This study considers the “big four” health behaviours, i.e. alcohol, diet, physical activity, which all have a strong impact on health. As a growing body of research suggests that these health risk behaviours typically cluster and do not occur in isolation [
21‐
23,
62], this trial can offer needed knowledge on the feasibility and effectiveness of an mHealth intervention targeting multiple behaviours, and to provide an mHealth platform where users can navigate freely using a personal dashboard that provides access to the four behaviour modules. The combination of recurring screening, text messages, and an interactive platform also mean a flexibility that encourage individual preferences. Findings in a systematic review and meta-analysis that evaluated the effectiveness of mHealth interventions targeting adolescents suggested that interventions using multiple mHealth solutions such as a combination of mobile applications, text messages and phone calls, have better potential than interventions that use single mode of delivery such as phone calls for instance. However, the results in the review was inconsistent between different outcomes which mean an uncertainty regarding the effectiveness of different mHealth solutions [
63]. Therefore, the current study might contribute to the understanding of the effectiveness of interventions that use a combination of a mHealth solutions such as automatic text messages and interactive dashboard platforms.
Limitations
While almost all of the trial processes are automated, one potential risk of detection bias is the use of follow-up by telephone among those not responding to initial automated attempts. While every effort will be taken to avoid prompting participants to reveal such information, participants may disclose their group allocation to research personnel at this stage. Overall, we believe that the benefits of decreasing follow-up attrition by calling non-responders reduces the risk of bias from missing data and outweighs this risk of detection bias.
We are expecting to have relatively low attrition, due to a scheme of collecting follow-up data which has been successful in our previous studies [
13,
33‐
35], despite not incentivizing participants. However, our power calculation is based on a Monte Carlo study, which takes into consideration uncertainty in our estimates, thus we expect that our calculations will be robust to slight deviations from assumptions.
The use of non-validated questionnaires for measuring mediators, acceptability, and experience of the intervention and control groups is also a limitation of this study. The decision to do so is based on reducing participant burden (to avoid attrition), but also to capture dimensions which are not present in validated measures. This does however limit both comparison with other studies, and the degree to which we can credit mediated effects to specific psychosocial constructs, as the face-valid single items are not validated to do so.
Finally, by randomising participants on individual level, rather than school level, we could potentially increase the risk of contamination between treatment groups. The risk of contamination is commonly present in digital intervention trials, as information is easily shared among participants. Cluster randomisation may reduce the risk; however, it may also create false confidence that the risk has been mitigated. High school students in Sweden are divided into something that resembles a traditional school class; however, students are mixed across classes and schools when attending courses. Young adults’ presence on social platforms also removes any geographical limitation that could be used for clustering; thus, there is no randomisation level that would sufficiently shield participants. In addition, distance learning has been implemented at times throughout the Covid-19 pandemic, which also limits interactions. Clustering would, therefore, in this case only accomplish a false sense of bias reduction. Analysis by treatment allocation, disregarding potential contamination, will bias estimates towards the null, potentially resulting in more conservative estimates than can be expected in a full-scale roll-out.