Trial results

Parts of this chapter have been reproduced with permission from Willis et al.1 and Jago et al.2 These are Open Access articles distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/. The text below includes minor additions and formatting changes to the original text.

Recruitment

Figure 1 presents the CONSORT diagram and the recruitment and retention to the trial.

FIGURE 1 The CONSORT flow diagram. (Figure not reproduced here; see the original article.)

School recruitment and randomisation

Of the 33 schools that were approached to take part, 22 provided a signed school agreement (20 plus two reserve schools). Two schools withdrew before baseline measures and the reserve schools were utilised, ensuring that 20 schools were included. The two schools that withdrew cited staff capacity as the reason for doing so. Schools were grouped into three regions for the purposes of randomisation: Avon (including the local authorities of Bristol, Bath and North East Somerset, North Somerset and South Gloucestershire), Wiltshire and Devon. Table 6 presents the baseline deprivation characteristics of the schools randomised to each trial arm.

Schools in the intervention arm had higher IMD scores than schools in the control arm, with a median and IQR of 3 (1–4) and 4.5 (3–5), respectively.

Participant recruitment

Of the 1848 girls who received the recruitment briefing across all 20 schools, 1558 (84.31%) consented to take part. The mean number of eligible girls in each school was 92.04 (SD 24.23) and the mean number of girls who consented was 77.90 (SD 22.84). Recruitment percentages ranged from 46 to 97, but were comparable between trial arms. Table 7 presents participation rates by school and trial arm. Two schools (one in each arm) had particularly poor consent rates owing to school-specific issues (see Chapter 5, School and participant recruitment). In total, 290 girls did not provide consent to take part and were excluded from trial measures at T0. Of these, 230 opted out (intervention schools, n = 108; control schools, n = 122) and 60 did not participate for other reasons, such as absence, long-term illness, non-attendance or having left the school (intervention arm, n = 38; control arm, n = 22). Table 8 presents reasons for non-consent by school and measurement time point.

Peer supporter recruitment

In the intervention arm, 191 girls were invited to be peer supporters and, of those, 166 (86.91%) consented to take part in the study. Peer supporter recruitment rates are presented by school (see Table 7). Recruitment rates of Year 9 girls in intervention schools ranged from 13.41% to 23.21%. Two schools failed to meet the 15% recruitment target: one school because of a miscalculation in the peer nomination process that reduced the number of girls who were invited to be peer supporters (school 15) and the other because of unavoidable absences (school 14). Reasons for non-consent were provided by 17 of the 25 non-consenters and included not wanting to miss important lessons (n = 10) or known absence on the scheduled training dates (n = 4). Only three nominated girls specified that they did not want to become a peer supporter: one did not feel confident enough to fulfil the role, one felt that she was not active herself and the other gave no reason.

Trainer recruitment

Trainers were recruited via local authority networks to utilise a local delivery model. In total, 18 trainers expressed an interest and nine were invited to receive training after availability and suitability checks were applied. Of those nine, two were based in Wiltshire, five were based in Avon and two were based in Devon. The two Devon-based trainers withdrew prior to the training, citing changes in circumstances as a reason, leaving seven participating trainers. Participating trainers were all female, with ages ranging from 23 to 45 years and a range of experience in youth work, drama, yoga, physical activity or a combination of these: five had experience working with young people in a youth work or physical activity context, and the other two had experience of delivering drama or yoga workshops with young people and had both previously worked on PLAN-A as trainers in the feasibility trial. All had or obtained enhanced disclosure and barring services (DBS) checks to work with children for the trial.

Data provision

Baseline and follow-up measures were completed 12 months apart and, at T1, schools were measured 12 months ± 2 weeks after T0 to ensure seasonal consistency across time points. The timing of the measures is displayed in Table 9. Baseline data provision rates were high. Of the 1558 girls who consented to take part in the trial, 1511 (96.92%) provided baseline accelerometer data and 1557 (99.94%) girls provided complete questionnaire data. These completion rates were consistent across intervention and control arms. Participants lost to follow-up totalled 172 (control arm, n = 100; intervention arm, n = 72). Reasons for drop out are presented by school (see Table 8). Follow-up data provision rates were lower, but still > 85% for both accelerometer and questionnaire data, which is better than the age-group norms for research projects using accelerometers.94 Accelerometer and questionnaire data provision rates by school at each time point are presented in Table 10.

At baseline, 36 accelerometers were not returned and 11 files were corrupt or recorded no data and were, therefore, labelled ‘invalid’ files. Of the 1511 files available at baseline, 1431 (94.70%) met the inclusion criteria of ≥ 500 minutes of valid wear time on at least two weekdays. At follow-up, 21 accelerometers were not returned and nine files were invalid. Of the 1356 files available at T1, 1219 (89.90%) met the inclusion criteria and were included in the final analysis. Table 11 presents the proportion of participants meeting the accelerometer inclusion criteria by arm at each time point. Missingness is displayed by trial arm in Table 12.

Baseline data

Participant-level baseline data split by treatment arm are presented in Table 13. Overall, the trial arms were well balanced; however, the distribution of participants between the three geographical areas differed slightly. IMD scores were slightly higher in the intervention arm than the control arm, and the proportion of students claiming free school meals was slightly larger, suggesting a slight imbalance in deprivation, favouring the control arm. Accelerometer-assessed weekday and weekend physical activity scores were very similar between arms, with both groups achieving, on average, 51 minutes of weekday MVPA and 31% of each group meeting the Chief Medical Officer (CMO)’s guidelines of an average of 60 minutes of MVPA on weekdays.9 Active travel to and from school was more prevalent in the control arm than the intervention arm. Psychosocial variables were balanced between arms. The KIDSCREEN-10 T-scores, reflecting HRQoL, were balanced between arms, but were notably lower than the UK norm values for girls aged 12–18 years.95

Table 14 presents baseline characteristics of the girls in all 20 schools who were nominated by their peers to be peer supporters, compared with the characteristics of the rest of the sample. Demographically, there was balance between peer supporters and non-peer supporters (those who were not nominated). Peer supporters recorded more daily MVPA across the week than non-peer supporters by ≈ 4 minutes per day (51.01 ± 17.99 vs. 47.11 ± 18.53 minutes); however, sedentary time was very similar. A larger proportion of nominated peer supporters than of non-peer supporters achieved the CMO guidelines on weekdays (35.84% vs. 27.70%). Travel mode to and from school was similar. Peer supporters reported slightly higher self-esteem, physical activity-based autonomous motivation, need satisfaction and KIDSCREEN-10 T-scores than non-peer supporters.

Baseline characteristics of participants who did or did not provide weekday physical activity data at T0 (Table 15) and T1 (Table 16) reflect that, in both comparisons, those who failed to provide usable weekday physical activity data had slightly higher IMD scores (a larger proportion were in the most deprived quintile and a smaller proportion were in the least deprived quintile, and more were claiming free school meals). Those missing weekday MVPA data at baseline presented very slightly, but consistently, lower scores on all psychosocial variables, except for physical activity-related amotivation. They also walked home from school more and used public transport for the journey home less often. The same pattern was observed among those for whom MVPA data were missing at T1.

Primary outcome

The primary outcome of this trial was accelerometer-measured minutes of weekday MVPA at T1, 5–6 months after the end of the 10-week intervention period. Table 17 presents the primary outcome analysis using an ITT regression approach. After adjustment for baseline weekday MVPA, the number of days of valid data and the stratification variables (i.e. area and weighted IMD), the difference between arms was estimated as –2.84 (95% CI –5.94 to 0.25; p = 0.071) minutes, meaning that no statistically meaningful difference was found between the two groups. After additional adjustment for imbalanced variables at baseline (i.e. free school meals, travel mode to and from school, and IMD quintile), the difference increased to –3.57 (95% CI –6.75 to –0.39; p = 0.028) minutes in a negative direction. When adjusting further for the month that MVPA measurements were taken to account for seasonality, the difference in means across the intervention and control arms decreased to –2.96 (95% CI –5.98 to 0.062; p = 0.055) minutes and is, once again, not statistically meaningful. With a CI spanning zero, this would suggest weak to no evidence of a negative treatment effect that is ameliorated by seasonality.

The primary analysis was rerun to assess the sensitivity of the results to missing data. As demonstrated in Table 16, most baseline characteristics were comparable between those with and without missing MVPA data at T1. The group that diverged furthest from the mean SEP (i.e. IMD score) was that with missing follow-up data; participants in this group were more likely to come from socioeconomically deprived areas. Missing MVPA at follow-up was estimated using MICE. The imputation model included treatment arm, baseline weekday MVPA, the number of days of valid data, region, school-level IMD score, whether or not the child received free school meals at baseline, how the child travelled to school at baseline and IMD score based on home postcode. When comparing the primary analysis using complete cases with the analysis using the imputed data (Table 18), the between-group difference was smaller in the latter analysis and its CI included the null. These results are, of course, subject to assumptions inherent to the multiple imputation approach: chiefly, that the incomplete MVPA data are missing at random such that the missingness is associated with the observed data only.

We can conclude that, in this trial, the PLAN-A intervention was not associated with an increase in weekday MVPA in 13- to 14-year-old girls compared with usual practice.

Secondary outcomes

This trial also set out to determine the effectiveness of PLAN-A to improve four secondary outcomes, as listed in Chapter 2, Analysis of intervention effect. No statistically meaningful difference was observed at T1 between arms in minutes of weekend MVPA or weekday or weekend sedentary time or self-esteem scores (Table 19). The normality of the weekend MVPA data was improved when data were log-transformed. Analysis of weekend MVPA was therefore conducted using log-transformed data. The ratio of geometric means is presented, with 95% CIs and p-values.

Mediation analyses

Table 20 presents the mean scores of each motivational construct considered as a potential mediator by arm and measurement time point. Potential mediators were balanced between arms and decreased equally between T0 and T1. No meaningful correlations between potential mediators and the primary outcome were observed (Table 21), nor was there evidence of mediation, as the CI of the average causal mediated effect (ACME) for each potential mediator included zero (Table 22).

The average direct effect is the effect of the intervention on weekday MVPA that is not explained by the mediators (i.e. the amount of the total effect that is not explained by the mediator). The ACME is the effect of the intervention on the weekday MVPA that is explained by the mediator (i.e. the amount of the total effect that can be explained by the mediator). For instance, the row ‘Physical activity motivation: autonomous’ in Table 22 indicates that 1% of the intervention effect on MVPA is likely to be mediated by autonomous motivation.

Subgroup analyses

Subgroup analyses were performed by including an interaction term with the treatment arm and the subgroup variable. The results of these subgroup analyses should be interpreted with caution because of a lack of statistical power.

Subgroup analysis showed that only two interaction terms warranted further exploration: nominated peer supporters (p-value = 0.047) and the proportion of sedentary time at baseline (p-value ≤ 0.001). However, the individual measures of SEP (i.e. proportion of free school meals, median distance of home to school, and school- and pupil-level IMD score) showed that no further exploration was required. Table 23 presents the subgroup analysis comparing peer supporters (i.e. those nominated by their friends across all 20 schools) between treatment arms and non-peer supporters between treatment arms. The treatment effect among nominated peer supporters was estimated to be –5.98 minutes of MVPA (95% CI –10.34 to –1.61 minutes of MVPA; p = 0.007), meaning that nominated peer supporters in the intervention arm accrued fewer minutes of daily MVPA at T1 than the nominated peer supporters in the control arm. This contrasts with the feasibility trial finding that the intervention had a beneficial effect on peer supporters’ daily MVPA.51 The treatment effect among non-peer supporters was estimated as –1.90 minutes of MVPA (95% CI –5.18 to 1.37 minutes of MVPA; p = 0.23), meaning that non-peer supporters in the intervention arm had fewer minutes of daily MVPA at T1 than non-peer supporters in the control arm; however, the CI spans zero, so there is no evidence of a difference among non-peer supporters in the intervention and the control schools.

The difference in treatment effects across peer supporters and non-peer supporters was –4.08 minutes of MVPA (95%CI –8.14 to –0.01 minutes of MVPA; p = 0.049), suggesting weak evidence of a difference in treatment effect in nominated peer supporters versus non-peer supporters, with the nominated peer supporters having a larger negative treatment effect than the non-peer supporters.

Because proportion of sedentary time at baseline is a continuous measure, the treatment effect is a function of the proportion of sedentary time at baseline. As displayed in Figure 2, for cases where baseline sedentary time was approximately > 85% (i.e. for 4.7% of the sample), the estimated treatment effect implies that the intervention arm had higher MVPA than the control group at T1. This can be interpreted as the fact that for ‘high-sedentary’ girls, being in the intervention arm resulted in higher physical activity. Those who had baseline sedentary time between 75% and 85% (i.e. ≈ 60.7% of the sample) had a treatment effect that showed no evidence of a difference in MVPA at T1. Those who had smaller proportions of sedentary time at baseline, that is sedentary time between 45% and 75% (i.e. 34.5% of sample), had a negative estimate, meaning that those in the intervention group achieved fewer minutes of MVPA than those in the control arm at T1. This suggests that the PLAN-A intervention is more effective in pupils who are highly inactive; however, the trial was not powered to detect differences in this subgroup analysis, so estimates and CIs should be interpreted with caution.

FIGURE 2 Treatment effect over differing levels of proportion of sedentary time at baseline. LB, lower bound; UB, upper bound. (Figure not reproduced here; see the original article.)

Adverse events

There were two minor adverse events during intervention delivery. Both were reported to the chairperson of the ethics committee, the TSC and the funder (for details, see Report Supplementary Material 12).

Post hoc analyses

All of the planned analyses were performed, and the results were presented to the TMG. The group discussed further areas of interest in the data and a few additional analyses were agreed to inform the results. These additional analyses included further subgroup analyses, analysis of additional outcomes, school-level data exploration and further investigation into the peer supporter treatment effect.

Additional subgroup analyses

Two additional subgroup analyses (i.e. pupils meeting CMO guidelines and pupils’ mode of transport from school to home) were performed. There was no evidence of a difference in treatment effect between those who did and those who did not meet CMO guidelines (p-value = 0.138), or between those who did and those who did not use active travel as a mode of transport from school (p-value = 0.658).

Repeating the primary analysis using counts per minute as the outcome

After adjustment, at T1, the counts per minute (CPM) (weekday, weekend and average across all valid days) were lower in the intervention arm than in the control arm. Mimicking the primary outcome analysis, there was some evidence of a difference in mean weekday CPM, favouring the control arm (difference in means –22.23, 95% CI –44.43 to –0.05; p = 0.05) (see Appendix 2).

School-level change in moderate to vigorous physical activity from time 0 to time 1

Figure 3 presents a school-by-school comparison of change in mean weekday MVPA from T0 to T1. This figure shows that the mean weekday MVPA for all schools decreased between the time points, with the only exception being school 6 (mean change 2.52 minutes, SD 18.13 minutes). The largest decline was observed in school 1 (mean change –11.81 minutes, SD 19.87 minutes) and school 2 (mean change –12.27 minutes, SD 17.53 minutes).

FIGURE 3 School-by-school change in the primary outcome between T0 and T1. Navy indicates the intervention; aqua indicates the control. Error bars depict 95% CIs. (Figure not reproduced here; see the original article.)

Peer supporters versus non-peer supporters in intervention schools only

Comparison of weekday MVPA data between peer supporters and non-peer supporters (Appendix 3), showed that there is very weak to no evidence that there is a difference in weekday MVPA minutes at T1 when comparing trained peer supporters with non-peer supporters (adjusted difference –2.25, 95% CI –5.27 to 0.22; p = 0.071). At baseline, peer supporters had a higher mean level of weekday MVPA than non-peer supporters, but both groups had similar levels at T1. Both groups showed a decrease in their MVPA between the time points, with peer supporters having a greater reduction in MVPA than non-peer supporters. The adjusted difference in T1 MVPA between peer supporters and non-peer supporters also showed weak to no evidence of an intervention effect.