Introduction

Adequate levels of physical activity (PA) and sedentary time reduces risk of non-communicable diseases, such as diabetes1, and enhances mental health2. PA levels have declined over the past 22 years, especially in adolescents3. Indeed, a large proportion of adolescents, especially girls, are far from reaching the PA guidelines, and activity levels decline further with age3,4. Schools are considered key settings for promoting child and adolescent PA and a wealth of initiatives have been instigated to date 5, yet lack of PA remains a threat to public health globally, as only a minor proportion of effective interventions are successfully implemented and sustained6.

In education outside the classroom (EOtC), teachers facilitate educational activities outside the school buildings during school hours on a regular basis. EOtC takes advantage of environments and places outside the school buildings, e.g., green spaces, societal institutions, or companies, to encourage teaching and learning activities where the body and senses are used in authentic situations to promote learning and wellbeing in agreement with curricular obligations. EOtC is characterised by a concurrent focus on place and pedagogy, which aims towards action-centred, collaborative, inquiry-based, and thematic learning processes7. Ideally, the pedagogy of EOtC provides pupil autonomy and an experience of competency in a close bond with teachers and peers, in agreement with Self-Determination Theory’s (SDT) description of how human wellbeing and motivation is supported8. EOtC commonly involves active transportation and learning activities that require pupils to move, such as measuring the volume of trees in mathematics, exploring local environments on foot in social studies to investigate how cites develop over time, immersing themselves in the everyday life of an Iron Age village using dramatisation in history lessons, or practicing school gardening in home economics9,10. Consequently, there is a close alignment between EOtC as a PA promoting initiative and the ‘core business’ of schools, i.e., learning and wellbeing. This promotes teacher acceptability and, in turn, the efficiency and longevity of PA promotion9. In Denmark, the practice of EOtC has increased noticeably in the past two decades from a few schools and teachers using the approach at the turn of the millennium to almost one fifth of Danish public and independent schools as shown in year 2014 and 2019 surveys11,12.

Recent evidence suggests that EOtC is an educational approach with potential to simultaneously improve pupils’ PA13,14 as well as school motivation15, social wellbeing16,17, and academic achevement18. Thus, EOtC have been characterised as an intersectoral19 and integrated ‘win–win’ opportunity for both health and education sectors, i.e., an ‘add-in’-approach to promotion of a healthy active school day9,20. To date, however, the causal evidence for the effects of EOtC on PA is scarce. Previous quasi-experimental evidence suggested that EOtC was positively associated with PA in children in grade 3–6 (ages 9–12 years); for boys, moderate-to-vigorous PA (MVPA) was higher on a weekly basis, and light-intensity PA (LPA) was higher among boys and girls on days with EOtC compared to normal schooldays without physical education lessons13,14. Furthermore, results also indicated that school days with EOtC in green settings were associated with more LPA for boys and girls compared to other school days with EOtC in non-green settings21. However, methodological limitations in the evidence base have been identified which calls for more rigorous studies10.

The objectives of the MOVEOUT study were to investigate the effects of an EOtC intervention on grade 4–10 pupils’ (aged 10–16 years) school-based and overall PA behaviours (see the following section for a description of the intervention). School-based PA comprised all hours that the pupils were in school during a five-day school week. Higher school-based PA contributes to overall PA and might support school-based learning via enhanced wellbeing and cognitive capabilities2. However, enhanced school-based PA might be accompanied by compensatory behaviours after school, i.e., less leisure time activity22,23. We therefore also included overall PA behaviours which refer to PA taking place during all waking hours across the entire week, including the weekend. This timeframe has primary interest for public health because ultimate health benefits will be derived from higher absolute levels of PA across both school and leisure time. Furthermore, if only school-based PA, and not overall PA behaviours, is affected by the intervention, the results would suggest a need to develop supplementing activities to avoid compensatory behaviours and thereby increase overall PA. The pupils’ overall and school-based PA behaviours are categorised, as traditionally done in PA research, according to time spent in different PA intensity domains (i.e., sedentary (SED), light (LPA), and moderate-to-vigorous (MVPA)), hereon after termed PA intensities. PA behaviours are also categorised by different types of everyday PA behaviours (i.e., running, walking, standing, and sitting) because these are more easily interpreted and translated to practice. These are hereon after termed PA types.

The MOVEOUT study also investigates effects on school-motivation, wellbeing, and academic achievement. Using a mix of observational and quantitative methods, the MOVEOUT study also investigates mechanisms by which EOtC impacts PA behaviours, school-motivation, wellbeing, and academic achievement. The aim for us is to identify components of EOtC implementation that results in improvements on one or more of the mentioned outcomes. However, in the context of this registered report, we focus on effects of the EOtC intervention on PA behaviours.

The overall research question of this registered report is:

What are the effects of the EOtC intervention on pupils’ school-based and overall PA behaviours over the course of one school year?

All research questions, hypotheses, and associated analyses are summarised in Table 1. For both school-based and overall PA behaviours, the first group of hypotheses is:

  1. (1)

    Pupils in the intervention group will retain higher PA intensities compared to the control group, specifically 1a) more MVPA, 1b) more LPA, 1c) less SED time.

Each of these hypotheses are based on previous quasi-experimental evidence that suggests that EOtC associates with enhanced MVPA, LPA, and reduced SED time. The associated null hypothesis is that the opposite or no effect is observed. For each PA behaviour, the difference in the so-called compositional product will be estimated (see the methods section for further description).

For both school-based and overall PA behaviours, the second group of hypotheses was:

  1. (2)

    Pupils in the intervention group will retain 2a) more running, 2b) more walking, 2c) more standing, and 2d) less sitting time compared to the control group.

There is no previous evidence that has investigated associations between EOtC and PA types. However, a key tenet of EOtC is that learning activities require pupils to move and we therefore expected that pupils will retain more running, walking, and standing, and less sitting compared to the control group. The associated null hypothesis is that the opposite or no effect is observed. The difference in the compositional product which is attributable to the intervention versus the control will be used to assess intervention effects.

Table 1 Design table.
Full size table

The MOVEOUT study builds on insights from previous EOtC research, in particular the TEACHOUT study24. In TEACHOUT, an EOtC intervention was developed, investigated, and manualised25 in Danish public schools (grade 3–6). In Denmark, as other Scandinavian countries, primary/elementary education is compulsory and for most children take place in publicly funded (municipal) schools with mixed gender classes from grade 0 (approximately 6 years) to 10th grade (approximately age 15). The EOtC intervention studied in the MOVEOUT study is based on the TEACHOUT intervention and the experiences and findings from this quasi-experimental study.

The EOtC intervention25 consisted of a two-day training course for teachers followed by the teachers applying EOtC for one school year in one or two weekly sessions during school hours, with a total weekly duration of at least five hours. The course consisted of a combination of talks, workshops, and plenum discussions as well as time for planning of local implementation. The teachers received a one-time remuneration of 1500 Danish kroner in return for their participation. As an adaptation from the previous applications of the EOtC intervention, the two-day training course was followed up by two one-hour webinars to support implementation. The webinars were placed in the first and second half of the school year. Each EOtC session could be conducted by one or more EOtC-trained teachers simultaneously, in one or a combination of multiple school subjects at various places and settings outside the school building. The intervention’s Theory of Change was well-established, and the components have previously been thoroughly described24. See Fig. 1 for the intervention Theory of Change.

Fig. 1
figure 1

Theory of Change of the EOtC intervention. The figure outlines the intervention inputs (here, referred to as intervention education components), activities, output, and outcomes (both primary and secondary outcomes, as well as long-term impact).

Full size image

Methods

Ethics information

The MOVEOUT study was assessed by the Regional Committee on Health Research Ethics, Capital Region (ID: 21,062,006) and deemed not eligible to undergo ethics review (documentation available on the Open Science Framework: https://osf.io/89pmf/?view_only=298a0d461e594bc3b2d73ace8d107882). By Danish law, only research projects of biomedical character or projects that involve risks for participants need to have their ethics reviewed by a Regional Ethics Board. All other research projects can not apply for formal ethical approval. The MOVEOUT study was assessed by University of Copenhagen internal ethics committee, The Research Ethics Committee for SCIENCE and SUND (ID 504–0292/21–5000), which concluded that the project is compliant with relevant Danish and international standards and guidelines for research ethics (documentation available on the Open Science Framework (see link above)). The MOVEOUT study was preregistered at clinicaltrials.gov (NCT05237674).

Parents or legal guardians of the involved pupils, and the teachers engaged in the project, were given written information about the study before decision on participation was made. All children received age-appropriate oral information about the study and were asked to provide their consent to participate in the study, if their parents or legal guardians provided a written consent allowing them to participate.

All data and other personal information collected during the study are processed in accordance with the Danish Data Protection Act. Raw data in digital format are pseudo-anonymised using ID codes and stored in a secure project folder. Keys that link ID codes with person identifiable data are stored in a separate secure project folder. Data can only be accessed by persons who are affiliated with the study and have a duty of confidentiality. Data used in the analysis are pseudo-anonymised and only reported on a group level.

Design

This study is a between-subject, cluster-randomised trial. Participating schools were allocated to either the intervention or a waitlist group after completion of pre intervention data collection in April 2022. Fifteen of thirty Danish public schools with one or more classes in grade 4–10 (pupils aged 10–16 years) were randomly assigned to the intervention (see Fig. 2 for illustration of the study design, and the overall MOVEOUT study protocol26 for this study’s integration in the overarching MOVEOUT investigation of various outcomes of EOtC). To ensure that the intervention and control groups have a balanced representation of grade levels, stratified block randomisation in a 1:1 ratio was used (i.e., schools with participating classes in grade 4–6 and schools with classes 7–10 were randomised block wise). The remaining 15 schools functioned as waitlist controls with teachers continuing their teaching as usual (see Fig. 1 for design and timeline). Teachers in the waitlisted control schools received the two-day training course on EOtC upon finalisation of the intervention and the post measurements. Data collection was not performed blind to the conditions of the experiments.

Fig. 2
figure 2

Study design and timeline. The figure shows the timing of study milestones including measurements and implementation monitoring (in green), randomisation, training courses for both intervention and waitlist control groups (in orange), and the intervention period, 2022–2023.

Full size image

Sampling plan

The number of participating pupils were ~ 1134 pupils (age 10–16 years) in 54 classes, grades 4-10th (1.8 classes per school, mean pupils in class = 21) from 30 Danish schools. The size of the sample reflects what is practically possible in terms of recruitment and funding. Based on the TEACHOUT study24, three classes were estimated to drop-out during the intervention. In each of the included classes, the estimated drop-out and rate of invalid accelerometer data were expected to reduce the sample by approximately 20%, corresponding to four pupils per class. The analytical sample was therefore expected to be ~ 30 schools, ~ 51 classes, ~ 867 pupils in total. Power calculations showed that with intraclass correlations (ICC) calculated from a similar dataset, from the TEACHOUT study, values of 0.046 within classes and 0.035 within schools and a power of 95%, one tailed test, this analytical sample allowed for an identification of a small to medium effect size with an analysis that accounts the structure of the data (e.g., linear mixed models) (Cohen’s d = 0.374 95% CI [0.187,0.561]). This corresponds to small to medium effect sizes observed in comparable research13,14, and was thus expected to meet the needs of our main analyses. Supplementary Figure S1 online shows how the minimum detectable effect size depends on how many schools are recruited and included in the analysis given three commonly used levels of statistical power. The power calculations were done with the function mdes.cra3r3 from the R package PowerUpR. This function was used to calculate minimum detectable effect-size in a three-level cluster-randomised trial. The complete R-code for the power calculation as well as creation of Supplementary Figure S3 can be found in the MOVEOUT study’s Open Science Framework folder (https://osf.io/jcrvh/).

The aim was to purposely sample 30 schools among schools in municipals of the Capital Region, Region Zealand and Southern Region of Denmark. In order to study whether the efficacy of the intervention is dependent on pupil’s socioeconomic background efforts were made to recruit and include schools located in municipalities characterised by higher than region average proportions of citizens with lower levels of education. Teachers were recruited by contacting municipalities and school heads via e-mail, and subsequently telephone conversation and/or online meetings with schools’ heads and/or teachers.

Inclusion criteria.

  • Non-special need Danish schools and school classes grade 4–10.

  • Classes not involved in other school development or research projects.

  • Pupils for whom parents or legal guardians provided written informed consent.

  • Pupils with valid data from at least one accelerometer pre intervention measurement.

Exclusion criteria

  • Classes not able to adhere to the intervention or control group conditions (i.e., intervention group classes: a school year average of > 150 min of weekly EOtC; control group: a schoolyear average of <  = 150 min of weekly EOtC) were excluded from per-protocol analysis but included in the intention-to-treat analysis.

  • Pupils with significant health problems, reported by their parents or legal guardian in the background questionnaire, or as judged by the investigators or teachers were excluded from the analysis.

Measures and procedures

Data on PA behaviours were obtained across four measurement periods of seven days each. The investigators collected data two times during winter and summer 2022 (pre) and two times during winter and summer 2023 (post). Data were collected in both seasonal periods to account for influences of the weather on PA behaviours. Each time, PA behaviour was assessed with accelerometry. Each pupil was be equipped with one Axivity® AX3 accelerometer (23 × 32.5 × 7.6 mm) which was affixed with skin tape on the left thigh. Two trained investigators (one male and one female) handled the affixation and removal of the devices during school hours in a separate classroom in groups of four. The male investigator was assigned to male pupils and the female investigator was assigned to female pupils. The pupils were instructed to wear the accelerometer for seven consecutive days including nights and water activities.

Data on socio-economic status (SES) were be collected from the parent or legal guardian pre intervention using the Danish Occupational Social Class Measurement27. Background data on child health status were collected from the parent or legal guardian pre intervention.

During the intervention period (the school year August 2022 to June 2023), the participating teacher in both the intervention group and control group answered a weekly SMS questionnaire that measures degree and characteristics of the EOtC implementation (and any non-intended EOtC practice in the control group) in terms of amount, place, mode of transportation, and school subjects taught during the EOtC session. The questionnaire is based upon the existing and validated EOtC implementation monitoring tool from the TEACHOUT study24.

Analysis plan

Pupils could not be blinded from group allocation. To eliminate researcher expectation biases from the analyses, allocation levels was coded before analyses and analysts was blinded from this coding. The analysts was not informed about the codes before the final modelling and analyses have been performed.

Processing accelerometer data

All accelerometery data were be processed in Matlab (Version 9.9.0 R2020b, Mathworks Inc., Natick, Massachusetts, US) which included resampling, generating ActiGraph counts28, identification of non-wear and summarising the pupils’ time spent in different PA intensity domains as well as time spent sitting, standing, walking, and running. Non-wear periods were identified from both acceleration and temperature data using the method described by Rasmussen et al.29. All periods identified as not worn were marked as missing data and not excluded from the subsequent analysis.

PA intensity was determined using the algorithm proposed by Brønd et al.30. Moderate and vigorous PA time was estimated using 10-s epochs accounting for the elevated post oxygen consumption during intermittent PA using the second-by-second epoch data to improve the assessment of vigorous activity31. The cut-points defining the intensity domains were determined using an internal calibration study conducted in a group of children at the age of nine to 13 years. The cut-points were determined using the following steps. The moderate intensity was determined as the average counts for walking at self-selected speed. The energy expenditure during walking at self-selected speed was set by stature and the intensity corresponds to 30–35% of the children maximal aerobic capacity32. The vigorous intensity threshold was determined as the counts level for which 95% of the intensities measured during running was included. Running is considered a vigorous activity and performed at > 60% of the children’s aerobic capacity. The thigh cut point was 4822 for moderate and 9143 for vigorous and 100 counts for the below 1.5 METs sedentary behaviour.

The PA types were determined in 1-s epochs from the raw acceleration using a simple decision-tree algorithm30. This method has been validated with a standardised protocol and provide a sensitivity and specificity > 99.3% for sitting and standing and > 85.8% for walking and running activities with children in the same age range as included in this study. No more than two hours of non-wear were allowed for a measured day to be valid.

All PA outcomes were averaged across the week requiring at least three valid schooldays and one valid weekend day and adjusted using a 5/7 weight for schooldays and 2/7 weight for weekend-days. The school hours was intended to be identified in the complete PA recordings using the individual school schedules, provided by the schools in digital format.

Compositional analyses

Since the outcomes in this study are time spent engaging in PA behaviours, which sum up to the total amount of time engaging in the PA behaviours, there cannot be a change in the amount of time spend in one behaviour without a ‘compensatory’ change of time used in another33. The important information is the relative, and not the absolute, amount of time spent engaging in the PA behaviours, calculated as the percentage of time spent engaging in school-based or overall PA behaviours. The collinearity of the time spent engaging in PA behaviours means that traditional multivariate models are inappropriate when the goal is to take the compositional nature of PA data into account. To accommodate this, compositional analyses are used34,35.

Descriptive analyses

Pre and post intervention PA is shown for intervention and control group as the compositional mean (sample center) and the variation matrix. The compositional mean describes the mean adjusting for compositional nature of the data, i.e. that the time spent engaging in each PA behaviour add up to the same total. For the two timeframes 840 min (14 h; waking minutes a day) and 420 min (7 h; a school day) respectively was intended to be used. Descriptive differences between the groups and between pre and post intervention PA behaviours are illustrated using geometric mean bar-plots. Biplots are not presented as they are not suitable for illustration of differences between groups34.

Missing data

To create a full analysis data set for analysis, missing data for the outcome were imputed using multiple imputation (MI) with 1000 samples. The model for MI imputed missing values of the isometric log-ratio transformed variables (see below). Some pupils did only have measurements for one of the two measurement periods at pre-test and/or at post-test (see Table 4). To use all available information about PA, the MI-model included variables for the logarithm to time spent engaging in each PA behaviour in either of the two measurement periods. Other independent variables included in the model (without missing values) were age and gender of the pupils. MI was handled separately for intervention/control as this is found to have greater robustness when the analysis model overlooks an interaction effect involving the randomised group36. Due to the design, the model for MI included information about classes and schools in a multilevel joint modelling multiple imputation. The R-package jomo was be used for MI.

Analysis of intervention effects

Estimation of intervention effect on both total-time and school-time were calculated using a ‘per-protocol’37 analysis, where all classes of the intervention group schools not adhering to the definition of ‘regular EOtC’ (a school year average of > 150 min of weekly EOtC) and all classes of the control group schools practicing too much EOtC (average of > 150 min of weekly EOtC) were excluded. These analyses were supplemented with a sensitivity analysis, following the principle of ‘intention-to-treat’37 where all schools including all pupils were included in the analyses regardless of their adherence to the intervention or control condition.

The PA behaviours were transformed into isometric log-ratio (ilr) coordinates to take the compositional nature of the PA behaviour data into account. This was be done using the function ‘acomp’ in the R package ‘Compositions’.

Inspired by the approach described by Martín-Fernández et al.34, to test if there was an overall effect of the intervention on PA behaviour, a dataset with the pre-test ilr-coordinates as one variable, the post-test ilr-coordinates as one variable and another variable (‘ilr-group’) containing information of what ilr-coordinate the data was coming from, was created. A linear mixed model was conducted with the ilr-coordinates for post-test as outcome and intervention/control per-protocol as a covariate adjusting for pre intervention ilr-coordinates, sex, and age as well as adjusting for the random effect of the ‘ilr-group’, schools, and classes. If the estimate of intervention/control per-protocol is significant in this model, it will be interpreted as an overall effect of the intervention on PA behaviour in general.

To investigate the effect of the intervention on PA intensities and PA types, a series of linear mixed models was conducted with each of the post-test ilr-coordinates as outcome and intervention/control as covariate while adjusting for the pre intervention ilr-coordinates. These models was also adjusted for sex and age as well as the random effect of schools and classes.

Using the emmeans package in R a table with means for the ilr-coordinates adjusted for pre intervention, sex, and age for both intervention and control was created. This table was backtransformed with the ilrInv function in ‘compositions’ providing a table of interpretable means in minutes/day. Based on these means, log-ratio differences between the intervention and the control group were calculated. To estimate the uncertainty of these log-ratio differences, bootstrap 95% confidence intervals were calculated and if theses intervals overlap zero, they were considered uncertain/insignificant.

Sensitivity analyses

A number of sensitivity analyses were be carried out to account for the risk of biases related to potentially high proportions of missing data in school-based trials. Analysis on complete data using 1) listwise deletion as well as 2) models where the intervention variable were included following the principle of ‘intention-to-treat’ (ITT) as opposed to ‘per-protocol’ (PP) was carried out as sensitivity analyses37. Besides these changes, the analyses was carried out with the same statical plan as indicated above. The analyses followed the same hypothesis as the main analyses (see Table 1). If there was no conflict between main and sensitivity analyses, we were confident that no attrition was biasing main analyses. If either of these analyses conflicted with the main analyses, this might suggest that attrition could be biasing the main analyses. Results that conflicted with the main analyses served to monitor the strength of the evidence of the main analyses, not interpreted unequivocally against the alternative hypothesis.

As final sensitivity analyses, models which do not take the compositional nature of PA data into account were carried out, and where time spent engaging in each of the PA behaviours was treated as dependent variables individually. These sensitivity analyses were conducted to check the robustness of the rather complex approach of the main analysis. In these analyses a series of linear mixed models were carried out using the raw seconds of time spent in each activity post intervention, adjusting for pre intervention values, sex, and age as well as the random effect of schools and classes. The decision to take the compositional nature of data into account in the main analyses is a debated topic in the scientific literature on objectively measured PA38. Because PA behaviours are used as outcomes, and not as exposures, the use of compositional data analysis is novel which means that it is useful to check the robustness of the produced results against more commonly practice used method. The analyses followed the same hypothesis as the main analyses (see Table 1). If there were no conflict between main and sensitivity analyses, we were confident that no statistical bias was affecting main analyses. Results that conflict with the main analyses served to monitor the strength of the evidence rather than be interpreted unequivocally against the alternative hypothesis. Other advantages of conducting these sensitivity analyses were that the results are simpler and therefore provide a more intuitive interpretation, and that they enabled comparison of the results to other studies.

Stage 2 alterations

The randomisation was performed between the first and the second pre-test and not after the second pre-test as indicated in Fig. 2. The timing of the randomisation was changed to ensure that the schools had sufficient time to plan their intervention participation.

The timeframe of the school week was altered to a shorter timeframe. It was set to 330 min (8:30 to 14:00 h), instead of the planed 420 min. This alteration was implemented due to inconsistencies in the provision of school schedules, and the need to ensure certainty that morning and afternoon transportation was not included in the school-based timeframe.

Results

Participants

Recruitment and enrolment commenced approximately two years prior to the intervention’s implementation and persisted until randomization (April 1st, 2022)26. Initially, schools across 42 municipalities in the Capital Region, Region Zealand, and Region of Southern Denmark were invited to participate, leading to the enrolment of 19 schools. Subsequently, invitations were disseminated nationwide through teacher networks, resulting in the enrolment of another six schools. A total of N = 983 pupils across 44 classes within the 25 enrolled schools were invited. Of those, n = 724 were given parental/guardian consent to participate. During the intervention, nine classes (n = 125 pupils) withdrew. Additionally, n = 96 pupils were excluded due to ineligibility reasons. Hence, n = 503 pupils from 35 classes were included in the ITT analysis. Of these, 11 classes (n = 156 pupils) did not comply with the protocol. Hence, 24 classes (n = 347 pupils) were included in the PP analysis. See Fig. 3 for the participants flowchart.

Fig. 3
figure 3

Participants flowchart. The figure shows the flow of schools, classes, and pupils from enrolment to the inclusion in intention-to-treat and per-protocol analysis.

Full size image

The pupils in the PP intervention group had a mean age of 12.3 years (SD, 1.1 years) with 48.3% being female, while those in the PP control group had a mean age of 12.4 years (SD, 1.3 years) with 56.0% being female. See Table 2 for descriptive statistics of the participants.

Table 2 Participants descriptive.
Full size table

Of the pupils enrolled in the PP analysis, the teachers assigned to the intervention had a mean work experience as teachers of 18.4 years (SD, 8.6 years), and a mean use of EOtC of 2.5 years (SD, 3.3 years). For the control group teachers, the mean work experience was 16.1 years (SD, 10.1 years), and a mean use of EOtC of 1.7 years (SD, 3.5 years) (see Supplementary Table S1 online for descriptors of enrolled class teachers). Among the teachers assigned to EOtC, 7.7% had previously participated in EOtC-relevant training, such as EOtC coordinator or friluftsliv, nature, or culture interpreter.

Intervention implementation

The two-day EOtC training course was conducted as intended, ultimo April 2022. Of the enrolled teachers in the intervention group, 85.7% participated in the course, with all classes represented by at least one teacher, with a mean of 1.5 teachers (SD, 0.5 teachers) from each enrolled class.

The use of EOtC took place in both intervention and control groups (see Supplementary Table S2 online). For the classes in the PP analysis, EOtC was implemented with a weekly mean of 238 min (SD, 50 min) and with a mean frequency of 1.17 sessions (SD, 0.37 sessions) per week, compared to the control group who had a mean weekly level of implementation of 77 min (SD, 34 min) of EOtC and 0.49 EOtC sessions (SD, 0.21 sessions). In general, teachers applied EOtC sessions to various subject domains of language arts, science technology and math, cultural studies, arts, and physical education. Evaluation of the implementation of EOtC and non-intended use of EOtC at PP was based on 91.4% (SD, 15.5 percentage points) of possible weekly reports in the intervention group, and 68.6% (SD, 40.3 percentage points) in the control group.

PA data

The sample that fulfilled the criteria for inclusion in the analysis by having valid data from at least one accelerometer pre-intervention measurement, i.e. one full day (24 h), comprised of 503 pupils (see Fig. 3).

At pre-test of the n = 247 pupils that was included in the PP analysis, n = 337 pupils met the school day analysis requirement (three valid school days), n = 331 met the full day analysis requirement (three valid weekdays and one valid weekend day). At post-test, n = 282 met the school day analysis requirement and n = 253 met the full day requirement. For full information on pupils meeting measurement requirements, see Supplementary Table S3 online. For compositional mean at pre- og post-test for intervention and control groups at PP, see Table 3.

Table 3 Compositional Mean at Pre- og Post-Test for Intervention and Control at Per-Protocol.
Full size table

The goal of having exactly one year (365 days) between measurement time points was considered achieved. The actual difference between the PA measurement periods of pre-test 1 and post-test 1 was on average 369.2 days (SD, 7.1 days) for the intervention group and 388.8 days (SD, 26.0 days) for the control group. The actual difference between pre-test 2 and post-test 2 was, on average, 365.9 days (SD, 2.2 days) for the intervention group and 375.6 days (SD, 15.7 days) for the control group.

The variation matrix indicated that both intensities (see Supplementary Table S4 online) and PA types (Supplementary Table S5 online) were, in general, similarly correlated across the intervention group and the control group as well as for the two different time periods. Generally, larger variability between the PA types were identified compared to the intensities. The largest variability between intensities were observed between MVPA and sedentary time in the control group at pre- og post-test with a variance of 0.26 and for PA types the largest variance were observed between sitting and running (variance = 1.44) at post-test.

Intervention effects

The two Linear Mixed Models (LMM) investigating the overall effect of the intervention on PA Confirmed that the intervention had an overall effect in the school day period (p < 0.001) but did not confirm a significant overall effect in the full day period (p = 0.091) (see Supplementary Table S6 online). Because of the complexity of the LMM used to calculate an overall effect, we encountered issues calculating degrees of freedom (df) when running the analyses. To solve these issues, we reduced the complexity by removing the random effect of the variable ‘ilrgroup’.

Geometric mean barplot showing the time spent in PA intensities (see Supplementary Fig. S2) and PA types (see Supplementary Fig. S3) were produced, the bars represent the logarithm to the ratio between the geometric mean in the intervention and control groups and the geometric mean in the whole sample. A ratio of 0 indicates that there is no difference between the geometric mean in the specific group and in the whole sample. Positive values indicate that the groups geometric mean is larger than in the whole sample and negative the opposite. Based on the logarithmic ratios shown in the figures, we can determine the actual comparison of the average activity levels of a specific group to the overall average. For instance, Fig. S2 shows a value of 0.02 for time spent in MVPA for the intervention group at post-test during the school day. This indicates that the pupils in the intervention group, on average, spend 2% more time in MVPA at school days compared to the whole sample at post-test. Overall, the differences between the intervention and control groups were quite small. However, the tendency was that the differences were larger during the intervention period (post-test) and that the differences were larger during the school week compared to during the full week.

School week

During the school week, the intervention group exhibited lower levels of Sedentary time compared to the control group, with an estimated marginal means (EMM) of 184.66 min per day (CI: 178.64–190.83) versus 198.74 min per day (CI: 193.59–204.05), resulting in a logarithmic difference (see Table 4 and Fig. 4) of − 0.074 (CI: − 0.109 to − 0.039). Conversely, LPA and MVPA were higher in the intervention group, with LPA showing an EMM of 123.24 min (CI: 118.14–128.26) against 112.91 min (CI: 108.65–117.04) in the control group (LogRatio: 0.087, CI: 0.042–0.134). MVPA was also significantly higher in the intervention group, registering 22.10 min per day (CI: 20.34–23.91) compared to 18.34 min per day (CI: 16.84–19.86) in the control group (LogRatio: 0.186, CI: 0.080–0.295).

Table 4 Difference between the intervention and control group at per-protocol.
Full size table
Fig. 4
figure 4

Log Ratio difference between intervention and control groups during a school week and during a full week per-protocol.

Full size image

With regards to the PA types, sitting, standing, moving, walking and running also showed significant differences. The intervention group was more active, with increases in the time spent standing (EMM: 50.70 vs. 44.30 min, LogRatio: 0.134, CI: 0.017–0.251), running (EMM: 2.74 vs. 2.14 min, LogRatio: 0.247, CI: 0.067–0.429), and walking (EMM: 57.98 vs. 50.07 min, LogRatio: 0.147, CI: 0.063–0.231) as well as less time sitting (EMM: 205.01 vs. 221.40 min, LogRatio: -0.074, CI: − 0.109 to − 0.039).

Full week

With regards to the full week, the differences between the intervention and control groups were minimal, suggesting that the EOtC intervention may not have significantly influenced daily activity patterns (see Table 4 and Fig. 4). For sedentary behavior, the intervention group showed an EMM of 509.15 min per day (CI: 499.15–519.12), compared to 518.11 min per day (CI: 509.22–527.01) in the control group. The logarithmic difference of -0.017 (CI: − 0.040–0.005) highlights that the CIs overlap zero, suggesting no statistically significant difference in sedentary behavior between the groups.

Similarly, LPA and MVPA showed minimal differences. LPA in the intervention group was 261.97 min (CI: 254.19–269.82) versus 257.79 min (CI: 251.07–264.51) in the control group, with a logarithmic difference of 0.016 (CI: − 0.018–0.050). MVPA exhibited a logarithmic difference of 0.072 (CI: − 0.006–0.151), with the intervention group engaging in 68.88 min (CI: 65.00–72.91) versus 64.10 min (CI: 60.27–67.99) in the control group. The overlapping CIs for both LPA and MVPA suggest that any differences observed are not statistically significant.

With regards to the PA types, sitting, standing, moving, running, and walking also presented overlapping confidence intervals, further underscoring the lack of statistically significant differences between the two groups for these activities over the full week. For example, the reduction in sitting time (EMM: 525.81 vs. 537.93 min, LogRatio: − 0.023, CI: − 0.051–0.005) and the increase in walking (EMM: 116.64 vs. 110.79 min, LogRatio: 0.051, CI: − 0.017–0.120) both involve CIs that include zero, indicating that the results could be attributed to chance variations.

Sensitivity analyses

Sensitivity analyses involved calculating group differences with an intention-to-treat method rather than per-protocol (for all ITT analysis outcomes, see Supplementary Table S7, Supplementary Table S8, Supplementary Table S9, Supplementary Table S10, Supplementary Fig. S4 and Supplementary Fig. S5 online), analyzing complete datasets instead of using multiple imputation (see Supplementary Table S11 and Supplementary Fig. S6 online), and running a series of LMM’s without treating the PA variables as compositional (see Supplementary Table S12). The results from these sensitivity analyses were generally consistent with those from the primary analyses, suggesting that neither attrition bias nor statistical bias significantly influenced the overall results.

Discussion

This randomized controlled trial of the TEACHOUT EOtC intervention showed that the intervention positively affected pupils’ school-based PA, whereas the impact on total amounts of PA over the week was not significant.

Implementing the EOtC intervention (and having > 150 min of EOtC per week) was associated with 20.4% more (LogRatio = 0.186) MVPA during school-time, when adjusting for time spent pre-test, sex and age as well as the clustered nature of data. For LPA, the difference was 9.1% more time spent (LogRatio = 0.087) and for sedentary time the difference was 7.1% less time spent (LogRatio = − 0.074). ITT analysis supported these effects of EOtC on school-time PA and so did the analysis of different types of movement behavior showing that participating in the EOtC intervention was associated with more of school-time spend running and walking and standing and less time sitting. Furthermore, the observed positive effects of the EOtC intervention on reducing sedentary behavior correspond with the findings of decreased sitting time.

However, these effects on school-based PA were not sufficiently large or consistent enough to significantly affect the pupils’ total amounts of PA over the week. The largest difference between EOtC and control group’s total amounts of PA was for time spend in MVPA, where the difference was 7.5% (LogRatio = 0.072), but the confidence intervals included negative values (− 0.006; 0.151) indicating that this estimate was not statistically significant. It is difficult to ascertain, whether this low and insignificant effect on total time PA is due to insufficient power in our sample size, due to a compensation effect or due to children’s total amount of PA over the week being dependent on so many other factors outside school that an average exposure of EOtC per week in the intervention group were too small a contribution to significantly affect the overall total amount of PA over a week.

The findings of this study makes a substantial contribution to the field by, with a stronger scientific design, adding to the findings in the previous quasi-experimental TEACHOUT EOtC study on children in grade 3–6 (ages 9–12 years) which found that participating in EOtC was associated with higher weekly amounts of MVPA13 and that the amounts of LPA was higher on days with EOtC compared to normal schooldays without physical education lessons14.

As the pupils involved in the study were 9–14 of age the findings must be considered in the context of recent meta-analysis and reviews of school-based PA promotion interventions for children and adolescents. Two recent reviews have shown no or weak overall effect on MVPA and inconclusive evidence on sedentary time among children39,40, whereas there is more support for the effectiveness for adolescents41.

The findings should also be considered in relation to the one-year implementation period of the intervention, and with the intervention group averaging 238 min weekly of EOtC compared to 77 min weekly in the control group. Many school based PA promoting interventions run for shorter time39. Maintaining a consistent high amount of EOtC every week over a whole school year might be difficult due to changing conditions such as weather, new extra tasks for the teachers, and staff/teachers turn over42. At the same time, EOtC is becoming a common teaching practice in Danish schools and occurs in small sporadic amounts in many school classes without an intervention from the outside12. This was indeed seen in the control schools in this study which may have contributed to smaller differences to the intervention classes. Similarly, more daily PA has been implemented in Danish schools following the Danish school reform in 201443, which may have resulted in fewer opportunities to implement additional PA initiatives.

As mentioned, the largest effect of the intervention was on school-time MVPA whereas the effects on LPA and sedentary behavior were smaller. This corresponds well with the finding that the type of movement behavior that were mainly affected by the intervention was running while the EOtC effects on time spend walking, standing, moving and sitting were smaller. This finding is both surprising and encouraging. We had anticipated that the largest effects would be on lower-intensity PA types, specifically an increase in time spend walking and a decrease in time spent sitting. This assumption was based on the fact that, in many cases, the necessary transportation to and from EOtC locations outside school premises inherently involves walking time, that replaces less active time during indoor classroom lessons21. Studies have showed that recess at school is associated with high levels of MVPA for ten-year-old Danish school pupils44. However, EOtC in outdoor environments settings may often offer more physical space and opportunities for breaks, that encourage running, which has been argued to be a central explanation for previously found associations between EOtC and time spent in MVPA45,46. From a health promotion perspective, it is encouraging, that the movement-behavior that was mainly affected by the EOtC intervention was running because running, as a form of vigorous PA, is considered an important aspect of children’s health47. However, as this study represents the first investigation into the effects of EOtC on various types of PA, further research is necessary to validate these findings.

Practical implications

While the findings of this study supports that the EOtC intervention had positive effects on school-based PA, it could not demonstrate a statistically significant effect on overall PA behaviors. To fully consider the EOtC intervention as an effective way to promote children’s general amount of PA, the intervention could perhaps benefit from supplementing it with other PA promotion activities during school, e.g., more time for recess or physical education lessons, PA activities in after-school programs, leisure time, or encouraging active transportation to avoid any compensatory behaviours and thereby increase overall PA40. The findings could also suggest that further adjustments are needed to refine the intervention activities to increase PA levels such as e.g. combine PA with academic content that varies the teaching and serves as a driving force for learning, such as play and games48, or to consider increased exposure of amount of EOtC to make sure that the impact is large enough to also influence the total amount of PA over the week. Optimizing implementation strategies might be needed to achieve a higher sustained dose of EOtC.

The value of implementing EOtC as a way to increase PA must also be considered in connection with other potential important outcomes, such as pupils’ learning, school motivation, and wellbeing. Not the least because such school-relevant benefits increase the relevance for educational institutions to implement the intervention. Previous non-randomized experimental studies, on the TEACHOUT EOtC intervention have not demonstrated either worsened or significantly improved reading competence18 or math skills49, but did indicate improvements in pupils’ motivation for school15 and some aspects of psychosocial wellbeing16,17. Studies on similar EOtC programs specifically highlight school wellbeing as a significant outcome10,50, yet the evidence remains inconclusive51,52. These impacts should and will be investigated further as part of the MOVEOUT study, which also includes an examination of the mechanisms by which EOtC impacts PA behaviours, school motivation, wellbeing, and academic achievement26.

Strengths and limitations

The study has strengths and limitations, which are discussed below.

The study is a cluster randomised trial, which enables more solid conclusions about the effects of EOtC by ruling out confounding influences such as teachers and pupils in the intervention and control groups being different. However, (as is often the case in education research) the intervention components could not be blinded to those delivering the intervention (the teachers) or to the intervention receivers (the pupils). The attrition of school classes during the study was higher than expected resulting in lower power than aimed for. This might have contributed with a higher risk of conducting a type 2 error.

The participation of schools was voluntarily. This introduces a risk of a selection bias where it is mainly schools and teachers initially interested in using EOtC that have been recruited and participated in the study. This makes it uncertain to what extent the results of the study apply if the EOtC intervention is attempted to be implemented at scale53 in a non-voluntary, and more top-down decision process e.g., mandated by law. As the recruitment of schools and teachers were based on their interest and willingness to participate it is likely that the participating teachers in both the intervention and the control schools had an interest in implementing EOtC in their teaching. However, being assigned to the control group when hoping to do the intervention may have resulted in a higher use of the EOtC in the control group than is generally the case in Danish schools, which is likely to have impacted negatively on our results and as such we are quite confident in the conclusion that implementation of EOtC has a positive impact on school week PA.

It is therefore an important strength of the study that data on weekly EOtC practice were collected among all classes in the project. This implementation data enabled us to avoid a black box evaluation with no knowledge of implementation extent54. It also enabled us to conduct a PP analysis of the effects when EOtC is implemented at a minimum of 150 min weekly.

The use of a device-based measure of PA behaviours circumvents detection biases such as recall and social desirability biases but relies heavily on equipment calibration and mathematical conversions that might obscure identification of true effects. Further, the study employs some quite complex and novel statistical approaches in the main PP effect analysis which takes the compositional nature of PA, clustering of participants into account, as well as imputation of missing data to reduce attrition bias. The use of complex statistical manipulation and multiple adjustments, however, might likewise mask true effects. As such, it is considered a strength that the robustness of these results is checked against the more commonly used methods in sensitivity analysis using ITT analysis and PP analysis on the full dataset. It must also be considered a strength that the overall results of the sensitivity analyses were consistent with the primary analysis.

Conclusion

The study shows that the TEACHOUT EOtC intervention positively affects pupils’ school-based MVPA, LPA, and decreases sedentary time. This is mainly due to children’s amount of running and to a lesser extent walking during the school week being increased. However, these effects on school-based PA were not sufficient to significantly increase the children’s total amount of PA over the week.