Effects of Unilateral Upper Limb Training in Two Distinct Prognostic Groups Early After Stroke

Design

We conducted a stratified, multicenter, observer-blinded randomized controlled trial with 8 repeated measurements within the first 26 weeks poststroke. After stratifying patients within the first 14 days after stroke to either the mCIMT trial (if they had voluntary FE of 10° or more) or the EMG-NMS trial (no voluntary FE), randomization to the experimental treatment or usual care was performed online by a restricted minimization randomization procedure. Randomization was executed by 2 staff members not involved in the treatment protocol (GK and HA).

Two trained observers (RN and CW) who were blinded to the treatment allocation performed all clinical measurements at baseline (before randomization), and at 2, 3, and 4 weeks of follow-up during face-to-face sessions at the location of initial admission and at the location of subsequent rehabilitation. Postintervention assessments were performed at 5, 8, 12, and 26 weeks after stroke. The same assessor performed the serial assessments for an individual subject in both trials. Patients were not blinded to the intervention, however they remained naïve as to the supposed efficacy of the 2 intervention conditions.

Both clinical trials were part of the EXPLICIT-stroke program. EXPLICIT-stroke is an acronym for EXplaining PLastICITy after stroke and was funded for 5 years by the Netherlands Organisation for Health Research and Development (ZonMw: No. 890000001). The study was approved by the Medical Ethics Review Committees of Leiden University Medical Center (No. P08.035) and the Dutch Central Committee on Research Involving Human Subjects (CCMO: No. NL21396.058.08), and was registered in the Netherlands Trial Register (NTR, www.trialregister.nl, TC1424).

Patient Selection

For both trials, we selected patients with (1) first-ever ischemic stroke in one of the cerebral hemispheres; (2) upper limb paresis according to National Institutes of Health Stroke Score (NIHSS) item 5; (3) baseline ARAT score of ≤53 on a maximum of 57 points; (4) ability to communicate and comprehend (Mini Mental State Examination ≥23 points on a maximum of 30 points); (5) ability to sit independently for at least 30 seconds; (6) 18 to 80 years of age; (7) no successful thrombolysis therapy resulting in upper limb motor recovery and attaining 0 points on NIHSS item 5 of the paretic arm; (8) no musculoskeletal impairments of the upper paretic limb; (9) no additional therapies such as botulinum toxin injections or medication intake that may influence upper limb function in the previous 3 months; (10) willing to participate in an intensive rehabilitation treatment program; and (11) written informed consent. At intake, patients were stratified to either the mCIMT trial if able to voluntarily extend the thumb and/or 2 or more fingers of the affected hand (10° or more), or to the EMG-NMS trial if they could not.

Interventions

Patients allocated to the mCIMT group daily received 60 minutes of supervised intensive graded practice focused on improving task-specific use of the paretic arm and hand, including enhancing FE. Therapy was delivered in either 1 session or split into 2 sessions of 30 minutes, depending on the available time and patients’ tolerance. Time between sessions per day was not controlled for. One hour of mCIMT therapy was chosen to not overload patients in the early phase poststroke.^16,17 Patients were instructed to wear a padded safety mitt (Sammons Preston #6727; Sammons Preston, Inc, Bolingbrook, IL, USA) for 3 hours per working day, during 3 consecutive weeks and lasting up until 5 weeks poststroke. Details of the mCIMT treatment protocol have been described previously. ¹⁷

Patients allocated to the EMG-NMS group received two sessions of 30 minutes stimulation of the finger extensors each working day, for 3 weeks, using the Stiwell-Med4-system (Ottobock Healthcare Products GmbH, Vienna, Austria). Active participation of the patient was required to reach the EMG threshold during the dorsiflexion movement of fingers and wrist in order to trigger the NMS. When triggered, the Stiwell-Med4-system stimulated for 5 seconds, followed by 25 seconds of rest. In case of absent EMG-activity, patients learned to facilitate extensor activity of wrist and fingers by simultaneously abducting their paretic arm (often described as Soques’s phenomenon) or by simultaneously EMG-triggering of extensors with the nonparetic arm. ¹⁵ EMG-NMS therapy was augmented by offering patients visual feedback presented in front of the patient on a computer screen. For this purpose, the amount of voluntary EMG-activity of finger extension was visualized in a computer game in which patients had to raise a hot air balloon above a mountain.

Usual upper limb therapy in both strata consisted of exercise therapy based on recommendations from current Dutch guidelines ¹⁸ applied face-to-face by a physical therapist or occupational therapist for 30 minutes per working day executed for 3 consecutive weeks. Patients in the control group with an unfavorable prognosis engaged in passive range-of-motion exercises and facilitation of voluntary movements, whereas those with a favorable prognostic group received exercise therapy according to Dutch guidelines. ¹⁸

All therapy sessions were performed in each participating center by two to three trained physiotherapists and/or occupational therapists. The content and duration of therapy was recorded by the therapist in patient diaries. After the 3-week intervention phase, all patients received usual care (ie, about 30 minutes arm-hand treatment per working day) during their stay in the rehabilitation center or nursing home for on average 2 months. The exact duration and type of postintervention usual care was not controlled for, and based on individual needs.

Outcomes

The primary outcome variable for both trials was the ARAT score. ¹⁹ The ARAT is a performance test that assesses the ability to perform gross movements and the ability to grasp, move, and release objects of various sizes, weights, and shapes. ¹⁹ The original test consists of 19 items rated on 4-point ordinal scales. Removing 4 items enabled the construction of a hierarchical 1-dimensional scale. ²⁰ The ARAT has been shown to be valid, reliable, and responsive. ²¹ It was applied according to the guidelines developed by Yozbatiran et al. ²² The minimal clinically important difference (MCID) was set at 10% of the range of the scale, that is, 6 points. ²³

Secondary outcome variables in both trials were the following Upper Extremity motor function section of the Fugl-Meyer Assessment (FMA-UE), Wolf Motor Function Test (WMFT), Motricity Index of the Upper Extremity (MI-UE), Erasmus modification of the Nottingham Sensory Assessment of the Upper Extremity (EmNSA-UE), Nine-Hole Peg Test (NHPT), Frenchay Arm Test (FAT), Motor Activity Log (MAL), and the Hand domain of the Stroke Impact Scale (SIS-Hand, version 3.0).

The FMA-UE is a reliable and valid motor function test^24,25 evaluating the ability to make arm movements outside the synergistic pattern. ²⁶ The WMFT is reliable and valid ²⁷ and consists of 2 strength items, 6 timed tasks, and 9 integrative functional tasks. Assessment involves performance time with a maximum of 120 seconds and a 6-point functional ability scale (FAS). ²⁷ The MI reliably and validly assesses upper limb strength by testing 6 functions rated from 0 to 100 points.^28,29 The EmNSA is a 3-point ordinal scale measuring sharp-blunt discrimination, 2-point discrimination, and proprioception. Except for the 2-point discrimination, intra- and interrater reliability are good to excellent (κ = 0.58-1.00). ³⁰ The NHPT is a reliable and valid test assessing manual dexterity^31,32 by measuring the speed with which a patient grasps and inserts (and removes) 9 pegs into a grid of vertical holes. The test was discontinued after 50 seconds if the patient was still unable to insert any pegs. ³² Reliability and validity have been assessed and norms are available.^31,32 The reliable and valid FAT measures dexterity using 5 functional tasks, scoring on a binominal scale (fail/pass). ³² We used a translated and adapted version of the MAL ³³ comprising the 14 original activities. It was used to independently rate how well (6-point quality of movement [QOM] scale) and how much (6-point amount of use [AOU] scale) the paretic arm was used spontaneously to accomplish the activities of daily living outside the laboratory. Reliability and validity of the MAL have been proved in a number of studies. ³³ The SIS-Hand was used to evaluate patient-perceived outcome for the paretic upper limb. Each item was scored on a 5-point rating scale from “not difficult at all” to “cannot do at all.” The SIS has shown excellent psychometric properties in terms of concurrent and construct validity, test-retest reliability and responsiveness.^34,35

Sample size estimation

Power analysis showed that a sample size of 60 patients (including 10% dropout) was estimated to be sufficient for the mCIMT trial, and 120 (including 10% dropouts) for the EMG-NMS trial. Details are described elsewhere. ¹⁵

Statistics

Successful blinding of the assessors for treatment allocation was tested by comparing assessors’ guesses with actual treatment assignment in both trials using a Cohen’s κ statistic. In both trials, we tested for differences in baseline values with Fisher’s exact test or the χ²-test for nominal outcomes, the Mann-Whitney U test for ordinal outcomes, and Student’s t-test for independent groups for other outcomes, assuming equal variances for interval or ratio scales. The assumption of normality was tested with Z-scores for skewness and kurtosis, and by visual inspection. To analyze time series of continuous outcomes, the generalized estimating equations model with a compound symmetry as a covariance structure was used to evaluate differences in overall effects in the experimental and control groups within each trial over the time course of 5, 8, 12, and 26 weeks poststroke. This model included time, group, and the interaction between group and time in the regression model. We added the baseline value of the dependent variable to the regression model. The proportional difference in the number of stroke patients showing some return of FE and dexterity in the EMG-NMS trial was assessed using a dichotomous generalized estimating equations model. Intention-to-treat analyses were performed on all patients, including those with incomplete sets of data. Missing values were not imputed. We calculated β-values and standard errors for the group × time interaction effects and, subsequently, used the Wald test to obtain P values. All hypotheses were tested 2-tailed, with an α = .05. All analyses were carried out in SPSS, version 20.

Strengths and Limitations

First, patient selection limits the generalizability to other stroke populations. Recruitment of patients with the same stroke type and no limiting comorbidities early poststroke is difficult as is illustrated by the proportion of patients eventually included 3.7%. The proportion of included patients was comparable to, for example, the 2.9% in the VECTORS-trial. ⁶ Despite this low recruitment rate restricted to patients with a first-ever ischemic hemispheric stroke, this study is the largest and earliest started upper extremity intervention trial conducted thus far. Fewer stroke patients were recruited for the EMG-NMS trial (n = 101) than originally planned (n = 120). ¹⁵ However, increasing the number of patients in this trial and, hence, its statistical power for the probability of regaining some upper limb capacity would most likely not have changed our conclusion because the current results do not show any trend toward positive, clinically significant effects of EMG-NMS in comparison with usual care. Omission of the stratification, by allocating all patients to 1 experimental (n = 79) and 1 control group (n = 80), would have resulted in a neutral trial due to lack of homogeneity between subjects. We believe this prognostic stratification is essential when performing rehabilitation trials, and may be a powerful tool to prevent type II errors.^2,43 Furthermore, in line with previous experience in rehabilitation trials^4,28 we applied a repeated measurements design to create individual time series in order to reduce the unexplained variance due to within-subject measurement error and to enhance the precision of estimating the added value of therapy relative to the underlying logistic pattern of spontaneous neurological recovery. Unfortunately, most rehabilitation trials so far did include patients at arbitrary time points in the first 3 months poststroke, ⁴³ whereas time-dependent spontaneous neurological recovery explains almost 70% of recovery.^45,46 Not stratifying patients according to the expected proportional recovery46, as well ass starting recruitment of subjects in trials at nonfixed moments poststroke, such as moment of admission in a rehabilitation center, will affect the variation in spontaneously driven gains between subjects. ⁴⁴ Patients recruited during the first days poststroke will show on average more change than those recruited after a number of weeks. We are of the opinion that differences in starting times poststroke and lack of stratification increases the within-group variance for measured improvement in trials, and that this will be at the expense of identifying therapy-induced effects between-groups in randomized clinical trials. An important next step is to figure out how differences in timing of recruiting patients for trials in the subacute phase poststroke will affect random as well as systematic error in parallel group designs by using simulation models.

Second, both the mCIMT and the EMG-NMS group received more therapy in comparison to the usual care groups, that is, the amount of therapy was not dose-matched. We therefore cannot answer the question whether the beneficial treatment effect of mCIMT was completely due to the content of the therapy itself or partly by the higher dose of therapy applied. Another limitation may be the choice of EMG-NMS as an intervention for patients with no voluntary control of finger extensors at baseline. Thirty-two patients with an unfavorable prognosis showed return of voluntary FE due to spontaneous neurological recovery within the first 5 weeks. Moreover, these patients showed more improvement in motor function than expected following the shoulder abduction and finger extension model.^4,47 Further research should try to identify these “false negatives” in order to improve stratification for intervention. Despite return of some FE, evidence-based therapies for this patient group with an unfavorable prognosis are lacking at this moment. From our data, it cannot be determined if a longer intervention phase, or applying EMG-NMS solely with task specific movements would have changed the current findings. There is some evidence that the sensory-motor integration during EMG-NMS of the paretic arm may activate neuronal networks^13,48 and increases the cortical perfusion of the ipsilesional sensorimotor cortex in chronic stroke. ⁴⁹ Our findings stress the need to explore the meaning of changes in neural networks, as revealed by imaging techniques such as functional magnetic resonance imaging and electroencephalography, regarding the impact of improved motor performance early poststroke. ⁴¹ Furthermore, the added value of combining task specific practice with interventions directly focused at the damaged brain should be investigated (eg, noninvasive brain stimulation and neuropharmacological therapies).

Interpretation

This is the first stratified randomized controlled trial of poststroke upper extremity rehabilitation based on functional prognosis using repeated measurements. Our results suggest that 1 hour daily supervised mCIMT for 3 weeks in those with some voluntary FE early poststroke is more effective than usual care alone leading to clinically meaningful effects on arm-hand capacity up to 3 months poststroke. We found no evidence that 1 hour daily EMG-NMS for 3 weeks in those without voluntary FE, influences the likelihood of return of any FE within the first 6 months after stroke. The present study therefore supports the clinical point of view that effects of upper limb training are restricted to patients with some voluntary motor control of finger extensors at baseline.^2,43 The literature indeed offers no examples of evidence-based therapies resulting in significant improvements in patients with an initial flaccid hand.^1,2,42,43,50 The preservation or return of FE early poststroke is assumed to reflect the integrity of the corticospinal tract. ³ Moreover, some innervation of end-effectors is needed to induce improvements by exercise therapy early poststroke. At the same time, we found no evidence that we were able to influence spontaneous neurological recovery of underlying impairments based on clinical scales, suggesting that functional improvements of the mCIMT group were based on adaptation strategies to use intact end-effectors in a more optimal way.^38,42