Recruitment and Eligibility
Between December 2016 to December 2017, 342 participants were recruited on a rolling basis. Recruitment was achieved via a combination of different recruitment strategies such as newspaper advertisement and word-of-mouth. Individuals who were interested were asked to contact the study coordinator for their eligibility. To be eligible for the study, participants must be: (1) Singaporean or Singapore Permanent Resident; (2) between 13–16 years old; (3) willing to wear a pedometer for 16 weeks; and (4) English-speaking, which is the dominant language spoken by this age group in Singapore.(25-28) Participants were also screened for exercise-related risks using Physical Activity Readiness Questionnaire (PAR-Q). A doctor’s approval was required if the participant answered “YES” to any of the items in PAR-Q.
Eligible participants were invited to attend a briefing session by the study team at Duke-NUS Medical School Singapore. Informed consent was obtained according to the National University of Singapore-Institutional Review Board (NUS-IRB) protocol, both from the participant and one of the parents or guardians. Consenting participants were then issued a pedometer wristband and asked to complete the baseline assessment that included wearing the pedometer for two weeks and answering the baseline survey. The trial lasted 16 weeks and participants were compensated for SGD 50 (~USD 36.23) upon completion of end-of-study survey.
Sample Size Calculation
The sample size calculation was powered to detect a mean difference of at least 0.25 standard deviations in the average number of steps taken by adolescents between the two study arms, accounting for the correlation between observations at the baseline and at 6 months. Based on 5% level of significance and a power of 80%, and by accounting for a potential 20% attrition, the target sample size was determined to be 312 (156 in each arm).
Randomization and Intervention Design
Randomization was performed after baseline assessment was completed. Of the 342 participants who registered for the study, 13 participants did not complete the baseline assessment and were excluded from the study. Eighteen of those who completed the baseline assessment were not contactable and were not included in the study. Participants were randomized based on gender and baseline step counts via block randomization. The randomization algorithm was pre-programmed by a statistician. The allocated arm was sealed in an envelope. A total of 311 participants who completed the baseline assessment (31 dropped out) were randomized to either Anonymous (n=155) or Onymous (n=156) Arms. The participants were informed of their allocated arms after they completed the baseline assessment. Since the aim of the intervention was to provide descriptive norm information about the participants, we could only start the intervention only when all 311 participants were recruited. However, since recruiting 311 participants could have taken us several months and we did not want participants who were recruited earlier in the study to lose interest, we decided to roll out the intervention in batches. A batch included one Anonymous Arm and one Onymous Arm, consisting of 12 participants in each arm. This method resulted in 13 batches in total. We chose 12 participants since we found this number to be large enough to create competition among group members, but small enough that participants can start the trial before they lose interest in the study. Figure 1 shows the recruitment and randomization process.
During the 16-week study period, both study arms received weekly notification of their physical activity performance from the past week via automated short message service (SMS). Participants were ranked according to the total step count accumulated in the past week within their group. For participants in the Anonymous Arm, the weekly information included step count ranked from the highest to lowest within all participants in the same group. Participants in the Onymous Arm received the same information plus the (real) full names of the participants next to the step count. To discourage lower levels of physical activity, those with a step count of zero were not included in the list for both arms.
Participants were asked to complete a survey at both baseline and end of the study. The baseline survey captured socio-demographic information and self-reported physical activity. Paediatric Quality of Life (PedsQL) scale, Asian Adolescent Depression Scale (AADS), Social Support and Exercise scale, Physical Activity Self-Efficacy (PASES) scale, and Physical Activity Enjoyment Scale (PACES-8) were administered in both surveys. The end-of-study survey also included questions about participants’ experience with the study and whether they knew participants in their group and in the study. The survey instruments developed for this study are provided as Additional File 1.
Steps (primary outcome): Steps were measured by a Fitbit FlexTM wireless pedometer. Fitbit devices have been validated in measuring step counts among healthy individuals.(29) The step counts recorded by the device can be easily visualized and monitored with a Fitbit account by synchronizing the data to an installed mobile phone or computer application. Average weekly number of steps was used for the analysis.
Pediatric Quality of Life (PedsQL) Inventory Score: Quality of life (QoL) was assessed using PedsQLTM 4.0 Generic Core Scales, a 23-item scale that was developed to evaluate QoL in teenagers, and has been widely used in other pediatric QoL studies.(30) The scores were transformed on a scale of 0 to 100. Higher score indicates better QoL.(31) License was obtained to use this scale for this study.
Asian Adolescent Depression Scale (AADS): Depressive symptoms were measured using AADS, a 20-item instrument that was developed in Singapore to assess depression among adolescents(32), and has been successfully used in other studies.(33) The total score is the sum of the 20 items. Possible score for this instrument ranges between 20 to 100. A higher score indicates higher level of depressive symptoms, a total score that exceeds 80 is an indicator of depression.(33)
Social Support and Exercise (SSE) Survey: Exercise-related support from family and friends was estimated using the Social Support and Exercise Survey.(34) This 13-item instrument evaluates the behaviour and attitude of family and friends toward their participation in exercise in the past six months with a 5-point Likert scale. Eleven of the 13 items measure supportive behaviour and attitude toward exercise, while two items measure negative social support associated with exercise. Reverse scoring was performed for the two items, measuring negative social support. A total score was generated by summing scores from both family and friends. Possible score for this scale ranges between 26 to 130. A higher positive support score reflects having received frequent positive support.
Physical Activity Self-Efficacy (PASES) Scale: This instrument contains 17 items that assess children’s self-efficacy in overcoming barriers to physical activity.(35) It is a dichotomous scale where “Yes” is coded as “1”, while “No” is coded as “0”. The total score ranges from 0 to 17. Higher score indicates a higher level of self-efficacy associated with physical activity.
Physical Activity Enjoyment Scale (PACES-8). Enjoyment of physical activity was assessed using a set of 8 questions derived from an original 18-item scale to measure enjoyment.(36) Participants were asked to rate their level of exercise-related enjoyment on a 7-point Likert scale. Possible scores for PACES-8 range from 7 to 57, with higher values reflecting greater enjoyment of physical activity.
User experience: Participants were asked to report their satisfaction with the intervention and level of participation. We also asked if they know anyone in the study or in their group, and their feelings when they ranked either the top or bottom five.
Analyses were done on an intention-to-treat basis for 311 adolescents who were randomized into one of the arms. The missing data (i.e., zero steps) for the number of steps were imputed based on 30 multiple imputed datasets by employing predictive mean matching.(37) Main analysis of primary outcome was conducted by comparing steps from the 155 participants in Anonymous Arm and 156 participants in the Onymous Arm. Secondary outcomes were analysed based on the completed surveys by the 138 participants in the Anonymous Arm and 144 participants from the Onymous Arm. For the secondary outcomes, scores were not computed if >50% of the items in the scale were missing.(31) The means of the completed items were imputed only if ≥50% items were completed.
We used a multilevel mixed-effects model to compare the differences between the two arms with random effects for groups and individual effects. In the mixed-effects model, we included the number of steps for the 16 time points, comparing each week in the trial to the baseline.
Group-based trajectory modelling was used to investigate the step trajectories over the trial period.(38) Following the methods from Nagin and Odgers (2010) (38) we identified 4 group trajectories based on the following considerations: 1) obtaining a minimal increase in the Bayes Information Criterion for an additional trajectory group, 2) obtaining for each trajectory group a close correspondence between the estimated probability of group membership and the proportion assigned to that group based on the posterior probability of group membership, 3) ensuring that the average of the posterior probabilities of group membership for individuals assigned to each group exceeds a minimum threshold of 0.7, and 4) ensuring that the odds of correct classification based on the posterior probabilities of group membership exceed a minimum threshold of 5. The functional form of the trajectory for each group was based on the significance of the polynomial terms by iteratively dropping non-significant terms. Dummy variables on gender (female=1, otherwise=0) and knowing other participants in the group (know someone=1, otherwise=0) were created. They were then interacted with the Onymous Arm dummy variable (Onymous arm=1, Anonymous arm=0) and were used as predictors of trajectory group membership. All statistical analysis was performed using Stata 15.(39)