Dopamine Augmented Rehabilitation in Stroke (DARS): a multicentre double-blind, randomised controlled trial of co-careldopa compared with placebo, in addition to routine NHS occupational and physical therapy, delivered early after stroke on functional recovery

Article history

The research reported in this issue of the journal was funded by the EME programme as project number 08/43/61. The contractual start date was in January 2010. The final report began editorial review in November 2015 and was accepted for publication in February 2017. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The EME editors and production house have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the final report document. However, they do not accept liability for damages or losses arising from material published in this report.

Declared competing interests of authors

Gary A Ford received personal fees from Lundbeck Ltd, Boehringer Ingelheim, Pfizer and AstraZeneca, and grants and personal fees from Athersys outside the submitted work. Alastair Cozens had equity in Skene Software Ltd and other financial activity outside the submitted work for SiLCK Clinical Solutions Ltd. David Meads is a member of the Health Technology Assessment (HTA) Elective and Emergency Specialist Care (EESC) Panel. Catherine M Sackley is a member of Health Services and Delivery Research researcher-led board. Amanda J Farrin is a member of the HTA Clinical Evaluation and Trials Board and the HTA Commissioning Strategy Group.

Copyright statement

© Queen’s Printer and Controller of HMSO 2019. This work was produced by Ford et 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.

2019 Queen’s Printer and Controller of HMSO

What is ‘rehabilitation’?

The World Health Organization (WHO) definition of rehabilitation focuses on the enablement of people in order to reduce the impact that a disabling condition may have on a person’s life. 8

The cornerstone of rehabilitation is the WHO’s International Classification of Functioning, Disability and Health9 (ICF). This model postulates that an underlying health condition (such as stroke) may cause impairment in bodily structures and function, limitation of activities and restriction of participation. 10 Clinically important impairments following stroke may include changes in gait pattern and intellectual functions, inability to sequence complex movements, and alterations in muscle tone and power. 11 Activity limitations and participation restrictions may include difficulties in problem-solving, transferring oneself, maintaining family and personal relationships, and acquiring and keeping a job. 11 However, an individual’s functional abilities are also profoundly influenced by wider contextual factors,8 both environmental and personal. Environmental factors comprise not only the built environment but also legislation and societal attitudes. 8 Depending on circumstances, they may serve as barriers to, or facilitators of, individual function. 8 Personal factors are intrinsic to the individual but may not be directly related to the underlying health condition. 8 They include sex, age, race and physical fitness. 8

Therefore, in ICF terms, ‘disability’ is constructed as an interaction between an individual and their environment and personal factors. 10,12 ‘Rehabilitation’ typically comprises a complex package of measures targeting multiple levels of the ICF. The precise nature of the intervention required varies depending on the extent of cerebral damage caused by the stroke and the patient’s goals, and with the passage of time from the stroke. Within the first few hours, optimum acute care minimises the extent of tissue injury and secondary complications (impairment in structure) and, therefore, maximises preservation of function. 13 As planning for discharge progresses, environmental assessment and, if necessary, provision of assistive technologies (contextual factors) may help to enhance safety and personal independence on leaving hospital. In the long term, a combination of rehabilitation interventions at impairment, activities and environmental levels may be necessary to address specific goals such as return to work (participation). 14 Many stroke survivors require not only intermittent discrete periods of time-limited rehabilitation to address particular functional goals, but also longer-term monitoring and support to prevent long-term complications and to mitigate the effects of changing disability (such as deterioration in mobility due to accelerated joint ageing). 13

Rehabilitation interventions may be restorative (e.g. encouraging use of a hemiplegic limb) or compensatory (such as the teaching of new techniques or provision of orthoses or assistive technologies). 14 Since the 1950s, a variety of rehabilitation approaches have been developed, each based on differing theories about how patients recover from stroke. 4 National Institute for Health and Care Excellence (NICE) quality standards15 state that a minimum of 45 minutes of each active therapy should be offered to stroke patients for a minimum of 5 days per week. The dose of active therapy is such to allow patients to meet their rehabilitation goals, and active therapy should continue as long as the patient benefits and can tolerate the therapy. More recently, the provision by the therapist of active feedback on the quality of movements has been emphasised. 16 A recent Cochrane review17 has demonstrated no clear evidence in favour of any one approach. Although many rehabilitation strategies are in clinical use,18 most restorative interventions share one crucial commonality: their reliance on the patient’s ability to learn (or to relearn) motor skills.

Brain structures involved in motor learning

Motor learning is fundamental not only to rehabilitation but also to daily life,19 from a baby learning to walk to a musician practising a symphony. Becoming skilled in a motor task requires us not only to learn the correct order of movements but also to develop an awareness of the sensory input that guides decisions, such as timing of the movement, trajectory and what force should be applied. 20 Often, acquiring a skill also requires the learner to manipulate or interact with objects in his or her environment. 21 Although there is no universally accepted definition of ‘motor learning’, it has been conceptualised as ‘a change in motor behaviour, specifically referring to the increased use of novel, task-specific joint sequences and combinations, resulting from practice and/or repetition’. 22

Many studies have sought to define the brain structures involved in motor learning. Whether or not it is reasonable to extrapolate theoretical frameworks for motor learning derived in healthy individuals to those with stroke is debatable. However, stroke patients participating in rehabilitation demonstrate increased activation of the ipsilesional primary motor cortex, pre-motor cortex and supplementary motor areas. 23 This suggests that functional reconfiguration is possible after stroke, allowing lost functions to be recovered using partially spared pathways. 23 The cellular basis for this is axonal remodelling, changes in the number and morphology of dendrites,24–26 and long-term potentiation or depression of synaptic transmission both adjacent to and remote from the original stroke,24,25 a process termed ‘neuroplasticity’.

Magnetic resonance imaging (MRI) has been the mainstay of attempts to understand learning processes, both in healthy participants20,21,27,28 and in those recovering from stroke. 23 Attention has focused on the role of the basal ganglia. These contain the striatum (caudate, putamen and nucleus accumbens) and the globus pallidus. 29,30 They are now known to form part of wider corticobasal and corticocerebellar loop circuits. The striatum receives white matter projections from almost all cortical areas. It sends projections to the thalamus and cerebellum, which, in turn, project to the cortex29–31 (Figure 1). Although the precise function of these pathways remains opaque, it is clear that they are crucial to a variety of cognitive and motor processes. 32–34 Lesions at any point in these pathways may mimic deficits resulting from cortical injury:31 a phenomenon that has been termed ‘disconnection syndrome’. 31 For example, discrete lesions to the basal ganglia may result in a spectrum of clinical manifestations more typically associated with frontal lobe insult due to the extensive projections these structures receive from the prefrontal cortex. 31

FIGURE 1.

Connectional pathway and neurotransmitter systems of the corticobasal ganglia thalamic pathway. GABA, gamma-aminobutyric acid.

The indirect pathway (light-green arrow) projects from the cortex via a complex web of connections between the globus pallidus externa and the subthalamic nucleus, to the substantia nigra and globus pallidus interna and, thence, to the thalamus. The direct pathway (light-blue arrow) projects from the cortex to the striatum, and thence to the thalamus. The hyperdirect pathway (grey arrow) bypasses the striatum and projects directly to the substantia nigra and globus pallidus interna. Dopaminergic projections from the substantia nigra (black arrows) exert an inhibitory effect on GABAergic interneurons of the indirect pathway, and an excitatory effect on GABAergic interneurons of the direct pathway.

The role of dopamine in learning processes

There has recently been a growing appreciation of the central role that dopamine plays in learning. Dopamine is a key modulator of striatal function; through its selective action on striatal projection interneurones, dopamine is able to inhibit the indirect striato-thalamic pathway and excite the direct pathway. 35 By directly influencing the activity of these pathways, dopamine may contribute to the selection and termination of motor programmes for skilled movements. 35,36 In conditioned learning, dopamine may potentiate drive and arousal, encode the ‘value’ of a reward, or ‘stamp in’ associations between stimulus and response. 37 More recently, it has been proposed that phasic dopamine release dopamine acts as an ‘alerting signal’, prompting the orientation of conscious attention and cognitive processing towards salient environmental cues and increasing general arousal and motivation. 38 Together with other neurotransmitters (noradrenaline, serotonin and acetylcholine), dopamine up-regulates excitatory glutamatergic transmission, which in turn stimulates activation of N-methyl-D-aspartate receptors. 39 It is the activity of these receptors that promotes long-term potentiation or long-term depression of synaptic efficacy. 39 These long-term synaptic changes are thought to mediate consolidation of memory traces (e.g. of the association between stimulus and reward,29 or motor memories40).

‘Pharmacorehabilitation’: a new frontier in stroke rehabilitation?

Taken together, these disparate findings present an intriguing possibility: might pharmacological manipulation of neurotransmitter systems be used to enhance the reacquisition of motor skills after stroke? Several drugs influence directly the levels of key neurotransmitters. Dextroamphetamine and metamphetamine inhibit reuptake of dopamine, noradrenaline and serotonin into the presynaptic terminal, thereby increasing their bioavailability within the synaptic cleft. 39 Methylphenidate, similarly, blocks presynaptic reuptake of dopamine and noradrenaline. 39 As all of these neurotransmitters are known to enhance excitatory glutamatergic transmission in the corticostriatal pathways, using drugs that act on these systems may present an opportunity to manipulate neuroplasticity, thereby enhancing the effectiveness of conventional rehabilitation. 39 An early-phase trial41 in rats with a unilateral experimental lesion of the motor cortex reported an immediate improvement the animals’ ability to run along a rotating beam following a single dose of dextroamphetamine. This finding has been replicated in other trials in both rats42–46 and primates47 (albeit with some dissenting findings). 48 The treatment response appears to depend on repetitive task practice during the drug’s period of action. 42–44,46

Encouraged by these findings, there have been several trials of amphetamines to enhance recovery in human stroke survivors. 39,49 Although there is a trend towards improved motor outcomes in those treated with amphetamine,39 many trials are of small scale, with considerable heterogeneity in the dosing regimens, timing between dose and therapy, follow-up period and outcome measures. 39 Several showed differences in the baseline characteristics of the intervention and placebo groups (age, consciousness level), which may account for a trend towards higher mortality rates in those treated with amphetamine. 39 Therefore, positive findings must be interpreted with caution39 and clinical use of amphetamine has been limited as a result of the concern about sympathomimetic effects. 49

Drugs that promote dopaminergic activity directly may be more appropriate targets for pharmacological enhancement of rehabilitation. 50 Levodopa, an orally administered precursor of dopamine, crosses the blood–brain barrier before being metabolised to dopamine centrally resulting in a rise in brain dopamine levels. 39 Around 5% of the levodopa that enters the central nervous system is further metabolised to noradrenaline, raising the possibility that its effects may be mediated at least in part by noradrenaergic transmission. 51 Co-careldopa (Sinemet®, Merck Sharp & Dohme Ltd) is a combined preparation of 100 mg of levodopa with a peripheral DOPA decarboxylase inhibitor, carbidopa. Carbidopa reduces peripheral levodopa metabolism, thereby maximising the central bioavailability of levodopa. 51 Peak plasma levels of levodopa occur between 30 minutes and 2 hours after a single oral dose of co-careldopa, with a plasma half-life of 1–3 hours.

Only one systematic review has examined the use of dopamine agonists to enhance motor recovery from stroke in humans. Berends et al. 39 assessed the evidence for a number of candidate drug classes, including amphetamine, selective serotonin reuptake inhibitors (SSRIs) and levodopa. Two papers52,53 concerning the use of levodopa met their inclusion criteria. The evidence of a positive treatment effect with this agent was judged to be lacking. 39 A number of other trials54–62 that they did not cite also address this question. These are of variable quality and report mixed results. Many were limited by small sample sizes56,58 or comparatively short follow-up,58,60 or administered only single doses of co-careldopa. 58,59 Some recruited patients months or years after stroke. 56,58 Several demonstrated benefit with dopamine on motor outcomes including the Barthel index54 (BI), the National Institutes of Health Stroke Scale,54 the Functional Independence Measure (FIM),55 walking speed and manual dexterity,56 procedural motor learning,57 motor memory,59 the Fugel-Mayer Assessment60 and the Rivermead Motor Assessment. 62 However, others have found no improvement in length of stay,61 cognitive and motor domains of the FIM61 or upper limb function. 58

A systematic search was carried out to identify relevant studies and emerging data as the trial progressed. An overview of identified randomised controlled trials (RCTs) is given in Table 1 and the search strategy is in Appendix 1.

These equivocal results may perhaps be explained by dopamine’s complex and multiple actions in vivo and the heterogeneous groups of patients involved in these trials. Any link between administration of dopamine and recovery of function may be attenuated by injury to pathways involved in motor learning and cognitive function, or by reduced motivation to participate in rehabilitation. However, it is clear that co-careldopa is well tolerated, with no serious adverse events (SAEs) reported with the dosing regimens used. Therefore, there is a need for a larger RCT to address the question of whether or not administering levodopa is beneficial in enhancing recovery from stroke.

The Dopamine Augmented Rehabilitation in Stroke (DARS) trial set out to evaluate the effect of administering co-careldopa for up to 6 weeks on patients’ ability to walk ≥ 10 metres independently at up to 8 weeks after stroke. 64

Aim and objectives

The aim of this trial was to determine if combining co-careldopa with routine PT and OT during early rehabilitation in people with new stroke admitted to a stroke unit enhances the effect of conventional rehabilitation treatments in terms of physical functioning.

Trial design

The DARS trial was a multicentre, prospective, randomised, double-blinded, placebo-controlled trial of NHS PT and OT treatment alone compared with NHS PT and OT treatment in addition to up to 6 weeks of co-careldopa treatment for patients admitted to acute stroke services after new or recurrent stroke. 64 Outcome measures were obtained at 8 weeks, 6 months and 12 months following randomisation.

Participants

Eligibility

Trial settings

A total of 51 NHS stroke services across the UK with an acute inpatient stroke rehabilitation facility and a service allowing rehabilitation treatments within the community setting that could consist of early supported discharge or community stroke teams/services obtained NHS permissions to take part in the DARS trial. A further five stroke services participated as repatriation sites for participants recruited from other services.

Face-to-face follow-up was conducted by a blinded researcher in the participant’s home, at hospital or at a community facility. When it was not possible to complete a face-to-face assessment, telephone follow-up was conducted.

Interventions

Patients were randomised to receive either co-careldopa (Sinemet) or a matched placebo tablet in addition to routine NHS PT and OT. The initial two doses of co-careldopa were 62.5 mg (50 mg of levodopa and 12.5 mg of carbidopa) and the remaining doses were 125 mg (100 mg of levodopa and 25 mg of carbidopa).

Patients were required to take a single oral tablet 45–60 minutes before PT or OT sessions (this also includes programmed rehabilitation delivered by rehabilitation assistants). Rehabilitation treatment appropriate for drug administration within the DARS trial was defined as active physical treatment (i.e. most PT and OT directed at motor skills such as walking, transfers and dressing, but not psychological input sessions or speech and language therapy).

The dose and timing of the medication reflects current evidence on the use of co-careldopa in this context. 52,57,76,77 A pragmatic approach to taking the IMP was chosen; although the IMP should optimally be taken 45–60 minutes prior to the rehabilitation treatment session, it was recognised that, on occasions, the therapist was unable to contact the patient to remind them to take the tablet or the patient forgot and it was acceptable for the tablet to be taken 0–15 minutes before the start of therapy in these situations. The reasons for any deviations from the optimal timing were recorded. If the participant was scheduled to have two therapy sessions either one directly after the other or within 3 hours of a dose then a repeat dose was not given before the second therapy session. If the participant was scheduled to have more than two PT or OT sessions, the dose was not administered more than twice during any one 24-hour period.

Co-careldopa or placebo in addition to NHS PT or OT was given for a maximum of 6 weeks and assuming a maximum of two sessions of therapy per day for 30 days over a 6-week period. The duration of treatment could be < 6 weeks if the patient was clinically deemed to not require further rehabilitation treatment. The decision about need for rehabilitation intervention was made by the treating clinicians, therapists and nurses in consultation with patients and families.

A pack of 62 tablets (two 62.5-mg tablets and 60 125-mg tablets), each with a unique kit number, was dispensed to each participant by the hospital pharmacy on the day of randomisation. The first two co-careldopa or placebo doses were administered in the hospital setting, allowing the participant to be observed in case of early AEs. Blood pressure was checked before administering the first dose and, if the systolic blood pressure was < 90 mmHg, the decision whether or not to administer the IMP was discussed with the local PI. Subsequent doses were administered in the hospital setting by the attending nursing staff or within the home setting post discharge.

Following discharge into the community, telephone reminders were given to the participant approximately 1 hour prior to the therapy visit by the treating community rehabilitation staff to remind the patient to take the medication.

Figure 2 provides an overview of the treatment pathways permitted within the DARS trial.

FIGURE 2.

Treatment pathways. a, Rehabilitation treatment appropriate for drug administration within DARS is defined as an active physical treatment (i.e. most physical and occupational therapy directed at motor skills such as walking, transfers and dressing, but not psychological input sessions or speech and language therapy, swallowing, splinting).

Adherence to medication

Stroke survivors may have significant residual impairments that may affect their ability to comply with trial medication, and ensuring pharmaco-adherence in the community setting is a challenge. To maximise compliance, patients, clinicians, pharmacy staff and the manufacturer were consulted during the design of the IMP packaging to ensure that it was usable by people with hemiparesis and included clear instructions as to when the trial medication should be taken in relation to the NHS PT and OT provision. A DVD was developed for participants and carers to use as a visual aid to complement the other patient information. This content was presented in a manner accessible to patients with aphasia or hemisensory neglect and used graphics to illustrate abstract concepts. A telephone reminder was also implemented, whereby the treating community rehabilitation staff contacted the participant approximately 1 hour prior to their community-based therapy session to prompt them to take the trial medication. 73

Outcomes

Data collection methods

Baseline data were collected by the clinical research team from clinical records and via face-to-face administration of the questionnaires, and participants also self-completed a baseline RMI measure. All baseline data were collected prior to randomisation.

Follow-up data were collected at 8 weeks, 6 months and 12 months via face-to-face administration by an independent blinded researcher in the participant’s home, at the hospital or at a community facility and documented on paper case report forms (CRFs) provided by the Clinical Trials Research Unit (CTRU). Completion of the primary outcome measure (RMI) was via telephone when it was not possible to arrange a face-to-face visit.

The researcher could choose to send the questionnaires to the patient in advance to allow time to prepare for the interview. Carer-completed outcomes were collected at the same time as the patient’s outcomes, when possible, and documented.

Assessment instruments

Table 2 summarises the measures used at each assessment time point and a summary of each instrument used is provided below:

• The RMI assesses the ability to walk independently. It was completed by researchers prior to randomisation to inform the stratification of patients and by participants at baseline and follow-up to see how they assessed their own mobility. It has 15 items (score range of 0–15 points), with higher scores indicating an increased ability to walk independently.

• The ABILHAND questionnaire measures upper limb impairment by asking participants to rate 23 items relating to their manual ability on a three-level scale (1, impossible; 2, difficult; and 3, easy). The raw scores are converted into a linear measure of ability using Rasch analysis. The scores are, thus, expressed as logits on an interval scale ranging from plus to minus with the centre of the scale set to zero; a higher number logit indicates greater patient-perceived ability. The baseline score in the DARS trial is based on participants’ own assessment of their manual ability in the month before their stroke and at the time of assessment at the follow-up.

• Physical and social independence were assessed using the NEADL. This assesses aspects of physical and social independence performance across 22 items (score range of 0–66 points) grouped into four categories [(1) mobility, (2) kitchen, (3) domestic and (4) leisure activities]. A higher score indicates greater independence. At baseline, participants were asked to consider their independence before their stroke and at the follow-up they were asked to consider their indepencence in the previous month.

• The GHQ-12 assesses emotional health. It contains 12 questions (score range of 0–36) addressing issues of decision-making, loss of sleep and confidence, feelings of strain, enjoyment of daily activities, confidence and happiness. A higher score indicates worse emotional health. The scores can also be categorised into no sign of psychological distress (score of ≤ 15 points), evidence of distress (score of > 15 points but ≤ 20 points), or severe problems and psychological distress (score of > 20 points).

• Activities of daily living, disability and mobility were assessed using the BI. Ten items (score range of 0–20 points) evaluate the patient’s functional ability; higher scores indicate a greater degree of functional independence.

• Joint, neck or back pain was measured using the MSK-SSP manikin. Combinations of locations of (1) pain in upper limbs, (2) pain in lower limbs, (3) central post-stroke pain, (4) thumb, hand, finger or wrist joint pain and (5) spinal pain were based on the previous work with MSK-SSP. 74,78

• The FAS contains 10 items (score range of 10–50 points); higher scores indicate more severe fatigue.

• The mRS is used to assess global outcome after stroke and is scored from 0 (no symptoms at all) to 5 (severe disability, bedridden); patients who die are given a score of 6.

• The MoCA is a researcher-administered instrument that assesses cognitive function. It contains 10 items (score range of 0–30 points); a score of < 26 points indicates cognitive impairment.

• The CBS is a 22-item scale (score range of 22–88 points) assessing various aspects of caregiver burden, including general strain, isolation, disappointment, emotional involvement and environment; higher scores indicates higher burden.

Strategies to maximise data collection

Site achievement reports (SARs) were developed to provide sites with clear communication of accomplishments against trial-specific targets to assist sites in understanding their own performance. SAR benchmarks were based on recruitment, screening, baseline and consent data, intervention delivery monitoring and return of data essential for primary end-point reporting. RAG ratings indicated predefined target completion. Site-specific SARs were sent to site researchers (nurses and PIs); monthly and regional summaries were sent to the National Institute for Health Research (NIHR) SRN managers. 80

Several methods were implemented to allow flexibility in the collection of follow-up data to maximise return rate. This included a research nurse visiting the participant’s home to complete the questionnaires or completion at a hospital visit. The research nurse could also choose to send the questionnaires to the participant in advance of the scheduled appointment to allow the patient some time to prepare for the visit.

To maximise completion of the primary end-point data, the research nurse could complete the RMI questionnaire via telephone when it was not possible to arrange a face-to-face visit.

Strategies to maximise recruitment

Recruitment in to a pharmacorehabilitation stroke trial is challenging and the barriers to achieving target recruitment are multifaceted. A responsive approach was implemented to ensure issues in recruitment and implementation were identified in a timely manner so that the trial could adapt processes accordingly. Table 3 provides a summary of the strategies implemented and the perceived benefit.

Randomisation

Blinding

The trial was double blind. Participants, clinicians, research staff and trial personnel at the CTRU involved in the day-to-day running of the trial were blinded to group allocation until the final database lock. Outcomes were collected by assessors masked to the treatment allocation. All analyses until the final analysis were undertaken blinded to treatment allocation. A trial safety team had access to the treatment allocation for the purposes of emergency unblinding and preparation of unblinded reports to the DMEC.

Withdrawal from treatment and data collection

Participants could withdraw from further trial treatment and/or data collection.

Participants did not continue to receive co-careldopa or the placebo after randomisation if they developed contraindications to the co-careldopa treatment, if the treating physician deemed that the patient was at a significant health risk from continued participation in the trial or if the participant decided that they no longer wished to take part in the trial.

If treatment was stopped, participants were still followed up unless they had withdrawn consent to follow-up. The primary reason for discontinuation was recorded.

Participants were made aware (through the information sheet and consent form) that, should they withdraw, safety data would still be collected after their last dose and all data collected prior to the withdrawal date would be used in the final analysis.

Patient and public involvement

We consulted with the Consumer Research Advisory Group (CRAG), which is a local Yorkshire-based group comprising stroke survivors and caregivers who give advice on research projects. Their advice was sought on the design of the IMP packaging and the proposal to invite patients to participate in more than one research project.

Mr Ossie Newell (Joint Chairperson of the Nottingham Stroke Research Consumer Group) reviewed the original research idea from a patient perspective and reviewed the patient information sheets and proposed outcome measures to ensure that they were appropriate for the patient population. Mr Michael Bonner, a member of the Trial Steering Committee (TSC), ensured that the patients’ perspective was considered during the progress of the trial. He also reviewed the patient information sheet and proposed outcome measures.

Statistical methods

The statistical analysis plan was finalised and approved by the Trial Management Group and TSC. Significance was tested at the 5% level for all analyses. Analyses were completed in SAS^® version 9.2 (SAS Institute Inc., Cary, NC, USA). Analyses were pragmatic, based on an ITT sample.

Radiological review

The influence of the stroke lesion itself (as seen via imaging) was explored as part of the secondary analysis. All available CT and MRI scans obtained during the index admission were obtained for participants who had provided informed consent. Participating sites copied the images to CD (compact disc) USB (universal serial bus) memory stick in Digital Imaging and Communication in Medicine (DICOM) format,85 and sent them to the CTRU for centralised review. Participating sites were instructed to remove patient-identifiable information (name, NHS number and treating clinician) from the scans before they were sent to the CTRU. If this was not possible, sites were required to encrypt the scan data. The possibility that scans would be incompletely anonymised was explained to participants in the patient information sheet and as part of the informed consent process.

The use of a standardised classification system for interpreting scans has been shown to minimise the risk of missing subtle changes. 86 A number of systems have been developed. 87–89 When selecting a system for the DARS trial, it was desirable to use an existing protocol with proven interobserver agreement that could be applied to both CT and MRI scans. As motor learning is thought to depend on the interaction of several cortical areas and the basal ganglia,20,21,28 the template was required to accurately code the location and extent of subcortical infarcts as well as the cortical lesions. The Acute Ischaemic Stroke Classification Template (AISCT)87 was developed by Professor Wardlaw et al. of the University of Edinburgh Brain Research Imaging Centre and is available to download from www.ed.ac.uk/files/imports/fileManager/CT%20and%20MR%20reading%20form.pdf (accessed 6 July 2018). It was used in DARS with permission of Professor Wardlaw (2015, personal communication).

The AISCT was developed in 1994 as a system that reliably codes both the site and size of an infarct, as well as the degree of swelling and any haemorrhagic transformation. 87 Following review of over 100 scans, Wardlaw90 constructed standardised templates to illustrate the different patterns of infarction and tissue swelling. The AISCT shows good interobserver agreement between experienced neuroradiologists for infarct size and type (K = 0.78), excellent agreement for infarct swelling (K = 0.8) and moderate agreement for haemorrhagic transformation of the infarct. 87 Among radiology trainees, it has moderate to good agreement for infarct size and site, fair to moderate agreement for infarct swelling and poor to fair agreement for haemorrhagic transformation. 87 The interobserver agreement of this system has subsequently been evaluated by the Acute Cerebral CT Evaluation of Stroke Study (ACCESS): a large online validation, which has now been ongoing for several years. 86,91 Neuroradiologists tended to spot more subtle signs of ischaemia than non-radiologists91 and took longer to read each scan than non-radiologists. 86 More severe ischaemia (hypodensity and swelling) was more reliably detected than subtle signs and a longer time from presentation to scan also improved detection rates for ischaemic change. 86 Detection of acute ischaemia was not influenced by scan quality or by the presence of an old ischaemic lesion. 86 The AISCT has been used in the Third International Stroke Trial (IST-3)92 and also a subsidiary study of IST-3 to determine if CT or MRI angiography might be used to guide administration of tissue plasminogen activator at up to 6 hours after stroke. 93

The AISCT begins with an overall subjective judgement on the quality of images (trichotomised as good, moderate or poor). The presence of any visible abnormality (an acute stroke lesion or any other pre-existing abnormality) is then documented. For those scans judged to be abnormal, the presence or absence and laterality of any ischaemic change is documented. Features of early ischaemia were classified on CT as loss of definition between the cortical grey matter and underlying white matter; loss of the outline of the basal ganglia; and frank hypodensity. 86,91,93 For MRI, signs of early ischaemic change were classified as a faint hyperintensity on diffusion-weighted imaging (DWI) sequences but no lesion visible on T2-weighted or fluid-attenuated inversion recovery (FLAIR) images; bright hyperintensity on DWI but no/pale lesion visible on T2 or FLAIR images; or a lesion clearly visible on T2 or FLAIR images as well as on DWI. 93 Acute swelling was classified using the AISCT framework and reference diagrams. 86 Middle cerebral artery (MCA) lesions were classified as involving less than or more than one-third of this territory. 86,88 Using the reference diagrams developed for the AISCT, MCA territory lesions were then further classified as small cortical; basal ganglia striatocapsular; lateral to ventricle striatocapsular; anterior cortical MCA territory; posterior cortical MCA territory; whole of cortical MCA territory; whole of cortical MCA territory with lateral part of basal ganglia; and whole MCA territory. 86 Lesions in the anterior cerebral artery and posterior cerebral artery territories were each defined as involving < 50% of that territory, > 50%, or complete. 86 Lacunar lesions were classified as involving the internal capsule or lentiform, internal border zone, centrum semiovale, or thalamus. 86 Infarcts involving the anterior and posterior border zones were noted. 86 Cerebellar lesions were classified as lacunar infarcts or as involving < 50% or > 50% of the hemisphere. 86 Similarly, brain stem lesions were classified as lacunar or as involving < 50% or > 50% of the brain stem. 86 The Alberta Stroke Proforma Early CT Score (ASPECTS)89 was recorded for all lesions involving the MCA territory. 86 Only the primary acute ischaemic lesion was coded in this way. When two acute ischaemic lesions were present, we utilised clinical judgement when deciding which was likely to be the most clinically significant. Any second (minor) acute ischaemic lesion was coded simply as present or absent. 93

The AISCT categorises the presence of haemorrhage as petechial, significant haemorrhagic transformation of an underlying infarct parenchymal haematoma with no infarct visible, parenchymal haematoma clearly remote from infarct, subdural haematoma, subarachnoid haemorrhage and extradural haemorrhage. 93 It also records the maximum diameter of the lesion (< 3 cm, 3–5 cm, 5–8 cm or > 8 cm). 93 If blood is present in more than one location (e.g. a primary parenchymal haematoma with rupture in the subarachnoid space), then the presence of both was recorded93 and a clinical judgement was exercised to place them in rank order of likely significance. 93 The following variables were documented for the DARS trial analysis: the location of haematoma (frontal, temporal, parietal, occipital, basal ganglia/thalamus, internal capsule, brain stem, cerebellum);94 the extent of midline shift (in millimetres); the presence or absence of hydrocephalus; and the presence or absence of intraventricular extension. Haematoma volume was calculated using the equation:

where A is the greatest diameter of the haemorrhage on axial imaging, B is the greatest diameter at 90° to A and C is the approximate number of axial imaging slices on which the haematoma is visible, multiplied by slice thickness. 95,96 Haematoma diameter and volume were calculated for confluent haematomas only and no attempt was made to assess these parameters for petechial haemorrhage or for subarachnoid blood.

The presence and extent of white matter lesions were documented using the same validated systems97,98 as were used by Wardlaw et al. 93 Old vascular lesions were classified as old cortical infarct(s), old striatocapsular infarct(s), old borderzone infarct(s), old lacunar infarct(s), old brain stem/cerebellar infarct(s) or probable old haemorrhage. 93 Finally, non-stroke lesions were classified as cerebral tumour, encephalitis, cerebral abscess and demyelination. 93 Brain atrophy was recorded as present or absent.

All scans were co-reported by the clinical research fellow and one of two experienced consultant neuroradiologists (Dr Jeremy Macmullen-Price and Dr Tufail Patankar) on a CRF using the standardised reporting system using AISCT or ASPECTS. Images were reviewed blinded to the original study report from the recruiting centre, clinical information (stroke laterality, administration of thrombolysis and Oxford Community Stroke Project clinical stroke syndrome),99 and trial treatment allocation. Images were reviewed in DICOM format. The time taken to review each case was not recorded; reviewers were free to take as long as they deemed necessary.

It was expected that all scans would be subjected to routine reporting by local radiologists in order to guide the clinical management of the patient. As the review of scans for the DARS trial could be delayed by several months from the point of a patient’s enrolment in the trial, the DARS trial analysis could not provide clinically significant information in a timely manner. It was also important that the trial team did not seek to influence or guide the clinical management of participants in any way. Clinical feedback was not provided to local sites. However, when clinically significant and unexpected findings arose, they were communicated to the local PI, who was then responsible for ensuring that they were noted and acted on if necessary. 100

Economic evaluation

Screening and recruitment

During the recruitment period of May 2011 to March 2014, 19,509 patients were screened for eligibility. Of those screened, 1547 (7.9%) were deemed eligible, 599 (3.1%) were consented and 593 (3.0%) were randomised. Figure 3 shows the monthly, cumulative and target accrual and Figure 4 shows the flow of patients through the trial. The number of participants recruited per site ranged from 1 to 50.

FIGURE 3.

Cumulative recruitment over the DARS trial period (with revised target).

FIGURE 4.

Screening flow diagram for the DARS trial. a, Initial screening category. Reproduced with permission from Ford et al. 119 © 2019 The Author(s). Published by Elsevier Ltd. This is an Open Access article under the CC BY 4.0 license (https://creativecommons.org/licenses/by/4.0/).

Reproduced with permission from Ford et al. 119 © 2019 The Author(s). Published by Elsevier Ltd. This is an Open Access article under the CC BY 4.0 license (https://creativecommons.org/licenses/by/4.0/).

A comparison of screening data for randomised and non-randomised patients is presented in Table 4. Compared with screened patients, randomised patients were younger and more likely to be male, of white ethnicity and screened at a later date post stroke.

Eligibility violations

Table 5 provides details of the 10 eligibility violations that were identified during the trial: five in each randomised group. In the placebo group, one participant was unable to comply with the treatment schedule and four were not randomised within the eligible time period post stroke.

Initially, the eligibility criteria stated that participants must be diagnosed with stroke within 2 weeks prior to randomisation, which was subsequently changed to 42 days. One participant violated eligibility early on in the trial by being diagnosed > 2 weeks before randomisation; the remaining were diagnosed > 42 days after randomisation.

In the co-careldopa group, one participant was taking part in another trial, two had known hypersensitivity or contraindications to co-careldopa and two had their diagnosis of stroke revised (one to demyelination and one to traumatic brain injury). With the exception of the participant who was unable to comply with treatment (because they were nil by mouth), all participants received some protocol treatment.

Participant baseline characteristics

Overall, patient baseline characteristics and questionnaire scores were balanced between the randomised groups, as shown by the demographic characteristics (see Table 6), hospital admission and current stroke data (see Table 7) and neuroimaging data (see Table 8).

The mean age was 67.5 years for co-careldopa participants and 69.6 years for placebo participants. More males than females were randomised: 187 (60.7%) in the co-careldopa group and 177 (62.1%) in the placebo group. The majority of participants were of white ethnic origin: 289 (93.8%) in the co-careldopa group and 270 (94.7%) in the placebo group. With regard to education, 134 co-careldopa participants (43.5%) and 139 placebo participants (48.8%) spent ≤ 12 years in formal, full-time education (Table 6).

Participants were randomised at around 2 weeks post stroke (median of 14 and 15 days in the co-careldopa and placebo groups, respectively). The cause of stroke was cerebral infarction in the majority of cases: in 270 participants (87.7%) in the co-careldopa group and in 238 participants (83.5%) in the placebo group. Thrombolysis was received by 62 co-careldopa participants (23.0%) and 59 placebo participants (24.8%) (Table 7). The proportion of patients with total anterior circulation infarcts was higher in the co-careldopa group than in the placebo group (36.3% vs. 26.5%, respectively; see Table 7), although baseline walking/mobility did not differ between the two groups (see Table 13). This difference was due to chance.

The majority of participants underwent digital imaging [263 co-careldopa participants (88.0%) and 250 placebo participants (89%)] and more participants underwent CT than MRI [299 co-careldopa participants (97.1%) and 281 placebo participants (98.6%)] (Table 8).

Overall, 46 patients (7.8%) had also participated in another study: 27 in the co-careldopa group (8.8%) and 19 in the placebo group (6.7%).

The majority of participants had at least one current or ongoing medical condition: 284 in the co-careldopa group (92.2%) and 265 in the placebo group (93.0%), with a mean of 3.5 and 3.3 conditions, respectively.

Carer baseline characteristics

Eighty-four carers (27.3%) of participants in the co-careldopa group and 81 carers (28.4%) of participants in the placebo group consented to participate in the trial, and their baseline characteristics are presented in Table 9.

Participant withdrawals

Ninety-one participants (51.3%) withdrew from the trial: 58 (18.8%) in the co-careldopa group and 33 (11.6%) in the placebo group (Table 10). The majority of withdrawals occurred within 8 weeks of randomisation. Patient withdrawal accounted for most of the withdrawals: 33 (10.7%) in the co-careldopa group and 12 (4.2%) in the placebo group. The PI withdrew 13 patients (4.2%) and nine patients in the co-careldopa (3.2%) and placebo groups, respectively. A participant’s health and/or an AE were the most common reasons given for withdrawal, although a large proportion did not provide a reason. Approximately half of the withdrawals from treatment were for a clinical reason, with a higher proportion in the placebo group (n = 7, 63.6%) than in the co-careldopa group (n = 8, 40.0%).

Carer withdrawals

Ten carers (1.7%) withdrew: eight (2.6%) were carers of co-careldopa participants and two (0.7%) were carers of placebo participants. The majority of withdrawals occurred between 6 and 12 months following randomisation (Table 11).

Participant assessments

The time between randomisation and completion of questionnaires is shown in Figure 5. The median time between randomisation and follow-up assessments was 56 days for the 8-week primary outcome, 183 days for 6-month secondary outcomes and 370 days for 12-month secondary outcomes. At 8 weeks, six participants completed questionnaires later than the agreed timelines (five in the co-careldopa group and one in the placebo group). At 6 months, this increased to 16 participants (11 in the co-careldopa group and five in the placebo group). At 12 months, 25 questionnaires were completed late (16 in the co-careldopa group and nine in the placebo group).

FIGURE 5.

Time between randomisation and completion of RMI at each follow-up assessment.

Table 12 shows the number of questionnaires completed at each follow-up assessment, together with reasons for non-completion. Loss to follow-up was balanced between treatment arms at 8 weeks and 12 months; however, at the secondary outcome assessment at 6 months, more participants in the co-careldopa group than in the placebo group did not complete the assessments (21.4% and 12.3%, respectively). A comparison between randomised groups of the characteristics of those who did not respond to the 6-month questionnaire revealed no differences in sex, age, stroke type, RMI score at baseline, ability to walk independently at 8 weeks or BI at baseline. However, among participants who did not respond to the 6-month questionnaire, those in the co-careldopa group had a statistically significantly higher NEADL mean score at baseline (59.6 points, 95% CI 57.4 to 61.7 points) than those in the placebo group (52.5 points, 95% CI 47.4 to 57.5 points; p = 0.012).

Table 13 summarises the participant assessments at baseline and 8 weeks, 6 months and 12 months post randomisation. Further description is provided in the following sections.

Carer assessments

Table 14 summarises the category and total scores for CBS at all time points. At the 8-week follow-up, carers of participants in both arms reported moderate to severe burden, although those in the placebo group reported higher mean total scores than those in in the co-careldopa group: 46.6 points compared with 43.0 points, respectively, with strain, emotional burden and disappointment accounting for the greatest differences between the groups. The score at 6 months increased slightly in both groups, to 44.6 points in the co-careldopa group and to 49.1 points in the placebo group, whereas, at 12 months, the mean score remained stable in the co-careldopa group (44.6 points) but increased slightly in the placebo group (51.8 points). The participant characteristics for the subsample with a consenting carer are shown in Table 15 and appear similar between arms.

Primary end point

Per-protocol analysis

The per-protocol population excluded the following participants: (1) non-eligible patients; (2) patients not receiving the treatment they were randomised to including treatment crossovers; (3) those receiving no IMP prior to rehabilitation therapy; (4) patients not strictly complying with protocol (i.e. randomised drug taken 45–60 minutes before therapy, involving ≥ 20 minutes of motor therapy for ≥ 80% of the sessions); and (5) patients completing the 8-week follow-up primary end point outside 7–9 weeks. Figure 6 shows the number of exclusions at each step. The final sample size of the per-protocol population was 54 (9.1% of the ITT population), representing 9.7% of the co-careldopa group and 8.4% of the placebo group. Table 18 presents the descriptive statistics of the per-protocol population. Owing to the small sample size, no analyses were conducted on this population.

FIGURE 6.

Flow chart of exclusions to obtain per-protocol population.

TABLE 18 - Descriptive statistics for the per-protocol population, by ability to walk independently at 8 weeks

Q1, quartile 1; Q3, quartile 3.

Participant characteristics	Able to walk independently at 8 weeks
	Yes (N = 24)	No (N = 30)	Total (N = 54)
Randomised allocation, n (%)
Co-careldopa	14 (46.7)	16 (53.3)	30 (100.0)
Placebo	10 (41.7)	14 (58.3)	24 (100.0)
Patient sex, n (%)
Male	13 (48.1)	14 (51.9)	27 (100.0)
Female	11 (40.1)	16 (59.9)	27 (100.0)
Type of stroke, n (%)
Infarction	22 (44.9)	27 (55.1)	49 (100.0)
Primary haemorrhage	2 (40.0)	3 (60.0)	5 (100.0)
RMI score at baseline (points)
Mean (SD)	2.8 (1.58)	1.8 (1.24)	2.3 (1.48)
Median (range)	3.0 (0.0–6.0)	2.0 (0.0–4.0)	2.0 (0.0–6.0)
Q1, Q3 (points)	1.5, 4.0	1.0, 3.0	1.0, 3.0

A number of sensitivity analyses were conducted to test the robustness of the assumption that patients who died or were lost to follow-up were categorised as unable to walk independently. Descriptive statistics and estimates with 95% CIs and p-values are presented in Appendix 3 for the following analyses:

1. Analysis with the assumption that patients who died, were lost to follow-up or their RMI score could not be imputed were able to walk independently at 8 weeks (see Tables 69 and 70).

2. Complete-case analysis (see Tables 71 and 72).

3. Using patient-completed RMI score at baseline: descriptive statistics as for sensitivity analysis (see Table 73).

4. Using researcher-completed RMI score at baseline in place of score at 24-hour randomisation (see Tables 74 and 75).

5. Including only participants for whom the 8-week questionnaire was received by 12 weeks post randomisation (see Tables 76 and 77).

6. Assuming participants with no primary outcome data because they were too unwell were unable to walk independently, and participants who had died, withdrawn, were unwilling to be visited, refused to completed the questionnaire, had moved out of the area, those we could not get hold of and questionnaires that were lost at site or in the post were classed missing (see Tables 78–80).

The results for a sensitivity analysis when centre was included as a fixed effect in the model is not presented as the model fit was poor.

None of the sensitivity analyses affected the estimates: the ORs were not statistically different from those in the primary ITT analysis.

Patient secondary end points

Estimates and 95% CIs for the patient secondary end points are presented in Tables 19–24. At all follow-up assessments, there was no evidence of statistically significant differences between the randomised groups in patient secondary end points, with the exception of GHQ-12 score. Participants in the co-careldopa group had a significantly lower GHQ-12 score than those in the placebo group at the 6-month follow-up.

Carer secondary end points

Owing to small numbers, a single-level model was conducted for carer secondary end points (i.e. centre was not fitted as a random effect). Adjusted MDs in CBS scores were statistically significantly higher in carers of those in the placebo group than in carers of those in the co-careldopa group (Table 26), indicating higher levels of burden in the placebo group. Female carers reported greater burden than males at each follow-up time and older carers had lower scores at 12 months.

Safety

Mortality

A total of 39 participants (6.6%) died within 12 months of randomisation: 22 (7.1%) in the co-careldopa group and 17 (6.0%) in the placebo group. More participants in the co-careldopa group died within 8 weeks than in the placebo group: six (1.9%) and one (0.4%), respectively. The highest proportion of deaths occurred between 6 and 12 months [nine (2.9%) in the co-careldopa group and nine (3.2%) in the placebo group], and the majority occurred in hospital [14 (4.5%) in the co-careldopa group and 12 (4.2%) in the placebo group] (Table 34). The cause of death of the seven participants who died during treatment is presented in Table 35 and a graphical representation of the time between randomisation and death is presented in Figure 7. There were no carer deaths.

TABLE 34 - Summary of participant deaths, by treatment group

Participants deaths	Treatment group, n (%)		Total (N = 593), n (%)
	Co-careldopa (N = 308)	Placebo (N = 285)
Number of deaths	22 (7.1)	17 (6.0)	39 (6.6)
Time death occurred
Up to 8 weeks	6 (1.9)	1 (0.4)	7 (1.2)
From 8 weeks to 6 months	7 (2.3)	7 (2.5)	14 (2.4)
From 6 months to 12 months	9 (2.9)	9 (3.2)	18 (3.0)
Place of death
Home	2 (0.6)	4 (1.4)	6 (1.0)
Hospital	14 (4.5)	12 (4.2)	26 (4.4)
Institutional care	5 (1.6)	2 (0.7)	7 (1.2)
Unknown	3 (1.0)	2 (0.7)	5 (0.8)

TABLE 35 - Cause of death for participants who died within 6 weeks of randomisation

NSTEMI, non-ST elevation myocardial infarction.

Randomised allocation	Cause of death
Co-careldopa	Cause of death for participants who died during treatment
	NSTEMI, respiratory arrest
	Pneumonia
	Deterioration from initial stroke
	Re-admitted to acute stroke unit from community rehabilitation unit with having had another stroke
	Peritonitis, bowel perforation
Placebo	Aspiration pneumonia

FIGURE 7.

Time between randomisation and death.

Content of occupational and physical therapy

Treatment compliance

Drug (kit) replacement

Four participants lost their drug kits and required a replacement: three in the co-careldopa group and one in the placebo group. No participant received the wrong drug.

Patient perspective regarding the use of the investigational medicinal product with rehabilitation treatment

Although many participants did not answer the questions relating to patient perspective regarding the use of IMP, those who did answer generally found it easy to remember to take the drug and thought that the package instructions were clear. In total, 21% of participants in both groups found it difficult to remove the IMP from the packaging and had to ask for help, but the majority did not have problems taking the drug (Table 48).

Blinding

Further analysis

Moderator analysis

The moderator analysis explored whether or not the size of the treatment effect (irrespective of whether or not the effect is significant) was influenced by baseline characteristics of patients. The following results are based on exploratory analysis and should be treated with caution.

Each moderator variable was tabulated by primary outcome category and treatment group (see Appendix 6, Tables 82 and 83). Mean baseline RMI score (regardless of reporting status) and the proportion of patients with no lower limb pain was higher among those walking independently at 8 weeks. There was no difference in baseline RMI score or lower limb pain between the treatment groups. There was a suggestion that both site of MCA lesions and haemorrhage location were associated with the primary outcome and treatment group. A higher proportion of those with no MCA lesion and those with a (subcortical) haemorrhage walked independently at 8 weeks and belonged to the placebo group, which fits with the ITT analysis that found that those in the placebo group have a higher odds of walking independently at 8 weeks (though not statistically higher than those in the co-careldopa group). Scan data were available for 472 out of 593 patients (79.6%) and patients with haemorrhage are those who had a primary haemorrhage stroke (rather than infarction). Those without scan data were more likely to be in the co-careldopa group (58.7% without scan data vs. 50.2% with scan data) and have fewer (≤ 12) years of education (53.7% without scan data vs. 45.2% with scan data). Hence, by not including those with scan data, the numbers in the co-careldopa arm reduces and the ITT analysis of the primary outcome shows an OR closer to 1 (OR 0.806, 95% CI 0.521 to 1.249).

With the exception of baseline RMI score, no other baseline characteristics showed potential moderation when accounting for covariates included in the primary analysis multilevel model (see Appendix 6, Table 84). There was weak evidence (p = 0.069) that baseline RMI score moderates the effect of treatment on walking independently at 8 weeks, such that there is a significant difference in the association of baseline RMI score with the primary outcome for placebo and co-careldopa. The effect of treatment on walking independently at 8 weeks changes as baseline RMI score increases (Figure 8). When the baseline RMI score is 0 points, those in the placebo arm have a significantly higher chance of walking independently at 8 weeks than those in the co-careldopa arm, but there is no evidence of a significant difference between the treatment groups at the mean baseline RMI score or as the baseline RMI increases towards a score of 6 points (Figure 9).

FIGURE 8.

Linear prediction plot for treatment group, by baseline RMI score.

FIGURE 9.

Predictive margins of treatment group with 95% CI.

Departures from randomised treatment as a mediator

All patients received their treatment in accordance with their randomised allocation. Of the 593 patients with primary outcome data, 536 had five or more therapy sessions and were included in descriptive compliance statistics. With the exception of drug intake compliance, there was no suggestion of a difference in the distribution of the other categories of compliance between the treatment groups (Table 54) or of a difference between the treatment groups in the outcomes of ‘non-compliers’, where ‘non-compliers’ are defined as those not satisfying the criteria for the compliance category (Table 55). Within the relaxed drug intake compliance category, the proportion of compliant patients was higher in the placebo group than in the co-careldopa group (76% vs. 68%, respectively) and the proportion of non-compliant patients walking independently at 8 weeks was higher in the placebo group than in the co-careldopa group (48% vs. 38%, respectively).

Mediator analysis

The details of the mediator analyses are shown in Appendix 7 (see Tables 85–89). In summary, there was no evidence that any of the variables investigated mediated the effect of treatment on walking independently at 8 weeks.

Sample

A total of 593 patients were recruited to the trial (308 patients to the co-careldopa group and 285 to the placebo group). Of these, 1% died between baseline and 8 weeks, 2% died between 8 weeks and 6 months and 3% died between 6 months and 12 months. Two-thirds of those who died before the end of the trial were from the co-careldopa group.

Missing data

A total of 122 patients had complete EQ-5D scores and costs at baseline and all follow-up periods. For the remaining 471 participants (53% from the co-careldopa group and 47% from the placebo group), either EQ-5D scores and/or cost data were missing for at least one of the follow-up periods. Indicatively, and as shown in Figure 10, 99 patients from the co-careldopa group had missing EQ-5D scores at 12 months, compared with only 19 patients at baseline; of these, 13 patients had all EQ-5D items missing at 12 months compared with 93 at baseline. The number of patients with missing EQ-5D items increased over time, and a majority of those with missing data did not complete the measure. Figure 11 shows the number of patients with missing resource use items and, hence, missing costs. For example, 180 patients in the placebo group had missing costs at baseline, compared with 228 patients at 12 months. The number of patients with missing costs increased over time and co-careldopa patients were more likely to have missing cost data. In addition, most of the missing costs were the result of at least one missing item rather than non-completion of the resource use questionnaire.

FIGURE 10.

Total number of patients with missing EQ-5D scores, by treatment group.

FIGURE 11.

Total number of patients with missing total resource use costs, by treatment group.

Missing utility and cost data were imputed in each period using, as predictors, health status measures (BI),120 mRS121 and GHQ-12,122 and baseline characteristics such as age, sex, ethnicity and type of stroke. Imputation was possible for 313 patients but not for 158 patients. The latter was due to missing scores over time for at least one of the predictor variables (BI, mRS and GHQ-12). Therefore, a total of 435 patients (216 placebo and 219 co-careldopa) were included in the primary base-case analysis.

The baseline characteristics of the patients included in the analysis sample are presented in Table 56. Patients in the placebo group were marginally younger than those in the co-careldopa group. In both arms, more than half of the patients were male. More than two-thirds of the patients in both arms were admitted with an infarction stroke at baseline but did not receive thrombolysis. These results seem to be consistent across samples (i.e. base case compared with complete case) with the exception that, in the complete case, patients in the co-careldopa arm were slightly older than those in the placebo arm.

Resource use and costs

As resource use questions for the last follow-up periods (6 months and 12 months) referred to the last 3 months, monthly face-to-face and telephone/e-mail contacts were calculated and then multiplied by 4 and 6, respectively, to cover the period between the follow-ups. Table 57 shows the average resource use per patient in each trial arm in the observed data only (i.e. complete case). Patients in the co-careldopa group were more likely than those in the placebo group to use the majority of services, and that pattern persisted over time. Any other conclusions should be tempered given the little reported resource use within the trial.

In terms of the total NHS resource use cost, the co-careldopa group used the highest number of resources and also had the highest cost. Mean total costs from the use of primary care and community health services (before imputation) were £5095.70 (SD £6555.99) for the placebo group and £7937.92 (SD £12,247.55) for the co-careldopa group. Mean total costs from the use of hospital and residential care services (before imputation) were also higher for the co-careldopa group [£15,595.72 (SD £20,182.44)] than for the placebo group [£13,619.56 (SD £14,950.22)]. There were no significant differences in costs between the two arms using Mann–Whitney U-tests. This was the case for both the complete-case sample and the sample after imputing for missing data. For this reason, no adjustments in costs for baseline differences were considered necessary. Table 58 reports resource use for complete cases only as the imputation generated total cost values rather than individual resource counts.

Quality-of-life data

Table 59 shows the mean EQ-5D scores for each period for the two arms of the trial when no scores are imputed (i.e. complete case) and when values are imputed (i.e. base case) (Figure 12 provides a graphical representation). In both arms, there was a considerable increase in EQ-5D scores from baseline to 8 weeks. There are small fluctuations in the utility values between 8 weeks and 12 months for both placebo and co-careldopa, but the improvement in utility scores is more or less maintained until the last follow-up period (12 months).

FIGURE 12.

Mean (unadjusted) EQ-5D scores by treatment group (base case).

Table 60 provides the mean EQ-5D change scores between baseline and follow-up points for the base case. All of the observed changes are above the estimated minimally important difference (range 0.08–0.12) for the EQ-5D. 123 Independent-sample t-tests showed that there were statistically significant differences over time in EQ-5D scores in both trial arms. Differences in EQ-5D scores between the placebo and co-careldopa arms were significant (at the 1% level) at baseline and at 8 weeks but not at any of the other follow-up times. This indicated that baseline adjustments are necessary and, thus, adjustments were made using an ordinary least squares (OLS) regression of QALYs on the treatment indicator and baseline EQ-5D. 124

Cost-effectiveness results

Table 61 shows the costs and QALYs for each of the two trial arms for the primary analysis (multiple imputation for missing data and adjustment for baseline EQ-5D differences). It also provides the incremental cost and benefit (expressed as QALY gains). On average, co-careldopa patients incurred higher costs and gained fewer QALYs than placebo patients. Therefore, co-careldopa is ‘dominated’ by placebo and, hence, the ICER was not calculated. The results indicate that co-careldopa is not cost-effective.

Figure 13 shows the cost-effectiveness plane for co-careldopa compared with placebo, based on bootstrapped estimates of costs and QALYs. The average costs from the bootstrapped estimates were £14,531.73 (SD £1130.00) and £16,640.91 (SD £1294.21) for the placebo and co-careldopa groups, respectively. The mean number of QALYs was 0.420 (SD 0.002) for the placebo group and 0.397 (SD 0.002) for the co-careldopa group. The simulation estimates are spread in the north-west and south-west quadrants, suggesting that co-careldopa is unlikely to lead to better health outcomes. They also indicate that placebo is less costly but more effective in terms of improving HRQoL. A small proportion of simulations is beneath the diagonal cost-effectiveness threshold (£20,000 per QALY) indicating cost-effectiveness. However, in these cases, co-careldopa is still less effective but less costly.

FIGURE 13.

Cost-effectiveness plane for co-careldopa compared with placebo (outcome measure: QALY, health-care provider perspective).

A CEAC is presented in Figure 14. At a threshold of £20,000 per QALY, co-careldopa has a 7% chance of being cost-effective, decreasing to 5% when the threshold λ = £30,000. There is a negative relationship between λ and co-careldopa’s chance of being cost-effective as placebo is more effective on average.

FIGURE 14.

Cost-effectiveness acceptability curve.

Net benefit regression

We calculated the NMB for each patient assuming a threshold λ = £20,000 and accounted for differences in EQ-5D scores between trial arms at baseline. The average NMB for both groups was negative. In detail, the NMB for the placebo group was –£6138.90 (SD £16,841.91) and for the co-careldopa group was –£9146.19 (SD £20,543.19). In Table 62 we present the results from the net benefit regression. We ran OLS regressions of the NMB on the treatment indicator and other controls such as age, sex, type of stroke, classification of stroke and mRS score at 8 weeks. Some of the models included also showed interactions between treatment and mRS, treatment and type of stroke, and treatment and stroke classification. In the majority of the models, treatment was not found to be a significant predictor of net benefit. On two occasions, the treatment coefficient was significant at the 10% level and negative, suggesting that co-careldopa was associated with lower net benefit than placebo.

As mentioned in the results section, a per-protocol analysis was not considered feasible because only a few patients met the strict compliance criteria (i.e. patient received tablet 45–60 minutes before therapy for ≥ 80% of the sessions) set by the trial team. However, as a sensitivity analysis, and to see whether or not patient non-compliance was a driver of results, we relaxed the strict compliance criterion (i.e. 80%). In particular, we looked at 75%, 65%, 55%, 50%, 45% and 40% compliance and ran OLS regressions of NMB on treatment and other controls, each time on the population that met the specific compliance percentage. The results are presented in Table 63. In all cases, except when compliance is 40%, the treatment coefficient is not significant, suggesting that the lower monetary benefits of co-careldopa than placebo were not due to the non-compliance of the patients.