and contributing to children’s mental and physical health

• convince everyone involved with education in schools of the value of gardening in developing active citizens and carers for the environment

• understand the importance of plants and show how gardening can contribute to a sustainable environment.

The Royal Horticultural Society-led intervention

The RHS Campaign for School Gardening consisted of two programmes. The RHS-led intervention schools received the following:

• a day visit from the RHS regional advisor each half-term to work in the garden with teachers and children (summer term 2010 to summer term 2011 inclusive)

• follow-up visits to aid planning by the teachers who were leading the gardening activity (autumn term 2011 to autumn term 2012)

• general ongoing advice on the school garden, and free seeds and tools

• one twilight teacher training session each term (summer term 2010 to summer term 2011 inclusive), based on seasonal tasks in the school garden (open to RHS-led schools’ teachers and others from local schools)

• free access to a wide range of teacher resources at www.rhs.org.uk/schoolgardening/.

The role of the regional advisor was to assist the schools in developing a successful garden, through working directly with teachers and pupils to give them support and practical advice (Figure 5). They were also expected to help overcome barriers to developing gardening within schools. The regional advisor had the expertise and experience to tie in gardening and growing activities with the national curriculum and to run staff training sessions for teachers. The key tasks of the regional advisor were to:

• deliver advice and support to schools in setting up school gardens and growing projects

• promote the RHS Campaign for School Gardening by contacting schools, local education authorities and partner organisations and by giving talks and demonstrations

• train teachers in practical skills to grow plants and harvest crops

• build community links and recruit volunteers to enable the wider community to support and get involved in school growing projects

• contact, advise and support schools within the region by means of visits, e-mails and phone calls

• make links with partner organisations and recruit volunteers to support schools in setting up school gardens and growing plants

• run termly twilight training session courses at 10 school venues throughout the year.

FIGURE 5.

The RHS regional advisor seed sowing at one of the RHS-led schools. Photograph © RHS. Reproduced with permission.

An example of some of the work conducted in one of the RHS-led schools is described below.

• Embedding gardening in the school in order to attain all the benefits which that brings (e.g. most pupils never have access to gardening, as they do not have gardens themselves).

• Establishing a community garden which helps to deal with some of the difficult issues faced in the ‘forgotten estate’.

• Redeveloping the school garden (to be used for class growing).

• Creating simple beds, paths, a fence, and later possibly a greenhouse.

• Digging a pit for the nursery to prevent the raised bed being ‘dug’.

• Clearing the community allotment garden (‘secret garden’). The community garden is to be used for project work, teaching (e.g. about life cycles in a wildlife area) and community beds, and for use by learning mentors to work with children who have learning difficulties and/or behavioural issues.

The two images in Figure 6 below demonstrate the before-and-after effect of the RHS-led intervention in one of the 10 RHS-led schools.

FIGURE 6.

Before-and-after images of the development of the school garden at a RHS-led school. Photographs © RHS. Reproduced with permission.

The teacher-led intervention (‘teacher-led schools’)

The teacher-led intervention schools worked with the RHS by attending termly twilight training at a nearby RHS-led school, to help support them in developing and using their school gardens. Unlike the RHS-led schools, the teacher-led schools did not have direct support from the regional advisor. The regional advisor ran these twilight sessions for them, and provided the teacher-led schools with advice as needed for their school gardens. The following is an example of some of the topics taught in the twilight sessions.

Data sources

The data used in this study came from the following sources.

School-level data

• School gardening level questionnaire, June 2010.

• Gardening in schools – process measures e-mail dated October 2010.

• Information collated from the RHS advisor on school gardening in the intervention schools.

The main outcome measurements were collected at baseline in May and June 2010, when the children were in Years 3 and 4 (aged 7–8 years). The follow-up measurements were collected between October 2011 and January 2012, when the children were in Years 5 and 6 (aged 9–11 years). A breakdown of the different phases of these two trials is illustrated in Figure 7.

FIGURE 7.

Trial phases.

Training the fieldworkers: nutrition students

The primary schools were spread throughout London, and therefore a large sample of undergraduate or masters nutrition students were recruited to undertake baseline collection. These sessions were designed and led by MSC with assistance from one of the research assistants on the trial. The students were recruited from King’s College London and Roehampton University. The students were offered £70 payment per school, and were informed that in order to participate it would be necessary for them to attend one of the two training sessions offered in London. The first training session was at Roehampton University on 9 April 2010; the second was at King’s College London on 12 April 2010. Baseline collection took place from mid-April to July 2010. The students were not informed as to which intervention group the schools they visited were allocated.

Most of the students who registered an interest in the study were dietetic students, who had little data collection experience. In order to ensure that the standard of data collected was consistently high, training was provided to the students to teach them how to complete not only the school and home food diaries, but the child knowledge and attitudes questionnaire as well.

An important quality needed to work with children is presentation skills, the ability to speak confidently in a room full of young children. To assess the students on their ability to complete the baseline collection, the first part of the training required them to introduce themselves and explain how to play one of their favourite childhood games.

The next component of the training was a presentation by MSC introducing the students to the study, and what exactly their tasks would be if they were involved in the data collection. This was the first time the students had seen the questionnaires, so each section was explained to them in detail to help them familiarise themselves with the questionnaires. They were also shown the instructional DVD. The main part of the training consisted of two activities which are explained in detail below.

Sample diet exercise

This exercise involved giving the students examples of children’s food intake for the whole day. The aim was for the students to correctly code each food and categorise it in the right section of either the school or home food diary. An example of a child’s diet is presented in Table 9, shown with the correct school food diary codes. There were always some challenging food items included, which were typical for children to eat but not adults, such as the Dairylea Lunchables and Dunkers ham wrap.

Right or wrong

In the second activity, the students were presented with 10 completed home food diaries and were asked to identify whether the diaries had been completed by the parents correctly or incorrectly. The aim of this exercise was to show the students what to expect on day 2 of the baseline collection, and to identify when it is necessary to take a recall from a child due to serious errors in completion of the home food diary.

At the end of the session, the students had the opportunity to ask questions and raise queries about completing the different questionnaires and the overall structure of the data collection process.

Blinding

The project statistician (CE) allocated a random code for the different intervention groups and the control group involved in both trials. This was done to blind MSC to the intervention allocation while she was conducting the data cleaning and initial primary analysis, to ensure that there was no bias in the data cleaning method. Once the primary analysis was completed, the project statistician informed MSC of the code, so that she could finalise the secondary outcomes and final results. The details of school allocation for both trials was sealed in an envelope and kept in the principal investigator’s office.

Food and nutrient data

Data from baseline and follow-up school and home food diaries, based on CADET, were entered by Swift Research Ltd. The dietary information in the diaries was converted to a Microsoft Access spreadsheet providing the number of portions of 95 food types consumed at each of seven possible meal/snack events (breakfast/before school, morning break, lunchtime, afternoon break, before tea/after school, evening meal/tea and after tea/during night). For example, on the diary a child could tick sugary cereals at breakfast time. The database manager used predetermined age-related portion sizes to estimate the weight of all food types consumed. The database manger then used established in-house software named DANTE, based on the composition of foods67 and using standard predefined algorithms, to convert weights of foods into total daily nutrient values for each child. The 42 nutrients included total energy intake, macronutrients, vitamins and minerals, of which only those associated with fruit and vegetable intake were analysed further. These included total energy, fats, sugars, carbohydrates, fibre (non-starch polysaccharides), carotene, vitamin C, folate, zinc and iron. The 115 food types were reduced further to 14 categories, one of which was fruit (group M) and one of which was vegetables excluding potato (group L). Fruit juice was categorised as one category of group A (drinks). The weights of all types of fruit were summed to give the total weight for fruit, in addition to the total number of portions of fruit (one portion = 80 g). The weights of all types of vegetables were summed to give the total weight for vegetables, in addition to the total number of portions of vegetables.

Each child was given a unique identification code containing information on the school and the child. Follow-up and baseline data were combined using the unique identification codes for the children; therefore, no names or identifying information were included. The database was password protected so that only the database manager, project assistant and administrator and MSC could access the data. Any Microsoft Excel spreadsheets (Microsoft Corporation, Redmond, WA, USA) with children’s names included (these were needed to identify the children for follow-up collection) also contained a password. Only MSC, the project assistant and administrator had access to this password.

Data cleaning

Values for non-dietary data collected at the follow-up phase were checked to ensure that all values were within plausible predetermined ranges. Out-of-range values were checked against original data to identify data entry errors. Errors due to data collection methods were recorded as missing.

Baseline and follow-up data were checked for completeness. Missing data for participants, such as date of birth and gender, were obtained from schools, where possible, by the project administrator. If these details were not available, children who had missing age data were given the mean age of children in their year group (Year 3, 4, 5 or 6). Where gender data were missing, they were given mean portion sizes, based on an average of boys and girls for that particular age group. Where both age and gender data were missing, then both steps above were applied.

The school and home food diaries were entirely tick box-based and were scanned; therefore, they were free of data entry errors. However, it was possible that there were scanning errors, such as diaries scanned the wrong way round or not lined up properly, or random marks mistaken for ticked boxes. Accurate scanning of diaries was initially checked by Swift Research Ltd. On arrival at the Nutritional Epidemiology Group, a random sample of the scanned diaries (approximately 10% of home diaries and 10% of school diaries) was inspected by MSC to provide a further check that the scanning process was accurate. Based on previous research into children’s diet diaries that have mean energy and/or total fruit and vegetable intake, ± three times the SD were identified as outliers and excluded.

Also, it was noticed from inspecting the baseline data that when a child ate fruit salad, several other types of fruit (more than three) were also ticked for that particular mealtime. It was decided to clean this data so that only fruit salad was recorded, as the fruit intake for that particular meal was considered too high for the majority of children.

Linear regression analysis

Linear regression analysis explores the dependency of one variable – in this case, total fruit and vegetables consumed – on one or more other variables, such as gender, by fitting a linear equation to the observed data. Although the fundamental principles of regression remain the same, owing to the cluster randomisation of participants by school, multilevel regression methodology was applied to all statistical analyses in this chapter. 84

Clustered multilevel regression analysis

Multilevel regression analysis is often used for education-based data as it takes into consideration the hierarchical structure of school data. 85 In this study, level one is the individual child and level two is the school. Level 1, the individual level, is considered to be nested within the higher level, i.e. the school. It is based on the theory that all children’s food consumption within a school is similar; for example, children who eat a school meal will all have the same options or choice on any given day at that particular school, and are therefore more likely to consume similar foods. The benefit of this technique is that the means and confidence intervals (CIs) for the different foods and nutrients will be more accurate, if there is variation at school level. As children within a school have more similar food consumption to each other, there will be less variability in the sample from each school compared with a random sample from the whole population. 86 Also, multilevel modelling is not focused on the individual schools within the sample, but on estimating the patterns of variation within the population of schools. 86 If a single-level model was used instead for this analysis, ignoring the hierarchical structures within the data, this would lead to inaccurate or misleading results. The CIs would be too narrow, potentially leading to different conclusions. 86

Study population

This study includes baseline measurements from the children in both trials. These were children from 52 primary schools in the following London boroughs: Wandsworth, Tower Hamlets, Greenwich, Sutton, Lewisham, Lambeth, Merton and Newham.

Variables

The descriptive analysis uses results from the CADET school and home food diaries to describe food and nutrient intakes.

Further analysis used questions in section 2 of the home food diary, which asked about the child’s fruit and vegetable intake and the home environment. The responses were completed by the parent or carer. These questions explored fruit and vegetable habits in the family home:

• Do you have different kinds of fruit/vegetables at home?

• Do you buy a specific fruit/vegetable because your child asks for it?

• Do you cut up fruit/vegetables for your child to eat?

• Do you (parents) eat fruit/vegetables every day?

• Do you eat fruit/vegetables together with your child?

• Do you have to ask your child to eat their fruit or vegetables?

• Do you allow your child to eat as much fruit/vegetables as she/he likes?

The responses to these questions were collected as ‘yes/always’, ‘yes most days/often’, ’sometimes’, ‘rarely’ and ‘never’. General summary statistics, including box plots and histograms of the different categories, were first analysed to identify the best method of coding the data. Based on the frequency of responses to these questions, they were then categorised ‘never/rarely’, ’sometimes’, or ‘always‘.

Four questions were designed to identify the factors associated with consumption habits of the family:72

• the money I have available to spend on fruit and vegetables

• the price of fruit and vegetables

• the time I have available to prepare fruit and vegetables

• likes and dislikes of my family for fruit and vegetables.

The responses to these questions were collected as ‘very important‘, ‘important‘, ‘neither important or unimportant‘, ‘unimportant’ and ‘very unimportant‘. Correlation tests indicated that these questions were highly correlated. These were recoded into a scale of 1 (not important at all) to 5 (very important).

In addition to these questions, the home inventory question ‘Please tick if you have any of the following fruit or vegetables in your fridge/freezer or cupboards’ was collected to identify the variety of fruit and vegetables in the home. The question ‘How many nights a week does your family eat together at a table?’ was asked to explore how the family meal habit affects children’s fruit and vegetable intake. As the response to this question can only be 0–7 it is considered a multinomial variable; therefore, it cannot be treated as a continuous variable. Total fruit and vegetable intake by the eight possible responses was explored. Owing to the similarity in total fruit and vegetable intake, in grams, for people who ate together at a table 1–6 nights a week, the data were recoded into the following: ‘never’ (0 nights a week), ’sometimes’ (1–6 nights a week) and ‘always’ (7 nights a week).

Basic characteristics

A total of 2529 children were asked to participate in the study, from 52 schools. After excluding school withdrawal and parents who did not consent, 2420 children received the intervention. After excluding children who did not complete both the home and school food diaries or who had a total energy and/or total fruit and vegetable intake more than three times the SD of the mean, the final sample size for baseline analysis was 2393, and the response rate was 94%. The mean age of the children (1188 girls and 1205 boys) was 8.3 years (95% CI 8.2 to 8.3 years). Of all the children in the sample, 29% received free school meals and 33% ate a packed lunch. English was spoken as an additional language by 46%, while 59% of children had a member of the family educated to degree level or higher. These results are presented in Table 10.

Children’s nutrient intake

The mean, standard error (SE) and 95% CI for key nutrient intakes for the whole sample are presented in Table 11. The only nutrient not above the recommended mean was vitamin A, which was 100 µg lower than the recommended intake (mean 406 µg, 95% CI 388 to 424 µg). The mean energy intake for all children was 2018 kcal (95% CI 1990 to 2047 kcal). Total fat was 13 g higher than the recommended intake (mean 81 g, 95% CI 79 to 83 g), and sodium was 1508 mg higher than the recommended intake for this age group (mean 2658 mg, 95% CI 2604 to 2711 mg).

Children’s key food and drink intake

The mean, SE and 95% CI for key foods for the whole sample are presented in Table 12. On average, children consumed 94 g of vegetables and 200 g of fruit, with a combined mean of 295 g of fruit and vegetables at baseline. Table 12 also shows the number (%) of children who consumed different foods and the mean intake of this subsample. From this analysis it is evident that 84% of the sample consumed some vegetables on the day of collection and 80% consumed some fruit, with 95% of the children eating either fruit or vegetables. The other most commonly consumed items were drinks; fizzy pop/squash was consumed by 53%, fruit juice by 51% and milk by 43% of the sample.

Fruit and vegetable intake by meal event

Further analysis was conducted to explore fruit and vegetable consumption by lunch type. These results are presented in Table 13 for the whole sample and for consumers only; 2269 children consumed fruit or vegetables during the day, meaning that only 124 children did not consume any at all. The most common times to consume fruit were lunch and before tea/after school, with the largest proportion of children, 38%, consuming fruit at lunchtime. Lunch was also one of the most common mealtimes to consume vegetables, with the largest proportion of children, 58%, consuming vegetables with their evening meal.

Difference in fruit and vegetables between packed lunches and school meals

At lunchtime, children can have either a school meal (provided by the school) or a packed lunch (provided by the parents). Table 13 displays the breakdown of fruit and vegetables based on lunch type. These results show that fruit intake was, on average, 42 g higher in children who had packed lunch meals compared with children who had school meals, and vegetable intake was 33 g higher in children who had school meals compared with children who had a packed lunch.

Differences in key nutrient intake between boys and girls

Multilevel regression analysis was conducted to explore the differences between boys and girls in this sample. Table 14 displays the means and SDs/SEs for boys and girls, and the unadjusted and adjusted regression results. These results identified that there is a significant difference between boys and girls for fibre, potassium, sodium, carotene and vitamin C, after adjusting for ethnicity and IMDS.

Differences in key food and drink intake between boys and girls

Further analysis was conducted only on boys and girls who consumed particular foods or drinks (Table 15). Girls, on average, consumed 20 g (95% CI 12 to 25 g) more vegetables, 14 g (95% CI 10 to 17 g) more dried fruit, 37 g (95% CI 20 to 54 g) more total fruit and vegetables (excluding pulses and beans), 19 g (95% CI 13 to 25 g) more nuts and 11 ml (95% CI –3 to 25 ml) more fruit juice. Boys, on average, consumed 5 g (95% CI 3 to 8 g) more sweets and 63 ml (95% CI 45 to 81 ml) more fizzy pop than girls.

Differences in fruit and vegetable intake by meal event between boys and girls

The differences, by meal events, between boys and girls who consumed fruit and vegetables are presented in Table 16. On average, girls consumed 7 g (95% CI 3 to 11 g) more vegetables at lunchtime and 10 g (95% CI 4 to 15 g) more vegetables with their evening meal than boys, after adjusting for ethnicity and IMDS.

Children’s fruit and vegetable consumption and the home food environment

Clustered (by school) multilevel regression models, with total fruit and vegetable consumption as the primary outcome, were conducted to explore how the home food environment affects children’s fruit and vegetable intake. Table 17 displays the results, unadjusted and adjusted for children’s gender, ethnicity and IMDS.

Mealtime behaviour

Children from families who reported ‘always’ eating a family meal together at a table consumed, on average, 125 g (95% CI 92 to 157 g) more fruit and vegetables than those from families who reported ‘never’ eating a meal together. Children from families who reported ’sometimes’ eating a family meal together ate, on average, 95 g (95% CI 57 to 133 g) more fruit and vegetables than those children who never ate a family meal together at a table.

Parental role modelling and fruit and vegetable consumption

The children of parents who eat fruit and vegetables every day consumed, on average, 87 g (95% CI 37 to 138 g) more fruit and vegetables than children whose parents never/rarely eat fruit and vegetables. Having different types of fruit and vegetables at home was also associated with increased fruit and vegetable intake. ‘Always’ having to ask a child to eat his or her fruit and vegetables had a non-significant inverse relationship with overall intake.

Provision of fruit and vegetables

Children whose parents always cut up fruit and vegetables for them consumed, on average, half a portion more (44 g, 95% CI 18 to 71 g), and those whose parents sometimes cut up fruit and vegetables an average of 21 g (95% CI –6 to 49 g) more, than the children of parents who never cut up their fruit and vegetables. There were no significant differences in fruit and vegetable consumption if parents bought specific fruit and vegetables for their children.

Clustered (by school) multilevel regression models, with total fruit and vegetable intake as the primary outcome, were conducted to explore the effect of the number of different types of fruit and vegetables that people had in their households on the questionnaire completion day. The results indicated that for every additional type of fruit or vegetable in the house, children’s fruit and vegetable intake increased by 5 g (95% CI 4 to 6 g, p < 0.001), after adjusting for sex, ethnicity and IMDS. Further analysis was conducted to explore whether or not there was an association with preparation time and cost of fruit and vegetables on a scale of 1 (unimportant) to 10 (very important). The models showed that there were no significant differences (preparation time: 3 g, 95% CI 0 to 6 g, p = 0.9; cost: 3 g, 95% CI –1 to 6 g, p = 0.9).

Children’s nutrient intake and key foods

Multilevel modelling was conducted to explore whether or not there was any difference in mean nutrient intake depending on family mealtime behaviour. These results are presented in Table 18. The results show that there was a significant difference in mean carbohydrates, fat, sugar, folate, carotene, vitamin C, fruit and vegetable intake and 5 A DAY portions, with higher intakes in families who reported always eating together. For families who reported always eating together at a table, children met the government recommendations for 5 A DAY (5.0 portions, 95% CI 4.8 to 5.2 portions), compared with families who reported sometimes eating together (4.6 portions, 95% CI 4.5 to 4.8 portions) and families who reported never eating together at a table (3.3 portions, 95% CI 2.8 to 3.8 portions).

Strengths and limitations

There were some limitations of this study. There were 887 parents (36%) who did not complete the additional questions, and of these, 23% did not return the home food diary; therefore, the results are potentially subject to response bias. However, no differences were found when analysing with or without the missing participants. The CADET questionnaire was completed by trained fieldworkers in school hours, and by parents for the evening meal and breakfast. Parents and children might be inclined to give socially desirable responses, leading to an overestimation of the association between the home food environment and children’s fruit and vegetable intake. Reverse causation is also possible, in that children with good behaviour at mealtimes may encourage family eating, whereas fussy children might be left to eat alone. This type of dietary assessment has limitations; the portion size assumed for each item in CADET is based on weighed intakes from UK children. A 1-day tick list may not reflect true nutrient intake in the longer term.

Nevertheless, this study is particularly interesting as its population is from London, a highly diverse population in terms of ethnicity and socioeconomic groups. Response at baseline was high at 92%. Although the London boroughs chosen to sample were some of the more disadvantaged, we found that 46% of families spoke English as an additional language (EAL) and 59% had a family member educated to at least degree level. This is higher than the London average, with 55% of primary school children in London speaking EAL in 2012 and 38% of families including someone with a degree. 123 The responding sample may be more advantaged than the general London population. This could have influenced the results obtained, with higher intakes of fruit and vegetables than those found in the NDNS. 18

The dietary data were collected using the previously validated 24-hour food tick list, CADET. The strength of the CADET diary is that it uses age- and gender-specific food portion sizes to calculate food and nutrient intake. A 1-day tick list is an economically effective way of gathering nutrient information from children. Furthermore, all the results were conducted using multilevel analysis. The benefit of this technique is that the means and CIs for the different foods and nutrients will be more accurate; children within a school are more similar to each other in terms of their food consumption, with less variability within the sample compared with a random sample from the whole population. 86,124 In addition to previous research using this tool, a DVD with instructions for completing the questionnaire was sent home for parents/carers and children to watch, and a trained fieldworker reviewed the diary with the children to improve the home food data quality.

Study population

Trial 1 included 23 schools from the following London boroughs: Wandsworth, Tower Hamlets, Greenwich and Sutton. Trial 2 included 31 schools from the following London boroughs: Lewisham, Lambeth, Merton and Newham.

Variables

The primary objective of the trials was to evaluate the RHS Campaign for School Gardening by measuring the change in mean intake of daily portions of fruit and vegetables, daily portions of fruit and daily portions of vegetables, using data derived from the school and home food diaries.

• All three variables are continuous and derived from the nutrient software dietary nutrition tool for evaluation (DANTE).

• These measurements were taken at baseline (April 2010) and again at follow-up (15 months later).

Secondary aims measures

Comparison of intervention and control groups at baseline

School-level baseline characteristics were compared between groups for trials 1 and 2. This was performed to confirm that randomisation had resulted in broadly similar groups, in terms of weights of foods and nutrients and individual and school-level characteristics. Balance of school/class- and child-level variables between the two intervention groups was assessed using the following variables.

Primary outcome analysis of the trial

The variability between the schools determined which type of model should be used for this analysis. The main analysis used a cluster randomised regression random effects model, with change in total fruit and vegetables as the primary outcome to explore the study aims and objectives; results were reported both as unadjusted and adjusted for baseline intake. Analyses using random effects models were used to determine any differences between schools. This analysis was based on the theory of intention-to-treat (ITT) analysis, where all participants are analysed based on their randomised condition at baseline. The output that was generated for the primary analysis included effect size, SE, 95% CIs and p-values, with a p-value of < 0.05% taken to represent statistical significance. Mean values are presented in the tables, in some instances rounding has occurred when differences are referred to in the text.

Description of means of food types and nutrients by intervention status

In addition to comparison of baseline variables, the mean weight (g) of fruit and vegetables consumed on the follow-up CADET data collection day, with SE and 95% CIs, was recorded for all children. This was reported both with and without adjustment for baseline fruit and vegetable levels.

Secondary outcome analysis of the trial

Subgroups were compared by gender, including as an interaction term. A p-value of 0.01 was used to take into account multiple testing. These analyses answer plausibility questions, i.e. whether or not the intervention effect differs by gender.

Sensitivity analysis

This is an epidemiology-based RCT, and therefore it is typical that dropout would occur; approximately 25% of the baseline sample did not complete the trial. Reasons why participants dropped out were described in Chapter 4. Sensitivity analyses were carried out using baseline data brought forward to explore the effect on the primary outcome.

Sample size

Ten schools were randomised to receive the RHS-led intervention and 13 schools to receive the teacher-led intervention in trial 1. In trial 2, 16 schools were allocated to receive the teacher-led and 15 to receive the comparison interventions. Our sample size at baseline for both trials (2529 children allocated to the intervention groups) was less than the original aim of 2900 children. The final sample size reduced to 1554, with only 641 children in total completing trial 1 (RHS-led: 312; teacher-led: 329); similar results were found in trial 2, with 916 children in total completing the trial (teacher-led: 488; control: 428). The response rate at follow-up for the two combined was 62%. This reduced the average group size to approximately 388, which was 94 children fewer than the proposed sample size of 482. This has reduced the power to detect the difference of one portion of fruit and vegetables from 90% to 83%.

The flow of schools and children through both trials is presented in the following four CONSORT diagrams (Figures 8–11).

FIGURE 8.

Trial 1: CONSORT flow chart of schools. CRB, Criminal Records Bureau; n/a, not applicable.

FIGURE 9.

Trial 1: CONSORT flow chart of children. n/a, not applicable.

FIGURE 10.

Trial 2: CONSORT flow chart of schools. n/a, not applicable.

FIGURE 11.

Trial 2: CONSORT flow chart of children. n/a, not applicable.

Regression assumptions

The primary analysis for these trials explored fruit and vegetable intake using multilevel regression analysis, which requires the primary outcome to be broadly normally distributed and the residuals of the regression to be normally distributed. For children, fruit and vegetable intake is rarely normally distributed, as often a percentage of children do not consume any fruit or vegetables on a particular day. This leads to a negatively skewed distribution. Figure 12 shows the possible transformations that might improve the distribution of combined fruit and vegetable intake at follow-up. It is evident from the transformation options that none of these improve the general distribution of follow-up fruit and vegetable intake. Please note that the histogram labelled identity is the distribution without any transformation.

FIGURE 12.

Output exploring if there is a suitable transformation for follow-up total fruit and vegetable intake (ftotalfv).

In addition to exploring the histogram of the distribution of follow-up fruit and vegetable intake, a plot of the residuals was explored to determine if it would be appropriate to use follow-up fruit and vegetable intake, adjusted for baseline fruit and vegetable intake, in the analysis. Figure 13 displays the plot of the residuals for follow-up fruit and vegetable intake from the primary multilevel regression analysis. From the figure it is evident that the distribution is skewed. Therefore, if the analysis was conducted using follow-up fruit and vegetable intake as the primary outcome, the regression assumptions would not be met.

FIGURE 13.

Residuals for total fruit and vegetable intake adjusted for baseline intake.

In an attempt to better meet the regression assumptions, a change in the fruit and vegetable intake (follow-up intake minus baseline intake) variable was created. Figure 14 displays the histogram of the mean change in combined fruit and vegetable intake. It is evident from the histogram that the distribution of change in fruit and vegetable intake is much closer to a normal distribution than follow-up fruit and vegetable intake.

FIGURE 14.

A histogram of mean change in fruit and vegetable intake.

Further analysis of the residuals of mean change in combined fruit and vegetable intake is presented in Figure 15. The plot of the residuals illustrates that change in mean difference in fruit and vegetable intake is broadly normally distributed, making it suitable for multilevel regression analysis.

FIGURE 15.

The residuals for change in mean fruit and vegetable intake.

Change at follow-up has been used before to analyse RCTs. However, it is necessary to assess if there is a baseline imbalance between the two groups in these trials, to determine if it is appropriate to use change instead of adjusting for baseline. As there appeared to be little imbalance at baseline for fruit and vegetables in either trial, change in fruit and vegetable intake was used to analyse the primary outcome for both these trials.

General descriptive

Table 19 describes the demographic details for the children who completed trial 1. The children’s age (RHS-led mean 8.2 years, 95% CI 8.1 to 8.4 years; teacher-led mean 8.1 years, 95% CI 8.0 to 8.3 years), percentages of boys and girls and ethnicity were very similar between the two intervention groups. There was a difference in the percentage of children eligible for free school meals; in the RHS-led group, 33% received a free school meal, compared with 24% in the teacher-led group. The percentage of children with EAL also differed.

Table 20 describes the demographic details for the children who completed trial 2. The children’s age (comparison mean 8.2 years, 95% CI 8.2 to 8.3 years; teacher-led mean 8.3 years, 95% CI 8.2 to 8.3 years), percentages of boys and girls and ethnicity were very similar between the two groups. Again, the ethnic diversity of this sample is illustrated by trial 2. In trial 1 it was evident that there was a difference in free school meal eligibility between the two groups; however, for trial 2 there is very little difference in percentage receiving free school meals, IMDS and percentage of children with EAL.

Table 21 shows the baseline nutrient and food intake for all children in trial 1, broken down by intervention allocation (RHS-led and teacher-led). At baseline, values for key foods, nutrients and energy were all closely matched across the two intervention groups; the mean energy intake for the RHS-led group was 2085 kcal (95% CI 1971 to 2103 kcal) compared with the teacher-led mean intake of 2046 kcal (95% CI 1987 to 2103 kcal). There was only 5 g difference in mean carbohydrates intake (RHS-led mean: 265 g, 95% CI 257 to 272 g; teacher-led mean: 270 g, 95% CI 263 to 277 g); and 5 mg difference in vitamin C intake (RHS-led mean: 108 mg, 95% CI 102 to 114 mg; teacher-led mean: 103 mg, 95% CI 97 to 108 mg). There was a very small difference in fruit and vegetable intake, with the teacher-led group consuming on average more vegetables (RHS-led mean: 86 g, 95% CI 78 to 93 g; teacher-led mean: 101 g, 95% CI 94 to 106 g) and more total fruit (RHS-led mean: 190 g, 95% CI 174 to 204 g; teacher-led mean: 201 g, 95% CI 195 to 224 g).

Table 22 shows the baseline nutrient and food intake for all children in trial 2 broken down by intervention allocation (teacher-led and comparison group). At baseline, values for key foods, nutrients and energy were all closely matched across the two groups. Compared with trial 1, there was a small difference in mean energy intake between the two groups, with the teacher-led group consuming on average 2034 kcal (95% CI 1979 to 2089 kcal) and the comparison group consuming on average 1970 kcal (95% CI 1917 to 2021 kcal). There was a small difference of 11 g in mean carbohydrates intake (teacher-led mean: 267 g, 95% CI 260 to 273 g; comparison group mean: 256 g, 95% CI 250 to 262 g), and a 2 mg difference in vitamin C intake (teacher-led mean: 115 mg, 95% CI 109 to 120 mg; comparison mean: 117 mg, 95% CI 111 to 121 mg). However, unlike the small differences in trial 1 for vegetable intake, in trial 2 there was almost no difference in consumption, with the teacher-led group consuming on average 93 g of vegetables (95% CI 86 to 99 g) and the comparison group consuming on average 98 g (95% CI 90 to 104 g). There was a small difference of 8 g in fruit consumption, with the comparison group consuming slightly more fruit than the teacher-led group (teacher-led mean: 204 g, 95% CI 190 to 216 g; comparison group mean: 196 g, 95% CI 183 to 208 g).

The nutrient and food intake for all children who completed trial 1, compared with children who did not complete the whole trial, is shown in Table 23. Overall, these results reveal that there was very little difference for key nutrients and foods between children who completed trial 1 baseline and follow-up and children who did not complete follow-up. The most noticeable difference was for mean energy intake, with children who completed the trial consuming, on average, 196 kcal less than children who did not complete the trial (completers: 1936 kcal, 95% CI 1879 to 1994 kcal; non-completers: 2090 kcal, 95% CI 2010 to 2169 kcal). However, there was very little difference in mean vitamin C intake (completers: 102 mg, 95% CI 97 to 107 mg; non-completers: 103 mg, 95% CI 97 to 110 mg) and mean vegetable consumption (completers: 91 g, 95% CI 85 to 97 g; non-completers: 93 g, 95% CI 85 to 100 g). There was a small difference of 11 g in mean fruit intake, with the children who completed the trial consuming, on average, more fruit than children who did not complete the trial (completers: 200 g, 95% CI 187 to 213 g; non-completers: 189 g, 95% CI 172 to 206 g).

Additional descriptive analysis was conducted to explore the baseline nutrient and food intake for children who did not complete follow-up by intervention allocation. These results again revealed very little difference for children who did not complete trial 1 by intervention allocation. As expected, there was a slight difference in energy consumption, with the teacher-led group consuming more than the RHS-led group (RHS-led mean: 2046 kcal, 95% CI 1922 to 2169 kcal; teacher-led mean: 2119 kcal, 95% CI 2015 to 2223 kcal). Similar findings were found for the primary outcome measures of fruit and vegetable intake, with the teacher-led group consuming slightly more (for vegetables, RHS-led mean: 85 g, 95% CI 71 to 97 g; teacher-led mean: 98 g, 95% CI 88 to 107 g, and for fruit, RHS-led mean: 167 g, 95% CI 140 to 192 g; teacher-led mean: 204 g, 95% CI 181 to 226 g).

The nutrient and food intake for all children who did not complete follow-up in trial 2 by intervention allocation revealed very little difference for children who did not complete the trial. There was, on average, only 10 kcal difference between the teacher-led and comparison groups (teacher-led mean: 2020 kcal, 95% CI 1912 to 2126 kcal; comparison group mean: 2030 kcal, 95% CI 1943 to 2116 kcal). Similar results were found for the primary outcome measures of fruit and vegetable intake, with the teacher-led group consuming slightly more (for vegetables, teacher-led mean: 95 g, 95% CI 85 to 104 g; comparison group mean: 87 g, 95% CI 74 to 99 g, and for fruit, teacher-led mean: 199 g, 95% CI 177 to 219 g; comparison mean: 195 g, 95% CI 170 to 218 g).

Mean nutrient and food intake at baseline for all children who completed trial 2 compared with children who did not complete the trial (Table 24) found similar results to trial 1. Overall, these results reveal that there was very little difference for key nutrients and foods between children who completed trial 2 baseline and follow-up and children who did not complete follow-up. Similar to trial 1, the most noticeable difference was for mean energy intake, with children who completed the trial consuming, on average, 135 kcal less than children who did not complete the trial (completers: 1891 kcal, 95% CI 1839 to 1942 kcal; non-completers: 2026 kcal, 95% CI 1959 to 2092 kcal). This difference, however, was smaller than the difference in kcal intake seen in trial 1. Again, for trial 2 there was very little difference in mean vitamin C intake, in this case only 1 mg (completers: 112 mg, 95% CI 107 to 116 mg; non-completers: 111 mg, 95% CI 105 to 117 mg); and there was no difference in mean vegetable consumption (completers: 92 g, 95% CI 86 to 98 g; non-completers: 92 g, 95% CI 84 to 99 g). There was, however, a small difference of 9 g in mean fruit intake, with the children who completed the trial consuming, on average, more fruit than children who did not complete the trial (completers: 190 g, 95% CI 179 to 201 g; non-completers: 199 g, 95% CI 182 to 214 g).

The baseline nutrient and food intake for all children who did complete baseline and follow-up in trial 1 showed small differences (Table 25), with a mean energy intake for the RHS-led group of 2034 kcal (95% CI 1956 to 2111 kcal) compared with the teacher-led mean intake of 1993 kcal (95% CI 1925 to 2059 kcal). There was only 2 g difference in mean carbohydrates intake (RHS-led mean: 265 g, 95% CI 256 to 273 g; teacher-led mean: 267 g, 95% CI 259 to 275 g) and 3 mg difference in vitamin C intake (RHS-led mean: 108 mg, 95% CI 100 to 115 mg; teacher-led mean: 105 mg, 95% CI 98 to 112 mg). There was a small difference in fruit and vegetable intake, with the teacher-led group consuming on average more vegetables (RHS-led mean: 87 g, 95% CI 78 to 95 g; teacher-led mean: 102 g, 95% CI 93 to 110 g) and more total fruit (RHS-led mean: 201 g, 95% CI 183 to 219 g; teacher-led mean: 214 g, 95% CI 195 to 232 g). This difference in intake was also noted in the 5 A DAY variable (RHS-led mean: 342 g, 95% CI 319 to 364 g; teacher-led mean: 374 g, 95% CI 347 to 382 g). The baseline nutrient and food intakes overall, however, are very similar in terms of levels of nutrients; this would suggest there was no evidence of imbalance between the groups.

Table 26 shows the baseline nutrient and food intake for all children who did complete baseline and follow-up for trial 2. At baseline, the values for key foods, nutrients and energy were all closely matched across the two groups. Similar to trial 1, there was a small difference in mean energy intake between the two groups, with the teacher-led group consuming, on average, 2039 kcal (95% CI 1974 to 2103 kcal) and the comparison group consuming, on average, 1932 kcal (95% CI 1867 to 1996 kcal). There was only 13 g difference in mean carbohydrates intake (teacher-led mean: 267 g, 95% CI 259 to 275 g; comparison group mean: 254 g, 95% CI 246 to 275 g) and no difference in vitamin C intake (teacher-led mean: 118 mg, 95% CI 111 to 124 mg; comparison mean: 118 mg, 95% CI 111 to 124 mg). Again, there were similar results for the groups in trial 2 for fruit and vegetable consumption, with only small differences between the groups. The teacher-led group consumed, on average, less vegetables (teacher-led mean: 95 g, 95% CI 87 to 102 g; comparison group mean: 100 g, 95% CI 90 to 108 g) and more fruit (teacher-led mean: 206 g, 95% CI 190 to 221 g; comparison group mean: 193 g, 95% CI 177 to 209 g).

Change in fruit and vegetable intake: trial 1

Table 27 displays the changes in fruit intake, vegetable intake and combined fruit and vegetable intake (follow-up minus baseline) and the intervention mean difference for trial 1, both unadjusted and adjusted for IMDS, age, gender and ethnicity. For both groups, there was a small decrease in fruit intake after adjusting for possible confounders (RHS-led mean: 8 g, 95% CI: –69 to 52 g; teacher-led mean: 20 g, 95% CI –36 to 77 g). For vegetable consumption there were no significant differences found between the unadjusted and adjusted models (intervention effect: –13 g, 95% CI –39 to 11 g). The teacher-led group did have, on average, a higher mean change in vegetable consumption, of 29 g (95% CI –6 to 66 g) compared with 16 g (95% CI –11 to 38 g) in the RHS-led group.

For combined fruit and vegetable intake there was a borderline significant difference in the unadjusted model (intervention effect: 40 g, 95% CI –1 to 80 g; p = 0.05), with the teacher-led group having a higher mean change of 8 g (95% CI –19 to 36 g) and the RHS-led group a mean change of –32 g (95% CI –60 to –3 g). However, after adjusting for possible confounders this difference was not significant (intervention effect: –42 g, 95% CI –88 to 1 g; p = 0.06).

The plot of the school residuals with their 95% confidence limits are presented in ascending order in Figure 16. All of the schools do pass through zero, indicating that the schools do not differ significantly from the average line at the 5% level. From the adjusted model, results state that 1.2% of the variance in change in mean fruit and vegetable intake can be attributed to the difference between schools.

FIGURE 16.

Plot of the school residuals for change in fruit and vegetable intake in trial 1, in ascending order with their 95% confidence limits.

Change in fruit and vegetable intake: trial 2

Table 28 displays the changes in fruit intake, vegetable intake and combined fruit and vegetable intake at baseline and follow-up and the intervention mean difference for trial 2, both unadjusted and adjusted for IMDS, age, gender and ethnicity. For mean change in fruit intake, the teacher-led group (mean change 44 g, 95% CI –28 to 118 g) increased their consumption, on average, by 22 g more than the comparison group (mean change 22 g, 95% CI –50 to 94 g). However, these differences in mean change for fruit intake were not significant in the unadjusted or adjusted models. For vegetable intake, the comparison group consumed, on average, more vegetables (mean change 17 g, 95% CI –30 to 21 g) compared with the teacher-led group (mean change 10 g, 95% CI –36 to 52 g). However, this difference was not significant.

As a result of having a higher intake of fruit, the teacher-led group consumed on average 15 g (95% CI –36 to 148 g) more fruit and vegetables than the comparison group. This difference was not significant in either the adjusted or the unadjusted model.

The plots of the residuals with their 95% confidence limits are presented in ascending order in Figure 17. It is evident that the majority of the schools do pass through zero, indicating that they do not differ significantly from the average line at the 5% level. The adjusted model of mean change in combined fruit and vegetable results shows that 7.3% of the variance in mean change in fruit and vegetable intake can be attributed to the difference between schools.

FIGURE 17.

Plot of the school residuals for change in fruit and vegetable intake in trial 2, in ascending order with their 95% confidence limits.

Differences in nutrients and key foods

For both trials, the differences in key nutrients and foods were explored to see if there was an effect of either intervention on their mean intakes. Results are presented for trials 1 (Table 29) and 2 (Table 30), both unadjusted and adjusted for age, gender, ethnicity and IMDS. Overall, there was very little difference in either trial for these key nutrients and foods. The mean differences were small for nearly all nutrients and foods, except for energy and carotene intake. Although there were differences in mean intakes for these two, they were not significant. The only significant difference was found in trial 1 for vitamin C intake in the adjusted model. Once the adjustments were made there was a 12.7-mg-per-day difference between the RHS-led and teacher-led groups, with the teacher-led group having a significantly higher intake of vitamin C.

Differences in food and drink intake

An additional analysis was conducted to determine if there were differences in non-essential food intake (sweets, toffees, mints, chocolate bars, crisps, savoury snacks) and commonly consumed drinks (milk; fizzy pop, squash, fruit drink and pure fruit juice). For both trials, no differences were found in intakes of non-essential foods or drinks, after adjusting for age, gender, ethnicity and IMDS. Overall, there was very little difference between the different intervention groups. In trial 1, the RHS-led group consumed, on average, 19 ml less milk (95% CI –49 to 11 ml) than the teacher-led group. However, this difference was not significant.

Baseline values brought forward

Sensitivity analysis was carried out using baseline data brought forward to explore the effect on the primary outcome. The results from this analysis are presented for trial 1 in Table 31 and for trial 2 in Table 32. The same methodology used to explore the intervention effect in the main analysis was applied to baseline values brought forward. There was very little difference in the results for baseline brought forward compared with the main trial analysis for trial 1. Instead of having a decrease in mean change in fruit intake, there is almost no change (2.0 g) for the RHS-led group and a change of 10.7 g for the teacher-led group. The mean difference in vegetable intake increases from 13 g to 35 g; however, after the adjustments are made this difference is not significant. The difference in combined change in fruit and vegetable intake was negligible between the main ITT model and baseline brought forward for trial 1.

The plots of the residuals with their 95% confidence limits are presented in ascending order in Figure 18. There was even less divergence from zero for all the schools, indicating that the schools do not differ significantly from the average line at the 5% level. From the adjusted model of the mean change in combined fruit and vegetable intake, results state that 0.1% of the variance in change in mean fruit and vegetable intake can be attributed to the difference between schools.

FIGURE 18.

Plot of the school residuals for change in fruit and vegetable intake in ascending order with their 95% confidence limits, for trial 1 baseline values brought forward.

For trial 2, displayed in Table 32, differences in the main analysis and the baseline brought forward are minor, with only a slight decrease in all three mean differences for fruit, vegetables and combined fruit and vegetable intake. Again, once adjusted for the covariates, these differences in mean intakes were not significant.

Figure 19 shows very little difference compared with the main analysis. The overall plot shows the majority of the schools do pass through zero, indicating that the schools do not differ significantly from the average line at the 5% level. From the adjusted model of the mean change in combined fruit and vegetable intake, results state that 3.8% of the variance in change in mean fruit and vegetable intake can be attributed to the difference between schools.

FIGURE 19.

Plot of the school residuals for change in fruit and vegetable intake in ascending order with their 95% confidence limits, for trial 2 baseline values brought forward.

Differences between boys and girls by intervention allocation

Differences in fruit and vegetable intake between boys and girls by intervention allocation were explored. There is very little difference between boys and girls in the RHS-led group, with both showing a mean decrease in fruit consumption (girls: –34 g, 95% CI –68 to –1 g; boys: –31 g, 95% CI –64 to 2 g) and for vegetable consumption, almost no difference (girls: –1 g, 95% CI –17 to 30 g; boys: 5 g, 95% CI –14 to 25 g) (Table 33). For the combined mean change in fruit and vegetables, boys in the RHS-led group decreased their consumption less than girls (girls: –37 g, 95% CI –76 to 1.2 g; boys: –26 g, 95% CI –65 to 12 g). Results for the teacher-led schools revealed that the girls tended to consume more vegetables than boys (girls: mean change 28 g, 95% CI 6 to 49 g; boys: mean change 5 g, 95% CI –16 to 26 g). This difference was also reflected in combined fruit and vegetable intake, with girls on average having a mean change of 15 g (95% CI –63 to 55 g) in fruit and vegetable consumption, compared with 2 g change for boys (95% CI –36 to 63 g).

For trial 2, there was very little difference between the two groups in either the teacher-led intervention or the comparison group, with the teacher-led girls having a slightly higher mean change of 32 g (95% CI –27 to 91 g) in fruit and vegetable intake, compared with the boys’ mean change of 27 g (95% CI –32 to 87 g) (Table 34). In the comparison group there was a difference in fruit intake between boys and girls, with the girls having a decrease in mean intake of 9 g (95% CI –46 to 28 g) and the boys having a mean change of –17 g (95% CI –50 to 14 g). However, their vegetable and combined fruit and vegetable intakes were very similar.

Differences between boys and girls interaction effect

Additional analysis was conducted to explore whether or not there was an interaction between gender and the intervention. The results from this analysis are presented in Tables 35 and 36. After adjusting for age, ethnicity and IMDS, no interaction effect of gender was detected.

Fruit and vegetable consumption

The results from these trials revealed that there was very little difference in children’s mean change in fruit, vegetable or combined fruit and vegetable intake. For both trials, the teacher-led group had slightly higher mean intakes of vegetables and combined fruit and vegetables than the RHS-led or comparison group; however, there was no significant intervention effect after taking into consideration the adjustment for possible confounders. Only five other studies measured the relationship between children’s fruit and vegetable intake and a gardening intervention. 14,47,48,125,126 The results from these five studies were mixed, with two studies revealing a significant difference for fruit and vegetable intake,14,48 one125 finding that boys had a higher consumption of fruit and vegetables compared with girls, and one reporting a significant increase in vegetable consumption,126 while the fifth study found no differences in fruit or vegetable intake (measured separately only). 47

Of the four studies that did show an effect on fruit or vegetable intake, two used self-selection to determine which school received the intervention. 48,126 In one study, the teacher was able to choose whether they received the intervention or not,126 and the other was based on existing gardening activities within the schools. 48 One study47 stated that the head teacher chose which classes would receive the intervention, and the fourth study14 used convenient sampling for the three schools involved; two of the three schools were randomly assigned, whereas the third school was assigned based on garden availability. However, for both of the current trials, gardening area or existing activities was not a requirement and all schools were randomly assigned to their intervention group.

Nutrient consumption

For both trials, the differences in key nutrients and foods were explored to see if there was an effect of either intervention on their mean intakes. Overall, there was very little difference in either trial for these key nutrients and foods. The only significant difference was in trial 1 for vitamin C intake in the adjusted model. Once the adjustments were made there was a 13.0-mg-per-day difference between the RHS- and teacher-led groups, with the teacher-led group having a significantly higher intake of vitamin C. A previous study14 explored key nutrients, identifying significant increases in dietary fibre and vitamins A and C in the gardening intervention group compared with the control group.

Potential barriers to changing children’s fruit and vegetable consumption

Changing children’s fruit and vegetable consumption patterns is a challenging task. The main barriers to increasing children’s fruit and vegetable intake are availability, convenience, taste preferences, peer pressure, parental/school support and knowledge. 49 Successful implementation of an intervention is often determined by the time allocated and the teachers’ and parents’ perceptions of its importance. The main barrier for teachers in implementing school-based interventions is preparation time. 127 The teacher’s willingness to teach the intervention and own beliefs in the importance of the garden could explain the current findings that, although not significant, the teacher-led intervention tended to have a higher increase in fruit and vegetable consumption compared with the RHS-led intervention and the comparison group. Another important geographical component to acknowledge when evaluating the success of a gardening intervention is that all of the successful interventions were located in the USA, in areas where fruit and vegetables could be grown all year round. In the current research, the length of the growing season may have had an effect on the outcome. Further analysis exploring how the delivery and implementation of the intervention may have affected the primary outcome is described in subsequent chapters.

Limitations and strengths

One of the disadvantages of the design of this research was having two trials instead of one. Therefore, the difference between the RHS-led intervention and the comparison group could not be analysed. However, for both trials the study design was robust, using randomisation to determine which schools received the different interventions. This is the first clustered RCT to evaluate the effectiveness of a school gardening intervention on children’s fruit and vegetable consumption. One of the main limitations of previous studies in this area was their study design and the use of convenience sampling. 41–43 The strength of the current study is that schools were randomised to either one of the intervention groups or the comparison group; therefore, there was no possibility of introducing selection bias into the study. One of the fundamental problems with previous research in this area is that schools were selected based on their having or not having a garden; those without a garden were used as control schools. 14,44 Other biases included constraints of the school district and characteristics such as pupil numbers,46 self-selection and teachers being given the option to choose their condition group. 47,48,126

The sample size at baseline for both trials was lower than anticipated, with the response rate at follow-up for both trials combined being 62%. This reduced the average group size to approximately 388, 94 children fewer than the proposed sample size of 482, reducing the power of the study. Small sample sizes can lead to an underestimation of the SEs and affect the sensitivity of the tests used to determine the statistical differences between the groups. The sensitivity analysis (baseline brought forward) was conducted to explore whether or not the reduced sample size had an effect on the primary outcome. The results were very similar to those of the main analysis, suggesting that the reduced sample size did not affect the primary analysis of the trial. Furthermore, these trials are the largest trials to evaluate school gardening to date. Some of the studies in this area had a very small sample size, with some only involving a few schools or one school implementing the intervention. 41,42,44,125,128,129 Furthermore, the current trials involved a highly diverse population in terms of ethnicity and socioeconomic groups.

The dietary data were collected using a validated 24-hour food tick list (CADET) for children aged 3–11 years. The strength of the CADET diary is that it uses age- and gender-specific food portion sizes to calculate food and nutrient intake. A 1-day tick list is an economically effective way of gathering nutrient information from children. However, the disadvantage of using a 24-hour food frequency questionnaire is that it uses pre-allocated portion sizes for each item in CADET, based on average weighed intakes from UK children. 62 A 1-day tick list may not reflect true nutrient intake in the longer term. This study attempted to improve the quality of the data by providing parents and children with an instruction DVD to help explain how to complete the CADET home food diary. In this study, the trained fieldworkers also collected and reviewed all home food diaries. This was for two reasons: to reduce errors in the data collected to make sure that children did consume everything ticked on the diary, but also to obtain a retrospective recall for children who did not return the home food diary. The CADET diary does avoid these issues with child self-reported food intake, and is less of a burden on participants than the most commonly used alternative, a weighed 4-day food diary.

All the results were analysed using a robust statistical methodology, multilevel analysis. The benefit of this technique is that the means and CIs for the different foods and nutrients will be more accurate, as the children within a school are more similar to each other in terms of their food consumption, with less variability within the sample, compared with a random sample from the whole population. 86,124 The primary outcome measuring children’s fruit and vegetable consumption, using multilevel regression analysis, was originally intended to explore differences in follow-up intake, adjusting for baseline intake. However, owing to the negative distribution of the residuals for follow-up fruit and vegetable intake, a change score was calculated by subtracting baseline fruit and vegetable intake from follow-up intake. For both trials there was no imbalance between intervention and comparison groups for baseline intake of fruit and vegetables, suggesting that this was an appropriate methodology. 130

Conclusion

This is the first clustered RCT to explore whether or not a gardening intervention can increase children’s fruit and vegetable intake. The results showed that there was no change in children’s fruit and vegetable intake after receiving either the RHS-led or the teacher-led intervention.

Fruit and vegetable knowledge

To assess knowledge of the 5 A DAY fruit and vegetable campaign, children were asked to circle on the child questionnaire a number between 1 and 8 in answer to the question ‘How many servings of fruit and vegetables do you think you should eat every day to stay healthy?’ To test children’s ability to recognise different fruit and vegetables, they were asked to draw a line from the names of 12 fruits and 16 vegetables to a colour photo of each item. Apple was provided as an example. All the fruits were listed and pictured on one page; these were raspberries, blackberries, pears, blueberries, plums, bananas, grapes, orange, pineapple, nectarine, watermelon and kiwi fruit. The following vegetables were listed on another page: courgettes, spinach, French beans, parsley, lettuces, parsnips, radish, sweetcorn, carrots, leeks, spring onions, broccoli, peppers, cucumber, tomatoes and garlic (see Appendix 1). For each item, correct responses were coded 1 and incorrect responses were coded 0.

Attitudes towards fruit and vegetables

To assess attitudes towards fruit and vegetables, the children were asked to circle responses indicating whether they agreed a lot, agreed a little, disagreed a little or disagreed a lot with 10 questions (Table 37), such as ‘I enjoy eating fruit’ or ‘I like trying new fruit’, which relate to perceived barriers to consumption. Self-efficacy was assessed using ‘I try to eat lots of fruit’ and ‘I’m good at preparing fruit and vegetables’. Perceived social influences and availability in the home environment were evaluated with the questions ‘My family encourages me to eat fruit and vegetables’ and ‘There’s usually lots of fruit and vegetables to eat at home’.

Gardening experience

To determine gardening experience, the children were asked to circle ‘yes’ or ‘no’ in answer to ‘We grow fruit and vegetables in our garden or allotment’. They were then asked ‘What fruit or vegetables have you grown?’. For each child the number of different types of fruit and vegetables listed were coded as two separate variables. Finally, they were asked ‘Have you tasted any fruit or vegetables from your garden or allotment?’ (‘yes’ or ‘no’), and ‘What fruit or vegetables have you tasted?’. Each child’s list of tasted items was compared with his or her list of own-grown fruit and vegetables, and recorded for analysis as ‘none’, ’some’ or ‘all fruit and vegetables grown’.

Response rate

In trial 1, 404 children (69%) from the teacher-led group and 373 (70%) from the RHS-led intervention attempted parts of both the baseline and follow-up child knowledge and attitudes questionnaire. In trial 2, 559 children (77%) from the teacher-led intervention and 541 (71%) from the control group attempted this. Not all of these children completed every section of the questionnaire. The numbers of children with completed dietary data were 329 (56%) from the teacher-led group and 323 (61%) from the RHS-led group in trial 1, and 500 (69%) from the teacher-led group and 431 (57%) from the comparison group in trial 2.

Attitudes towards fruit and vegetables

In relation to children’s attitudes towards and perceptions about fruit and vegetables, > 90% of the children from the two trials at both baseline and follow-up agreed that eating vegetables every day kept them healthy and that their parents encouraged them to eat these. Over 90% of the children at both baseline and follow-up agreed that they enjoyed eating fruit, whereas 60–70% agreed that they enjoyed eating vegetables, and only 50–60% agreed that they liked trying new vegetables (see Table 37). In trial 2 (Table 38), children in the gardening intervention group were significantly more likely to agree that they enjoyed eating vegetables at follow-up compared with the control group (69.5% vs. 61.7%), even after adjusting for baseline answers (OR = 1.3, 95% CI 1.0 to 1.8); however, this was not significant after adjusting for gender, ethnicity or IMDS (OR = 1.2, 95% CI 0.9 to 1.6, p = 0.1). There were no other significant differences in trial 2 for this section of the questionnaire, and there were no significant differences relating to vegetables in trial 1 with regards to answers at follow-up. However, children in the RHS-led group in trial 1 were significantly less likely to agree that they tried to eat lots of fruit or liked to try new fruit than those in the teacher-led group, even after baseline adjustments (OR = 0.4, 95% CI 0.2 to 0.8, p = 0.009 and OR = 0.5, 95% CI 0.2 to 0.9, p = 0.05 respectively). In addition, after further adjustment for sociodemographic factors (including deprivation score), children in the RHS-led group were significantly less likely than those in the teacher-led group to agree that there were lots of fruit and vegetables to eat at home (OR = 0.4, 95% CI 0.2 to 0.9, p = 0. 02).

At baseline a high proportion of children (> 67%) knew that five servings of fruit and vegetables should be eaten every day to stay healthy. Of the children who answered this question at both baseline and follow-up, there was no significant difference in the proportion giving correct answers at follow-up between the RHS- and teacher-led groups in trial 1 (79% vs. 79%). However, there was a significant difference between the intervention groups in trial 2 at follow-up, with the teacher-led group giving more correct answers than the comparison group (79% vs. 68%). From the multilevel logistic regression analyses, a significant difference remained (OR = 1.7, 95% CI 1.1 to 2.6, p = 0.006) after adjusting for baseline answers (which were significantly different between groups) and also after further adjustment for sociodemographic factors (OR = 1.7, 95% CI 1.1 to 2.5, p = 0.004).

Additionally, there was no evidence that the school gardening interventions significantly increased the likelihood of children tasting their own-grown fruit and vegetables.

The children’s ability to recognise different fruit was already very good at baseline. Table 39 reports the mean number of fruit and vegetables recognised at baseline and follow-up for both trials. It is evident when comparing the change in total fruit recognised from baseline to follow-up that there was no significant difference between intervention groups, in either trial 1 or trial 2, in the unadjusted independent t-test analyses or after adjustment for sociodemographic variables in multilevel analyses. Similarly, there was no significant difference in the change in total vegetables recognised between intervention groups in trial 2. However, in trial 1 there was a significantly larger increase in the number of different vegetables recognised from baseline to follow-up for the RHS-led group compared with the teacher-led group (a mean increase of 2.44 vs. 1.65 out of a total of 16 vegetables). This was statistically significant (p = 0.03) in multilevel analysis after additionally adjusting for sociodemographic factors.

Similarly, in trial 1 there was a significantly larger increase in the total number of fruits and vegetables recognised from baseline to follow-up for the RHS-led group compared with the teacher-led group (p = 0.007 in the t-test), but this was not significant after adjusting for sociodemographic variables in multilevel models.

In both trials (Tables 40 and 41; Figures 20–23), ≥ 80% of children were able to identify each type of fruit on the questionnaire, apart from blackberries, blueberries, plums and nectarines, which were identified by only ≥ 64% of children. Over 90% of the children could identify pears, bananas, grapes, oranges, pineapples and watermelons. In contrast, the ability to recognise vegetables was more varied, with 90% of children recognising sweetcorn, carrots, peppers and tomatoes, but only 50% identifying spinach, parsley, leeks and spring onions.