• No results found

Statistical methods

A detailed statistical analysis plan was written by the team statisticians and approved by the TSC (statistical analysis plan version 1.0, dated 15 March 2018) prior to trial database lock.

Analytical approach

All analyses were undertaken in accordance with appropriate analytical and reporting guidelines.116Primary

analysis (in the form of summary statistics, not formal/inferential analysis) was undertaken on an intention- to-treat basis, whereby participants were analysed according to their allocated group, regardless of adherence to the protocol or lack of participation or completion if allocated to the intervention group.

Statistical significance levels

As this was a feasibility trial, no inferential between-group comparisons were undertaken (i.e. there was no between-group hypothesis testing). When presented, confidence intervals are at the 95% level, unless otherwise stated.

Interim analysis

There was no planned interim analysis for this trial.

Time points of statistical analysis

Statistical analysis was undertaken once the final group of participants completed the final assessment at 27 (±1) weeks post randomisation and the database was locked following final approval and sign-off of the statistical analysis plan by the TSC.

Data sources and data quality

The data from this trial came from information entered onto CRFs completed by a blinded research therapist at baseline, 15 (±1) and 27 (±1) weeks post randomisation. All participants were asked to complete a 2-weekly self-reported pre-formatted paper diary. In addition, intervention participants were asked to complete an online exercise diary to record their adherence to the programme. Attendance at the BRiMS face-to-face sessions and the number of log-ins to the online exercise portal were also recorded.

Missing data

One of the objectives of this feasibility trial was to assess the completeness of potential outcome measures for the definitive trial, at the level of both item and outcome measure. Missing outcome data were noted and used to inform the likely pattern of missing data in a full-scale trial.

Imputation methods

For the validated outcome measures MSWS-12vs2,61MSIS-29vs261,72,73and FES-I,101the established

methods for imputing missing item-level data were implemented, when the minimum number of items required to impute the missing data was met.

If the participants completed at least 11 of the 14 Mini-BEST test components, the final score was imputed by replacing missing values with the mean of the non-missing test component scores.

The mean of the three Functional Reach Test values (forwards and lateral) was calculated and analysed. If participants were missing any of the three repeated test components of the Functional Reach Tests, the mean of the successful attempts was used.

If there were up to four missing items from the FES-I score, the total score was imputed by replacing the missing items with the mean score.117

A validated imputation method was not available for any CPI score. Therefore, if a participant was missing at least one item, he or she was excluded from the analysis.

Statistical software

The statistical analyses were undertaken using Stata/SETMversion 14 (StataCorp LP, College Station, TX, USA),

supplemented, where required, by R (The R Foundation for Statistical Computing, Vienna, Austria).

Statistical analyses

As this was a feasibility trial, it was not powered to be able to support or justify any conclusions regarding treatment effectiveness and efficacy realised from hypothesis testing,23and, indeed, that was not the

purpose of the trial. As such, the analysis of the results did not involve formal/inferential statistical

comparisons between groups, but rather it was descriptive with the view to informing the design of a fully powered BRiMS RCT.

Continuous measures were summarised as means, SDs and ranges when the distribution appeared normal, and as medians, interquartile ranges (IQRs) and ranges when the distribution appeared otherwise. Categorical data were summarised by frequencies and percentages. When appropriate, parameter estimates (e.g. between- group differences) were presented with 95% confidence intervals (CIs). With the exception of the falls diary analyses (seeAnalysis of patient-reported and clinician-rated outcome measures), any potential outliers were identified and reported but not removed from the descriptive statistics of this feasibility trial unless stated. Analyses of quantitative data were conducted to summarise feasibility outcomes (objectives i–vi), evaluate acceptability and adherence to BRiMS (objective xii), and the completion and summary statistics of the planned primary and secondary patient-reported and clinical outcomes measures (objective vii). In addition, appropriate plots were used to illustrate key data and assess potential relationships.

Trial population

Data from the screening process through to the completion of the trial were recorded and presented in a CONSORT-style flow diagram.116

Baseline characteristics and demographics

Baseline characteristics, collected before randomisation, were summarised by allocated group to informally check for balance between groups (by visual inspection) and provide an exploratory overview of the trial population.

DOI: 10.3310/hta23270 HEALTH TECHNOLOGY ASSESSMENT 2019 VOL. 23 NO. 27

© Queen’s Printer and Controller of HMSO 2019. This work was produced by Gunnet al.under the terms of a commissioning contract issued by the Secretary of State for Health and Social Care. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

Analysis of randomised groups at baseline is not good practice118and so this was not undertaken, but any

considerable imbalances were noted to inform the design of the full trial.

Participants who discontinued, withdrew or were lost to follow-up

It was possible that participants would withdraw consent part-way through the trial, or that their treatment would be discontinued for medical reasons. It was unlikely that a participant would be discontinued on medical grounds (in either allocated group), but for reasons such as injury, some participants may not have been able to complete the trial. Participants who discontinued were categorised as follows:

l continued to consent for follow-up and data collection

l consented to use pre-collected data only

l complete withdrawal of consent to use any data.

Reasons for withdrawal or loss to follow-up were summarised, where these were reported, at each stage of the process. These included‘participant withdrew before randomisation’,‘participant did not receive their allocated treatment’,‘participant did not complete treatment’and‘participant was lost to follow-up’. Participants who withdrew from the trial, or whose treatment was discontinued on medical grounds, were not replaced. No participant who withdrew from the trial requested that their previously collected data be removed from the trial database. The extent of discontinuation, withdrawal and loss to follow-up will be used to inform the design of the anticipated fully powered trial, predominantly to ensure a sufficiently powered trial after allowing for losses to follow-up.

Trial feasibility outcome analyses

In addition to the summary statistics detailed inStatistical analyses, data pertaining to a range of feasibility issues were summarised, including the time to recruit each block of individuals; the number of completed assessments within the pre-defined assessment window; a detailed breakdown on attendance at each BRiMS face-to-face session; and the recorded web-based log-ins and diary completions.

Analysis of patient-reported and clinician-rated outcome measures

Summary statistics were calculated for each of the patient-reported and clinician-rated outcome measures at each time point, including CIs for the SDs. Between-group differences at 15 weeks (±1 week) and 27 weeks (±1 week) post randomisation were calculated, together with 95% CIs (nop-values are presented). The correlation between baseline and follow-up scores was calculated across all participants with available data, with corresponding CIs, for each of the candidate primary outcome measures for use in future sample size calculations.

Visual displays, such as box plots and scatterplots, with point and interval estimates, were used to identify any baseline characteristics that have a strong association with each or all of the candidate primary outcomes.

Analysis of EuroQol-5 Dimensions, five-level version, data

The advice from NICE is to use the EQ-5D-3L rather than the EQ-5D-5L,119so the EQ-5D-5L was mapped to

the EQ-5D-3L using the‘crosswalk’technique.120Therefore, we quote the EQ-5D-3L in the results tables

throughout.

Analysis of activity monitor (activPAL) data

Data cleaning

Initial data cleaning was undertaken using visual inspection of each activity summary sheet to remove any incomplete days of data at the start and end of the recording period (e.g. data from the assessment day or the day of the removal of the activPAL).

Decisions relating to the classification of incomplete days at the end of the recording were informed by reference to the typical daily activity patterns recorded by the individual participant (through visual inspection of the summary sheets), and lack of event recordings in the following 24-hour periods

(indicating prolonged non-use). Any uncertainties were addressed by checking appointment dates for the individual participant to inform the scheduled removal date.

Data analysis

All complete days of activity data were included. Initial analysis reported the time spent in the three activity classifications (sitting/lying, standing/incidental stepping, and purposeful stepping), plus step count and sit-to-stand transitions per day averaged over the number of full days of collected data.

Analysis of falls diary data

Participants returned data reporting falls and related injuries every 2 weeks for the duration of the study. As the diary return rate and completeness of diary returns were below what we expected, these data were analysed and presented in three ways.

Analysis 1 (reported)

Falls/injurious falls rate was calculated using the actual number of days of data available as the denominator (i.e. valid days only):

Number of falls or injurious falls

Actual number of days of data available× 365 (1)

Analysis 2 (expected)

Falls/injurious falls rate was calculated according to the number of days available had all those who submitted any diary entries (n= 48) done so fully (i.e. returned 100% of their expected diaries):

Number of falls or injurious falls

Number of days, assuming a 100% diary return rate for all participants who submitted diaries× 365 (2)

Analysis 3 (randomised)

Intention-to-treat analysis: falls/injurious falls rate was calculated according to the number of days available had all those randomised (n= 56) done so fully (i.e. returned 100% of their expected diaries):

Number of falls or injurious falls

Number of days, assuming a 100% return rate for all randomised participants× 365 (3)