Nurseries and children
In 2002 we invited 124 nurseries to participate in the movement and activity Glasgow intervention in children (MAGIC) trial. Eligible nurseries had at least 12 children in their preschool year. We randomly selected 36 of the 104 nurseries willing to participate. To ensure comparability of intervention and control groups, nurseries were stratified and pairs of nurseries randomly selected from the same stratum, one randomly allocated to intervention and one to control. Stratification was carried out simultaneously according to three characteristics that might have affected the intervention or study outcomes: type of nursery (school, class, extended day, private sector); size of nursery (area and number of children); and socioeconomic status of the area. All families with children in their preschool year attending the 36 nurseries were invited to participate. Parents gave informed written consent to participation
The intervention had nursery and home based elements and is described in detail elsewhere (www.gla.ac.uk/developmental/research/activities/Exercise%20&%20Metabolism/Magic/index.html).
—The nursery based element was an enhanced physical activity programme consisting of three 30 minute sessions of physical activity each week over 24 weeks. To deliver the intervention two members of staff from each intervention nursery attended three training sessions. A researcher unblinded to allocation (AW) carried out a monitoring visit to assess implementation. The nursery based element of the intervention was intended to increase levels of physical activity and children's fundamental movement skills10
and meet the requirements of the “physical development and movement” component of the nursery curriculum in Scotland. Nurseries experience several barriers to meeting this curriculum requirement, including lack of space, and limited competence of staff in physical education.12
The nursery element of the intervention was also intended to be inexpensive and therefore generalisable (capital cost <£200, €297, $377).
—The home based element of the intervention had two parts: each participating family received a resource pack of materials costing £16 (€24, $30), with guidance on linking physical play at nursery and at home, and two simple health education leaflets (one on opportunities for increasing physical activity, summarising our recent evidence that levels of physical activity in preschool children in Glasgow were low9
; the second encouraged families to seek opportunities to reduce the time spent watching television). For six weeks during the intervention, each intervention nursery also displayed posters focused on increasing physical activity through walking and play.
Control group—In the control group, nurseries continued with their usual curriculum and headteachers agreed not to enhance their physical development and movement curriculum.
Objective and outcome measures
We tested the efficacy of the intervention by comparing nurseries allocated to intervention with those in the control group at six months after the start of the intervention (when all children were still in their final year at nursery) and 12 months after the start (when 99% of children had gone to school). All primary and secondary outcomes were measured less than one week apart in pairs of nurseries, and outcome data were analysed and presented at the cluster (nursery) and child level.
Primary outcome measure
—Our primary outcome was body mass index expressed as a standard deviation score.14
This was calculated at baseline and at six months (mean 24 weeks, SD 2) and 12 months after the start of the intervention (mean 52 weeks, SD 4). To obtain the score LK, who was blinded to group allocation, measured height to 0.1 cm and weight to 0.1 kg in duplicate using a portable stadiometer (Leicester Height Measure, Child Growth Foundation, London) and portable scales (TANITA 300GS, Cranlea, Birmingham) with children in light indoor clothing and no shoes.
Secondary outcome measures
—We measured habitual levels of physical activity and sedentary behaviour objectively over six days with accelerometry at baseline and at six months15
using the CSA/MTI WAM-7164 accelerometer (Manufacturing Technology, Fort Walton Beach, FL, USA). Activity data were summarised as total physical activity (accelerometer count per minute) and proportion of waking hours in moderate or vigorous physical activity (accelerometer count >3200 per minute)15
and in sedentary behaviour (no trunk movement; accelerometer count <1100 per minute).16
Because accelerometers and staff time were limited, in nurseries with more than 15 participating children we randomly selected up to 15 children per nursery for accelerometry. We objectively assessed performance in fundamental movement skills at baseline and six months using the movement assessment battery, which has high validity and reliability in preschool children.17
The battery provides a global motor skills score of 0-15, which is a composite of performance in jumping, balance, skipping, and ball exercises. AF carried out all assessments and was blinded to group allocation.
Sample size and power
We originally intended to recruit a sample large enough to detect a reduction in standard deviation score of 0.25 with power 80% at a significance of 5%. Making conservative assumptions about the variance of the paired difference in the standard deviation score, the correlation within nurseries (and hence the loss of efficiency because of clustering), and the attrition rate at 12 months, we set out to recruit at least 400 children from at least 30 nurseries. As we were able to exceed these numbers, post hoc analysis suggests we had power of 80% to detect a reduction in score of just 0.125.
Sequence generation and blinding
All 36 participating nurseries were allocated to group in advance in one operation with stratified random sampling. Allocations were concealed by carrying out randomisation of the 36 nurseries at the same time and informing the liaison researcher and nurseries together.
The researchers who made the outcome measures were blinded to nursery allocation with the exception of the statistician who carried out the allocation (JHM) and the contact between the research team and nurseries (AW).
Statistical analysis and evaluation
We used multi-level (or hierarchical) modelling for all statistical analysis (MLwiN version 1.10), the iterative generalised least squares method for model fitting, and Wald tests to obtain P values. We analysed and compared baseline and later results using two level models, level 1 being the individual child and level 2 the nursery (cluster). To achieve approximate normality of errors at both levels, we transformed mean accelerometry count in counts per minute and proportion of time spent in moderate or vigorous physical activity by taking the natural logarithm. The modelling for each variable at baseline began by entering the random effects at nursery and child level (which are required because of the cluster design) along with five fixed effects: an “intercept” term, a slope with respect to age (years), a slope with respect to date of baseline activity measurement (days from start of study), and “dummy” variables for female sex and intervention group. For all the outcome variables that were not derived from measurements of physical activity we introduced a sixth term—namely, a slope with respect to log counts per minute. The modelling of the fixed effects proceeded in a backward stepwise manner until we obtained a final model in which all fixed effects were significant. P values were obtained for retaining the effect in the model, at the point at which it was a candidate for removal; this is in the final model for the effects that are significant, but in an intermediate model (not shown) for the other effects. For the random effects, variance components, and their estimated standard errors are listed. We calculated the intraclass correlation to compare the variation between clusters to the total variation; this is measured on a scale from 0 to 1, with a value close to 0 indicating that the clusters were all “similar.”
The modelling of follow-up data at six and 12 months proceeded in a similar manner, with the follow-up measurement itself (rather than the difference between follow-up and baseline) being used as the response. The fixed effects included an “intercept,” slopes with respect to the corresponding measurement at baseline and age at measurement, and “dummy” variables for female sex and intervention group. In models where the response was not derived from physical activity, we included a slope with respect to log (mean counts per minute) at six months as a further fixed effect. Where the effects of a “dummy” variable and slope were both significant, the interaction term between them was considered for the final additive model as another fixed effect. In no case was such an interaction significant at the usual 5% significance level.
We assessed the process of implementation of the intervention by requesting that nurseries record each session of physical activity delivered and attendance by children. We also ensured that nursery staff distributed home based equipment and educational materials to all participating families in the intervention group.