JOURNAL OF THE ROYAL SOCIETY OF MEDICINE
Biofeedback therapy in stroke rehabilitation:
Morton Glanz MD MSc Sidney Klawansky MD PhD
J R Soc Med 1 997;90:33-39
INTRODUCTION
The history of biofeedback therapy in neuromuscular
rehabilitation began with the discovery in 1830 that
electrical events emanated from the surface of a muscle
during its contraction1. The prototype forerunner of the
current electromyogram (EMG) appeared in 19282. Reports
of the application of biofeedback (BFB) techniques for the
treatment of various neuromuscular disorders began to
appear in the late 1950s and early 1960s3-5. In the next
three decades, enthusiastic case series, and later randomized
controlled trials, were published. The results of the latter
more rigorous trials were mixed in their appraisal of its
efficacy68. Finally over the last three years, quantitative
literature overviews, or meta-analyses, have been published
in an attempt to elucidate the potential role of this modality
more adequately9-11. Although the various reports have
described its use in a wide variety of neuromuscular
disorders including spinal cord injury12, spastic torticollis
and other dystonias13, cerebral palsy and peripheral nerve
damage14, by far the greatest area of investigation and
concomitant utilization has been post-stroke rehabilitation.
This paper will describe some of the basic physiological
constructs of BFB, and the functional aspects of applying it in
the clinical setting. The special problems related to research
in stroke rehabilitation will be reviewed with special
reference to BFB. However, the main focus and purpose
of this work will be to provide the reader with an assessment
of the overall efficacy of this modality in stroke
rehabilitation.
METHODS
A MEDLINE search was conducted by use of the keyword
'Biofeedback'. This indexes articles dating back to 1966,
well before the first clinical trials on BFB began to appear.
One thousand four hundred and nineteen items were cited.
Abstracts of all articles relevant to the use of biofeedback in
stroke rehabilitation were reviewed. Particular attention was
paid to randomized controlled trials, meta-analyses,
critiques of research or applications of BFB, and recent
reports on technological advances in the field. The tables of
Technology Assessment Group, Department of Healfth Policy and Management,
Harvard School of Public Health, Room LL-A7, 677 Huntington Ave, Boston, MA
02115, USA
Correspondence to: Morton Glanz
Volume 90 January 1997
a review
Thomas Chalmers MD
contents of retrieved articles were reviewed for other
relevant articles.
BACKGROUND
Cerebrovascular disease remains an extremely important
health problem. In the United Kingdom alone, each year
130000 new strokes occur, and 64000 individuals die as a
result of this ailment15. The actual incidence of stroke is 2.2
per 100016. Despite advances in the treatment of important
risk factors such as hypertension, the incidence of stroke has
not been declining of late, perhaps due to the aging of the
population in general, a problem which is expected to
progress'7. Each health district can expect to care for 1500
episodes annually at a cost of £3 million18. In 1988 the
King's Fund Consensus Statement was published. This
document recognized at once the importance of stroke in the
United Kingdom and the lack of organization and
standardization of the delivery of services for this disorder18.
As a result stroke rehabilitation in particular began to be
widely discussed in the health districts, and it received a
significant boost. Subsequent papers have delineated the lack
of objective evidence for the efficacy of various neuromus-
cular rehabilitative techniques in stroke and the need for
further research in the area'9.
Normal physiology
The EMG enabled researchers to monitor the fundamental
functional substrate of neuromuscular function, the motor
unit. This represents all of the muscle fibres which are
innervated by the branches of a single axon. It was
subsequently discovered that subjects could learn specifically
to gain control over the selective activation of a single such
motor unit, an ability that is the basis for all motor training,
which involves the progressive activation and inhibition of
larger groups of motor units. Selective alteration in the rate
of firing of individual units is another important aspect of
normal physiology, leading to increased strength of
contraction, and this too is amenable to BFB modulation20.
Altered physiology in stroke
The physiological impairment and consequent functional
limitation in stroke involves both decreased force of muscle
contraction and overactive but dysfunctional contractions
known as spasticity. These exist in opposing groups of 33
 JOURNAL OF THE ROYAL SOCIETY OF MEDICINE
muscles, resulting in impaired volitional movement. An
important additional phenomenon is that of muscle synergies
or the stereotypical synchronous contraction of several
closely grouped muscles, instead of the finely controlled
selective contraction and relaxation of individual muscles
which is important for normal mature motor function.
Analysis of the stroke rehabilitation literature is rendered
difficult by the interchangeable use of imprecise terms such
as function, limitation, impairment, and disability. Similarly,
the lack of standard staging taxonomy has hindered the
critical evaluation of rehabilitation modalities in stroke, and
certainly may explain a lot of the disparity of results in the
BFB trials. In a recent review, Duncan has proposed looking
at stroke disability at three levels: physiological impairment
(e.g. loss of range of motion), functional limitation (e.g.
inability to walk) and actual disability, such as the inability to
dress oneself. She has proposed a staging system based upon
these descriptive elements which is currently being
validated21.
Three basic physiological constructs have been the target
of BFB treatment in stroke. These are the functions of the
shoulder, wrist-hand, and ankle joints, as detailed by
Basmajian20. As a result of weakness of the rotator cuff,
levator scapulae, and teres minor muscles, the shoulder
joint, which is dependent on these muscles for anatomic
integrity during normal movement, becomes lax with
eventual downward subluxation of the humeral head. This
in turn renders the joint functionally useless. Weakness of
extensors, but more importantly spasticity of flexor
musculature, is of major significance in hand and wrist
function. Loss of ankle range of motion is the key factor in
analysing gait disability in stroke. In the swing phase of gait,
during which the foot must dorsiflex, the combination of
weakness of the anterior tibial musculature and spasticity of
the gastrocnemius and soleus muscles impairs this process.
PRINCIPLES OF BIOFEEDBACK THERAPY
The mechanism by which BFB may help in the rehabilitation
of the stroke patient is not clear. Wolfe has noted the
various possibilities, concluding that all represent con-
jecture6. No definitive studies since the time of his
observation were uncovered in the process of this review.
A favoured explanation is that the patient, through the
artificial proprioception provided by the BFB apparatus, is
able to gain conscious control over subliminal yet
undamaged upper neuron pathways which are in turn able
to subserve the missing functions responsible for the altered
physiology.
Biofeedback units range in size and complexity from
small individual portable units, suitable for home use, to
complex models with more advanced capabilities. All share
34 certain basic elements. These include one or more channels
Volume 90 January 1997
from which emanate bipolar electrodes capable of recording
activity in individual muscles, mechanisms for amplifying and
integrating the recorded EMG signal, and a scheme to
display the output, which can be of an auditory or visual
nature. Newer more complex models contain advanced
integrative and memory capabilities22.
In practice BFB is combined with traditional
physiotherapy, with the two modalities being integrated
by the therapist in complementary fashion. Many studies
evaluating the efficacy of BFB have focused the research
question in terms of BFB versus traditional physiotherapy.
Wolfe has argued against this artificial dichotomy, citing
the co-dependence of the modalities23. In an actual session
the therapist will apply the recording electrodes to one or
more muscles, asking the patient to try to activate (in the
case of paresis) or relax (in the case of spasticity or
undesirable synergy) the respective muscle. Judgement is
therefore required regarding such factors as the location
and spacing of electrodes, reflecting the therapist's
assessment of the altered physiology. The artificial
proprioceptive feedback provided by the auditory or
visual output of the unit enables the patient to become
'aware' of volitional changes in motor unit activity and
thus restore functional patterns of movement. How
various parameters of BFB can affect the outcome of
treatment with special reference to the reported clinical
trials will be discussed below. Additional parameters
include the frequency and duration of individual sessions,
the type of feedback rendered (continuous versus
bandwidth)24, and whether simple or more complex,
purposeful movement patterns are utilized25. These factors
amongst others are important in considering the
heterogeneity of treatment effects reported in the various
trials of BFB efficacy.
In a representative Boston rehabilitation hospital BFB-
assisted physiotherapy is billed at $160 per hour, while
routine physiotherapy is billed at $148 per hour. As is often
the case with charges they may not reflect the true resource
cost26. Nevertheless the similarity of the charge regardless of
whether BFB is utilized in the treatment session suggests that
BFB is not labour or resource intensive. The basic BFB unit is
cheap ($2000) and durable. Finally the fact that therapy can
be applied at home, by patient or family, lessens the drain on
a national health budget.
THE EFFICACY OF BIOFEEDBACK IN STROKE
The initial case reports and small uncontrolled series on
BFB therapy in stroke were quite enthusiastic, often citing
rather dramatic and remarkable recoveries from what
were thought to be permanent disabilities. Case series
described instances where patients, thought to have long
since plateaued in their rehabilitation course, discarded
 JOURNAL OF THE ROYAL SOCIETY OF MEDICINE
their cumbersome foot braces and walked away free14.
However, as noted previously, subsequent controlled
series were more mixed in their conclusions. There are
five potential sources to explain disparate conclusions in
clinical trials. These include differences in the treatment,
populations exposed, outcome measures, random varia-
tion, and systematic variation also known as bias. It is
instructive to examine how these facets might relate to
the BFB studies.
Treatment
There is no standard way to admninister BFB in the
treatment of the paresis and spasticity of stroke patients22.
By nature the treatment requires ad hoc decisions by the
therapist regarding electrode placement and joint position-
ing, based on the perceived status of the abnormal
physiology. The way traditional physiotherapy is inte-
grated into the BFB training depends upon the attitude
and skill of the therapist regarding both modalities. The
individual trial can specify the frequency, duration, and
total number of training sessions, though these parameters
will obviously vary between trials. Additionally the way
the feedback is provided to the patient (continuous versus
bandwidth) can also affect the success of motor learning24.
Finally the use of purposeful activity (walking) versus
undifferentiated activity (range of motion of the ankle
joint at rest) has been shown to influence motor
training25. This situation is in stark contrast to a
chemotherapy trial where a standard dosage schedule of
a specific drug can be uniformly applied, and differences
across trials can be more easily scrutinized28.
Patient population
This is perhaps the greatest potential source of heterogeneity
among trials. Wolfe has enumerated the critical aspects of
stroke patients which are of concern in this regard6. These
include the time post-stroke of treatment application, since
less and less spontaneous recovery can be expected after 3
months29. If treatment and control groups are not evenly
matched for this parameter, an obvious bias can therefore be
introduced. The actual territory and extent of the stroke
needs to be considered. For example there is some evidence
that patients with middle cerebral artery occlusion fare less
well in rehabilitation. Only one study has attempted to
account for this variable in its planning30. Associated
neurological features such as density of paresis, dominant
versus non-dominant side involvement, proprioceptive loss,
aphasia and cognitive deficits can impact on the success of
any rehabilitation programme. Age, associated affective
disorder, and motivational factors are additional features to
consider in this regard. Exactly how these factors relate to
success in BFB for stroke is unclear and controversial31'32. In
Volume 90 January 1997
these cited studies Wolfe has presented some evidence that
time post-stroke did not influence the likelihood of success
with BFB, at least for the lower extremity. As noted above,
the development of a proper staging modality could greatly
facilitate research in this area.
Outcome measures
The endpoints utilized in the various clinical trials on BFB
therapy in stroke have ranged from the simple quantification
of the integrated EMG signal33 to the performance of
complex tasks such as drinking from a glass25. The
arguments regarding what constitutes an appropriate
endpoint in stroke rehabilitation are complex and have been
reviewed by Ernst34. While it seems obvious that a decrease
in the level of disability (as per Duncan's scheme) would be
the ultimate goal of any therapy, such measurements are not
easy to obtain in a valid and reliable manner35'36.
Additionally, they are reported in an ordinal or discontin-
uous integral fashion and thus do not lend themselves to
parametric statistical analysis. A frequently utilized end-
point, range of motion, is considered critical to the recovery
of joint usage37, though not representing, of itself, a
'function'. It has the statistical advantage of measurement on
a continuous numerical scale, but has been criticized as
lacking in reproducibility35'36. Wolfe has suggested the use
of speed, distance and force as endpoints that circumvent
such concerns6. When assessing the overall efficacy of BFB in
stroke rehabilitation, the observer must be aware of these
considerations.
Random variability
All of the stroke-related BFB trials have been small. Even if
BFB had a specific positive treatment effect, the results of
individual small trials might point in different directions.
Such differences are accountable by measurement error or
day-to-day biological variability in the subjects. Stroke
patients are noted to display such random variability in the
degree of functional impairment exhibited6. In general the
small size of individual trials in the face of such variability
would tend to widen the confidence intervals around a
calculated positive treatment effect, and hence reduce the
level of statistical significance. This phenomenon, frequently
designated as type 2 error, is explored in the quantitative
overview or meta-analytic technique discussed later in this
paper.
Systematic error
Multiple forms ofbias which can affect the validity of a clinical
trial have been well described38. Most ofthese relate to the fact
that patients, planners and observers may have a vested
interest in a particular outcome. The double-blind randomized 35
 JOURNAL OF THE ROYAL SOCIETY OF MEDICINE
controlled clinical trial is the study structure of choice to
minimize such bias. No trials are of this arcadian design,
however, on BFB therapy of stroke. In fact of the trials
completed from 1960-1982 and critiqued by Wolfe, only
42% contained a control group at all6. He does acknowledge
the tremendous operational difficulties of carrying out a
double-blind study on this subject in the rehabilitation setting.
Furthermore, most of the controlled trials do not even utilize
sham treatment for the control group, decreasing the strength
of design and clouding the issue of whether BFB, even if it is
'effective', works through a specific neurophysiological
mechanism or merely provides a nonspecific motivating
influence, engendered in part by the seductive influence of
'high tech' factors22. As might be expected, the randomized
controlled trials in general have been less enthusiastic in their
reporting of the efficacy of BFB.
Meta-analyses of BFB therapy for stroke
Just as the double-blind randomized controlled trial is the
paradigm for the ideal study, a well conducted meta-analysis
embodies the essential elements of the 'state - of the art'
review article3940. This discipline, developed initially for
application in the social sciences, was first applied in the
medical sciences by Chalmers to address the avalanche of
(often conflicting) clinical trial data. Its strength lies in the
ability to assess the significance of ostensibly small but
potentially important clinical results demonstrated in the
individual trials, by the pooling of outcomes. The fact that
randomized trials in general are too small to demonstrate
conclusively such benefits has been the impetus behind the
formation of the Cochrane Collaboration41. The question of
validity is approached by systematically including only high
quality studies, or at least considering the effect on the
analysis of including trials of lesser merit. Quality in this
sense generally refers to the extent of proper randomization
and blinding.
In the last 2 years, three meta-analyses on the efficacy of
BFB therapy in stroke rehabilitation have been published. By
critically examining the methodology and conclusions of
these papers, one can hope better to assess the efficacy
question as well as generate hypotheses regarding modulat-
ing factors. Table 1 summarizes some salient features of
these three analyses.
It is notable that, even though all three studies set out to
ask the same research question, there was little overlap in
the included trials. Only one study was used in all three
analyses. The greatest number of trials common to any two
analyses was three. This lack of congruence was related to:
(1) endpoint(s) chosen; (2) stringency of validity criteria for
a study to be included; (3) comprehensiveness of the search
36 process.
Volume 90 Jan uary 1 997
The Schleenbaker analysis looked at functional outcomes
rather than more basic physiological measures, with the
rationale that only such endpoints are of ultimate significance
in the natural history of stroke recovery. In doing so they
combined different endpoints all representative of 'function'
using the effect size method. The study by Glanz et al. utilized
the most consistent endpoint, i.e. change in range of motion.
The rationale is that a meta-analysis is on strongest grounds
when the critical elements of the included studies (patients,
exposures, outcome variables) are as similar as possible39. The
Moreland analysis dichotomized its selected studies into
groups representative of functional and more basic physio-
logical constructs (termed 'impairment'). The analysis was
limited to outcomes expressed for upper extremity alone,
while the former two looked at both upper and lower
extremity endpoints. As noted, the upper extremity is more
recalcitrant to rehabilitation efforts in stroke.
The concept of validity in a clinical trial relates to how
well the design translates into an objective study of the
research question. This signifies the avoidance of systematic
error or bias. The methods of Oxman42 or Chalmers43 can
be used to assess the individual validity of studies to be
included in a meta-analysis. In this regard the Moreland
study is the most rigorous of the three, selectively
excluding studies which did not measure up to a
predetermined matrix of criteria, and also measuring the
agreement amongst the authors of their study regarding the
parameters of merit. The Glanz study measured quality
according to the method of Chalmers. The scores were all
within a low range and studies were not excluded or
otherwise weighted on this basis. (A previous paper which
looked at this system found no correlation between
outcomes of randomized controlled trials and the quality
score. A possible reason proffered was that the strength of
the randomized controlled design supersedes the impor-
tance of the detailed quality scoring system44.) The
Schleenbaker analysis was least rigorous in this regard, in
that non-randomized trials were included45'6. Chalmers
had previously shown that non-randomization tends to bias
results toward the treatment group47. It is notable that the
studies with the largest effect sizes in the Schleenbaker
analysis were non-randomized.
The difficulty in retrieving all relevant trials for a meta-
analysis has been well documented. In one study only 29%
of the trials were elicited by use of a standard computerized
data base48. The Moreland search protocol seems the most
exhaustive in this regard, even including a search for
unpublished studies.
The validity of a meta-analysis, as with an individual
trial, relates to the adherence to a predetermined protocol
which is designed to minimize bias. Again the Moreland
study, as far as can be discerned from the individual papers,
seems mos ngorous.
 JOURNAL OF THE ROYAL SOCIETY OF MEDICINE Volume 90 J a n u a ry 1 9 9 7

Table 1 Meta-analyses of Biofeedback therapy in stroke

Total
Study number Discussion Included
Literature selection Strength of studies! Type 2 of clinical study
Study search criteria protocol Endpoints patients Results error relevance overlap*

Schleen- ++ + + Tests of function Effect size NA + 0.50

baker (upper and lower 8/192 0.81(95%CI:0.5, 1.12)
extremity combined)
Glanz + ++ ++ Joint range of motion Effect size
upper extremity 3/62 2.3 (95%CI: -1.06, 0.73 + 0.50
5.66)
lower extremity 5/118 1 .5(95%CI: -0.59, 0.71
3.59)
Moreland +++ +++ +++ Odds ratio Not
Functional construct 5/135 2.16(95%CI:0.82, calcul- + 0.50
(upper extremity) 5.79) ated
Impairment construct 3/84 1.29(95%CI:0.43,
(upper extremity) 3.99)

*Proportion of included studies found in either of the other two meta-analyses

Regarding data analysis the Schleenbaker and Glanz
papers used the effect size method of Glass49. This is a
parametric technique, thus presupposing that the data are in
the form of continuous variables in a roughly normal
distribution. (Parametric analysis of discontinuous variables
may be acceptable only with very large sample sizes.) The
former analysis does incorporate some results with
discontinuous scales. Additionally, when more than one
outcome was available in a study for inclusion in the analysis,
it is not clear how the selection was made. Neither study
justified the assumption of normality in individual study data,
for example by commenting on the rough appearance of the
distributions. The Moreland study acknowledged the
difficulty of using a parametric technique given the above
concerns and therefore organized the data into a
dichotomous format based upon the presence of improve-
ment in the designated endpoint. This approach led to the
loss of some quantitative information, but subverted the
stated problem of combining discontinuous and somewhat
dissimilar variables, as well as concerns over the nature of
the distribution of the data set.
All three studies reported a pooled mean outcome in
favour of BFB therapy for stroke although only in the
Schleenbaker study was the result statistically significant at the
P <0.05 level, one-tailed. The mean effect sizes noted (0.81
for Schleenbaker and 1.50 for Glanz) would be considered
'large' by convention50. In fairness it should be noted,
however, that such qualitative statements about effect size
magnitudes and their resultant clinical significance are
controversial51. The pooled odds ratio for the Moreland study
(2 .2) was interpreted by the author, using the 'number needed
to treat' format of Sackett38, as being small: seven need to be
treated to help one patient. It is not clear, in view of the
potential impact of stroke disability on quality of life, that this
represents an unacceptable level of efficacy. To claim such
would probably require a degree of analysis beyond the scope
of this paper. The Glanz study did calculate the magnitude of
type 2 error, which quantifies the probability that a true
treatment effect was present but the sample size was too small
to demonstrate it. This was equal to 0.73 for the upper
extremity and 0.71 for the lower, which are in fact quite large.
In planning a clinical trial an investigator generally aims for a
type 2 error of 0. 10 or less. Moreland was unable to calculate a
type 2 error for technical reasons. One randomized controlled
trial has been reported since the publication of these meta-
analyses52. This was a positive study in favour of BFB, but as it
contained only 16 patients, we doubt if it would have
materially reduced the type 2 error ofthe Moreland and Glanz
studies.
CONCLUSION
Research efforts into limiting neuronal damage in the acute
phase of stroke are still at an embryonic stage, so the major
focus remains on rehabilitative efforts. Research and
organization of rehabilitation services including BFB have
been desultory, and these have been earmarked for
upgrading18. This is an area where small improvements,
such as a few extra degrees of range of motion in a paretic
ankle joint, can make a big difference in level of disability,
perhaps the difference between independence and institu-
tionalization. Although a broad-brush statement for the
efficacy of BFB therapy for stroke cannot be made at this
time, certain considerations merit emphasis. 37
 JOURNAL OF THE ROYAL SOCIETY OF MEDICINE
There is a rich tradition of basic neuroscience research
behind BFB therapy. This is complemented by applied
research, especially the efforts of Basmajian and Wolfe. That
randomized controlled trials were not as enthusiastically
supportive of efficacy as earlier case series or historically
controlled studies could be said of almost any treatment47.
The multiple layers of difficulty involved in stroke
rehabilitation research in general, and biofeedback in
particular, would make this a 'Gordian knot' for even the
most resolute of clinical trialists. Three quantitative
overviews recently published in an attempt to summarize
the scientific clinical experience of BFB in stroke have
arrived at different yet potentially consistent conclusions:
one with a statistically significant positive result; another
with a large mean effect size not statistically significant but
with a very large type 2 error; and a third, concentrating on
the more problematic upper extremity, with a positive
pooled result but not significant at the P<0.05 level. It is
certainly plausible that inadequate sample size and a failure
properly to stratify trial enrollees with reference to the
potential for neurorecovery can explain the lack of statistical
significance of the latter two meta-analyses. Furthermore,
most of the trials are old and their results alone do not
reflect subsequent technological advances and the learning
curve which accompanies them. The least that can be said
for BFB therapy in stroke rehabilitation is that it offers hope
for a proportion of those with this affliction, and that
controlled research in this area should be strongly
encouraged.
In summary, despite the burgeoning of clinical trials and
other scientific evidence there is paradoxically insufficient
evidence to guide the majority of clinical decisions. Even the
best planned studies rarely give enough information to apply
the overall conclusions to individual subsets of patients.
Given cost and resource considerations, it is unlikely that
such information will be forthcoming.53 As a recent
publication noted, clinicians will often have to make
decisions based upon '. . a considered review of the
totality of evidence'54.
Acknowledgments The authors wish to thank Jason
Glanz, BS and Edward Andrews, MPA for their assistance.
REFERENCES
1 Gersh MR. Electrotherapy in Rehabilitation. Philadelphia: FA Davis
Company, 1992
2 Adrian ED, Bronk DW. The discharge of impulses in motor nerve fibers.
J Physiol 1928;66:81-101
3 Andrews JM. Neuromuscular reeducation of the hemiplegic with the aid
of the electromyograph. Arch Phys Ed Rehabil 1964;45:530
4 Mims HW. Electromyography in clinical practice. South Med J
1 956;49:804
5 Marinacci AA, Hornande M. Electromyogram in neuromuscular
38 reeducation. Bull Neurol Soc 1960;25:57
Vol ume 90 January 1 997
6 Wolf SL. Electromyographic biofeedback applications to stroke
patients: a critical review. Phys Ther 1983;63:1448-55
7 DeWeerdt WJG, Harrison MA. The efficacy of electromyographic
feedback for stroke patients: a critical review of the main literature.
Physiotherapy 1986;72:108-18
8 Ince LP, Leon MS, Christidis D. EMG biofeedback with the upper
extremity: a critical review of experimental foundation of clinical
treatment of the disabled. Rehabil Psychol 1987;32:77-91
9 Schleenbaker RE, Mainous AG. Electromyobiofeedback for
neuromuscular reeducation in the hemiplegic stroke patient: a meta-
analysis. Arch Phys Med Rehabil 1993;74:1301-4
10 Moreland J, Thomson MA. Efficacy of electromyographic biofeedback
compared with conventional physical therapy for upper extremity
function following stroke: a research overview and meta-analysis. Phys
Ther 1994;74:534-47
11 Glanz M, Klawansky S, Stason W, et al. Biofeedback therapy in
poststroke rehabilitation: a meta-analysis of the randomized controlled
trials. Arch Phys Med Rehabil 1995;76:508-15
12 Klose KJ, Needham BM, Schmidt D, Broton JG, Green BA. An
assessment of the contribution of electromyographic biofeedback as an
adjunct therapy in the physical training of spinal cord injured persons.
Arch Phys Med Rehabil 1993;74:435-56
13 Cleeland CS. Biofeedback and other behavioral techniques in the
treatment of disorders of voluntary movement. In: Basmajian JV ed.
Biofeedback: Principals and Practice For Clinicians, 3rd edn. Baltimore:
Williams & Wilkins, 1989
14 Basmajian JV. Biofeedback in rehabilitation: a review of principles and
practices. Arch Phys Med Rehabil 1981;62:469-75
15 Bamford J, Sandercock P, Dennis M, et al. A prospective study of
acute cerebrovascular disease in the community: the Oxfordshire
community stroke project, 1981-86. 1. Methodology, demography
and incident cases of first ever stroke. J Neurol Neurosurg Psychiat
1988;51 :1373-80
16 Oxfordshire Community Stroke Project. Incidence of stroke in
Oxfordshire: first year's experience with a community stroke
register. BMJ 1983;287:713-17
17 Office of Health Economics Statement. The Treatment of Stroke. London:
OHE, 1988
18 King's Fund Consensus Conference. Treatment of stroke. BMJ
1988;297: 126-28
19 Dennis M, Warlow C. Strategy for stroke. BMJ 1991;303:636-8
20 Basmajian J. Research foundations of EMG biofeedback in rabilitation.
Biofeedback Seff-Reg 1988;13:275-97
21 Duncan PW. Stroke disability. Phys Ther 1994;74:399-407
22 Marzuk PM. Biofeedback for neuromuscular disorders. Ann Intern Med
1985; 102:854-8
23 Wolfe SI, Edwards DI, Shutter LA. Concurrent assessment of muscle
activity (CAMA): a precedural approach to assess treatment goals. Phys
Ther 1986;66:218-24
24 Winsteen CJ. Knowledge of results and motor learning-implications
for physical therapy. Phys Ther 1991;71:140-9
25 Wissel J, Ebersbach G, Gutjahr L, Dahlke F. Treating chronic
hemiparesis with modified biofeedback. Arch Phys Med Rehabil
1989;70:612-17
26 Finkler SA. The distinction between cost and charges. Ann Intern Med
1982;96: 102-9
27 Basmajian J. Biofeedback in rehabilitation: a review of principles and
practices. Arch Phys Med Rehabil 1981 ;62:469-75
28 Dennis M, Langhorne P. So stroke units save lives: where do we go
from here? BMJ 1994;309:1273-7
29 Hewer RL. Rehabilitation after stroke. QJ Med I1990;76:659-74
30 Shahani BY, Connors L, Mohr JP. Electromyographic audiovisual
feedback training effect on the motor performance in patients with
 JOURNAL OF THE ROYAL SOCIETY OF MEDICINE
lesions of the central nervous system. Arch Phys Med Rehabil
1977;58:519
31 Wolf SL, Baker MP, Kelly KL. EMG biofeedback in stroke: effect of
patient characteristics. Arch Phys Med Rehabil 1979;60:96-102
32 Wolf SL, Baker MP, Kelly KL. EMG biofeedback in stroke: 1-year
follow up on the effect of patient characteristics. Arch Phys Med Rehabil
1980,61:351-5
33 Hurd WW, Pegram V, Nepomuceno C. Comparison of actual and
simulated biofeedback in the treatment of hemiplegic patients. AmJ Phys
Med 1980;59:73-82
34 Ernst EE. A review of stroke rehabilitation and physiotherapy. Stroke
1990:21:1081-5
35 Fernando CK, Basmajian JV. Biofeedback in physical medicine and
rehabilitation. Biofeedback Se!f-Reg 1978;3:435-55
36 Keefe FJ, Surwit RS. Electromyographic biofeedback: behavioral
treatment of neuromuscular disorders. J Behav Med 1978; 1: 13-24
37 Duncan PW, Badke MB. Stroke Rehabilitation. Chicago: Year Book
Medical Publishers Inc, 1987
38 Sackett DL, Haynes RB, Guyatt G, Tugwell P. Clinical Epidemiology: a
Basic Science For Clinicians, 2nd edn. Boston: Little Brown & Co, 1991
39 Oxman AD, Guyatt GH. Guidelines for reading literature reviews. Can
Med Assoc J 1988; 138:697-703
40 Mulrow CD. The medical review article: state of the science. Ann Intern
Med 1987; 106:485-8
41 The Cochrane Collaboration. Oxford, England: UK Cochrane Centre,
National Health Service Research and Development Programme,
1994
42 Oxman AD, Sackett DL, Guyatt GH, for the Evidence-Based Medicine
Working Group. Users' guide to the medical literature, I: how to get
started. JAMA 1993;270:2093-5
Volume 90 January 19971
43 Chalmers TC. A method of assessing the quality of randomized control
trials. Control Clin Trials 1981 ;2:31-49
44 Emerson JD, Burdick E, Hoaglin DC, Mosteller F, Chalmers TC. An
empirical study of the possible relation of treatment differences to
quality scores in controlled randomized clinical trials. Control Clin Trials
1990;11:339-52
45 Wolf SL, Binder-Macleod SA. Electromyographic biofeedback
applications to the hemiplegic patient: changes in upper extremity
neuromuscular and functional status. Phys Ther 1983;63:1393-403
46 Wolf SL, Binder-Macleod SA. Electromyographic biofeedback
applications to the hemiplegic patient: changes in lower extremity
neuromuscular and functional status. Phys Ther 1983;63:1404-14
47 Chalmers TC, Celano P, Sacks HS, Smith H, Jr. Bias in treatment
assignment in controlled clinical trials. N EnglJ Med 1983;309:1358-61
48 Dickersin K, Hewitt P, Muth L, et al. Perusing the literature:
comparison of MEDLINE searching with a perinatal trials database.
Control Clin Trials 1985;6:306-17
49 Glass GV, McGaw B, Smith ML. Meta-analysis in social research. Beverly
Hills, CA: Sage, 1981
50 Rossi P, Wright S. Evaluation research: an assessment of theory,
practice and politics. Eval Q 1977;1 :5-52
51 Pillemer DB. Conceptual issues in research synthesis. J Special Ed
1984;18:27-40
52 Intiso D, Santilli V, Grasso MG, Rossi R, Caruso I. Rehabilitation of
walking with electromyographic biofeedback in foot-drop after stroke.
Stroke 1994;25: 1189-92
53 McDonald CJ. Medical heuristics: the silent adjudicators of clinical
practice. Ann Intern Med 1996;124:56-62
54 Borzak S, Ridker P. Discordance between meta-analyses and large scale
randomized controlled trials. Ann Intern Med 1995;123:873-7
39