A group-based exercise and behavioural maintenance intervention for adults over 65 years with mobility limitations: the REACT RCT

Article history

The research reported in this issue of the journal was funded by the PHR programme as project number 13/164/51. The contractual start date was in September 2015. The final report began editorial review in January 2020 and was accepted for publication in November 2021. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The PHR editors and production house have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the final report document. However, they do not accept liability for damages or losses arising from material published in this report.

Copyright statement

Copyright © 2022 Stathi et al. This work was produced by Stathi et al. under the terms of a commissioning contract issued by the Secretary of State for Health and Social Care. This is an Open Access publication distributed under the terms of the Creative Commons Attribution CC BY 4.0 licence, which permits unrestricted use, distribution, reproduction and adaption in any medium and for any purpose provided that it is properly attributed. See: https://creativecommons.org/licenses/by/4.0/. For attribution the title, original author(s), the publication source – NIHR Journals Library, and the DOI of the publication must be cited.

2022 Stathi et al.

Scientific background and explanation of rationale

Healthy ageing is defined as ‘the process of developing and maintaining the functional ability that enables wellbeing in older age’. 2 Functional ability comprises the intrinsic capacities, both mental and physical, that people can draw on, relevant environmental characteristics and demands, and how people respond to these demands.

With increasing age, there is a population-wide decline in physical function. 3–5 Frailty and associated comorbidities compromise quality of life for older adults and contribute major societal costs directly to people who live with frailty, to friends and family providing care and losing productivity, and to health-care and social care services. 6,7 The impact of this transition is further heightened by an ever-increasing ageing population in the UK (18.2% in 2017 over the age of 65 years, rising to 20.7% by 20278) that is also reflected worldwide. 9,10

There is strong evidence that physical activity has a positive impact, slowing or preventing disability in later life. 11–14 A fit and active older person has 36% lower risk of developing functional limitations and 38% lower risk of hip fracture than inactive older people. 15 In the UK-based OPAL PLUS cohort study,16 older people who undertook at least 25 minutes of moderate to vigorous physical activity (MVPA) per day at baseline received fewer prescriptions and were less likely to be admitted to hospital in an emergency 4–5 years later. 16 Despite these significant benefits, as people age they engage in less physical activity and spend more time being sedentary,13,17 with 31% of 65- to 74-year-olds reporting < 30 minutes of MVPA per week, rising to 53% of people aged ≥ 75 years. 5

This toxic mix of reduced physical activity leading to compromised physical function and increasing pressure on health-care and social care support services has shifted the focus towards supporting the maintenance of functional capacity in later life, with healthy ageing identified as a key public health priority. 18,19 Clinical trials have provided robust evidence that physical features of frailty, such as reduced muscular strength or endurance, can be reversed or their progression slowed by undertaking an appropriate exercise programme. 14,20,21 According to data from the Health Survey for England, 31% of women and 22% of men aged ≥ 65 years report needing help in the last month with one or more activities of daily living (ADLs), such as getting up and down stairs, dressing, getting around indoors or shopping. 5 These people are in transition from independence to frailty and have a great deal to gain if loss of function can be reversed and independence maintained.

Physical inactivity and mobility limitations in older people are more prevalent in socioeconomically deprived sectors of the population. 22,23 Ethnically diverse groups in the USA have a significantly greater risk of developing a range of physical and mental health problems than their white counterparts, and subsequently suffer higher rates of morbidity and premature mortality. 24,25 Self-reported data from the Health Survey for England and the Active Lives survey indicate that older (≥ 55 years) Bangladeshi, Pakistani and Indian adults in England are less likely than their white counterparts to meet physical activity guidelines. 26–28 Thus, exercise interventions that can successfully engage and retain inactive and ethnically diverse groups of older adults could contribute to the reduction of health and social inequalities.

Lifestyle Interventions and Independence for Elders (LIFE) was a landmark US study in the field of physical activity promotion in older adults. 21 LIFE was a multicentre randomised controlled trial (RCT) comparing the effects of a physical activity programme with a successful ageing educational programme in more than 1600 functionally impaired older persons, over an average follow-up period of 2.6 years. The intervention reduced the risk of developing major mobility disability (defined as the inability to complete a 400-metre walk test within 15 minutes) by 18% relative to the control group [hazard ratio (HR) 0.82, 95% confidence interval (CI) 0.69 to 0.98]. It also reduced the risk of persistent mobility disability by 28% (major mobility disability at consecutive time points) (HR 0.72, 95% CI 0.57 to 0.91). The intervention group maintained an increase of 40 minutes per week (95% CI 29 to 52 minutes; p < 0.001) in objectively assessed lifestyle intensity activity (≥ 760 counts/minute compared with the control group at 24 months of follow-up). These estimates are likely to be conservative because the study utilised an active control group that received a substantial health education/lifestyle intervention, including weekly workshops for 6 months and monthly sessions for a further 18 months. Being an efficacy trial, however, LIFE was highly resource intensive, and there was no post-intervention long-term follow-up. The challenge now is to build on this evidence base and develop affordable, scalable interventions that are suitable for delivery in specific contexts and demonstrate maintenance of effect in the long term.

Building on the findings of the LIFE RCT, we designed REtirement in ACTion (REACT), a pragmatic effectiveness trial to test whether or not the LIFE intervention could be adapted into an effective real-life, community programme for older adults at high risk of mobility limitations in the UK. To the best of our knowledge, REACT is the first large-scale UK-based study of its kind. If effective and cost-effective, this programme would offer important health-care and social care benefits, while sustaining health and independence among older adults.

Objectives and hypotheses

The REACT study aimed to test the effectiveness and cost-effectiveness of a community, group-based physical activity and behaviour maintenance intervention based on the US LIFE programme for reducing or reversing the progression of functional limitations in older people who are at high risk of mobility-related disability.

A nested substudy, led by the Wellcome Centre for Integrative Neuroimaging, University of Oxford, tested the hypothesis that the REACT intervention slows the rate of brain atrophy and cognitive function decline. Measures included a brief battery of tests (paper and pencil and computerised) to assess memory, attention and executive function; structural and functional brain magnetic resonance imaging (MRI) measures; and gait analysis for a subsample of participants. This substudy was funded by the National Institute for Health and Care Research (NIHR) Oxford Biomedical Centre, University of Oxford, and is reported elsewhere.

Secondary objectives

To compare intervention and control groups in terms of:

• minutes of moderate-intensity physical activity, as measured by accelerometer data

• sedentary time and breaks in sedentary time

• self-reported physical activity

• hand-grip strength of the dominant hand

• performance on a brief test of cognitive function

• the rate of brain atrophy and performance on more detailed tests of cognitive function and gait analysis tests (functional MRI substudy)

• mental and social well-being, energy, sleep quality and pain

• health-related quality of life

• ADL scores.

To conduct a full economic evaluation to estimate the incremental cost-effectiveness of the REACT intervention compared with the control group. To conduct a multimethod process evaluation to evaluate the feasibility of intervention implementation and inform future implementation and possible refinements of the intervention; to evaluate the intervention delivery and inform conclusions about intervention effectiveness; to investigate the proposed mechanisms of change outlined in the REACT logic model (Figure 1) and seek alternative explanations if this model is not supported; and to understand the role of context to inform whether or not and how the findings can be generalised.

FIGURE 1.

The REACT logic model and associated data collection. CV, cardiovascular; PA, physical activity.

Ethics and governance

The REACT study was reviewed and approved by the NHS South East Coast – Surrey Research Ethics Committee (15/LO/2082).

A substantial amendment (approved 7 March 2016) was submitted to change the SPPB inclusion criteria from 4–8 to 4–9; to change the randomisation process in the internal pilot from 1 : 1 to 2 : 1; and to streamline the recruitment process.

A second substantial amendment (approved 28 October 2016) was submitted to revert to 1 : 1 randomisation in the main trial without balancing the sample imbalance owing to 2 : 1 randomisation in the pilot; to make a change to the randomisation process to balance sites on key variables; to make a change to the process of informing MRI study participants of incidental findings; to include an additional case report form to be used with a REACT group funded by Bristol Ageing Better (Bristol, UK); to add questions to the Telephone Screening Questionnaire; to randomise couples together; and to make an addition to the consent form regarding audio-recording of REACT sessions. Finally, we refined the recruitment materials and processes after recommendations by participants during the internal pilot study.

A third substantial amendment (approved 2 October 2017) was submitted to add socioeconomic status (SES), marital status and caring-related questions to the Telephone Screening Form; and to enable the requesting of participants’ e-Frailty data from general practitioners (GPs) and also summary data from GP mailing databases to allow comparison of responders with non-responders.

A fourth substantial amendment (approved 9 January 2018) was submitted to gain approval for documentation relating to session leader interviews and to exclude planned hospitalisation from serious adverse event (SAE) reporting.

A fifth substantial amendment (approved 16 May 2018) was submitted because three co-applicants changed institution, and a small change was made allowing disclosure of a MRI scan to a GP at the participant’s request. No further protocol changes were made. All research and development approvals were sought and obtained for each site and amendment. The study is registered as a current RCT (ISRCTN45627165).

Public involvement and engagement

The REACT study was built on several years of multidisciplinary work by the study team aimed at understanding influences on the adoption and maintenance of physical activity in community-dwelling older adults. Our Avon Network for the Promotion of Active Ageing in the Community (AVONet) [Medical Research Council (MRC) Lifelong Health and Wellbeing – Collaborative Development Network; ref 90543] used focus groups and workshops with service providers, older people, international experts and service commissioners to assess the needs of older people and their communities for physical activity promotion. 30 The REACT study was considered by our AVONet service user, service provider and commissioner stakeholders to be suitable for delivery across a range of socioeconomic and cultural populations. The REACT protocol was developed based on this input. The Trial Management Group (TMG) was closely involved in the development of the study protocol, and all co-applicants, the trial manager and three people from our service user advisory group (research partners) formed part of that committee. One of the research partners also served on the Trial Steering Committee (TSC). The draft protocol was open to consultation by our service user representatives, our public health expert and members of community organisations. In addition, our Advisory Group (six members) reviewed REACT study materials prior to use.

Trial design

The REACT study was a multicentre, pragmatic, two-arm, single-blind, parallel-group RCT with an internal pilot phase that incorporated comprehensive process and economic evaluations. Following identification and recruitment, 777 participants who met the study inclusion criteria were randomised to either the intervention arm or the control arm. The internal pilot phase assessed the feasibility of recruitment methods (allowing for some refinement) and confirmed that the prespecified criteria for retention to the intervention sessions and the study were met, prior to progressing with the main trial. A nested substudy, led by the Wellcome Centre for Integrative Neuroimaging at the University of Oxford, employed MRI to test the hypothesis that the REACT intervention slows the rate of brain atrophy and decline in cognitive function. This substudy was funded by the NIHR Oxford Biomedical Centre, University of Oxford, and is reported elsewhere. 31 Outcome data were collected at baseline and 6, 12 and 24 months. The protocol of this study was published as an open access publication in 2018. 1

Changes to the trial design

A substantial amendment was accepted (7 March 2016) to change the SPPB inclusion criterion (described below) from scores 4–8 (inclusive) to 4–9 (inclusive) out of 12. This change was deemed to be appropriate to support the recruitment efforts and to adopt the same inclusion criterion as the US LIFE trial. 21

A further substantial amendment was accepted (28 October 2016) to (1) revert to 1 : 1 randomisation in the main trial without balancing the sample owing to 2 : 1 randomisation in the internal pilot study and (2) clarify the randomisation process regarding balancing sites on key variables [study site, age group, sex and initial functional ability (SPPB)].

These changes were discussed and agreed by the TMG and the TSC.

Participants

The eligibility criteria were intended to identify sedentary, community-dwelling, older people aged ≥ 65 years, who were not in full-time employment and had functional limitations (i.e. who were at risk of major mobility limitations), but who were still ambulatory (i.e. they could still walk). This was measured using a battery of objective physical function tests, the SPPB, a group of measures that combines the results of the gait speed, chair rise and balance tests:

• gait speed test – a timed walk over 4 metres at normal walking pace (minimum score 0, maximum score 4)

• chair rise test– time taken to stand up and sit down five times from a dining/kitchen-style chair, arms crossed across the chest (minimum score 0, maximum score 4)

• balance test – unsupported balances for 10 seconds in three positions – feet side by side, feet in semi-tandem and feet in full tandem position (minimum score 0, maximum score 4).

Summed, these tests generate a physical function score from 0 to 12. Older adults with scores of 4–9 (inclusive) out of 12 were eligible to take part in REACT. This criterion was based on data showing that older adults with SPPB scores of ≤ 9 have substantially higher risk of major mobility disability 3 years later [odds ratio (OR) 8.3, 95% CI 3.3 to 20.67] than those with a score of 12. 21,32 Furthermore, the European Medicines Agency has published specific cut-off points for SPPB to identify three stages of frailty (< 7, frail; 8–9, pre-frail; and 10–12, fit/normal function), which support the REACT study inclusion criterion. 33

During recruitment, participants were informed that they would receive £15 shopping vouchers for attending each of the 6-, 12- and 24-month assessment sessions.

Eligibility screening

The eligibility of respondents was assessed in a three-step sequential screening process:

1. Initial self-selection. The participant approach letters and participant information sheet (PIS) clearly stated that we were recruiting people who were still able to carry out, but had some difficulty with, daily activities, such as walking, climbing stairs and getting out of a chair.

2. Phone-based screening. After gaining verbal consent, a preliminary phone screen checked inclusion and exclusion criteria that could be assessed by telephone, including a check on medical exclusion criteria. We excluded people who were unable to walk across a room, who were living in residential care, who were awaiting hip or knee surgery, who were receiving radiation therapy or chemotherapy, who had received recent heart or spinal surgery, or who had severe illness that would prevent participation, including severe arthritis, diagnosed dementia, severe kidney disease, unstable heart disease and severe psychiatric illness. Availability to attend intervention sessions was also checked. Participants who did not meet the eligibility criteria were thanked for their time and provided with the Age UK (London, UK)/NHS guide to healthy ageing and other sources of further advice and information. 34

3. Face-to-face screening sessions. Potentially eligible participants were invited to a group-based assessment session at a local community centre, at which they could ask questions about the study and were asked to give written informed consent. The SPPB gait speed test was conducted first, and those who failed to complete the 4-metre walk or did not meet the SPPB inclusion criterion (a score of 4–9 inclusive) were thanked for their time and provided with an information pack (as above). Participants who met the eligibility criteria were invited to take part in the remainder of the baseline assessments.

Full inclusion and exclusion criteria are summarised in Table 1.

Consent

Older adults who were willing to take part in REACT were asked to provide verbal informed consent at the beginning of the telephone screening call and written informed consent prior to commencement of the face-to-face screening sessions.

Study settings

The study was conducted in Bath, Bristol, Birmingham and Devon, UK, allowing recruitment of a socioeconomically and ethnically diverse sample, comprising participants from urban, rural and semirural locations.

A range of recruitment strategies to identify suitable participants was employed.

Study intervention

Intervention delivery

REACT session leaders were qualified to at least REPS Level 3 (Exercise Referral Diploma or equivalent) and were experienced in delivering safe and effective exercise sessions to older adults. Behavioural maintenance sessions were usually delivered by the same session leader but were occasionally delivered by other staff from the same organisation. The REACT co-applicants provided a 2-day intervention delivery training to session leaders, including detailed session plans to ensure consistency in and fidelity to programme delivery based on a written programme of materials and manuals.

Intervention fidelity

We included a range of the strategies outlined by the National Institutes of Health Behaviour Change Consortium to reinforce intervention fidelity. 42 We (1) ensured ‘design fidelity’ by building our intervention around a clear logic model (see Figure 1); (2) recruited REACT session leaders with appropriate skills and experience; (3) developed an accessible, standardised intervention manual; (4) implemented the standardised REACT session leader training programme; (5) trained more REACT session leaders than needed to accommodate illness or withdrawal; (6) monitored delivery fidelity by recording consultation meetings for a sample of three or four sessions per intervention provider and by applying a fidelity checklist; and (7) checked for intervention ‘receipt’ and ‘enactment’ of appropriate levels of physical activity outside the REACT sessions by checking participant understanding of the correct performance of exercises and regularly reviewing progress in the behavioural maintenance sessions. In addition, fidelity was enhanced by incorporating a gradual transition to daily activity within the structure of the REACT intervention (i.e. withdrawal of one session per week after 12 weeks, reduction of behavioural maintenance sessions to once per month after the first 6 months, with targeted planning of ongoing daily lifestyle physical activities around each transition).

Health economics

Full details of the REACT economic evaluation are given in Chapter 5, but, in summary, we used data collected during the trial to estimate the resource use and costs associated with the delivery of the intervention and the wider NHS, social care and participant-level resource use and costs over a 24-month follow-up. The primary economic outcome measure is the quality-adjusted life-year (QALY) gain, derived from participant-reported EuroQol-5 Dimensions, five-level version (EQ-5D-5L), data.

Given the long-term nature of the potential benefits from the REACT intervention, we conducted decision-analytic modelling to assess the longer-term (lifelong) consequences of the intervention compared with the control, including consequences in terms of health-care and social care costs.

Process evaluation

The REACT study follows the principles of the UK MRC guidance on process evaluation. 43 For full details of the process evaluation see Chapter 4; however, the aims and methods are briefly summarised below. The purpose of the process evaluation was to:

• evaluate the feasibility of implementation, including barriers to and facilitators of implementation, to inform future implementation and possible refinements of the intervention

• evaluate the quality and quantity of intervention delivery to inform conclusions about intervention effectiveness

• investigate the proposed mechanisms of change outlined in the REACT logic model (see Figure 1) and seek alternative explanations if this model is not supported

• understand the role of context to inform whether or not and how the findings can be generalised.

Methods used in the process evaluation included:

1. a mixed-methods assessment of intervention fidelity (quality of intervention delivery) using a checklist applied to audio-recordings of the REACT social education sessions

2. a qualitative analysis of semistructured interviews conducted with participants and staff providing the intervention

3. quantitative testing of hypotheses derived from the logic model, using study data on demographics and outcomes, as well as a set of questionnaires to measure processes of behaviour change proposed by the REACT logic model.

Sample size calculation

Power calculations for the primary outcome (SPPB score) at 24 months were based on the published definition of a clinically minimum meaningful change in SPPB score of 0.5 points,21,44 an expected SD for change in SPPB scores from baseline to 2 years of 2.2,21 a two-sided significance level of 0.05 and an expected cumulative loss to follow-up of 12.5% per year. To provide 90% power, this required a sample size of 384 participants per arm (768 in total).

To detect a change of 0.5 points with a SD of 2.0, assuming that loss to follow-up accumulates at 12.5% per year throughout REACT’s 2-year follow-up period, the required sample size was 384 participants per arm for 85% power using two-sided 5% significance. The REACT study, therefore, looked to recruit a total sample of 768 participants. This sample size also provides 90% power to detect a difference in moderate-intensity physical activity of 50 minutes per week [standard deviation (SD) 185 minutes per week] with 5% significance. A 2 : 1 randomisation process was applied in the internal pilot phase. A 1 : 1 randomisation process without rebalancing was applied in the main trial, resulting in an allocation ratio of 1.11 : 1. On the assumption that the effect size, the dropout rate and the significance level that we were interested in remained unchanged, our power reduced to 84.9% (from 85% power using two-sided 5% significance if the sample was rebalanced).

Interim analyses and stopping guidelines

The TSC, with advice from the Data Monitoring and Ethics Committee (DMEC), assessed the feasibility of the trial during the internal pilot phase, taking into account findings on the acceptability of trial procedures, intervention adherence and recruitment rates. After 6 months, recruitment data were reviewed by the TSC and, as outlined in the study protocol,1 when recruitment rates were found to be less than predicted,29 the research team took actions to increase them (increasing the number of people approached and adapting recruitment procedures). The TSC was happy with the impact on recruitment rates and recommended that we proceed to the main trial. Retention rates (proportion of people providing follow-up data) were also checked at 6 months. Receipt of strong negative feedback from the majority of either participants or intervention providers about the intervention or trial methods would have been considered a stopping criterion. No such negative feedback was received.

Randomisation

Blinding

It is not possible to blind study participants to treatment allocation in behavioural intervention studies, and this is not a problem in pragmatic trial designs, which aim to estimate the benefits of the intervention over and above usual or standardised care. However, we took steps to ensure that data collectors, statisticians and the research team remained blinded to allocation, excluding one RA at each site. The chief investigator was unblinded when needed to allow assessment of SAEs. At follow-up data collection visits, participants were asked not to reveal which arm they were in.

Data were coded so that those undertaking the statistical and economic analyses were blinded. Given the study design, we did not anticipate a substantial risk of contamination (i.e. exposure of the control arm participants to the REACT intervention). However, as part of their briefing on entry to the study, participants in the intervention arm were asked not to share or discuss the content of the intervention sessions with any control arm participants whom they may be in touch with for the duration of the study. The possibility of contamination of control arm participants by REACT session leaders was minimised by giving them clear instructions not to provide intervention materials or information to any participants not assigned to the intervention arm. Attrition bias was minimised by having robust trial procedures to prevent data loss and also by analysing the data using an intention-to-treat (ITT) approach.

Case report forms did not contain any data that would enable the identification of the participant, so staff entering data remained blinded.

Unblinding

The DMEC undertook regular safety data reviews after recruitment began, and all SAEs were reported to it. The DMEC was responsible for identifying any need for unblinding and periodically reviewed unblinded safety data to determine patterns and trends of events, or to identify safety issues, which would not be apparent on an individual case basis.

Statistical analyses

The REACT analysis populations were as follows:

• Population 0. The primary analysis was performed on an ITT basis including all participants consented and randomised.

• Population 1. Analysis of all consented participants who completed ≥ 50% of the intervention (minimum required dose analysis).

• Population 2. Analysis of all consented participants who completed ≥ 75% of the intervention (high adherers analysis).

There is no published consensus on what constitutes a minimum dose of attendance for behaviour change interventions. Our choice of 50% as a minimum/sufficient dose and 75% as a high dose was based on (1) attendance rates for other successful behaviour change interventions promoting physical activity range from 36%47 to 61%,21 and (2) expert opinion within the TMG.

Primary and secondary outcome data analysis

Analyses were prespecified in the published protocol. Primary outcome analysis was undertaken blinded to group allocation. Primary comparative analysis was undertaken using the principles of ITT with due emphasis placed on CIs. Using appropriate descriptive statistics, we assessed any imbalance between the trial arms at baseline and described the characteristics of participants. The comparison of primary interest was the difference between the intervention and the control arms in SPPB score at the 2-year follow-up (24 months after randomisation). Factors in the model comprised baseline age, sex and study site (Bath/Bristol, Devon or Birmingham). Baseline SPPB scores were added as a covariate in the model and, in addition, we adjusted the estimates for clustering by exercise group within the intervention arm, with control arm participants entered as individual groups following the method cited by Flight et al. 48 Analyses were conducted in Stata^® SE version 15.0 (StataCorp LP, College Station, TX, USA). The randomisation stratification factors (site, age and sex) were entered into the analysis model as categorical variables.

Secondary outcome analyses were undertaken using the same approach as for the primary analysis (excluding the sensitivity analyses), using the baseline, 6-month, 12-month and 24-month follow-up data and linear mixed-regression models. Health-related quality of life, as assessed by EuroQol-5 Dimensions (EQ-5D), will be reported elsewhere as part of the health economic evaluation.

As an exploratory analysis, the effect of several predefined factors was further investigated and is presented. These included the stratification variables [age categories (65–74 years and ≥ 75 years), sex and study site (Bath/Bristol, Devon, Birmingham)], as well as comorbidity levels at baseline (none or one chronic medical conditions vs. two or more chronic medical conditions), socioeconomic subgroups (using education, home ownership and quintiles of area deprivation), history of falls (recorded fall or not during 6 months prior to baseline) and the uptake of any co-interventions during the 24-month study period. Health-related quality of life, as assessed by EQ-5D, is reported in Chapter 10.

Intervention costs were estimated by identifying key resources (programme co-ordination, session leader time and expenses, venue hire, equipment, consumables, and programme-specific training) and assigning values to the resources used (see Chapter 5). The data were collected by the REACT session leaders and trial manager.

Adjustment in the models

The intention was to adjust the models for the four stratification factors and the intervention group (clustered owing to the group nature of the intervention) and the intervention arm as follows: age group and sex were included as fixed effects, site was included as a random effect, baseline functional ability was included as a covariate, the intervention group was included as a random effect and the intervention arm (intervention or control) was included as a fixed factor. The intervention group effect is the effect due to the intervention being delivered to groups of individuals. In the control arm, the clusters are specified by the individual. 48

Subgroup analyses

To examine the association between dose and response, we conducted subgroup analyses comparing participants attending ≥ 50% and ≥ 75% of sessions with (all) controls (the three defined analysis populations above). Mirroring the primary analysis, variables representing age group, sex, baseline SPPB and exercise group (within the intervention arm) were entered into the model.

Missing data

We expected a relatively low level of dropout and missing data. Therefore, the primary analysis was undertaken without imputation of missing values. However, a comparison of baseline covariates between completers and non-completers was undertaken to assess the impact of dropouts on the results.

Multiple testing

The primary analysis (SPPB at 24 months in the ITT population) is a single test, and, therefore, adjustment for multiple testing was not considered to be appropriate.

The analyses of the secondary outcome measures and the primary outcome at other time points were considered as exploratory analyses, and, therefore, there was no adjustment to account for multiple testing. The results of these analyses were interpreted in the light of the potential for an increased risk of making a type 1 error.

Safety data

The safety population was all trial participants (see Chapter 3, Main outcome results, for the analysis population for the primary outcome).

All SAEs were reported regardless of relatedness to the trial. Non-SAEs (regardless of relatedness) were not reported. All reportable events were followed up until resolution, where possible, or until the end of the data collection period.

No formal comparisons between groups were made because the numbers of events were relatively small. Full details of SAEs are presented in Chapter 8.

Amendments to the statistical analysis plan

Following the provision of baseline scores, but prior to the primary analysis, concerns were raised by members of the TMG about the inclusion of the adjustment for clustering in the primary analysis. Given the group nature of the intervention, there were concerns that adjusting for the clustering (given the lack of clustering in the control arm) would reduce the apparent significance of the intervention effect. Two teleconference meetings were held between the TSC and the TMG: one on 8 January 2020 and one on 27 January 2020. This addendum reflects additions to the analysis following those discussions.

This addendum is an addition to the main REACT statistical analysis plan v5. The statistical analysis plan outlines the primary analysis, including (but not limited to) adjusting the primary analysis for the group nature of the delivery of the intervention in the intervention arm.

In the submitted proposal30 and the protocol paper,1 no consideration was given to the clustering in the intervention arm. Therefore, by adjusting for the clustering in the intervention arm in the analysis, there would be a potential reduction in the power of the study. It was also agreed that the clustering in the intervention arm is a potentially important aspect to the structure of the data and, therefore, not including this may result in an increased risk of making a type 1 error. The impact of this, potentially, would be to marginally increase the size of the standard errors associated with the treatment effect estimates. Flight et al. 48 provided three case studies suggesting that for moderate levels of intracluster correlation these are unlikely to affect the conclusions. However, this may be the case where a result is borderline significant or where there is a strong clustering effect. 48

Sensitivity analysis for the primary outcome data analysis

The structure of the study was such that participants in the intervention arm met in groups whereas participants in the control arm had minimal interaction. Therefore, in the primary analysis, the primary outcome (SPPB at 24 months) was modelled using the approach recommended by Flight et al. 48 for a partially clustered design; an analysis of covariance (ANCOVA) was undertaken, adjusting for the site as a fixed effect (Birmingham, Bath/Bristol, Devon) and group as a random effect (the groups within the intervention arm that met for exercise). Following this primary analysis, an additional sensitivity analysis was undertaken without the inclusion of the treatment groupings in the model. This analysis is in line with the analysis proposed in the protocol paper. 1

The sensitivity analysis was further enhanced by an investigation of the intracluster correlation at each data collection period once other factors (primarily site) had been accounted for. Investigating how the amount of clustering (represented by the intracluster correlation) changes from baseline to 6 months (peak intervention intensity) to 12 and then 24 months may help to indicate the extent to which the intervention itself was associated with any clustering effects (e.g. owing to treatment groups forming close activity-supporting bonds or having strong leadership that may cause the outcomes to cluster by intervention delivery group). This could aid in our interpretation of any discrepancies between effectiveness estimates produced by the clustered and the un-clustered analysis.

Statistical significance

For the hypothesis test, two-tailed p-values of ≤ 0.05 are considered statistically significant. Given that the additional sensitivity analyses outcomes here are exploratory in nature, there was no adjustment for multiple testing.

Model assumptions

For all methods outlined, underlying assumptions were checked using standard methods (e.g. residual plots). If assumptions were violated, alternative methods of analysis were sought. In particular, the underlying assumption of the linear impact of baseline covariates was assessed to ensure that baseline covariates were categorised and fitted as factors in the model. The assumption of the consistent clustering effect between the intervention and the control arm was evaluated.

Outcomes

Changes to the study outcomes during the course of the study

There were no changes to the study outcomes during the course of the study.

Participant flow

Between June 2016 and September 2017, 3116 people were telephone screened (of whom 1077 were not eligible and 825 declined to participate further) and 1214 attended for baseline screening. Of these, 804 were found to be eligible and 777 were randomised (intervention, n = 410; control, n = 367). Throughout the study, 39 couples or pairs of close friends who were both eligible were randomised together to minimise contamination between study arms. The number of participants included in the primary analysis at 24 months was 628 (80.8%): 334 (81.5%) in the intervention arm and 294 (80.1%) in the control arm. Figure 2 shows the flow of participants through the study.

Recruitment

The trial was successful in recruiting. In fact, slightly more participants were recruited than had been planned (target, n = 768; actual, n = 777). Between February 2016 and September 2017, we contacted 25,559 people (via 35 GP practices, community partners and a PR campaign). A total of 3116 people responded and were telephone screened by the local site RA, 1214 people were screened face to face at local community centres and 777 were randomised (intervention, n = 410; control, n = 367), slightly exceeding the recruitment goal of 768 participants. 29

Baseline data

Baseline characteristics were similar between the two study arms (Table 2). The mean age of the participants was 77.6 years (SD 6.8 years) and 66.2% were female. The majority of participants (95.1%) were Caucasian, 1.2% were Asian, 3.0% were African/Caribbean and 0.8% were of other/mixed ethnicity. The mean SPPB score was 7.37 (SD 1.56) and the mean body mass index (BMI) was 29.25 kg/m² (SD 5.71 kg/m²). Just over half of participants were educated beyond secondary school level (n = 417; 53.7%) and the majority were overweight/obese (n = 588; 76.5%). The study aimed to broadly represent the diversity of deprivation and ethnicity for individuals over 65 years of age within the UK population. Comparisons of the REACT cohort and the population aged over 65 years in England and Wales are shown in Appendix 1, Table 21. 62–64 In total, 11.1% of REACT participants fell within quintile 1 (most deprived) of the Index of Multiple Deprivation (IMD), compared with 14.3% of the general UK population aged over 65 years. In quintile 2, these figures were 20.2% and 17.6%, respectively. In terms of ethnicity, REACT under-recruited Asian participants (2.6% in the UK population aged over 65 years, 1.2% in the study) but over-recruited African/Caribbean participants (1.3% in the UK population aged over 65 years, 3.0% in the study).

The proportion of Caucasian/white participants was slightly lower than in the general population (95.1% vs. 95.5%, respectively) while the proportions of other/mixed ethnicities were very similar (0.8% vs. 0.7%, respectively). In terms of sex, 45.6% of the over-65 years population of England and Wales are male, compared with the 33.85% of REACT participants. However, compared with the over-65 years population of England and Wales, the REACT cohort was skewed towards the older age ranges, where the proportion of females increases. 29

Main outcome results

At the 24-month follow-up, the mean SPPB score (adjusted for baseline SPPB, age, sex, study site and exercise group) was significantly higher in the intervention arm (mean 8.08, SD 2.87) than in the control arm (mean 7.59, SD 2.61), with an adjusted mean difference of 0.49 (95% CI 0.06 to 0.92; p = 0.014). Only one instance of unblinding was reported during the collection of data at 24 months.

The primary and secondary outcomes at 24 months are presented in Table 3, and all outcomes at 6 and 12 months are reported in Appendix 1, Tables 25 and 26. The SPPB score was significantly higher in the intervention arm than in the control arm at 6 months (adjusted mean difference 0.68 points, 95% CI 0.39 to 0.96 points; p < 0.001) and 12 months (adjusted mean difference 0.77 points, 95% CI 0.40 to 1.14 points; p < 0.001). Self-reported physical activity was significantly higher in the intervention arm than in the control arm at 6 months (adjusted mean difference in PASE score of 16.3 points, 95% CI 6.78 to 25.9 points; p = 0.001), 12 months (adjusted mean difference 10.8 points, 95% CI 3.18 to 18.5 points; p = 0.006) and 24 months (adjusted mean difference 10.7 points, 95% CI 2.62 to 18.8 points; p = 0.010). Self-reported engagement in muscle-strengthening exercise showed a similar pattern, with highly significant differences (p < 0.001) at all three follow-up time points.

Accelerometer data indicated a significant difference favouring the intervention group at 12 months for total MVPA (adjusted mean difference 3.11 minutes per day, 95% CI 0.00 to 6.23 minutes; p = 0.05) and MVPA accumulated in bouts of at least 10 minutes (1.24 minutes per day, 95% CI 0.22 to 2.26; p = 0.018). This equates to a difference of 22 minutes per week of MVPA lasting ≤ 10 minutes. Significant differences favouring the intervention arm were also observed in the SF-36 physical component score (at 6 and 12 months), hand-grip strength (at 12 months) and the MAT-SF self-reported lower limb physical functioning scale (at 6, 12 and 24 months).

Numbers analysed

All data were analysed based on the participants’ originally assigned groups (ITT). The number of participants included in the analysis of each outcome measure is shown in the outcome tables (see Table 3; see Appendix 1, Tables 25–26).

Losses and exclusions

Over the 2-year measurement period, we had a relatively low level of withdrawals from the study. Only 14.3% (n = 111) of participants withdrew from the study: 59 from the intervention arm and 52 from the control arm. The reasons were personal choice (30.6%), health issues (25.6%), unknown reasons (18.9%), death (14.4%), family health issues (7.2%) and being excluded by the intervention delivery organisation (3.6%) (see Appendix 1, Table 22). Final 24-month data collection was completed in October 2019. Given that the number of missing data did not differ substantially between the intervention and the control arm, the primary analysis was undertaken without imputation of missing values.

Data for physical function (SPPB score) at 24 months (primary outcome) were available for 628 participants (80.8%): 334 (81.5%) participants in the intervention arm and 294 (80.1%) participants in the control arm. Compared with the predicted loss to follow-up of 12.5% per year (25% cumulative for two years), the actual loss to follow-up over the 2 years of the study was 19.2% (see Appendix 1, Table 23).

Programme adherence

Of the 410 participants allocated to the intervention arm, 16.1% did not engage with any of the intervention sessions (non-starters), 19.0% attended < 50% of the sessions offered, 20.2% attended 50–74% of the sessions offered and 44.6% attended ≥ 75% of the sessions offered. Among all participants allocated to the intervention arm (including the non-starters), the mean percentage of sessions attended was 56.8% (95% CI 53.6% to 60.1%). Among participants allocated to the intervention arm who engaged with the programme (starters only), this figure was 67.7% (95% CI 65.1% to 70.4%). An association between dose and response was observed (see Appendix 1, Table 28), with an adjusted mean SPPB difference of 0.64 (95% CI 0.23 to 1.05; p = 0.002) for those attending ≥ 50% of intervention sessions and 0.81 (95% CI 0.38 to 1.23; p < 0.001) for those attending ≥ 75% of intervention sessions.

Sensitivity analysis

Sensitivity analyses, including imputation of missing values and not adjusting for clustering by exercise group, did not significantly change the above results (see Appendix 1, Table 29). The intracluster correlation coefficient for SPPB scores relating to clustering by exercise group within the intervention arm was 0.02 (95% CI 0.0085 to 0.129). Subgroup analyses for age, education levels and SES (key inequality populations) or other characteristics found no significant interactions with study arm, indicating that the intervention worked equally well for all of the pre-identified subgroups (see Appendix 1, Table 29).

Health economic results

The full 12-month REACT programme, as delivered in the trial, was estimated to cost £9466 per group, an average of £622 per participant. For more details on health economic analysis, see Chapter 5.

Non-adherence to the protocol

As determined in the REACT statistical analysis plan, the following protocol violations were considered:

1. Enrolment protocol violations were considered to occur if a member of the research team failed to appropriately apply the study’s eligibility criteria, resulting in the enrolment of an inappropriate patient into the trial.

2. A randomisation protocol violation was defined as a technical or human error leading to the violation of the intended randomisation sequence or any attempts to subvert allocation concealment.

3. A study intervention protocol violation was defined as a delivery error (incorrect number of sessions delivered) in the study intervention attributable to members of the research team. The research team included members of the study co-ordinating centre, site investigators and research co-ordinators.

4. Data collection protocol violations encompassed errors in which the research team failed to comply with specific trial guidelines for data collection and/or outcome evaluation for avoidable reasons.

No type 3 or 4 violations occurred. One type 1 error occurred when a participant scored 3 on the SPPB (inclusion criterion: SPPB score of 4–9) but was mistakenly included in the trial and randomised to the control arm. Data for this participant were included in the analysis. Three type 2 errors occurred. Prior to a protocol change to allow close friends to be randomised together (to avoid contamination), two pairs of friends were randomised separately and allocated to different arms. After discussions with the TMG and the TSC, it was agreed to submit an ethics amendment for the protocol change and both pairs of friends were allowed to attend the intervention programme. There was also one erroneous allocation letter sent to a participant informing them that they were allocated to the control arm when, in fact, their allocation was to the intervention arm. In all cases, analysis was conducted on the basis of original allocation (ITT), and the flow diagram (see Figure 2) reflects this.

Harms

In the 56 months of the study, 93 events were classified as SAEs: 59 from the intervention arm and 34 from the control arm (see Appendix 1, Table 24). SAE data were collected when inviting participants to attend assessment sessions, at assessment sessions (6, 12 and 24 months), when inviting control arm participants to attend one of their three social and education sessions and via session leaders reporting SAEs among intervention participants. The somewhat larger number of SAEs reported for intervention arm participants is likely to be a result of the higher frequency of contact with this group by session leaders, providing increased opportunity for reporting illness, injury and hospitalisation.

In the early phases of the study, all hospitalisations were reported as SAEs, including hospitalisation for planned procedures. On 9 January 2018, a substantial ethics amendment was approved to exclude planned hospitalisation from SAE reporting. From that date, any adverse event or adverse reaction was regarded as serious if it:

• resulted in death

• was life-threatening

• required non-elective hospitalisation, prolongation of existing hospitalisation or elective hospitalisation that may be related to taking part in the study

• resulted in persistent or significant disability or incapacity.

In line with the REACT protocol, SAEs were reported within 24 hours to the chief investigator, the Data Management and Ethics Committee chairperson for consultation on the relatedness to the study, and the sponsor of the trial. Ninety of these SAEs were classified as events unrelated to the study and, therefore, were not reported for further consultation with the TSC. One case was related to the study, and two cases considered to possibly have been related to the study were investigated and deemed not to be related to the study. These were reported to the appropriate regulatory body, the sponsor of the REACT trial and the DMEC, as per trial protocol.

Discussion

In support of the primary hypothesis, older adults with mobility limitations who received the 12-month REACT programme experienced statistically and clinically significant improvements in lower limb physical function compared with control arm participants at the 24-month follow-up (12 months after the end of the intervention) and also at 6 and 12 months, indicating a sustained benefit over time. Higher intervention effects were associated with increased programme attendance.

The baseline SPPB scores were almost identical to the LIFE study population,21 enabling comparison. The observed difference in SPPB score of 0.49 at 24 months was three times larger than the between-group difference reported in the LIFE trial. In the LIFE trial, this smaller difference in SPPB was sufficient to reduce the subsequent risk of major mobility disability (defined as the objectively assessed inability to walk 400 metres) by 18% and the risk of persistent mobility disability (defined as two consecutive major mobility disability assessments or assessment of major mobility disability followed by death) by 28%. The ability to walk a distance of 400 metres strongly relates to maintenance of independent living and reduced risk of mortality. 65 These meaningful impacts from lower levels of SPPB change in the LIFE study suggest that the minimum clinically meaningful difference in SPPB may be considerably lower than the difference of 0.50 used in calculating our sample size. Indeed, other evidence suggests that changes in SPPB score of ≥ 0.28 are meaningful in frail or pre-frail older adults (SPPB score of 4–9). 66

At the completion of the intervention (12 months post baseline), significant differences were observed in SF-36 physical component score, MAT-SF, MVPA, self-reported physical activity and adherence to muscle-strengthening exercise, and hand-grip strength. These are consistent with the idea that the intervention increased engagement in muscle-strengthening, balance and endurance exercise that mediated the observed effects on physical functioning.

The strong association between session attendance and increased lower limb physical function (see Appendix 1, Table 28) also suggests that the intervention worked through engagement in the REACT programme. However, causality cannot be implied from this association and there may be other explanations; for example, older adults who became frailer owing to injuries or life events, or who did not feel they were benefiting from the intervention, may have been less likely to attend the intervention.

It is worth noting the methodological issue of how to choose cut-off points representing a ‘sufficient’ or minimum dose for behavioural interventions, and this study provides data that may be informative for future studies of group-based exercise interventions. Our data show that attending ≥ 50% of sessions was associated with a clinically meaningful effect on SPPB (0.64 points) and that a minimum of 75% attendance was associated with a considerably stronger effect (0.81 points).

The increase in MVPA (9 minutes per week bouted or 22 minutes per week unbouted) was small, although it is worth noting that any increase in MVPA has effects on health67 and these increases must be set against very low levels of initial MVPA in this sample of frail or pre-frail older people (the baseline level of bouted MVPA was just 41 minutes per week). The 2.6-point change in SF-36 physical component score was small (a clinically meaningful difference being cited as around 4 points),68,69 as was the change in hand-grip strength (0.8 kg compared with a clinically important difference of 5.0 kg). 70 This demonstrates that functional benefits of the REACT programme were specific to the focus of the exercise intervention programme (lower limb mobility) and did not generalise to upper body physical functioning.

At 24 months, of the secondary outcomes, only changes in self-reported physical activity, muscle-strengthening exercise and MAT-SF were sustained. Other differences found at 12 months were reduced (17 minutes per week of unbouted MVPA, 1.5 points in SF-36 physical component score) but fell below the level of significance. This may reflect deterioration of the effects on exercise behaviours over time, as well as a lack of statistical power to detect smaller differences.

To the best of our knowledge, the REACT trial is the first trial targeting physical function with a long-term (12-month) intervention and a 24-month follow-up in this population. It was a well-powered definitive study, being the largest of its kind conducted in the UK. It was a robustly designed and conducted study, with low attrition rates (19% at 24 months) and high adherence.

Although only 3% of those invited to take part were recruited, it should be noted that the REACT study invited everyone in our study area aged ≥ 65 years and then applied a two-stage screening process. Based on the screening data from the study, we estimate that over 80% of those invited were likely to be ineligible because their SPPB score was outside the target range or because they met other exclusion criteria. On this basis, the response rate among the eligible population was 17%. Although this still leaves some uncertainty around the generalisability of the findings, it is reassuring that the recruited sample was representative of the UK population aged over 65 years in terms of deprivation and ethnicity, except for an under-representation of South Asian older adults. 29,65

The main limitation was that, as with other studies of behavioural interventions, blinding of participants to the study arm was not possible, introducing the possibility of social desirability bias in patient-reported measures. However, the primary outcome here consisted of a battery of physical tests assessed by independent observers, with the data collectors blinded to study arm allocation. Given that the secondary outcome analyses were exploratory, there was no adjustment for multiple testing, and the significance of the analyses needs to be interpreted accordingly. A further limitation is that it was not possible to explore variation in outcomes by ethnicity owing to insufficient numbers of ethnic minority participants.

Measurement of physical activity using accelerometers is held to be a gold standard for large-scale field trials and superior to self-report questionnaires. However, in REACT we observed that, despite improved physical function, Muscle-Strengthening Exercise Questionnaire (MSEQ) and self-reported PASE scores, there were no meaningful improvements in accelerometer-measured physical activity over the course of the study. The lack of agreement between objectively measured and self-reported measures could be explained by (1) wrist-mounted accelerometers lacking the sensitivity to detect muscle-strengthening and balance exercise and (2) the fact that the MSEQ is designed to capture muscle-strengthening activity and the PASE includes an item that captures muscle-strengthening exercises and weights it heavily in its scoring algorithm. Furthermore, a focus on MVPA using absolute cut-off points derived from healthier populations may not be appropriate for a pre-frail older adult population whose capacity for aerobic physical activity is likely to be lower than that of healthier, younger adults. Further work is, therefore, needed to understand how valid measures of physical activity and of engagement in muscle-strengthening exercise/activities can be derived from accelerometer data, particularly in older adults.

The subgroup analyses suggested consistency of intervention effects across different subgroups of the population, including both sexes and people of different ages, education levels and SES (as assessed by area deprivation). This indicates that, if rolled out, the intervention has potential to reduce health inequalities, if targeted towards underserved populations.

Aims

The purposes of the process evaluation in the REACT trial were to:

1. explore participants’ and facilitators’ experiences of the intervention to identify possible refinements of the intervention for future implementation

2. evaluate the quality and quantity of intervention delivery to inform conclusions about intervention effectiveness

3. investigate proposed mechanisms of change outlined in the REACT logic model and seek alternative explanations if this model is not supported

4. understand the role of context (e.g. setting, area deprivation, ethnicity and sex) to inform whether or not and how the findings can be generalised.

The dose of the intervention delivered and the associations of outcomes with dose are reported in Chapter 3. Subgroup analyses exploring differential impacts of the intervention in different population subgroups are also reported in Chapter 3, but are combined with the data presented in this chapter to help to inform refinements of the REACT logic model.

The REACT logic model

The logic model for REACT is shown in Figure 1. It identifies:

• the REACT intervention components and how they were intended to be delivered to participants

• hypothesised mechanisms of action of the REACT intervention (the model incorporates a number of causal assumptions about the process by which the intervention was intended to effect change in health behaviours and outcomes)

• hypothesised contextual variables that might moderate or mediate mechanisms of change in motivations and behaviour

• hypothesised interactions between participation in the intervention, delivery quality, motivation, behaviour and outcomes.

Study 1: intervention fidelity

Parts of this section have been reproduced from Cross et al. 71 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/. The text includes minor additions and formatting changes to the original text.

Study 2: qualitative evaluation of intervention processes

Results

Six-, 12- and 24-month semistructured interview findings

Participant characteristics

At the 6-month interviews, the study sample comprised 12 women (71%) and five men (29%). The baseline age of the participants ranged from 68 to 88 years. The SPPB score was relatively balanced across the sample (47% frail, 53% pre-frail), as was attendance (47% low attendance, 53% high attendance).

At the 12-month interviews, as it was not possible to contact four women and one man from the 6-month cohort, three additional participants were recruited and the sample comprised 11 women (73%) and four men (27%). The age of the participants ranged from 68 to 89 years. At the 12-month interviews, most participants were classified as pre-frail (73%). Programme attendance was relatively balanced across the sample (40% low attendance, 60% high attendance).

At the 24-month interviews, the study sample comprised 13 women (68%) and six men (32%). The age of the sample ranged from 68 to 89 years. At the 24-month interviews, the majority of the participants were classified as being pre-frail (79%) and most (63%) were classified as having high attendance. Table 7 summarises participant characteristics and reports baseline and 24-month SPPB scores and programme attendance.

TABLE 7 - Participant characteristics

Attendance of sessions is ≥ 70% of planned sessions (n = 64).

Pseudonym	Group	Sex	Age (years) (baseline)	SPPB score			Attendance	Interview
				Baseline: frail	Baseline: pre-frail	24 months		6 months	12 months	24 months
Dorothy	1	Female	68	7		12	72%a	✗	✗	✗
Cordelia	1	Female	88	4		5	27%	✗
Etta	1	Female	70		8	10	98%a	✗	✗	✗
Anandi	1	Female	76		9	9	49%	✗	✗	✗
Darsha	1	Female	77		9	8	81%a		✗	✗
Iris	2	Female	89		9	8	79%a	✗	✗	✗
Mary	2	Female	76	4		4	35%	✗
Cecil	2	Male	69	4		12	65%	✗
Frederick	2	Male	87		8	9	90%a	✗	✗	✗
Geraldine	3	Female	78		8	10	61%	✗	✗	✗
Arman	3	Male	71		8	9	95%a	✗	✗	✗
Valerie	3	Female	86		8	9	61%		✗	✗
Rita	3	Female	76		8	10	75%a		✗	✗
Beatrice	4	Female	80		9	9	96%a	✗	✗
Alvita	4	Female	76	5		8	91%a	✗	✗
Arthur	4	Male	76		8	9	65%	✗	✗
Roger	4	Male	84	5		2	87%a	✗	✗
Ann	5	Female	74	7		6	78%a	✗	✗
Evelyn	5	Female	69	7		5	35%	✗
Flora	5	Female	79		8	5	30%	✗
John	4	Male	83	6		1	56%			✗
Sam	2	Male	88	5		2	51%			✗
Betty	4	Female	88	7		7	98%a			✗
Timothy	6	Male	70		9	7	5%			✗
Angelina	6	Female	73		9	9	65%			✗
Mark	7	Male	73		9	9	63%			✗
Jane	2	Female	84		8	10	35%			✗
Rachel	4	Female	83		6	11	82%a			✗
Eleanor	6	Female	68		8	12	30%			✗

Participant characteristics: session leaders

Following completion of the study, all (n = 15) session leaders and organisation providers were approached and asked to participate in either an interview or a focus group regarding their experiences of REACT. Twelve agreed to participate (Bath/Bristol, n = 4; Birmingham, n = 4; Devon, n = 4) but three had left their organisations and were uncontactable. In addition, one research team member who had delivered some of the health behaviour maintenance sessions with three groups (Devon) and two members of staff from one provider organisation (Bath/Bristol) were also interviewed. The total number of interviewees was 15 (male, n = 7; female, n = 8).

Thematic tree at 6, 12 and 24 months

The four predetermined central themes were supported by participant descriptions of their experiences of the REACT study and participation in daily physical activity. This section presents the four central themes, higher-order themes (HOTs) and subthemes (Figures 3 and 4) and the comparison of responses among participants classified as either pre-frail or frail and participants with low and high attendance, and presents these findings in the context of the REACT logic model.

FIGURE 3.

Thematic trees identified from participant experiences: reasons to engage in the REACT study and benefits of participation.

FIGURE 4.

Thematic trees identified from participant experiences: barriers to and enablers of participation.

Reasons to engage in REACT

Where relevant, and in an effort to further understand themes generated by participant response, we triangulate these data with relevant data provided by the session leaders.

The central theme, ‘reasons to engage in REACT’, comprised six HOTs: seeking better health, seeking social connectedness, appeal of research, appeal of physical activity, being influenced by others and having no reason not to take part in the 6-month interviews. Four HOTs persisted at the 12-month interviews: seeking better health, seeking social connectedness, appeal of research and appeal of REACT physical activity. This central-order theme was made up of 15 subthemes at the 6-month interviews and seven subthemes at the 12-month interviews. All 12-month subthemes remained from the 6-month interviews, with no new subthemes identified. A total of 19 subthemes remained at the 24-month interviews, with one new subtheme identified.

Reasons to engage in REACT for people classified as frail and pre-frail

Reasons for engagement in REACT were very similar between people classified as frail and people classified as pre-frail. At the 6-month interviews, health was the predominant reason to engage in REACT for those classified as frail:

And then I thought about it and I thought ‘well give it a try’, anything that will help me to walk better.

Cordelia, 6 months, SPPB score of 4, attendance 72%

By contrast, people classified as pre-frail based on their SPPB scores were more concerned with the social aspect of engaging in REACT and the appeal of taking part in research:

I thought it was interesting that they were studying older people to find out if they do have enough exercise ’cos I don’t think they do . . . I’ve done these sort of things before [participate in research] . . .

Iris, 6 months, SPPB score of 9, attendance 79%

At the 12-month interviews, participants classified as frail or pre-frail were concerned with losing the physical benefits that they experienced during the REACT programme:

I just wanted to carry on with what we’d been doing, because like you know I could see the benefits.

Beatrice, 12 months, SPPB score of 9, attendance 96%

Reasons to engage in REACT for people with low and high attendance

At the 6-month interviews, the key difference between people with low and people with high attendance was that those with high attendance reported that increasing their activity levels was a key reason to participate in REACT:

. . . just hoping to get a bit of exercise out it . . . yes, I think I’ve always wanted to be more active really . . .

Beatrice, 12 months, SPPB score of 9, attendance 96%

This was less commonly reported by those with low attendance, who more often reported improving health and avoiding age-related decline as key reasons for participation. At the 12-month interviews, those with high attendance had multiple reasons to attend, whereas those with low attendance tended to report fewer reasons to continue attending, citing social connectedness as one of the most important reasons.

Findings in the context of the REACT logic model

Participants reported concerns about their poor physical health affecting their independence and social connectedness. Consequently, they sought to participate in exercise that they perceived to be potentially beneficial. The group-based delivery was particularly appealing. Social interactions are proposed by the REACT logic model as a means of maximising enjoyment. At the 6-month interviews, participants confirmed the importance of social interaction and reported that it could be beneficial to their well-being and a key reason to engage in REACT. This was further confirmed at the 12-month interviews, at which participants highlighted the supportive role of this enhanced social connectedness in forming social support networks. This was reported as a motivator of continued participation and confirmed the logic model, which anticipated that changes in relatedness would mediate short-term REACT attendance. The appeal of physical activity reported at 6 months was strengthened with the perceived benefits of participation in REACT reported at 12 months, which motivated participants to maintain these benefits in the long term. This supports the logic model, which highlights that perceived benefits from participation in the REACT sessions would influence attendance of REACT sessions and levels of daily physical activity (see Figure 2). One finding, which was not identified in the logic model, was the appeal of participating in research. Initial curiosity in participation translated to appreciation of the value of the research beyond individual gain and a sense of obligation to continue to participate in REACT. These findings in the context of the logic model are shown in a modified REACT logic model (see Appendix 2, Figure 28).

Benefits of participation in REACT

Under the central theme ‘benefits to participation in REACT’, participants reported five main benefits: improved physical health, improved social connectedness, changes in knowledge and attitudes, better mental well-being and positive behaviour change. These benefits (HOTs) were reported at both the 6-month and the 12-month interviews. They were made up of 23 subthemes at the 6-month interviews, with 19 subthemes identified at the 12-month interviews: four were new subthemes and 15 remained from the 6-month interviews. A total of 16 subthemes remained at the 24-month interviews, with the emergence of one new subtheme (see Figure 3).

At the 6-month interviews, those classified as pre-frail reported experiencing more physical health benefits, whereas those classified as frail seemed to experience more benefits related to mental well-being and social connectedness. Similarly, at the 12-month interviews, participants classified as pre-frail reported more physical health improvements than participants classified as frail:

Yes it was . . . very interesting, sometimes challenging uh, it was interesting to see what your body could do and how you improved, you know we all improved balance wise.

Beatrice, 12 months, SPPB score of 9, attendance 96%

At the 24-month interviews, participants identified four main categories of physical activity participation:

1. walking (for leisure or active travel)

2. everyday activity (housework, gardening)

3. structured exercise (group or gym based)

4. independent exercises (prescribed by a health-care professional, learned through the REACT programme).

Although not described as physical activity, social activities reported by participants included meeting friends, caring for family and friends, being church members and being members of community groups.

At 24 months, physical activity maintenance was related to specific benefits, including increased mobility and ability to carry out ADLs. Experiencing physical gains from physical activity promoted feelings of competence, which further facilitated physical activity maintenance:

My health has got better, I’m more motivated and I can get around a lot better. I don’t use my stick now whereas I always had my stick. I just got better and better [during REACT], now I walk all round the cycle track.

Geraldine, 24 months, SPPB score of 10, attendance 64%

Four participants reported increased physical activity levels post REACT. Three attributed improved motivation to be active directly to REACT. They described multiple benefits gained physically, developing new friendships and engaging in new activities (e.g. flower arranging and bowling). Two individuals wanted to capitalise on the benefits gained:

I think because I enjoyed REACT so much, and because before I used to do a tiny bit of exercise, but I enjoyed it so much and I just thought it’s a waste if I just leave it.

Dorothy, 24 months, SPPB score of 12, attendance 72%

People with both high and low SPPB scores reported experiencing improved self-efficacy for physical activity, improved social connectedness and increased physical activity levels:

I think what we did at REACT programme did give me bit of more confidence in walking.

Darsha, 12 months, SPPB score of 9, attendance 81%

Being supported to reach achievable goals enhanced participants’ confidence. One participant was spurred on to do more vigorous physical activity (running):

I was really getting better. So, I better do something with my good health . . . They [REACT] have opened my mind on what I can do. I cannot think of not going to exercise or not go for a walk or anything. I just have to go and do it and I am happy to do it.

Angelina, 24 months, SPPB score of 9, attendance 65%

REACT prompted participants to self-monitor and seek opportunities for exercise outside the REACT programme, in which support from session leaders was important to maintain self-efficacy:

It’s giving me an incentive to be more aware of what I need to do . . . I don’t exercise but I do stretching exercises . . . I never did it before . . . this basically has to do with REACT encouraging me. In a way you know, otherwise I wouldn’t be doing it because I never did.

Eleanor, 24 months, SPPB score of 12, attendance 30%

I said [to session leader] that I hadn’t really enjoyed it so I wanted to do something much more active, so he said, come to the gym, and then I found out I was in the gym on my own and I didn’t like that. So he then said well do aqua aerobics on a Friday.

Jane, 24 months, SPPB score of 10, attendance 35%

Benefits of participation in REACT findings and the REACT logic model

Participants experienced physical, social, emotional and behavioural benefits that supported the mediating role of perceived benefits on REACT attendance and physical activity outcomes, as stated in the REACT logic model. Physical activity outcomes or positive behaviour change in the form of increased physical activity levels and trying new exercise classes were mediated by perceptions of improved self-efficacy for physical activity and improved motivation for physical activity, with participants feeling more capable and motivated to exert autonomy over their choice to participate in daily physical activity and new exercise classes. The logic model proposed that relatedness would act as a mediator of physical activity outcomes. Participants indicated that improved social connectedness not only was a benefit of participation in REACT, but also enhanced enjoyment of REACT and acted as a driver for continued participation. The findings suggested that connectedness, shared identity and sense of relatedness were moderated by the specific intervention groups that participants were members of. The logic model currently identifies site characteristics that could moderate REACT attendance and physical activity outcomes; however, the moderating role of the specific intervention group that participants join on social connectedness and relatedness was not specifically identified. Group dynamics and coherence may differ among different groups and their members, and these processes need more attention and targeting. These findings are shown in the context of the logic model in a modified REACT logic model (see Appendix 2, Figure 29).

Barriers to participation in REACT and daily physical activity

Under the central theme ‘barriers to participation in REACT and daily physical activity’, participants identified nine main barriers (HOTs): being time poor, declining mental well-being, physical health barriers, REACT programme features, difficulty accessing transport, cost of daily physical activity, negative life experiences, bad weather and proposed reasons for dropouts at the 6-month interviews. These HOTs comprise 24 subthemes at 6 months. At 12 months, all nine HOTs identified at 6 months remained, in addition to three new HOTs: not perceiving benefits, barriers in the local area and lack of past physical activity. The 12 HOTs at 12 months comprised 32 subthemes, nine newly identified at 12 months, while 23 remained from the 6-month interviews. At the 24-month interviews, 12 subthemes remained, with the emergence of one new subtheme. Although all HOTs and subthemes within this central theme are presented in Figure 4, only the key themes in terms of addressing the research questions of this evaluation are discussed below. Furthermore, we distinguish between barriers that influenced attendance at REACT sessions and those that affected daily physical activity participation.

Barriers to participation in REACT and daily physical activity and the REACT logic model

Although people classified as either frail or pre-frail experienced physical health barriers to REACT and daily physical activity, those classified as frail experienced several health problems simultaneously:

My muscles, because I have this muscle weakness . . . because like sitting down I’m fine but getting up I struggle . . . I used to fall over quite often . . . I’d lose my balance like . . . And then I’ve seems to got no muscles to hold me back . . . I’m just blop! [laughs] . . . that’s my big problem.

Alvita, 12 months, SPPB score of 5, attendance 91%

In addition to physical health barriers, those classified as frail also reported experiencing other barriers to participation, such as having other commitments and difficulty accessing transport. A difference in physical health barriers experienced between people classified as frail and people classified as pre-frail was less evident in the 12-month interviews, with both groups reporting similar physical health barriers. Although the barriers to physical activity for participants classified as frail were dominated by physical health barriers in the 12-month interviews, participants classified as pre-frail reported barriers that related to the REACT programme features, including issues with some venues, some REACT session leaders, the delivery of the health behaviour maintenance programme, some faulty pedometers and moving from two sessions per week to one session per week from week 12 onwards:

I’m a bit disappointed in we had it [health behaviour maintenance programme] a couple of times when it first started but most of the people don’t seem to want to stay behind you know and have a coffee and a chat . . . They seem to have other things they want to do.

Cecil, 6 months, SPPB score of 4, attendance 65%

One participant recounted there being no formal discussion of goal-setting:

. . . we have talked about the step counter [pedometer] uh but we’ve sort I don’t think we’ve never talked about the actual, about any actual goals you know.

Beatrice, 6 months, SPPB score of 9, attendance 96%

The barriers that were related to REACT programme features were confirmed in session leader interviews, in which it was reported that, although the use of pedometers was popular in some groups, a number of groups faced difficulties with poor-quality pedometers that did not accurately record steps.

Participants classified as frail reported experiencing numerous and chronic health problems that remained throughout the REACT programme. By contrast, those classified as pre-frail often reported experiencing either acute illness or injuries that were relatively short-lived or health concerns that were stable or they were accustomed to dealing with, such as diabetes. At 12 months, both participants classified as frail and those considered pre-frail shared their experiences of the REACT programme and identified where they faced challenges. These experiences also varied depending on the site of the intervention group and how some session leaders delivered different components of the programme. Reports from interviews with session leaders also noted the challenges involved when working with older adults experiencing numerous physical health barriers, which often included adapting exercise:

When you’re working with 15 people with multiple health conditions you need to be flexible with your delivery.

At the 24-month interviews, physical health barriers were the most cited reasons for lapses in physical activity participation, together with leading busy lives and having competing commitments. Participants who articulated strategies to overcome ‘tensions’ to physical activity maintenance had both (1) lived active lives and (2) been engaged in a wide array of activities prior to REACT. Self-regulatory strategies to overcome challenges, such as declining health, competing commitments, motivational barriers and adverse weather, included adjusting expectations or actual physical activity, self-monitoring and making plans. These are discussed in more detail in Enablers of participation in REACT and daily physical activity.

Barriers to participation in REACT and daily physical activity for people with low and high attendance

Although the responses between the two groups were similar in the 6-month interviews, in the 12-month interviews there was a distinction between participants with high attendance scores, who tended to experience a single barrier to participation (predominantly physical health), and participants with low attendance scores, who reported multiple barriers to participation (predominantly physical health barriers and having other commitments). When considering participants who were in both the low SPPB score and the low attendance strata, it is important to note that two of the three participants interviewed at 6 months dropped out shortly before the 6-month interview, having experienced multiple barriers to participation relating to physical health, time commitments and travel difficulties. The third participant dropped out a few months after the 6-month interview, having experienced significant physical health barriers and reporting having other commitments. None of the three was available for follow-up at the 12-month interviews.

Being time poor and its associated subthemes – having competing commitments and not having the time – were expected to be a barrier to physical activity, especially in the short term; however, BCTs were incorporated into the REACT programme to help participants to identify and resolve sources of tension around increasing physical activity. Reports that time was a barrier to participation remained throughout the 12 months, indicating that participants were not able to successfully utilise BCTs, the BCTs were not suitable for resolving these tensions or the BCTs were not delivered as planned. The intervention fidelity study, which assessed delivery fidelity of health behaviour maintenance sessions (not exercise sessions), triangulates this finding, as it demonstrated that BCTs, such as managing setbacks and problem-solving, were delivered with low fidelity. Furthermore, reports from session leader interviews at follow-up indicated that reasons for poor delivery fidelity could include (1) not feeling comfortable delivering the content or (2) participants not being engaged with the health behaviour maintenance sessions. This, together with real life issues (e.g. lack of access to transport or competing interests) that participants may have faced, may explain why issues with time availability and tensions with having competing commitments remained at 12 months:

I mean I have been playing indoor bowls quite recently up and till about a month ago but I’ve had to give up my car so I can’t get around quite as much so it looks like I am not able to drive. And also I’ve got to find someone to take me.

BT2013, 24-month interview

I didn’t pursue any others. I should have done, I think, but it may be difficult to explain this to you but despite my age I’m very busy, I don’t have much spare time. I just deferred starting my last book due to lack of time, I’m going to write it some time. I’ve written 16 altogether but . . . I’ve got one more I want to write but certainly things get in the way, but I’ll get there. I’ve got to make sure I live long enough to do it!

AT2624, 24-month interviews

Although REACT programme features such as issues with the REACT session leader and issues with REACT structure and content were not reported to have affected REACT attendance, REACT venue issues were prohibitive of full engagement in REACT sessions. This shows how context in the form of intervention provider and site can, by moderating the way in which interventions are implemented, affect the way that participants experience an intervention and engage with it, as illustrated in the logic model. For example, in the case of issues with REACT structure and content, reports suggest that in some cases BCTs (managing setbacks and problem-solving/goal-setting and action-planning) were not discussed widely in some groups, suggesting a lack of competent delivery of health behaviour maintenance sessions on behalf of the REACT session leader, which confirms findings from the intervention fidelity study and reports from session leaders that some could not remember content from health behaviour maintenance training sessions. The intervention fidelity study showed that the delivery of action-planning and goal-setting was rated as needing improvement (mean score 2.2, SD 1.14), while managing setbacks and problem-solving were delivered with low fidelity (mean score 1.9, SD 0.81).

Participants did not report the delivery fidelity of these BCTs during health behaviour maintenance sessions as being prohibitive of either daily physical activity or REACT attendance; however, intervention outcomes may indicate whether or not this influenced long-term physical activity outcomes as the logic model anticipates it could. Participants with low motivation for physical activity or low self-efficacy for physical activity did not report this as a barrier to REACT attendance, which suggests that participants had the motivation for specifically engaging with REACT and felt a sense of self-efficacy and competence in the REACT context. It is expected that this would not necessarily translate to physical activity in other contexts. This supports the logic model, which anticipated that changes in intrinsic motivation and competence/self-efficacy would mediate changes in physical activity outcomes, in either a negative or a positive feedback loop. Participants reporting physical health barriers seemed to overcome them during REACT classes. This is perhaps because feelings of low motivation for physical activity or low self-efficacy for physical activity were reported more in relation to daily physical activity rather than in REACT itself (i.e. physical health barriers were not further compounded by low self-efficacy for physical activity and low motivation for physical activity). Many participants reported that having the autonomy to manage their own exercise throughout the class enabled them to overcome physical health barriers. Furthermore, the fact that session leaders were able to adapt exercises for those struggling with physical health barriers supported the mediating effect that participants’ autonomy over their route of progression and ability to carry out an adapted exercise had on REACT attendance.

Participants and session leaders reported difficulty in accessing transport as an important contextual variable affecting REACT participants, who used a range of strategies to overcome this barrier, such as carpooling and accessing community transport.

The organisation of carpooling is an example of participants building support networks to enable REACT attendance. The logic model anticipated that this could be a determinant of physical activity outcomes and could be mediated by feelings of relatedness among REACT peers. Findings from ‘barriers to participation in REACT and daily physical activity’ in the context of the logic model are shown in a modified REACT logic model (see Appendix 2, Figure 30).

Enablers of participation in REACT and daily physical activity

Within the subtheme ‘enablers of participation in REACT and daily physical activity’ there was a clear distinction between pre-existing motivation and components of the REACT project targeting participant motivation. This was reflected in the two HOTs, characteristics of REACT and personal and psychosocial characteristics of participants, including 10 subthemes. At the 12-month interviews, all 10 subthemes remained, with the addition of a new subtheme. At the 24-month interviews, 11 subthemes remained, with the emergence of two new subthemes (see Appendix 2, Figure 31).

Enablers of participation in REACT and daily physical activity for people classified as frail and pre-frail

At the 6-month interviews, participants classified as either frail or pre-frail reported similar enablers of participation to REACT. However, both groups demonstrated a shift in importance of enablers from the 6-month to the 12-month interviews. At 6 months, the predominant enabling factor was the support received from REACT session leaders; however, although this is also discussed at the 12-month interviews, the emphasis was on enjoyment of REACT and supportive group dynamics:

Just being with everybody and seeing how we could all do, and everybody said ‘oh well done’ . . . we all encouraged each other I think you know.

Beatrice, 12 months, SPPB score of 9, attendance 96%

This suggests a transition from REACT session leader-supported participation to a group network of support for physical activity. When focusing on daily physical activity, participants reported that their REACT session leaders’ encouragement and support of other activities within the community were important factors in their continuation of activity independent of REACT. Furthermore, having motivations to return to previous physical activity behaviours seemed to be enabling of daily physical activity for both frail and pre-frail participants. Previous physical activity behaviours were also important at the 24-month interviews, with participants who overcame barriers to physical activity participation being those who had been historically active and engaged in a variety of activities before enrolling in REACT.

In contrast to reports at the 6-month and 12-month interviews, in which self-regulatory techniques were not commonly cited, at the 24-month interviews self-regulatory techniques were reportedly used in an effort to overcome challenges such as declining health, competing commitments, motivational barriers and adverse weather. These self-regulatory strategies included:

• Adjusting expectations of or actual physical activity in accordance with their health limitations and other commitments –Valerie, 24 months, SPPB score of 9, attendance 61%

  If I can’t [attend exercise class] then I shall do what I do now which is to do light exercises for my back before I get out of bed in the morning and walk and [partner’s] corridor is about the same length as my path so we walk up and down that several times . . .

• Walking indoors was a coping strategy for staying active when the weather was poor –Darsha, 24 months, SPPB score of 8, attendance 81%

  I make a habit of a daily walk you know, of course if the weather is very very bad. Like it was so windy I try to go but then I just walk up and down the corridor.

• Doing ‘something rather than nothing’ was another coping strategy –Frederick, 24 months, SPPB score of 9, attendance 90%Rachel, 24 months, SPPB score of 11, attendance 82%Frederick, 24 months, SPPB score of 9, attendance 90%

  . . . now instead of just going and getting my paper and coming straight back, I could do a circuit if it’s a nice day, you know? I’m not counting them [steps] now but it’s still reasonable.

  You have to make yourself sometimes, you know I think to myself, ‘come on get up get up’ and I do I reckon every day I reckon to go out somewhere, even if I only walk up the village and back down.

  I’m pretty active in silly ways. It sounds nothing, but I don’t use the lift . . . it is a small thing . . . so not to use the lift, I have to walk to the far end of the building . . . Go down the stairs and walk back half-way through the building. It’s far enough to be an activity.

• Repeating the same walking route enabled participants to self-monitor their progress –Mark, 24 months, SPPB score of 9, attendance 63%

  That’s a favourite walk because we can sort of tell where we were last year and last week.

• ‘Habit’, ‘routine’ and ‘making plans’ were key enablers for physical activity, whereas participants did stress that not being as able as before was evident in generally being slow with all activities, which led to frustration –Dorothy, 24 months, SPPB score of 12, attendance 72%

  I make a plan to walk from home to (city centre) and back . . . and then at home I walk, because I have a lot of stairs, I walk up and down, up and down until I get tired.

Post REACT at the 24-month interviews, individuals who reported frequent physical activity engagement described gaining pleasure and enjoyment from physical activity:

I love seeing things, I love being out in the fresh air I love noticing things. I don’t get up and say right I’ve gotta do a couple of miles this morning, no I go when I want to enjoy myself.

Jane, 24 months, SPPB score of 10, attendance 35%

Another participant described how physical activity was an enabler to participation in meaningful activities:

I’ve got no real interest in . . . just exercise as exercise. My interest was always in exercise really incidental to some proper activity, and that would be a social activity among people I liked.

Frederick, 24 months, SPPB score of 9, attendance 90%

Despite the importance of a supportive group network as an enabling factor during the REACT programme, post REACT, social support was a less prominent theme in explaining physical activity maintenance. None of the interviewed participants had utilised support from community agencies (e.g. health visitors, health friend scheme):

I personally did not need it, because you know I am quite fulfilled with the activities that I am doing, and then I go on holiday and visit my daughters and granddaughters.

Arman, 24 months, SPPB score of 8, attendance 95%

For some participants, friendships developed during REACT and motivational support from friends and spouses facilitated continued physical activity engagement:

Well [friend] bullied me . . . she went and it [attending REACT] was a good way of meeting up and perhaps getting a coffee after . . . and then it developed from there . . .

Valerie, 24 months, SPPB score of 9, attendance 61%

I’m not very good at self-motivation . . . if you said, oh, you know, go for a run every morning I wouldn’t, but if the whole group of us were meeting and I said, ‘oh we’re going to go for a short walk meet at 10 o’clock at [location]’ then I’d go.

Rita, 24 months, SPPB score of 10, attendance 75%

Illness also brought about changes in some friendships and had an impact on people’s social activities:

It was quite hard for me because she [friend] used to come and ring me up every day and we’d meet for an evening meal nearly every day, and all of a sudden it was gone!

Rita, 24 months, SPPB score of 10, attendance 75%

Social isolation and lack of support can have detrimental effects, as vividly reported by one participant:

I’m less active than I ever was in the past and it’s getting worse because I have no friends, no relatives. I don’t have children . . . They’ve all died or they’ve got hip or knee problems . . . or they’re involved with family, you know, so I have to create things to do to keep myself occupied but in [city] it’s a big fat zero. I have nothing, no one, and I don’t want to go walking by myself because it can be dangerous. If I fall there’s no one you see?

Eleanor, 24 months, SPPB score of 12, attendance 30%

Enablers of participation in REACT and daily physical activity of people with low and high attendance

It was apparent from both the 6-month and the 12-month interviews that people with high attendance were more motivated by REACT than people with low attendance. This indicates that motivation could act to mediate REACT attendance but barriers could moderate the positive loop between motivation and REACT attendance, as we observed that people in the low-attendance group experienced more physical health barriers. The role of supportive group dynamics was reported more often in the 12-month interviews by people in the high-attendance group. Three participants in the group with low attendance and low SPPB score who reported supportive group dynamics at the 6-month interviews still dropped out, showing that, although supportive group dynamics may be a strong enabling factor, it might not be sufficient to overcome the multiple barriers (physical health, having competing commitments and difficulties accessing transport) that these participants may had faced before they dropped out. People with high attendance reported more participation in previous physical activity behaviours, suggesting a moderating role of previous physical activity behaviours in the attendance of REACT sessions in the short term; however, at the 12-month interviews, this relationship was no longer evident, suggesting that previous experiences may not be a moderator of long-term REACT attendance.

Enablers of participation in REACT and daily physical activity findings in the context of the REACT logic model

The logic model anticipated that enjoyment of REACT would mediate REACT attendance and physical activity outcomes. Participant reports suggest that enjoyment was a key reason to participate in REACT and start new physical activity classes beyond REACT. Participants reported being motivated by REACT, deriving motivation from their REACT session leaders who supported participants’ competence and sense of autonomy, and by watching their REACT peers improve their fitness levels. Participants’ perceived competence mediated both short-term and long-term physical activity outcomes. High levels of competence and self-efficacy at baseline, mainly among participants who had previously been active, may influence outcome expectations and need to be accounted for by the REACT logic model (see Appendix 2, Figure 31).

Key REACT programme components

The key programme components responsible for influencing participants’ lifestyles at 24 months are summarised below.

Behaviour change techniques

Techniques for managing slips/lapses, supporting habit change and resolving sources of tension around increasing physical activity were utilised by participants to support physical activity at 24 months. Participants were able to overcome ‘tensions’ to physical activity by employing self-regulatory strategies (e.g. incorporating more physical activity into daily life, finding alternative physical activity and monitoring techniques). Routine was an enabler for maintaining PA.

REACT facilitator delivery style

REACT session leaders used a person-centred delivery style to build autonomy/intrinsic motivation. Individuals with formerly low physical activity levels expressed increased motivation to be active through mastery and improved self-efficacy. This was sustained post REACT.

Group-based physical activity

The group-based format of the REACT programme promoted social connectedness among participants. The social benefits of physical activity were recognised by most participants. Individuals with established social networks reported more physical activity. The success of REACT in supporting sustainable social networks varied between groups. Where this was most successful, social contact continued post intervention.

Themes identified from session leader and organisation provider interviews

The REACT programme session leaders and organisation providers were interviewed at follow-up (12 months) to explore whether or not the intervention was delivered as planned and what, if any, changes could be made to the REACT programme for future delivery (research questions 4 and 5). The analysis of session leader interview data identified 10 themes, which address whether or not the programme was delivered as planned and propose further improvements to the programme components. The programme components presented are manual and training, one-to-one sessions and group exercise classes, health behaviour maintenance sessions, group size/common start time, equipment/recording attendance, ambassadors programme, non-REACT activities, continuation session, transport and cost/funding.

Manual and training

The REACT manual was a substantial document and was given to session leaders at the training session. It was designed to be used as a reference document throughout the 12-month programme. The manual could appear daunting at first sight, but the training session explained/eased its use and session leaders were very positive about using it:

. . . it scares a lot of people . . . when you’ve done a course you realise it’s actually quite simple.

Exercise specialist, AT9001, Male

Another session leader recounted:

I do refer to the training manual, I have it with me at all times.

Exercise specialist, AT9002, Male

Feedback on the REACT 2-day training course delivered by three or four members of the research team was largely positive, with session leaders stressing that it informed them of what they needed to deliver. Some session leaders thought that the sessions focusing on the exercise component of the programme might have been too detailed given that the leaders were well qualified. However, others thought that these sessions were a good refresher:

I found it very, in depth and possibly . . . I didn’t need it to be that in depth. But I didn’t mind sometimes you need reminders.

Exercise specialist, AT9002, Male

Suggestions for improvements included a practical demonstration of how an entire class would run, and provision of online materials for use during the programme delivery:

. . . maybe a demonstration of classes . . . this is what session one would look like.

Exercise specialist, AT9004, Female

Another suggestion for improvement was the provision of ‘an online resource; an online clip of an hour session can you just watch it when you got time to kind of format and layout’ (exercise specialist, AT9001, male).

Feedback was less detailed about the health behaviour change training. Some leaders did not recall the training session, and some recalled only some aspects:

I’m not quite sure what you mean. I’ve got to be honest I don’t.

Exercise specialist, BT001, Female

One-to-one sessions and group exercise classes

Feedback about these sessions, at which participants met with their session leader on a one-to-one basis, was overwhelmingly positive. One session leader reported that they had taken their participants on a familiarisation tour of the facility as part of the one-to-one session.

There was positive feedback about the structure of the classes, especially the warm-up and games, although ‘when you’re working with 15 people with multiple health conditions you need to be flexible with your delivery’ (exercise specialist, AT9003, male). Some sessions leaders reported that they had to make some adjustments, including a lengthier warm-up.

Reducing the exercise sessions from two sessions per week to one was perceived as challenging by some, who reported that participants did not want this reduction, as they were engaged with the programme and were making good progress. Some session leaders agreed that moving to one session reduced the rate of progression of the participants.

A change in session leader was challenging for the incoming leader, who reported that the groups were attached to their session leaders and were not happy with changes:

I took on the first group probably about 4/5 months into it, and I think it was quite hard to engage with them as an outside instructor coming into an already established group.

Exercise specialist, AT9003, Male

Health behaviour maintenance sessions

Feedback about the health behaviour maintenance sessions was mixed, with some session leaders reporting that they did not deliver these sessions as planned because they were uncomfortable with either the content or the format of the sessions:

I ended up just incorporating what needed to be covered within the classes ’cause I found it really hard to structure specific sessions for that.

Exercise specialist, AT9002, Male

Alternatively, session leaders reported that they did not deliver sessions as planned because their participants did not want to engage with them:

I’ve got to say a lot of the time they just wanted to have a coffee and a chat as a group, it was quite difficult to, you know, get all their attention.

Exercise specialist, ET9001, Male

However, some session leaders reported that, although it was challenging to deliver these sessions, on the whole participants enjoyed them:

. . . they had a few problems with them but on the whole they enjoyed that.

Exercise specialist, BT9004, Female

Group size/common start time

The fact that all REACT participants joined the group at the same time for the 12-month duration was viewed as both a positive (‘being in a group starting off starting off them as a group to get them to know one another which is fantastic’; exercise specialist, BT9003, male) and a negative (‘if a few people drop out . . . if you don’t have a flow of people to keep fresh blood, it’s just not sustainable’; exercise specialist, AT9003, male).

Equipment/recording attendance

Everyone was happy using Therabands (Akron, Ohio, USA), but getting ankle weights on and off participants was an issue: ‘they took quite a long time to get them on an off so we were kind of losing a bit of time’. Using pedometers as part of the health behaviour maintenance programme was popular, but some groups experienced issues with poor-quality pedometers that did not accurately record steps:

They really enjoyed having that kind of visual tool [map] to do it and they got really on board with counting their steps and I think that was good motivator.

Exercise specialist, AT9003, Male

There were issues around recording attendance at sessions. Some session leaders did not have access to online systems in the exercise room, so needed to keep paper-based registers. Several session leaders suggested the provision of an online register in future REACT delivery.

Ambassadors’ programme

Most session leaders found it difficult to engage participants in a formal ambassador programme because participants ‘didn’t want that full-on responsibility’. However, in many cases, some elements of the ambassador roles were taken up by participants in a more informal manner:

There wasn’t a specific ambassador . . . they just naturally started to pick their own roles.

Exercise specialist, BT9002, Male

I think it’s a really good idea because it just gives them that more sort of ownership over the programme. But it just didn’t work.

Exercise specialist, ET9001, Male

Non-REACT activities

A main focus of the REACT programme was to encourage participants to engage with other community activities. This was particularly the case when the frequency of sessions changed from twice per week to once per week. Session leaders reported some level of success with some groups, but some participants ‘were not easily persuaded and were very wedded to their group and session’ (exercise specialist, AT9002, male). However, ‘one or two of them jumped on quite quickly were very eager to do it’ (exercise specialist, AT9001, male). Despite some having concerns about new activities, ‘a lot of them were a bit worried about going out and trying other things’ (exercise specialist, AT9001, male).

Continuation sessions

Many groups did not want the sessions to end after the 12-month programme:

When I had to leave the group, it’s been a bit traumatic. They were really really upset. So the majority have continued (in some form) with participants paying to attend.

Exercise specialist, AT9002, Male

As a result, additional sessions were created by trusts for the REACT participants:

There was so much demand for the trust to continue with some form of exercise class that we created the [name] weekly exercise class [REACT continuation].

Exercise specialist, AT9008, Male

Transport

Despite the fact that during the REACT screening process potential participants were given specific information about the location of the REACT classes and asked if they would be able to attend sessions, some still faced transport issues. These were usually solved by community transport or, more commonly, through lift-sharing.

Cost/funding

Some session leaders reported that the programme was a large investment for a small number of people, but they acknowledged the importance of free sessions to improve engagement with exercise sessions:

But I think you need to hook them with the free yeah just to get them settled.

Exercise specialist, ET9001, Male

Study 3: quantitative evaluation of intervention mechanisms

Methods

Design

A detailed set of hypotheses (outlined in the following section) were derived from the logic model and quantitatively tested using a range of process measures, including questionnaires and accelerometer data collected at baseline and 6 (short term), 12 (medium term) and 24 months (long term). The hypotheses were stated and published ‘a priori’, prior to unlocking or any analysis of the REACT data set. 1 Some minor edits were made prior to analysis (e.g. owing to non-availability of some data), and these are indicated in footnotes to the hypotheses below.

Hypotheses for quantitative testing of the REACT logic model

The hypotheses were derived from the stated and implied assumptions in the REACT logic model. The time from baseline to 12 months represents the intervention period (0–6 months was the ‘initial change phase’ and 6–12 months was the ‘supported maintenance phase’) and the 12- to 24-month time frame represents the post-intervention (unsupported maintenance) phase.

Hypothesis 1 was that being in the intervention arm would lead to changes in physical activity (MVPA, steps and sedentary time) and engagement in muscle-strengthening exercise from 0 to 6 months.

Hypothesis 2 was that increased exposure to the intervention would correlate with increased change in physical activity (MVPA, steps and sedentary time) and engagement in muscle-strengthening exercise from 0 to 6 months.

Hypothesis 3 was that increased exposure to the intervention would correlate with increased change in physical activity (MVPA, steps and sedentary time), engagement in muscle-strengthening exercise and SPPB score from 0 to 12 months.

Hypothesis 4 was that increased exposure to the intervention would lead to increased maintenance of physical activity and engagement in exercise. Therefore, within the intervention arm, intervention dose would correlate negatively with decreases in physical activity, engagement in muscle-strengthening exercise and SPPB score from 6 to 12 months (during the supported maintenance period) and from 12 to 24 months (the unsupported maintenance period).

Hypothesis 5 was that exposure to the intervention would lead to changes in key psychosocial determinants of physical activity and exercise from baseline to 6 and 12 months:

• Compared with the control arm, the intervention arm would experience increases in autonomy, competence (self-efficacy), relatedness for MVPA and exercise, perceived intrinsic benefits of exercise (social, physical and emotional) and enjoyment of physical activity and exercise from 0 to 6 months. An originally intended measure of physical activity-related self-concept was removed from the questionnaire pack to reduce patient burden following review of the process evaluation measures by the TMG and patient and public involvement members. Hence the words ‘self-concept’ have been removed from the original wording of this hypothesis.

• Compared with the control arm, the intervention arm would experience increases in physical activity-related autonomy, competence (self-efficacy), relatedness for physical activity and exercise, perceived intrinsic benefits of physical activity and exercise (social, physical and emotional) and enjoyment of physical activity and exercise from 0 to 12 months. As with the hypothesis above, the words ‘self-concept’ have been removed from the original wording of this hypothesis.

• Increased exposure to the intervention (total contact time from baseline to the relevant time point) would correlate with increased change in the above determinants (and in the expected direction).

Hypothesis 6 was that the intervention effect on SPPB (intervention vs. control) would be mediated by changes in muscle-strengthening exercise and by changes in MVPA, changes in lower intensity physical activity or walking activity (Figure 5). The amount of variance in SPPB explained by the different types of activity/exercise will be of interest. The words ‘changes in balance and co-ordination exercise’ have been removed from the original wording of this hypothesis, as this could not be distinguished from muscle-strengthening exercise by the measure used.

FIGURE 5.

Causal diagram for mediators and moderators of intervention effects on physical function.

Hypothesis 7 was that the intervention effect on SPPB score and on physical activity and exercise from 0 to 6 and 0 to 24 months may be moderated by a number of potential moderating variables, including age, sex, ethnicity, baseline physical activity and SPPB, co-interventions, comorbidities, BMI, mental health, SES, education level (see Figure 5, and Appendix 2, Figure 30). The words ‘and on PA and exercise’ have been added to this hypothesis, as the moderating influence of contextual variables on changes in physical activity behaviours (as well as physical functioning) is of key scientific interest. The time point 0–12 months was removed to reduce the number of analyses. The moderation of short-term (0–6 months) and long-term (0–24 months) effects was considered to be of the most interest from the three time points available.

Participants

We applied analyses to the whole sample where data were available; however, some hypotheses applied only to the intervention arm.

Measures

Brief questionnaires assessing the following mechanisms of change suggested by SCT and SDT (the main theoretical underpinnings of the intervention model) were administered at baseline and 6, 12 and 24 months (see Table 2 for full details):

• autonomy in relation to physical activity

• competence for physical activity

• relatedness for physical activity

• enjoyment of physical activity

• perceived intrinsic benefits of physical activity (social, physical and emotional)

• autonomy for strength-building exercise

• competence for strength-building exercise

• relatedness for strength-building exercise

• enjoyment of strength-building exercise

• perceived intrinsic benefits of strength-building exercise (social, physical and emotional).

To test the hypotheses, we also used measures of intervention engagement (session attendance) and intervention effectiveness [SPPB scores, physical activity/time doing MVPA (by accelerometer and PASE questionnaire), walking (PASE walking subscale), sedentary time (accelerometer), self-reported engagement in muscle strength exercise (MSEQ adherence score)] and secondary outcomes of interest (cognitive function, falls, pain, sleep quality, well-being and grip strength).

Variables used to define individual context were age, sex, ethnicity, baseline physical activity, baseline SPPB, co-interventions, physical comorbidities, BMI, mental health status (any diagnosed mental health conditions in the 6 months prior to baseline), IMD (from postcode) and education level. All of the measures are described in detail in Chapter 2.

Analysis

The above hypotheses were tested using a variety of statistical analyses, as described in Table 10. To maximise sensitivity to change and avoid difficulties in constructing the mediation analyses (which require specialist software),97 the analyses testing hypotheses 1 to 19 have been constructed without accounting for clustering by exercise group within the intervention arm.

TABLE 10 - Mechanistic hypotheses and analysis methods used to test them

Hypothesis	Analysis
Hypothesis 1: the intervention will increase physical activity and muscle-strengthening exercise from 0 to 6 months	Comparison of change scores between intervention and control arms, using univariate ANCOVA models with the baseline value of the outcome, age (continuous variable), sex (male, female) and study site (Devon, Bath/Bristol, Birmingham) entered as covariates
Hypothesis 2: intervention exposure will correlate with change in physical activity and muscle-strengthening exercise from 0 to 6 months	Bivariate correlation between the outcome variable and intervention exposure (% of sessions attended) within the intervention arm only. Given the skewed distribution for intervention exposure (46/409 intervention participants did not attend any sessions), we categorised attendance as high (≥ 75%), medium (51–74%) or low (0–50%) and used ANCOVA analysis to look for a linear trend in the mean outcome scores for these three groups. These models also controlled for age, sex and study site (the covariates used in the primary trial analysis)
Hypothesis 3: intervention exposure will correlate with change in physical activity, muscle-strengthening exercise and SPPB from 0 to 12 months	The analyses were the same as for testing hypothesis 2, except for changes to the dependent variable (changes from 0 to 12 months rather than from 0 to 6 months)
Hypothesis 4: intervention exposure will correlate with maintenance of physical activity and engagement in exercise from 6 to 12 and from 12 to 24 months	The analyses were the same as for testing hypothesis 2, except for changes to the dependent variable
Hypothesis 5: intervention will change key psychosocial determinants of physical activity and exercise from baseline to 6 and 12 months. Such increases will correlate with intervention exposure within the intervention arm	Comparison of change scores between intervention and control arms, using univariate ANCOVA models with the baseline value of the outcome, age (continuous variable), sex (male, female) and study site (Devon, Bath/Bristol, Birmingham) entered as covariates
Hypothesis 6: the intervention effect on SPPB (and on key secondary outcomes) will be mediated by changes in muscle-strengthening exercise and in physical activity	Mediation analyses were conducted using linear regression models set up in SPSS v26, using the PROCESS v3.4 macro.97 Model 4 was selected, which uses a single regression model to estimate coefficients for (1) the direct effect of the independent variable (intervention arm) on the dependent variable (e.g. SPPB) and (2) the indirect effect of the independent on the dependent variable when mediated by a third variable (e.g. physical activity). Bootstrapping with 5000 samples was used to estimate CIs. Age, sex and study site were entered into the models as covariates
Hypothesis 7: the intervention effect on physical activity and exercise from 0 to 6 and from 0 to 24 months may be moderated by a number of contextual variables (11 variables specified)	Linear regression models were used to estimate the interaction effects of individual moderators with trial arm. We followed procedures for moderation analysis recommended by Hayes.97 Any significant interactions were probed by calculating point estimates at the 16th, 50th and 84th centiles for continuous variables (equivalent to the mean plus or minus one SD for normally distributed data) or at distinct category values where categorical moderators were included. Any significant interactions were visualised to aid interpretation and checked for floor or ceiling effects. The analyses were conducted in SPSS V26 using the PROCESS macro.97 The p-value representing significance was not adjusted for multiple testing as these analyses are considered to be exploratory (as indicated by the words ‘may be’ in the hypothesis). It is worth noting that adjusting for the 11 tests conducted at each time point/for each outcome would yield a p-value of 0.0045. Any findings not meeting this criterion are, therefore, potentially vulnerable to analytic bias owing to multiple testing. Note that, for these analyses, ethnic groups were collapsed into two groups owing to numbers being too small to allow finer-grained analysis: (1) black, Asian, Chinese and other ethnic groups and (2) white

Results

Hypothesis 1 to hypothesis 5: effects of the intervention on mediators of lower limb physical function

Full data tables are available in Report Supplementary Material 2. Key statistics are provided in the following text for significant findings only.

Hypothesis 1

Hypothesis 1 was that being in the intervention arm would lead to changes in physical activity (MVPA, steps, sedentary time) and engagement in muscle-strengthening exercise from 0 to 6 months.

There was mixed evidence in support of hypothesis 1, with measures of self-reported physical activity, walking and muscle-strengthening exercise appearing to be more sensitive to the intervention than accelerometer measures. Unbouted MVPA seemed to be more sensitive than bouted MVPA to the intervention for the accelerometer measures. Hence, unbouted MVPA was used to assess changes in physical activity in testing the following hypotheses (hypotheses 2–7). Based on accelerometer data, there were no significant differences between the intervention arm and the control arm in bouted MVPA (in bouts of at least 10 minutes) or sedentary time at 6 months, controlled for baseline MVPA, site, sex and age. However, there was a close to significant difference of 20.2 minutes of unbouted MVPA per week (2.88 minutes per day, 95% CI 5.82 to –0.06 minutes per day; p = 0.054) favouring the intervention arm.

Using questionnaire data, significant differences favouring the intervention arm were observed in self-reported physical activity (PASE total score) (mean difference 16.54 points, 95% CI 7.89 to 25.20 points; p < 0.001), self-reported walking (PASE walking outside items: days per week × hours per day) (mean difference 0.727 hours per week, 95% CI 0.293 to 1.162 hours per week; p = 0.001) and self-reported muscle-strengthening exercise (MSEQ adherence score) (mean difference 0.73 scale points, 95% CI 0.42 to 1.04 scale points; p < 0.001).

Hypothesis 2

Hypothesis 2 was that increased exposure to the intervention would correlate with increased change in physical activity (MVPA, steps, sedentary time) and engagement in muscle-strengthening exercise from 0 to 6 months.

Hypothesis 2 was broadly supported, with a consistent pattern of data across all measures (albeit with two out of the five measures achieving only marginal significance in the linear trend analysis). There were significant, weak correlations between intervention attendance and changes from 0 to 6 months in unbouted MVPA (R = 0.147, p = 0.012, n = 292), PASE total score (R = 0.136, p = 0.012, n = 339) (see Appendix 2, Figure 34), hours per week spent walking outside (from PASE walking items) (R = 0.165, p = 0.002, n = 332) and change in MSEQ adherence score (R = 0.122, p = 0.029, n = 320), but not in sedentary time (R = –0.063, p = 0.290, n = 281).

Significant linear trends, whereby positive changes in outcomes were associated with increased session attendance, were identified for unbouted MVPA (Figure 6; difference between low and high attenders 9.8 minutes per day, p = 0.002 for the linear trend, n = 265), PASE walking outside (see Appendix 2, Figure 32) (difference between low and high attenders: 1.24 hours per week, p = 0.011, n = 302) and MSEQ adherence score (see Appendix 2, Figure 33) (difference between low and high attenders 0.82 points, p = 0.049, n = 293). Linear trends that were in the predicted direction were observed for 0- to 6-month changes in sedentary time (difference between low and high attenders 28.7 minutes per day, p = 0.056, n = 255) and PASE total score (difference between low and high attenders 17.8 points, p = 0.051, n = 308).

FIGURE 6.

Changes in daily unbouted MVPA from 0 to 6 months for different levels of attendance within the intervention arm (error bars: 95% CI).

Hypothesis 3

Hypothesis 3 was that increased exposure to the intervention would correlate with increased change in physical activity (MVPA, steps, sedentary time), engagement in muscle-strengthening exercise and SPPB score from 0 to 12 months.

Hypothesis 3 was broadly supported, although not with respect to accelerometer-assessed sedentary time and with two out of the five measures achieving only marginal significance in the linear trend analysis. There were significant, weak correlations between intervention attendance and changes from 0 to 12 months in unbouted MVPA (R = 0.174, p = 0.004, n = 278), PASE total score (R = 0.134, p = 0.015, n = 329), PASE walking score (R = 0.158, p = 0.004, n = 321), MSEQ adherence score (R = 0.252, p < 0.001, n = 315) and SPPB score (R = 0.175, p = 0.001, n = 346), but not in sedentary time (R = –0.076, p = 0.215, n = 268).

Significant linear trends, whereby positive changes in outcomes were associated with increased session attendance, were identified for PASE total score (see Appendix 2, Figure 35) (difference between low and high attenders 17.9 points, p = 0.043 for the linear trend, n = 301), MSEQ adherence score (Figure 7) (difference between low and high attenders 1.28 points, p = 0.001, n = 292) and SPPB (Figure 8) (difference between low and high attenders 0.89 points, p = 0.004, n = 316). A linear trend that was in the predicted direction was observed for 0- to 12-month changes in unbouted MVPA (difference between low and high attenders 5.1 minutes per day, p = 0.051, n = 256) and for the PASE walking outside score (difference between low and high attenders 0.805 hours per week, p = 0.083, n = 295). No significant trend was observed for sedentary time.

FIGURE 7.

Changes in muscle-strengthening exercise (MSEQ adherence score) from 0 to 12 months for different levels of attendance in the intervention arm (error bars: 95% CI).

FIGURE 8.

Changes in SPPB total score from 0 to 12 months for different levels of attendance in the intervention arm (error bars: 95% CI).

Hypothesis 4

Hypothesis 4 was that increased exposure to the intervention would lead to increased maintenance of physical activity and engagement in exercise. Therefore, within the intervention arm, intervention dose would correlate negatively with decreases in physical activity, engagement in muscle-strengthening exercise and SPPB score from 6 to 12 months (during the supported maintenance period) and from 12 to 24 months (the unsupported maintenance period).

Hypothesis 4 was not supported by the data (in 13 out of 14 analyses). These analyses were applied in the intervention arm only. No significant correlations were found between percentage session attendance and changes from 6 to 12 months in bouted MVPA, unbouted MVPA, sedentary time, PASE total score, PASE walking score, MSEQ adherence score or SPPB score. There was a weak negative correlation between percentage session attendance and changes in bouted MVPA from 12 to 24 months (R = –0.125, p = 0.047, n = 252). No other changes in outcomes from 12 to 24 months were significantly correlated with session attendance. No significant linear relationship were found for any of the outcomes across the three attendance groups, controlling for age, sex, study site and baseline SPPB score.

Hypothesis 5

Hypothesis 5 was that exposure to the intervention would lead to changes in key psychosocial determinants of physical activity and exercise from baseline to 6 and 12 months.

Hypothesis 5a

Hypothesis 5a was that, compared with controls, the intervention arm will experience increases in autonomy and competence (self-efficacy); relatedness for MVPA, exercise and perceived intrinsic benefits of exercise (social, physical and emotional); and enjoyment of physical activity and exercise from 0 to 6 months.

Hypothesis 5a was broadly supported: we saw an intervention effect on most psychosocial determinants specified by the REACT logic model at 6 months. However, no signal was detected in relation to autonomy (for either muscle-strengthening or MVPA) or for enjoyment of muscle-strengthening exercise.

In relation to MVPA, at the 6-month follow-up, there was a significantly greater increase in the intervention arm compared with the control arm for perceived competence (mean difference 0.343 points, 95% CI 0.148 to 0.539 points, p = 0.001, n = 627), relatedness (mean difference 0.278 points, 95% CI 0.138 to 0.419 points, p < 0.001, n = 610) and enjoyment of MVPA (mean difference 0.149 points, 95% CI 0.015 to 0.282, p = 0.030, n = 627). However, there was no significant difference in autonomy related to MVPA (mean difference 0.065 points, 95% CI –0.040 to 0.170 points, p = 0.223, n = 625).

In relation to muscle-strengthening exercise, at the 6-month follow-up, there was a significant increase in the intervention arm compared with the control arm for perceived competence (mean difference 0.705 points, 95% CI 0.490 to 0.921 points, p < 0.001, n = 625) and relatedness (mean difference 0.283 points, 95% CI 0.040 to 0.525 points, p = 0.023, n = 254). The perceived intrinsic benefits of exercise were higher in the intervention arm than in the control arm at 6 months (mean difference 0.862, 95% CI 0.672 to 1.053, p < 0.001, n = 609). However, there were no significant differences in enjoyment (mean difference 0.185 points, 95% CI –0.037 to 0.407 points, p = 0.103, n = 261) or in autonomy related to muscle strengthening (mean difference –0.042 points, 95% CI –0.199 to 0.115 points, p = 0.597, n = 257).

Hypothesis 5b

Hypothesis 5b was that, compared with control subjects, the intervention arm will experience increases in physical activity-related autonomy and competence (self-efficacy); relatedness for physical activity, exercise and perceived intrinsic benefits of exercise (social, physical and emotional); and enjoyment of physical activity and exercise from 0 to 12 months.

Hypothesis 5b was broadly supported: we saw an intervention effect on most psychosocial determinants specified by the REACT logic model at 12 months. However, no signal was detected in relation to autonomy (for either muscle-strengthening or MVPA) or enjoyment of MVPA.

In relation to MVPA, at the 12-month follow-up, there was a significant increase in the intervention arm compared with the control arm for perceived competence (mean difference 0.279 points, 95% CI 0.073 to 0.486 points, p = 0.008, n = 615) and relatedness (mean difference 0.211 points, 95% CI 0.064 to 0.357 points, p = 0.005, n = 594). However, there was no significant difference in autonomy related to MVPA (mean difference –0.052 points, 95% CI –0.064 to 0.168 points, p = 0.378, n = 613) or in enjoyment of MVPA (mean difference 0.083 points, 95% CI –0.051 to 0.218 points, p = 0.225, n = 612).

In relation to muscle-strengthening exercise, at the 12-month follow-up, there was a significant increase in the intervention arm compared with the control arm for perceived competence (mean difference 0.625 points, 95% CI 0.403 to 0.847 points, p < 0.001, n = 609), relatedness (mean difference 0.438 points, 95% CI 0.196 to 0.679 points, p < 0.001, n = 265) and enjoyment (mean difference 0.250 points, 95% CI 0.023 to 0.476 points, p = 0.031, n = 268). The perceived intrinsic benefits of exercise were also higher in the intervention arm than in the control arm at 12 months (mean difference 0.757, 95% CI 0.565 to 0.950, p < 0.001, n = 598). However, there were no significant differences for autonomy related to muscle-strengthening (mean difference –0.023 points, 95% CI –0.174 to 0.128 points, p = 0.769, n = 268).

Hypothesis 5c

Hypothesis 5c was that increased exposure to the intervention (total contact time from baseline to the relevant time point) will correlate with increased change in the above determinants (and in the expected direction).

Hypothesis 5c was largely supported in relation to muscle-strengthening exercise but not in relation to MVPA.

In relation to MVPA, at the 12-month follow-up, we identified one significant linear trend, whereby an increase in MVPA relatedness was associated with increased session attendance (Figure 9) (difference between low and high attenders 0.564 points in the predicted direction, p = 0.002 for the linear trend, n = 295). No significant trend was observed for MVPA autonomy, MVPA competence or MVPA enjoyment.

FIGURE 9.

Changes in relatedness in relation to MVPA from 0 to 12 months for different levels of attendance in the intervention arm (error bars: 95% CI).

In relation to muscle-strengthening exercise, at the 12-month follow-up, we identified significant linear trends whereby positive changes in psychosocial determinants were associated with increased session attendance for muscle-strengthening competence (Figure 10) (difference between low and high attenders 0.633 points in the predicted direction, p = 0.021, n = 298), relatedness (Figure 11) (difference between low and high attenders 0.619 points in the predicted direction, p = 0.041 for the linear trend, n = 149), enjoyment (Figure 12) (difference between low and high attenders 0.745 points in the predicted direction, p = 0.014, n = 153) and perceived benefits of muscle-strengthening exercise (Figure 13) (difference between low and high attenders 1.142 points in the predicted direction, p < 0.001, n = 291). No significant trend was observed for autonomy for muscle-strengthening exercise.

FIGURE 10.

Changes in competence in relation to muscle-strengthening exercise from 0 to 12 months by level of attendance in the intervention arm (error bars: 95% CI).

FIGURE 11.

Changes in relatedness for muscle-strengthening exercise from 0 to 12 months by level of attendance in the intervention arm (error bars: 95% CI).

FIGURE 12.

Changes in enjoyment in relation to muscle-strengthening exercise from 0 to 12 months by level of attendance in the intervention arm (error bars: 95% CI).

FIGURE 13.

Changes in perceived benefits of muscle-strengthening exercise from 0 to 12 months by level of attendance in the intervention arm (error bars: 95% CI).

Hypotheses 6 and 7 mediation and moderation of intervention effects on lower limb physical function

Hypothesis 6

Hypothesis 6 was that the intervention effect on SPPB (intervention vs. control) will be mediated by changes in muscle-strengthening exercise and by changes in MVPA, lower-intensity physical activity or walking activity.

Lower-intensity physical activity was assessed using our accelerometer measures of sedentary time (also called ‘very low physical activity’).

Hypothesis 6 was largely supported by the data, with the proviso that accelerometer-based measures did not follow the same pattern as the self-reported measures. The intervention effect on SPPB score was partly or completely mediated by changes in self-reported muscle-strengthening exercise and self-reported changes in physical activity (both total activity and walking) between 0 and 12 months and between 0 and 24 months. No accelerometer measures of physical activity significantly mediated the intervention effect on SPPB score at any time point. The standardised mediation effects for muscle-strengthening exercise were generally higher than those for physical activity, and walking seemed to account for the majority of the mediating effect identified by the PASE questionnaire (Table 11).

TABLE 11 - Summary of mediation of intervention effects on SPPB by physical activity variables

VLPA, very low physical activity.

Note

Orange shading represents no significant mediation. Light-purple shading represents partial mediation. Aqua shading represents total mediation.

Mediating variable	Standardised coefficient (95% CI); n
	6 months	12 months	24 months
Muscle-strengthening (MSEQ)	–	0.051 (0.018 to 0.092); 579	0.053 (0.020 to 0.094); 569
Total activity (PASE)	–	0.025 (0.004 to 0.054); 616	0.039 (0.008 to 0.075); 598
Walking (PASE)	–	0.017 (0.000 to 0.041) 598	0.029 (0.004 to 0.059); 590
MVPA (accelerometer)	–	–	–
VLPA (accelerometer)	–	–	–

Hypothesis 7

Hypothesis 7 was that the intervention effect on SPPB score, physical activity and exercise from 0 to 6 months and from 0 to 24 months may be moderated by a number of variables, including age, sex, ethnicity, baseline physical activity and SPPB score, co-interventions, comorbidities, BMI, mental health, SES and education level.

In the main trial analysis (see Chapter 3 and Appendix 2, Figure 34), intervention effects were found for PASE total score and muscle-strengthening exercise (MSEQ adherence score), as well as the primary outcome (SPPB score) at both 6 and 24 months. Report Supplementary Material 2 shows the moderation effects of all of the variables specified in the hypothesis of these three variables. The text below gives an overview of these data, highlighting any significant associations.

Intervention effects on SPPB, physical activity and engagement in muscle-strengthening exercise (MSEQ) were not significantly moderated by age, sex, baseline SPPB score, co-morbidity (number of chronic physical conditions at baseline) or BMI. However, the following significant interactions were observed (Table 12).

TABLE 12 - Summary of moderation of intervention effects from 0 to 12 months and from 0 to 24 months on MSEQ, physical activity (PASE total) and SPPB by demographic variables

SE, standard error.

Figures are standardised coefficients for binary variables and R² change for adding the trial arm × interaction term for variables with more than two categories. Blank cells represent no significant moderation.

Moderating Variable	SPPB		Muscle-strengthening (MSEQ)		Total activity (PASE)
	6 months	24 months	6 months	24 months	6 months	24 months
Age
Sex
Ethnicity	–1.426, SE 0.714; p = 0.048				–69.519, SE 24.500; p = 0.005
SES (deprivation quintile)	R2 change 0.022, F = 3.818 (df = 4, 649); p = 0.0045		R2 change 0.017, F = 2.729 (df = 4, 587); p = 0.029			R2 change 0.017, F = 2.715 (df = 4, 604); p = 0.029
Baseline physical activity					0.188, SE 0.079, p = 0.018
Baseline SPPB
Co-interventions						26.741, SE 11.664, p = 0.022
Comorbidities
BMI
Mental health			–2.335, SE 0.972; p = 0.017	–2.450, SE 1.057; p = 0.021
Education level				R2 change 0.016, F = 2.447 (df = 4, 572); p = 0.045

At 6 months, the estimated effects of the intervention on SPPB score and the PASE total score were higher for ethnic minority participants than for white participants [1.43 points higher for SPPB score (beta 1.43; p = 0.048); 69.52 points higher for PASE total score (beta 69.52; p = 0.005)]. The interactions are visualised in Report Supplementary Material 2. These results need to be treated with considerable caution owing to (1) multiple testing (the Bonferroni adjusted p-value for multiple testing would be 0.0045) and (2) unbalanced group sizes (31 ethnic minority participants vs. 628 white participants for SPPB score, and 25 ethnic minority participants vs. 625 white participants for PASE total score).

Socioeconomic status (deprivation quintile) significantly moderated the effects of the intervention on SPPB score (p = 0.0045) and MSEQ score at 6 months (p = 0.029), and on PASE total score at 24 months (p = 0.029) (see Table 12). The interactions are visualised in Report Supplementary Material 2. For SPPB score at 6 months, the intervention worked best in the highest deprivation group (estimated mean difference 1.78 points, 95% CI 0.88 to 2.67 points, n = 70) and worst in the lowest deprivation group (estimated mean difference –0.13 points, 95% CI –0.67 to 0.41, n = 191), with a general pattern of increasing effect size as area deprivation decreased. However, there was no clear pattern of effect for MSEQ score or physical activity. These results need to be treated with considerable caution owing to (1) multiple testing (the Bonferroni adjusted p-value for multiple testing would be 0.0045) and (2) unbalanced group sizes (ranging from 63 to 171).

Physical activity level at baseline significantly moderated the effects of the intervention on physical activity at 6 months (beta 0.188; p = 0.018), with intervention effects being greater in those who were more physically active at baseline. The interaction is visualised in Report Supplementary Material 2. This result needs to be treated with caution owing to multiple testing (the adjusted p-value would be 0.0045).

The intervention effect from 0 to 24 months on self-reported physical activity (PASE total score) was significantly moderated by engagement in co-interventions during the 24 months of the study (beta 26.741; p = 0.022). The intervention effect on physical activity for people who engaged in additional physical activity interventions during the course of the study was 26.741 PASE points higher than for the subgroup who did not engage in such interventions. This result needs to be treated with caution owing to multiple testing (the adjusted p-value would be 0.0045).

Mental health status at baseline significantly moderated the effects of the intervention on MSEQ score at both 6 months (beta –2.335; p = 0.017) and 24 months (beta –2.450; p = 0.021), with intervention effects being greater in those who had experienced no diagnosed mental health condition in the last 6 months. The interactions are visualised in Report Supplementary Material 2, which shows that the effects in both cases are largely driven by large increases over time in MSEQ adherence for people with mental health conditions in the control arm. This may reflect a regression to the mean effect in participants who had a transient mental health condition at baseline (they may have been less engaged in exercise at baseline), or it may reflect some kind of responding bias (i.e. a tendency for people with a mental health condition to under-report activity at baseline). This result needs to be treated with considerable caution owing to (1) multiple testing (the adjusted p-value would be 0.0045) and (2) unbalanced group sizes (29 participants with a recent mental health condition at baseline vs. 567 participants with none).

The intervention effect from 0 to 24 months on MSEQ adherence score was also significantly moderated by education level. The explanatory value of adding the arm × education interaction term to the regression model (R² change) was 0.016 (p = 0.045). The interaction was largely driven by an unexpected increase in MSEQ score from 0 to 24 months in the least educated control arm participants. This result needs to be treated with caution owing to multiple testing (adjusting for this would require a p-value of 0.0045 to denote significance).

Within-trial economic evaluation

Results

Intervention costs

The cost of the REACT intervention was estimated to be £9465.69 per arm, or £621.83 per participant (see Appendix 3, Table 35). The cost was somewhat sensitive to underlying assumptions, with the cost per participant ranging from £554.54 (–11%) to £775.13 (+25%) across the sensitivity analyses (see Appendix 3, Table 36). The methods for estimating these costs are described in detail in Appendix 3.

Resource use and utility values

A total of 367 and 410 participants were randomised to the control arm and the intervention arm, respectively. Resource use at baseline was generally well balanced, but there was some evidence of higher baseline resource use in the intervention arm (see Appendix 3, Table 37).

The average total value of NHS and PSS resource utilisation in the 6 months prior to randomisation was £653 and £797 in the control arm and the intervention arm, respectively. After multiple imputation and adjusting for covariates, the difference in pre-randomisation resource utilisation was £107 (95% CI –£50 to £263). Over the course of the REACT study, we estimate that NHS and PSS resource use was £724.74 lower in the intervention arm (95% CI £25.79 to £1423.69) (see Appendix 3, Table 38). There was generally a trend towards greater resource use over time, and the difference in costs was greatest in the first 6 months. Expanding to a societal perspective moved the expected cost difference towards zero (£645.65), meaning that the cost difference was no longer statistically significant (95% CI –£63.73 to £1355.02).

Table 14 explores which cost categories lead to savings between the intervention arm and the control arm. In available case analyses and MI analyses, the category ‘other hospital’ (which includes day-case procedures, outpatients and A&E attendances) was the primary category in which intervention arm participants had lower costs. The difference between MI and CCAs appears to be driven by the difference in hospital overnight stay costs: the unadjusted difference in mean costs is £72.80 among available cases compared with –£78.62 with MI, and this difference may be exacerbated by adjustment.

TABLE 14 - Breakdown of costs (£) across different categories (excluding intervention costs)

Societal perspective.

Cannot be estimated owing to convergence issues.

Note

Costs are from NHS/PSS perspective unless otherwise specified. Model for estimating adjusted difference includes only baseline resource use as a covariate owing to convergence issues.

Cost category	Mean value (SD); n		Difference
	Intervention arm	Control arm	Unadjusted	Adjusted (95% CI)
Available case analysis
Primary care	820.49 (680.47); 299	838.12 (704.30); 260	–17.63	–14.40 (–129.17 to 100.37)
Hospital overnight stay	1104.98 (3670.8); 289	1032.18 (3282.4); 246	72.80	254.25 (–358.29 to 866.80)
Other hospital	1372.61 (1799.1); 299	1856.01 (3281.2); 263	–483.40	–484.43 (–931.70 to –37.16)
Day care	1.78 (13.53); 276	7.71 (95.31); 232	–5.93	–8.74 (–31.88 to 14.40)
Convalescent	27.18 (199.77); 299	24.66 (312.97); 263	2.52	26.74 (–7.47 to 60.95)
Other	26.16 (332.18); 410	10.66 (187.38); 366	15.49	b
Day carea	2.65 (20.20); 276	11.51 (142.26); 232	–8.85	–13.04 (–47.58 to 21.49)
Convalescenta	40.56 (298.17); 299	36.80 (467.12); 263	3.76	39.91 (–11.14 to 90.96)
Informal carea	169.93 (590.27); 253	128.99 (364.44); 217	40.94	46.69 (–40.82 to 134.19)
Othera	46.56 (501.27); 410	22.86 (285.68); 366	23.70	9.96 (–39.66 to 59.58)
MI analysis
Primary care	848.02 (685.56); 410	853.80 (717.95); 367	–5.79	–25.36 (–127.81 to 77.08)
Hospital overnight stay	1064.98 (3440.2); 410	1143.60 (3492.6); 367	–78.62	–107.76 (–628.11 to 412.60)
Other hospital	1441.98 (2039.3); 410	1836.89 (3255.0); 367	–394.91	–462.03 (–861.98 to –62.07)
Day care	3.72 (37.73); 410	6.64 (82.25); 367	–2.92	–4.79 (–21.69 to 12.12)
Convalescent	69.34 (674.84); 410	35.57 (354.83); 367	33.77	16.02 (–87.99 to 120.04)
Other	26.16 (332.18); 410	10.64 (187.13); 367	15.52	b
Day carea	5.56 (56.32); 410	9.91 (122.76); 367	–4.35	–7.15 (–32.38 to 18.09)
Convalescenta	103.50 (1007.2); 410	53.09 (529.60); 367	50.40	23.92 (–131.33 to 179.17)
Informal carea	190.51 (589.62); 410	163.07 (451.56); 367	27.44	23.79 (–54.63 to 102.21)
Othera	46.56 (501.27); 410	22.80 (285.29); 367	23.76	9.30 (–37.78 to 56.37)

Participants randomised to receive the intervention had marginally better EQ-5D utility values than those randomised to the control arm at baseline (0.689 vs. 0.677, respectively) and this difference diminished after MI and adjustment (0.007, 95% CI –0.016 to 0.029). However, for SF-6D utility values, there was no difference between the arms at baseline (0.000, 95% CI –0.014 to 0.014) (Table 15).

TABLE 15 - Health-related quality of life measured in the REACT study

HRQoL, health-related quality of life; MCS, mental component score; PCS, physical component score.

HRQoL	Mean value (SD); n
	Intervention arm				Control arm
	Baseline	6 months	12 months	24 months	Baseline	6 months	12 months	24 months
EQ-5D value
Crosswalk to EQ-5D-3L	0.689 (0.158); 397	0.708 (0.167); 346	0.705 (0.170); 337	0.686 (0.200); 330	0.677 (0.165); 352	0.670 (0.181); 299	0.680 (0.183); 293	0.661 (0.195); 302
EQ-5D-VAS	70.6 (17.3); 399	72.6 (16.7); 349	71.4 (17.8); 338	70.4 (18.4); 329	72.1 (16.9); 362	72.0 (17.3); 298	70.6 (17.0); 294	69.4 (19.2); 301
EQ-5D-5L value set	0.789 (0.149); 397	0.805 (0.160); 346	0.801 (0.158); 337	0.782 (0.180); 330	0.781 (0.152); 352	0.770 (0.177); 299	0.785 (0.162); 293	0.767 (0.174); 302
SF-36
PCS	29.7 (11.0); 393	32.8 (11.5); 342	31.9 (11.5); 334	30.9 (12.0); 326	30.0 (10.6); 352	30.6 (10.9); 293	29.8 (10.9); 293	29.2 (10.8); 295
MCS	54.6 (8.3); 393	54.4 (8.6); 342	54.0 (8.8); 334	54.3 (8.6); 326	53.8 (8.7); 352	54.0 (9.1); 293	54.3 (9.4); 293	54.5 (9.1); 295
SF-6D	0.622 (0.095); 365	0.637 (0.103); 323	0.637 (0.098); 313	0.630 (0.105); 312	0.623 (0.089); 326	0.621 (0.103); 280	0.622 (0.100); 280	0.619 (0.096); 283

The resulting predicted mean QALYs (calculated from the EQ-5D and SF-6D) and incremental QALYs are shown in Table 16. The intervention is associated with increased utility values at all follow-up time points, and this difference is statistically significant at 6 months using EQ-5D and at 12 months using SF-6D. The within-trial QALYs estimated using EQ-5D were 1.354 for the intervention arm and 1.314 for the control arm, a statistically significant difference of 0.040 (95% CI 0.006 to 0.074).

TABLE 16 - Quality-adjusted life-years in the REACT study

Timepoint	EQ-5D utility (crosswalk)				SF-6D utility
	Intervention arm	Control arm	Difference	95% CI	Intervention arm	Control arm	Difference	95% CI
Baseline	0.685	0.678	0.007	–0.016 to 0.029	0.622	0.622	0.000	–0.014 to 0.014
6 months	0.703	0.668	0.035	0.012 to 0.058	0.634	0.621	0.012	–0.003 to 0.027
12 months	0.692	0.676	0.016	–0.007 to 0.039	0.635	0.617	0.018	0.004 to 0.031
24 months	0.676	0.657	0.019	–0.007 to 0.046	0.627	0.616	0.011	–0.002 to 0.024
Within-trial QALYs	1.354	1.314	0.040	0.006 to 0.074	1.241	1.216	0.025	0.006 to 0.044

Sensitivity analyses

Varying the unit costs for NHS resources each by ± 20% did not result in a large impact on total costs (Figure 14). Four NHS resources had the potential to affect incremental total costs by more than £20:

1. Increasing the cost of a day-case procedure (base-case value £752) by 20% increased the costs for the intervention relative to the control arm by £39.

2. Increasing the cost of a hospital inpatient bed-day (base-case value £504) by 20% decreased the costs for the intervention relative to the control arm by £32.

3. Increasing the cost of an intensive care unit stay (base-case value £3532) by 20% increased the costs for the intervention relative to the control arm by £25.

4. Increasing the cost of an outpatient appointment (base-case value £127) by 20% increased the costs for the intervention relative to the control arm by £24.

FIGURE 14.

Tornado diagram for one-way sensitivity analyses of resource values (NHS/PSS perspective).

For costs affecting the societal perspective only, one-way sensitivity analyses showed very limited sensitivity to unit costs. Only the cost of a friend/relative taking a day off work and the cost of a chiropractor affected relative costs by more than £5 when varied by ± 20%.

The sensitivity analysis for the plausible extreme values of the societal value of an hour of help around the house revealed some sensitivity in costs from a societal perspective. When the value was reduced to £0.35 per hour, the total incremental costs for the intervention compared with usual care became more negative (to –£541, from a base case of –£458), while when the value was increased to £335 per hour the difference in costs narrowed (to –£347).

Cost-effectiveness

Cost-effectiveness estimates include intervention costs (£621.83 per participant), as well as the impact of REACT on within-trial resource use, SPPB score and QALYs, as described above. Considering cost per 1-point increase in SPPB score at 24 months, the intervention is dominant in the base case (saves £103 and results in an increase in mean SPPB of 0.49 points).

When using QALYs as the measure of health benefit, as shown in Table 17, the intervention is cost-effective, either being dominant or having ICER below the £20,000 per QALY cost-effectiveness threshold. In the base case, the 95% upper confidence limit is below £20,000 per QALY, suggesting reasonable confidence that the intervention is cost-effective within the trial time horizon. However, uncertain results are found when there is no adjustment for covariates in the MI analysis or when only participants with complete data are included in the analysis adjusting for baseline covariates, because the upper 95% confidence limit for the ICER extends beyond the £20,000 per QALY cost-effectiveness threshold in these analyses.

TABLE 17 - Within-trial cost-effectiveness results

Note

ICERs reported to nearest £50.

Trial arm	Costs (£)	QALYs	Including costs (£)		Including QALYs		ICER
			Estimate	95% CI	Estimate	95% CI	Estimate	95% CI
Base case: MI adjusted
Control	4046	1.314
Intervention	3943	1.354	–103	–695 to 489	0.040	0.009 to 0.071	Dominant	Dominant	16,950
Scenario analysis
MI unadjusted
Control	3887	1.319
Intervention	4076	1.364	189	–401 to 778	0.045	–0.001 to 0.091	4200	Dominant	Dominated
CCA adjusted
Control	3573	1.323
Intervention	3871	1.372	297	–421 to 1016	0.049	0.010 to 0.089	6000	Dominant	71,250
CCA unadjusted
Control	3609	1.325
Intervention	3856	1.389	247	–498 to 992	0.063	0.005 to 0.122	3900	Dominant	18,000

The uncertainty in cost-effectiveness is demonstrated in cost-effectiveness scatter plots (Figure 15) and cost-effectiveness acceptability curves (Figure 16). The proportion of estimates cost-effective at £20,000 per QALY is 98% in the base case, 84% in the MI unadjusted analysis, 89% in the CCA adjusted analysis, and 98% in the CCA unadjusted analysis.

FIGURE 15.

Cost-effectiveness scatter plot for within-trial analysis. a, CCA adjusted; b, CCA unadjusted; c, MI adjusted; d, MI unadjusted.

FIGURE 16.

Cost-effectiveness acceptability curves for within-trial analysis.

Scenario analyses

In Table 18, a number of scenarios were explored. Removing the resource use impact observed in the REACT study makes the intervention less cost-effective but is not enough by itself to make the intervention not cost-effective at a £20,000 per QALY threshold. If the resource use impact is removed and SF-6D utilities are used to estimate QALYs (instead of EQ-5D), then the ICER for the intervention increases to £25,050 per QALY, which would not be considered cost-effective at a £20,000 per QALY threshold but would be considered cost-effective at a £30,000 per QALY threshold.

TABLE 18 - Cost-effectiveness scenario analyses (within-trial economic evaluation)

ICERs reported to nearest £50.

Scenario	Cost (£)			QALYs			ICER
	Control arm	Intervention arm	Difference	Control arm	Intervention arm	Difference
Base case	4046	3943	–103	1.314	1.354	0.040	Dominant
No resource use impact	0	622	622	Unchanged from base case			15,650
Societal perspective	4240	4216	–24	Unchanged from base case			Dominant
SF-6D QALYs	Unchanged from base case			1.216	1.241	0.025	4150
No resource use impact and SF-6D QALYs	0	622	622	1.216	1.241	0.025	25,050
Group size 12	4046	4097	50	Unchanged from base case			1250

Subgroup analyses

Subgroup analyses were conducted by refitting models for within-trial costs and QALYs, including interaction terms for the subgroup of interest with the group allocation variable. The REACT study appears to be cost-effective across all subgroups, except when the trial population is divided by baseline SPPB score. Participants with a baseline SPPB score of 4–7 do appear to obtain some health benefit in terms of QALYs, but there is no beneficial effect of the intervention on NHS/PSS resource use (see Appendix 3, Table 39).

Model-based economic evaluation

A pragmatic review identified two economic evaluations of physical activity interventions for older adults at risk of frailty (see Appendix 3, Figure 39). In both cases, the authors concluded that the intervention was cost-effective; however, both studies have certain limitations. The study by Alhambra-Borrás et al. 124 is at high risk of treatment selection bias, and the authors may have introduced further bias into their modelling by estimating costs and utility values separately for the intervention and control arms. Their modelling approach was based on dividing participants into two states based on the risk of falls/frailty, but this was poorly reported and was, therefore, of little value for informing the development of an economic model for REACT.

In the cost-effectiveness analysis study by Groessl et al. ,125 uncertainty is introduced on the impact of health-care resource utilisation, which in turn makes the cost-effectiveness of the LIFE study uncertain; the authors also did not attempt to estimate costs and QALYs beyond mean follow-up from the trial.

Although there is some evidence that group-based physical activity interventions can be cost-effective within their own settings, these findings are unlikely to be applicable to the REACT intervention owing to differences in the intervention designs (e.g. lasting 12 months compared with 36 months in the LIFE study).

Methods

A new decision-analytic model was developed in R (version 4.0.0) using the heemod package126 (version 0.12.0.9000).

Model structure and methodology

The decision analytic model focuses on projecting mobility over time and the consequences of mobility on utility values and costs for the NHS and PSS. The cohort of REACT participants enter the model at the end of the trial follow-up, distributed across a set of mutually exclusive mobility states defined by SPPB score. The following year, the patients face the probability of remaining in their end of trial mobility state, of moving to a worse or better health state, or of dying. This annual cycle process is repeated separately for intervention and control arms of the trial, until all cohort members are dead. At the end of the lifetime simulation, the costs and QALYs are aggregated and compared between trial arms to assess the differences in expected costs and QALYs and cost per QALY gained by receiving the intervention over usual care.

In this Markov cohort simulation model, the transition probabilities are allowed to vary with time but the model retains the Markov memoryless property,127 with 13 health states reflecting different levels of mobility (SPPB score of 0 to SPPB score of 12) and a death state. SPPB score has been shown to be a prognostic for loss of ability to walk 400 metres32 and development of mobility disability and disability in ADL. 128 We have further found it to be correlated with utility values and NHS/PSS resource use within the REACT study.

The utility values and NHS/PSS costs are related to the SPPB score based on statistical models fitted to the control arm of the REACT study. Transition probabilities between the mobility states are related to the current SPPB score, age and sex, estimated using statistical models fitted to the control arm of the REACT study. It conservatively assumed that mobility does not affect the mortality rate, which means that the model predicts only differences in QALYs based on utility values. Studies have shown an association between SPPB and mortality;129 however, it was considered plausible that long-term health conditions could affect both mobility and mortality, and that mobility has not been established sufficiently as a surrogate for mortality risk. 130

The model has a cycle length of 1 year and runs for 35 cycles, at which point the cohort is aged 100–130 years, depending on the starting age, which ranges from 65 to 95 years in 5-year intervals. The model is run separately for men and women. In the first two annual cycles, the costs and QALYs are taken directly from the REACT study, after which the distribution of SPPB values in the cohort is established based on predicted values from the REACT study according to age and sex.

Model inputs

Population

In the base-case analysis, we assume a population of women aged 75 years given that women were the majority of participants (66%) in REACT and the age 75 years is close to the modal age in the study of 77 years for men and women. We also analyse cost-effectiveness for men and women aged 65–95 years in 5-year intervals.

The distribution of SPPB values (after 24 months) for each combination of age, sex and group (intervention or control) is estimated from the REACT study using ordered logistic regression, as shown in Figure 17. Specifically, an ordered logistic regression was fitted to trial-end SPPB score, with intervention allocation, age, sex, SPPB score at screening into the study and study site as predictors, and predicted probabilities of each trial-end SPPB score category were computed based on age, sex and intervention allocation using the margins command in Stata.

FIGURE 17.

Expected trial-end (24-month) SPPB score by age, sex and group (error boxes show ± 1 standard error). a, male; b, female.

To estimate the average population costs and benefits of REACT, we use two reference populations:

1. the REACT study participants

2. UK population tables. 131,132

In each case, we estimate weights based on the number of individuals of a given sex whose nearest age is 65, 70, 75, . . . and 95 years.

Intervention effectiveness

The effectiveness of the intervention is incorporated in two components:

• the directly observed effect of the intervention on costs (excluding the intervention cost) and QALYs within the REACT study (i.e. up to 24 months after randomisation)

• the SPPB score distribution realised at the end of the trial period.

The first of these components is taken from the analysis in which MI is used to account for missing data and bootstrapping is used to characterise uncertainty. This analysis (see Results) produces cost estimates of £4043 [standard error (SE) £254] in the control arm and £3319 (SE £173) in the intervention arm (excluding the cost of delivering the intervention), and produces QALY estimates of 1.348 (SE 0.012) and 1.388 (SE 0.014) in the control arm and the intervention arm, respectively.

The second of these has been described above (see Figure 17).

Extrapolation of the intervention effectiveness on health outcomes and costs is modelled as a function of the primary outcome in the trial, SPPB, under variable scenarios about the rate at which the effect of REACT on this outcome wears off.

Natural history

There are two natural history components in the model:

1. the evolution of mobility with age

2. the changing mortality rate with age.

These components do not interact, that is there is no differential mortality rate according to mobility level. Both components apply equally to the intervention and control arms. This means, for example, that there is no assumption that participants in the intervention arm are better able to maintain their mobility after the 24-month trial period. The natural history model is fully described in Appendix 3.

Costs

NHS and PSS (annual) costs were estimated as a function of mobility (SPPB score, categorised as 0–3, 4–7, 8–9 or 10–12) from the control arm of the REACT study. For each time point, we estimated the current cost accrual rate based on the cost of recalled health-care resource usage in the 6 months leading up to the time point, which we denote x_m for the m-month follow-up time point. At baseline, we estimated costs to be accrued at a rate of:

⁽¹⁾ r 0 ≈ x 0 + x 6 .

We further estimate:

⁽²⁾ r 6 ≈ x 6 + x 12 r 12 ≈ x 6 + 2 x 12 + x 24 2 r 24 ≈ 2 x 24 .

A linear regression with dummy variables for the SPPB categories was fitted, adjusting for clustering within participants, as shown in Figure 18.

FIGURE 18.

Cost by SPPB score range.

The model shows that costs generally increase with decreasing mobility, but there is considerable uncertainty about costs for those with very poor mobility (SPPB score of < 4); this is partly because the REACT study selected participants with a SPPB score of ≥ 4, so there are fewer data available (n = 48 observations with SPPB score of < 4 out of 1269 SPPB observations in the control arm).

Utility values

Utility values were estimated from SPPB using a linear regression with a restricted cubic regression spline for SPPB. 133 Observations from control arm participants were included, and the analysis was adjusted for clustering within participants (Figure 19).

FIGURE 19.

Predicted EQ-5D utility according to SPPB score.

Given that SPPB score declines with age (see Appendix 3), this means that utility values also decline with age. In the REACT study, a Pearson’s chi-squared test showed significant associations of baseline SPPB score with baseline EQ-5D self-care (p < 0.001), usual activities (p < 0.001) and pain/discomfort (p < 0.001) scores. In addition, age was not negatively associated with EQ-5D utility before or after adjusting for baseline SPPB score.

Analysis of uncertainty

Deterministic one-way sensitivity analyses of model parameters and probabilistic sensitivity analysis (PSA) were considered. Correlations between model parameters in the PSA are incorporated through sampling from suitable multivariate normal distributions or sampling from bootstrap estimates. We assume that the cost of the intervention follows a gamma distribution with coefficient of variation of 20%.

We also consider a number of scenario analyses:

• all model inputs re-estimated using CCA rather than MI

• societal costs instead of NHS/PSS costs

• SF-6D utilities instead of EQ-5D

• four alternative models for the evolution of SPPB (as described in Appendix 3).

A summary of all model parameters is provided in Appendix 3, Tables 45–48.

Results

Base case

The model predicts a decline in SPPB over time for a woman aged 75 years with no difference in mortality between the two arms, as shown in Figure 20.

FIGURE 20.

Mean SPPB score and proportion alive in the Markov model.

In the lifetime simulation, the cohort receiving the intervention spends more time in higher SPPB states (8–12) but less time in lower SPPB states (< 8), and there is no difference in total life years lived. Spending more time in higher SPPB states results in accruing more QALYs and incurring lower costs (see Appendix 3, Table 48).

In our base-case analysis, a woman aged 75 years receiving the intervention is expected to gain 0.072 discounted QALYs over her lifetime and to reduce lifetime discounted costs by £290 (see Appendix 3, Table 49). This compares with the trial-end incremental costs of –£103 and 0.040 QALYs. The intervention is both more effective and less costly than usual care and, therefore, cost-effective regardless of the threshold at which the NHS is willing to pay for a QALY.

Figure 21 shows the cumulative incremental discounted QALYs over time. More than half of the lifetime incremental discounted QALYs come in the first 2 years. After 10 years, there are very limited further incremental QALYs, which reflects the fact that SPPB score distributions have largely converged by this point and an increasing proportion of the population is dying and, hence, unable to accrue further QALYs.

FIGURE 21.

Cumulative incremental discounted QALYs over time.

Figure 22 shows the cumulative incremental discounted costs over time. In contrast to QALYs, the majority of the lifetime cost savings do not come in the first 2 years, although, similar to QALYs, there is very limited change after 10 years. The intervention is predicted to lead to QALY gains and cost savings and, therefore, be cost-effective at any time horizon beyond 2 years.

FIGURE 22.

Cumulative incremental discounted costs over time.

Heterogeneity analysis

As shown in Figure 23, it is expected that discounted lifetime costs and QALYs are higher for younger participants than for older participants, and are higher for women than for men. This is consistent with the fact that life expectancy is greater for women than for men.

FIGURE 23.

Heterogeneity analysis, absolute lifetime discounted costs and QALYs. a, Lifetime costs; b, lifetime QALYs.

Analysis of the incremental costs and QALYs reveals that the REACT programme is most beneficial to those aged 80–85 years and is more beneficial for women than for men (Figure 24). The REACT programme is expected to reduce lifetime costs for participants of any age and sex (Figure 25), which means that it is dominant across the population. The incremental net monetary benefit (INMB) at a willingness-to-pay threshold of £20,000 per QALY is shown in Figure 26, which demonstrates that the cost of the intervention could increase by £1350 (to around £1970) and still be marginally cost-effective for 65-year-old men and cost-effective for men of other ages and women across the age range.

FIGURE 24.

Heterogeneity analysis: incremental QALYs.

FIGURE 25.

Heterogeneity analysis: incremental costs.

FIGURE 26.

Heterogeneity analysis: INMB.

The base case produces similar incremental costs and QALYs to the REACT study participants, on average. When the reference population was the UK general population aged 65–100 years, REACT was still dominant but produced slightly smaller cost savings and incremental QALYs (see Appendix 3, Figure 41).

Sensitivity analyses

A PSA was conducted with 500 iterations. The 95% credible interval for INMB for the base case was £659 to £3550 at a willingness-to-pay threshold of £20,000 per QALY. The 95% credible interval for the ICER was –£14,626 (dominant) to £5537 per QALY. Further heterogeneity analysis is presented in Appendix 3, Figure 42.

One-way sensitivity analyses

One-way sensitivity analyses were conducted for all model parameters (Figure 27). The results show that the within-trial QALY and cost estimates strongly influence cost-effectiveness overall, as do the effects of age on utility values and SPPB evolution. There are no parameters that independently cause the intervention to have an ICER above £20,000.

FIGURE 27.

Tornado diagram for one-way sensitivity analyses.

Scenario analyses

The results of the scenario analyses with a population of women aged 75 years (base-case analysis) are shown in Table 19.

TABLE 19 - Scenario analyses for model-based cost-effectiveness analysis

Scenario	Cost (£)		QALY		INMB (£)	ICER (£/QALY)
	Control arm	Intervention arm	Control arm	Intervention arm
Base case	20,627	20,338	7.111	7.183	1735	Dominant
Inputs from CCAs	19,580	19,690	7.177	7.259	1528	1345
Societal perspective	21,168	20,891	7.111	7.183	1723	Dominant
SF-6D utilities	20,627	20,338	6.579	6.623	1165	Dominant
Mobility model 2 (simplified ordered logistic)	21,057	20,690	7.044	7.126	2008	Dominant
Mobility model 3 (static mobility distributions)	20,502	19,838	7.171	7.302	3281	Dominant
Mobility model 4 (alternative method for inclusion of REACT SPPB changes)	23,843	23,268	6.749	6.855	2697	Dominant
Mobility model 5 (artificial transition matrix with 38% annual convergence)	23,192	22,932	6.834	6.896	1501	Dominant

All model inputs were re-estimated using CCAs rather than MI analyses. This was the only scenario in which lifetime costs in the intervention arm exceeded those in the control arm (although costs for both arms were lower than in the base-case analysis). The intervention is predicted to be cost-effective in this scenario because the ICER is well below £20,000 per QALY.

Societal perspective on costs had a limited effect on cost-effectiveness. QALYs were unchanged and costs in both arms increased, although incremental costs were marginally smaller.

The SF-6D utilities had a moderate effect on QALYs, reducing lifetime QALYs in both arms and reducing the difference in QALYs between the arms. Costs were unchanged and REACT was still the dominant strategy.

The finding that the intervention produced cost savings and more QALYs than usual care remained in all scenarios analysis of structural uncertainty associated with the model extrapolation.

Key findings and interpretation

The primary REACT hypothesis was supported. Older adults with mobility limitations who received the 12-month, community-based group physical activity and behavioural maintenance programme had significantly better lower limb function at the 24-month follow-up than participants who attended a socioeducation-only control programme (adjusted mean difference 0.49 SPPB points, 95% CI 0.06 to 0.92 SPPB points). The difference in lower limb function between participants in the intervention and control arms was clinically meaningful at 6, 12 and 24 months. The intervention was also found to have significantly increased self-reported physical activity and performance of muscle-strengthening activity, and decreased perceived joint pain, after 24 months’ follow-up. However, there was no significant effect on accelerometer-measured physical activity or other secondary outcomes, and some differences found at 12 months were reduced at 24 months. This may reflect deterioration of the effects of exercise behaviours over time.

In qualitative interviews, participants reported that they enjoyed the programme and being part of a research study, and they reported better mental and social well-being, especially higher physical confidence, improved motivation and feeling more outgoing. Improved social connectedness and bonding with REACT groups were key outcomes for participants in the intervention arm, who also cited numerous physical health benefits, including improvements in mobility, strength, balance, walking, fitness, sleep and physical independence. Themes identified at 24 months largely mirrored those reported at 6 and 12 months. However, whereas the 6- and 12-month interviews found that social support was a key reason for engaging in REACT, at 24 months individual-level factors, such as perceived benefits, were more prominent themes in explaining physical activity maintenance. Key components of the REACT programme that seemed to positively influence maintenance of physical activity at 24 months were (1) techniques for managing slips/lapses, supporting habit change and resolving sources of tension around increasing physical activity; (2) the person-centred delivery style to build autonomy/intrinsic motivation; and (3) the group-based delivery promoting social connectedness.

The quantitative process evaluation confirmed that, compared with the control arm, the intervention arm reported experiencing greater benefits from exercising in terms of their physical, mental and social well-being (see Chapter 4, Results). In particular, the hypothesis that increased exposure to the intervention will be associated with positive changes in competence, relatedness, enjoyment and perceived benefits (hypothesis 5c) was largely supported in relation to muscle-strengthening exercise only. Increased exposure to the intervention was associated with positive changes in psychosocial determinants for muscle-strengthening exercise 12 months after baseline, but not with changes in determinants of MVPA. Moderation analyses (see Chapter 4, Results) and subgroup interactions (see Appendix 2, Table 33) showed that REACT appeared to work equally well with men and women, and the positive effects were consistently seen irrespective of age, SES (assessed by education, home ownership and quintiles of area deprivation), baseline physical function, comorbidities and study site. However, the findings in relation to ethnicity and people with mental health problems need to be treated with caution owing to small group sizes.

The intervention fidelity analysis confirmed good delivery of the exercise programme. However, it identified considerable scope for improvement in the delivery of the behaviour maintenance sessions, particularly in terms of support for monitoring progress/eliciting benefits of physical activity, action-planning/goal-setting, modelling of behaviour, supporting competence/self-efficacy, supporting relatedness and managing setbacks/problem-solving.

Taken together, the qualitative and quantitative process evaluations broadly supported the logic model for the REACT intervention. They identified several ways that the intervention and its implementation could be improved. This included possible changes to the logic model (from the qualitative and quantitative studies) and changes to delivery processes (from the intervention fidelity and qualitative studies). Recommendations for refinement/implementation of the REACT intervention are listed in the discussion sections of the relevant chapters.

Our health economic analyses indicated that REACT offers good value for money. The intervention plus usual care was cost-effective compared with usual care alone over the 2-year time period of the REACT intervention. In the base-case scenario, the intervention saved £103 in NHS/PSS costs per participant, with a QALY gain of 0.04 within the 2-year trial window. Lifetime horizon modelling estimated that further cost savings and QALY gains were accrued up to 15 years post randomisation. This is primarily because the intervention reduced frailty and its associated costs to the NHS and PSS, as well as improving health-related quality of life.

Stronger intervention effects were associated with a higher frequency of session attendance. An association between dose and response was observed, with an adjusted mean SPPB score difference of 0.64 points (95% CI 0.23 to 1.05 points; p = 0.002) for those attending ≥ 50% of intervention sessions and of 0.81 points (95% CI 0.38 to 1.23 points; p < 0.001) for those attending ≥ 75% of sessions. This strong dose–response relationship could simply reflect the effectiveness of the intervention or it could be explained in terms of participants being less likely to attend if their physical function deteriorated over the course of the study (e.g. owing to injury or other life events).

To situate our findings in the context of existing evidence, we searched Google Scholar (Google Inc., Mountain View, CA, USA), the Cochrane Database of Systematic Reviews, the NIHR Journals Library and PubMed, focusing on recent systematic reviews and meta-analyses from inception to 2014, using terms representing ‘older people’, ‘physical activity’, ‘physical function’ and ‘randomised controlled trials’.

Based on these searches, the REACT study findings are consistent with a robust existing body of evidence demonstrating the positive impact of physical activity on physical functioning,11–14 independent living, mobility-related disability, falls, hospital admissions and mortality. 16,134,135

For example, the largest efficacy trial worldwide in this population, the US LIFE efficacy trial, has already shown that an exercise and behavioural support intervention can reduce the risk of developing major mobility disability by 18%. However, prior to the publication of the current findings, there has been a lack of evidence regarding the effectiveness and cost-effectiveness of long-term (≥  1 year) exercise programmes administered at a community level in a real-world setting and tailored for a UK population of older adults at risk of mobility disability. Furthermore, our database searches found no long-term, community-based interventions reporting effects for longer than 12 months post intervention.

The REACT study, therefore, provides robust evidence that a relatively low-resource, 1-year exercise intervention can improve physical functioning in real-world community settings in the UK, with clinically meaningful benefits that are sustained over at least 24 months.

The baseline SPPB scores were almost identical with the baseline score of the LIFE study, which enables comparison. The observed difference in SPPB score of 0.49 points at 24 months was three times larger than the between-group difference reported in the LIFE trial. 21 The REACT intervention cost an estimated £622 per participant, which also compares favourably with the cost of the LIFE intervention (£1770 per participant).

Strengths, limitations and generalisability

To the best of our knowledge, the REACT trial is the first trial in the UK to target physical function with a long-term exercise-based intervention in a population of older adults who have restricted mobility. It was a robustly designed and conducted study with low attrition rates (19% at 24 months) and a comprehensive health economic evaluation. The comprehensive process evaluation increases our understanding of the mechanisms of engagement and behaviour change underpinning the REACT intervention. This will inform further refinement and implementation of the intervention and its training, as well as further development of the logic model and behaviour change theory. As well as informing any future implementation of the REACT intervention, this may help to design other interventions promoting active ageing.

Attendance compared favourably with that reported in other exercise programmes. Of the 83.9% who started the programme, 67.7% (95% CI 65.1% to 70.4%) of participants attended ≥ 50% of the sessions. However, it is notable that 16.1% of participants allocated to the intervention arm did not attend any intervention sessions, so there is scope to improve initial engagement in the programme. The intervention fidelity analysis also identified scope for improvement in several aspects of delivery of the health behaviour maintenance sessions. Taken together, these intervention delivery deficits may have led to underestimation of intervention effects.

There are some limitations that need to be acknowledged. As with other studies of behavioural interventions, blinding of participants to the study arm was not possible, which introduces the possibility of social desirability bias in patient-reported measures. 136 However, the primary outcome was assessed by independent observers who were blinded to study arm allocation, so this is unlikely to affect the primary outcome.

The generalisability of the findings to a wider UK population is also a key issue. The REACT study aimed to test a programme that was acceptable to as wide a range of the older adult population as possible and to recruit a diverse population in terms of sex, area deprivation and ethnic background. Our attempt to recruit a representative cohort for REACT in terms of deprivation and ethnicity was reasonably successful. Across the IMD quintiles, we recruited 3.2% less than the general population of those most deprived (quintile 1), but 2.6% more than those in quintile 2 (see Table 2). Overall, we avoided a substantial skewing towards affluence, which has proved challenging with this age group in previous research. 23 The sample recruited was also representative of the older adult UK population in terms of deprivation and ethnicity, except for an under-representation of South Asian older adults. 29

Although only 3% of those invited to take part were recruited, the response rate among the eligible population was 17% (see Chapter 3). Given this, and the small numbers of ethnic minority participants recruited, the generalisability of the findings (particularly) to ethnic minority populations should be explored further in future studies or as part of a REACT implementation evaluation study.

The mixed findings on the secondary outcomes indicate that the effects of the intervention were limited to lower limb physical function and did not extend to substantial increases in physical activity or other domains of physical function (e.g. grip strength). The secondary outcome analyses were exploratory, with no adjustment for multiple testing, and should be interpreted accordingly.

Furthermore, the trial analyses did not show an impact on quality of life. However, the more sophisticated, time-integrated approach used in the health economic analysis revealed a significant difference in EQ-5D (as well as a saving in health-care costs). Indeed, the health economic analyses indicated that the increases in physical function observed were associated with substantial quality-of-life and health economic benefits, both within the 24-month trial window and across a lifetime horizon.

Although the overall results for REACT were positive, the process evaluation indicated, as with most service-based interventions, that there was considerable scope for improvement by session leaders in the facilitation of important self-regulation processes and social/relatedness-building processes during the delivery of the behavioural maintenance programme. To some extent, this may have been mitigated by mutual support among participants and self-delivery of some of the intended processes within the groups during the exercise sessions. However, future implementations of the REACT intervention should aim to improve the training and delivery of the programme accordingly.

Considerations for further research

A key consideration for scaling up REACT and for future research in this population is how to identify people who would benefit from interventions (those with SPPB scores of 4–9). Although we met our recruitment target, from invitation to randomisation we had a success rate of only 3%. This recruitment yield is comparable to that of other behavioural interventions targeting older populations. 29 However, finding ways to make recruitment more efficient and more inclusive would be of benefit to future research in this field.

One means by which this might be achieved is via the UK’s electronic Frailty Index (which is recorded in GP databases) or similar schemes being tested in Europe. 33 However, evidence of the electronic Frailty Index’s ability to identify people with mobility limitations is required to reveal whether or not its use would be beneficial in recruitment for studies targeting improved mobility. Another option might be to develop an easy-to-use, parsimonious self-report measure that has strong predictive value for lower limb physical function. One candidate measure is the MAT-SF measure (a video-scenario-based self-assessment tool designed to assess lower limb physical function) that we used as a secondary outcome of REACT. 53 Further analyses of the REACT (and other) data will reveal whether or not this might prove to be a useful screening tool for use in practice and in future studies. A third option would involve relationship-building with multiple charities, local GPs and social prescribers, and community groups to help identify suitable candidates for intervention.

More research is needed in ethnic minority communities, particularly South Asian communities, to identify strategies for greater inclusivity and any necessary programme modifications to optimise the effectiveness of the REACT programme for ethnic minority populations. Further evaluation of the effectiveness of the programme for people from diverse ethnic backgrounds is also warranted.

The exclusion criteria for REACT meant that some groups of older adults or those with particular conditions were not able to participate, despite the fact that their needs suggest that they may have benefited. Examples include people with diagnosed dementia or people living in a residential care home. Future consideration should be given to strategies and conditions for their inclusion and for assessing the effectiveness of the REACT intervention in these populations.

Further research is also needed to optimise the implementation of REACT at scale and further evaluate and extend its reach, effectiveness and cost-effectiveness.

Using comprehensive participant and public involvement techniques to engage with session leaders, older adults eligible for the REACT programme, together with the research team, will help to fine-tune the behavioural maintenance programme (in terms of both content and training) and its delivery to ensure better fidelity of delivery when REACT is implemented at scale. The examples of good practice identified in the intervention fidelity analysis and the feedback from the qualitative process evaluation will be helpful in refining the REACT training materials.

It may be possible and synergistic to integrate the REACT intervention with existing mobility-related prevention and rehabilitation services, and provide support for participants who do not meet the criteria for specific programmes targeting falls. Given their cost-effectiveness, research into the cost-effectiveness of ongoing/continuous exercise programmes may also be fruitful; this may further improve the maintenance profile of the intervention, thereby increasing the available cost savings. There is also scope for longer-term health outcomes of the REACT programme to be evaluated. For example, maintained or increased mobility in older adults may influence longer-term mental well-being and delay onset of or deterioration through other diseases and conditions, such as dementia, diabetes and heart disease. Finally, the estimated effects of the programme on long-term health service costs, such as medications, primary care visits and emergency secondary care usage, should be directly measured to verify the conclusions of the health economic modelling, particularly to identify more definitive estimates of the cost consequences of delivering the REACT intervention at scale.

The REACT study provides robust evidence that the trajectory of declining physical functioning with age is modifiable and may even be reversible for many adults. It is particularly timely, allowing for some optimism post the COVID-19 pandemic, during which deconditioning was one of the key health challenges that older people experienced.

Future research will need to examine the feasibility of new ways to increase access to the programme, including virtual attendance, which might be useful in future lockdown/pandemic situations for people who are housebound, who may live more rurally or who prefer to engage in home-based interventions in later life. Similar to other programmes,137 REACT has the potential to be delivered online in its entirety, but we need to evaluate the reach/accessibility of this delivery style and its impact on physical and psychological well-being.

Implications for decision-makers and for practice

The REACT study has provided robust evidence that a group-based exercise programme with long duration (12 months) that encourages participants to build strong social networks and the habit of exercising in a safe and fun environment can improve their prognosis for maintaining mobility and save costs in both the short and the long term. This adds to the existing literature indicating that physical activity is crucial for older people’s physical functioning and health. It should strengthen the case for consideration by funders and policy-makers to support exercise programmes for the critical at-risk population of older people with lower limb frailty or pre-frailty (as defined by a SPPB score of 4–7).

The intervention delivered positive results on mobility across different settings in three different geographical areas of the UK, using a range of session leaders who were already active in the field of exercise programme delivery. For this reason, the REACT intervention is suitable for delivery by existing service providers, including city council providers, voluntary sector organisations, commercial organisations, the leisure industry and the private sector; therefore, the results should generalise well to other real-world settings. However, attention needs to be paid to maintaining (or improving) delivery fidelity when rolling out complex interventions. 43 Therefore, it may be helpful to consider what organisation-level governance or performance-monitoring systems might support high-quality delivery of the REACT intervention.

Given that the intervention appeared to work equally well irrespective of age, gender, SES (assessed by education, home ownership and quintiles of area deprivation) and baseline physical function, REACT may have the potential to reduce health inequalities if delivery is targeted towards underserved populations.

The REACT exercise intervention also provides important evidence supporting both the US and the UK physical activity recommendations for multimodal exercise programmes. 138,139 In particular, it demonstrates the beneficial effect of engaging in at least one exercise session per week. This is a strong public health message for older adults in terms of the level of commitment that they need to demonstrate to achieve maintenance of their lower limb physical function. Contrary to the widespread belief that ageing causes an inevitable decline in physical functioning, the REACT study has proved that this decline can be delayed or reversed with relatively modest lifestyle changes.

The health economic analysis indicates that REACT provides a feasible and cost-effective means of helping older people stay mobile, and this will enhance their chances of remaining independent for longer. Based on the moderation and subgroup analyses presented above, it also has potential to help reduce health inequalities among the elderly. Given this and the cost-saving nature of the intervention, REACT should be given serious consideration by health and public health commissioners.

Although recruitment of older participants to a research study that includes a RCT is much more of a commitment than recruiting to an exercise programme offered by leisure or community services, many lessons have been learned that may help to inform real-world implementation. These lessons are available through our recruitment publication. 29 Furthermore, given that the NIHR Public Health Research programme does not fund intervention costs, the researchers needed to work closely with multiple partners to deliver the REACT intervention programme, which was an excellent test of programme pragmatism and sustainability. The delivery of the REACT study provides an important template for this partnership model.

Two further implications that may be useful for future practice and/or decision-making are that (1) the REACT exercise intervention seems to work even with attendance as low as 50% and a lower-than-intended quality of delivery of the behavioural maintenance programme, and (2) there is scope for improvement of delivery quality through refinements to the behavioural maintenance programme and its training course. This may lead to increased effectiveness and better maintenance of effects in future implementations of the programme.

Finally, it is important to position the REACT programme within the COVID-19 pandemic context and the potential for REACT delivery in the post-COVID-19 era. The COVID-19 pandemic has resulted in the target population of the REACT intervention following stay-at-home guidance during lockdowns and, in some cases, shielding outside lockdowns. During these periods, it would not have been possible to deliver the REACT intervention as originally designed and evaluated. It is unclear whether or not a digital alternative to REACT, delivered in people’s homes or online, would be feasible, safe and effective.

As the UK exits from COVID-19 restrictions and a large proportion of the target population have been vaccinated, it is likely that the REACT intervention will once again be deliverable. There may be some concerns around travel to the REACT venues (if the target population are less comfortable taking public transport than previously), which could affect take-up. On the other hand, the demand for appropriate venues may have dropped owing to an increasing number of people working from home, which could lead to a reduction in the cost of delivering REACT (in the short to medium term).

Considering the possibility of future pandemics (e.g. further novel coronaviruses or influenza pandemics), the REACT intervention may be expected to allow more people to maintain their mobility and continue to live in private homes, rather than entering residential or nursing homes because of frailty. A significant burden of COVID-19 in the UK resulted from SARS-CoV-2 spreading through care homes and leading to the hospitalisation and/or death of residents; this may give an additional incentive for local governments and the NHS to fund interventions that stave off frailty.

Conclusion

The REACT study provides robust evidence that a relatively low-resource, 1-year exercise intervention delivered alongside a behaviour maintenance programme helped UK older adults to retain their lower limb physical function over a 24-month period. The REACT intervention is cost saving from an NHS/PSS perspective within a 2-year window, with further cost savings and QALY benefits estimated in the longer term. The results indicate that the well-established trajectory of declining physical functioning in older age is modifiable and in some cases reversible. Following further refinements guided by the findings of the process evaluation, the REACT intervention is suitable for large-scale implementation. The findings also suggest that there is potential for improvement in the design and delivery of the REACT intervention, and implementation should be evaluated to confirm the generalisability of these findings to the wider population, especially in ethnic minority populations.

Some text in this chapter has been reproduced from Stathi et al. ,1 Withall et al. 29 and Cross et al. 71 These are Open Access articles distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/. The text includes minor additions and formatting changes to the original text.

Non-author contributors

We wish to express our thanks to the entire REACT research team, the Clinical Research Networks at each REACT site and all of the general practices and community organisations that supported REACT recruitment. Delivery of the REACT programme was possible only because of the support of our partners Bath and North East Somerset Council; Exeter and Solihull City Councils; Westbank Charity, Exminster; St Monica Trust, Bristol; Bristol Ageing Better; St John’s Hospital, Bath; Age UK, Birmingham; Agewell, West Midlands; Sandwell and West Birmingham Hospitals NHS Trust; the Portway Lifestyle Centre, Oldbury; and Solihull Borough Council, Birmingham.

Our Trial Steering Committee, chaired by Professor Yoav Ben-Shlomo, Data Management and Ethics Committee, chaired by Professor Dawn Skelton, and independent statistician, Professor Peter Thomas, provided hugely valuable expert guidance and support.

Dr Andrew F Hayes of The Ohio State University Department of Psychology provided valuable advice on the conduct and interpretation of the moderation analyses.

Finally, we would like to thank the 777 REACT participants without whom this research would not have been possible.

Contribution of authors

Afroditi Stathi (https://orcid.org.uk/0000-0003-2162-777X) (Professor of Physical Activity and Community Health) obtained funding for the research, acted as chief investigator and principal investigator at the Bath site, led the design of the study and drafted the report manuscript.

Janet Withall (https://orcid.org.uk/0000-0003-0196-5309) (Trial Manager, Active Ageing) led study co-ordination, contributed to the design of the study, was involved in the collection and processing of data and drafted the report manuscript.

Colin J Greaves (https://orcid.org.uk/0000-0003-4425-2691) (Professor of Psychology Applied to Health, Health Psychology) obtained funding for the research, acted as principal investigator at the Devon site, contributed to the design of the study, led the process evaluation and drafted the report manuscript.

Janice L Thompson (https://orcid.org.uk/0000-0002-7312-3226) (Professor of Public Health Nutrition and Exercise) obtained funding for the research, acted as principal investigator at the Birmingham site, contributed to the design of the study and reviewed the report manuscript.

Gordon Taylor (https://orcid.org.uk/0000-0003-1715-0913) (Professor of Medical Statistics) obtained funding for the research, contributed to the design of the study, developed the statistical analysis plan and undertook the analyses, and reviewed the report manuscript.

Antonieta Medina-Lara (https://orcid.org.uk/0000-0001-7325-8246) (Associate Professor in Health Economics) obtained funding for the research, contributed to the design of the health economic study, estimated the cost of the intervention and reviewed the report manuscript.

Colin Green (https://orcid.org.uk/0000-0001-6140-1287) (Honorary Professor of Health Economics) obtained funding for the research, contributed to the design of the study and reviewed the report manuscript.

Tristan Snowsill (https://orcid.org.uk/0000-0001-7406-2819) (Senior Research Fellow, Health Economics) was involved in the collection and processing of data, estimated the cost of the intervention and reviewed the report manuscript.

Heidi Johansen-Berg (https://orcid.org.uk/0000-0002-4134-9730) (Professor of Cognitive Neuroscience) obtained funding for the research, contributed to the design of the study (MRI substudy) and reviewed the report manuscript.

James Bilzon (https://orcid.org.uk/0000-0002-6701-7603) (Professor of Human and Applied Physiology, Rehabilitation Medicine) obtained funding for the research, acted as principal investigator at the Bath site, contributed to the design of the study and reviewed the report manuscript.

Selena Gray (https://orcid.org.uk/0000-0001-6257-4024) (Professor of Public Health) obtained funding for the research, contributed to the design of the study and reviewed the report manuscript.

Rosina Cross (https://orcid.org.uk/0000-0002-2548-2398) (Postdoctoral Research Fellow, Primary Care) was involved in the collection and processing of data and contributed to the process evaluation report manuscript.

Max J Western (https://orcid.org.uk/0000-0003-1107-8498) (Lecturer, Behavioural Science) was involved in the collection and processing of data and reviewed the report manuscript.

Jolanthe L de Koning (https://orcid.org.uk/0000-0001-6811-5726) (Research Assistant, Active Ageing) was involved in the collection and processing of data and reviewed the report manuscript.

Peter Ladlow (https://orcid.org.uk/0000-0002-9891-9714) (Higher Scientific Officer, Strength and Reconditioning/Rehabilitation) contributed to the design of the study, led the design of the exercise intervention and reviewed the report manuscript.

Jessica C Bollen (https://orcid.org.uk/0000-0002-6437-5923) (Research Assistant, Active Ageing) was involved in the collection and processing of data and reviewed the report manuscript.

Sarah J Moorlock (https://orcid.org.uk/0000-0003-4473-7226) (Research Assistant, Active Ageing) was involved in the collection and processing of data and reviewed the report manuscript.

Jack M Guralnik (https://orcid.org.uk/0000-0001-5108-7263) (Professor in Epidemiology and Public Health, Epidemiology and Ageing) obtained funding for the research, contributed to the design of the study and reviewed the report manuscript.

W Jack Rejeski (https://orcid.org.uk/0000-0003-2281-4649) (Professor of Health and Exercise Science, Behaviour Change and Disability in Ageing) obtained funding for the research, contributed to the design of the study and reviewed the report manuscript.

Melvyn Hillsdon (https://orcid.org.uk/0000-0003-2818-3278) (Associate Professor in Physical Activity Epidemiology and Public Health, Physical Activity and Health) contributed to the processing and analysis of accelerometer data and reviewed the report manuscript.

Kenneth R Fox (https://orcid.org.uk/0000-0003-2216-4287) (Emeritus Professor of Exercise and Health Sciences, Physical Activity and Health) obtained funding for the research, contributed to the design of the study and reviewed the report manuscript.

Publications

Stathi A, Withall J, Greaves CJ, Thompson JL, Taylor G, Medina-Lara A, et al. A community-based physical activity intervention to prevent mobility-related disability for retired older people (REtirement in ACTion (REACT)): study protocol for a randomised controlled trial. Trials 2018;19:228.

Withall J, Greaves CJ, Thompson JL, de Koning JL, Bollen JC, Moorlock SJ, et al. The tribulations of trials: lessons learnt recruiting 777 older adults into REtirement in ACTion (REACT), a trial of a community, group-based active aging intervention targeting mobility disability. J Gerontol 2020;75:2387–95.

Cross R, Greaves CJ, Withall J, Rejeski WJ, Stathi A. Delivery fidelity of the REACT (REtirement in ACTion) physical activity and behaviour maintenance intervention for community dwelling older people with mobility limitations. BMC Public Health 2022;22:1112.

Demnitz N, Stathi A, Withall J, Stainer C, Seager P, de Koning J, et al. Hippocampal maintenance after a 12-month physical activity intervention in older adults: the REACT MRI study. Neuroimage Clin 2022;35:102762.

Snowsill T, Stathi A, Green C, Withall J, Greaves C, Thompson J, et al. Cost-effectiveness of a community-based physical activity and behaviour maintenance intervention for preventing decline in physical functioning in older people: an economic evaluation of the REACT (Retirement in Action) intervention. Lancet Public Health 2022;7:E327–34.

Stathi A, Greaves C, Thompson J, Withall J, Ladlow P, Taylor G, et al. Effect of a physical activity and behaviour maintenance programme on functional mobility decline in older adults: the REACT (REtirement in ACTion) randomised controlled trial. Lancet Public Health 2022;7:E316–26.

Data-sharing statement

All data requests should be submitted to the corresponding author for consideration. Access to anonymised data may be granted following review.

Disclaimers

This report presents independent research funded by the National Institute for Health and Care Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, the PHR programme or the Department of Health and Social Care. If there are verbatim quotations included in this publication the views and opinions expressed by the interviewees are those of the interviewees and do not necessarily reflect those of the authors, those of the NHS, the NIHR, the PHR programme or the Department of Health and Social Care.

Intervention fidelity