Description of both projects
The Pro Children Study is a cross- European study on F&V intake among primary schoolchildren (age; 10–11 years old at baseline) . For the present analyses, we only included the Dutch intervention part of the Pro Children Study, which was implemented in Rotterdam, one of the major cities in the Netherlands.
The Pro Children Study addressed a wide range of important determinants of F&V intake based on a previously published theoretical framework . This theoretical framework recognizes the role of the physical environment, such as the availability and accessibility of foods , as well as social environmental factors . Among children, the role of family environment is of specific importance. Furthermore, children spend a considerable amount of their time at school, and the school environment may also importantly influence nutrition and physical activity behaviors .
The Schoolgruiten Project is also a Dutch intervention study among primary schoolchildren (age; 9–10 years old at baseline). This project was implemented in two intervention cities in the Netherlands: in The Hague, one of the major cities in the west part of the Netherlands, and in Almelo, a medium sized city in the east part. Similar to the Pro Children Study, the main strategy within the Schoolgruiten Project was targeting taste preferences, availability and accessibility.
For the present study, we only included children from the intervention schools, since these children are more likely to show changes in potential determinants of F&V intake, as a consequence of the intervention activities [20, 21]. The data was used as observational longitudinal cohort data.
Design of the studies
The baseline survey of the Pro Children Study was conducted in September 2003. First follow-up was performed nine months after the baseline measurement (May 2004) and second follow-up was performed exactly one year after first follow-up (May 2005).
The baseline survey of the Schoolgruiten Project was conducted in The Hague in the spring of 2003 and in Almelo in the autumn of 2003. First follow-up was conducted in both cities exactly one year later and second follow-up was conducted exactly two years later.
During the intervention period the intervention schools of the Pro Children Study were provided with a piece of fruit or ready-to-eat vegetables (cherry tomatoes, baby carrots) for free during a fruit break twice a week. In addition, a classroom curriculum was implemented, which consisted of worksheets and a web-based computer tailored feedback tool . Furthermore, parents were encouraged to be involved in the project by means of their children's homework assignments, parental newsletters, and a parent version of the web-based computer tailored tool that enabled them to get personalized feedback on their own F&V intake levels.
The children of the intervention schools of the Schoolgruiten Project received also a piece of fruit or ready-to-eat vegetables (cherry tomatoes, baby carrots) for free during a fruit break twice a week. All children ate the piece of fruit or vegetable in their own classroom. Apart from increasing availability and accessibility, this F&V scheme was also supposed to increase the children's exposure to F&V, which in turn can influence the children's taste preferences .
Additionally, a school curriculum, developed and carefully pre-tested by the Netherlands Nutrition Center Foundation, aiming at increasing knowledge and skills related to F&V intake was offered to the intervention schools. The intervention schools were not obliged to use this curriculum, but they were strongly encouraged to do so.
Schoolgruiten Project was approved by the Netherlands Organization for Health Research and Development (ZonMw) Program for Prevention and by The World Cancer Research Fund. The Pro Children Study was approved by the medical ethical committee of the Erasmus University Medical Center. For the Pro Children Study, the parents provided written informed consent for themselves and their child. For the Schoolgruiten Project the informed consent was authorized by a legal representative (the school board).
Recruitment of the schools and study sample
For the Pro Children Study, 76 primary schools were initially sampled of which 24 schools with 735 eligible students (both intervention and control schools) agreed to participate in the total study. These 24 schools were randomly assigned to an intervention (12 schools) or a control group (12 schools).
As mentioned before, only the data of the twelve intervention schools are included in the present analyses. At the start of the study, 410 pupils were eligible for participation. Because of absence at the day of data collection or because of lack of informed consent (n = 41), 369 children actively participated at baseline. Since one of the study purposes was to investigate whether positive changes in or maintenance of high scores on the potential determinants of F&V intake were associated with positive changes or maintenance of favorable levels in F&V intake frequency later in time, only the children that had valid data on all three measurements were included. A total of 258 children were finally included for analyses.
Children with complete self-reported data on F&V intake frequency at baseline but not at first or at second follow-up were considered as dropouts. Dropout was due to children who moved to other places or schools, did not graduate to the next grade, were absent on the day of administration at first or at second follow-up (n = 87), or had missing F&V reports at one of the three measurements moments (n = 24).
The Schoolgruiten Project included 31 intervention schools. All fourth grades from primary schools in the intervention cities were eligible for participation and schools were randomly approached by phone, and invited to participate in this survey. All children (n = 693) who were present on the day of administration completed the questionnaires. Two schools were not willing to participate anymore at first follow-up, resulting in fewer children (n = 613) at first follow-up. Six schools were not willing to participate anymore at second follow-up, again reducing the number of children (n = 504) in the sample. Only the children that completed all questions on fruit intake frequency or all questions on vegetable intake at all three measurements were included in the analyses. Finally, a total of 344 children were included in the analyses.
Again, children with valid self-reported data on fruit and/or vegetable intake frequency at baseline but not at first or at second follow-up were considered as dropouts. Dropout was due to the loss of eight schools (n = 146), because children moved to other places or schools, did not graduate to the next grade, were absent on the day of administration at first or at second follow-up (n = 190), or had missing F&V reports at one of the three measurements moments (n = 13).
Since some children had missing data for some of the variables, the number differed slightly between different analyses, as indicated in the relevant tables.
In both studies schoolchildren received questionnaires, which they completed during one-school hour. Questionnaire administration was according to a written protocol. In the Schoolgruiten project, questionnaire administration was led by the teacher, while in the Pro Children Study children completed the questionnaires in the presence of a project worker.
Children of both studies received another questionnaire to take home for completion by one of their parents. Responses were treated anonymously and confidential.
All data of the Pro Children Study were entered and cleaned in the national centers according to a standardized protocol. All national data sets were pooled and further data processing and quality control was carried out in the Data Management Center at the University of Vienna (for more information on protocols and data management, as well as the questionnaires visit the website of Pro Children ).
For both projects two questionnaires were developed, a parental and a child version. In the Pro Children Study the questionnaires were pilot-tested for validity and reliability. Specific information on the development, reliability and validity of the questionnaires has been published previously [25–27]. Briefly, correlations between the frequency questionnaire answers regarding total F&V intake and the reference method (a 7-day food record) varied between 0.38 and 0.53. Correlations for total F&V intake between the test and the re-test measurement varied between 0.47 and 0.76 .
The internal consistency of the scales and the test-retest reliability and predictive validity of the behavior theory-based constructs measuring the personal, social and environmental correlates of F&V intake of the Pro Children questionnaire is measured in five European countries in 10–11-year-old children . The test-retest reliability was good to very good (intra-class correlation coefficient (ICC > 0.60) for 12 out of the 15 fruit constructs and also 12 out of the 15 vegetable constructs. Acceptable ICCs, ranging between 0.50 and 0.59, were found for the remaining constructs. Cronbach's α values were moderate to high (range 0.52 to 0.89) with the exception of the general self-efficacy scale, which had a value below 0.50 for both fruit (α = 0.42) and vegetables (α = 0.49). Spearman correlations with intake ranged between -0.16 and 0.54 for personal determinants and between 0.05 and 0.38 for environmental determinants. Compared with other studies, predictive validity can be considered moderate to good .
The child and parent questionnaires developed for the Schoolgruiten Project were based on the two Pro Children questionnaires. A more detailed description of the questions and answer alternatives of the Schoolgruiten questionnaire has been published previously .
For both studies, child reported F&V intake frequency and their potential determinants were used. For both studies, general demographic information, like information on the parent's country of birth, level of education and child's age, was retrieved from the parent's questionnaire.
F&V intake frequency
Both the Pro Children Study and the Schoolgruiten Project used food frequency questions to assess usual frequency of intake of F&V. This measure is very useful and often used in studies on correlates, predictors or determinants of food intake. This measure assesses individual usual intake frequency. More specific measures such as 24 H recall methods do not provide valid assessments of usual intake.
In the Pro Children Study, usual daily intake frequency of F&V was assessed with different food frequency questions . Frequency of fruit intake was assessed by one question: "How often do you usually eat fresh fruit". Frequency of salad and grated, other raw and cooked vegetables intake was measured separately by three questions with eight response alternatives ranging from "never" (0) to "every day, more than twice per day" (7). Mean intake in grams per day was calculated by the sum of frequency of intake of salad/grated, raw and cooked vegetables multiplied by a standard portion size (60 gram for cooked vegetables, 40 gram for salad, and 50 gram for raw vegetables ). This questionnaire is validated by a 7-day food record; the first day was a weighed record and the following 6 days were estimated records .
In the Schoolgruiten Project, we calculated average daily fruit intake (in pieces per day) as the number of days when fruit was eaten multiplied by the number of pieces eaten per day, divided by seven. We calculated average daily vegetable intake (in grams per day) as the number of days when vegetables were eaten multiplied by the amount of vegetables shown on the indicated photographs, divided by seven. The photographs were based on the validated Dutch EPIC Food Frequency Questionnaire . The pictures represent a combination of standard portions, four plates with standard portions of boiled vegetables, and four plates with standard portions of composite dishes. A composite dish contains at average a third of mixed vegetables . For calculations we combined these portion sizes by adding up these amounts and dividing by two. For calculation of raw vegetable consumption, 35 gram was considered a standard child portion size. This is half of the standard adult portion size (70 gram) according to the Netherlands Nutrition Center Foundation . Total frequency of vegetable intake was the sum of boiled and raw vegetables. A more detailed description of the questions and answer alternatives has been published elsewhere .
Potential determinants of F&V intake
In the Pro Children Study a wide range of potential personal, social and environmental determinants related to F&V intake were measured . For the present study we made a selection of these variables that were measures in the two studies, based on the literature and previous results of cross-sectional multivariate associations reported for the Pro Children Study [10, 12]. In these studies the authors concluded that liking of F&V, knowledge of recommended intake levels of F&V, self-efficacy for eating F&V, availability of F&V at home, and parental influences were the most important potential determinants of F&V intake. Therefore, we included these variables in the present study.
All these potential determinants were measured separately for both F&V intakes. All factors, except knowledge of recommended intake levels, were assessed using a 5-point Likert scale: fully disagree (-2) to fully agree (+2). To assess knowledge of recommended intake levels, children were asked on an eight-point scale how much fruit or vegetables they should eat every day. Response options ranged from "no fruit or no vegetables" (0) to "5 pieces or portions per day or more" (7). This was subsequently recoded into a dichotomous variable (less than the recommended intake levels versus the recommended intake levels or more).
Also in the Schoolgruiten Project different potential determinants of fruit intake were measured. Based on the same arguments as described before, we included the following determinants in the current study: liking and knowledge of recommended intake levels, both for F&V intakes, and accessibility and availability at home, only for fruit intake. All these determinants were assessed with questions similar to those used in the Pro Children Study .
General demographic information
For both studies, distinctions were made between children of Dutch, non-Western, and non-Dutch Western ethnicity (Europe (excluding Turkey), North America, Oceania, Indonesia or Japan), according to the definition of the Dutch Institute of Statistics . When at least one of the parents was born in a non-Western country the child was considered as of non-Western ethnicity.
Family educational level was used as a measure of socio-economic position. For both studies, parents responded to questions regarding their educational level. Educational level was treated as a categorical variable, using three categories based on the highest educational level of one of the parents (primary school or pre-vocational training = low; high school or medium level vocational training = medium; high level vocational training, college or university training = high).
Means, standard deviations and percentages were calculated to describe the key variables.
Selective dropout was assessed by logistic regression analyses with gender, parental educational level, ethnicity, region of residence of the children (only for Schoolgruiten study) (categorical variables), and intake frequency of fruit or vegetable at baseline (continuous variables) as independent variables and dropout (1 = yes, 0 = no) as the dependent variable.
As suggested by Twisk and Proper, associations between changes in potential determinants and changes in F&V intake frequency were assessed by means of multilevel multinomial logistic regression analyses . This method takes into account that change can either be increase or decrease, or no change (stable). Furthermore, it accounts for the phenomenon that children with high intake levels at baseline, are less likely to increase their intake, and are more likely to report less extreme values at follow-up (i.e. regression to the mean) . For these analyses newly constructed categorical dependent variables were created, describing change in a specific determinant. The categories were: the 'decreasers' group (= the reference group (0)), the 'stable low' (SL) group (1) and the 'stable high' (SH) group, which was merged with the increasers group (2). These two groups were merged together because both outcomes were a positive outcome. To describe the positive change or maintenance of favorable levels in fruit or vegetable intake frequency in the first and second time lapse, we used a relative measure, to overcome the phenomenon that children of this study tended to overestimate their FV intake frequency at baseline, as published previously . The phenomenon of over-reporting by younger children was also observed by Reinaerts et al. . For the relative measure, we constructed quartiles of intake frequency at all time points and analyzed whether children changed their relative position. This resulted in a dichotomous variable: 'the SL' and 'the decreasers' (0), (negative outcome) and 'the SH' and 'the increasers' (1) (positive outcome). All children who complied with the Dutch daily recommendations for fruit or vegetables intake were also assigned to the 'SH/increasers' group, because these outcomes were still a positive outcome.
A multilevel analysis was used to take into account the nested design of the study (pupils were nested within schools). Analyses were further adjusted for children's age, gender, parental education level, ethnicity, and region of residence (only for the Schoolgruiten study).
According to the aims of the present study, we performed a series of analyses (see Figure 1) assessing; A) if positive changes or maintenance of favorable levels of F&V intake frequency in the first time lapse, was associated with higher odds of having positively changed or kept high scores of the specific potential determinants in the same time lapse (association A in Figure 1); B) if positive change of maintenance of favorable levels of F&V intake frequency later in time was associated with higher odds of having positively changed or kept high scores of the specific potential determinants in the previous time lapse (association B in Figure 1); 3) if positive changes or keeping high scores of the specific potential determinants later in time were associated with higher odds of having increased F&V intake frequency in the previous time lapse (association C in Figure 1).
Associations were estimated by odds ratios (ORs), which reflect the odds for the group that increased or maintained favorable levels of F&V intake frequency of being in the specific category (for change in determinant) compared to being in the reference category (= decreasers group). When cells for the multinomial logistic regression analyses include a small number, no reliable ORs can be estimated. Five percent of the total sample or less was considered as a small number and in that case three categories were merged into two categories to solve this problem.
The data analyses were performed using SPSS 11.0 (SPSS Inc., Chicago, IL, USA, 1999). The multi-level analyses were conducted using MLwiN software (Version 2.01) . The significance level was set at p < 0.05.