Abstract
Background
Limited evidence exists to support very early intensive aphasia rehabilitation after stroke. VERSE is a PROBE trial designed to determine whether two types of intensive aphasia therapy, beginning within 14 days of acute stroke, provide greater therapeutic and cost-effectiveness than usual care.
Objective
To publish the detailed statistical analysis plan for the VERSE trial prior to unblinding. This statistical analysis plan was based on the published and registered VERSE trial protocol and was developed by the blinded steering committee and management team, led by the trial statistician. This plan was developed using outcome measures and trial data collection forms.
Results
The VERSE statistical analysis plan is consistent with reporting standards for clinical trials and provides for clear and open reporting.
Conclusions
Publication of a statistical analysis plan serves to reduce potential trial reporting bias and outlines transparent pre-specified analyses. Australian New Zealand Clinical Trials Registry (ANZCTR) Registration number: ACTRN12613000776707; Universal Trial Number (UTN) is U1111-1145-4130.
Study design
Overview
VERSE 1 is a prospective, parallel group, assessor blinded, randomized, multicenter, international clinical trial. Ethics and local governance approval was obtained from all sites participating. Version 3 of the protocol is current at the time of publication. Written, informed consent is obtained from each participant or their legal representative using full hospital-approved consent processes supported by aphasia friendly consent documents. All participants in VERSE receive usual ward-based care. Participants are randomized in a ratio of 1:1:1 to one of three groups (i) usual ward care alone (UC); (ii) usual ward care plus high-intensity aphasia therapy (UC-Plus); or (iii) usual ward care plus high-intensity prescribed and structured aphasia therapy (VERSE). The distinguishing features of each of these arms are described in the section Intervention.
Aims and hypotheses
The primary aim of the VERSE trial is to determine whether very early daily aphasia therapy (UC-Plus and VERSE groups) is superior to UC at 12 weeks post stroke. The secondary aim is to test whether a standardized intense aphasia therapy regimen (VERSE) is more effective and cost effective than the non-standardized intense aphasia therapy (UC-Plus). Participants are recruited within the first 14 days following stroke. For the purposes of this trial, aphasia therapy refers to direct aphasia treatment designed to change the communication performance of the person with aphasia. All analyses herewith include only direct aphasia therapy which does not include patient/family counselling and education, case management or documentation.
Primary hypothesis
Very early aphasia therapy will result in a greater improvement in communication at 12 weeks post stroke, as measured by the Western Aphasia Battery–Revised; Aphasia Quotient (WAB-R (AQ)). 2
Secondary hypotheses
VERSE aphasia therapy will result in a greater improvement in communication than UC-Plus aphasia therapy at 12 and 26 weeks post stroke, as measured by the WAB-R (AQ).
2
VERSE aphasia therapy and UC-Plus will each result in greater improvements in communication than UC alone, as measured by the WAB-R (AQ),
2
at 12 and 26 weeks post stroke. VERSE aphasia therapy will result in greater improvements in connected speech than UC-Plus, as measured by discourse analysis, at 12 and 26 weeks post stroke. VERSE aphasia therapy and UC-Plus will each result in greater improvements in connected speech than UC alone, as measured by discourse analysis, at 12 and 26 weeks post stroke. Very early aphasia therapy (UC-Plus and VERSE) will result in better quality of life at 12 and 26 weeks post stroke than UC control. Very early aphasia therapy (UC-Plus and VERSE) will be more cost-effective than UC at 12 and 26 weeks post stroke.
Patient population
Participants include patients with acute stroke and aphasia screened using the shortened Frenchay Aphasia Screening Test. 3 Participants are identified, recruited, randomized, and assessed within 14 days of stroke. Therapy commences the day after this assessment.
Inclusion criteria
Patients are eligible if the following inclusion criteria are met:
The patient has acute aphasia of any type resulting from an acute stroke (ICD 10 codes 161–164, i.e., no TIA, SAH, or SDH); The patient has a score of less than 93.7 of the Aphasia Quotient of the Western Aphasia Battery–Revised
2
; The patient is deemed medically stable at recruitment; The patient can maintain a wakeful alert state for 30 consecutive minutes within 14 days of stroke onset; The patient has normal or corrected hearing and vision; The patient is at least 18 years old; there is no upper age limit for inclusion; Informed consent is obtained from the patient or person responsible.
Exclusion criteria
Patients are excluded from the study if they have any of the following:
Pre-existing aphasia before the stroke event; Have suffered a head injury or have had or require neurosurgery (clot retrieval surgery is not an exclusion if patient is stable post operatively and meets the above criteria); Pre-existing clinical diagnosis of dementia; Diagnosis or treatment of major depression at the time of enrolment; Concurrent progressive neurological disorders; Unable to participate in English-based therapy due to English being a second language; Is participating in another intervention trial, which in the opinion of the VERSE trial manager, may interfere with the outcome of the VERSE trial.
Randomization
Participants are randomized to one of three groups (UC, UC-Plus, and VERSE) via a central allocation system in a 1:1:1 ratio. Randomization is stratified by aphasia severity on the WAB-R (AQ) 2 (severe: WAB-R (AQ) 2 0–31.2; moderate: WAB-R (AQ) 2 31.3–62.5; mild: WAB-R (AQ) 2 62.6–93.7). A remote, web-based, computer-generated randomization procedure is used. All online data submissions are secured by the use of password site entry and data encryption procedures. Once the site staff submits patient recruitment data, the randomization allocation is immediately provided back to the investigator.
Intervention
Participants randomized to one of three arms of therapy receive direct aphasia therapy with a qualified speech pathologist for a maximum period of 25 working days. The three arms are:
UC: Usual ward-based aphasia care provided as per healthcare site standards. UC-Plus: Usual ward-based care for aphasia is provided but using a defined therapy regimen of daily sessions of 45–60 min; each participant receives 20 sessions (15–20 h of direct aphasia therapy) in addition to usual care. VERSE intervention: The intensity of this arm of therapy will match the usual care-Plus arm, (20 daily sessions of 45–60 min duration; 15–20 h of direct aphasia therapy) but the speech and language training is prescribed and standardized according to a separate VERSE intervention protocol, in addition to usual care.
The intervention period for the trial concludes at a maximum of 50 days post stroke or when 20 sessions of aphasia therapy are completed (UC-Plus and VERSE) – whichever comes first. Participants are followed up at 12 and 26 weeks post stroke.
Details of the intervention are provided in the VERSE Trial protocol. 1
Baseline and follow-up assessments
All stroke patients with aphasia, who are at least 18 years old, are screened for inclusion in the study. Those screened and deemed ineligible for the study are entered onto a screening log which outlines the reason for non-inclusion. Details of the screening and recruitment procedures, as well as the assessment schedule, are included in the main trial protocol.
The assessment timeline is outlined in Figure 1. Baseline assessment occurs no earlier than two days from the onset of stroke and no later than 14 days from the onset of stroke. Participants undergo the (AQ) component of the WAB-R (AQ).
2
Participants are also assessed for their medical and stroke history. Baseline data collection includes the collection and documentation of:
Demographic details including age, gender, past medical history, languages spoken, education level, handedness, employment, Aboriginal or Torres Strait islander status, and living arrangements. Stroke type and severity (type, hemisphere, first or recurrent stroke, National Institute of Health Stroke Scale, (NIHSS
4
), modified Rankin Scale (mRS
5
), Oxfordshire Community Stroke Project Classification (OCSP
6
), Royal Brisbane Hospital Outcome Measure for Swallowing (RBHOM
7
), and Australian Therapy outcome measures (AusTOMS
8
). Additional baseline assessments including the Boston Naming Test (BNT
9
), the “clock drawing” cognitive sub test, and discourse collection. Time of assessment and aphasia therapy commencement. Trial timeline.
Therapy commences the day after the baseline assessment and is completed within a maximum of 25 working days and must be completed by day 50 post stroke.
The participants are followed up at their place of residence or in the hospital clinic by a blinded assessor, 12 and 26 weeks after the date of the stroke. Patient assessments at these follow-up visits include:
Sample size considerations
The study is powered to detect a difference of 4.4% in communication improvement, measured as percent of maximal improvement (%MPR) on the WAB-R (AQ). 2 This is based on the smallest difference between VERSE and UC-Plus estimated from our pilot studies. We have consulted international aphasia experts who have reported a previous consensus that a five-point difference on the WAB-R (AQ) 2 is clinically meaningful improvement in communication. It is important to note this relationship is not linear and will vary based on aphasia severity. We believe the 4.4% change in communication status, which is marginally more conservative than the five-point difference reported in consensus is reasonable, given this is the difference between the two intensive arms of this trial. The required sample size was estimated under the assumption that the effectiveness hypotheses would be analyzed using a longitudinal regression model (generalized estimating equations or linear mixed model), with outcomes measured at three times (at baseline, 12 and 26 weeks).
A sample of 246 participants (82 per arm) will provide 80% power to detect this difference at a (two-sided) significance level of α = 0.05, after adjusting for a 20% loss to the analysis (lost to follow-up due to death, dropout and non-adherence).
Our pilot data show similar effect sizes for differences between the VERSE and UC-Plus interventions for discourse measures and communication outcomes. Therefore, we believe that the sample size will provide sufficient power to assess differences in discourse outcomes. Pilot data on the quality of life for this population were not available at the trial inception; however, we expect to observe larger effects on quality of life than on communication or discourse and therefore expect to have sufficient power to assess difference on the quality of life.
Other data collection
During the intervention period, all usual care speech therapy is documented in medical notes. The detail of all additional trial-related aphasia therapy is documented separately on the electronic Case Report Form (eCRF) as part of the clinical trial. A daily speech therapy log is completed for each trial participant which includes content, duration of session, and frequency of sessions. Where usual care sessions with participants have occurred prior to the participant’s enrolment in the trial, these sessions are retrospectively collected from the participant’s medical record to ensure all therapy provided to the participant enrolment is captured. Deviation from the prescribed therapy intensity protocol (UC-Plus and VERSE) is documented. Reasons for the withdrawal of therapy such as health-related complications, patient withdrawal, and death are recorded. At the end of the treatment period, a participant diary is provided to the patient to capture resource utilization information.
Participants are assessed and questioned about general health to determine if there have been any adverse events (AEs) or serious adverse events (SAEs) during the study period.
The assessment visits (Baseline, Week 12 and Week 26) are completed as a single session. If the participant is unable to complete assessment tasks in a single session, they are completed on the same day if possible (i.e., over a morning and afternoon session). If this is not possible, the assessments take place on consecutive days and a protocol deviation is recorded.
Information relating to resource utilization is obtained by the blinded assessor from the participant and the participant diary and includes general practitioner (GP) visits, hospital and rehabilitation admissions, outpatient and community service use, medication use for anxiety and depression, speech therapy, use of speech aids and devices, employment status, and respite care and informal care as a result of stroke. Where possible, resource utilization data are verified in medical or other hospital records.
Blinding
VERSE is a complex intervention, and therefore, it is difficult to blind the participants to receipt of intense treatment or usual care. Similarly, it is not possible to blind the clinician providing therapy to the participant’s treatment allocation. Every effort is made to ensure only minimal staff are aware of randomization allocation, and all outcome assessments are conducted by assessors who are blinded to allocation. In addition, chief investigators and trial statistician are blinded to treatment allocation.
Unblinding
Only the Data and Safety Monitoring Committee (DSMC) have access to progressive data. The DSMC Chair is Professor Richard Lindley, University of Sydney, NSW, Australia. The DSMC review unblinded data in accordance with the DSMC Charter (Version 1.0, 25 February 2015).
Definition of the outcomes
Primary outcome
The primary outcome is improvement in communication at 12 weeks after stroke. Communication outcome is measured by the WAB-R (AQ). 2 Improvement will be assessed using the percentage of maximal potential recovery achieved (%MPR). 19
A formal definition of %MPR 19 is provided in the section Primary outcome measure.
Secondary outcome measures
The secondary outcome measures include effectiveness, quality of life, safety, and cost effectiveness measures. These are:
Communication improvement, assessed as %MPR
19
at 26 weeks; Improvement in connected speech at 12 and 26 weeks on discourse measures;10–16 Quality of life measured by the SAQOL
17
scale; Depression measured at 12 and 26 weeks by the ADRS18; AEs and SAEs as defined in the trial protocol; Cost effectiveness of VERSE and UC-Plus therapies compared to UC. The economic analysis will be outlined separately in an economic analysis plan.
Statistical plan
Analysis principles and general considerations
All outcomes and analyses are prospectively characterized as primary or secondary. Differences in all endpoints between the three arms of the trial (VERSE, UC-Plus, and UC) will be tested independently at the two-tailed 5% significance level. All estimates of treatment effects will be presented with 95% confidence intervals. No formal adjustments will be undertaken to constrain the Type I error associated with planned secondary or exploratory analyses. The information provided by analyses is designed to supplement the evidence from the primary analyses; it will provide a more complete characterization of the treatment effects. The analyses for all outcome measures (effectiveness, quality of life and safety) will be conducted on an intention the treat (ITT) basis, i.e., all patients will be analyzed as members of the group to which they were randomized, irrespective of whether they received the allocated treatment or not. The ITT strategy for VERSE is based on the following principles: ○ All available outcome data are collected on all randomized patients. ○ All participants are analyzed in the groups to which they are randomized. ○ All available outcome data will be used in the primary analyses. The primary analyses will be reported without missing data imputed. ○ A separate analysis with missing data imputed will also be reported. Missing data will be assumed to be missing at random (see Treatment of missing values section) ○ A sensitivity analysis including all randomized individuals will be conducted. The sensitivity analysis will consider alternative assumptions about data missing not at random (MNAR) (see Sensitivity analyses section). A per-protocol analysis will be conducted separately, but will not be included in the primary results for this trial. The per-protocol analysis will be described in greater detail when it is published. However, it will be based on the following principles: ○ The per-protocol cohorts will be based on whether or not participants received the planned minimum 15 h of direct aphasia therapy as the intervention dose. ○ The per-protocol cohort will include: ▪ VERSE participants who receive at least 15 h of direct language therapy over the intervention period; ▪ UC-Plus participants who receive at least 15 h of direct language therapy over the intervention period; ▪ All UC participants who are randomized to the UC group. For primary and secondary analyses, the treatment effects for the primary effectiveness outcomes will be adjusted for baseline aphasia severity, measured by the WAB-R (AQ)
2
and baseline stroke severity measured with the NIHSS.
4
Unadjusted analyses will be reported separately from these pre-specified analyses. Subgroup analyses will be carried out irrespective of whether there is a significant treatment effect on the primary outcome. Their purpose is to supplement evidence from the primary analyses to help to fully characterize the treatment effect. Results from subgroup analyses will be interpreted in this context. Analyses will primarily be conducted using the R Statistical Programming Language
20
and supplemented with SPSS
21
to generate formatted tables.
Interim analysis and stopping rules
There are no formal interim analyses planned for this trial. The DSMC periodically reviews data for the primary outcome measure and safety and will advise the chair of the Executive committee if, in their view, the randomized comparisons have provided both (i) ‘proof beyond reasonable doubt’ that very early aphasia rehabilitation (VERSE and UC-Plus groups) is clearly indicated or clearly contra-indicated because of safety concerns and (ii) evidence that might influence future patient management. The DSMC are guided by Haybittle–Peto boundaries in making this determination. That is, they work on the principle that a difference of at least 3 standard errors in the analysis of effectiveness or SAEs (e.g.,– death from all causes, aspiration pneumonia within the first 50 days post stroke) may be needed to justify halting, or modifying the study before the planned recruitment is completed.
Trial profile
The trial will be reported in accordance with the CONSORT statement for non-pharmaceutical trials 19 and the VERSE therapeutic protocol conforms to the SPIRIT statement.22–24 The report will include the number of screened patients who met the inclusion criteria, the number included, and the major reasons for exclusion of eligible patients. At follow-up, the number of patients withdrawn, lost to follow-up, and the number who died within that period will be reported.
Patient characteristic and baseline comparisons
Baseline participant and stroke characteristics will be presented for each of the three groups: UC, UC-Plus, and VERSE (Table 1). These will include: age, gender, geographic region (Australia/NZ), stroke risk factors, stroke type, stroke severity (NIHSS), 4 mRS, 5 time to randomization, and baseline assessment. Baseline communication, cognition, and swallowing characteristics will be presented in Table 2 and will include Aphasia severity, as measured by the WAB-R (AQ), 2 clock drawing test, discourse scores (words, correct information units),11–16 Australian Therapy Outcome Measures (AuSTOMS), 8 and Royal Brisbane Hospital Outcome Measure for Swallowing (RHBOMS). 7
Discrete variables will be summarized as frequencies and percentages. Unless otherwise indicated in the tables, percentages will be calculated according to the number of patients for whom data are available. If there are more than 5% missing values, the denominator will be indicated in the corresponding summary table. Continuous variables will be summarized by the mean and standard deviation (SD) or by the median and interquartile range (IQR). Durations and time intervals will be summarized by medians and IQRs.
Primary outcome: Communication recovery at 12 weeks by intensity
Primary outcome measure
The primary outcome is improvement in communication at 12 weeks after stroke. Communication outcome will be measured by the WAB-R (AQ).
2
The WAB-R (AQ) is a comprehensive measure incorporating production of spoken language and auditory comprehension. Improvement will be assessed using the percentage of maximal potential recovery achieved (%MPR),
19
which is calculated as the ratio of the difference in WAB-R (AQ)
2
at 12 weeks and WAB-R (AQ)
2
at baseline, divided by the difference of maximal attainable score on WAB-R (AQ)
2
and the WAB-R (AQ)
2
at baseline. That is
Statistical hypothesis
The primary research hypothesis is that:
Compared to UC, providing very early daily aphasia therapy (UC-Plus or VERSE) will result in greater improvements in communication ability at 12 weeks post stroke. This will be tested by refuting the null hypothesis that there is no difference in communication recovery, between the usual care group (UC) and the groups receiving very early intensive therapy (UC-Plus or VERSE) at 12 weeks post stroke. For purpose of testing this intensity hypothesis, the UC-Plus and VERSE groups will be combined into a single high-intensity group.
Treatment of missing values
The primary analyses will be presented both with and without imputation of missing data. The imputation will treat missing values as missing at random (MAR), i.e., it will be assumed that the missing data are related to observed data, but not related to the value of the missing data itself. That is, it is assumed that the values of the missing data may reasonably be predicted from all observed data. In particular, it will be assumed that missing values of the primary outcome measure (WAB-R (AQ) 2 at three months) may be estimated from variables on which data have been collected (e.g., baseline aphasia severity, baseline stroke severity, age, gender, lesion size, and location), and on the observed values of WAB-R (AQ). 2 Multiple imputations will be conducted using chained equations. 22 A separate model will be developed for each imputation. The pooled result of these imputed models will be reported and compared with the primary model (without imputed data).
Based on monitoring by the DSMC, 13% of the data for the primary outcome measure in the VERSE trial is missing at the time of publication of the statistical analysis plan. This is substantially less than the previous aphasia trials and less than the 20% missing data that had been planned for in the planning of this trial. 1 The assumption that the data are missing at random was discussed by the VERSE Executive Committee as well as the DSMC and considered to be plausible based on the information available at the time. Sensitivity analyses that consider various other plausible assumptions about missing data will be presented (see Sensitivity analyses section).
Analysis method
The primary aim of this trial is to test the effect of intensity of therapy in early stroke recovery. This is achieved through the primary effectiveness hypothesis which will be analyzed using a linear mixed effects regression model with %MPR 16 as the outcome measure. The UC-Plus and VERSE groups will be combined into a single high-intensity group. This combined intervention group will be compared to the UC group on the primary outcome measure (%MPR 16 at 12 weeks). The model will adjust for differences in baseline aphasia severity and baseline stroke severity by including the baseline WAB-R(AQ) 2 score and the baseline NIHSS 4 score as covariates in the model. The effect of hospital site will be controlled for by including hospital site as a random effect. The treatment effect will be reported as difference in %MPR with the corresponding 95% confidence interval.
Since publication of the trial protocol 1 five new sites have been added to address low participant recruitment. As a result, the Executive Committee decided there was a need to include hospital site as a random effect. This required the adaptation of our original statistical plan from using General Estimating Equations model (GEE) to a Linear Mixed Effect Regression models.
Subgroup analyses
The linear mixed effects regression model will be modified to analyze the difference in the primary effectiveness outcome (%MPR
16
at 12 weeks) between
the VERSE group and UC group and the UC-Plus group and the UC group
As with the primary analysis, the model will adjust for differences in baseline aphasia severity and baseline stroke by including the baseline WAB-R(AQ) 2 score and the baseline NIHSS 4 score as covariates in the model. The effect of hospital site will be controlled for by including recruiting site as a random effect. The main between-group difference at each time point will be assessed through the interaction effect of group and time point. The treatment effect for each subgroup analysis will be reported as difference in %MPR 16 with the corresponding 95% confidence interval.
No corrections for multiple testing in subgroup analyses will be undertaken for these planned analyses. The unadjusted p values will be reported together with the number of subgroup analyses undertaken.
Secondary effectiveness hypotheses
The following secondary effectiveness hypotheses will be assessed:
VERSE (standardized, daily very early) aphasia therapy will result in greater improvements in communication ability than UC-Plus (non-standardized daily very early), as measured by the WAB-R (AQ),
2
at 12 and 26 weeks post stroke. VERSE (standardized, daily very early) aphasia therapy and UC-Plus (non-standardized daily very early) will each result in greater improvements in communication ability than the UC alone, as measured by the WAB-R (AQ),
2
at 12 and 26 weeks post stroke. VERSE (standardized, daily very early) aphasia therapy will result in greater improvements in connected speech than UC-Plus (non-standardized daily very early) as measured on discourse analysis10,11 (number of words, number of Content information units; % Content information units), at 12 and 26 weeks post stroke. VERSE (standardized, daily very early) aphasia therapy and UC-Plus (non-standardized daily very early) will each result in greater improvements in connected speech than UC alone, as measured by discourse analysis10–16 (number of words, number of Content information units; % Content information units), at 12 and 26 weeks post stroke.
Secondary statistical hypotheses
The specified set of secondary effectiveness hypotheses involve the assessment of the following statistical hypotheses:
VERSE (standardized, daily very early) aphasia therapy will result in greater improvements in communication ability than UC-Plus (non-standardized daily very early), as measured by the WAB-R (AQ),
2
at 12 weeks post stroke. VERSE (standardized, daily very early) aphasia therapy will result in greater improvements in ability than UC-Plus (non-standardized daily very early), as measured by the WAB-R (AQ),
2
26 weeks post stroke. VERSE (standardized, daily very early) aphasia therapy will result in greater improvements in communication ability than UC alone, as measured by the WAB-R (AQ),
2
12 weeks post stroke. VERSE (standardized, daily very early) aphasia therapy will result in greater improvements in communication ability than UC alone, as measured by the WAB-R (AQ),
2
26 weeks post stroke. UC-Plus (non-standardized daily very early) will result in greater improvements in communication ability than UC alone, as measured by the WAB-R (AQ),
2
12 weeks post stroke. UC-Plus (non-standardized daily very early) will result in greater improvements in communication ability than UC alone, as measured by the WAB-R (AQ),
2
26 weeks post stroke. VERSE (standardized, daily very early) aphasia therapy will result in greater improvements in connected speech than UC-Plus (non-standardized daily very early), as measured by discourse analysis,10–16 12 weeks post stroke. VERSE (standardized, daily very early) aphasia therapy will result in greater improvements in connected speech than UC-Plus (non-standardized daily very early), as measured by discourse analysis,10–16 26 weeks post stroke. VERSE (standardized, daily very early) aphasia therapy will result in greater improvements in connected speech than UC alone, as measured by discourse analysis,10–16 12 weeks post stroke. VERSE (standardized, daily very early) aphasia therapy will result in greater improvements in connected speech than UC alone, as measured by discourse analysis,10–16 26 weeks post stroke. UC-Plus (non-standardized daily very early), will result in greater improvements in connected speech than UC alone, as measured by discourse analysis,10–16 12 weeks post stroke. UC-Plus (non-standardized daily very early), will result in greater improvements in connected speech than UC alone, as measured by discourse analysis,10–16 26 weeks post stroke.
Analysis methods
The outcome measure for all effectiveness analyses that involve WAB-R (AQ)
2
will be %MPR.
16
For outcomes assessed at 12 weeks post stroke, this will be %MPR
16
at 12 weeks post stroke, as defined in the section Primary outcome measure. For outcomes assessed at 26 weeks post stroke, this will be %MPR
16
at 26 weeks as measured at baseline. That is
The outcome measure for all efficacy analyses involving discourse analysis10,11 will be %Correct Information Units (CIUs).
Comparison between VERSE and UC-Plus groups
Communication outcomes (WAB-R (AQ)) 2
A longitudinal linear mixed model will be used to assess difference between the VERSE and UC-Plus groups on %MPR 16 at 12 and 26 weeks. The main between-group difference at each time point will be assessed through the interaction effect of group and time point. The model will adjust for baseline aphasia severity by including the WAB-R (AQ) 2 score at baseline as a covariate. Baseline stroke severity will be controlled for by including the NIHSS 4 at baseline as a fixed factor in the model. Recruitment site will be included as a random effect.
Connected speech outcomes (discourse analysis)
A similar longitudinal linear mixed model will be used to assess difference between the VERSE and UC-Plus groups on %CIUs10–16 at 12 and 26 weeks. The main between-group difference at each time point will be assessed through the interaction effect of group and time point. The model will adjust for baseline aphasia severity by including the WAB-R (AQ) 2 score at baseline as a covariate. Baseline stroke severity will be controlled for by including the NIHSS 4 at baseline as a fixed factor in the model. Recruitment site will be included as a random effect.
Comparison with UC group at 12 and 26 weeks
Communication outcomes (WAB-R (AQ)) 2
A longitudinal linear mixed model will be used to compare each of the VERSE and UC-Plus groups to the UC group on %MPR 19 at 12 and 26 weeks. The main between-group difference will be assessed through the interaction effect of group and time point. The model will adjust for baseline aphasia severity by including the WAB-R (AQ) 2 score at baseline as a covariate. Baseline stroke severity will be controlled for by including the NIHSS 4 at baseline as a fixed factor in the model. Recruitment site will be included as a random effect.
Connected speech outcomes (discourse analysis)
A similar longitudinal linear mixed model will be used to compare each of the VERSE and UC-Plus groups to the UC group on %CIUs10–16 at 12 and 26 weeks. The main between-group difference at each time point will be assessed through the interaction effect of group and time point. The model will adjust for baseline aphasia severity by including the WAB-R (AQ) 2 score at baseline as a covariate. Baseline stroke severity will be controlled for by including the NIHSS 4 at baseline as a fixed factor in the model. Recruitment site will be included as a random effect.
Handling of missing data
Missing data for secondary analyses will be handled as per the section Treatment of missing values.
Other secondary outcomes: quality of life
It is hypothesized that:
Very early aphasia therapy (UC-Plus and VERSE) will result in better quality of life at 12 and 26 weeks post stroke than UC control.
Outcome measure for the quality of life
Quality of life will be assessed using the SAQOL scale. 17
Statistical Hypotheses for the quality of life
The specified quality of life hypotheses involve the assessment of the following statistical hypotheses:
Very early daily aphasia therapy (UC-Plus or VERSE) will result in higher quality of life than UC alone at 12 weeks post stroke. VERSE aphasia therapy will result in higher quality of life than UC alone at 12 weeks post stroke. UC-Plus will result in higher quality of life than UC alone at 12 weeks post stroke. VERSE aphasia therapy will result in higher quality of life that UC-Plus at 12 weeks post stroke. Very early intensive aphasia therapy (UC-Plus or VERSE) will result in greater improvements in communication ability at 26 weeks post stroke. VERSE aphasia therapy will result in higher quality of life than UC alone at 26 weeks post stroke. UC-Plus will result in higher quality of life than UC alone at 26 weeks post stroke. VERSE aphasia therapy will result in higher quality of life that UC-Plus at 26 weeks post stroke.
Analysis method
Separate linear mixed effects regression model will be used to assess quality of life as measured by the SAQOL
17
at 12 and 26 weeks post stroke. Each model will compare the effects of:
VERSE therapy to UC alone UC-Plus therapy to UC alone
The model will adjust for baseline aphasia severity by including the WAB-R (AQ) 2 score at baseline as a covariate. Baseline stroke severity will be controlled for by including the NIHSS 4 at baseline as a fixed factor in the model. Recruitment site will be included as a random effect. The treatment effects will be reported as difference in SAQOL 17 (between VERSE and UC and between UC-Plus and UC); the corresponding 95% confidence intervals will be reported.
Handling of missing data
Handling of missing data will be similar to the detail in the section Treatment of missing values.
Other secondary outcomes: Cost effectiveness
It is hypothesised that:
Very early intensive aphasia
Analysis method
Details of the cost effectiveness analyses will be published separately and are not included here. Results from the effectiveness analyses will be the same for describing the effectiveness component of the incremental cost effectiveness calculation.
Secondary outcomes: Safety IMEs, AEs, and SAEs
IMEs, AEs, and SAEs are defined in the main trial protocol. These are expected to be rare occurrences. Aphasia therapy is not expected to have an effect on these events. Therefore, no formal hypotheses have been stated about these outcomes. All IMEs, AEs, and SAEs will be reported by therapy group.
As rare events, counts of IMEs, AEs, and SAEs are expected to have a Poisson or negative binomial distribution. The distributions of these events will be modelled and appropriate generalized linear mixed models will be developed to assess differences between groups if the data suggest that there is a between-group difference of greater than 3 standard deviations in any of these event types.
Sensitivity analyses
Sensitivity analysis for the primary outcome will be conducted under various assumptions about the missing data. The main analysis is planned under an assumption of missing at random; therefore, the sensitivity of the results to plausible departures from MAR will be explored as a part of an intention-to-treat analysis strategy.25–27
The 2010 National Research Council Panel on the Handling of Missing Data in Clinical Trials 27 outlined two frameworks for modelling missing data: selection models and pattern-mixture models. The selection model framework models the probability of nonresponse (data being missing) rather than the distribution of the outcomes. In contrast, the pattern-mixture model framework models the distribution of the outcomes, under the assumption that the distribution of outcomes is different for responders and non-responders. The pattern-mixture model framework admits a transparent and easily interpretable method for conducting a sensitivity analysis which involves adding a parameter (delta) to the mean response. The parameter, delta, measures the degree of departure from missing at random. In the VERSE trial, we propose using a pattern-mixture model to conduct a sensitivity analysis that assesses sensitivity of the results to plausible departures from the MAR assumption. If the inference about the treatment effects can be overturned by plausible values of the delta parameter, then the results of the trial will be considered equivocal.
We note that the gain in transparency and interpretability of conducting sensitivity analyses for pattern-mixture models may be offset by the computational complexity in averaging across large numbers of patterns in models with a large number of repeated measures. We do not anticipate facing this issue in the VERSE trial since outcomes are measured at only two time points after baseline (at primary endpoint and at follow-up). However, if for any reason, we are unable to conduct a sensitivity analysis under the pattern-mixture model framework, the sensitivity analysis will be conducted under an appropriate selection model framework.
Tables and figures for the main paper
The proposed tables and figures for the main results are presented in Appendix A1. Table A1 will report the main baseline demographic and stroke characteristics by participant group. Table A2 will report baseline communication, cognition, and swallowing characteristics. Table A3 will report key summary data about the timing, frequency of therapy and amount of therapy provided in each group, length of hospital stay, and discharge destination. Tables A4 and A5 will report the primary and main secondary outcomes at 12 and 26 weeks post stroke and Table A6 will report deaths and SAEs. Figure A1 will be the CONSORT 22 diagram. Figure A2 will be a bar chart of each grade on the mRS 5 in each treatment group at 12 and 26 weeks post stroke. Figure A3 will be a forest plot of the treatment effect on the primary outcome among different subgroups at 12 and 26 weeks post stroke.
Approval
The final version of this statistical analysis plan was approved by the Executive Committee on 20th December 2017.
Footnotes
Declaration of conflicting interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study is funded by the National Health and Medical Research Council (NHMRC), Australia (Project Grant Number 1044973), The Tavistock Trust for Aphasia, United Kingdom and Edith Cowan University, Australia. The funders of this study have no influence on the study design, data collection, data interpretation, writing of the protocol or statistical analysis plan.