Manually controlled instrumented spasticity assessments: a systematic review of psychometric properties


  • Lynn Bar-On,

    Corresponding author
    1. Clinical Motion Analysis Laboratory, University Hospital Leuven, Belgium
    2. KU Leuven Department of Rehabilitation Sciences, Belgium
    • Correspondence to Lynn Bar-On, Clinical Motion Analysis Laboratory, University Hospital Leuven, Weligerveld 1, 3212 Pellenberg, Belgium. E-mail:

    Search for more papers by this author
  • Erwin Aertbeliën,

    1. KU Leuven Department of Mechanical Engineering, Belgium
    Search for more papers by this author
  • Guy Molenaers,

    1. Clinical Motion Analysis Laboratory, University Hospital Leuven, Belgium
    2. KU Leuven Department of Development and Regeneration, Belgium
    3. Department of Orthopedics, University Hospital Leuven, Belgium
    Search for more papers by this author
  • Bernard Dan,

    1. Department of Neurology, University Children's Hospital Queen Fabiola, Université Libre de Bruxelles, Brussels, Belgium
    Search for more papers by this author
  • Kaat Desloovere

    1. Clinical Motion Analysis Laboratory, University Hospital Leuven, Belgium
    2. KU Leuven Department of Rehabilitation Sciences, Belgium
    Search for more papers by this author



The first aim of this study was to systematically review and critically assess manually controlled instrumented spasticity assessment methods that combine multidimensional signals. The second aim was to extract a set of quantified parameters that are psychometrically sound to assess spasticity in a clinical setting.


Electronic databases were searched to identify studies that assessed spasticity by simultaneously collecting electrophysiological and biomechanical signals during manually controlled passive muscle stretches. Two independent reviewers critically assessed the methodological quality of the psychometric properties of the included studies using the COSMIN guidelines.


Fifteen studies with instrumented spasticity assessments met all inclusion criteria. Parameters that integrated electrophysiological signals with joint movement characteristics were best able to quantify spasticity. There were conflicting results regarding biomechanical-based parameters that quantify the resistance to passive stretch. Few methods have been assessed for all psychometric properties. In particular, further information on absolute reliability and responsiveness for more muscles is needed.


Further research is required to determine the correct parameters for quantifying spasticity based on integration of signals, which especially focuses on distinguishing the neural from non-neural contributions to increased joint torque. These parameters should undergo more rigorous exploration to establish their psychometric properties for use in a clinical environment.




Minimally important change


Modified Tardieu Scale


Modified Ashworth Scale


Root mean square electromyography


Range of motion


Smallest detectable change


Surface electromyography


Stretch reflex threshold


Upper motor neuron

Excessive and uncontrolled spasticity causes pain, limits functional recovery, and is thought to cause secondary complications such as contractures and bone deformities.[1] It appears in conditions such as upper motor neuron (UMN) syndrome, and is the most common neurological feature in persons with cerebral palsy (CP). Despite the impact of spasticity and the many therapeutic paradigms aimed at treating it, there are few clinically suitable, reliable methods for its assessment. One reason for the lack of consensus on the assessment method is the absence of a commonly accepted definition for spasticity.[2]

In 1954, Tardieu et al.[3] described the phenomenon of a ‘spastic catch’ as ‘a sudden reactive resistance to a fast passive stretch of a spastic muscle’. In 1980, Lance[4] was the first to define spasticity as ‘a velocity-dependent increase in tonic stretch reflexes with exaggerated tendon jerks, resulting from hyper-excitability’. Although Lance's definition is most commonly cited, in routine clinical practice it is nearly impossible to distinguish this definition of spasticity from other positive symptoms of the UMN syndrome. For example, other reflex mechanisms (e.g. cutaneous or nociceptive) could also contribute to increased muscle activation, and are difficult to distinguish from the proprioceptive reflex mechanisms described by Lance.[5]

Sanger et al.[6] defined spasticity as ‘resistance to an externally imposed movement that rises with increasing speed of stretch or rises rapidly above a threshold speed or joint angle’. Here too, however, distinguishing the resistance caused by pathological muscle activation due to a hyperactive stretch reflex from the increased resistance due to passive stiffness is clinically very challenging. Non-neural muscle and tendon alterations also contribute to reactive resistance, especially in persons with UMN syndrome.[7] Changes in the viscoelastic properties of these structures will determine both the stiffness and the velocity dependence of a movement. Thus, it appears that ‘observed’ spasticity encompasses multiple phenomena and is not a single pathophysiological entity. In line with this finding, the SPASM consortium (established to develop standardized measures of spasticity) introduced a broader definition for spasticity: ‘a disordered sensorimotor control, resulting from an upper motor neuron lesion, presenting as intermittent or sustained involuntary activation of muscles’.[5]

The complexity of distinguishing spasticity from other positive symptoms highlights the challenges in developing suitable measurement systems. Firstly, a distinction has to be made between measurements that assess spasticity in a relaxed muscle, or during activity. In most clinical settings, spasticity is measured using subjective, easy-to-use ordinal scales that assess the level of resistance felt by the examiner during a passive muscle stretch. Examples of such scales include the Modified Ashworth Scale (MAS)[8] and the Modified Tardieu Scale (MTS).[9] The MTS is considered more valid for the assessment of spasticity as defined by Lance[4] as the resistance is compared during stretches at different velocities. However, lack of standardization of stretch velocity and the subjective nature of both scales has resulted in poor interrater reliability[10, 11] and, for the MTS, inaccuracies in determining the correct catch angle.[12] In light of the above-mentioned difficulty of isolating spasticity, these tests also greatly oversimplify the phenomenon. It is therefore not surprising that many studies have shown poor correlations between the clinical measures (MAS, MTS) and objective indicators of pathologically increased muscle activity during passive stretch.[7, 13-15] For example, some individuals, who have been found to have spasticity during a clinical examination as indicated by increased resistance to passive stretch, lack any signs of hyperactive H-reflexes.[16] In these cases, increased resistance to passive stretch may have been as a result of non-neural causes.

Therefore, it is now acknowledged that quantified, instrumented methods should be used to provide a more accurate and valid evaluation of spasticity.[17] In 2005, spasticity assessment measures were comprehensively summarized by the SPASM consortium in three review articles.[18-20] These reviews identified and categorized a large number of non-invasive, instrumented applications for quantitative spasticity assessment into biomechanical and neurophysiological methods.[18, 19] They concluded that these methods are complementary, and should be used simultaneously to differentiate sufficiently between neural and non-neural causes of increased resistance.[20] Biomechanical devices record joint angular characteristics and/or resistance around a joint during passive stretching.[18] They include, for example, motor-driven or hand-held dynamometers. Neurophysiological methods measure muscle activity using, for example, electromyography (EMG) during passive movement or nerve stimulation.[19] Furthermore, the consortium stressed that collecting experimental data in a highly technical and controlled environment would greatly improve the modelling of the complex pathophysiology. However, combining these recommendations in view of a clinical application requires some compromise. A suitable method should, on the one hand, be more valid and reliable than the current clinical tests, and on the other hand, remain clinically feasible in different patient populations, including children. For example, while some motor-driven isokinetic devices that measure limb resistance to passive movement have great reliability, due to the limb being moved at a controlled velocity,[21-24] they are often bulky and difficult to apply to children in high-velocity stretches.[20] In addition, a stretch reflex may be easier to excite by a transient acceleration, which is robotically more difficult to apply.[25] A manually controlled displacement method offers a clinically applicable alternative.[26-28] However, to ensure accuracy, manually controlled displacement methods must follow standardized protocols, and the psychometric properties need to be defined before they can be used in clinical practice.[20] A recent review of spasticity assessments for children and adolescents with CP highlighted the insufficient psychometric soundness of spasticity evaluation tools.[29] However, this review did not emphasize the need to integrate biomechanical and electrophysiological signals, as is recommended for valid spasticity assessment.[20] Therefore, the researchers' conclusion that using electrophysiological methods to assess spasticity demonstrates the most promising results in terms of reliability and discriminant validity, may have been misleading.

The aim of the current study was twofold. First, we wanted to systematically and critically assess clinically applicable spasticity measurement methods that adhere to the recommendations of the SPASM consortium.[20] Following these recommendations, any developed spasticity measurement method should (1) be able to make measurements at variable velocities of displacement; (2) incorporate simultaneous recording of EMG and torque; and (3) include a clearly defined protocol. To ensure a similar conceptualization of spasticity across reviewed articles (i.e. the definition of spasticity as offered by Lance[4]), only measurements during passive conditions were to be included. Secondly, we aimed to extract a set of quantitative parameters that measure spasticity based on the reviewed articles.


Search strategy

A single reviewer (LB) performed a web-based search for relevant literature using the following electronic databases: Science Direct (, MEDLINE (, and EMBASE ( Only full-paper articles published in English, in peer-reviewed journals, and performed on human participants, were included. Keywords included (‘All fields’ and MeSH): (1) spasticity, (2) tone, (3) cerebral palsy, (4) stroke, (5) spinal cord injury, (6) upper motor neuron, (7) measure, (8) evaluation, and (9) assessment. The following word combinations were implemented: 1 or 2; AND 3 or 4 or 5 or 6; AND 7 or 8 or 9.

Study selection

Two reviewers (LB and RDB) independently selected the studies for inclusion in the review. Firstly, titles and abstracts were screened for eligibility. Secondly, the full text of potentially relevant papers was read to ascertain whether the study met all selection criteria, that is the article had to describe a method to quantitatively assess spasticity by recording both biomechanical and electrophysiological signals during manually applied passive muscle stretches. Studies were excluded if the method (1) only assessed spasticity based on subjective measurements, including Ashworth- and Tardieu-like scales;[11] (2) applied only a motor-controlled device or a pendulum-like test[30] to stretch the muscle; (3) was limited to collecting either biomechanical or electrophysiological signals; (4) applied a passive stretch at only one velocity; or (5) assessed spasticity during function or active movements. Use of the tendon and Hoffmann reflexes as a means to assess spasticity has been extensively studied.[19] However, their clinical applicability and relevance is limited. Therefore, studies that also applied excitation of these reflexes or electro-stimulation as a neurophysiological means to assess spasticity, were excluded from the current review. Finally, in those cases where more than one article was published by the same research group with the same methodology, the most recent publication was selected for review, unless older articles investigated different psychometric properties. The bibliographic details of excluded studies were listed and reasons for exclusion noted. Any discrepancies regarding final selection were resolved by consensus, and if necessary, by consulting a third reviewer (KD).

Data extraction and quality assessment

Selected studies were read by two independent reviewers (LB and KD) to extract information on study populations, methodology, study design, outcome parameters, results, and conclusions. Both reviewers independently evaluated the quality of the psychometric properties of the described method using the COSMIN checklist.[31] The COSMIN checklist consists of 12 domains. It offers common terminology and definitions for psychometric properties.[32] For each study included in the current review, only domains relevant to the investigated psychometric properties were checked. The relevance of each domain and the interpretation with respect to spasticity measurements were discussed before commencing. Six domains were considered relevant (see Table SIA, online supporting information): two were used to determine whether a study met the methodological quality criteria in terms of reliability and measurement error; two assessed the method's content and construct validity (including hypothesis testing); one assessed the responsiveness of the method; and lastly, one determined the interpretability. Generalizability was determined for each of the previous domains. The following domains from the COSMIN checklist were not considered relevant for spasticity assessment: Item Response Theory, internal consistency, structural validity, cross-cultural validity, and criterion validity. Reasons for not assessing these properties are described in Table SIB (online supporting information).

Each of the six domains (and generalizability) were rated by both assessors independently on a four-point scale according to the COSMIN guidelines.[33] ‘Excellent’ quality was assigned if all relevant COSMIN items within a domain were scored as adequate. ‘Good’ quality was assigned to those studies that lacked some aspects, although it could still be assumed that the items were acceptable. ‘Fair’ quality was assigned if the measurement property was underrepresented or explored in a moderate sample size, or if there were other minor flaws in the design or statistical analyses. ‘Poor’ quality was assigned if there were major flaws in the design or statistical analyses. Finally, in each article the statistical findings per domain were rated according to quality criteria provided by Terwee et al.[34] as positive, indeterminate, negative, or no information available (Table SIA, online supporting information). For each domain, all items, resulting scores, and statistical ratings were then discussed by the reviewers and any discrepancies resolved by consensus.


A flow chart of the selection process is shown in Figure 1. After filtering the databases on keywords and screening relevant titles and abstracts, 158 potential full-text articles were found. Further examination of these full-text articles revealed that 33 papers did not apply an objective measurement method, 39 used a robot to displace the limb, 27 applied electro-stimulation, and 38 articles measured either a biomechanical or an electrophysiological signal in isolation. One article measured both signals, but did not use the biomechanical parameters as a means to quantify spasticity. One article was excluded as the limb was displaced at only one velocity. Lastly, three articles were excluded as their methodology was reported in more recent studies by the same research groups. Therefore, 15 studies were identified as meeting all the inclusion criteria. The data extracted from these are summarized in Tables 1-4 and in Tables SII and SIII (online supporting information).

Figure 1.

Flow chart of article search and selection strategy.

Table 1. Characteristics of included studies: study populations and protocol design
First authorStudy populationProtocol design
Participants n Age (y)Diagnostic detailsFunctional levelMain selection criteriaAgonists tested for spasticityAntagonistsInstrumentsType and trajectory of stretchStretch velocitiesNumber of repetitions per velocityRest period between repetitionsComparator tests
  1. AS, Ashworth Scale; MAS, Modified Ashworth Scale; GMFCS, Gross Motor Function Classification System; MACS, Manual Ability Classification System; ARAT, Action Research Arm Test; NR, not relevant; CP, cerebral palsy; SDR, selective dorsal rhizotomy.

Lamontagne[43]Spinal cord injury9Mean 41, SD 111–5y post injury; complete (n=8); incomplete (n=2); traumatic (n=8); ischaemic (n=1)C6 (n=1); T5–6 (n=1); T5 (n=3); T7 (n=1); T8 (n=1); T10 (n=2)MAS score ≥1; no fixed contractures or deformities in lower limbs; no history of fracture or thrombophlebitisSoleusTibialis anteriorHand-held dynamometer; electrogoniometer and potentiometer; surface electromyography; metronomeRamp movement from −35° plantarflexion to 5° dorsiflexionLow-velocity average: 3.3°/s (SD 3.4°/s)51sKin-Kom isokinetic dynamometer
High-velocity average: 311.1°/s (SD 380°/s)
Wu[27]CP10Mean 10, SD 31 quadriplegia; 6 right hemiplegia; 3 left hemiplegia. Movement disorder (spasticity, dystonia, ataxia)GMFCS level: I (n=2); II (n=3); III (n=2); IV (n=2); V (n=1) MACS level: II (n=5); III (n=4); V (n=1)Not mentionedBiceps brachiiTriceps brachiiTorque sensor, potentiometer, surface electromyographyRamp movement from full elbow flexion to full elbow extension30, 90, 180, 270°/s1 at 30°/s, 3 at 90°/s, 180°/s, and 270°/s1minMAS score
Typical development10Mean 10, SD 3NRNR
Voerman[35]Stroke12Mean 57, SD 9First stroke: 9 left hemiplegia; 3 right hemiplegiaARAT: (scored for six participants) 0 (n=3); 2 (n=1); 5 (n=1); 6 (n=1)MAS score 1–3 in wrist and finger flexors, >20° pain-free wrist extension, 5° active wrist flexion, able to communicate, no history of serious medical, psychological or cognitive impairmentWrist flexorsWrist extensorsHand-held dynamometer, potentiometer, surface electromyography, electronic metronomeSinusoidal wrist movement from neutral to extension and back to neutral30, 60, 90 cycles/min (180, 360, 540°/s)5–7NoneMAS score; ARAT; wrist rig
Healthy participants11Mean 57, SD 8NRNRNot mentioned
Van der Salm[42]Spinal cord injury9Mean 35, SD 7Minimum 6mo after injuryC5 (n=1), C6 (n=2), C6–7 (n=1), T4 (n=1), T5 (n=1), T4–5 (n=1), T8 (n=1), T11 (n=1)MAS score ≥1, >18y, absence of voluntary movements in triceps surae, tibialis anterior can contract using electrical stimulation, no fixed ankle contractureTriceps suraeNoneCalibrated strain gauge dynanometer, potentiometer, gyroscope, surface electromyographyRamp movement across full ankle range of motionRandom from 30 to 150°/s30–405sMAS score
Bar-On[45]CP28Mean 10, SD 5Spastic CP; 3 right hemiplegia; 5 left hemiplegia; 19 diplegia; 1 quadriplegiaGMFCS level: I (n=10), II (n=12), III (n=5), IV (n=1)Age 5–18y; spastic CP; no ankle or knee contractures, no previous orthopaedic surgery, no intrathecal baclofen pump; no SDR; no botulinum toxin A in last 6moLateral gastrocnemius; Medial hamstringsTibialis anterior; Rectus femorisTorque/force load-cell; inertial measurement units, surface electromyographyRamp movement across full ankle or knee range of motion

Average low velocity: gastrocnemius 22.5°/s (SD 7.2°/s); hamstrings 35.2°/s (SD 7.5°/s)

Average high velocity: gastrocnemius 202.1°/s (SD 54.2°/s); hamstrings 317.7°/s (SD 47.7°/s)

47sMAS score
Typical development10Mean 11, SD 6NRNRNot mentioned
Bar-On[46]CP31Mean 9, SD 2Spastic CP; 6 right hemiplegia; 5 left hemiplegia; 17 diplegia, 1 triplegia; 2 quadriplegiaGMFCS level: I (n=12), II (n=12), III (n=6), IV (n=1)Age 3–18y; spastic CP; no ankle or knee contractures, no previous orthopaedic surgery, no intrathecal baclofen pump; no SDRMedial hamstringsRectus femorisTorque/force load-cell, inertial measurement units, surface electromyographyRamp movement across full knee range of motionAverage low velocity: 75.48°/s (SD 17.31°/s); Average high velocity: 288.44°/s (SD 54.11°/s)47sMAS score; Modified Tardieu Scale score
Pandyan[13]Stroke14Median 61, interquartile range 52–63Median 48mo post stroke (interquartile range 32–60), 6 left hemiplegia; 8 right hemiplegiaNot mentionedClinical diagnosis of spasticity, capable of providing written informed consentBiceps brachiiTriceps brachiiForce transducer, electrogoniometer, surface electromyographyRamp movement across full elbow range of motion with humerus abducted to 90°Slow, fast (median difference: 34°/s, interquartile range 20–46°/s)1 slow stretch, 1 fast stretchNot mentionedMAS score
Lebiedowska[47]Stroke3Mean 65, SD 8Not mentionedNot mentionedNot mentionedMedial hamstrings, Rectus femorisRectus femoris, Medial hamstringsHand-held stain gauge dynanonmeter, potentiometer, surface electromyographyRamp movement from neutral to knee extension and from neutral to 142° knee flexion0.2–1.5 rad/s (11.5–540°/s)Several (not reported in detail)Not mentionedNone
Adults with CP4Mean 35, SD 123 diplegia; 1 right hemiplegia
Children with CP13Mean 13, SD 410 diplegia; 2 right hemiplegia; 1 left hemiplegiaNR
Healthy participants19Mean 13, SD 8
Fleuren[7]Stroke18Mean 57, SD 13–16Not mentionedNot mentionedSelf-reported spasticity, no contractures, no severe pain, able to understand simple commandsBiceps brachii, Brachioradialis, Rectus femoris, Vastus lateralisNoneHand-held dynanometer, electrogoniometer, surface electromyographyRamp movement across full elbow and knee range of motion (subject side lying)Slow, fast (median velocity: 76.6°/s for elbow flexors, 85.2°/s for knee extensors)One at slow velocity, two at fast velocityNot mentionedAS score
Spinal cord injury2
Neuromuscular disease4
Malhotra[37]Stroke100Median 74, inter quartile range 43–91Average of 3wks post first stroke (range 1–6wks), 52 right hemiplegia, 48 left hemiplegiaARAT 0 (n=97), 1 (n=2), 3 (n=1)Within 6wks of first stroke, score of 0 on grasp section of ARAT, no wrist contractures, no major illnessLong wrist flexorsLong wrist extensorsForce transducer, electrogoniometer, surface electromyographyRamp movement across full wrist range of motionSlow, fast (mean difference between velocities: 87°/s, SD 36°/s, range 10–190°/s)Two at each velocityNRMAS score, ARAT, Brunnstrom Index
Chen[38]Stroke10Mean 57, SD 12Average of 38mo (SD 27mo) post stroke, 3 right hemiplegia; 7 left hemiplegiaBrunnstrom Index: III (n=4); IV (n=2); V (n=4)At least 6mo post stroke, no elbow contractures, no severe cognitive or affective dysfunction, Brunnstrom Index ≥IIIBiceps brachiiTriceps brachiiAir bags, differential pressure sensor, angular velocity sensor, surface electromyographySinusoidal movement from 120˚ to 60˚ elbow flexion1/3, 1/2, 1, 1.5 Hz (120, 180, 360, 540°/s)Not mentioned≥30sMAS score
Turk[39]Stroke12Mean 62, SD 12Average of 6y (SD 4y) post stroke, 4 right hemiplegia; 8 left hemiplegiaMean ARAT 18.8 ± 11.5At least 3mo post stroke, some active wrist movement, no wrist contractures, no neglect or major illnessFlexor carpi ulnaris, Flexor carpi radialisExtensor carpi radialis longusStrain gauges (force sensor), potentiometer, surface electromyographySinusoidal movement across full wrist range of motionSlow: 0.04 or 0.08Hz (14.4 or 28.8°/s)2 at slow velocity followed by fast sinusoidalNot mentionedMAS score, ARAT
Fast: 1.5Hz (540°/s)
Healthy participants12Mean 51, SD 20NRNRNot mentioned
Alhusaini[44]CP27Mean 7, SD 2Not mentionedGMFCS levels I and IISpastic CP, GMFCS level I to II, no severe cognitive dysfunction, no orthopaedic surgery, or anti-spasticity treatment in previous 5moMedial gastrocnemius, SoleusTibialis anteriorLoad cell, potentiometer, surface electromyographyRamp movement across full range of motionAs slow as possible, slow, fastAt least 3Not mentionedMAS score, Tardieu Scale score
Ada[40]Stroke14Mean 65, SD 9Hemiplegia; 5–10mo post stroke≥3 on motor assessment scale≥3 on motor assessment scale; sufficient cognitive abilityMedial gastrocnemiusNoneLoad cell, potentiometer, surface electromyographySinusoidal between 10° plantarflexion and 10° dorsiflexion0.5, 1, 1.5, 2Hz (180, 360, 540, 720°/s)Each velocity trial was performed during 25sNoneNone
Healthy participants15Mean 52, SD 6NRNRNeurologically typical
Vattansilp[41]Stroke30Mean 68, SD 92–5y post stroke, 12 right hemiplegia, 18 left hemiplegiaNot mentionedCalf muscles diagnosed as clinically stiff, AS score ≥2, sufficient cognitive ability, no other problems interfering with ankle motionMedial gastrocnemiusNoneLoad cell, potentiometer, surface electromyographyRamp movement across full range of motion, sinusoidal between 10° plantarflexion and 10° dorsiflexionUndefined velocity for assessing contracture, 2°/s for assessing thixotropy, and at 2Hz (720°/s) for assessing spasticity1 at undefined velocity; 2 at 2°/s and during 30s at 2Hz (720°/s)NoneAS score
Healthy participants10Mean 59, SD 8NRNRNot mentioned
Table 2. Outcome parameters from instrumented tests developed from different signals at different stretch velocities
First authorPositionTorqueSurface EMG
  1. RMS-EMG, root mean square electromyography; ROM, range of motion.

Lamontagne[43]Low velocity: average angular velocity at −5° plantarflexionLow and high velocity: average torque at −5° plantarflexionHigh velocity: EMG onset was defined when EMG was greater than 2SD above the mean baseline level preceding onset
High velocity: maximum angular velocity
Wu[27]30°/s: ROM, angle of catch defined as the angle at maximum derivative of torque with respect to time, ratio between angle of catch and ROM30°/s: slope of torque–angle curve at 70° elbow flexion, energy loss (area between ascending and descending limbs of torque–angle curve), torque at 45°, 60°, 75° elbow flexion90°, 180°, 270°/s: EMG onset angle. EMG onset was defined when EMG was greater than the mean plus 3SD above the background level recorded during rest
At 90°, 180°, 270°/s: slope of peak torque vs. three stretch velocities, peak torque, maximum derivative of torque with respect to time
Voerman[35]Low velocity: passive wrist ROM30, 60, 90 cycles/min (180, 360, 540°/s): slope of torque-angle curve from neutral to full wrist extension30, 60, 90 cycles/min (180, 360, 540°/s): average RMS-EMG from neutral to full wrist extension
30, 60, 90 cycles/min (180, 360, 540°/s): passive wrist extension ROM, angular velocity
Van der Salm[42]<70°/s: ROM<70°/s: average torque in three zones over the full ROM50°, 75°, 100°/s: average RMS-EMG over 100ms window after EMG onset (>3SD) plotted against stretch velocities, exponential fit over 30–45 values, angle and angular velocity at EMG onset, slope values of angle/velocity onsets, angle at 100°/s = reflex initiating angle
High velocity: ROM, average maximum angular velocity
Bar-On[45]Low velocity: ROM, average maximum angular velocityLow and high velocity: change in average torque at maximum velocity between velocities, change in average integral of torque–angle curve from maximum velocity to 90% ROM between velocitiesLow and high velocity: change in average RMS-EMG in maximum velocity zone (200ms before maximum velocity to 90% ROM) between velocities (expressed as percent of peak value of three maximum voluntary isometric contractions), EMG onset defined as time of first muscle activity according to the method of Staude and Wolf[63]
Bar-On[46]Low velocity: ROMLow and high velocity: change in average torque at 70° knee flexion between velocities, change in average integral of torque–angle curve from maximum velocity to 90% ROM between velocitiesLow and high velocity: change in average RMS-EMG in maximum velocity zone (200ms before maximum velocity to 90% ROM) between velocities
High velocity: average maximum angular velocity, angle of catch defined as the angle corresponding to the time of minimum power after maximum power during the first high velocity stretch (expressed as percent of the ROM)High velocity: minimum power after maximum power in the first high-velocity stretch
Pandyan[13]Low and high velocity: ROM, average angular velocityLow and high velocity: change in slope of force–angle curve between velocities over full ROMLow and high velocity: change in RMS-EMG over full ROM between velocities
Lebiedowska[47]0.2–1.5rad/s (11.5–540°/s): passive ROM0.2–1.5rad/s (11.5–540°/s): slope of torque–angle curve during initial increase, integral of torque–angle curve over full ROM0.2–1.5rad/s (11.5–540°/s): maximum value of RMS-EMG over ROM, slope of RMS-EMG velocity curve
Hypertonia of neural origin: RMS- EMG ≥ mean 3SD before movement began in slow and fast velocity stretches.
Hypertonia of non-neural origin: RMS-EMG < mean 3SD before movement began in low- and high-velocity stretches
Fleuren[7]Low velocity: passive ROMLow and high velocity: integral of torque–time curve over full ROMLow and high velocity: average RMS-EMG over full ROM
Malhotra[37]Low and high velocity: ROMLow and high velocity: slope of force–angle curve 10–90% ROMLow and high velocity: average RMS-EMG over full ROM
Shapes of force–angle curves:Patterns of muscle response:
Slope of force–angle curve <0.7N/°: negative stiffness.No/negligible muscle response
Slope of force–angle curve >0.7N/° and R2 >0.6: linear stiffnessPosition dependent: muscle response independent of stretch velocity
Slope of force–angle curve >0.7N/° and R2 <0.6: non-linear stiffness (catch or clasp-knife)Velocity dependent: negligible muscle activity during slow stretch, increased activity during fast stretch
Position and velocity dependent
Early catch: early muscle activation reducing as the muscle lengthens
Chen[38]1/3, 1/2, 1, 1.5Hz (120, 180, 360, 540°/s): ROM1/3, 1/2, 1, 1.5Hz (120, 180, 360, 540°/s): velocity-dependent viscous component of torque (see appendix in[38]), Slope of viscosity–velocity graph (see[26])1, 1.5Hz (360, 540°/s): angle at EMG onset
Turk[39]0.5Hz (28.6°/s): tracking index –ability to accurately follow tracking signal, ROM0.04Hz (14.4°/s): force/torque angle index, average change in force/torque between 0° and 30° wrist extension1.5Hz (540°/s): stretch index – average RMS-EMG minus resting EMG during wrist extension
Alhusaini[44]Low velocity: ROMLow velocity: contracture, angle <10° dorsiflexion at 4.6Nm of forceHigh velocity: average, normalized RMS-EMG
Ada[40]None0.5, 1, 1.5, 2Hz (180, 360, 540, 720°/s): change in torque over 20° interval0.5, 1, 1.5, 2Hz (180, 360, 540, 720°/s): gain in RMS-EMG over ROM (μV/°)
Vattansilp[41]Undefined low velocity: ROM2Hz (720°/s): change in torque over 20° interval2Hz (720°/s): gain in RMS-EMG over ROM (μV/°)
Table 3. COSMIN scores and reasoning for scores on the reliability of included studies
First authorInterrater reliabilityIntrarater reliabilityMeasurement error
  1. For an extended version including statistical findings, see Table SII (online supporting information). NA, not applicable; MIC, minimally important change; ARAT, Action Research Arm Test; ICC, intraclass correlation coefficient; SDC, smallest detectable change.

Lamontagne[43]Not performedWithin one session, 1s between repetitionsWithin one session, 1s between repetitions
COSMIN scoreNAPoorPoor
Only the biomechanical parameter was assessed for reliabilityThe absolute measurement error was not provided
Short time interval between repetitionsOnly the biomechanical parameter was assessed for reliability
No information on missing valuesNo information on missing values
Wu[27]Not performed1d between measurementsNot calculated
Only the biomechanical parameter was assessed for reliability
Reliability was measured only in children with typical development (the results cannot be generalized to persons with disabilities)
Voerman[35]1d between measurements10min between measurementsNot calculated
COSMIN scoreFairGoodNA
Small sampleUnclear whether administrations were independent
Participants were missing ARAT score
Van der Salm[42]Not performedWithin one session, 5s between repetitionsNot calculated
Short time interval between repetitions
Only one parameter was assessed for reliability
Bar-On[45]Not performedAverage of 13d (SD 9d) between measurementsAverage of 13d (SD 9d) between measurements
COSMIN scoreNAGoodGood
Small sample sizeSmall sample size
No indication if participants were stable in the interim periodNo indication if participants were stable in the interim period
MIC not reported
Bar-On[46]Not performed (0)Average of 13d (SD 9d) between measurementsAverage of 13d (SD 9d) between measurements
COSMIN scoreNAGoodGood
No indication if participants were stable in the interim periodNo indication if participants were stable in the interim period
MIC not reported
Turk[39]Immediately following the assessment by the first raterInterval of one measurement procedure (time not specified)Interval of one measurement procedure (time not specified)
COSMIN scoreGoodGoodGood
ICC values were not calculatedTime interval between administrations unknownFor some parameters the average difference between persons with disabilities and comparison participants were larger than the SDC
MIC not reported
Pandyan[64]Not performedWithin one session, 10–15s between repetitionsNot calculated
Only the biomechanical parameter wasassessed for reliability
Short time interval between repetitions
ICCs were not calculated
Table 4. COSMIN scores and reasoning for scores on the validity of included studies
First authorContent validityConstruct validity/hypothesis testingResponsivenessInterpretability
  1. For an extended version including results and statistical findings, see Table SIII (online supporting information). CP, cerebral palsy; SDC, smallest detectable change; MIC, minimally important change; ARAT, Action Research Arm Test; ROM, range of motion; RMS, root mean square; EMG, electromyography.

Lamontagne[43]Not measuredComparison with motor-controlled deviceNot measuredMeans and SDs of outcome parame-ters provided
COSMIN scoreNAFairNAGood
Small sampleSmall sample
The high-velocity stretches were not comparable between the hand-held dynamometer and the motor-controlled deviceSDC and MIC not reported
A description of the parameters from the motor-controlled device were missingLimited focus
Small sampleSmall sample
Wu[27]Relation between signalsMatched to comparison groupNot measuredMeans and SDs of outcome parameters provided
Relation of signals to velocityComparison with clinical scales
COSMIN scoreGoodGoodNAFair
The type of CP (spastic, dystonia, etc.) was not mentionedParametric statistics performed to compare groups while the sample size was relatively small and the data distribution was not reportedNo description of missing data
No description of missing dataSDC and MIC not reported
Type of CP and the study setting not mentionedThe type of CP and the study setting was not mentionedThe type of CP and the study setting was not mentioned
Voerman[35]Relation of signals to velocityMatched to comparison groupNot measuredMeans and SDs or medians and ranges of outcome parameters provided
Comparison with motor-controlled device
COSMIN scoreFairGoodNAGood
Theoretical framework described but statistical comparisons not performedThe sample size used for the correlations with ARAT was smallSDC and MIC not reported
The measurement properties of the motor-controlled device/comparator instrument were not described
No data on content validity availableFewer participants were tested with ARAT and with the motor-controlled deviceFewer participants were tested with ARAT and with the motor-controlled device
Van der Salm[42]Relation between signalsNot measuredNot measuredMeans and SDs of outcome parameters provided
Relation of signals to velocity
COSMIN scoreFairNANAGood
Torque measured in only four participants and only at low velocitySmall sample SDC and MIC not reported
The characteristics of excluded participants were missingTorque was measured in only four of the participants
Bar-On[45]Relation of signals to velocityMatched to comparison groupNot measuredMeans and SDs of outcome parameters provided, SDC could be calculated
Comparison with clinical scales
COSMIN scorePoorGoodNAGood
No statistical tests performedHypotheses not explicitly statedMIC not reported
Limited data on content validity available
Bar-On[46]NAComparison with clinical scalesTreatment with botulinum toxin AMeans and SDs of outcome parameters provided, SDC provided
COSMIN scoreNAExcellentExcellentGood
MIC not reported
Pandyan[13]Relation between signalsComparison with clinical scalesNot measuredMedians and ranges of outcome parameters provided
Relation of signals to velocity
COSMIN scoreExcellentGoodNAGood
No description of how missing data were handledSDC and MIC not reported
Lebiedowska[47]Comparison between signalsMatched with comparison groupNot measuredMeans and SDs of outcome parameters provided
Relation of signals to velocity
COSMIN scoreFairFairNAFair
See comments on relation of signals to velocity in Table SIII (available online)The excluded participants' characteristics were not describedSubgroup comparisons were based on small samples
Statistical comparisons were carried out on small samplesElectromyography data were not normalizedSDC and MIC not reported
No diagnostic information, indication of spasticity severity, or functional level providedNo diagnostic information, indication of spasticity severity, or functional level providedSubgroup comparisons were based on small samples
Influence of heterogeneity between participants was not checked forInfluence of heterogeneity between participants not checked for
Fleuren[7]Relation between signalsComparison with clinical scalesNot measuredMeans and SDs of outcome parameters were not provided
Relation of signals to velocity
COSMIN scoreGoodGoodNAPoor
The instrumented parameters were correlated to the velocity of stretch with the intention of explaining the variability in performance rather than to test content validityThe instrumented parameters were correlated to the Ashworth Scale with the intention of explaining the variability in performance rather than of testing construct validityNo instrumented data on spasticity presented
SDC and MIC were not reported
Muscle activity from antagonist muscles not measuredLarge influence of rater on multivariate mixed linear model with Ashworth Scale score as dependent variable
Disease characteristics not reportedDisease characteristics not reportedDisease characteristics not reported
Malhotra[37]Relation between signalsComparison with clinical scalesNot measuredMeans and SDs of outcome parameters provided
Relation of signals to velocity
COSMIN scoreExcellentGoodNAGood
No information on missing dataSDC and MIC not reported
Chen[38]Relation of signals to velocityComparison with clinical scalesTreatment with botulinum toxin AMeans and SDs of outcome parameters provided
COSMIN scorePoorPoorPoorFair
No statistical tests carried outNo statistical tests carried outThe EMG parameter was compared pre and post treatment on individual participant data rather than with group analysisNo information on missing data
No analysis of subgroups
No information on missing dataSome comparisons were made using independent, rather than dependent group analysesImportant statistical flaws
SDC and MIC were not reported
No information on missing dataNo information on missing dataNo information on missing data
Turk[39]Not measuredMatched to comparison groupNot measuredMeans and SDs of outcome parame-ters provided. SDC can be calculated
COSMIN scoreNAGoodNAExcellent
The magnitude of expected differences between groups was not included in the hypotheses
Alhusaini[44]Relation of signals to velocityComparison with clinical scalesNot measuredMeans and SDs of outcome parameters were not provided
COSMIN scorePoorGoodNAPoor
No statistical tests were carried outNo description regarding missing dataNo values from the instrumented test were reported
The magnitude of expected correlations were not included in the hypothesesSDC and MIC were not reported
Stretch velocities were not reported
Ada[40]Relation between signalsMatched to comparison groupNAMeans and SDs of outcome parameters provided
COSMIN scoreFairGoodNAGood
Subgroup analyses were based on small samplesSome missing statistical resultsSDC and MIC not reported
Some missing statistical resultsNo hypotheses on expected resultSome samples were too small
No information on how missing values were handledThe percentage of responders who had the lowest/highest possible scores was not reported
Vattansilp[41]Relation between signalsMatched to comparison groupNANot all means and SDs of outcome parameters were provided
Comparison with clinical scales
COSMIN scoreGoodPoorNAPoor
Spasticity was not definedThe velocity of stretch used to evaluate ROM was not definedMissing some descriptive statistics related to contracture and spasticity
Gain in RMS-EMG not compared between groupsSDC and MIC were not reported
Change in torque was assessed only at high velocity
Parameter values were not compared with clinical scores
Gender of included participants was not reportedGender of included participants was not reportedGender of included participants was not reported
Place from which participants were recruited was not mentionedPlace from which participants were recruited was not mentionedPlace from which participants were recruited was not mentioned

Study populations and muscles tested

Information on participants, instrumentation, and protocol details is summarized in Table 1. Seven of the 15 articles studied spasticity in adults post stroke.[35-41] Two articles included persons with spinal cord injury,[42, 43] and four reported on children with CP.[27, 44-46] One study included adults post stroke, and adults and children with CP, within the study population.[7] One article included adults post stroke, with spinal cord injury, and with CP.[47] Eight studies additionally included a healthy comparison group of participants.[27, 35, 39-41, 45-47] Six articles studied spasticity in upper limb muscles,[13, 27, 35, 37-39] eight in lower limb muscles,[40-47] and one in both upper and lower limb muscles.[7]

Instruments and protocols

Angular position and velocity were recorded in most studies using calibrated potentiometers or electrogoniometers,[7, 13, 27, 35, 37, 39-44, 47] in two studies using inertial sensors containing an accelerometer and a gyroscope,[45, 46] and in one study a velocity sensor was used.[38] Forces and/or torques exerted at the joint when manually displacing the segment during passive stretch were measured with different devices. Most often, force measurements were carried out using single- or multiple-axes force transducers,[7, 13, 35, 37, 39-41, 43-47] or differential pressure sensors.[38] Forces were then recomputed to torques based on measurements[39-41, 44-47] or estimations[38, 43] of moment arms. Three studies directly measured torque near the joints[42] in order to account for the torques applied by the examiner on the handle of the sensor.[45, 46] All studies used surface EMG (sEMG) to record agonist muscle activity and eight studies additionally measured the antagonist muscle activity.

All studies assessed spasticity during passive ramp stretches of the spastic agonist muscles, except for three studies that analysed passive sinusoidal movements,[35, 38, 39] and two studies that did both.[40, 41] Stretches were performed at two velocities (slow and fast),[7, 13, 37, 39, 41, 43, 45, 46] at three velocities,[35, 44] or at four or more velocities.[27, 38, 40, 42, 47] Stretch velocities ranged from 2° to 720°/s. One study did not report the applied stretch velocity.[44] Within each velocity, stretch repetitions were applied immediately or with rest intervals up to 1 minute.

In addition to instrumented spasticity tests, 12 of the 15 studies assessed spasticity with the MAS[7, 13, 27, 35, 37, 38, 41, 42, 44-46] and two studies also used the MTS.[44, 46] Three studies in adults post stroke also examined the relation between spasticity indicators and upper limb function.[35, 37, 39]

Study design and data analysis

Although most authors failed to mention how spasticity was defined in their study, the majority followed the reasoning that velocity-dependent hyperactivity of the stretch reflex, causes a pathological augmentation in muscle activity.[4] Slow stretching was performed at a velocity below the stretch reflex threshold (SRT), whereby it was hypothesized that non-neural elastoviscous muscle properties accounted for any increased force or torque measured over the range of motion (ROM). During a high-velocity passive stretch, activation of the muscle additionally influenced any increase in torque. The amount of gain in muscle activity, its timing, and the amount of torque produced at different stretch velocities constituted some of the possible quantifiable measures of spasticity. A summary of the main outcome parameters developed by each study to quantify spasticity can be found in Table 2. A distinction is made in Table 2 between parameters that mostly reflect angular position/velocity, forces and/or torques, or muscle activity. The velocity at which each parameter was examined is also specified however, most studies combined different signals and velocities to develop their outcome parameters.

For the angular position/velocity parameters, all but two studies[40, 43] measured the available ROM during a passive stretch performed at a velocity below the SRT. Therefore, any decrease in ROM or catch angle[27, 46] during a higher velocity stretch was presumed to be caused by increased muscle activity. Often referred to as either resistance[37, 40, 41, 47] or stiffness,[27] the slope of the torque–angle curve was the most common measure of increased torque. This parameter was calculated over the entire ROM,[13] or over a section of the ROM[35, 37, 39-42, 47] and compared between velocities[13, 27, 35, 37, 40, 47] or between positions.[42] Four studies examined the torque value at a specific joint angle at different velocities.[27, 43, 45, 46] Four studies also examined the integral of the torque–position graph to quantify the amount of work needed to stretch the examined muscle,[27, 45-47] and one study calculated the integral of the torque–time graph.[7] When stretches were performed against the force of gravity and the mass of the displaced segment was not negligible,[7, 27, 38, 42, 44-47] five studies subtracted the effect of inertia from the resulting measured torque.[27, 38, 45-47]

Nine of the 15 articles quantified sEMG amplitude by calculating the average root mean square of the sEMG signal (RMS-EMG) over a particular interval,[7, 13, 35, 37, 39, 42, 44-46] two by examining the gain in RMS-EMG over the ROM,[40, 41] and one by calculating the maximum value of the RMS-EMG.[47] Similarly to the biomechanical parameters, average RMS-EMG was often calculated over a specific portion of the ROM and compared between velocities. Two articles normalized the RMS-EMG amplitude value to maximum voluntary isometric contraction.[44, 45] Three articles recorded and analysed either the angle or the velocity at EMG onset.[27, 38, 42] Two articles identified different types of spasticity based on sEMG parameters.[37, 47]

Psychometric properties


The COSMIN scores of those studies examining reliability can be found in Table 3. For an extended version of this table, also containing the methodological and statistical results and scores, readers are referred to Table SII. Six studies[27, 35, 39, 42, 43, 45] explored the intrarater reliability of some, or all of the outcome parameters from the instrumented tests and two studies[13, 46] referred to previously collected reliability results. Of these eight studies, only four examined the reliability of electrophysiological parameters in addition to biomechanical parameters in persons with disabilities,[35, 39, 45, 46] and two studies also assessed interrater reliability.[35, 39] The methodological quality of studies ranged from poor to good as study samples tended to be small or the interval between repeated measurements was inappropriate. Reliability results were generally better among persons with disabilities than among comparison groups. Biomechanical parameters tended to have higher relative reliability than electrophysiological parameters.[35, 39] Turk et al.[39] reported on the measurement error of the parameters in their study, which ranged from 40% to 77% of the mean values of those parameters in the participant sample. Several parameters from the studies by Bar-On et al.[46] were found to have an absolute measurement error small enough to distinguish between groups[45] and detect change as a result of treatment. The minimally important change (MIC) was not identified in any study.


The COSMIN scores on the validity of the different studies are summarized in Table 4. The methodological quality of the included studies ranged from poor to excellent, with the main weaknesses being uncertainty of statistical strength and limited analyses, mainly for content validity. Reasons for score allocation per domain, together with methodological and statistical scores can be found in Table SIII.

Content validity

Content validity was evaluated by a comparison of biomechanical to electrophysiological parameters,[7, 13, 37, 40, 47] or by a comparison of parameters between stretch velocities.[7, 13, 27, 37, 40-42, 45-47] Pandyan et al.[13] and Fleuren et al.[7] reported conflicting results regarding the correlation between RMS-EMG and the slope of the torque–angle curve in spastic elbow flexors.[7, 13] In the soleus of participants post stroke, higher torque values were associated with hyperactive stretch reflexes,[40] and the gain in EMG accounted for 27% of the variance in the measured torque.[41] Associations between patterns of muscle activity and the biomechanical parameters during high-velocity passive stretches could not be demonstrated in the wrist[37] or the knee flexors.[47] Electrophysiological[13, 27, 37, 40, 42, 45, 46] and biomechanical[7, 37, 38, 45, 46] parameters however, often changed with increasing stretch velocity. Two studies reported no increase in the slope of the torque–angle curve between velocities.[13, 40]

Construct validity and hypothesis testing

Constructs or hypotheses were tested in 12 studies by comparing persons with disabilities with a comparison group,[27, 35, 39-41, 45, 47] or by comparing the instrumented method with a clinical spasticity test,[7, 13, 27, 35, 37, 41, 44-46] or with a motor-driven test.[35, 43] In those studies, comparing persons with disabilities to those without disabilities, average RMS-EMG parameters were always able to distinguish between groups.[35, 39, 46, 47] In contrast, only in four studies, and only in some muscles, were biomechanical parameters able to distinguish persons with disabilities from comparison participants.[27, 40, 45, 47] Conflicting results were found when outcome parameters were related to the scores of clinical spasticity tests. Two studies reported good significant correlations (r=0.64) between RMS-EMG and MAS scores for some muscles[7, 35] while others reported low associations (r=0.06[13], kappa=0.09[44]). RMS-EMG parameters were significantly higher in the hamstrings of children with CP with high MAS scores (2–3) than in the muscles of those with low MAS scores (1–1+), but this was not the case for the gastrocnemius.[45] Similarly, for the MTS, conflicting results were found for the calf muscles of children with CP, with one study reporting good agreement (kappa=−0.48) between the angle of response as measured by the Tardieu scale and RMS-EMG,[44] and another, only poor to fair (r=0.2) correlations.[45] In five studies, ROM and biomechanical parameters were strongly correlated to MAS scores[7, 27, 35, 41, 45] and in one study to the Tardieu Scale.[44] However, Malhotra et al.[37] found that their biomechanical parameters did not change with increasing MAS scores. Bar-On et al.[46] found that the instrumented assessment identified significantly more responders to treatment with botulinum toxin A (BoNT-A) injections in the hamstring muscles than the MAS, but not more than the MTS. However, a combination of several baseline parameters from the instrumented test could better predict the effect of treatment than the baseline MTS alone.[46] Parameters from a manual device were compared with those from a motor-driven device and showed very good correlations (r=0.86–0.94).[35] On the other hand, Lamontagne et al.[43] detected fewer participants with hyperactive stretch reflexes using the motor-driven system than with the hand-held device, although, in this study, stretch velocities were not comparable.

Responsiveness and interpretability

Responsiveness to anti-spasticity medication was evaluated by only two studies. However, the conclusions of one study were weakened as the methodology did not fulfil all criteria for high quality.[38] No study provided MIC values. In three studies,[39, 45, 46] the smallest detectible change (SDC) values could be calculated from the reported absolute measurement errors. Bar-On et al.[46] identified EMG and torque-related parameters that, relative to the SDC, decreased post treatment. No study investigated all aspects of content validity, construct validity, and responsiveness as relevant to spasticity measurement.


The goal of this systematic review was to identify instrumented spasticity assessment methods that could be used as viable alternatives to commonly applied clinical evaluations, such as the MAS. Fifteen instrumented spasticity assessment methods developed following the recommendations by the SPASM consortium[20] were identified. These methods are manually controlled ensuring their ability to be translated to clinical settings and measure both electrophysiological and biomechanical signals.

In comparison with previous reviews,[17-20, 29, 30] the current paper covered a narrower scope of spasticity assessments by reporting on the measurement of passive-state spasticity only. This focus ensured that the concept of spasticity was similarly defined in all of the included studies, namely the definition of spasticity as offered by Lance.[4] A wider definition of spasticity includes spasticity as manifested during active conditions.[5] The exact pathophysiology of spasticity during active motion remains debatable[48] and consequently, the literature related to its impact on function is divided.[40, 49] Whereas in the passive state enhanced muscle activity is primarily pathological, in the active state it is more difficult to discern reflex-mediated activity from voluntary activation. In persons with UMN syndrome, activation is also influenced by other impairments such as sensorimotor control problems and weakness. Whether one can apply a theory developed for measurement of a phenomenon in the passive state to the complex activation occurring during activity is, therefore, subject to speculation.[50] Although it is acknowledged that spasticity affects activity, we believe that accurate assessment methods need first to be developed for passive and active situations separately in order to deconstruct the multifactorial phenomenon.

Overall, the findings of the current review show that manually controlled instrumented spasticity assessments that are clinically applicable are available. Those developed for assessing spasticity in the hamstrings in children with spastic CP have so far, undergone the most rigorous clinical assessments.[45, 46] However, no developed method has been sufficiently assessed on all the required psychometric properties. Several UMN syndromes were assessed in the included studies showing that spasticity can be quantified in a variety of different pathologies. Most studies on this subject however, have been carried out in adults post stroke and the number of muscles investigated remains limited. This indicates that instrumented spasticity assessment in other areas still requires much development. Similar to the findings of Flamand et al.,[29] only six studies quantified spasticity in children with CP and information on absolute reliability and responsiveness was limited to work by only one research group.[45, 46]

Most of the reliability findings were limited to biomechanical parameters, with only four studies including a reliability analysis of RMS-EMG parameters among persons with disabilities.[35, 39, 45, 46] Since no or little electrophysiological response is expected when passively stretching healthy muscles, it was not surprising that relative reliability in comparison participants was poor. However, also among persons with disabilities, the electrophysiological response was occasionally found to be variable and unstable.[39] To reduce the variability inherent in RMS-EMG and to be able to compare between participants, signals can be normalized to a maximum voluntary contraction, as was done in two of the reviewed studies.[44, 45] However, this normalization technique in persons with co-contraction and weakness is debatable.[51] EMG can also be normalized to an M-wave during a supramaximal stimulation[52] although more studies are required to assess the clinical applicability of such a method. As an essential start, better protocol standardization is required to reduce the variability of RMS-EMG parameters. On the other hand, the variability in response may also be a true phenomenon of spasticity which on its own, is worthy of further investigation.

Quantifying the measurement error of an instrument is important when analysing reliability and responsiveness. Calculation of the standard error of measurement was carried out in three of the reviewed studies.[39, 45, 46] This permits the calculation of the SDC, which is the value of the amount of change that falls outside the measurement error of an instrument.[53] This is essential for a methods application as an evaluative measure in intervention studies, and without it clinical practice is limited. In a study by Bar-On et al.,[46] three parameters were identified that on average, decreased more than the SDC post treatment with BoNT-A. In addition, the baseline values of these parameters were able to predict the response post treatment.[46] The MIC refers to the change that is considered to be minimally important by patients and clinicians.[53] The MIC differs from the SDC as it cannot be statistically determined. Instead, it requires large, in-depth intervention studies, often in combination with clinical consensus. Such methodology was not applied in any of the reviewed studies, which resulted in limited scores on the interpretability item of the COSMIN checklist.

To be clinically applicable, an assessment also needs to be compact and easy to administer. Although clinical feasibility and utility were not systematically assessed in the current review, the choice to only include manually controlled assessment methods partially covered this issue. In children and particularly during high-velocity displacements, a motor-driven device may prevent the participant from being sufficiently relaxed. Manually controlled assessments on the other hand, are better tolerated, allow the examiner to have more control over the state of the participant, and are also transportable. For example, in the study of Malhotra et al.,[37] the assessments were performed at the patient's bedside.

The compromise between accessibility and accuracy is also challenged by the need to record and synchronize both electrophysiological and biomechanical signals. Fortunately, technological advancements have improved the accuracy, synchronization capabilities, and portability of equipment. For example, wireless inertial measurement units are reliable and valid in motion analysis[12] and have recently been combined with EMG sensor technology.

Recording kinematic data is essential for comprehensive spasticity assessment. Firstly, it ensures the consistency of stretch performance and allows for interpretation of data in accordance with the velocity of stretch. Secondly, with advances in musculoskeletal modelling, kinematic data can be used to calculate muscle lengths and lengthening velocities,[54] essential for spasticity interpretation. While all of the reviewed methods acknowledged the need to assess spasticity at various muscle lengthening velocities, only eight studies integrated the information from EMG and torque with velocity. Even fewer explored both signals relative to joint position or muscle length.[37, 42] Evaluating EMG response to both increasing muscle length and lengthening velocity, allows identification of SRTs, which have been found to be reduced in individuals with UMN syndrome.[55] Studies in adults suggest that decreased SRTs may be related to spasticity severity,[56] type of motor deficit,[55] and risk of developing contractures.[37] Investigating both the dynamic and static SRTs in elbow flexors, Jobin and Levin[57] found more velocity dependence of the SRTs in children with CP than in adults with stroke. Van der Salm et al.[42] highlighted position-dependent activation in persons with spinal cord injury in which the joint angle, rather than the angular velocity, was the trigger of the neurological response. These findings were supported by two more studies that identified either position- or velocity-dependent muscle activation patterns among different participants.[37, 47] Chen et al.[38] reported an increase in the dynamic SRT post BoNT-A treatment. However, identification of SRTs is highly dependent on the performance of controlled, yet variable stretch velocities, which may be more difficult to achieve with manual stretches.[58] Nevertheless, as protocols become more standardized, the reliability of acquiring these parameters with a manual test is worth further investigation.

Several studies were able to show that average RMS-EMG, whether over the full ROM, over a specific interval, or as a function of velocity, was distinguishable between those with disabilities and comparison participants without disabilities.[19, 39, 45, 47] That said, only three studies showed that some of the developed biomechanical parameters, namely the slope of the torque–velocity curve,[27] and the integral of the torque–angle curve,[45, 47] were higher in people with disabilities than in comparison participants. Results on content validity showed only moderate correlations between torque–angle curves and RMS-EMG.[13] Chen et al.[38] found that the velocity-dependent viscous component calculated from the torque–velocity curve during a sinusoidal motion was sensitive to treatment with BoNT-A. When interpreting these results together, it is possible that a parameter based on torque and velocity best corroborates the velocity-dependent nature of spasticity, while the slope of the torque–angle curve is better used as a measure of non-neural related stiffness. The lack of agreement on which parameter best quantifies the biomechanical effect of spasticity may be solved by better differentiation of the neural and non-neural components of increased torque. Models that differentiate into components such as reflex-mediated torque, stiffness, and viscosity have mostly been validated on data collected in research settings using motor-driven devices.[24, 59] Proponents of motor-driven spasticity assessment devices argue that by allowing a robot to control the displacement, the limb dynamics of the experimenter can be avoided, allowing for accurate modelling of the participant's passive state. Nevertheless, as was partially shown by two of the included studies, by improving the performance standardization of manual tests, a distinction can be made between an increase in torque which is aggregated by muscle activity, and an increase in torque of non-neural origin, for example contracture.[37, 47] Future work should focus on validating the different components and checking their responsiveness to treatment.

Although comparison of an instrumented test with a clinical comparator was indicated in the current review as constituting a part of construct validity, multiple studies have shown the inadequacy of clinical tests, such as the MAS and MTS, in assessing spasticity.[10, 11, 45, 60, 61] Therefore, it was not surprising that in general, the articles reviewed reported poor correlations between the electrophysiological findings of the instrumented tests and the scores of the MAS and MTS. This finding confirms the inadequacy of the clinical tests, rather than highlighting the construct validity of the instrumented alternatives. The MAS and MTS may be useful for diagnosing spasticity and for categorizing muscles into broad severity categories.[45, 62] However, for a comprehensive picture of the problem and better differentiation of mid-range severities, the clinical examinations should be supported by more rigorous instrumented assessments, especially for those undergoing treatment.[46]

In conclusion, the search for a clinically applicable instrumented spasticity assessment is still ongoing as the translational capabilities from research to clinical setting are unnecessarily lagging behind. Some promising developments of instrumented spasticity assessments that integrate signals have been found. However, more consensus is required on the optimal parameters that quantify spasticity, provide insight into its nature, and differentiate it from non-neural-related increases in torque. Parameters based on RMS-EMG fulfil aspects of validity in adults post stroke[13, 37] and in children with CP.[27, 45] However, the interrater reliability of these parameters remains unexplored, and responsiveness studies should be expanded to more muscles and different patient populations. Most importantly, for a parameter based on RMS-EMG to be used as a quantifiable measure of spasticity, methods should aim at standardizing tests to ensure adequate reproducibility. Few developed torque-related parameters possess convincing content or construct validity to be used as clinical measures of spasticity. By improving the joint torque models, however, and differentiating between the components of increased torque, this could be achieved. Simple, yet accurate, instrumented spasticity assessments will greatly advance clinical practice in terms of treatment planning and outcome evaluation. In parallel, collection of instrumented data will help define and classify different aspects of spasticity, providing insight into the many paradigms related to its pathophysiology.


Lynn Bar-On is supported by a grant from the Doctoral Scholarships Committee for International Collaboration with non EER-countries (DBOF) of the University of Leuven, Belgium. This work was further supported by a grant from for Applied Biomedical Research from the Flemish agency for Innovation by Science and technology (IWT-TBM: grant number 060799). The authors would like to thank Riet De Bondt for her help in article selection and Simon Schless for proof reading the manuscript. The authors state that that they had no interests that might be perceived as posing a conflict or bias.