Study design

The cluster-RCT recruited adolescents aged 13–14 years from 38 state secondary schools/academies (approximately 2,500 participants in the South of England). Each school was randomly allocated to either ‘control’ or ‘intervention’ status. The study was approved by the Research Ethics Committee of the Faculty of Social Sciences, University of Southampton (ERG reference: 7892 amendment 1 (10/10/13), ERG reference: 12328 (14/11/14) and amendment 18817 (14/01/2016)). Written consent was obtained from participants and their parents. The study followed the CONSORT guidelines for the design and reporting of clinical trials (S2 CONSORT Checklist), and the CONSORT flow diagram for the study is shown in Fig 1. The full trial protocol has been published [41]. The concept for this project was informed by our pilot trial [23] and developed in 2014. In the initial phases, as this was an education RCT, the requirement to register as a clinical trial was not appreciated. Consequently, although the trial registration process was initiated in September 2014, final confirmation of registration did not happen until March 2015. The trial is registered with International Standard Randomised Controlled Trials, number ISRCTN71951436, the authors confirm that all on-going and related trials for this intervention are registered.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. CONSORT diagram to show participant flow through the trial.

https://doi.org/10.1371/journal.pone.0250545.g001

Participants

All state secondary schools within a two-hour travel radius of UHS were eligible. Though, we excluded independent, grammar and special schools. At an individual level, there were no exclusion criteria for students.

Participating schools were required to allocate a minimum of three ‘middle ability’ classes (~90 students). To minimise loss of participants to follow-up, student lists for each participating class were requested. Parent and student information sheets were provided for the schools to disseminate. Two versions of the information sheets were produced—‘intervention’ and ‘control’.

For intervention schools, parental consent was opt-in. For the control schools, consent was opt-out at the request of the schools to place less of a burden on them. In all cases, student assent was built into the questionnaire and, at any point, a student or parent could request that their information be withdrawn from the trial.

Recruitment of 38 schools was carried out between 1^st July 2014 and 22^nd October 2015. The first participant completed the baseline questionnaire in October 2014 and the last in July 2016. Follow-up took place between October 2015 and July 2017.

Randomisation and masking

Schools were recruited via a number of routes: information flyers and targeted letters sent to Headteachers and Heads of Science; presentations at local network and leadership meetings; and word of mouth. In total 112 schools were made aware of the trial. We aimed to recruit the first 32 schools to respond but, over the time-period when recruitment was open, 38 responded and were therefore included. The distribution of schools which responded included a range of schools from across the South of England and hence were broadly representative of the general population. Of the schools recruited, 70% had been judged as Good or Outstanding at their most recent Ofsted inspection, slightly below the national average of 74%. Prior to baseline data collection, groups of recruited schools were randomised in blocks of even numbers to the control or intervention arms. The blocks were not randomly selected as we needed to accommodate the needs of the schools to be randomised quickly. We obtained the largest even number of schools we could in a short space of time and then randomised. The block sizes varied between 2 and 12. Once two schools had been recruited, the schools were numbered, documented and the date of return of Headteacher’s agreement letter noted. Recruiting schools was an on-going process, the smallest block size being two and the largest twelve. Randomisation was conducted off-site to conceal allocation; researchers were unaware of the procedure, ensuring that prediction of allocation would not be possible. The randomisation process was conducted by a statistician at the MRC Lifecourse Epidemiology Unit with no knowledge of the schools in question, using computer-generated sequences. The nature of the intervention meant that it was not possible for researchers delivering the intervention, school staff, parents or students to be masked to group allocation.

Procedures

In supervised class time at school, typically a science lesson, web-based questionnaires were administered to control and intervention participants at baseline and again approximately 12 months later. The LifeLab team supported teachers, from all schools during the process of administering the questionnaire; a member of the LifeLab team was present, but the students completed the questionnaires independently on their computers. During follow-up data collection, support was again provided as requested by the class teacher. The questionnaire took on average 20 minutes to complete. Students either completed the questionnaires on iPads brought into school as part of the data collection exercise, on school laptops or devices, or in a school computer room. Typically, classes consisted of 30 students, with the class teacher and a researcher present. Students with literacy issues were supported by the class teacher or researcher reading out the questions to them. A written script was read aloud to explain the process to the students, ensuring consistency across all control and intervention schools. Shortly after baseline data were collected, intervention schools began the programme, as set out in Box 1. Control schools received normal schooling, they were offered the chance to attend LifeLab in the subsequent year, with a different cohort of students.

Outcomes

The primary outcome was a theoretical health literacy score. Following the approach devised by Guttersrud et al. [22], but guided by our previous pilot studies and PPI input from young people themselves, we identified a series of health literacy questions, based on those used in the pilot RCT [23], which assessed adolescents’ knowledge of the way lifestyle choices can impact on health throughout the life course, and on the health of future generations. These were a series of five questions, four of which used a Likert scale, with five responses ranging from ‘strongly disagree’, (scored 1), to ‘strongly agree’ (scored 5), the fifth question “At what age do you think our nutrition starts to affect our future health?” has responses from before birth then in decades up to > 60 years. These were coded as >30 years = 1, 20–30 = 2, 10–20 = 3, 0–10 = 4, Before birth = 5. Responses to the individual health literacy questions were coded and totalled, and the scores ranged from 6 to 30, with 30 being the highest score possible. The baseline scores were then standardised to produce a normally distributed theoretical health literacy score with a mean of 0 and a standard deviation of 1, with higher scores indicating better health literacy. The follow-up scores were standardised to the mean and standard deviation of the baseline scores to allow change to be assessed. Secondary outcome measures were assessed with questions about students’ reported health behaviours, and perceptions of the healthiness of their lifestyle, including understanding of cardiovascular disease and cancer risk. ‘Prudent’ diet score for each individual was an additional secondary outcome measure, generated using principal components analysis of data from the food frequency questionnaire on consumption of 15 specific food items. The second principal component gave positive weights to the foods and drinks that are generally considered healthy (e.g. fruit, vegetables, fish) and negative scores to those considered less healthy (e.g. biscuits, sweetened drinks, sausages). The score from this component at baseline was standardised as a prudent diet score in the same way as in previous studies [42]. The coefficients from this score were applied to the reported food and drinks consumption at follow-up (standardised to the mean and standard deviation of their values at baseline) to obtain a second prudent diet score for comparison with the baseline scores [42].

For the purposes of trial monitoring and reporting, an adverse event was considered to include any sign of distress related to diet, weight, or body image during the intervention or questionnaire completion, as reported by the school teachers, LifeLab teachers, researchers, parents or students themselves. None were reported.

Statistical analysis

Using data (z-scores) from our previous research in young adults in and around Southampton [18], we estimated an intra-cluster correlation of 0.035 from previous data; to be conservative we rounded this up to 0.04. With three Year 9 (aged 13–14 years) classes from each school averaging 90 students per school and using 90 as our cluster size we estimated that 14 clusters in each group had 80% power at the 5% level of significance to detect a difference in change in primary outcome score of 0.25 SDs over a 12-month period between the intervention and control schools.

Descriptive statistics, including chi-squared tests and independent t-tests, were used to test differences at baseline between the control and intervention groups in school characteristics, participants’ demographic characteristics (age, gender, and home deprivation score) and the primary outcome (theoretical health literacy score). Descriptive statistics were tested for differences between participants who dropped out and those who completed questionnaires at 12-month follow-up. To assess the level of deprivation of the area in which participants lived, the Income Deprivation Affecting Children Index (IDACI) [43] was obtained from home postcode, with higher scores indicating greater deprivation. An intention-to-treat analysis was performed using multi-level models to account for clustering [44]. Multiple linear regression analysis was used to compare theoretical health literacy and continuous secondary outcomes, in the intervention and control groups, adjusting for baseline values, gender and IDACI score. The categorical outcome variables were dichotomised and general linear mixed modelling was used to obtain prevalence rate ratios for the outcome in relation to the intervention. We used a Poisson model with robust standard errors and a log link function. Outcome variables were the results at the 12-month follow-up with adjustment for baseline responses and IDACI deprivation score included in all models. We also adjusted for gender since single-sex schools were included in the recruitment process and analysis showed small differences in gender distribution of the schools in the two arms of the trial. Planned subgroup analyses focused on whether there were different effects for boys and girls, and for more and less disadvantaged students (based on IDACI score). All models included schools as a random effect, assumed to be independently normally distributed. To assess the fit of the model, standardised residuals were plotted against fitted models and quantile-quantile (q-q) plots of the standardised residuals and the best linear unbiased predictors against the quantiles of the normal distribution were examined. All analyses were conducted in STATA.

Comparison with other studies

The development of LifeLab drew on the experience of a similar laboratory-based facility for schools in Auckland, New Zealand, known as LENScience. Many children from across the country visited LENScience, and a before- and after- comparison was undertaken, albeit without a control group. Similar to our findings, the students’ knowledge increased, but they also showed some limited impact on behaviour change [21]. However, conducting such studies at a time when adolescents are undergoing rapid changes in their behaviours means that the lack of a contemporaneous control group makes interpretation of such findings difficult. This was the rationale for the present study.

Strengths and limitations

Historically, recruitment of schools has been a concern for educational RCTs [52, 53]. We have shown that providing opportunities linked to the National Curriculum, and which meet schools’ needs, can provide a route to successful engagement; we over-recruited to this study. This is also a consequence of the project being embedded within the local education community and good working relationships being cultivated over many years. The overwhelming majority of parents supported the intervention, with more than 99% giving consent for their children to participate. The retention rate was high (85%), showing continued engagement from the schools in both the intervention and control arms.

Educational interventions delivered in school have the potential to interact with a large number of students, providing a means for public health interventions to have a wide reach. To increase the chance of effectiveness, however, an educational intervention needs to be delivered by those professionals who understand best how to engage with the appropriate age groups. Our intervention, which took into account the subject-specific constraints and other pressures under which teachers work, included as a key component the provision of face-to-face Healthy Conversation Skills training for teachers. This gave them communication skills to support students to reflect on their current health-related behaviours and plan changes to these. Behaviour change frameworks highlight the importance of a trusted ‘facilitator’ to deliver the intervention [54] and earlier research shows that school students prefer interventions delivered by teachers [8, 55]. Systematic reviews have shown that training for teachers prior to delivery of interventions is associated with effectiveness of the intervention [56] and Jacobs et al., under review.

A limitation of the study is that it was not possible to mask the schools, teachers, young people or intervention delivery staff to allocation, raising the risk of ascertainment bias. This is common in public health interventions. Standard protocols were followed for both the data collection and intervention delivery and a process evaluation was conducted alongside the intervention to determine both fidelity of implementation of the intervention and to document any indicators of bias. The RCT design was informed by discussion with school leaders, who deemed it necessary to have different consent procedures for the control and intervention arms. However, very few parents refused to give consent for their children to participate in either arm of the study, so we do not believe that this affected recruitment or led to differences between trial arms. Although we took a cluster randomisation approach, the relatively small numbers to include as individual clusters (n = 38) led to some imbalances, notably the follow-up was poorer in intervention schools which were in more deprived areas and this has to be considered a possible bias in interpreting the results. There were also more girls in the intervention arm. Our analysis was therefore adjusted for gender and deprivation level and took account of the baseline levels of outcome measures. There was a high retention rate for the follow-up measurements, though the participants lost to follow-up due to repeated absences at school or educated off-site may represent the most vulnerable participants with the worst health behaviours; our analysis of those who were retained for follow-up suggested that those lost were more likely to be from areas of higher deprivation.

Finally, it was not possible to randomise the schools after baseline data collection, which would have been ideal for minimising risk of bias. Schools require timely information for planning their annual timetable and allocating curriculum time and the short time between baseline data collection and intervention delivery necessitated that intervention schools were made aware of their curriculum commitment as soon as possible.

Due to the paucity of data of adolescent health literacy and lack of validated measures for adolescent health literacy, we followed the approach devised by Guttersrud et al. [22], but considering issues that were more appropriate for adolescents, we identified a series of health literacy questions, used previously in feasibility and pilot studies [18, 23]. These have been shown to be acceptable and understood by this age group, and give measurable differences over time. Acknowledging this as a limitation of this current trial, but questioning what the gold standard against which such a scale could be validated, our on-going work aims to develop and validate an appropriate tool.

This intervention showed change in knowledge, which, in the outcome measures selected did not translate into behaviour change; this limitation of the current trial could reflect a need to include additional intervention elements which support participants to effect change in behaviour.