Development and Evaluation of an Adaptive Mobile Health Physical Activity Intervention Post-stroke: a Sequential Multiple Assignment Randomised Trial

Stroke is the second leading cause of death and disability globally and the absolute number of people who have had a new stroke, died, survived or remained disabled from stroke has almost doubled between 1990 and 2017 (Krishnamurthi, Ikeda, & Feigin, 2020). Meta-analytic evidence demonstrates that the 1- and 5-year risk of recurrent stroke is 11.1% and 26.4%, respectively (Mohan et al., 2011). Recurrent stroke can be prevented by improving identified, modifiable risk factors including clinical conditions (hypertension, hyperlipidaemia, atrial fibrillation, diabetes, and obesity) and lifestyle factors (smoking, physical inactivity, unhealthy diet, and excess alcohol consumption) (Bridgwood et al., 2018). Despite advances in acute stroke intervention, secondary prevention strategies are lacking and require urgent attention (McElwaine, McCormack, & Harbison, 2016). Physical activity is the second-largest predictor of stroke (O'Donnell et al., 2016). Recent secondary prevention guidelines state that people with stoke, when able, should aim to achieve population-based recommendations for physical activity (40-minute sessions, 3 to 4 times per week of moderate to vigorous-intensity aerobic activity) (Kleindorfer et al., 2021). People with stroke have additional barriers to physical activity, such as muscle weakness, sensory dysfunction, reduced balance, and fatigue (Billinger et al., 2014). Consequently, physical activity levels of community-dwelling people with stroke remain lower than their age-matched counterparts (English et al., 2016), and people with stroke typically sit for more than 10 hours per day (Saunders, Mead, Fitzsimons, Kelly, van Wijck, Verschuren, Backx, et al., 2021). Given the heterogeneous nature of stroke, physical activity interventions which are adaptive to individual performance are recommended (Billinger et al., 2014). A Cochrane review on the effectiveness of interventions to reduce sedentary behaviour (any waking behaviour characterized by an energy expenditure <1.5 METS while in a sitting, reclining or lying posture [Ainsworth 2011]). post-stroke found that all included studies were at high or unclear risk of bias (Saunders, Mead, Fitzsimons, Kelly, van Wijck, Verschuren, Backx, et al., 2021). The analysis found that interventions to reduce sedentary behaviour do not reduce cardiovascular other adverse events or amount of time spent sitting. The authors highlight that the current World Health Organization guidelines recommend reducing the amount of sedentary time in people with disabilities, in general (Saunders, Mead, Fitzsimons, Kelly, van Wijck, Verschuren, & English, 2021; WHO, 2020). They conclude that interventions for people with stroke should focus on other therapeutic targets, such as increasing physical activity, to reduce sitting time (Saunders, Mead, Fitzsimons, Kelly, van Wijck, Verschuren, & English, 2021). The effectiveness of exercise interventions on death, disability and adverse events among people with stroke compared to treatment as usual control was examined in a Cochrane review including fifty-eight trials (n= 2797) (Saunders et al., 2020). Disability outcome showed moderate improvement after cardiorespiratory training (SMD)= 0.52, 95% CI=0.19-0.84; p=0.002) and by a small amount after mixed training (SMD=0.26, 95%=CI 0.04-0.49; p=0.02). Most of the studies were small and of moderate quality, demonstrating high and unclear risk of bias throughout. The variability and quality of the included trials and lack of data prevents conclusions about other outcomes and limits generalisability of the observed results. A systematic review investigated the effects of interventions to promote participation in lifestyle physical activity in community-dwelling stroke survivors (Morris, MacGillivray, & Mcfarlane, 2014). Two intervention types were identified: individualised tailored counselling with or without supervised exercise (n=6 studies) and supervised exercise with advice (n=5 studies). Interventions to improve participation in physical activity should incorporate physical activity-specific tailored counselling based on sound behavioural theory to promote long-term participation in physical activity post-stroke. A systematic review examining the study characteristics and the promise of interventions targeting free-living PA and/or sedentary behaviour in people with stroke included nine RCTs (n=717)(Moore et al., 2018). Moore et al. (2018) identified nine BCT's most likely to increase physical activity in people with stroke: action planning, goal setting (behaviour), credible source, social support (unspecified), problem solving, biofeedback, feedback on behaviour, information about health consequences, and information about social and environmental consequences. A systematic review and qualitative synthesis of users' experiences of using mHealth applications to promote physical activity was conducted in October 2017 (Carter, Robinson, Forbes, & Hayes, 2018). Studies were limited to qualitative research and the experiences of adults who had used mHealth to promote PA. The search yielded 4,420 studies, of which 16 studies (n=361 participants) were included. Six themes related to the experience of using mHealth to promote physical activity were generated inductively: (a) increased physical activity as a result of mHealth, (b) motivation, (c) self-monitoring, (d) personalised/ tailored mHealth, (e) technical issues, and (f) social features. Overall, mHealth applications were perceived positively by participants. However, some noted negative experiences, particularly in the face of technical issues, competition or anxiety around self-monitoring. The results highlight the paucity of qualitative research in this area post-stroke and indicate that further qualitative research is warranted to guide the design of future mHealth interventions post-stroke (Carter et al., 2018). The findings of a recent Cochrane review demonstrate that there is currently not enough evidence (four small RCTs with 274 participants) to support the use of activity monitors to increase PA after stroke, with authors outlining the need for further research (Lynch et al., 2018). Effective clinical management of stroke often requires a sequence of treatments, each adapted to individual response, and hence multiple treatment decisions throughout the course of an individual's rehabilitation (Murphy, 2005). However, there is a lack of empirical data on the optimum sequence of these treatments. Due to the once-off randomisation of participants at baseline, a standard Randomised Controlled Trial (RCT) design, the gold-standard for the examination of causal relationships, does not enable the required flexibility to deliver and evaluate an adaptive intervention that is responsive to patient needs. RCTs are not designed to assess the effectiveness of individual intervention components. Adaptive RCT designs, such as Sequential Multiple Assignment Randomised Trials (SMART), can identify non-responders and allow for the empirical adaptation of subsequent physical activity treatments to realise larger benefits for some and reduce the use of less-effective therapies for many. SMARTs have been developed for the purpose of designing optimal adaptive interventions (Murphy, 2005). SMARTs are factorial designs in a sequential setting and can be described as multi-stage randomised controlled trial designs (Almirall, Nahum-Shani, Sherwood, & Murphy, 2014; Murphy, 2005). Each stage corresponds to a decision stage. All SMART participants are randomised at least once and some or all participants may be randomised more than once throughout the course of the trial. By enabling repeated randomisations of participants to treatments, the current study will use a SMART to design and evaluate an optimum adaptive physical activity intervention for people in the first six months post-stroke after discharge from formal rehabilitation. This adaptive intervention will include two intervention components: Structured Exercise and Lifestyle PA to increase physical activity, in addition to combinations of these treatment types. The objective of this SMART is to construct an adaptive intervention that will subsequently be evaluated against treatment-as-usual using a standard two-arm trial design. This SMART will determine the optimum sequence of embedded treatments to improve physical activity at 3-month follow-up.

Methods Design This trial will follow a SMART design (insert Figure of SAMRT design here). Participants will be randomized into one of two programme components; Structured Exercise or Lifestyle PA. At week 6, participants will be classified as either responders or non-responders according to their step count. Participants classified as responders will continue in their respective groups until the end of the trial. Participants classified as non-responders will be re-randomised to alternative treatment options for the second 6 weeks.

Defining responders and non-responders Classification of participants as responders or non-responders will be based on participants' 7-day mean step count at week six compared to their baseline 7-day mean step count. A relative improvement of at least 20% of each participant's baseline step count will determine them as a responder. For example, if a participant demonstrates a baseline mean steps/day of 3000 steps, an improvement of less than 600 mean steps/day will classify them as a non-responder. There is limited evidence available regarding anticipated improvements in step count in this population (Lynch et al., 2018). A relative improvement of at least 20% was chosen based on data from a feasibility trial of a smartphone-delivered, PA intervention for people with mild-to-moderate stroke living in the community (Paul et al., 2016). At the end of the intervention the experimental group has increased their average daily step count by 39.3% (4158 to 5791 steps/day). Taking a conservative approach, the cut-off for non-responders was set at approximately half of this or 20% in the current SMART.

Randomisation All participants will be randomised at baseline to either the Structured Exercise or Lifestyle PA groups. A simple equal allocation randomisation at the individual level using a computer-generated randomised list will be used. This will be concealed from the researchers involved in trial enrolment by using a centralised online service. At 6 weeks non-responders will be re-randomised following the same procedure.

Selection criteria.