Study design
A 2x2x2 factorial randomized controlled trial was carried out to evaluate the efficacy of three BCTs (action planning, coping planning and self-monitoring) and their combinations (27). The protocol was preregistered (32). Participants were randomly allocated to one of eight experimental groups, with each experimental group receiving a unique combination of either none, one, two or all three BCTs (see Table 1). The CONSORT 2010 checklist and TIDieR checklist were taken into account (33, 34); the completed checklists are included as Additional file 1 and 2.
Table 1. Combinations of the three BCTs for each experimental group.
|
Action planning
|
Coping planning
|
Self-monitoring
|
Group 1
|
+
|
+
|
+
|
Group 2
|
+
|
+
|
-
|
Group 3
|
+
|
-
|
+
|
Group 4
|
-
|
+
|
+
|
Group 5
|
+
|
-
|
-
|
Group 6
|
-
|
+
|
-
|
Group 7
|
-
|
-
|
+
|
Group 8 (control group)
|
-
|
-
|
-
|
A plus (“+”) symbol indicates that the experimental group received that BCT while a minus (“-“) symbol denotes that the experimental group did not receive that technique. All eight possible combinations of the three BCTs were considered.
Participants and procedure
Adults (18-70 year) were recruited either in person by researchers at the city library of Ghent (which is also a meeting place and strongly focusses on innovation and research in collaboration with Ghent University), or through social media (i.e. Facebook) between February 2018 and December 2018. Only Dutch-speaking adults who had internet access at home and/or work and owned a smartphone (iOS or Android) were eligible. In addition, individuals had to complete the ‘Physical Activity Readiness Questionnaire’ (PAR-Q) (35) as a screening instrument to detect individuals at risk for adverse effects when being more physically active (e.g. individuals with heart problems). The PAR-Q is a self-report instrument consisting of seven items with ‘yes’ or ‘no’ answers. Only those who answered ‘no’ to all items were eligible for the study. Eligible participants were then provided with more information about the study either in person (for those recruited at the library) or via email (for those recruited via social media).
The intervention lasted for 5 weeks. It consisted of 5 weekly website sessions and a mobile application, which could be used at any time during these 5 weeks. The procedure of the study is shown in Figure 1. Participants were measured at 2 time points; pre-test (T0) and post-test (T1). The pre-test measurements were conducted via an online survey tool (Limesurvey, 2017), and included: 1) an agreement requiring them to provide informed consent, 2) sociodemographic information, 3) information about the difference between PA and SB, followed by questions about whether they wanted to improve PA or SB, and 4) questions assessing their current PA or SB level respectively (depending on their response to the previous question). Subsequently, participants were instructed on how to download the “MyPlan” (i.e. the English translation of “Mijn Actieplan”) app on the iOS AppStore or the Android PlayStore. Thereafter, participants were randomly assigned by a researcher (HS) to one of the eight experimental groups using block randomization (on the website www.randomization.com) (32). The study was single-blinded: participants did not know which experimental group they were allocated to, but the researcher did (32). After allocation, participants received an email with a link to the intervention and further instructions.
One week after each intervention session, an automatic email to encourage people to start with the following weekly session was sent. Participants who did not complete a session, were sent a reminder. If they had not completed a session after two weeks, they were phoned by the researcher (HS). When there was no reaction after 3 weeks, the participant was considered as a drop-out. Approximately one week after completing all 5 sessions, an email for post-test measurements (T1) was sent to the participants. Post-test measurements included the same questions to assess their PA or SB level (depending on what they had chosen at the pre-test measurement).
As stated in the protocol, the a priori power analysis suggested a sample size of 260 participants. Taken attrition into account, we aimed to recruit approximately 480 participants (32). The factorial trial was approved by the Committee of Medical Ethics of the Ghent University Hospital (ID number: NCT03274271).
Intervention MyPlan 2.0
MyPlan 2.0 is an e-and m-health intervention consisting of a website and a mobile application to increase PA and reduce SB. MyPlan 2.0 is based on a previous e-health intervention MyPlan 1.0 which was found to be effective for improving health behaviour (36, 37), but had high attrition rates (78.2%) (38). Therefore, in MyPlan 2.0 a number of changes were implemented based upon users’ needs and feedback (38, 39). MyPlan 2.0 is based on the Health Action Process Approach model (19), and includes BCTs such as goal setting, providing information on consequences of behaviour, providing feedback on performance, social support, action planning, coping planning, self-monitoring and reviewing behaviour goals. In this study, only the effects of action planning, coping planning and self-monitoring were experimentally investigated (Table 1) because they are well known to address post-intentional processes by bridging the intention-behaviour gap (19). Before a participant started with the intervention, they already chose at pre-test measurement which of the two behaviours (PA or SB) they wanted to improve (i.e. goal setting). This was done to create a feeling of ‘goal-ownership’ (20, 40). Depending on the chosen behaviour, the participant was directed to MyPlan 2.0 either targeting PA or SB. The two versions of the programme have an identical structure and offer the same BCTs, only the content was adapted to either PA or SB. Additional file 3 and 4 provide screenshots of the website and the mobile app.
The website
In brief, the website consisted of 5 weekly sessions (Figure 1). In the first session, users had to register with their email-address and create a profile. Every participant received a basic intervention to which the delivery of the BCTs action planning, coping planning and/or self-monitoring was experimentally manipulated (Table 1). The basic intervention consisted of 1) The possibility to fill in a quiz with information about the benefits about the selected target behaviour (increasing PA or decreasing SB) (i.e. providing information on consequences of behaviour), 2) Tailored feedback on the current state of their chosen behaviour (i.e. providing feedback on performance), 3) The possibility to read about how they could obtain social support from their partner, friends, family or colleagues.
With the BCT “action planning ” participants were prompted to make a plan by specifying how they wanted to reach their goal, what they wanted do to, where and when they wanted to do it. For example a set of possible answers to the questions for a participant who chose PA was: 1) “How do you want to be more physically active this week?” and his/her answer could be: “I will be physically active in my leisure time”, 2) “What do you want to do to be more physically active in your leisure time?” Answer: “I will go swimming”, 3)“Where do you want to do it?” Answer: “At the swimming pool 2 blocks from my house”, 4) “When do you want to do it?” Answer: “On Wednesday and Friday evening”.
With the BCT “coping planning” participants had to identify difficult situations or barriers they anticipated to experience while changing their behaviour over the course of the intervention. Participants were then given a list of possible solutions for each barrier, from which they could choose a relevant solution for their specific situation. For example a participant who chose PA might have been asked: “What could be a barrier for you this week to be more physically active?” and the answer could be: “It’s very busy at work this week, so I will have no time”. In response a list of possible solutions adjusted to the barrier were offered (e.g. “I will block a moment in my agenda”).
With the BCT “self-monitoring” participants were prompted to monitor their behaviour. They were offered several possibilities (e.g. the app, a diary, a calendar, an activity tracker). For example a person who chose PA may have been asked: “How do you want to keep track of how physically active you are this week?” and an answer could be: “Via the app”.
At the end of the first session, the completed plan was summarised in a printable form. After one week, users received an email to start the following session. There were four follow-up sessions (session 2-5) in which all participants, as part of the basic intervention, were required to reflect on their progress of behaviour change of the past week by evaluating their PA or SB goal (i.e. reviewing behaviour goals). Furthermore, participants could see their plan from the previous week and were prompted to adapt or maintain their action plan, coping plan and/or self-monitoring method depending on the experimental group they were assigned to.
The mobile application
The application was intended to provide support to the participants on a daily basis. It consisted of different modules through which the users could freely navigate. Again, every participant received a basic intervention with 1) a module consisting out of a quiz regarding the benefits of more PA or less SB, 2) a module where they could collect medals for completing website sessions and quizzes. In the other modules, the experimental manipulated BCTs were included as a function of the experimental group they were assigned to.
The BCT “action planning” was implemented in the app by providing the users with the option to review and change their plan throughout the week. Furthermore, a message was sent to remind participants about their plan of the day. If a person planned to go swimming on Wednesday and Friday, a message was sent those days as a reminder.
The BCT “coping planning” was implemented by providing a list of barriers participants could encounter. When selecting a barrier, a list of specific solutions for this barrier appeared.
The BCT “self-monitoring” supported users to monitor their behaviour. At the end of each day, participants received a message where they could rate on a scale from 0 to 5 if they succeeded in their plan of the day. They could also assess an overview of their responses of the week which were displayed in a graph.
A more detailed description of the intervention can be found in the study protocol paper of “MyPlan 2.0” (32). No changes regarding bug fixes, downtimes, or content changes to the MyPlan 2.0 intervention occurred after trial commencement.
Measurements
At pre-test (T0), the following sociodemographic variables were assessed: age, gender, weight, height, educational level (4 categories), occupation (7 categories) and marital status (4 categories). Weight and height were self-reported.
Primary outcomes (PA or SB) were assessed via self-report at pre-test measurement (T0) and post-test measurement (T1). If participants chose to increase their PA, the Dutch long version of the ‘International Physical Activity Questionnaire’ (IPAQ) (41) was used to measure total PA and moderate-to-vigorous-intensity PA (MVPA) in four domains (i.e. work, home, leisure time, transport), as stated in the protocol (32). However in this study, only MVPA in minutes per week (across all domains) is the outcome for PA. MVPA is considered the most important indicator of PA and PA guidelines for adults only address activities of at least moderate to vigorous intensity (7). The IPAQ has good reliability (ICC range from 0.46 to 0.96) and a fair-to-moderate criterion validity (ρ = 0.30-0.37). As the IPAQ overestimates PA (42), the data were truncated following the method described by Dubuy et al. (43). If participants chose to reduce their SB, the Dutch 7-days sedentary behaviour self-report questionnaire (SIT-Q-7d) (44) was used to measure sedentary time in five domains (meals, transportation, occupation, non-occupational screen time and other sedentary time). Also, here, only sedentary time in hours per day (i.e. the sum of sedentary time spent across all domains) is the outcome for SB, as existing SB guidelines focus on the total time spent sedentary. The SIT-Q-7d shows moderate-to-good test-retest reliability (ICC= 0.77) and moderate-to-good criterion validity (ρ = 0.49) for total sitting time on an average day (45). The data were truncated at a maximum of 16 hours of sedentary time a day (46).
Statistical analysis
In the protocol we stated that the observations would be combined by analyzing PA and SB together by using a standardized PA/SB score, then using the choice of behaviour (i.e. PA or SB) as a moderator (32). However, in this paper, we performed the analyses separately for PA and SB because of the following reasons: 1) It is complex to interpret the results of standardized scores, 2) A certain increase of PA (e.g. 30 min/week) does not have the same impact on health outcomes as the same decrease in SB (e.g. 30 min/week), regardless of whether standardization is used or not, and 3) PA and SB may have different working mechanisms when changing behaviour (47).
Descriptive statistics are provided for the total sample and for each of the 8 experimental groups. We first checked whether there were significant differences in the sociodemographic variables and the baseline levels of PA or SB between the experimental groups using one-way ANOVA (for the continuous variables) or chi-square tests (for the nominal variables). We then checked whether there were significant differences in the sociodemographic variables between participants who chose PA or SB. Lastly, we checked whether there were significant differences in the sociodemographic variables and the baseline levels of PA or SB between those who completed the study and those who dropped out (before week 5 or at post-test) using t-tests (continuous variables) or chi-square tests (nominal variables).
Linear mixed effect models with random intercepts for the participant (48) were fitted to the observed data using the lmer function from the ‘lme4-package’ (49) in R (version 3.2.5) (50) to investigate the effect of three BCTs (i.e. action planning, coping planning and self-monitoring) and their combinations on either PA or SB (48). The p-values were obtained using the ‘lmerTest-package’. Linear mixed effect models can account for missing data in repeated measures (51) provided that data are assumed to be missing at random.
There were separate analyses for MVPA and SB. In a first step, for each of the three BCTs (i.e. action planning, coping planning, self-monitoring) the interaction between time (T0=0, T1=1) and BCT (absent=0, present=1) was analysed. This interaction effect examines the efficacy of each BCT, independent from the other BCTs, and provides evidence for the overall efficacy of each of the three techniques (52). In a second step, a model for MVPA (and SB, respectively) was fitted with all possible interactions: time (T0=0, T1=1) x action planning (absent=0, present=1) x coping planning (absent=0, present=1) x self-monitoring (absent=0, present=1). Based on this model, pairwise comparisons between different combinations of the BCTs were analysed. By doing so, we examined to what extent adding a BCT increased efficacy (e.g. the effect of action planning alone can result in a different outcome than for example action planning combined with coping planning). The pairwise comparisons were calculated using the linear hypothesis function of the ‘car-package’ in R (53). These ‘follow-up’ or ‘post-hoc’ tests compare the different combinations of the techniques (represented by the 8 experimental groups). The codes for the linear hypotheses, can be found in Additional file 5. Effect sizes (with corresponding 95% confidence intervals) are expressed as the difference in change from pre-test to post-test between combinations.
In order to assess the effects of all possible combinations we had to perform multiple statistical tests. The multiple testing problem can be avoided by adjusting the cut-off score α using the Bonferroni correction to with n being the number of tests performed (54, 55). However, we chose not to make such adjustments for multiple comparisons because this 1) will lead to fewer errors of interpretation and 2) will not lead to missing possibly important findings (54, 55). As such, p-values below .05 were considered statistically significant.