A loyalty scheme to encourage physical activity in office workers: a cluster RCT

Physical activity levels

Physical inactivity is estimated to cause 6–10% of deaths worldwide from non-communicable diseases, such as coronary heart disease, type 2 diabetes mellitus and certain cancers. 5 On a global scale, estimates have highlighted that physical inactivity costs 53.8B (INT$, 2013) worldwide in 2013. 6 The impact of increasingly inactive lifestyles is thought to cost the NHS > £1B annually. 7 Within the UK, adults are recommended to undertake at least 150 minutes per week of moderate to vigorous physical activity (e.g. cycling, brisk walking or running). 8 The current guidelines also include recommendations to reduce sedentary behaviour across the day by taking regular breaks at work and incorporating active travel. 9 Despite government guidelines highlighting the health benefits of regular physical activity,7 > 80% of adults in the UK9 and > 60% of adults in Northern Ireland10,11 are still not meeting current recommended levels.

Within the most recent National Institute for Health and Care Excellence (NICE) guidelines,12 employers are encouraged to promote physical activity in the work place by identifying workplace ‘champions’ or by implementing workplace walking and cycling programmes. As a high percentage of the adult population worldwide is in employment and most of the waking day is spent at work, workplace-based interventions are practical settings for physical activity promotion and have the potential to significantly contribute to habitual physical activity and other beneficial health behaviours. With increasing numbers of inactive office-based occupations, improvements in physical activity levels may also contribute positively to well-being, mental health and productivity. 13 It has been suggested that there is a strong business case for investment in the health and well-being of the workforce. 14 In a report published by the Department for Work and Pensions,15 it was estimated that for every £1 invested in workplace health and well-being there is a potential return of > £4 as a consequence of improvements in absenteeism and productivity, which are thought to cost employers in the region of £30B annually. 13

Workplace-based physical activity interventions

Current evidence to support the effectiveness of workplace-based physical activity interventions is mixed,16 with reviews2–4 calling for more robust research on workplace-based interventions and well-designed RCTs. Previous meta-analyses of workplace-based physical activity interventions have shown small, positive short-term effects3,4 on levels of walking but little long-term effectiveness was evident. 12 Thus there is a recognised need to develop workplace-based interventions purposefully designed to encourage physical activity behaviour change maintenance.

More recently, a systematic review and meta-analysis of 26 intervention studies published in 201617 identified three main intervention strategies to reduce sedentary time and increase physical activity within the workplace: (1) educational or behavioural change programmes (e.g. goal-setting, motivational interviewing), (2) environmental changes (e.g. sit–stand workstations) and (3) multicomponent interventions (e.g. environmental changes coupled with behaviour change techniques). The review found that workplace-based interventions, especially multicomponent interventions (implementing both an educational and an environmental aspect to the intervention), effectively reduced workplace sitting time. However, most of these studies demonstrated behaviour change in the short term, with the longest follow-up from the included studies recorded at 12 months.

Financial incentives

Financial incentives have been proposed as a method of promoting healthy lifestyles by the UK government,18–20 but evidence in this field is relatively limited to date and considered controversial by some. 21,22 Financial incentives have been shown to be effective in the short term in smoking cessation studies23 and have also proved effective in encouraging discrete health behaviours, such as attendance at vaccination programmes. 20

In a systematic review published in 201524 that examined the effectiveness of financial incentives in changing health-related behaviours in general, it was reported that behaviour change was maintained until 18 months post baseline and 3 months after incentive removal. Effectiveness measured at > 6–12 months post baseline was moderated by the participants’ socioeconomic circumstances. In a recent trial,25 rebates were offered to participants when they purchased fruit and vegetables and this led to improvements in Healthy Eating Index-2010 scores. 26 The follow-up in this trial was conducted at 4–6 months and at 9–11 months after implementation, so only relatively short-term effects have been demonstrated.

Giles et al. 27 reviewed the effectiveness of financial incentives for behaviour change (17 reports of 16 studies) in a range of health behaviours, including smoking cessation, vaccinations/screening and physical activity, and found some evidence that effect sizes decreased as post-intervention follow-up period and incentive value increased.

Relatively little research has been conducted to date with regard to the provision of financial incentives in the promotion of physical activity. A recent systematic review’s results,28 based on three studies only, suggested short-term effectiveness, with the authors, unsurprisingly, calling for more research in this area. A recent trial conducted by Shin et al. 29 sought to assess the effectiveness of combining an activity tracker with financial incentives in a group of male students; the receipt of financial rewards was contingent on participants achieving their daily physical activity goals (process incentive) and achieving their target weight (outcome incentive). The trial concluded that the addition of a financial incentive was effective in increasing physical activity levels. Christian et al. 30 also sought to explore the feasibility of using financial incentives to increase physical activity levels in a teenage population (13.3 ± 0.43 years). The intervention involved provision of £25 of vouchers (five vouchers in increments of £5) per month for 6 months. The vouchers were provided to single participants. The authors found that not only did the vouchers encourage an increase in physical activity and significant improvements in the boys’ fitness, but they also reduced levels of sedentary behaviour in both sexes and encouraged friends to socialise.

Incentives have been shown to improve participant engagement and, as the aforementioned studies demonstrated, this can lead to increased levels of physical activity. However, as Mantzari et al. 24 reported, financial incentives can increase attainment of the target levels of behaviour change from the start until the end of the incentives’ offer, but subsequently there is a monotonic trend and a weakening effect over time. With such mixed findings and of which few provide longer-term follow-up data, we cannot yet be certain that increased levels of physical activity are maintained in the longer term, the behaviour change is maintained once the incentive is removed and new habits are formed. 24,31,32

Cost-effectiveness of incentive-based interventions

To date, studies have explored the cost-effectiveness of incentives across a range of health behaviours. The cost-effectiveness of financial incentives for smoking cessation in pregnancy is encouraging. Boyd et al. 33 conducted a smoking cessation trial in which women (n = 612) were randomised to usual cessation support with or without financial incentive vouchers up to the value of £400. The findings suggested that the financial incentive for smoking cessation in this population was highly cost-effective, with an incremental cost per quality-adjusted life-year (QALY) of £482.

Paul-Ebhohimhen and Avenell34 reviewed the literature on the use of financial incentives in the treatment of overweight and obesity and found that few studies had conducted cost-effectiveness analyses. The lack of evidence on cost-effectiveness poses a distinct limitation on financial incentives’ utility for policy-making.

Cost-effectiveness of incentive-based interventions for physical activity

Although the evidence for the cost-effectiveness of incentive-based interventions is limited, studies on interventions that focus on the promotion of physical activity, particularly within the workplace, are even more sparse. 35 However, encouragingly, a cost-effectiveness analysis performed in our pilot trial showed that the PAL scheme was potentially cost-effective from both a health-care and an employer’s perspective. 1

In summary, financial incentives alone may not be sufficient to promote maintained physical activity behaviour change, but may do so when embedded within an evidence-based, theoretically driven intervention. 36 Furthermore, financial incentive interventions designed using behavioural economic concepts have been shown to be effective for changing behaviour. 37 There is also limited evidence for the cost-effectiveness of interventions promoting physical activity in the workplace36,38 and for the cost-effectiveness of financial incentive-based interventions. 31 Therefore, to address such gaps in the evidence base and following the successful completion of a pilot study,1 we aimed to conduct a cluster RCT of a complex intervention (the PAL scheme), incorporating financial incentives to encourage physical activity and maintained behaviour change.

Study population

Randomisation, concealment and blinding

Clusters were the smallest work groups or units (e.g. a large open-plan office) within each participating organisation. A random allocation sequence was drawn up by the trial statistician and group allocation was stratified to ensure a similar number of clusters in both the intervention group and the control group. Research staff were blinded to group allocation until after data collection was completed. The outcome of the randomisation was communicated to participants by e-mail after baseline measurements were complete.

Intervention group

The PAL scheme is a complex multicomponent intervention based on concepts similar to those that underpin a high-street loyalty card aimed at encouraging repeated behaviour (i.e. loyalty) and is designed to incorporate a range of behaviour change techniques. Components include the provision of ‘points’ and rewards (financial incentives) contingent on meeting targeted physical activity behaviour goals (extrinsic motivation, goal-setting). Participants were encouraged to undertake 150 minutes per week of physical activity, which is in line with current guidelines. The PAL scheme integrated a novel physical activity remote tracking system with web-based monitoring and evidence-based behaviour change tools (i.e. self-monitoring, goal-setting). We carefully considered the design of this complex intervention in line with Medical Research Council’s guidelines. 45 We have followed advice from the MRC framework in our work, including theoretical work, development of the logic model, mapping of the intervention onto determinants and mapping of measures onto hypothesised mechanisms, pilot/feasibility study including qualitative work.

The 6-month intervention involved placing sensors (WiFi beacons) in the vicinity of participating workplaces at specific locations to encourage physical activity within a 2-km radius of participants’ worksites (i.e. including the provision of prompts/cues to facilitate habit formation). The WiFi beacons were placed at locations along footpaths, in local parks, leisure centres, shopping malls, bus stops and train stations. Maps of various walking routes and details about physical activity opportunities tailored to the workplace (and instructions on how to perform behaviour) were provided on the study website. Participant’s activity was logged when they passed within an approximate 25-m radius of the WiFi sensors with their PAL key fob when undertaking physical activity (e.g. walking). This logged the place, date and time of the bout of physical activity. Participants could log onto their account on the study website and receive real-time feedback on the number of minutes of physical activity logged by the tracking system. Minutes were converted to ‘points’ (10 ‘points’ for 1 minute of activity recorded), and collected ‘points’ were redeemable for rewards (i.e. downloadable retail vouchers) sponsored by, and redeemable at, local businesses. To reduce the risk of ‘gaming’, a daily ‘points’ cap was implemented and the transit times between sensors checked for anomalous values. Bonus rewards and ‘Double Points Days’ were offered when participants met specific weekly physical activity targets.

To tailor the intervention, a purposive sample of employees (both men and women across a range of ages) participated in three focus groups (maximum 10 participants per group) prior to the intervention to explore aspects such as the availability of, and preferences for, opportunities for physical activity in proximity to the workplace. Sensor locations were determined from the feedback received in the focus groups regarding popular walking routes. Employees’ opinions on the website interface and the rewards redemption function, for example, were also considered and this permitted the final tailoring of its functionality. To determine incentive levels prior to the intervention, stated preferences of the participants from the DCEs were used to assess their mean WTA, willingness to pay (WTP) and the trade-offs they would make for the attributes of the incentive programme. This information helped determine the level of the rewards available for earned ‘points’. In addition to the financial incentive element, the intervention had several other components designed to enhance the effectiveness of the incentives. These components were delivered via the study website and designed to have multiple effects: (1) to increase use of the study website, (2) as effective behaviour change techniques (BCTs) in their own right and (3) as techniques designed to aid the transition from more extrinsically motivated behaviour to more intrinsically motivated habitual behaviour. The techniques included the provision of regular tailored motivational e-mails, tailored feedback, information on walking routes in the vicinity of the participating workplaces and links to other resources, such as physical activity advice and healthy eating guidelines. They also included self-regulation techniques of goal-setting, self-monitoring and prompts to behaviour.

Underpinning theoretical framework

The financial reward component of the intervention was based on principles of learning theory46 by providing an immediate reward (extrinsic motivation) for behaviours that offer health gains in the future. The financial reward component of the intervention also contained elements based on other approaches, such as goal-setting, prompts, self-monitoring and habit formation that fit within a self-regulation control theory framework, motivational messages (persuasion) and social support (vicarious experience), all of which should increase self-efficacy according to social cognitive theory. 47 Social cognitive theory also holds that satisfaction with the consequences of behaviour change can act as a reinforcing mechanism, in addition to the reinforcement of financial rewards. 47 Therefore, the financial incentive component was embedded in a complex intervention containing evidence-informed BCTs. Figure 1 presents the logic model underpinning the development of the intervention.

FIGURE 1.

Logic model of the PAL scheme. Reproduced with permission from Hunter et al. 39,48 This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated. CVD, cardiovascular disease.

Reproduced with permission from Hunter et al. 39 This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated. Reproduced with permission from Hunter et al. 48 This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Outcomes and measures

Outcome measures are grouped into primary and secondary outcomes as displayed in Table 1. In summary, the primary outcome was physical activity behaviour change (mean number of steps per day) measured using a sealed pedometer. Participants were asked to wear the pedometer for 7 consecutive days on the waistband (dominant hip). They were asked to complete a wear time diary providing details on the dates and times that they removed and wore the monitor. Participants were advised to remove the monitor when showering, bathing or undertaking any water-based activities. Pedometer data were considered valid if the participant provided ≥ 250 steps per day for ≥ 3 days at each data collection period. 70 Pedometers were sealed at all times to blind participants to the output and prevent reactivity, which is standard practice. This device was used as a measurement tool only (i.e. the pedometer was not part of the intervention) and a standardised measurement protocol was used for the intervention and control groups. We followed a standard protocol that has been well validated71–75 and successfully employed as a measurement tool in numerous intervention studies.

Secondary outcomes included measures of health and well-being, work-related impacts, proposed mediators of behaviour change, compensatory behaviours, a range of variables for use in the economic evaluation and a range of variables for use in the process evaluation.

All outcome measures were analysed and, when required, collected by a postdoctoral research fellow blinded to group allocation. Self-reported outcome measures were collected at baseline and at 6 months (unless otherwise stated) via online questionnaires distributed by e-mail and automatically collated via Qualtrics.

Data collection

Sensitivity analysis

For a preliminary assessment of possible non-response bias, analysis of the primary outcome (i.e. 6-month pedometer steps per day, 12-month pedometer steps per day) was repeated after imputation of missing data using multiple imputation by chained equations (MICE) with 20 imputations. The MICE procedure operates on the assumption that data are missing at random. 76 Baseline variables were inspected for potential predictors of missingness of the 6-month outcomes and predictors of the outcome itself. Baseline ‘physical activity self-efficacy’,56 ‘intrinsic motivation’,59 ‘habit’,17 ‘recovery self-efficacy’20 and ‘maintenance self-efficacy’20 scores in addition to marital status (0 = married/cohabiting, 1 = other) significantly predicted missingness of the 6-month outcome. Baseline ‘physical activity self-efficacy’,56 ‘intentions’,61 ‘habit’,17 ‘outcome expectations’,55 ‘identified regulation’,59,60 ‘integrated regulation’,59,60 ‘intrinsic motivation’,59,60 ‘financial motivation’,57 ’outcome satisfaction’,55,65 ‘planning’,58 ‘social norms’,16 group assignment (0 = control group, 1 = intervention group), BMI, income (0 = ≤ £20,000, 1 = > £20,000) and season (1 = December 2015–April 2016, 2 = July–August 2016) significantly predicted the 6-month and 12-month outcome (i.e. mean number of pedometer steps per day). These variables were used in the imputation model to make predictions about the values of missing data. We then repeated the primary outcome analyses using imputed values and pedometer steps per day at 6 and 12 months were compared between the intervention and control groups using ANCOVAs adjusting for baseline values, randomisation stratum and season with SEs corrected for clustering. Owing to large seasonal effects, results are presented before and after adjusting for season. These analyses were carried out using the ‘mi estimate’ command in Stata. Rubin’s rules were used to obtain the final estimates from the estimates derived in the 20 imputed data sets. 77

Ethics and consent

Ethics approval was sought from and granted by the Office of Research Ethics Committees Northern Ireland (ORECNI) prior to the start of the study (reference number 14/NI/0090). Fully informed consent was obtained from all participants prior to their inclusion in the study. Participants were asked to confirm that they had read and understood the information sheet prior to agreeing to participate. Participants were also given an opportunity to ask any questions and ensure that these were answered satisfactorily prior to completing the consent form. Research governance approval was granted by the South Eastern Health and Social Care Trust and Belfast Health and Social Care Trust.

Methods

Participants’ experiences of being involved in the PAL scheme, including barriers to and facilitators of engagement with the intervention, were explored using focus groups (lasting approximately 1 hour) at 6 months (n = 9 focus groups) and 12 months (n = 6 focus groups). Purposive sampling was undertaken to ensure a representative sample of participants within each focus group (a maximum of 10 participants per group). A schedule of open-ended questions was employed to elicit information about reactions to the intervention, barriers to physical activity and suggestions for a future roll-out of the intervention if proven effective. At 12 months, additional focus groups were also undertaken with 17 of the same individuals who participated in the 6-month focus groups (n = 6 focus groups). On this occasion, the open-ended questions focused on physical activity behaviour change maintenance. Semistructured interviews were also undertaken with senior managers of six participating employers (n = 7 managers) and participating retailers (n = 4 retailers) to explore their perceptions of being involved in the study. Discussions were facilitated by AG, Doctor of Philosophy (PhD) (female), and analysis of the data was undertaken by LP, PhD (male), both of whom have prior focus group and qualitative methodology experience. The researchers did not have a relationship established with participants prior to study commencement.

Qualitative analysis

Focus groups and semistructured interviews were audiorecorded and transcribed verbatim. Transcripts from focus groups and semistructured interviews with retailers and managers were analysed using thematic analysis. 78 The thematic analysis followed six key steps, including (1) familiarisation with the data, (2) systematic coding, (3) grouping of codes to form potential themes, (4) reviewing themes and (5) formally naming and (6) defining themes. 78 All transcripts were independently coded by two members of the research team (AG and LP). Meaningful quotes from participants were extracted to highlight typical responses.

Methods

Statistical analysis

Methods

Systematic reviews suggest that financial incentive interventions have become increasingly popular in promoting health behaviour change. 97 Although the idea of using financial incentives to encourage healthier behaviours seems acceptable to the public in the UK context,98 the levels of incentives utilised in previous studies (i.e. the monetary amount thought necessary to trigger behaviour change) were not based on prior research and, consequently, led to uncertainty about the effectiveness and cost-effectiveness of financial incentives as an intervention tool. To investigate the optimal levels of financial incentives required to encourage changes in physical activity, all participants in our study were sent an online questionnaire at baseline to assess the average level of incentives that participants would be WTA to increase their physical activity overall as well as for increasing different types of physical activity. The values obtained in the preliminary contingent valuation analysis were later used to fix the level of the rewards available for earned ‘points’.

Although in its infancy, behavioural economics is helping our understanding of the relationship between behaviour change, time discounting and incentives, and is providing new insights that have sparked the interest of public health researchers and policy-makers. 99 The principle behind financial incentives is to capitalise on the notion of time preference (i.e. the tendency for an individual to place greater value on present benefits relative to future benefits). Time preference is typically elicited through a binary question that asks an individual to choose between an immediate, smaller reward and a delayed, larger reward. Individuals who choose the smaller, immediate reward are said to have a high time preference and, thus, it is hypothesised that they are likely to act on the present immediate benefits of a behaviour (unhealthy) at the expense of the perceived future benefits (of healthy behaviours). Systematic reviews have shown extensively that time preference plays an important role in shaping some unhealthy behaviours (e.g. smoking and obesity). 100,101

Two other key concepts possibly related to behaviour change are risk preference and loss aversion. The former refers to the participants’ willingness to take risks whereas the latter refers to the tendency of participants to prefer avoidance of losses to winning equivalent gains. A recent advance in the methodology of eliciting time preferences is the joint elicitation of risk and time preferences. Andersen et al. 102 have shown that when discount rates and risk preferences are elicited and analysed together the discount rates are significantly reduced and are more accurately defined. Major behavioural economic theories have suggested that people are more sensitive to losses than to gains. For instance, prospect theory has demonstrated that the utility foregone from a unit monetary loss is twice the utility gain from a unit monetary gain. 103 How loss aversion is related to unhealthy behaviours is rarely investigated. As a result, all participants remaining in our study were invited at the 6-month follow-up to participate in an incentive-compatible behavioural economic field experiment that was scheduled at their workplaces to elicit their personal behavioural economic traits (i.e. time discounting, risk preference and loss aversion).

Apart from the baseline contingent valuation survey that was necessary to inform plausible levels of incentives, the subsequent economic experiments were conducted on smaller subsamples of the population during follow-up (to keep the participant burden and costs to a minimum). This was planned as ancillary to the main trial analysis, with the primary purpose of shedding light on the possible behavioural economic moderators supplementary to the main mediation–moderation analysis conducted for the trial as a whole.

Statistical analysis

The statistical package Stata was used to carry out all analyses. For the contingent valuation, a conditional log-likelihood function for the set of DBDC responses was constructed and a maximum-likelihood estimation was used to predict the mean WTA. In the DCE analysis, the responses to a choice set were taken as a specific observation. A generalised multinomial logit model108 was used that assumed a specification that nests both scale heterogeneity and parameter heterogeneity. The participants who chose the opt-out options for all choice sets, indicating that they did not believe in the proposed hypothetical system of rewards, were excluded from the analysis. Interaction terms were used to explore whether or not preferences for attributes varied by the participants’ physical inactivity status at baseline. Next, we estimated a double-hurdle model109 that assumed that the decision to use the physical activity monitoring system and how long to use it for were independent decisions so that the determinants of the two decisions are allowed to differ. The estimation was accomplished by using the Stata module ‘craggit’. Missing data were excluded from the analysis. Furthermore, another double-hurdle model was estimated to investigate the determinants for the decision to increase physical activity and the amount by which physical activity had increased at the 6-month follow-up compared with the baseline (i.e. the mean number of steps per day at the 6-month follow-up was larger than the number of baseline mean steps per day), respectively. For the analysis of the behavioural economic experiments, we simultaneously elicited present bias, the discount rate, risk preference and loss aversion in an integrated framework using a hierarchical Bayesian methodology. 110 More specifically, a higher discount rate implies a higher level of impatience. And a present-bias parameter of significantly smaller than one implies a decreasing impatience over time (i.e. a participant has long-term goals that could benefit them in the future but no present actions will be taken to achieve these goals because of self-control problems). The risk aversion and loss aversion parameters take a value bigger than 0 and a higher value implies a higher degree of risk aversion and loss aversion, respectively. The estimated behavioural economic parameters were later entered into a double-hurdle model to investigate their influences on the decision to increase physical activity and the amount by which physical activity had increased at the 12-month follow-up compared with the 6-month follow-up, respectively.

Methods

Constructs hypothesised to mediate the initiation of physical activity behaviour were collected at baseline and at 4 weeks and included outcome expectations,55 physical activity self-efficacy,56 intention,61 planning,58 financial motivation,57 self-determined motivation (i.e. identified regulation, integrated regulation and intrinsic motivation)59,60 and social norms. 16

Constructs hypothesised to mediate the maintenance of physical activity behaviour were collected at baseline and at 6 months and included planning,58 self-determined motivation (i.e. identified regulation, integrated regulation and intrinsic motivation),59,60 habit,17 recovery and maintenance self-efficacy,20 outcome satisfaction,55,65 social norms16 and workplace norms. 16

Statistical analysis

All questionnaire items were scaled so that lower values indicated lower levels of the mediator/outcome. The mediation analyses are exploratory and should be interpreted with caution because of issues associated with multiple testing and because the study was powered for the primary physical activity outcome and not to detect change in mediators. The level of significance was a p-value of  < 0.05 for all analyses. Analyses were carried out using Stata.

Deviations and rationale

On 31 August 2015, the research team corresponded with the NIHR Public Health Research secretariat to indicate that the initial recruitment target was not met, explaining a variety of reasons why the original target proved to be problematic. The significant restructuring of a number of the public sector organisations that were to host the intervention had a severe impact on participant recruitment and the study team undertook various actions to mitigate the shortfall (e.g. by extending the number of sites and taking on advice from patient and public involvement members on our Trial Steering Committee regarding alternative recruitment methods within the host organisations).

Because recruitment numbers were lower than anticipated from our initial baseline data collection phase in Lisburn (from the restructured local authority and health-care trust), the intervention phase started later than scheduled, in May 2014. It was decided that recruitment would continue across other agreed sites in Belfast [Queen’s University Belfast (offering a pool of approximately 3000 staff) and the Stormont Civil Service Estate (offering a pool of approximately 3000 staff)] to attempt to meet the recruitment target and that the intervention would be implemented on a rolling basis thereafter. Therefore, randomisation and the intervention phases were implemented in July and August 2015 in the Queen’s University Belfast campus and Stormont Civil Service Estate, respectively.

Recruitment continued to be slower than anticipated, even with the addition of our new sites at Queen’s University Belfast and the Stormont Civil Service Estate. This was a result of unforeseen circumstances (austerity-related restructuring of local authorities and restructuring within the civil service) and, therefore, we sought approval to recruit from two final sites within the Belfast Health and Social Care Trust (Royal Victoria Hospital and Belfast City Hospital). Subsequently, we undertook an intense recruitment period in early September 2015. Participant recruitment was completed in January 2016 following the addition of a new worksite (n = 853 participants in total).

Professor Chris Patterson (study statistician) made a revised power calculation that took account of more recent literature on effect size estimates and the baseline data on cluster size, cluster variation and intraclass correlation coefficient. As reported from the previous Trial Steering Committees, the revised sample size requirement demonstrated a minimal impact on power (see Chapter 2, Sample, for further details).

Furthermore, to maximise trial retention, all participants received a £10 gift card at the 6- and 12-month stage for the full completion of study outcomes (funded from internal sources). The trial was originally designed to include the follow-up of participants up to 18 months post intervention. However, with unavoidable delays in initial recruitment attendant on the restructuring of the local authorities and in the light of the findings at the 6-month follow-up, it was agreed with NIHR that follow-up should stop at 12 months rather than at 18 months as was originally planned.

Project team meetings

From the outset of the trial, the research team has attended and fully engaged in the project team meetings with staff from the health trusts and organisations participating in the trial. This forum has maintained a schedule of twice-yearly meetings throughout the lifespan of the trial. The aim has been to help influence the research at an early stage of development to raise awareness of it and support the organisations’ involvement in the process.

Stakeholder members of the Trial Steering Committee

Alongside the project team meetings, key stakeholder (comprising staff from the health trusts and participating organisations) also contributed directly to the overall study evaluation as members of the Trial Steering Committee. Critically, this included contributing to the emerging interpretation of the findings and the development of the dissemination strategy.

Dissemination events

Alongside the dissemination of the key findings of this study at academic conferences, a number of non-academic dissemination activities were planned. Such activities have included the distribution of two annual study ezines to all host organisations, study participants and to the UK Clinical Research Collaboration Centre of Excellence (Northern Ireland) partners. Dr Aisling Gough (key co-ordinating postdoctoral research fellow for the study) attended a Northern Ireland Public Health Research Network workshop to update the network on the PAL scheme (February 2017). The offer of a health and well-being workshop and a PAL study debriefing was extended to the ‘champions’ or key contacts at each of the participating worksites. The research team also received interest from a number of organisations that expressed an interest in adopting the study’s bespoke web platform for promoting physical activity in the workplace, and the study team are currently in talks with the Health and Safety Executive of the South Eastern Health and Social Care Trust and Sport Northern Ireland regarding this. Talks have also been held with the Queen’s University Belfast Research and Enterprise team regarding the licensing of the web platform for such purposes. Furthermore, plans have been made to archive the results of dissemination activities on the British Science Association’s Collective Memory database (see http://collectivememory.britishscienceassociation.org/).

Process evaluation

The aim of the process evaluation was to examine (1) what the participants’ exposure was to the intervention; (2) the extent to which the intervention was implemented across the participating organisations; (3) how, for whom and under what circumstances the intervention could bring about behaviour change; (4) how, for whom and under what circumstances the intervention might maintain behaviour change; and (5) whether or not there were any unintended consequences of the intervention.

As described earlier in this chapter, the process evaluation involved in-depth engagement with all of the key stakeholders to ascertain their experiences of and perspectives on the programme. The qualitative insights gained from the participants and also from the employers are demonstrated in Chapter 4, Senior managers of participating employees.

Withdrawals

Participants noted as withdrawals were those who formally, via e-mail, by telephone or in person, expressed their wish to no longer take part in the trial. At 6 months post baseline, 154 participants (n = 71 participants from the intervention group) had withdrawn from the study. At 12 months post baseline, a further 45 participants had withdrawn (n = 22 participants from the intervention group). The stated reasons for withdrawals included lack of time and/or interest.

Lost to follow-up

Participants noted as ‘lost to follow-up’ were those who were unable to be contacted as a result of moving worksite location or those who did not reply to e-mail/telephone contact by the research team. At 6 months post baseline, 71 participants were lost to follow-up (n = 49 participants from the intervention group). Across the two groups, 69 participants were unable to be contacted and two participants had moved location. At 12 months post baseline, a further 47 participants (n = 30 participants from the intervention group) were lost to follow-up, of whom five had retired or were off sick, 39 had moved location and a further three were unable to be contacted by the research team.

Harms and adverse effects

There were no harms or adverse effects reported by any participant throughout the conduct of the trial.

Primary outcomes

At 6 months post baseline, the mean number of steps per day were 6990 (SD 3078) steps for the intervention group and 7576 (SD 3345) steps for the control group. There was a significant difference between the intervention and control groups in the mean number of steps per day (b = –336 steps, 95% CI –612 to –60 steps; p = 0.02) after adjustment for baseline values, randomisation stratum and season, and correction for cluster effects (Table 5). Possible reasons for this result are explored in Sensitivity analysis.

At 6 months post baseline, there was a significant difference between the intervention group compared with the control group for minutes per week of self-reported work-related physical activity (b = –33.3 steps, 95% CI –65.44 to –1.24 steps; p = 0.04). There was no significant difference between the groups for minutes per week of self-reported moderate- to vigorous-intensity physical activity (b = 4.12 steps, 95% CI –47.07 to 55.31 steps; p = 0.88).

Exploratory subgroup analyses

Moderation analyses in Table 7 showed that perceptions of the availability of physical activity opportunities in the workplace environment moderated the relationship between group assignment and 6-month mean number of steps per day, as indicated by a significant coefficient for the interaction term (b = –280 steps; 95% CI –538 to –23 steps; p = 0.03). This indicated that the negative effect of assignment to the intervention group on 6-month mean number of steps per day was greater at higher levels of workplace environment availability. Therefore, there was a greater decrease in the mean number of steps per day (between the baseline and the 6-month follow-up) for the intervention group participants than control group participants when perceptions of the availability of physical activity opportunities in the workplace environment were higher. It should be noted that, owing to issues with multiple testing, this result should be interpreted with caution. The results of all other moderation analyses were non-significant.

Primary outcomes at 6-month follow-up (12 months post baseline)

At 12 months, the mean number of steps per day were 7790 (SD 3462) steps for the intervention group and 8203 (SD 3,401) steps for the control group. There was a non-significant difference between the intervention and control groups in the number of steps per day (b = –570 steps, 95% CI –1267 to 127 steps; p = 0.11) after adjustment for baseline values, randomisation stratum and season, and correction for cluster effects (Table 8).

At 12 months, there were no significant differences between the intervention and control groups for minutes per week of self-reported work-related physical activity (b = 7.00, 95% CI –12.55 to 26.56 minutes; p = 0.48) and for minutes per week of self-reported moderate- to vigorous-intensity physical activity (b = 77.04, 95% CI –7.94 to 162.0 minutes; p = 0.08) (see Table 8).

Exploratory subgroup analyses

Moderation analyses (see Table 7) showed that perceptions of the safety of the workplace environment moderated the relationship between group assignment and the 12-month mean number of steps per day, as indicated by a significant coefficient for the interaction term (b = 294.2 steps, 95% CI 55 to 534 steps; p = 0.02). This indicated that the negative effect of assignment to the intervention group on the 12-month mean number of steps per day was greater at lower levels of workplace environment safety. Therefore, there was a greater decrease in the mean number of steps per day (between the baseline and the 12-month follow-up) for intervention group participants compared with control group participants when perceptions of the safety of the workplace environment were lower. Moderation analyses also showed that the ‘Time discount 3’ variable (high discount rate vs. low discount rate) moderated the relationship between group assignment and the 12-month mean number of steps per day, as indicated by a significant coefficient for the interaction term (b = –1216, 95% CI –2226 to –205; p = 0.02). This indicated that the negative effect of assignment to the intervention group on the 12-month mean number of steps per day was greater when discount rates were higher. Therefore, there was a greater decrease in the mean number of steps per day (between the baseline and the 12-month follow-up) for intervention group participants than with control group participants when participants’ discount rates were higher (i.e. participants prefer an immediate reward with a lower monetary value rather than having to wait to get a reward with a slightly higher monetary value tomorrow). Owing to issues with multiple testing, these results should be interpreted with caution. The results of all other moderation analyses were non-significant.

Participation and reach

A total of 9031 participants were invited to participate across four main sites (i.e. Lisburn, Stormont Estate, Queen’s University Belfast and Belfast City Hospital), as shown in Figure 2. In total, 853 participants were recruited and randomised. The demographic characteristics of the study population are shown in Table 4.

Participants

In total, nine focus groups (FGs) were conducted at the end of the intervention period across the study sites: Lisburn (n = 3 FGs; n = 21 participants), Stormont Estate (n = 2 FGs; n = 12 participants), Queen’s University Belfast (n = 2 FGs; n = 16 participants) and Belfast City Hospital (n = 2 FGs; n = 16 participants). Sixty six participants took part in the FG discussions [n = 52, 78% female; mean age 43.8 (SD 9.0) years]. At 12 months, 17 participants who had taken part in the 6-month FGs also took part in the 12-month FG discussions [n = 9, 53% female; mean age 46.1 (SD 9.3) years].

The following coding was applied to quotes to distinguish between the different FGs and semistructured interviews undertaken: FG denotes the focus group sessions (1–9) at 6 months (e.g. FG1, FG2); FG12 denotes the focus group session (12.1–12.6) at 12 months (e.g. FG12.1, FG12.2); MR denotes male respondent; FR denotes female respondent; R denotes retailer; SM denotes senior manager.

General perceptions of the intervention

The overall feedback from participants about the PAL scheme was positive. Many participants cited how the scheme had helped them engage in more physical activity and develop habits that could be maintained once the intervention had ended:

It’s been a very positive thing to get people out walking and to encourage them to walk more, with all the wee perks and to do a bit extra every day.

FG3, FR

So it’s been a brilliant success to get me walking and it’s now part of my daily routine.

FG8, FR

Perceptions of intervention components and behaviour change techniques

The key themes in relation to participants’ thoughts on the different intervention components of the PAL scheme for behaviour change are summarised in Table 10. These include the use of incentives, sensors, maps, feedback, fobs, e-mail and social support.

With regard to the sensors, participants reported some issues that affected the day-to-day running of the intervention, including issues with the accuracy, placement and restrictiveness of the sensors (see Table 10). The participants had mixed opinions on a number of website components, including the maps and e-mail prompts. The feedback element of the website was a positive feature for many, with participants highlighting that the visual feedback often prompted them to be more active.

A number of themes emerged in relation to rewards (retail vouchers) offered to participants in exchange for points they had accumulated during their participation in the PAL scheme. Some participants felt that the vouchers did not provide a sufficient reward for the effort and activity undertaken:

You’ll get incentives joining gyms anyway, so there wasn’t really any incentive in the incentives, really, unless it was a free cup of tea. Walking ten miles for a cup of tea. It needed to be more; the incentives just weren’t enough for me.

FG5, FR

Meanwhile, other participants felt that the vouchers were limited:

The vouchers were an issue, when you go to get them there wasn’t the ones you wanted, such as Argento. Just there was only so many available.

FG3, FR

Participants noted that the issues around the format and availability of the vouchers had an impact on their interest in other elements of the intervention:

And I found then that I lost interest because the points weren’t really going anywhere. And then whatever points I had, the incentive wasn’t good enough for the points I had.

FG5, FR

Senior managers of participating employees

Semistructured interviews were also undertaken with senior managers of six participating employers (n = 7 managers) and four participating retailers. Feedback about the intervention was positive from employers, with themes emerging from the interviews about employee benefits, both by means of getting active (a focus on physical and mental health) and relating to workplace productivity. Senior managers’ perceptions of the intervention were aligned with participants’ perceptions, regarding time as a key barrier to physical activity in the workplace.

Retailers

Semistructured interviews were undertaken with four participating retailers. A number of key themes emerged with regard to why retailers became involved with the PAL scheme, including benefits for their business and the broader benefits of the scheme for participants. Of the participating retailers interviewed, all cited the potential benefits being involved in the PAL scheme could have for their business.

Intervention costs

The intervention costs are detailed in Table 11. Intervention costs included the retail vouchers, sensors and key fobs, electrical engineering staff and travel for maintaining the sensors, and a computer scientist for monitoring and managing the data from the sensors and maintaining the website. The cost for the placement of the sensors on the lampposts was not included as it was refundable by local government. The vouchers cost £21,000 in total. These voucher costs could be greatly reduced if there was more meaningful engagement with the local authority/city management groups to facilitate better deals with the retailers. Therefore, the costs of vouchers were varied in sensitivity analysis. The lifespan of the sensors was estimated to be 2 years because the screw bosses were made of plastic, which limits the number of battery replacements. With upgraded screw bosses, the lifespan can be prolonged to over 5 years. The battery used in the key fobs cannot be replaced so the lifespan for the current key fobs in the trial is the life of the battery (i.e. 9 months). This short lifespan will be extended to 5 years with a battery replaceable design. Therefore, in sensitivity analyses, the costs for sensors with upgraded screw bosses and battery-replaceable key fobs are considered to explore the influence of upgraded technology on the cost-effectiveness of the intervention. The cost per participant for each disaggregable item is shown in Figure 3, from which the retail voucher and sensor together accounted for > 90% of the total cost.

FIGURE 3.

Cost of the PAL scheme over 6 months (per participant).

Missing data

The number and percentage of missing data for each collected NHS resource use item and EQ-5D-5L questions at baseline and at the 6-month assessment point are shown in Table 12 for intervention and control groups, respectively. Baseline for the resource use refers to the 6-month period before randomisation completed by participants retrospectively at baseline. All of the missing data for the EQ-5D-5L responses were caused by whole questionnaires missing, which can be seen from the uniform proportion of missing data for the five dimensions. A similar proportion of missing data was observed across the randomised groups and for both cost and EQ-5D-5L data. Approximately 15% of data were missing at baseline and 40% of data were missing at the 6-month follow-up.

Resource use and costs

Table 13 reports the use of each resource item (mean number of visits, SD, median, minimum and maximum) accrued over the 6-month primary follow-up period in the intervention and the control group. Overall, there were no statistically significant differences in the use of each resource item between the groups.

Table 14 shows other services used by the participants in each group. Other services and medications are not included in the total resource use owing to the similar amount of service use between groups, their randomness (i.e. they are most likely unrelated to the intervention) and the difficulty in matching unit cost with the free-text inconsistent reporting.

Table 15 compares the cost of service use between the groups with four regression approaches, as specified in Chapter 2, Health economics, Methods. There is no statistically significant difference in resource use between the groups. Despite this, the decrease of resource use in the intervention group compared with the control group was larger (approximately £60) with the gamma distribution adjusting for covariates than the OLS (approximately £15) unadjusted model, in both complete case and multiply imputed data.

Health-related quality of life

The EQ-5D-5L health utility values and visual analogue scale (VAS) for each group at baseline and 6 months are shown in Table 16 (complete case) and Table 17 (after imputation). The completeness at the 6-month follow-up was 57% (262/457) for the intervention group and 52.3% (239/457) for the control group. This was lower than the rate of completeness at baseline, which was 85.3% (390/457) and 74.2% (339/457) for the intervention and control group, respectively. A total of 45.0% (328/729) of participants recorded a state of ‘full health’ at baseline and 29.1% (146/501) of participants recorded the state of ‘full health’ at 6 months. The utility values for both groups declined over the 6 months from approximately 0.89 to 0.83. There was no statistically significant difference between groups at both baseline and 6 months.

Cost–utility base-case analysis

The cost-effectiveness results for the intervention are presented in Table 18. Overall, the intervention was approximately £25 more costly, but had no effect on QALYs. The average cost per participant was £253.5 (95% CI £188.4 to £318.6) in the intervention group and £227.64 (95% CI £170.9 to £284.4) in the control group. Mean QALYs accrued over the 6-month trial period were 0.4157 (95% CI 0.4077 to 0.4238) for the intervention group and 0.4158 (95% CI 0.4057 to 0.4260) for the control group, leading to a 0.0000891 (95% CI –0.008 to 0.008) lower QALY gain in the intervention group than in the control group.

The cost-effectiveness plane for the base-case analysis is shown in Figure 4. The dyads come from the 1000 bootstrap iterations. The x-axis represents the bootstrapped incremental QALYs between the randomisation groups and the y-axis represents the incremental costs. The bootstrapped pairs of cost difference and QALY difference were across the four quadrants, indicating greater uncertainty around the estimate of ICER. The bootstrapped pairs were almost evenly (47% vs. 53%) distributed across the y-axis in the first and second quadrant, caused by the close to zero mean incremental QALY. The existence of dyads in the third and fourth quadrants indicates that there was a small possibility (i.e. 13.4%) that the intervention is cost saving. The proportion of the simulated pairs distributed in each quadrant is presented in Table 19.

FIGURE 4.

Cost-effectiveness plane representing 1000 bootstrapped cost difference and QALY difference pairs.

The CEAC for the base-case analysis is shown in Figure 5. The probability that the intervention was cost-effective at the £30,000 threshold was 34.6%. This corresponds to the proportion of dyads on the right side of the sand colour line in the first, third and fourth quadrant of the cost-effectiveness plane in Figure 4.

FIGURE 5.

Cost-effectiveness acceptability curve.

Sensitivity analysis

Results for the sensitivity analysis are presented in Table 20.

Cost–benefit analysis employing net-cost model

It was reported earlier in Table 5 that the absolute absenteeism hours over a 4-week period for the intervention and control groups was 4.04 (66.11) and 7.01 (51.40), respectively (p = 0.62). The difference was estimated to be 2.59 hours over a 4-week period after adjusting for stratum, season and cluster (p = 0.62). This equates to 15.54 hours pro rata for a 6-month time period. After attaching the hourly salary values, the avoided cost of absenteeism and the net cost of intervention for each representative salary grade is shown in Table 21. This estimated cost saving for the employers ranged from £66 to £735 depending on the wage rate employed.

Given the difference in absolute absenteeism was not statistically significant, Figure 6 presents the uncertainty of the potential economic benefit from the employer’s perspective for each representative salary grade. At current intervention cost (= £55.68), the probability that the PAL intervention is cost saving from employer’s perspective is 64% for the high-salary group, 62% for the middle-salary group, and 57% for the low-salary group.

FIGURE 6.

Uncertainty surrounding the potential economic benefit from the employer’s perspective: net cost model result.

Initiation of physical activity

At 4 weeks post baseline, there were significant differences between the intervention and control groups for intentions (b = 0.29, SE = 0.13; p = 0.02), social norms (b = 0.23, SE = 0.08; p< 0.01), identified regulation (b = 0.14, SE = 0.06; p = 0.01), integrated regulation (b = 0.23, SE = 0.07; p < 0.01), and intrinsic motivation (b = 0.18, SE = 0.06; p < 0.01). There were non-significant between-group differences for PA self-efficacy (p = 0.31), outcome expectations (p = 0.58), financial motivation (p = 0.46) and planning (p = 0.75) (see Table 27). Thus, the intervention group participants showed increases above the control group participants in intentions, social norms, identified regulation, integrated regulation and intrinsic motivation between baseline and 4 weeks, approximately equivalent to 0.29, 0.23, 0.14, 0.23 and 0.18 points, respectively, on their original scale. Sensitivity analyses conducted using logistic regressions showed that the significance level changed for some mediators. Specifically, logistic regressions indicated that there was a significant between-group difference for financial motivation [exp(b) = 0.37, SE = 0.15; p = 0.02] and no significant between-group difference for social norms (p = 0.11).

Maintenance of physical activity

At 6 months, there were significant differences between the intervention and control groups for identified regulation (b = 0.11, SE = 0.05; p = 0.02), integrated regulation (b = 0.26, SE = 0.08; p < 0.01), intrinsic motivation (b = 0.17, SE = 0.06; p < 0.01), and habit (b = 0.48, SE = 0.12; p < 0.01). There were non-significant between-group differences for planning (p = 0.18), social norms (p = 0.42), workplace norms (p = 0.06), recovery self-efficacy (p = 0.80), maintenance self-efficacy (p = 0.80), and outcome satisfaction (p = 0.56) (see Table 27). Thus, the intervention group participants showed increases above the control group participants in identified regulation, integrated regulation, intrinsic motivation and habit between baseline and 6 months, approximately equivalent to 0.11, 0.26, 0.17 and 0.48 points, respectively, on their original scale. Sensitivity analyses conducted using logistic regressions showed that the significance level changed for some mediators. Specifically, logistic regressions indicated that there was a significant between-group difference for workplace norms [exp(b) = 0.37, SE = 0.18; p < 0.05].

Single mediator models

Table 28 shows the results of single mediator models based on the product-of-coefficients approach with CIs for the indirect effect formed using the bias-corrected bootstrap, and adjustment for randomisation strata, season, baseline values of the mediator and baseline pedometer steps per day, with SEs and p-values corrected for clustering. An explanatory diagram for single mediator models is presented in Figure 7.

FIGURE 7.

Explanatory diagram for single mediator models showing an indirect effect of group assignment on the 6-month mean number of steps per day through a 4-week/6-month mediator.

Single moderated-mediation models

Single mediator models including 6-month scores of integrated regulation, intrinsic motivation, and habit (with 6-month mean number of steps per day as the dependent variable) were tested for moderation of the a-path and b-path respectively by perceptions of workplace environment availability. Significant a-path moderation was found for the single mediator model with integrated regulation as the mediator (b = –0.10, SE = 0.05; p < 0.05), and the single mediator model with habit as the mediator (b = –0.17, SE = 0.05; p < 0.01). There was a non-significant moderator-by-group interaction coefficient (i.e. a-path moderation) for the single mediator model with intrinsic motivation as the mediator (p = 0.19). There was a non-significant moderator-by-mediator interaction coefficient (b-path moderation) for the single mediator models with integrated regulation (p = 0.36), intrinsic motivation (p = 0.06) and habit (p = 0.44) as the mediator. SRMR values ranged from 0.00 to 0.05, indicating a good fit,122 and CD values ranged from 0.66 to 0.77. Thus the apparent mitigation of the intervention effect, by virtue of higher levels of intrinsic motivation and habit was dependent on participant’s perception of workplace environment availability for PA. It is unclear if those with higher levels of such constructs benefited more with regard to PA because they had more PA opportunities near their workplace or if this was as a result of their perception of such opportunities. This is representative of an ‘intervention challenge’, meaning future intervention studies should give consideration to how to increase levels of integrated regulation and habit for participants with higher perceptions of the availability of PA opportunities in the workplace environment.

Engagement and non-use attrition

Table 29 shows descriptive statistics for 6-month engagement and non-use attrition. Engagement variables that were significant predictors of 6-month mean number of steps per day in univariable analyses were included in a multivariable model, which showed that frequency of hits on the monitoring and feedback component of the website across the 6-month intervention period (b = 50.2, SE = 24.5; p = 0.04). The percentage of earned points redeemed across the 6-month intervention period (b = 9.1, SE = 3.3; p < 0.01) were positively related to the 6-month mean number of steps per day, whereas frequency of hits on the discussion forum component of the website across the 6-month intervention period (b = –69.3, SE = 26.6; p < 0.01) was negatively related to the 6-month mean number of steps per day (Table 30). None of the other variables was significant predictors of the 6-month mean number of steps per day in univariable analyses.

The percentage of days (Table 31) that participants were recorded walking for at least 10 minutes captured via the PA monitoring system was a significant predictor of integrated regulation (b = 0.007, SE = 0.002; p < 0.01), intrinsic motivation (b = 0.003, SE = 0.001; p = 0.04), habit (b = 0.007, SE = 0.003; p = 0.01) and outcome satisfaction (b = 0.005, SE = 0.02; p = 0.01). The percentage of weeks that participants logged onto the website at least once was a significant predictor of integrated regulation (b = 0.004, SE 0.002; p = 0.02), intrinsic motivation (b = 0.002, SE 0.0010; p = 0.04), habit (b = 0.004, SE 0.002; p = 0.03) and outcome satisfaction (b = 0.004, SE 0.001; p = 0.01). The percentage of earned points redeemed was a significant predictor of 6-month social norms (b = 0.003, SE 0.001; p = 0.01).

Frequency of hits (Table 32) on the monitoring and feedback component of the website was a significant predictor of 6-month planning (b = –0.04, SE 0.01; p < 0.01), identified regulation (b = 0.02, SE 0.01; p = 0.04) and integrated regulation (b = 0.03, SE 0.01; p = 0.02). Frequency of hits on PA health information was a significant predictor of integrated regulation (b = 0.11, SE 0.05; p = 0.03). Frequency of hits on discussion forums was a significant predictor of 6-month social norms (b = –0.03, SE 0.01; p = 0.03) and integrated regulation (b = –0.02, SE 0.01; p = 0.03). Number of sections accessed on the website at least once was a significant predictor of identified regulation (b = 0.07, SE 0.02; p < 0.01) and integrated regulation (b = 0.09, SE 0.03; p < 0.01).

The estimated median lifetime usage (i.e. after which 50% of participants stopped use) was 25.5 days for use of the PA monitoring system to record daily activity and 12.5 days for use of the website. Non-use attrition of the PA monitoring system to record daily activity occurred for 89% of participants (n = 375), and website non-use attrition occurred for 96% of participants (n = 403). The results of univariable and multivariable Cox regression analyses examining baseline measures of sociodemographic variables, psychosocial variables, environmental variables and PA as predictors of non-use attrition are presented separately for use of the PA monitoring system to record daily activity and use of the website. The multivariable analysis for use of the PA monitoring system to record daily activity showed that having higher levels of identified regulation at baseline (hazard ratio 0.88, 95% CI 0.81 to 0.97), having higher levels of recovery self-efficacy at baseline (hazard ratio 0.88, 95% CI 0.80 to 0.98), or having a lower perception of the safety of the workplace environment for PA at baseline (hazard ratio 1.06, 95% CI 1.02 to 1.11) reduced the risk of attrition. The multivariable analysis for website use showed that being older (hazard ratio 0.99, 95% CI 0.98 to 1.00), unmarried/not co-habiting (hazard ratio 0.79, 95% CI 0.63 to 1.00), having higher values on the EuroQol (EQ-5D) weighted health index (hazard ratio 0.32, 95% CI 0.12 to 0.86), having higher levels of financial motivation at baseline (hazard ratio 0.93, 95% CI 0.87 to 0.99), having a lower perception of the attractiveness of the workplace environment for PA at baseline (hazard ratio 1.06, 95% CI 1.01 to 1.11), or having a higher perception of the availability of PA opportunities in the workplace environment at baseline (hazard ratio 0.94, 95% CI 0.90 to 0.98) reduced the risk of attrition.

Sensitivity analysis results

The sensitivity analysis showed that when an alternative business model was used or the cost of sensors was reduced, the incremental cost of the intervention became negative. The probability of the intervention being cost-effective increased to around 60%, but this may not be considered as a dominant scenario given the ‘benefit’ of the intervention on utility was still negative and unclear.

A potential ceiling effect of EQ-5D-5L was observed in this population, with > 40% and > 30% of participants responding ‘full health’ at baseline and 6 months, respectively. This may limit the sensitivity of this instrument for assessing the potential benefit of interventions such as this in public health settings, as there is little room for utility values to get higher in a generally healthy population.

Cost–benefit analysis/net cost model results

Adopting an ‘employee’ perspective, there was a net cost saving for the employers due to reduced absenteeism in the intervention group. This net cost saving ranged from £66 to £735 depending on the wage rate employed. By exploring the uncertainty surrounding this net cost saving estimate, we determined that the probability that the intervention is cost saving at current intervention cost (£55.68) is highest (= 64%) in the high-salary group and lowest (= 57%) in the low-salary group. If the intervention cost reduces in the future, with the aforementioned alternative business model or upgraded technology for the sensors, the probability to be cost saving will increase and be capped at 65.4% (the intercept of the three lines with y-axis in Figure 6). This capped value is owing to the great uncertainty (p = 0.62) surrounding the benefit in terms of the difference in absolute absenteeism rates between the groups.

A number of issues arising from the study merit further research from a health economic perspective. Although we used utilities directly elicited from participants using the EQ-5D, we observed small positive benefits in terms of well-being scores in the intervention group, which were at odds with the negative impact on the primary outcome. Studies with longer-term follow-up could usefully explore further how to aggregate or reconcile the various health and productivity effects of PA programmes in workplace settings, especially because, in the longer term, they may have an impact on the individual and the employer in different ways. To date, there have been few attempts to model the impact of workplace wellness programmes that are based on financial incentives. Basu and Kiernan150 constructed a novel modelling framework that identified how to optimise the marginal return on investment from workplace programmes that were based on incentivizing behaviour change (including those targeting PA). By integrating ‘demand curves’ that captured individual differences in response to any given incentive level with employee demographic and risk factor data (from the USA) they demonstrated that an employer (with ≈3000 employees) would see optimal return on investment by offering incentives of US$367 per year (i.e. somewhat larger than in the PAL scheme) but that this return on investment (ROI) was very sensitive to self-selection (if already active participants enrolled). Even with the optimally designed programmes, the actual degree of behaviour change varied widely, with most participants still undertaking low to moderate levels of PA despite the incentive, and > 70% of participants not engaging in PA on a weekly basis (in other words, they would be intermittent participants). Thus, it is right that future research in this area take a broad perspective and attempts to integrate individual and employer perspectives.

Perception of the availability of physical activity opportunities in the workplace environment

Perception of availability of PA opportunities in the workplace environment was a significant moderator of the relationship between group assignment and PA change between baseline and 6 months. Specifically, higher levels of perceived availability of PA in the workplace environment were associated with a greater decrease in PA between baseline and 6 months for an intervention group participant than for a control participant. Perceptions of workplace environment safety was a significant moderator of the relationship between group assignment and PA change between baseline and 12 months, with lower levels of perceived safety associated with a greater decrease in PA between baseline and 12 months for an intervention group participant compared with a control group participant. Previous studies have also supported the idea that perceptions of the environment are important moderators of PA behaviour change. 167,168 However, the direction of the moderating effect of perceptions of availability of PA in the workplace environment on PA at 6 months may appear counterintuitive (i.e. the intervention’s negative effect on PA was stronger when perceptions of availability were high). Perhaps participants who perceived that there were plentiful opportunities for PA around their workplace felt less reliant on the intervention to identify opportunities for PA or were less willing to confine their PA within the limited geographical range of the study’s sensors. This result merits exploration in future research.

Participant discount rate

Participant discount rate was a significant moderator of the relationship between group assignment and PA change between baseline and 12 months, with higher (vs. lower) discount rates associated with a greater decrease in PA between baseline and 12 months for an intervention group participant compared with a control participant. Thus, when participants were more impatient (i.e. preferred an immediate reward with a slightly lower monetary value to a reward with a short delay and a slightly higher monetary value) there was a greater decrease in PA for intervention group participants compared with control participants. This pattern of results may make intuitive sense for the PA outcomes at 12 months (if, for a subgroup at least, increases in intrinsic motivation were not sufficient to compensate over the longer follow-up period when the extrinsic rewards had been removed after the 6-month intervention) but cannot explain the decline in PA at 6 months (the primary outcome).

Financial incentive and motivation

Self-determination theory (SDT) posits that increasing intrinsic motivation (i.e. based on an individual’s own values) leads to more sustainable behaviour change than can be achieved by inducing extrinsic motivation. 169 The finding of increases in intrinsic forms of motivation (i.e. identified regulation, integrated regulation and intrinsic motivation) for intervention group participants compared with control group participants at 4 weeks and 6 months, and no intervention effect for financial (extrinsic) motivation at 4 weeks, has important implications for some of the main contentious issues highlighted in the literature on the use of financial incentives for achieving behaviour change. First, extrinsic (e.g. financial) rewards are not thought to produce long-term behavioural changes because withdrawal of the reward causes the behaviour to be extinguished. 170 Second, the use of financial incentives has been thought to have a ‘crowding out’ effect on intrinsic motivation for behaviours that are already internalised. 169 For example, the qualitative data from our study showed that some participants, who were already physically active, reported that they joined the scheme to receive rewards for behaviour that they already engaged in. Based on this argument, we may have expected to observe decreases in intrinsic motivation for intervention group participants versus control group participants. Conversely, the review conducted by Promberger and Marteau169 suggested that there was no current evidence to support the notion that extrinsic rewards or incentives undermine any intrinsic motivation in relation to health-related behaviours. In line with the findings of that review, our results suggest that using financial incentives in a behaviour change intervention does not necessarily diminish intrinsic motivation and can actually increase it if rewards are delivered as part of a complex behaviour change intervention with multiple components. 32

The consensus is that financial incentives may be useful in acting as a ‘kick-start’ to PA but are less useful for promoting long-term behaviour change. The qualitative work in our study would attest to this, with some participants commenting on the positive aspects of getting people out walking, and to do a little extra every day. Under Learning Theory, previous behaviour is predictive of subsequent behaviour. 46,171 This may be explained by the persistence of factors (e.g. thoughts or rewards) that cause the behaviour to be repeated, or by the formation and strengthening of habits (i.e. learned behavioural responses to certain stimuli and repetition in similar contexts). 162 Thorndike’s Law of Exercise states that the strength of the connection between a behavioural response and a given situation is directly related to the number of times the connection has occurred previously and to the potency and duration of those connections. 172 By the use of contingent reinforcement it was hypothesised that repetition of the behaviour-reward cycle would lead to workplace PA becoming a ‘learned’ behaviour. According to SDT, external regulation occurs when the individual engages in the behaviour to gain reward (e.g. financial incentives) or to avoid punishment by others. 173 However, the reward offered must be deemed worthwhile by them personally. For example, previous studies have found that monetary value138 or type (e.g. individual versus group based)174 of the reward can impact its effectiveness for behaviour change. There was no overall effect of group assignment on changes in financial motivation for participants in this study and focus group interviews indicated that, for some participants at least, vouchers failed to live up to expectations (e.g. insufficient monetary value, lacking in variety, undesirable goods or services, inconvenient expiration dates, or that desirable vouchers ‘ran out’). Therefore, the incentive may not have been attractive enough to entice participants to start (and by repetition, to ‘learn’) PA behaviour in the first place. This is despite the fact that these aspects of the study were designed using ‘best practice’ of pre-intervention development focus groups with participants to discuss all important features of the intervention (e.g. types of rewards, appropriate level of rewards, and placement of sensors).

Intrinsic motivation (integrated regulation and intrinsic motivation) and physical activity

The results of mediation analyses suggested that intrinsic forms of motivation (i.e. integrated regulation and intrinsic motivation) were important mediators of the relationship between group assignment and PA behaviour change at 6 months. SDT, for example, suggests that individuals experience greater vitality, self-motivation and well-being when the basic psychological needs for autonomy, competence and relatedness to others are fulfilled within the social context. 175 Focus groups highlighted that participants wanted more options with respect to rewards, and how and when they engaged in PA. Perhaps participants did not perceive a high degree of autonomy support (i.e. opportunities for choice with minimal use of pressure or demands)176 in the scheme to fully allow internalisation of PA behaviour.

Habit formation and physical activity

Mediation analyses indicated that the strongest indirect effects of group assignment on 6-month PA occurred through habit formation. The emphasis placed on the intervention’s self-regulation techniques supports the observation that people develop habits after a period of successful self-regulation. The process of habit formation incorporates an exchange between automatic and self-regulatory processes within the context of a range of external and internal stimuli. 164 Habits are formed when the behaviour becomes regular, automatic and ingrained in the self-concept. 17,159 Prompts in the social (e.g. e-mail reminders sent by the study team and having work colleagues participating in the intervention) and physical (e.g. seeing sensors surrounding the workplace) environments could have aided habit formation. Moreover, focus groups at 12 months post baseline sought to establish if participants were still engaging with PA and, if so, for what reason. The underlying rationale was to establish if behaviour change maintenance had occurred. Many of those who participated in the focus group discussion at 12 months commented about how they had continued to incorporate PA into their working day, from utilising the walking routes provided by the PAL scheme, to walking to meetings in short proximity rather than driving and generally looking for new ways to fit PA into their day. This is also in line with several habit theories,177–179 which hypothesise that environmental cues associated with prior and new behaviours determine whether or not the new behaviour is maintained. Moderation analyses also indicated that participants’ perceptions of the environment had an impact on the effectiveness of the intervention in line with theory that suggests that supportive social and physical environments facilitate behaviour change maintenance by lowering the opportunity cost of the behaviour. 164 When participants utilised the scheme’s self-regulatory, social and environmental aspects to increase their PA habits, this seemed to mitigate the overall negative impact on PA.

Discount rates and intervention effectiveness

Moderation analyses also indicated that participants’ derived discount rates had an impact on the ‘effectiveness’ of the intervention, with higher discount rates (i.e. higher levels of impatience) associated with greater decreases in the primary outcome for intervention group participants. This is line with the findings of the study’s additional behavioural economics analyses (see Key findings) and also with the qualitative work from which we noted how vexed some participants were on not achieving enough points to redeem a certain voucher, or that a priority voucher had been time expired (and withdrawn from offer) before they had collected enough points and so they subsequently ‘gave up’. Although we cannot infer causality, these associations suggest that when designing financial incentive based interventions, future studies should consider using information on participants’ discount functions to tailor incentive structures (e.g. timing, contingency and value) in order to achieve behavioural initiation and maintenance. 180 Other intervention strategies (e.g. goal-setting,181 provision of health information182) could also be usefully targeted depending on information available on individual participants’ time preferences.

The concept of adaptive versus static goals has also been explored in recent PA literature, whereby Zhou et al. 183 employed automated machine learning-based personalised daily step goals delivered via a mobile phone app. Their study sought to differentiate PA behaviours in those who received personalised daily step goals compared with a control group who had a constant step goal of 10,000 steps. Although the PAL scheme incorporated personalisation via the website where users could track and monitor progress, a more advanced method for deriving adaptive goals might be beneficial for PA. Adams et al. ,184 for example, also found that dynamic adaptive goal-setting may outperform static approaches for goal-setting for PA (steps), with slower decreases in PA steps from baseline to intervention end. The use of adaptive goals may be particularly useful in a workplace setting. Where some employees may be able to incorporate PA into their commute to work, others – as we identified from the qualitative work – may find the workplace to be their only opportunity to be physically active. Recent research on goals does not generally account for the potential time discounting confound. ‘Nearer’ outcomes are more motivating, and are proposed to explain differences in motivation over time in human behaviour as well. 185 One interpretation is that more temporally distant rewards are less motivating specifically because delayed rewards are discounted (and potentially also riskier). In this view, goal-gradient behaviour (exerting more effort for sooner rewards, depending on how far away you are from your goal) is another source of evidence for time discounting. Urminsky et al. 186 reported discount rates that were lower when separately accounting for goal gradient effects. With this in mind, the effects of adaptive goals, attendant on differing feedback schedules, could potentially be more or less motivating in people with different discount rates and may cater for patterns of behaviour resulting from changes in day-to-day circumstances. This could usefully be the subject of future research.

Physical activity self–efficacy

Although constructs such as PA self-efficacy are frequently shown to be important mediators or predictors of PA behaviour187–192 PA self-efficacy and phase-specific self-efficacy (i.e. maintenance and recovery self-efficacy) were not shown to be significant mediators in the present analysis. There is emerging literature disputing the causal role of self-efficacy in behaviour change193 and several systematic reviews show that the supporting evidence is lukewarm at best. Rhodes and Pfaeffli194 found limited evidence for self-efficacy and outcome expectations, but stronger support for self-regulatory processes and social support as mediators of behaviour change (focusing on initiation) in PA intervention studies targeting adult non-clinical populations. Notably, although self-regulation constructs have not been examined as putative mediators in this study, the qualitative analysis revealed that participants appreciated the self-regulation (e.g. self-monitoring, planning, goal-setting) and social aspects of the PAL scheme. Another review conducted by Teixeira et al. 158 revealed that PA self-efficacy was related to short-term PA outcomes (i.e. at < 12 months) in 10 out of 15 tests (67%), while outcome expectations were not related to PA outcomes in one test. Importantly, these results did not rely on formal mediation tests, highlighting the need for robust mediation analysis in PA intervention studies.

Engagement and non-use attrition

The results of GLS regressions showed that engagement in specific components of the intervention (i.e. the frequency of using the website self-monitoring and feedback component, the percentage of earned points redeemed, and the frequency of using website discussion forums) was significantly related to PA at 6 months. These findings suggest that when participants focused more on the intervention’s positive aspects (e.g. self-monitoring), this mitigated the intervention’s overall negative impact on PA behaviour. Previous research also shows that self-regulation techniques and self-monitoring are useful strategies for changing PA behaviour. 2,38,194–198 Redeeming a higher proportion of earned points for rewards was also associated with slightly higher PA levels at 6 months. Therefore, it appears that when participants found the financial incentives desirable and redeemed their accumulated points, this also mitigated the overall negative impact of the scheme on PA behaviour. Previous studies have also shown that in order for financial incentives to be successful in inducing behaviour change, the reward on offer must be deemed to be valuable to the individual participant. For example, monetary value138 or type (e.g. individual vs. group based)174 of reward can have an impact on effectiveness for behaviour change, and this was alluded to in feedback from focus groups. Examination of the relationship between engagement in different aspects of the intervention and PA at 6 months showed that a higher frequency of accessing the discussion forum component of the website was associated with a significant decline in PA at 6 months. Discussion forums were included on the PAL scheme website as a means of providing social support for behaviour change. For example, it was expected that participants would use these forums to contact researchers and interact with other participants to support behaviour change. However, this component of the scheme was primarily used by participants as a vehicle to report technical issues and queries, rather than as a means of engaging with other users in a support network. Therefore this finding may indicate that participant frustration with some of the intervention’s perceived limitations contributed to the overall negative impact on PA. One previous study that also found negative intervention effects on PA behaviour concluded that reduced support for the intervention over time was a contributing factor. 199

Intervention fidelity

The literature highlights the importance of intervention fidelity and of ensuring that variations from the original design can be assessed since non-significant findings may be attributable to intervention delivery rather than intervention design. 200 Carroll et al. ’s theoretical framework provides the means of measuring intervention fidelity and understanding its role in the process of intervention implementation. 201 For example, the authors highlight that several potential moderators of the fidelity with which an intervention is implemented include intervention complexity, use of facilitation strategies, quality of intervention delivery and participant responsiveness. Future intervention studies could potentially improve intervention outcomes by improving intervention fidelity with reference to such theoretical frameworks. Finally, systematic reviews also show that the effectiveness of web-based interventions may be enhanced by including various modes of intervention delivery (e.g. text messages, e-mail),201,202 which the PAL scheme did (e.g. through the use of e-mail prompts). However, we have been unable to include data on whether or not participants received and read these e-mails in our engagement analysis. Future work may seek to delve into further analyses around the delivery, open and click rate of e-mails relating to an intervention.

The GLS regressions also showed that use of particular intervention components was associated with increases in some mediators between baseline and 6 months for Intervention Group participants. A higher frequency of using the PAL scheme PA monitoring system (i.e. behavioural practice) and a higher frequency of accessing the study website was associated with higher levels of integrated regulation, intrinsic motivation and habit between baseline and 6 months (i.e. the constructs for which mediation analyses showed significant positive indirect effects). This suggests that a higher level of overall engagement with the intervention (as opposed to engagement with specific intervention components) was related to higher levels of these constructs for intervention group participants. Accessing a higher number of intervention components on the website at least once was associated with increases in identified regulation and integrated regulation, although there was no association with PA levels. Previous work has also identified that overall intervention engagement, and also greater use of intervention components was associated with increases in the primary outcome. 202,203

Further evidence for the importance of the intervention’s self-regulation component was provided by the results of the study’s engagement analysis (i.e. a higher frequency of accessing the website’s feedback and monitoring component was also associated with increases in integrated regulation and intrinsic motivation) which broadly support the findings of the mediation and qualitative analyses. Previous systematic reviews of web-based interventions also highlight web-based self-monitoring as a potentially effective technique. 203 In comparison, engagement with the financial incentive component of the intervention (i.e. redeeming a higher proportion of earned incentives for rewards or a higher frequency of accessing the website’s rewards component) was not found to be related to levels of identified regulation, integrated regulation and intrinsic motivation. Thus, we have further supporting evidence to conclude that the use of financial rewards does not necessarily ‘crowd out’ or diminish intrinsic motivation when delivered as part of a complex multicomponent behaviour change intervention.

Scheme non-use attrition

High levels of non-use attrition in the present study may, to some, reflect dissatisfaction with the intervention’s perceived limitations contributing to the decline in PA outcomes at follow-up. It should also be noted that the definition of non-use attrition that we employed (i.e. occurring at the time of the first 2-week lapse from intervention use) in the present study may be viewed as somewhat arbitrary. Although several studies of web-based PA interventions have adopted this definition, it may be less applicable to the analysis of non-use attrition in workplace interventions for which a 2-week lapse from intervention use may occur for many different reasons (e.g. if a participant is on annual leave, out of the office on work-related travel, or is absent from work for ≥ 2 weeks through illness). Time to non-use attrition for use of the PA monitoring system and use of the website was shown to be significantly predicted by some of the baseline sociodemographic, psychosocial and environmental variables (e.g. age, marital status, health status, financial motivation, identified regulation, recovery self-efficacy and perceptions of the workplace environment) in multivariable Cox regression models. These results are consistent with findings in previous intervention studies, demonstrating that older participants (vs. younger),117,204–208 unmarried participants (vs. married/co-habiting)209 and participants with higher reported health status at baseline (vs. lower health status)210 have decreased risk for non-usage. Individuals with higher levels of financial motivation may have been encouraged to continue participation in the intervention over time in order to continue benefiting from financial rewards. Existing theory suggests that identified regulation and recovery self-efficacy are constructs which are important for long-term behavioural maintenance. 175,176,210–214 For example, it makes sense that individuals with higher recovery self-efficacy should experience a longer period of consistent intervention use before encountering their first 2-week break as they are quicker to recover from a lapse. Notably, the results of our behavioural economics analysis (see Chapter 6, Results from the behavioural economic analysis) indicated that individuals who were more present-biased (i.e. who had less self-control) were also at higher risk for stopping intervention use. Furthermore, some subgroups of participants who were found to be less present-biased in the behavioural economics analysis were also found to be at reduced risk for non-use attrition in survival analyses (e.g. older participants, single participants). This may point to possible interactions between age or marital status and levels of present-bias.

Mediation and moderation analyses: implications for future research

Future PA intervention studies should make use of self-regulation techniques (e.g. self-monitoring, goal-setting), with social and environmental prompts. This may promote habit formation as the new PA behaviour, originally carried out consciously, becomes more automatic. This should also allow for the internalisation of the behaviour, enabling it to be more sustainable long-term. Monitoring of behavioural outcomes is an additional self-regulation technique which should be explored, as it can potentially encourage participants to focus on their satisfaction with more internal outcomes of behaviour change (and, therefore, to maintain it long-term). As Bouton aptly summarised, ‘successful learning of a new behaviour does not permanently replace an earlier one’. 215 Moreover, behaviour change context is also key to maintained change and as such, new behaviours (in the case of the current work, increased PA) may be easily disrupted. Although the PAL scheme utilised cues with regard to e-mails and environmental cues with the placement of sensors, more specific work on the types of contextual cues to aid with behaviour change should be explored.

To promote the internalisation and maintenance of PA behaviour change, future intervention studies should support all of the basic psychological needs posited as important in SDT (i.e. competence, autonomy, relatedness to others). 175,176,212 Studies should also examine how the social and physical environment can influence participants’ experience of PA interventions, and behaviour change (e.g. enjoyment, autonomy). Finally, behaviour change theories should be developed which accurately capture the process of PA behaviour change for specific individuals. This is reliant on well-designed intervention studies investigating causal mechanisms by means of mediation and moderation analyses, and exploring how these constructs operate together.

Engagement and non-use attrition: implications for future research

Participant engagement and non-use attrition are challenges faced in public health intervention research. Therefore, future intervention studies should measure levels of engagement and non-usage with a view to making recommendations for retaining groups of participants who are at the highest risk for non-usage and lack of intervention engagement. This is particularly important for studies of web-based interventions which are known to be particularly susceptible to non-usage and lack of participant engagement. 114 Better guidelines on how to measure intervention engagement are needed. Future intervention studies should also consider how to ensure that those with high risk for non-use attrition keep engaging in the intervention. This is important to ensure participants receive the full potential benefits of the intervention.

Our engagement analysis indicated that when participants found the financial incentive attractive and redeemed their points for rewards, there was a positive impact on PA. This suggests that offering financial incentives can potentially encourage participants to increase their PA behaviour. However, future research is warranted investigating the ability of financial incentives to initiate PA and facilitate long-term behaviour change. There is little evidence on optimal monetary values, varieties (e.g. retail vouchers, more generic gift cards, charity donations, group rewards) or timing of incentives, for achieving health behaviour change. Interventionists should consider how to retain participants who are initially less financially motivated in studies whose main component is the offer of financial rewards. Moreover, where trials of complex interventions such as the PAL scheme result in a negative primary finding, future work needs to determine which element of the intervention ‘does not work’ or whether there is a ‘failure’ that signifies suboptimal fidelity, dose, implementation or of the underlying programme logic model. For studies of PA interventions requiring behavioural practice in the outdoor environment (as was the case for the current study), participants’ perceptions of the environment with respect to PA are important influencing factors that should be mitigated to optimise engagement and minimise non-usage.