Multifactorial falls prevention programme compared with usual care in UK care homes for older people: multicentre cluster randomised controlled trial with economic evaluation

Abstract Objectives To determine the clinical and cost effectiveness of a multifactorial fall prevention programme compared with usual care in long term care homes. Design Multicentre, parallel, cluster randomised controlled trial. Setting Long term care homes in the UK, registered to care for older people or those with dementia. Participants 1657 consenting residents and 84 care homes. 39 were randomised to the intervention group and 45 were randomised to usual care. Interventions Guide to Action for Care Homes (GtACH): a multifactorial fall prevention programme or usual care. Main outcome measures Primary outcome measure was fall rate at 91-180 days after randomisation. The economic evaluation measured health related quality of life using quality adjusted life years (QALYs) derived from the five domain five level version of the EuroQoL index (EQ-5D-5L) or proxy version (EQ-5D-5L-P) and the Dementia Quality of Life utility measure (DEMQOL-U), which were self-completed by competent residents and by a care home staff member proxy (DEMQOL-P-U) for all residents (in case the ability to complete changed during the study) until 12 months after randomisation. Secondary outcome measures were falls at 1-90, 181-270, and 271-360 days after randomisation, Barthel index score, and the Physical Activity Measure-Residential Care Homes (PAM-RC) score at 91, 180, 270, and 360 days after randomisation. Results Mean age of residents was 85 years. 32% were men. GtACH training was delivered to 1051/1480 staff (71%). Primary outcome data were available for 630 participants in the GtACH group and 712 in the usual care group. The unadjusted incidence rate ratio for falls between 91 and 180 days was 0.57 (95% confidence interval 0.45 to 0.71, P<0.001) in favour of the GtACH programme (GtACH: six falls/1000 residents v usual care: 10 falls/1000). Barthel activities of daily living indices and PAM-RC scores were similar between groups at all time points. The incremental cost was £108 (95% confidence interval −£271.06 to 487.58), incremental QALYs gained for EQ-5D-5L-P was 0.024 (95% confidence interval 0.004 to 0.044) and for DEMQOL-P-U was 0.005 (−0.019 to 0.03). The incremental costs per EQ-5D-5L-P and DEMQOL-P-U based QALY were £4544 and £20 889, respectively. Conclusions The GtACH programme was associated with a reduction in fall rate and cost effectiveness, without a decrease in activity or increase in dependency. Trial registration ISRCTN34353836.

• A care home Falls Champion, with roles for awareness raising and liaison between staff and the NHS Falls Lead, was identified in each care home. No additional training was provided.
• Trained staff were taught to complete the GtACH screening and assessment tool with every resident, within four weeks after training, in private, and discussed with family, friends and other care home staff. Completed GtACH documentation was to be placed in the resident's care records and expected to contribute to the care plan for each resident. Re-assessment was expected if the resident developed a new health condition, or fall, or every three-six months.
Item 6: Mode of delivery • Training was given face to face. Item 7: Where the training occurred • Training was provided in a room within the care facility which was deemed appropriate for training by the care facility manager/owner.

Item 8: Duration/Intensity
• Each session was one hour. As it was not feasible for all care home staff to attend a single training session, repeated training sessions were offered according to care home staff availability. Staff were not remunerated, their roles were not back filled, repeated sessions maximized opportunity for all staff to attend. Item 9: Tailoring • Completed GtACH documentation was to be placed in the resident's care records and expected to contribute to the care plan for each resident. Re-assessment was expected if the resident developed a new health condition, or fall, or every three-six months.

Item 10: Modifications
• The training period for care homes was extended from two weeks to four weeks (post randomisation) to enable fall leads time to train very large homes (+50 staff members).

Item 11/12 : Fidelity
• Fidelity of the intervention is reported separately in the process evaluation which was conducted in parallel to the trial.

Supplementary material S2 -Statistical Analysis Plan
Section 1: Introduction

Background and rationale:
Preventing falls and injuries in those over 65 years of age is a public health priority (RoSPA 2013) and The King's Fund recommends structured patient-centred care in care home settings (Naylor 2013).
The recently published NICE Quality Standard 86, 'Falls in older people: assessment after a fall and preventing further falls' (NICE, 2015), recommends that all health and social care practitioners involved in assessing, caring for and treating older people who experience a fall should have sufficient and appropriate training and competencies to deliver the actions and interventions.
Community fall prevention interventions reduce falls by about 30%, but literature to date has found no conclusive reduction in falls in care homes (Gillespie 2012, Cameron 2012. To decrease fall rates it has been suggested that interventions need to be targeted at high risk groups such as elderly care home populations and include specific components (Close, 2005). They need to be delivered by the whole team (Bouwen 2008, Jensen 2004 to the whole environment.
The GtACH intervention aims to reduce fall rates in care homes by facilitating change in practice of care home staff. It was co-produced by a group of care home staff, clinicians, researchers, public, voluntary and social care organisations and includes care home staff training, support and documentation.

Objectives
The primary objective of the FinCH trial is to compare the rate of falls per participant in the 2 trial arms (GtACH arm and usual practice control arm) during the 3 month period comprising 4,5 and 6 months post randomisation.
Secondary objectives listed in the protocol are as follows:: i. Comparison of fall rates between the two groups during the 3-month period comprising 7, 8 and 9 months post randomisation. ii.
Comparison of fall rates between the two groups during the 3-month period comprising 10, 11 and 12 months post randomisation. iii.
Comparison of frequency of falls injuries between the two groups for falls occurring between baseline and 6 months, and between 7 and 12 months post randomisation. iv.
Comparison, between the two groups, of frequency and type of fractures occurring between baseline and 6 months, and between 7 and 12 months post randomisation. v.
Comparison, between the two groups, of physical activity , measured using PAM-RC, at 6 months, and 12 months post randomisation. vi.
Comparison, between the two groups of functional ability, measured using Barthel index, at 6 months, and 12 months post randomisation. vii.
Comparison, between the two groups, of quality of life, measured using DEMQOL and EQ5-D at 6 months, and 12 months post randomisation. viii.
Comparison, between the two groups, of medication use between baseline and 6 months, and between 7 and 12 months post randomisation. ix.
Comparison, between the two groups, of number of days in hospital between baseline and 6 months, and between 7 and 12 months post randomisation. x.
Comparison, between the two groups, of percentage of deaths

Changes from the protocol
The following objective has been added, on the recommendation of the DMC i Comparison, between the two groups, of percentage of residents falling, between 3 and 6 months post-randomisation, and between 7 and 12 months post-randomisation Changes to original secondary objectives i original objective viii (Comparison, between the two groups, of medication use between baseline and 6 months, and between 7 and 12 months post randomisation) has been amended to Comparison, of risk of falling, among those on 4 or more medications at baseline Full details of the trial are given in the protocol.

Trial Design
FinCH is a cluster randomised controlled, 2 arm, parallel group trial comparing the GtACH fall prevention intervention against usual care for people living in care homes in England (with and without nursing). Care home is the unit of randomisation.

Randomisation
Care homes are randomised on a 1:1 basis to one of two parallel arms: intervention (GtACH fall prevention programme) or control (usual care).
Randomisation is based on a bespoke computer generated pseudo-random code using variable block randomisation (block sizes 2 and 4) within strata (site [Lincolnshire, Derby, Northumbria, Leicester, Stafford, Norwich, Nottingham City, Nottinghamshire, Bradford and Solent] and care home type [nursing/residential/dual registration]) provided by the Norwich CTU via a secure web based randomisation service.

Sample size
The most recent sample size calculation was based on a falls rate of 2.5 falls per person per year (0.625 falls in 3 months) with 80% power and a two-sided significance level of 5%. Based on an average cluster size of 19 (SD 9.5) residents per cluster, a coefficient of variation (CV) of 0.5 and allowing for 16% attrition, the sample size target was 78 care homes -1482 residents in total. Full details of this sample size calculation, and the original calculation, may be found in sections 9.1.1, 9.1.2 and 9.1.3 of the FinCH protocol.

Framework
The FinCH trial is to determine whether fall rates are reduced following the implementation of the GtACH intervention and therefore is testing for superiority. Secondary outcomes will also be tested for superiority.

Statistical interim analyses and stopping guidance
There are no interim analyses planned for FinCH.

Timing of final analysis
All analyses will take place at one time point once all data are cleaned and locked.

Timing of outcome assessments
Data on falls, medication use, time in hospital and use of primary care and community services for the previous 3 months ("baseline") are collected at the time of resident recruitment, then at 3 months following randomisation, and at 3 monthly intervals thereafter, with the final collection of these items taking place at approximately 12 months post-randomisation. Data on Physical activity (PAM-RC), Activities of Daily Living (Barthel), Health related Quality of Life in Dementia (DEMQOL-U-5D / DEMQOL-P-4D, EQ-5D-5L/EQ-5D-5L proxy) will be collected from participating residents at the same 3 month intervals. Frequency and type of fractures occurring during the period of baseline and outcome data collection will obtained from Hospital Episode Statistics at the end of the 12 month follow-up. Full details of timing of collection of outcome measures are included in the FinCH trial protocol.

Confidence intervals and p-values
All statistical tests will be 2-sided and performed using a 5% significance level. All confidence intervals presented will be 95% and 2-sided. No adjustment for multiplicity is planned.

Adherence and protocol deviation
Compliance with the intervention is based on the percentage of care giving staff in each care home trained to use the GtACH tool. It is calculated as follows: Percentage compliance will be calculated and presented for each home in the intervention arm; average compliance for all intervention care homes will also be presented.

Analysis populations:
Analyses will be undertaken on an Intention to Treat basis in which care homes (and corresponding participating residents) will be analysed in the group to which they were allocated regardless of their compliance with the intervention). Those who died between care home randomisation and the three month follow-up data collection will be regarded as having been exposed to the intervention (GtACH/control), recorded as lost to follow-up at the time of death, and included in the consort diagram. The consort diagram will include the number of people who were randomised in error -eg those who were randomised but did not fulfil all the eligibility criteria.

Screening data
The following summaries will be presented for all screened care homes, both overall and by study site: Number of screened care homes The number of care homes not recruited for the following reasons: would not benefit the care home; have not got the time; not interested in research; other

Eligibility
Care homes: the number of ineligible care homes will be reported, along with reason(s) for ineligibility Residents: the number of eligible and ineligible residents will be reported, along with reason(s) for ineligibility A list of inclusion and exclusion criteria for care homes and residents is provided in section 7.5.2 of the protocol.

Recruitment
The Consort diagram will be used to summarise the following information:

Baseline characteristics
Baseline characteristics of randomised care homes and participating residents will be presented according to Table 1 and Table 2 respectively, both overall and by randomised group. No formal hypothesis tests will be undertaken. Categorical data will be summarised by numbers and percentages; continuous data will be summarised by mean, SD and range if data are normal, and median, IQR and range if data are skewed.  -Dementia -Diabetes -Stroke -CHD Number of falls during period 3 months prior to baseline data collection Number of Medications in period 3 months prior to baseline data collection -None -One to three medications -Four or more medications Physical activity (PAM-RC) score at baseline: mean(SD) Activities of Daily Living (Barthel) score at baseline: mean(SD) DEMQOL-U-5D at baseline DEMQOL-P-4D at baseline EQ-5D-5L at baseline EQ-5D-5L proxy at baseline

Section 6: Analysis Outcome definitions i) Falls rates: Primary outcome
The primary outcome is the rate of falls per participating resident during the 3 month period comprising months 4, 5 and 6 post-randomisation. For the purpose of the analyses, a month will be taken to consist of 30 days. Baseline, or time zero, for each care home will be the date of randomisation of that care home. For participating residents at each care home, month 4 will therefore start 91 days after randomisation, and month 6 will finish 180 days after randomisation. Therefore for each participating resident the numerator of the falls rate will be the number of falls occurring in the care home between 91 and 180 days. The denominator will be 90 days, for those participating residents who have lived at the care home for the entire 90 day period; for those participating residents who have spent time in hospital or elsewhere during the 90 day period, the denominator will be the number of days during that 90 day period spent as resident in the care home. The falls rate will be expressed as the number of falls per 1000 participating resident days for each group. For individuals who stopped participating in the trial before 180 days, data will be included up until the date of withdrawal.

Secondary outcomes
Rate of falls occurring during the 3 month period comprising months 7, 8 and 9 post-randomisation, and the 3 month period comprising months 10, 11 and 12 post-randomisation. As above, a month will be assumed to consist of 30 days. The 3 month period comprising months 7, 8 and 9 will therefore commence on day 181 and end on day 270. Likewise the 3 month period comprising months 10, 11 and 12 will commence on day 271 and end on day 360. Falls rates will be calculated as described for the primary outcome, and expressed as the number of falls per 1000 participating resident days for each group.

ii) Other secondary outcomes:
The number and percentage of residents having one or more falls, between 3 and 6 months postrandomisation, and between 7 and 12 months post-randomisation.
The number and percentage of residents having no medications at baseline, who have one or more falls, between 3 and 6 months post-randomisation, and between 7 and 12 months postrandomisation.
The number and percentage of residents having 4 or more medications at baseline, who have one or more falls, between 3 and 6 months post-randomisation, and between 7 and 12 months postrandomisation.
Fractures (from HSCIC data) occurring between baseline and 6 months and between 7 months and 12 months: • Number of fractures of any type, per resident • Number of residents having one or more fractures of any type • Number of hip fractures, per resident • Number of residents having one or more hip fracture • Number of wrist fractures, per resident • Number of residents having one or more wrist fracture Physical activity (measured by the PAM-RC questionnaire) at 3, 6, 9 and 12 months following randomisation. The PAM-RC questionnaire comprises 5 questions addressing ability (mobility and balance), and activity (walking frequency, wandering and outdoor mobility). Each question has graded responses, scored between 0-3 and 0-6. The total PAM-RC will be calculated by summing the scores for each question, giving a range of scores between 0 and 21, with 0 indicating no activity, and 21 indicating that the resident is fully mobile in all aspects of the questionnaire.
Activities of daily assessment (measured by the Barthel Index) are recorded at baseline, 3, 6, 9 and 12 months following randomisation. The Barthel Index of Activities of Daily Living comprises ten questions concerning resident's ability to self-care. Each item is scored between 0 and 1, 2 or 3, with minimum possible score 0 and maximum possible score 20; lower values indicating less independence with respect to Activities of Daily Living. Total ADL scores will be treated as continuous data.
Quality of life in dementia, measured using DEMQOL-U for residents with capacity and DEMQOL-Proxy for residents lacking capacity, is recorded at baseline, and at 3, 6, 9 and 12 months following randomisation. DEMQOL comprises 28 questions and DEMQOL-Proxy 31 questions, each scored between 1 and 4. Scores are summed to give total DEMQOL and DEMQOL-Proxy scores with possible ranges 28 to 112 and 31 to 124 respectively; higher scores imply better Quality of Life. DEMQOL and DEMQOL-Proxy also each include a single global health question -this is not included in the total DEMQOL or DEMQOL-Proxy score. Total DEMQOL and DEMQOL-Proxy scores will be analysed as continuous data.
Health status measured using EQ-5D-5L (where the participating resident has capacity), and EQ-5D-5L proxy (where the participating resident does not have capacity), is recorded at baseline and at 3, 6, 9 and 12 months post-randomisation. At each timepoint responses to the 5 component questions of the EQ-5D_5L/EQ5D-5L proxy will be converted to a weight. The values of weights range from -0.285 to 0.950, with higher weights indicating more favourable health states. A further question asks residents / proxy respondents to score how well or poorly they rate their health on that particular, on a scale ranging from 0 (worst imaginable health state) to 100 (best imaginable health state).
The number of days spent by residents in hospital (from HSCIC data) between baseline and 6 months, and between 7 months and 12 months post-randomisation The number of residents who die (from HSCIC data)

Primary outcome
The average number of falls occurring during the 3 month period comprising months 4,5 and 6 post randomisation (ie between 91 and 180 days post-randomisation) per participating resident, and the average rate of falls, occurring during the 3 month period comprising months 4, 5 and 6 post randomisation (again number of days spent as resident in the care home between 91 and 180 days post-randomisation, excluding any days spent in hospital during this period), per participating resident, will be presented by treatment group. Falls rates will be expressed as the number of falls per 1000 participating resident days. The primary effect estimate will be Incidence Rate Ratio, reported with 95% confidence interval.
The number of falls per resident will be compared between groups using a random effects/hierarchical two-level Poisson model with resident at level one and care home at level two, with length of residence in care home as an offset. The primary analysis will adjust for type of care home (residential, nursing, dual registration) and site.

Adjustment for covariates
Two further models will adjust for i) baseline fall rate; ii) baseline fall rate and other variables that are associated with falling, in addition to adjusting for care home type and site. Baseline fall rate will comprise information collected on falls in the care home for the 90 days prior to the date of baseline data collection. In the case of participants resident in the care home for fewer than 90 days prior to the baseline data collection, the fall rate will be calculated for the period in which data are available. For residents not fulfilling a minimum period of four weeks of baseline data collection in the care home, missing data will be imputed.

Methods used for assumptions to be checked for statistical methods
A negative binomial model will also be fitted, and goodness of fit will be compared between the two models, and the best-fitting model will be reported.

Details of alternative methods to be used if distributional assumptions do not hold
If neither Poisson nor negative binomial models are appropriate then alternative models will be explored, or two-stage bootstrap will be used. Alternatively cluster level analyses could be performed.

Secondary outcomes
Falls rates for the three month period comprising 7,8 and 9 months post randomisation and for the 3 month period comprising 10, 11 and 12 months post-randomisation will be analysed in a similar way to the primary outcome. Frequency and type of fractures and days in hospital between baseline and 6 months (180 days), and for the period between 6 months (181) and 12 months (360 days).
For other secondary outcomes, groups will be compared using multi-level regression analysis for continuous outcomes (DEMQOL, EQ-5D-5L, medication use, Barthel ADL and PAM-RC) and multilevel logistic regression for binary outcomes (death, at least one fall or not).

Missing data
A multiple imputation using iteratively chained equations [1] will be used to account for incomplete data and missing data in other outcomes. A random effect will be used to account for the clustering by care-home. The total number of imputations will be approximately the same as the percentage of cases that are incomplete, up to a maximum of 20 imputations. The estimates from imputation will be combined using Rubin's equations. The imputation model used will include all outcome measures and any baseline covariates which are associated with loss-to-follow-up and treatment group.
If any of the variables have a skewed distribution, then transformations will be attempted, however if none are found then predictive mean matching will be used for these variables.
The imputation model may be adapted during the analysis if a) Perfect prediction is observed. This occurs when the variance-covariance matrix is singular. If this occurs then 'augmenting' will be attempted, but if this does not resolve the issue then variables may be removed from the imputation model. b) If the imputation procedure does not converge then it will be necessary to remove variables from the imputation model. c) If the imputation model includes too many variables then instability in the imputation may occur. In this case it may be required to remove variables from the imputation model.
In order to avoid potential issues in model mis-specification, the imputation may not be attempted for outcomes with more than 50% missing data. Multiple imputation will only be used for the intention-to-treat analysis.

Details of any additional statistical analyses required
No other analyses are planned at this time

Safety Analyses
No additional analyses will be conducted other than those specified

Statistical software
The analysis will be carried out using standard statistical software, either Stata, SAS or R.