Care home-level data

• Care home demographics.
• Staffing levels.
• Resident profile.
• Anonymous data relating to hospital admissions, general practitioner (GP) call-outs, mortality rates and falls in the previous 3 months.
• At baseline screening process: anonymous PAM-RC (Dr Julie Whitney, personal communication) and BI^48,49 scores for all residents.

Staff-level data (for those with face-to-face contact with residents)

• Basic demographic data.
• Person-centred care provided in the CH [Person-centred Care Assessment Tool (P-CAT)].⁶⁹

Resident-level data

Accelerometer

Daily wear time was calculated as the time between the first non-zero and last non-zero reading of the day, minus any periods within that time when there was ≥ 120 minutes of consecutive zero counts (classified as ‘non-wear’). Longer periods of non-wear of 180 minutes and 240 minutes were also examined to reduce misclassification of sedentary activity. Valid wear time for analysis was defined as ≥ 8 hours 25 minutes on ≥ 4 days (which could be non-consecutive days). For residents who provided valid data, the amount of time spent in various PA categories⁷⁶ and in sedentary behaviour was identified using cut-off points applied to the vertical axis accelerometer counts per minute (cpm) (see Forster et al.⁶⁶).

Estimation of recruitment and follow-up

Overall, 392 CHs were screened between June 2015 and September 2016: 13 (7.6% of those eligible) provided consent and 12 (7.0% of eligible) were randomised (see the CH recruitment paper by Ellwood et al.⁷⁷). A range of CHs were recruited in terms of size, location, ownership and provision (see Appendix 7, Table 23). Five CHs were randomised to the MoveMore intervention plus UC and seven were randomised to UC only.

The two strategies used to recruit CHs provided different success rates, as well as time scales to recruitment. A total of 377 CHs in West Yorkshire were screened via the CQC database. Of these, only 164 (43.5%) passed initial eligibility checks and were contacted by a researcher, and 11 (2.9% of those screened) consented. In contrast, 15 CHs in North Yorkshire were screened via the ENRICH network, 12 (80.0%) were contacted by researchers and two (13.3% of those screened) consented to participate.⁷⁷

Care home and staff characteristics

Resident screening and recruitment rates

Three hundred residents in the 12 CHs were screened for participation. Of these, 278 (92.7%) were eligible, 159 consented/had consultee agreement (57.2% of eligible; 53.0% of those screened) and 153 were registered to take part in the study (55.0% of eligible; 96.2% of those consenting/consultee agreement) (Figure 10).

FIGURE 10

Flow of residents through screening, consent, baseline assessment and registration. a, Reasons not mutually exclusive.

Forty-three (59.7%) of 72 residents with mental capacity were recruited. Of the 203 residents with doubtful or who were lacking mental capacity, 187 had a potential PC; 73 (57.9%) of the 126 PCs who agreed to act agreed for the resident to take part in the study. A NC was used for 76 residents, 43 (56.6%) of whom agreed for the resident to take part in the study.

There was variation across CHs in the proportion of eligible residents who were registered, but the overall proportions were similar between the arms: 62 out of 113 (54.9%) eligible residents were registered in the MoveMore arm, compared with 91 out of 165 (55.2%) in the UC arm (see Appendix 7, Table 27). On average, 12.4 residents per home were registered in the MoveMore arm, and 13 residents per home were registered in the UC arm, although, in one CH, only six residents were registered in each arm (see Appendix 7, Table 27).

The average time for recruitment at the CHs was 64 days; the shortest time was 50 days and the longest was almost twice this (97 days), although these times also included approximately 3 weeks per CH for baseline data collection (depending on the number of participants in each CH).

Follow-up rates

Care home and resident retention during the study period was high, with no CH or resident protocol violations or withdrawals, although one CH in the UC arm temporarily withdrew from researcher visits at 3 months because of renovations in the home (Figure 11).

FIGURE 11

The CONSORT diagram of the flow of CHs and residents through follow-up by randomised arm.

There was some variation in the proportion of residents followed up across the CHs (50.0–90.9% at 9 months) (see Appendix 7, Table 27), with a lower overall proportion in the MoveMore arm (69.4%) than in the UC arm (76.9%) (difference –2.65, 95% CI –17.50 to 12.20) (see Appendix 7, Table 28).

Losses to follow-up were due to deaths and movement of residents out of the CH (see Appendix 7, Table 29). Deaths and resident movement were also similar between arms at the CH level (see Figure 11 and Appendix 7, Table 28).

Screening characteristics

Overall, the consenting residents had similar characteristics to the whole screened population, and therefore can be considered a representative sample (see Appendix 7, Tables 29–31). However, a larger proportion of residents in the MoveMore arm than in the UC arm had had a previous stroke, and residents in the MoveMore arm had lower mean BI and PAM-RC scores (see Appendix 7, Table 30). In the intervention arm, the care of a higher proportion of residents was funded through the local authority and a lower proportion of residents were self-funded (see Appendix 7, Table 30).

There was similarity in screening characteristics between those eligible who consented and those who did not, with the exception of a higher proportion of residents with a diagnosis of dementia consenting (see Appendix 7, Table 31).

Resident baseline characteristics

Baseline characteristics for registered residents reflected the differences between the MoveMore and UC arms in the physical function and history of stroke that was observed at screening (see Appendix 7, Table 32).

In addition, those in the MoveMore arm also had greater cognitive impairment, as measured by the 6-CIT (see Appendix 7, Table 32), than those in the UC arm and a higher proportion had at least one comorbidity (see Appendix 7, Table 33).

A comparison of the baseline characteristics between those who were and those who were not followed up indicated that those not completing follow-up were more likely to be male, have dementia, have no history of stroke, have lower physical function and have greater cognitive impairment (see Appendix 7, Table 34). This is consistent with a higher (although not significant) proportion of residents in the MoveMore arm being lost to follow-up and the residents in this arm having a lower physical function and greater cognitive impairment.

Implementation of MoveMore components

A series of three interactive workshops were delivered individually to each of the five intervention CHs, with one CH repeating workshops 1 and 3 to ensure that all those potentially responsible for intervention implementation were included (see Appendix 7, Tables 35–37). Workshops took place at the CHs, with the exception of one CH, for which the workshops took place at its head office.

The implementation of the intervention was designed to be led by a team involving staff, residents (if possible) and their relatives/friends, facilitated by a senior member of CH staff (i.e. the MoveMore lead). The MoveMore lead was present for workshops 1 and 3 at all CHs, and for workshop 2 for four of the five CHs (see Appendix 7, Tables 35–37). The attendees were mainly (permanent) carers (see Appendix 7, Tables 35–37), although staff in other roles attended, particularly in CH5, including administration, ancillary, manager/deputy manager, team leader and specialist non-clinical roles, such as activity co-ordinator. With the exception of one CH, there were at least two staff members from each CH who attended all three workshops. One resident/relative attended workshop 1; in another CH, the executive director attended workshop 2. All workshops were delivered by researchers with key input from an experienced CH physiotherapist and an artist; workshops ranged in duration from 45 minutes to 2 hours 30 minutes, generally decreasing in duration with each workshop (see Appendix 7, Tables 35–37).

Time scales for intervention delivery

There were some challenges delivering the initial workshops for three CHs because of staff sickness, leave and temporary closure. Thus, in practice, the initial workshop was delivered 4–11 weeks after randomisation (Table 4). There were also some delays in subsequent workshops. Only CH5 completed all three workshops within 3 months of randomisation; for three CHs, the workshops spanned > 6 months (see Table 4).

TABLE 4

Timing of intervention workshops

The process of organising and scheduling the workshops was refined as the trial progressed and became more efficient.

Enhancing implementation

The amount and type of contact that the researchers had with the CHs varied. An initial meeting was introduced (after randomisation of the first CH) to set the scene for the intervention implementation. Initial contact was focused around administrative details (organising workshops, etc.); thereafter, further support was provided by the research team, with at least one visit to each CH to support intervention delivery (Table 5).

TABLE 5

Enhancing implementation

Completeness of MoveMore documentation

The methods used to capture adherence to the intervention paperwork outside the workshops were not successful, as the intervention folders that contained the relevant data were not always available to the researchers in the CHs.

Interestingly, five of the seven UC CHs wished to take up the intervention at the end of the trial. Of these, three went on to have the full series of workshops, one had one workshop before the manager disengaged and one wanted the workshops but was unable to release staff.

The PE (see Process evaluation and Appendix 9) provides further context.

Care home staff

At each time point, almost all CHs experienced staff leaving and joining the home during the preceding 3 months, but there was no suggestion of a difference in staff movement between the arms (see Appendix 7, Table 38).

Accelerometer data

For a full description of the accelerometer findings, see Appendix 8.

Accelerometer wear

The proportion of available residents who agreed to wear the accelerometer was similar across arms (88–97%) until the 9-month time point, at which time the proportion agreeing to wear the accelerometer was greater in the MoveMore arm (see Appendix 8, Table 54).

The number of days residents wore the accelerometers was similar between the arms and, within wear days, residents wore the accelerometers for ≈ 10–11 hours (see Appendix 8, Table 54).

The proportion of residents wearing the accelerometer who met the minimum wear criteria for analysis (≥ 8 hours 25 minutes per day for ≥ 4 days) differed between the arms at each time point (see Appendix 8, Table 54). Even so, at each time point, the mean number of days the accelerometer was worn for was > 6 in both arms and mean daily wear time was > 10 hours 18 minutes (see Appendix 8, Table 54).

A preliminary review suggests that the data recorded on the activity logs provided incomplete information on when the accelerometer had been put on and taken off each wear day. Hence, the planned comparison between accelerometer and wear logs was not conducted.

Levels of physical activity and sedentary behaviour

Patterns of physical activity and sedentary behaviour

Completeness of follow-up questionnaire data

Questionnaires completed by the researchers with staff informants (i.e. the FAC, the BI and the PAM-RC) had a high completion rate, and completion levels were similar between arms at all time points (96–100%) (see Appendix 7, Tables 40 and 41). The time taken to complete these questionnaires was constant across arms and time points, with each questionnaire taking, on average, 1–3 minutes.

The EMS was completed by the researchers with staff informants, but also included two physical assessments (timed walk and functional reach). Completion levels for the EMS were variable (see Appendix 7, Tables 40 and 41). At each time point, the two physical assessments had lower completion rates (86–97%) than the other questions in the EMS (95–100%). The most common reasons for non-completion of these two physical assessments were that the resident declined to do them, the resident did not understand what they were being asked to do or was unable to follow instruction, and asking the resident to do these was not suitable because they were very frail, unwell or too tired. Hence, the pattern of completion observed between the arms was consistent with residents in the MoveMore arm having lower cognitive ability at baseline and those with higher cognition being included in the follow-up in both arms.

Among those questionnaires completed by the researcher with the resident (i.e. the GDS, the WHOQOL-OLD, the DEMQOL, the EQ-5D-5L and the 6-CIT), the level of completion was variable, ranging from 33% to 74% (see Appendix 7, Tables 42 and 43). Completion differed between the study arms: for all questionnaires at all time points, completion levels were higher in the MoveMore arm. The most commonly reported reasons for non-completion of the resident-completed questionnaires included staff members and NCs advising that the resident would struggle to understand questions or would be unable to communicate verbally, lack of co-operation or response to the questions, and the resident not being well enough to be approached.

The DEMQOL and the 6-CIT appeared to have slightly poorer completion rates than the GDS, the WHOQOL-OLD and the EQ-5D-5L (completion rates: DEMQOL, 34.1–64.0%; 6-CIT, 32.9–62.0%; GDS, 40.0–66.0%; WHOQOL-OLD, 34.3–69.8%; EQ-5D-5L, 42.9–74.0%).

As would be expected, staff proxy-completed questionnaires had a higher rate of completion than the resident-completed versions (completion rates for resident-completed vs. proxy-completed: DEMQOL, 34.1–64.0% vs. 82.4–100.0%, respectively; EQ-5D-5L, 42.9–74.0% vs. 91.4–100.0%, respectively) (see Appendix 7, Table 44).

The completion of the staff P-CAT was variable in both arms at all time points, with scores available for all but one staff member of those who returned the questionnaire booklets in the MoveMore arm at 3 months (see Appendix 7, Table 45). At baseline, completion levels were similar between the arms, but, at 6 and 9 months, rates of completion were lower in the UC arm. Examination of the item-level completion indicated that the questions on whether or not the ‘environment feels chaotic’ (question 8) and ‘assessment of residents’ needs is undertaken on a daily basis’ (question 11) had lower levels of completion in both arms at all time points (see Appendix 7, Table 45).

Assessment of physical activity outcome

There is limited research available on PA in this population to inform choice of the cut-off points for low-intensity PA or sedentary outcomes. We therefore planned to use the data generated to determine appropriate PA end point(s) for future use in a definitive trial. This decision was to be made following a review of the summary statistics for each of the levels of PA and emerging evidence on physiological effects (e.g. that increasing the number of breaks in sedentary behaviour is more important than reducing the total amount of sedentary behaviour), and would also reference information on (1) numbers of missing data, (2) ceiling and floor effects and (3) sensitivity to change.

Based on the design of the REACH study, the most appropriate analysis and corresponding coefficient to assess sensitivity to change is a correlation of change scores with residents’ retrospective global ratings of change. However, this study was not designed to assess sensitivity to change; therefore, we do not have a global rating of change measure. In addition, although the number of residents wearing the accelerometer was much higher than in other research conducted in this population, the numbers of registered residents wearing an accelerometer and achieving the minimum wear time criteria required for analysis was relatively low for informing a primary outcome. Hence, it was not feasible to investigate this to inform the most appropriate PA end point.

Estimation of effectiveness

Following discussions that it was not appropriate to make a statistical decision on the most appropriate PA end point(s) for future use in a definitive trial, planned analyses on the preliminary estimate of effectiveness were not conducted. Furthermore, it was also inappropriate to conduct any further formal cluster-level analysis on the accelerometer data. At 9 months, the number of residents per cluster ranged from 2 to 14 (mean 7.0) for sedentary outcomes, whereas, for the light-intensity outcomes, there were fewer than four homes per arm; therefore, a formal cluster-level analysis would not produce any meaningful conclusions. Cluster-level analysis is the only recommended approach to analysis with a small number of clusters per arm.

Assessment of questionnaire outcomes (see Appendix 7, Table 46)

There was no evidence of a significant difference in EMS scores between the two groups, although residents in the MoveMore arm had lower EMS scores at baseline, indicating lower mobility. During the study, scores decreased in both arms, with evidence of a difference emerging between the arms from 6 months, with lower scores in the MoveMore arm (difference 3.23, 80% CI 0.25 to 6.22), but caution is required in interpretation given the differing response rates. The BI and PAM-RC scores were significantly lower in the MoveMore arm at baseline, indicating less self-care, physical ability and PA. During the study, scores fluctuated in the MoveMore arm and decreased in the UC arm, such that, by 9 months, there was no evidence of a difference between the arms (BI: 2.55, 80% CI –0.43 to 5.52; PAM-RC: 2.06, 80% CI –0.28 to 4.41).

Sample size estimation

To inform the sample size estimation for a definitive trial, information is required on the average cluster size and coefficient of variation, as well as an estimate of the ICC for the proposed outcomes.

The mean cluster size, as defined by the number of registered residents per CH, was 12.4 (SD 4.51) in the MoveMore arm (62 registered residents; five CHs) and 13 (SD 5.69) in the UC arm (91 registered residents; seven CHs). These correspond to a coefficient of variation in cluster size of 0.36 in the MoveMore arm and 0.44 in the UC arm.

As outlined in Variation in outcomes between care homes, the ICC estimates for the PA outcomes, particularly at 9 months, were not reliable based on such a small number of CHs and residents in CHs. The ICC estimates and 95% CIs for the various resident questionnaires are provided in Appendix 7, Tables 46 and 47. However, as with the PA outcomes, the ICC estimates for the questionnaires completed with the resident were not reliable because of the reduced numbers of residents completing these questionnaires.

Falls

At baseline, a higher proportion of residents at the CH level in the MoveMore arm than in the UC arm experienced falls in the previous 3 months, and the average number of falls per resident was also higher, although not significantly (see Appendix 7, Table 48). Estimates at the CH level were consistent with the proportions and average fall rate observed among registered residents. The average number of falls per resident was also higher (MoveMore arm mean 2.50, 95% CI 2.00 to 3.00; UC arm mean 1.62, 95% CI 1.10 to 2.13), although not significantly (see Appendix 7, Table 49).

Over the study period, at the CH level, the average number of falls per resident declined in both arms, although not significantly, and there was no difference at 9 months between the arms (see Appendix 7, Table 48).

Among registered residents, over the study follow-up, the proportion of registered residents who fell was similar to that observed at baseline (MoveMore arm, 43.5%; UC arm, 38.5%). In both arms, there was, however, an increase in the average number of falls per resident (MoveMore arm mean 3.37, 95% CI 2.03 to 4.71; UC arm mean 2.00, 95% CI 1.57 to 2.43), although the CIs indicate that these increases were not significant, compared with baseline, and there was no difference in the average fall rate between arms (see Appendix 7, Table 49). In both arms, the reasons for falls among registered residents were largely unknown, as residents were commonly found on the floor. In cases for which the reasons were known, the majority had fallen from their bed.

Hospitalisations

At the CH level, at baseline, the number of registered residents with hospitalisations and the number of hospitalisations per resident over the previous 3 months were similar between arms [number of residents with hospitalisations: MoveMore arm, 8.8% and UC arm, 9.2%; number of hospitalisations per resident: MoveMore arm mean 1.64 (95% CI 1.33 to 1.95) and UC arm mean 1.47 (95% CI 1.34 to 1.61] (see Appendix 7, Table 50). At follow-up, there was an increase in the proportion of residents admitted to hospital in both arms, whereas the number of hospitalisations per resident decreased in the MoveMore arm and increased in the UC arm. Hence, the proportion of residents with hospitalisations and the average number of hospitalisations per resident was lower in the MoveMore arm at follow-up: approximately one-quarter of residents were hospitalised with, on average, 1.60 and 1.27 hospitalisations per resident in the UC and MoveMore arms, respectively (see Appendix 7, Table 51).

When examining the NHS Digital data for hospital admissions among registered residents (see Appendix 7, Table 50), there are marked differences, compared with CH reporting, particularly in the MoveMore arm (see Appendix 7, Table 51). Both the number of hospitalisations and the number of residents with hospitalisations were under-reported via the CH, compared with NHS Digital (proportion of residents with hospitalisations: MoveMore arm, 38.7%; UC arm, 35.2%), although, at the CH level, there were a few homes across both arms for which the number of hospitalisations agreed. In the UC arm, the number of hospitalisations was similar from both sources, but the number of residents with hospitalisation was higher according to NHS Digital. Hence, in contrast to the CH reports, the average number of hospitalisations per resident was higher in the MoveMore arm and in line with the average number of hospitalisations per resident at the CH level. The number of unplanned admissions at the CH level was also under-reported via the CH, compared with NHS Digital. However, this may be because CH-level data from NHS Digital were obtained via the CH postcode, and therefore may include data from those not resident at the CH.

Accident and emergency visits

The mean number of accident and emergency (A&E) visits at the CH level over the study period for each arm was as follows: MoveMore arm, 1.45 (95% CI 1.27 to 1.64); UC arm 1.57 (95% CI 1.40 to 1.74) (see Appendix 7, Table 50). At follow-up, the mean number of visits was 1.22 (95% CI 0.88 to 1.56) for the MoveMore arm and 1.26 (95% CI 0.99 to 1.53) for the UC arm (see Appendix 7, Table 52).

The mean number of A&E visits per registered resident for each arm at baseline was as follows: MoveMore arm: 1; UC arm 1.07 (95% CI 0.92 to 1.23). At follow-up, the mean number of visits was 1.22 (95% CI 0.88 to 1.56) for the MoveMore arm and 1.26 (95% CI 0.99 to 1.53) for the UC arm. The number of residents with an A&E visit appeared to be lower in the MoveMore arm [n = 5 (8.1%)] than in the UC arm [n = 14 (15.4%)] (see Appendix 7, Table 52). The number of A&E visits reported via NHS Digital was much higher than that reported via the CHs. (see Appendix 7, Table 52).

Deaths

The number of deaths was similar between both arms over the study period: MoveMore arm, 29 (6.9%); UC arm, 31 (6.5%) (see Appendix 7, Table 50).

Among registered residents, a similar proportion of deaths occurred in both arms: MoveMore arm, 19.4%; UC arm, 17.6% (see Appendix 7, Table 53).

Researcher unblinding

There were 11 instances of researcher unblinding. One instance occurred between baseline and the 3-month data collection via telephone calls from CH staff requesting to speak to an unblinded researcher, and another at the 9-month data researcher visit. Two instances occurred during the monthly safety calls, during which the researchers were informed of the allocation by CH staff. The remaining instances occurred at the 6-month researcher visit: allocation was revealed during discussions with staff (three instances), by residents (two instances) and via the display of the intervention materials in the CH (two instances).