2.1 Study design and participants
The PPI was a two-arm randomised-controlled trial (RCT) that recruited Singaporeans aged 40-65 between April and December 2016 during freely available community-based health screenings in the northern part of Singapore. Participants satisfied several inclusion criteria (2), including no prior medical conditions preventing engagement in PA. Following baseline assessments, 160 participants were randomised in a 1:1 ratio into the intervention or the control group.
The process evaluation consisted of a mixed-methods study with an explanatory sequential design (27). The qualitative focus group data was collected primarily to explain the quantitative results after the intervention was completed. This design, along with the fact that participants were recruited and started the intervention in stages, also allowed an interactive approach in the process evaluation where group exercise participation data informed the selection of regular group exercise participators as well as non-participators for separate FGDs to understand similarities and differences in their experiences of the PPI.
2.2 Intervention components of the evaluated trial
Participants in the intervention group received face-to-face counselling on PA, during which they also completed a park prescription sheet with a trained study team member. The prescription sheet outlined a goal they committed to specifying the frequency, intensity, time and location of exercise in parks. Participants subsequently received a sheet to plan their weekly park PA and information brochures about parks in their neighbourhood. The first brochure was one developed for the trial, providing specific information on parks in the northern part of Singapore (within communities where participants resided) and their features, including walking trails, their difficulty level and locations of fitness corners. The second was an existing brochure produced by the National Parks Board, Singapore, containing a map and information on the Northern Explorer Loop (a series of parks in Singapore’s north connected by a network of walking and cycling paths). The participants also received a planning sheet, where they filled in the types of activities they aimed to do each week throughout the trial. Half-way through the trial, a trained study team member provided a brief counselling phone call, which assessed participants’ progress towards their set goals and included modification of those goals if necessary. In addition, participants were invited to join in a weekly one-hour outdoor structured and supervised PA program in the park for the entire intervention period of six months. Each one-hour session comprised moderate intensity aerobic activity and strength and balance exercises. To provide options in timing, two sessions of the structured PA program were organised in selected public parks located in the participants’ neighbourhood each week, one on a weekday evening and the other on Sunday mornings. The sessions utilised different areas and features of the parks, including walking trails and open spaces, to maximise participants’ exposure to greenery. To encourage attendance, participants received SMS reminders prior to each weekly exercise session.
Participants in the control group continued with their daily routine. They received standard PA promotion materials that were not related to exercise in parks, which were existing publications by the Health Promotion Board, Singapore. In addition, they received all the information materials after the intervention group completed the study and they were also invited to join ongoing exercise classes upon study completion.
2.3 Data Collection and Measures
Table 1 shows the baseline (T0), three-month follow-up (T1) and six-month follow-up (T2) measurements of this study which related to the implementation and mechanism of impact functions of the process evaluation. The measures for each function, time points, source and instruments are described further in this section and a full description is provided elsewhere (20).
2.3.1 Implementation measures
Measures for implementation dose included intervention group participation rates collected via participation records for the five intervention components. Participation in the initial counselling including the actual park prescription and providing program materials (planning sheet and brochures on local parks) was recorded at T0. Participation in follow-up counselling was recorded at T1 and participation in the group exercise was recorded throughout the 26-week intervention. The intervention group rated their satisfaction level and the quality of the PPI elements via two follow-up surveys (at T1, T2). In the survey at T1, participants were asked to rate the prescription, the prescriber and program materials. The survey at T2 included questions on quality of the phone follow-up counselling and satisfaction with the intervention overall. Focus groups, conducted after the intervention was completed (i.e. after T2), explored whether participants valued the intervention and each of its components and included in-depth discussions on the barriers and facilitators to participation.
2.3.2 Measures to assess the intervention’s mechanisms of impact on outcomes
The intervention’s mechanisms of impact were explored via survey measures administered at T1 and T2 amongst the intervention and control groups; the accelerometer at T2; and, via focus groups with intervention participants upon completion of the intervention. Group exercise participation measured each week over the 26-week intervention divided intervention participants into four subgroups: 0% participation (n=18), >0-35.9% participation (n=18), >35.9-67.9% participation (n = 17) and >67.9% participation (n = 18,). Subgroup analyses explored whether levels of group exercise participation were related to MVPA time in minutes per week at T2 in the intervention group. This was measured over seven days, with three valid days being required as a minimum for data to be included in the analysis. Focus groups conducted at completion of the intervention explored whether the experiences of participants in the intervention group differed by level of participation.
The four outcomes measured by the self-report survey which improved significantly in the PPI intervention group compared to the control group at T2 were included in the mediation analysis: psychological quality of life (Domain 2 of WHO QoL BREF) (28), recreational MVPA (weekly total time spent in MVPA during recreational activities, using Global Physical Activity Questionnaire [GPAQ] instrument) (29), park time in the last month and park PA time in a typical month (Outcomes paper, under review). Based on formative research (30) we hypothesised five T1 survey measures as mediators of the outcomes at T2: motivation to engage in PA (based on Behavioural Regulation in Exercise Questionnaire- 2 questionnaire) (31), social support for PA, recreational MVPA (from GPAQ), time spent in parks in the last month and time spent doing PA in parks in a typical month. The most common barriers to PA generally and park PA specifically identified in the formative research were being “too busy” and feeling “too tired”. Social support for PA in the form of group exercise was identified as a strategy to address these barriers, so we hypothesised social support for PA measured at T1 would be a mediator of outcomes at T2. Central to the logic for the intervention achieving outcomes is the idea that exposure to greenery via exercise in parks may have health benefits beyond exercise alone, and emphasising the restorative effects of nature during the counselling components of the intervention was identified as a strategy to address the barriers to PA. Therefore, park use and park PA measured at T1 were also hypothesised as mediators of the intervention’s effect at T2. Since the intervention included several strategies to increase motivation to engage in PA generally, and the formative research identified goal setting and planning for PA to increase motivation specifically, motivation to engage in PA measured at T1 was also hypothesised as a mediator of outcomes achieved at T2.
2.4 Process evaluation focus group methodology
For the process evaluation, the FGDs methodology was guided by phenomenological interpretive approaches (32, 33), which examined the experience of the park prescription trial from participants’ perspectives. Intervention group participants were purposefully sampled to join focus groups based on their distinctly high and low group exercise session participation rates. For the first focus group (FGD1), participants classified as regular participators (attended 68-100% of sessions) were approached during an exercise session as well as via text message and the first 12 who replied were included in the group. For the second (FGD2) and third FGDs (FGD3) a letter of invitation was sent to intervention group participants who were classified non-participators (attended 0% of sessions). FGD2 comprised nine non-participators and FGD3 comprised five non-participators and two participants who attended only three group exercise sessions. The FGDs took place in a quiet room, with each discussion lasting 60-90 minutes. Participants were offered $20 to participate. The discussion was organised based on topic areas which mostly reflected the research questions of this process evaluation - (1) Perceived value of the program and its components, and (2) Barriers and facilitators to intervention participation. FGDs were conducted by a trained facilitator, whilst an observer took notes and managed the audio-recording. FGDs recorded amongst Mandarin speakers were transcribed to English (First and second FGD). Each section of the text was matched to a code representing each individual participant.
2.5.1 Describing implementation
Quantitative analysis of implementation includes descriptive statistics to summarise participation rates in each of the five elements of the intervention (i.e. dose) and the ratings for quality of and satisfaction with these elements.
2.5.2 Sub-group analysis to assess group exercise participation as a mechanism of impact
A comparison of the group exercise participation rates with the MVPA/week was made within the intervention group. The 71 intervention participants who completed the trial were grouped by the previously described (section 2.3.2) quantiles of exercise participation rates. Among the 71 participants, 62 provided complete accelerometer data for this analysis (n = 15, 13, 17 and 17 respectively for the four participation sub-groups). Linear regression compared accelerometer measured MVPA/week at T2 among different participation groups, with and without adjusting for baseline self-report PA.
2.5.3 Mediation analysis to assess mechanisms of impact
Mediation analysis involved all intervention and control participants with complete survey data at T2. As a significant effect of the intervention on a mediator is commonly required to establish mediation (34-37), we first conducted simple linear regression analyses to evaluate the impacts of the intervention on the five hypothesised mediators, where the significant ones remained as potential mediators to assess their mediating effects. An alpha of 0.1 was used for exploratory purposes in this step instead of the default 0.05 elsewhere.
To evaluate the effect of each potential mediator, we performed mediation analysis using structured equation modelling (SEM) for each of the four outcomes. Figure 1 illustrates the path diagram of single-mediator SEM: the nodes represent the variables included in the model, while the arrows indicate relationships between variables and the corresponding direction. Three types of effects of the intervention on the outcomes were quantified: indirect effects, direct effects and their sum - total effects. Indirect effects refers to the portions of total effects of the intervention that function through the mediator of interest, whereas direct effects account for the remaining part of the total effect (36). Additionally, 95% CIs of the indirect effects were obtained from bootstrapping with 10 000 iterations. These SEMs were modelled using the R package ‘psych’ (version 1.7.2).
2.5.4 Qualitative analysis to explore implementation and mechanisms of impact
Framework analysis was conducted to identify themes and explore similarities and difference in feedback from participants with distinctly high and low group exercise participation (38). After familiarising themselves with the data, two analysers (N.P. and a Master of Public Health student who had completed an advanced qualitative methods in public health unit) created codebooks before meeting to discuss them and agreeing on a final set of codes. These codes were sorted into themes, then the analysers reviewed the raw data under these themes to ensure coherence. For the mapping step of framework analysis, the analysers aligned identified themes with qualitative research questions. Framework matrix charts containing data for each research question, with themes in columns and data beneath it divided in two rows of ‘participators’ (FG1) and ‘non-participators’ (FGD 2 and 3) were exported from NVivo (Pro version 12) to individual worksheets in a spreadsheet. This was done to explore similarities and differences in the discussion from participants with different rates of participation, which aligns with research question 3 of this process evaluation and complements the sub-group analysis (section 2.5.2). In the interpretation step, the two analysers reviewed the charts to make individual summaries of the findings. This process was both inductive (i.e. explanations were derived primarily from a close reading of the data, without trying to fit the data to pre-existing concepts) and deductive (e.g. trying to explain what was discussed by relating what is being said to socio-cultural phenomena or evidence from other literature). Following this, a third researcher (F. M.-R.) met with the analysers to discuss the summaries and agree on the findings.