Strength muscle power and rapid force production assessment after stroke

Executive conclusions

Strength assessment after stroke is clinically useful when the task, limb, apparatus and phase of recovery are explicit. The most defensible clinical applications are to describe voluntary force capacity, identify a potentially treatable limitation, choose exercise loading and monitor change using a compatible protocol. The evidence is less mature for forecasting an individual's walking recovery, future falls or participation from a single strength or power result.

The central recommendation for rehabtools is a small set of well-described measurement options with transparent interpretation, rather than a single stroke strength score. Handgrip, handheld dynamometry, fixed dynamometry, isokinetic torque, machine one-repetition maximum, mechanical power and rapid force production measure related but distinguishable properties. Selection should follow the clinical decision and the person's capacity to complete the test.

Five conclusions are especially important.

Both limbs require interpretation. The less affected limb is an informative comparator, but it is not a healthy control. Bilateral deficits, differential recovery, premorbid asymmetry and compensatory loading can all alter the denominator of a paretic-to-nonparetic ratio. A favourable ratio can coexist with severe bilateral weakness. Conversely, increased nonparetic strength can make the ratio look worse despite paretic improvement. Preserve both raw limb values alongside every ratio.

High ICCs do not guarantee small individual error. Well-standardized grip and dynamometry protocols often rank participants reproducibly. However, error differs substantially by muscle, device, contraction mode, trial rule and interval. Some recent papers give attractive change thresholds that cannot be accepted at face value: an acute handheld study has unreconciled dispersion and MDC calculations, while a 2025 grip paper appears to reinterpret the standard error of the mean as measurement error. [1, 2]

Rapid force production is promising but technically demanding. A best 200-ms slope anywhere on a handheld torque trace is not the same outcome as the first 200 ms from contraction onset on a rigid dynamometer. Those approaches have materially different reliability and construct meaning. In the larger handheld study, rapid torque development added no significant explanatory value to strength for current fast walking speed. This is not evidence that rapid force is unimportant for every functional task, but it argues against assuming its superiority. [3, 4]

Prospective strength prognosis exists, but deployment evidence is limited. Admission grip has been associated with later inpatient ADL outcomes, and discharge lower-limb torque has been associated with physical quality of life and community reintegration six months later. These findings are different from the much larger cross-sectional literature. They do not yet supply a universally calibrated discharge, walking or fall-risk calculator. Model performance, where available, belongs to the complete model rather than the grip or torque measure alone. [5–9]

Stroke-specific interpretation cannot be borrowed from Parkinson’s disease or healthy ageing. Sarcopenia grip thresholds, healthy reference percentiles, early-stroke important-change estimates and laboratory MDC values answer different questions. None should be relabelled as a generic stroke prognosis cutoff. Assessment should remain possible for people with low force, while documenting inability to understand, grip, stabilize, tolerate or safely perform a task separately from a genuinely observed zero.

Scope and measurement framework

What this report covers

This report addresses assessment of maximal voluntary strength, muscle power and rapid force production after established stroke. It includes paretic and nonparetic handgrip, selected upper-limb strength tests, lower-limb handheld and fixed dynamometry, isokinetic testing, machine strength, and experimental activation or force-control measures where they clarify interpretation. Walking, transfers and ADL are considered as functional correlates and later outcomes. Detailed dexterity, comprehensive upper-limb recovery batteries and treatment-effectiveness reviews are outside the central scope. Chair-rise tests and derived chair-rise power belong primarily in the companion chair-rise report.

The synthesis is a critical narrative review, not a registered systematic review. It prioritizes appraisal of what original methods and data permit a clinician or developer to conclude. It does not treat every published statement that a variable “predicts” function as prospective prognostic evidence.

Stroke phase and eligibility

For orientation, the Stroke Recovery and Rehabilitation Roundtable distinguishes the hyperacute first 24 hours, acute period through the first week, early subacute period from one week to three months, late subacute period from three to six months, and chronic period beyond six months. These are research conventions rather than abrupt biological boundaries. Individual studies sometimes use different definitions; this report therefore retains their actual time since onset and explicitly notes important differences. [10]

Eligibility is equally important. Ability to walk without an aid, produce any force, grasp and release, tolerate prone lying or follow spoken commands can exclude much of the population seen in rehabilitation. A protocol validated in mild stroke should not be described as validated in severe aphasia, neglect, substantial pain or complete paralysis without evidence. The absence of such participants is an applicability limitation, not proof that they cannot be assessed with an appropriately adapted method.

Define the physical quantity before choosing the instrument

Maximum isometric strength is the greatest voluntary external force or joint torque attained under defined static constraints. A force reading in newtons is not automatically torque in newton metres. Torque calculation requires the perpendicular moment arm and consistent alignment; a change in pad position changes the force needed to produce a given joint torque. Device displays labelled kilograms generally represent kilogram-force rather than mass. Conversion to newtons is possible, but it cannot correct posture, fixation or leverage differences.

An isokinetic dynamometer constrains angular velocity within the usable movement range. It does not abolish acceleration and deceleration phases, eliminate gravity, or guarantee that a very weak participant achieves the target velocity. The measured torque is net joint torque, influenced by agonist drive, opposing muscle activity, passive tissues and test geometry. Isometric and eccentric tests answer different questions from concentric tests and should not be pooled into one strength number.

One-repetition maximum is the greatest load successfully moved once through a stated range on a stated machine. Its unit may be kilograms of stack mass, but that does not make it comparable across machines with different lever systems. It is useful for machine-specific loading decisions. It does not isolate an anatomical muscle group merely because the machine is described as a leg press or knee extension device.

Mechanical power is the rate of doing work, commonly force multiplied by velocity or torque multiplied by angular velocity. External mechanical power in a strictly isometric task is zero at the stationary interface, even when force rises rapidly. RFD and RTD describe force-time or torque-time slopes and require a defined time interval. A 0–50-ms slope, a 0–200-ms average slope, a maximum moving-window slope and time to 90% maximum are different variables. A derivative-based metric is particularly sensitive to acquisition, filtering, onset selection and mechanical compliance. [3, 4, 11–14]

Force steadiness, target tracking, relaxation and selective joint control are additional constructs. They may be impaired even when the person can generate substantial maximum force. They should be reported in their own units and tested under a defined target force and feedback condition. They are not interchangeable with power or the Fugl-Meyer motor score.

Table 1 Constructs and outputs

Distinguish voluntary weakness from resistance to passive movement

Spasticity and voluntary strength should be recorded separately. A person can have weak voluntary plantarflexion while displaying increased resistance during passive dorsiflexion. A dynamometer used to measure that passive resistance is not thereby measuring voluntary strength. Passive tissue stiffness, stretch-reflex activity, background contraction and speed-dependent responses may all contribute to the resistance. A Modified Ashworth score does not decompose those contributions.

Likewise, low voluntary torque is not a direct assay of muscle mass or a pure measurement of central drive. Pain, understanding, attention, apprehension, fatigue, motor-unit activation and antagonist activity can reduce the observed value. Burst-superimposition methods can investigate latent force-generating capacity and voluntary access to it, but their corrections and stimulation assumptions introduce additional uncertainty. Surface EMG amplitude alone cannot establish how much of the anatomical maximum force remains inaccessible. [15, 16]

The practical consequence is to combine force measurement with a short clinical description: passive range, pain, tone, selective movement, sensation and ability to follow the task. A motor-impairment scale may provide valuable context, but its ordinal movement categories cannot replace a continuous measurement of a specific action.

Separate measurement claims from outcome claims

Reliability describes reproducibility under particular conditions. ICCs are influenced by between-person variability; broad strength ranges can produce excellent ICCs despite clinically relevant within-person differences. Absolute error should therefore accompany relative reliability. The SEM of measurement, SD of repeated-measure differences, limits of agreement and MDC have different definitions from the standard error of a sample mean.

For two comparable measurements with independent, similarly distributed errors, a conventional MDC95 is 1.96 × square root of 2 × SEM. Equivalently, when appropriate assumptions hold, the half-width based on repeated differences is 1.96 × SD of those differences. Neither expression proves clinical importance. An MDC calculated for an averaged score should not automatically be used with one trial, a different device or a different assessor. Systematic retest bias should remain visible rather than being silently folded into a symmetric “real change” rule.

An anchor-based MIC or MCID asks how much change is important under a stated external criterion. That criterion may be the patient's perception of improved arm function rather than a measured change in isolated force. It is expected to depend on baseline severity, direction of change, phase, context and the anchor. Where the anchor sample is small, a single point estimate is a provisional guide rather than a biological boundary.

Concurrent association asks whether force and function measured at broadly the same time vary together. Genuine prognosis requires the assay to precede a later outcome. Change–change association across a rehabilitation interval is longitudinal but still does not establish that the baseline test predicts the later outcome. Prediction-model usefulness further requires validation, calibration, uncertainty and evidence of added decision value beyond readily available clinical information.

Evidence sources and review landscape

Search coverage and its limits

A native Scopus search of stroke or poststroke or hemiparesis combined with dynamometry, grip, power or rapid-force terms and reliability, validity, prediction or prognosis, with an explicit publication-year restriction below 2027, returned 243 unique records across ten pages. DOI-or-normalized-title reconciliation across these main searches yielded 889 records. Search results were evaluated as available on the search date, with online and issue years distinguished when known.

These are bounded, reproducible searches, not an exhaustive claim about the field. General uses of “stroke” returned swimming, engine and cardiovascular-risk records; incident-stroke studies in initially stroke-free populations were not treated as prognosis after stroke. Focused searches, public repositories and backward and forward citation chasing supplemented the main queries. The early strength review, early grip study and handheld RTD study were used as citation leads. Citation retrieval was provider-bounded; one version-specific RFD DOI was not resolved by OpenAlex. No claim of complete citation-network coverage is made.

Primary full methods and results were preferred. Public full text was obtained through Europe PMC/PMC, original publishers or institutional repositories. Important subscription sources still represented by abstracts or publisher extracts are explicitly labelled in the reference list and matrices. Those sources support the reported high-level findings, not unreported protocols, covariate sets, calibration or numerical individual-change rules.

What the existing reviews add

Kristensen and colleagues' review targeted criterion isokinetic dynamometry, rather than the complete clinical strength-assessment landscape. The paper appeared online in 2016 and in the 2017 journal issue; its reported 3 February 2015 search-end date was not independently verified. Twenty studies with 316 participants supported a broad pattern of paretic weakness and generally high relative reliability. Its abstract's 7–20% SEM statement should not be converted into a universal individual MDC. Full review tables were not available in this appraisal, so exact subgroup claims were checked against accessible originals. [17]

Beckwée and colleagues synthesized repeated muscle measurements within the first three months in 38 studies involving 1,097 participants. Strength generally improved while muscle thickness could decline, with smaller changes on the nonparetic side. That discordance cautions against equating stronger voluntary performance with restored muscle tissue. Recovery, rehabilitation exposure and testing effects coexist. Group-average change did not identify distinct individual recovery trajectories. [19]

The muscle-power review by Knight, Saunders and Mead covered six studies and 171 participants, with searches ending in July 2012. It established an early signal of bilateral power impairment and possible walking associations. It was not a contemporary validation of one standardized power test. Subsequent instrumented leg-press, cycling and rapid-force studies substantially expand the methods while leaving prospective clinical utility uncertain. [20]

Mentiplay and colleagues' gait review focused on univariate associations with isometric strength. Its stronger ankle-dorsiflexor than knee-extensor association after considering study size and quality is clinically informative, but univariate correlations do not establish incremental prognostic value or identify the best treatment target for every patient. Muscle choice, walking speed instruction, assistive devices and severity differ across studies. [21]

Handgrip and clinically relevant upper limb strength

What grip can and cannot represent

Grip is accessible, quantitative and often feasible when a full multijoint assessment is impractical. Paretic grip captures an important component of hand force. Less affected grip can describe an additional reserve relevant to transfers, aids and ADL, while also reflecting age, body size, premorbid health and global illness burden. It should not be described as an uncontaminated measure of general muscle health after a unilateral stroke.

Hand function requires more than squeeze force. Opening the hand, regulating force, individuating fingers, transporting the arm, sensing contact and using the limb in daily life can each limit activity. Conversely, a person with little measurable paretic grip may accomplish some tasks using proximal support or compensation. Correlation of grip with an arm score does not make grip a complete replacement for motor control or dexterity testing.

The floor of the chosen device matters. An instrument that cannot register a weak grip is not evidence that all grip force is absent. Record inability to hold or close around the handle, pain-limited effort and comprehension failure separately. A credible measured zero should be distinguished from an untestable value; the reason for missingness carries clinical information and affects the population to which reliability findings apply.

Early stroke and the problem of selective completion

Bertrand and colleagues followed 34 completers from 48 enrolled participants through 12 weeks, using a Jamar, handle position two and the mean of three trials. Paired sessions one or two days apart occurred at weeks 2, 4, 8 and 12. Paretic reliability analyses excluded unmeasurable grip: the available samples grew from 21 to 26. Paretic MDD95 fell from 4.11 kg at week two to 2.97 kg at week 12. In the subgroup below 20 kg, week-two MDD was smaller in kilograms but 41% of the subgroup mean, showing why a good ICC can coexist with substantial relative uncertainty. [22]

Sixteen participants had no measurable grip at week one; eight had regained some by week eight and nine by week 12. The study therefore supports continued reassessment rather than a fixed prognosis from an early floor. It also recorded small calibration offsets in some devices. Keeping the same dynamometer within each weekly pair protected short-interval reproducibility, but does not erase possible cross-device effects across clinical settings. Attrition, the relatively young sample and ongoing therapy limit generalization. [22]

Aguiar and colleagues provide openly inspectable grip, pinch and trunk dynamometry data in 32 people 3–6 months after stroke. Their definition of “subacute” is narrower and later than many contemporary studies. Although 24 returned, only 13 contributed paretic handgrip retest data. The sample was predominantly mildly impaired. First-trial and averaged values had similar group means, but first-trial paretic grip MDC95 was 4.64 kg, compared with 2.63 kg for the mean of two trials and 3.69 kg for the mean of three. Similar means therefore do not demonstrate equal individual precision. Familiarization and an independent recorder strengthen standardization; continued rehabilitation over a 1–2-week interval complicates the stability assumption. [23]

The defensible clinical lesson is not that every patient needs exactly three trials or that one trial is always sufficient. A small, tolerable number of standardized trials can reveal inconsistency and permits a declared summary rule. If burden requires one trial, the interpretation should use evidence for that rule rather than an error threshold established for averages.

Chronic stroke reliability and device specificity

Ekstrand and colleagues assessed 45 people with mild-to-moderate upper-limb paresis more than six months after stroke. They could bring the hand to the forehead and grasp/release a block, limiting application to severe paresis. Grippit testing used three 3-second efforts with 60-second rests, retained the maximum and was repeated one week later. Grip ICCs were 0.96 for the more affected hand and 0.95 for the less affected hand. Corresponding SRDs were 61.8 and 73.3 N, or 25.6% and 21.0% of cohort means. These are protocol-specific error estimates, not Jamar thresholds or MICs. [24]

The same study found high reliability for fixed isometric shoulder and elbow torque and isokinetic elbow torque. Some dynamic measurements showed retest gains and larger relative errors. Thus, adding an upper-limb torque assay may help answer a specific proximal weakness question, but it also introduces positioning, gravity correction, movement and familiarization requirements. It should not be added solely to generate a larger battery.

Chen and colleagues' earlier mixed stroke sample illustrates another modifier: hypertonicity. The summary reports grip SRDs of 2.9 kg and 4.7 kg for more and less affected hands, with larger relative errors for some pinch tests and greater error in participants with hand spasticity. Those values should not be selected merely because they are smaller than another study's estimates; instrument, impairment, sample and interval differ. [25]

Important change in early stroke

Lang and colleagues estimated anchor-based important change in 52 participants early after stroke, using mean Jamar grip with a fixed handle position and a patient rating of change in the affected arm. Assessments averaged 9.5 and 25.9 days after stroke. The often-cited 5.0-kg and 6.2-kg estimates refer to the affected dominant and affected nondominant hands, respectively. They are not affected-versus-unaffected thresholds. Only 12 participants occupied the minimally meaningful improvement category, and change distributions overlapped across perceived-change groups. These estimates should inform discussion of early improvement, not trigger a universal binary result across chronic stroke, devices or both hands. [26]

A practical report can state both observed change and whether it exceeds a compatible error estimate, then discuss patient-important function separately. If the compatible MDC is uncertain or larger than the proposed MIC, say so. A change can be important to the patient while remaining difficult to separate from measurement error in one measurement pair.

Newer digital grip devices need the same scrutiny

The 2026 InGrip comparison included 50 stroke outpatients with mainly mild upper-limb weakness. Two efforts within the same session produced high affected-side ICCs for InGrip and Takei, with SRDs around 3.3 kg. Yet between-device limits of agreement extended approximately from −6.09 to 4.36 kg despite a small mean bias. This is insufficient evidence for freely switching devices when tracking modest individual change. Standing extended-elbow measurement was changed to seated 90-degree elbow flexion for participants unable to stand, and longer-term reliability was not established. [27]

The 2025 Biometrics study included 100 people with chronic first ischaemic stroke and basic gripping ability. Its short-interval ICCs were high, but its tables label SEM as the standard error of the mean; the displayed values match SD divided by the square root of 100. The discussion then treats these as measurement SEMs and derives an MDC using 1.96 × SEM. That does not establish the conventional two-measurement MDC. The proposed 2.55-kg grip-change rule should therefore not be implemented. The directly reported Bland–Altman limits remain more informative, although they concern five-minute repeats and right/left groups rather than a clear paretic-side analysis. [2]

Device accuracy, clinical reliability and interchangeability are separate questions. Bench calibration is necessary but cannot reproduce variable human grip, altered handle contact or position. A device's marketing resolution should never be substituted for clinical measurement error.

Table 2 Selected grip reliability and change evidence

MDD, MDC and SRD are retained as the original authors label them. They concern detectable change, not clinical importance. Units are source units, not converted automatically.

Lower limb dynamometry and maximum dynamic strength

Handheld testing is a protocol rather than a device category

Handheld dynamometry offers portable quantitative assessment, but the examiner must maintain the intended geometry and resist the participant's force. A make test asks the person to push against a stationary sensor. A break test progressively overcomes the held position and introduces a different mechanical task. Their results should not be treated as interchangeable.

Fixation can reduce examiner limitations, but a belt-fixed protocol is not validated simply because an otherwise similar handheld protocol has a high ICC. Pad shape, placement, force direction, belt anchorage, joint angle, support surface and instruction all matter. For strong plantarflexors or knee extensors, examiner capacity may constrain the observed maximum. For weak muscles, gravity, inadvertent assistance and sensor activation thresholds can become more important. The appropriate response is to standardize and document the setup, not to assume the same error across the force range.

Force normalized to body mass remains force per kilogram; it is not torque per kilogram. If a force is measured more proximally during one session, normalization by body mass does not remove the leverage change. For a digital record, store the raw force, body mass, lever arm and calculated torque as separate fields, with the calculation version visible.

Supine testing in acute stroke

Yen and colleagues tested a wedge-supported supine protocol in 15 patients within seven days of stroke. Hip flexion was 45 degrees, knee flexion 90 degrees and the ankle positioned in slight plantarflexion. Three maximal efforts were averaged, with trained raters and vital-sign monitoring. Participants needed selective lower-limb control and command-following ability; this was not an unselected severe ICU population. [1]

The reported high ICCs support feasibility in that selected sample. However, its absolute-error table requires caution. For tester 1 knee extension, the printed individual data imply a pooled SD of approximately 14.70 lb, rather than the reported 4.31. Applying the stated formula to that pooled dispersion and the rounded ICC gives a substantially different value from the reported 4.60-lb MDC. This independent reconstruction is a reporting-consistency check, not a replacement clinical threshold. The acute 24-hour stability assumption is also uncertain. The safe conclusion is to retain the protocol as preliminary feasibility evidence and withhold its published MDC from automated interpretation pending clarification. [1]

A supine result should not be compared directly with a seated reference. The advantages of testing someone who cannot sit independently are real, but so are differences in muscle length, gravity and stabilization. Document changes in test position during recovery as a break in the measurement series or obtain an overlap assessment when clinically feasible.

Contemporary chronic stroke handheld error estimates

Itoh and colleagues studied 42 chronic-stroke outpatients, with an inter-rater subset of 12. Two three-second efforts were performed on the same day, with sensor removal and repositioning between trials, practice and repeats when compensation occurred. Force was divided by body mass. Reported intra-rater ICCs were 0.989–0.998; MDC95 values were 0.24 N/kg for hip flexion and knee extension, 0.21 for hip abduction, 0.15 for hip adduction and ankle dorsiflexion, and 0.20 for plantarflexion. [28]

Those estimates address short-interval repeatability under the studied conditions. They do not establish between-day rehabilitation monitoring error. Participants unable to generate the relevant movement were excluded from reliability calculations, and severity subgroups were sometimes tiny. The lack of a statistically significant correlation between error and motor score is not proof that error is invariant across all severities. The paper's table headings and some tabulated values also require care. The practical value is a contemporary protocol-specific estimate, with its time interval and eligible population displayed prominently. [28]

Fixed and isokinetic systems improve control but add their own constraints

Eng and colleagues examined 20 chronic-stroke participants across hip, knee and ankle actions. Peak and average torque derived from three consistent curves were highly reproducible, but second-session gains remained. Most participants could not complete ankle testing at the initially selected faster velocity and were tested more slowly. This is an important reminder that “isokinetic testing” may encompass participant-specific speeds, preloads and movement ranges. Apparatus control does not eliminate task learning or eligibility restrictions. [29]

Flansbjer and colleagues provide more informative absolute-error data in 50 community-dwelling participants who could walk at least 300 m, six to 46 months after stroke. Biodex testing included concentric knee extension/flexion at 60 and 120 degrees/s and eccentric extension at 60 degrees/s, usually one week apart. Paretic concentric extension SRD was approximately 33% at 60 degrees/s and 31% at 120 degrees/s; paretic flexion required approximately 48% and 55%. Eccentric extension had smaller relative SRD, around 25%, but fewer participants completed it. A high ICC thus coexisted with substantial uncertainty for small individual changes, especially in knee flexion. [30]

These findings favour matching the measurement to the expected change and clinical question. If an intervention is expected to alter a weak muscle only slightly, a highly reproducible rank ordering is insufficient. An assay with large error may still be useful to identify a gross deficit or prescribe loading, while being poorly suited to detect a small week-to-week improvement.

Gravity correction and axis alignment should be explicit. For an isokinetic test, report contraction mode, angular velocity, range used for analysis, torque extraction rule, preload and whether the participant actually reached the target speed. The highest torque from one repetition is not the same score as peak torque from an ensemble-averaged curve. Absent this information, a published Nm/kg cutoff cannot be safely transferred to a different dynamometer workflow.

Table 3 Lower limb strength evidence for measurement error

One repetition maximum and clinical loading

The stroke-specific 1RM reliability literature retrieved by the focused search was sparse. General-population or Parkinson-disease reliability estimates should not fill that gap. A stroke study that successfully uses 1RM to set power-test loads establishes feasibility in its selected participants; it does not necessarily validate between-day 1RM error or a patient-important change.

Stavric and McNair used unilateral 1RM on a modified supine leg press, then assessed power on a separate day at 30%, 50% and 70% of that maximum. Their participants could walk ten metres and lift the machine's minimum load. Familiarization, progressive load adjustments, rest and technical success defined the result. This is a useful research example of a machine-specific assessment and loading sequence, not a general prescription for all stroke survivors or all leg-press machines. [11]

For rehabtools, a 1RM module is most defensible as a structured record of an assessment chosen and supervised by a suitably qualified clinician. It should retain machine identity, unilateral/bilateral mode, range, assistance, technical failure criteria and load increments. An estimated 1RM from repeated submaximal lifts should be labelled as an estimate with its equation, rather than recorded as a directly observed maximum. A simple ordinal strength grade cannot be converted into a 1RM.

Mechanical power assessment

Bilateral deficits can be larger than the asymmetry suggests

The modified leg-press study by Stavric and McNair included 29 stroke participants and 29 matched controls. Peak power was measured from force and displacement at 1,000 Hz during explosive pushes. Power was lower in both stroke limbs, with the paretic limb most impaired. The largest values occurred at the lowest of the three tested loads. Importantly, the instrument permitted projection from the footplate, so its power values should not be transferred to a machine requiring deliberate deceleration before end range. Reliability came from a pilot of only eight participants and was expressed as ICCs without an accompanying individual MDC. [11]

The study does not establish that training at 30% 1RM is universally optimal. The load producing the largest test value, the load revealing the largest group deficit and the load producing the best functional training adaptation are different decisions. A finite comparison of 30%, 50% and 70% also does not locate the complete power–load optimum for every patient.

Reporting the raw control and limb means is safer than repeating loosely worded percentage comparisons. A percentage decrease from control to stroke has a different denominator from a percentage increase from stroke to control. This seemingly small distinction can substantially distort the apparent severity of impairment.

Cycling power is a coordinated whole task

Kostka and colleagues assessed 67 walking-capable participants two weeks to three months after stroke and 67 age- and sex-matched controls. Instrumented cycling used two eight-second sprints and a fitted power–velocity relationship. Stroke participants generated roughly half the control group's body-mass-normalized maximum power; power correlated with concurrent TUG, Rivermead Motor Assessment and Barthel Index. Selection excluded severe spasticity, restricted range and inability to complete testing. Adjustment was limited, particularly for neurological severity and premorbid fitness. [12]

Although described as quadriceps or knee-extensor power, this cycling result includes bilateral coordination, cadence and the ability to accelerate the ergometer. It does not isolate the paretic knee's maximum power. This is not a flaw if the intended construct is coordinated lower-limb performance; it becomes a problem only when the label suggests an isolated muscle property. Similarly, the fitted optimal cadence is an external task velocity rather than a directly measured fascicle-shortening velocity.

The 2024 repeatability study included 50 stroke participants, mostly one to three months after stroke. Across two within-session sprints, power ICC was 0.93 but SEM was 10.6% and MDC95 46.6 W, or 29.3%; optimal cadence MDC95 was 15.4 rotations/min, or 31.7%. The stroke group had the largest relative error of the clinical and healthy groups studied. Testing compared sprints at different loads within a session, rather than a standard identical-protocol between-day retest. These values should not be presented as validated long-term rehabilitation responsiveness. [13]

Leg extension rigs and asymmetry findings do not all agree

Dawes and colleagues' small pilot used a Nottingham leg-extensor power rig in 14 walking-capable stroke participants, with only nine in the retest subset. Greater between-leg power asymmetry was associated with poorer walking performance. Saunders and colleagues' larger baseline analysis of 66 independently ambulatory participants instead found that power in either limb, rather than relatively small residual asymmetry, explained current activity limitations. The difference is clinically plausible given different samples and ranges, and it prevents a universal claim that symmetry is the central determinant of walking. [31, 32]

Both studies concern association with present performance. Neither establishes a prospectively validated fall-risk threshold or shows that reducing the asymmetry itself causes better function. A ratio may show a strong relationship in a small heterogeneous sample and weak information in a more recovered group. Absolute bilateral capacity, task demands and motor control must remain visible.

Fujita and colleagues' nine-second modified Wingate study provides another distinct assay: recumbent cycling mean power. Its high retest ICC in 28 participants and functional correlations in a smaller validity sample are encouraging. Its correlation with chair-rise time does not convert the test into a chair-rise power equation. These instrument-specific measures should retain their names and protocols. [33]

Rapid force production and activation

The most clinically accessible stroke RTD study

Mentiplay and colleagues studied 63 adults more than three months after stroke who could walk without aids or orthoses. Seven muscle groups were tested using a handheld make test, with the highest of two values retained and torque normalized by body mass. RTD was the greatest slope over any successive 200-ms window, deliberately avoiding contraction-onset detection. Twenty-eight participants returned after intervals ranging from two to 69 days. RTD ICCs were 0.88–0.97; however, absolute error and MDC were not reported. Thirteen participants could not tolerate prone hip-extension testing. [3]

In adjusted concurrent models, RTD did not add significant information beyond strength for fast barefoot walking speed. Strength contributed beyond RTD for most muscle groups. The study explicitly recognized low sampling rate and foam compliance as limitations. Consequently, the result supports a specific practical trace-based measurement in independently walking participants, with a clear functional comparison. It does not validate onset-based early RTD, prove precision for small individual changes, or establish prospective walking recovery. [3]

A further nuance is the reliability coefficient's specification: the paper reports an average-measure ICC notation while selecting maximum trial values. This does not invalidate the results, but it cautions against inferring the precision of a single unsupervised contraction or calculating a new MDC from the published ICC alone.

Rigid dorsiflexion testing measures a different early force construct

Olsen and colleagues analysed two baseline sessions of an experimental dataset. Thirteen participants with chronic stroke contributed usable data; their mobility ranged from unaided walking to wheelchair use. The foot was fixed in 25 degrees of plantarflexion, signals were acquired at 1,961 Hz, onset was identified using a baseline threshold with visual review, and RFD was calculated over the first 200 ms. Mean-of-three MVC had ICC 0.97 and SEM 7%, whereas early RFD had ICC 0.79, a lower confidence bound of 0.48 and SEM 24%. The best-trial RFD was less reliable. [4]

The difference from the handheld study should not be reduced to “rigid is worse” or “handheld is better.” The participants, joint angle, sampling, onset rule and outcome all differed. Resampling a 40-Hz trace cannot restore information never acquired at higher frequency. Conversely, a robust moving-window slope may deliberately sacrifice information about initial activation. Choose the outcome according to the hypothesis, then validate its repeatability and usefulness in that population.

The Olsen dataset also shows why surface EMG requires separate interpretation: its relative error was larger than that of peak force. Electrode placement and biological activation variability affect the signal. A highly repeatable peak force does not automatically validate EMG-derived change or an explanation of its cause. This was a secondary analysis of an experimental cohort, so it should not be counted as an independent replication of that cohort's intervention findings. [4]

Impairment evidence is broader than clinical prediction evidence

Shimose and colleagues found lower knee-extensor RFD in 31 independently ambulant acute-stroke patients compared with 54 controls, despite no significant between-group MVC difference. The reductions were greater in the first 50 ms than over 200 ms and occurred on both sides. This is useful known-groups evidence that rapid force can reveal deficits overlooked by maximum force in a mildly affected sample. It supplies neither an individual diagnostic threshold nor a forecast of future walking or falls; the full original acquisition details remain an access limitation here. [34]

Silva and colleagues' 29-person chronic-stroke study reported larger concurrent walking-speed R-squared values for RTD than peak knee torque, but differences between correlations were not statistically significant. Adding assistive-device use and motor function increased explained variance. Its title's “excellent predictor” language should therefore be read as an exploratory concurrent model, not evidence of externally validated prognosis or proven superiority over maximum strength. [35]

Mechanistic observations strengthen biological plausibility without resolving clinical thresholds. Chou and colleagues recorded motor units during dorsiflexion in a very small chronic-stroke sample, with rapid-pulse analysis available for only four participants. Paretic discharge modulation and rapid force were impaired. Such invasive recordings explain why activation speed merits study, but their sampling, motor-unit selection and participant restrictions preclude a routine clinical cutoff. [16]

Voluntary activation measures may explain why a muscle is weak

The CEDRS portable plantarflexor system combines torque measurement with electrical burst superimposition. Its correction equation was developed in young neurotypical participants; 26 chronic-stroke participants then demonstrated feasible testing and lower paretic central-drive estimates. Associations with six-minute walking performance were concurrent. The authors themselves identify prospective response prediction as future work. [15]

For clinical translation, several limitations matter: the device-comparison accuracy study was not a stroke reliability study; calibration of the activation equation in healthy adults does not guarantee identical validity in altered stroke muscle; some trials were rejected because voluntary force was not steady; and test eligibility excluded several common clinical problems. A correction equation validated by leave-one-out analysis in its development sample is still not an externally validated prognostic model.

This approach is best regarded as a specialized adjunct when differentiating voluntary access from force-generating capacity will change a clinical or research decision. It should not be used to reinterpret every low handheld result as central activation failure, nor should the numerical activation ratio be labelled a percentage of neurological recovery.

Table 4 Power and rapid force protocols

Normalization asymmetry and clinical interpretation

Preserve the denominator and the bilateral values

Four common summaries answer different questions: raw paretic force, raw nonparetic force, their sum, and their ratio or difference. A ratio of paretic to nonparetic force is directionally different from a deficit percentage defined as 100 × (nonparetic − paretic)/nonparetic. A stronger-to-weaker ordering discards the anatomical side and can switch labels over time. Each definition should be visible beside the result.

As an illustrative calculation, paretic/nonparetic values of 20/40 and 10/20 yield the same ratio, yet the second person has half the measured bilateral force. If a person changes from 10/20 to 12/24, both limbs improve while the ratio is unchanged. These are arithmetic examples, not clinical cutoffs. They show why a symmetry-only interface can conceal meaningful recovery or bilateral deconditioning.

Nonparetic muscles may be weaker than healthy controls, but the direction is not uniform across all actions and contraction modes. Chronic-stroke studies have also described relatively preserved eccentric torque and compensatory strength patterns. “Unaffected” should therefore remain a conventional anatomical label, not an assertion of normal physiology. [36, 37]

Body-mass normalization can help relate force or power to the mass a person must move, but it does not remove sex, stature, limb length or body-composition effects. A person losing body mass may improve W/kg or N/kg without gaining absolute power or force. Report both. Scaling by BMI or muscle mass changes the question again and needs explicit justification. A population sarcopenia threshold and a within-person recovery measure should not be merged simply because both include grip.

Do not turn exploratory walking cutoffs into independence rules

A 28-person chronic-stroke study examined paretic and nonparetic isokinetic knee torque, their sum and their difference. It used stepwise polynomial models and data-selected walking-performance percentiles. AUCs around 0.75–0.76 concern classification within this small cross-sectional sample, not subsequent recovery of independent walking. [38]

The original has important internal inconsistencies: the displayed AUC confidence interval does not encompass its point estimate, reported six-minute-walk percentile thresholds differ between text and table, and the side-difference cutoff's units require clarification. These concerns, together with extensive model searching relative to sample size, preclude implementation of the published thresholds as clinical rules. The conceptual lesson that bilateral capacity and asymmetry may carry different information survives; the numerical calculator does not. [38]

This distinction should be explicit in rehabtools. A score can be associated with a walking category without being the threshold at which walking becomes safe or independent. Independence also depends on balance, perception, endurance, aids, environment, supervision and the definition of the outcome. Restricting an assessment to people already able to walk ten metres cannot establish its performance in those who are currently unable to walk.

What strength predicts over time

Separate the timing of the test from the timing of the outcome

For early recovery, record the day after stroke on which strength was measured, not just “admission.” Admission to an acute ward at day two and admission to rehabilitation at day 35 are different biological and clinical starting points. Likewise, “discharge” is an outcome time influenced by length of stay, health-system practice, complications, support and progress. It is not a fixed 90-day outcome.

A useful temporal evidence table has separate columns for assay time, outcome time and analysis type. The same rehabilitation dataset may support a reliability analysis, an impairment association and a later-outcome model. It should not be counted as three independent cohorts. When the original paper gives insufficient information to resolve overlap, label it uncertain rather than assuming independence.

Table 5 Verified cohort overlap and dependence

Strength recovery is not identical to functional recovery

Sunderland and colleagues enrolled 38 recent stroke patients and assessed grip using a sensitive electronic instrument; 31 completed six-month follow-up. The presence of voluntary grip at one month was associated with some functional arm recovery by six months. This historically important observation concerns a selected cohort, a sensitive assay and a broad recovery statement. It is not a modern individualized probability or a justification for limiting rehabilitation when grip is absent early. [39]

Suzuki and colleagues measured bilateral elbow and grip strength repeatedly in 21 inpatients beginning approximately one week after stroke. A logarithmic model using early observations fitted subsequent strength recovery well. That is a model of the measured strength trajectory, not a validated forecast of independence, participation or future falls. The small cohort and short horizon make broad individual extrapolation inappropriate; detailed validation beyond the accessible abstract remains unverified. [40]

Bertrand also provides genuinely time-ordered, although unadjusted, associations: grip measured at week one correlated with week-12 CAHAI at 0.76 and ABILHAND at 0.71. These are stronger evidence of temporal association than same-day correlations, but are not validated prediction models. Later grip assessments were closer to the outcome and cannot be compared as if they were equally early prognostic tests. The ABILHAND recall instruction was modified to the post-stroke period, which also matters when interpreting the outcome. [22]

These studies, together with early change in the muscle review, support repeated assessment during recovery. They do not imply that every patient follows the same smooth trajectory. Inattention, medical complications, measurement floors, pain, fatigue and treatment exposure can all alter observations. A changing ratio may reflect recovery on both sides rather than only neurological restitution in the paretic limb. [19]

Admission grip and later inpatient ADL

Yi and colleagues retained 127 patients from 1,065 screened admissions after excluding recurrent/bilateral stroke, missing early data and severe cognitive impairment. Grip was the mean of three efforts with 30–60-second rests. Rehabilitation admission occurred on average 15.7 days after stroke, with MBI reassessment about 15 days later. This is short-horizon, retrospective prognosis rather than a long-term prospective cohort. [5]

In the final discharge-MBI model, less affected grip had an unstandardized coefficient of 0.35 points/kg (SE 0.15, p=0.019), alongside age, NIHSS, lesion side and admission MBI; adjusted R-squared was 0.59. In the gain model its coefficient was 0.55 (SE 0.15, p<0.001), but adjusted R-squared was only 0.14. These complete-model values cannot be attributed to grip alone. Backward selection after univariate screening, the highly selected sample, absent hand-dominance data and lack of external validation or calibration limit deployment. The result supports a short-term association after adjustment, not a treatment effect or individual recovery guarantee. [5]

Matsushita and colleagues analysed 699 patients aged at least 65 years in one Japanese rehabilitation hospital. Grip was measured seated with the arm straight using a Takei Smedley instrument, retaining the higher of two attempts from the dominant hand or nonparalyzed hand in hemiplegia. Admission was typically three to four weeks after onset. There were 448 home discharges, with median stays near three months. Grip remained associated with discharge FIM and home discharge in sex-specific adjusted analyses. [6]

The reported home-discharge cutoffs were 15.1 kg for men and 9.5 kg for women, with derivation AUCs of 0.789 and 0.819. Reported sensitivity/specificity were 0.573/0.869 and 0.672/0.822. These estimates were selected using the same sample and are not externally validated or calibrated rules. Importantly, adjusted models included an energy-intake average using admission and discharge data. That covariate is unavailable at admission, so those fitted models cannot be presented as deployable baseline-only prognostic tools. Other concerns include exclusion tied to obtaining body-composition data, incomplete adjustment for neurological severity and cognition, and the social determinants of home discharge. [6]

The correct clinical interpretation is that less affected grip may contribute information about available reserve and illness severity. It is not interchangeable with paretic motor recovery. Nor can a home-discharge association be converted into a rule for denying community rehabilitation, allocating support or predicting safety at home.

Recent grip cohorts confirm association more readily than incremental value

A 2026 single-centre cohort of 180 ischaemic-stroke rehabilitation patients measured both hands at admission and discharge. Less affected grip was associated with discharge FIM in sex-stratified models, explaining approximately 13–19% of variance. Crucially, those models did not establish added value beyond admission FIM, neurological severity and cognitive impairment. Hand dominance was unavailable. Applying age- and sex-specific healthy grip percentiles described relative weakness, but did not create a validated stroke prognosis threshold. [8]

This is a useful confirmation of an association, especially because both limbs were measured using a stated protocol. Its regression equations should not be deployed as comprehensive prognostic models. Presenting a later outcome on the left side of an equation does not eliminate confounding, guarantee calibration or quantify what grip adds to routine clinical assessment.

Miyazaki and colleagues' 2026 analysis offers a larger, more developed modelling example. It used 980 rehabilitation patients and 33 candidate variables, including bilateral grip, FIM items, motor assessments and nutrition. A held-out 20% test set followed extensive development resampling. Best model-level R-squared values were 0.776 for discharge motor FIM and 0.758 for total FIM; performance was lower for gains. Less affected grip had feature importance, but these are complete-model results, not grip-alone accuracy. [7]

The model's limitations remain material: one centre, missing-data exclusions, exclusion of acute transfers and variable discharge timing. The age range extended below adulthood, which should be remembered when describing the sample. SHAP values describe contributions within the fitted model; they do not establish a causal effect, a treatment target or an externally calibrated probability. A model with and without grip, evaluated on a new population with calibration and decision analysis, would better establish incremental clinical value. [7]

Later physical quality of life and community reintegration

Cohen and colleagues recruited 75 adults within one month of discharge home and assessed six-month outcomes. Isokinetic torque and power were among a larger group of candidate physical measures. Total paretic lower-limb torque with walking measures explained part of later physical quality of life and physical community reintegration: adjusted R-squared values were 0.30 and 0.47. Explained variance for mental quality of life or social reintegration was much smaller. [9]

This study supports a genuine temporal association extending beyond inpatient discharge. It also shows why participation should not be reduced to impairment. Social opportunities, environment, mood and support may limit reintegration even when strength improves. The sample was already discharged home, selection was stepwise, and the accessible abstract cannot resolve complete adjustment, attrition or validation. No individual participation probability or power-specific prognostic benefit can be inferred from those model-level results. [9]

Current walking associations should remain in their own category

The study of 53 subacute patients by Hyun and colleagues is illustrative. Paretic knee-extensor torque and balance were related to concurrent walking endurance, while balance measures were more prominent in the walking-speed models. All assessments were cross-sectional. The findings identify impairment–function relationships and possible assessment priorities; they do not forecast later recovery. Subgrouping people by their present walking speed also changes the range over which correlations can be observed. [41]

Similarly, the RTD papers and bilateral knee-cutoff study cannot establish walking independence at a future time. In particular, comfortable speed, fast speed, endurance, need for assistance and real-world community walking are different outcomes. A model of one should not be relabelled as the others.

Falls and other unsupported prognostic extensions

No independently validated, calibrated standalone dynamometric strength or power threshold for future post-stroke falls is supported by the sources appraised here. This is a statement about the assembled evidence, not proof that strength has no role in fall mechanisms or that no relevant study exists. Stroke fall prediction frequently relies on balance, prior falls, assistance and motor scales, with strength included as one candidate rather than a validated isolated output.

A strength difference between people with and without a past fall does not establish future-event prediction. A physiological profile containing knee strength does not validate knee strength alone. A treatment study reporting improved balance or a laboratory “fall-risk” score has not necessarily shown fewer prospectively observed falls. General ageing, Parkinson-disease and mixed-population falls findings should not supply missing stroke thresholds.

The same caution applies to mortality, recurrent vascular events and sarcopenia composites. Population grip studies that predict the first stroke in initially stroke-free adults address a different question. A composite of muscle mass, strength and function may be prognostic without establishing the independent contribution of a particular grip protocol.

Table 6 Timing and prognostic meaning

R² refers to the complete reported model, not the isolated strength assay. No source supports replacing a multifactorial clinical assessment with a single cutoff.

Practical assessment and interpretation for rehabtools

A decision sequence for choosing the test

Begin by asking what result could change the plan. If the question is whether a hand is gaining usable force, select a reproducible grip assay and pair it with relevant activity assessment. If the question is why walking or transfers remain limited, choose lower-limb actions linked to the observed task problem rather than assuming grip stands for the leg. If loading on a specific machine is the decision, a supervised machine-specific strength assessment may be more direct. If a research question concerns rapid activation, use a time-resolved protocol with sufficient acquisition and explicit signal processing.

Next establish whether the person can safely and meaningfully complete the task. Check clinical stability, pain, range, ability to assume the position, comprehension and the need for assistance. Standardize communication and demonstration; do not mistake aphasia for inability to understand. Severe neglect, perceptual problems or fluctuating arousal may invalidate a seemingly simple maximal-effort instruction. Record the reason for a non-completed test, rather than silently excluding it or forcing a numerical value.

Then choose an established protocol as a whole. Keep the instrument, handle setting or pad location, joint angle, stabilization, trial number, rest, instruction and score rule stable. Deviations may be necessary and clinically sensible, but they should be visible. When recovery permits a better position, do not conceal that methodological change inside a continuous trend line.

Finally interpret the result in layers: observed raw force or torque; normalized and bilateral context; change relative to compatible error; relevance to the patient's functional goal; and, only when supported, prognostic information. The measurement can be useful even when no validated cutoff exists.

What every measurement record should preserve

The minimum record is more than a value and date. Store stroke onset date, test date, anatomical side and affected status separately. Include premorbid hand dominance where relevant. Retain device model, calibration/check status, force unit, posture, joint positions, fixation, contact location and lever arm. For each contraction store the value, successful completion and quality note, plus the declared maximum/mean/other score rule.

For grip, record handle setting and whether the device was supported. For handheld testing, distinguish make from break and manual from external fixation. For isokinetic testing, include mode, target speed, achieved usable range and gravity correction. For power, include load, velocity source, calculation method and whether output is peak, mean or model-derived. For RFD/RTD, add native sampling rate, filter, onset rule, window, pretension rule and trace-quality decisions.

Record pain, tone, fatigue, comprehension support and reasons for incomplete testing. Walking outcome records should separately state aids, orthoses, physical assistance, footwear and speed instruction. The strength test does not inherit those walking conditions automatically.

Table 7 Minimum reproducible clinical protocol record

Sensible interpretation of change

A change report should first show the actual difference, with both session values and units. If a matching measurement-error estimate exists, identify its source and scope. “Exceeds the reported MDC for this protocol” is more accurate than “clinically significant improvement.” If the protocol or population differs materially, use the evidence as context rather than an automated pass/fail boundary.

Do not compute a patient-specific percentage MDC simply by applying a cohort SRD percentage to any baseline value. Those relative values may have been calculated from a cohort mean, not from a proportional-error model. For very small baseline force, percentage change can become unstable; preserve the absolute change. A transition from unmeasurable to measurable should be explicitly described rather than assigned an infinite percentage gain.

Revisit the raw trace when a surprising change occurs. Check pad placement, lever arm, unit conversion, trial quality and whether the person developed a different compensation. Confirming measurement conditions is preferable to immediately attributing a sudden increase to neural recovery or a decrease to deterioration.

A repeated baseline can be useful when learning, fatigue or an unfamiliar machine makes the first session atypical. It should be planned proportionately; repeated maximal testing itself may create burden or practice effects. The evidence does not support unlimited repetitions in pursuit of a desired score.

What can be implemented now

A clinically useful first version can support bilateral grip and a selected set of lower-limb dynamometry records with explicit protocol templates, raw results and notes. It can provide educational descriptions of compatible reliability and error evidence. It should allow reporting without a categorical interpretation when the evidence does not match.

Advanced power and RTD modules should be instrument-specific. Importing a trace should not silently label every steep slope as early RFD, infer high-frequency accuracy from interpolated data, or apply a generic filtering recipe. A quality-controlled analysis should retain the raw trace and the processing specification so results can be reviewed or reproduced.

Prognostic summaries should distinguish an independent association, a development model and an externally validated clinical model. A source's term “predictor” should not determine the product label. The current evidence supports contextual education about later ADL and participation associations; it does not support a universal automated probability of walking, home discharge or falling from one strength result.

Training context should remain separate

Weakness and power deficits are potential rehabilitation targets, but a training response is not a measurement-validation study. A larger mean force after training does not, by itself, establish individual responsiveness, MIC or transfer to everyday function. Improvement on the trained machine may partly reflect task learning and does not imply equivalent improvement in walking, hand use or participation.

Equally, an observational association between grip and discharge outcome does not establish that adding grip exercise will change that outcome. Intervention choice should integrate the person's goals, motor control, medical status and the relevant treatment literature. This report supports assessment decisions and interpretation, not an individualized exercise prescription.

Evidence gaps and research priorities

The clearest gap is the distance between reproducible measurement and useful individual decisions. Future studies should recruit across severity and report how many people cannot complete each assay, why they cannot complete it and how those values are handled. Acute and subacute reliability studies need an explicit strategy for separating genuine recovery from measurement noise. Chronic protocols need replication beyond highly selected walking-capable samples.

Absolute error should be reported with a reproducible variance model and checked against raw repeated differences. Investigators should distinguish the SEM of measurement from the standard error of a mean, provide uncertainty, disclose systematic retest changes and ensure units and denominators are internally consistent. Grip and lower-limb MIC research should use credible anchors and sufficiently large minimally changed groups, with separate analysis of improvement and deterioration where possible.

For power and RTD, head-to-head work should hold the participants and mechanical task constant while changing the instrument or analysis rule. This would clarify whether differences reflect technology, severity, onset detection or an intentionally different construct. More reliable measurement is not necessarily more functionally informative; both need evaluation.

Prognostic work should prespecify the clinical use, predictor time and fixed outcome horizon. It should report complete adjustment, missingness, event counts where relevant, discrimination, calibration and validation in a new setting. Grip or torque should be evaluated for added value beyond baseline disability, neurological impairment and readily available clinical factors. Treatment-response prediction should test an interaction with treatment rather than assuming that a baseline association identifies who benefits most.

The immediate clinical position is therefore measured but positive: strength and selected power assessments can make rehabilitation reasoning more precise. Their value increases when the protocol is explicit and the interpretation is modest. The most consequential mistake would be to turn technically impressive ICCs, convenient ratios or exploratory cutoffs into certainty about an individual person's recovery.

Primary study evidence matrix

The following grouped tables summarize the original studies underlying the assessment and outcome chapters. Access labels distinguish full original appraisal from abstract-supported findings. Repeated reports and multiple analyses of one cohort are not independent replication.

Table 8 Grip and upper limb studies

Table 9 Lower limb strength studies

Table 10 Mechanical power studies

Table 11 Rapid force and activation studies

Table 12 Temporal and functional association studies

Search strategy and bounded retrieval

Dynamometry grip and assessment or prognosis

(stroke[Title/Abstract] OR poststroke[Title/Abstract]) AND (dynamomet*[Title/Abstract] OR "grip strength"[Title/Abstract] OR "muscle power"[Title/Abstract] OR "rate of force"[Title/Abstract]) AND (reliab*[Title/Abstract] OR valid*[Title/Abstract] OR predict*[Title/Abstract] OR prognos*[Title/Abstract])

Provider total 323; 7 pages requested.

Lower limb strength and assessment or prognosis

(stroke[Title/Abstract] OR poststroke[Title/Abstract] OR hemipar*[Title/Abstract]) AND ("knee strength"[Title/Abstract] OR "knee extensor"[Title/Abstract] OR "plantar flexor"[Title/Abstract] OR "muscle strength"[Title/Abstract]) AND (reliab*[Title/Abstract] OR predict*[Title/Abstract] OR prognos*[Title/Abstract])

Provider total 328; 7 pages requested.

Power and rapid force

(stroke[Title/Abstract] OR poststroke[Title/Abstract] OR hemipar*[Title/Abstract]) AND ("rate of force"[Title/Abstract] OR "rate of torque"[Title/Abstract] OR "muscle power"[Title/Abstract] OR "leg power"[Title/Abstract] OR "rapid force"[Title/Abstract])

Provider total 189; 4 pages requested.

Scopus complementary assessment search

TITLE-ABS-KEY((stroke OR poststroke OR hemiparesis) AND (dynamometry OR "hand grip" OR handgrip OR "rate of force" OR "rate of torque" OR "muscle power") AND (reliability OR validity OR prediction OR prognosis)) AND PUBYEAR < 2027

Provider total 243; 10 pages requested.

Reconciliation and selection

The PubMed pages overlapped, returning fewer unique identifiers than their advertised totals. Official NCBI ESearch and EFetch recovered all 840 query-specific records, yielding 728 unique PubMed records. The ten Scopus pages returned 243 unique records. DOI-or-normalized-title reconciliation yielded 889 records across the main searches. Additional focused searches were discovery checks rather than claims of complete coverage.

Eligibility focused on established stroke and objective voluntary strength, power or rapid-force assessment. General-population prediction of incident stroke, swimming and engine-stroke records, nonhuman work and intervention papers without relevant assessment information were not used as post-stroke prognostic evidence. Full-text appraisal was prioritized for clinically consequential measurement and outcome claims. Selection was targeted and not independently duplicated; the review does not claim a formal pooled estimate of all eligible evidence.

Citation checking included the references of the criterion-dynamometry review and early-grip study, and forward citations of the handheld RTD study, using a maximum of 100 records per provider. These are bounded citation samples, not complete networks. The version-specific Olsen DOI did not resolve through OpenAlex. Full-text retrieval used lawful public and institutional routes. An unavailable original remained abstract-labelled, and no missing method or validation result was inferred from its title or citation count.

References

References are numbered in first citation order. Study specific source descriptions identify the material examined and do not constitute a study quality rating. Links identify the original publication or the explicitly named primary source version.

1. Yen HC, Luh JJ, Teng T, Pan GS, Chen WS, Hsun CC, et al. Reliability of lower extremity muscle strength measurements with handheld dynamometry in stroke patients during the acute phase: a pilot reliability study. Journal of physical therapy science. 2017;29(2):317. DOI 10.1589/jpts.29.317 Source examined: Full text. Europe PMC; MDC calculation unresolved.

Source note: SRC-2b5111db4813 Yen HC 2017

2. Leszczak J, Pniak B, Baran J, Guzik A. Reliability of biometric devices for measuring hand grip and finger pinch strength in stroke patients over 50: a prospective observational study. Scientific reports. 2025;15(1):29188. DOI 10.1038/s41598-025-12712-1 Source examined: Full text. Europe PMC; measurement error interpretation disputed.

Source note: SRC-b7ebe194ca66 Leszczak J 2025

3. Mentiplay BF, Tan D, Williams G, Adair B, Pua YH, Bower KJ, et al. Assessment of isometric muscle strength and rate of torque development with hand-held dynamometry: Test-retest reliability and relationship with gait velocity after stroke. Journal of biomechanics. 2018;75:171. DOI 10.1016/j.jbiomech.2018.04.032 Source examined: Full text. UniSC open access accepted manuscript PDF.

Source note: SRC-4f8fe562a351 Mentiplay BF 2018

4. Olsen S, Taylor D, Niazi IK, Mawston G, Rashid U, Alder G, et al. Reliability of ankle dorsiflexor muscle strength, rate of force development, and tibialis anterior electromyography after stroke. F1000Research. 2023;12:423. DOI 10.12688/f1000research.132415.3 Source examined: Full text. Europe PMC version 3; indexed under 2023 with version/revision later.

Source note: SRC-7734cfa97b2b Olsen S 2023

5. Yi Y, Shim JS, Oh BM, Seo HG. Grip Strength on the Unaffected Side as an Independent Predictor of Functional Improvement After Stroke. American journal of physical medicine & rehabilitation. 2017;96(9):616. DOI 10.1097/phm.0000000000000694 Source examined: Full text. Licensed original Ovid PDF recovered through institution.

Source note: SRC-61cb6702e8ab Yi Y 2017

6. Matsushita T, Nishioka S, Yamanouchi A, Okazaki Y, Oishi K, Nakashima R, et al. Predictive ability of hand-grip strength and muscle mass on functional prognosis in patients rehabilitating from stroke. Nutrition (Burbank, Los Angeles County, Calif.). 2022;102:111724. DOI 10.1016/j.nut.2022.111724 Source examined: Full text. Licensed publisher HTML recovered through institution; unpaginated.

Source note: SRC-90b715e954a9 Matsushita T 2022

7. Miyazaki Y, Oba J, Ishikawa T, Kawakami M, Kondo K, Hirabe A, et al. Explainable machine learning prediction of functional independence measure scores and gain in subacute stroke survivors. Journal of neuroengineering and rehabilitation. 2026;23(1). DOI 10.1186/s12984-026-01963-2 Source examined: Full text. Europe PMC main article; no external validation.

Source note: SRC-edc500ad00b2 Miyazaki Y 2026

8. Lileikyte, Eidmantė, Lukaševič, Karolina, Savickas, Raimondas, Petrusevičienė, Laura. Ipsilesional Handgrip Strength and Functional Outcomes After Stroke: A Retrospective Cohort Study. Journal of clinical medicine. 2026. DOI 10.3390/jcm15155925 Source examined: Full text. Europe PMC and official PMC HTML.

Source note: SRC-50ca6283ae1b Lileikyte 2026

9. Cohen JW, Ivanova TD, Brouwer B, Miller KJ, Bryant D, Garland SJ. Do Performance Measures of Strength, Balance, and Mobility Predict Quality of Life and Community Reintegration After Stroke? Archives of physical medicine and rehabilitation. 2018;99(4):713. DOI 10.1016/j.apmr.2017.12.007 Source examined: Abstract only; public and institutional search found no licensed full-text route.

Source note: SRC-71a51e031c87 Cohen JW 2018

10. Bernhardt J, Hayward KS, Kwakkel G, Ward NS, Wolf SL, Borschmann K, et al. Agreed definitions and a shared vision for new standards in stroke recovery research: The Stroke Recovery and Rehabilitation Roundtable taskforce. International Journal of Stroke. 2017;12(5):444–450. DOI 10.1177/1747493017711816 Source examined: Full text. Official institutional-repository PDF; duplicate consensus publication not counted separately.

Source note: SRC-fda47248659a Bernhardt J 2017

11. Stavric VA, McNair PJ. Optimizing muscle power after stroke: a cross-sectional study. Journal of neuroengineering and rehabilitation. 2012;9:67. DOI 10.1186/1743-0003-9-67 Source examined: Full text. Europe PMC.

Source note: SRC-47b87fe024bc Stavric VA 2012

12. Kostka J, Niwald M, Guligowska A, Kostka T, Miller E. Muscle power, contraction velocity and functional performance after stroke. Brain and behavior. 2019;9(4):e01243. DOI 10.1002/brb3.1243 Source examined: Full text. Europe PMC.

Source note: SRC-90ab045a2193 Kostka J 2019

13. Kostka T, Kostka J. Feasibility and Reliability of Quadriceps Muscle Power and Optimal Movement Velocity Measurements in Different Populations of Subjects. Biology. 2024;13(3). DOI 10.3390/biology13030140 Source examined: Full text. Europe PMC.

Source note: SRC-821284d8d3c5 Kostka T 2024

14. Maffiuletti, Nicola A, Aagaard, Per, Blazevich, Anthony J, Folland, Jonathan, Tillin, Neale, Duchateau, Jacques. Rate of force development: physiological and methodological considerations. European journal of applied physiology. 2016. DOI 10.1007/s00421-016-3346-6 Source examined: Abstract and methodological context. General RFD methods source; no stroke threshold transferred.

Source note: SRC-3fc01b123ef6 Maffiuletti NA 2016

15. Collimore AN, Alvarez JT, Sherman DA, Gerez LF, Barrow N, Choe DK, et al. A Portable, Neurostimulation-Integrated, Force Measurement Platform for the Clinical Assessment of Plantarflexor Central Drive. Bioengineering (Basel, Switzerland). 2024;11(2). DOI 10.3390/bioengineering11020137 Source examined: Full text. Europe PMC.

Source note: SRC-b5c41e8f69f2 Collimore AN 2024

16. Chou LW, Palmer JA, Binder-Macleod S, Knight CA. Motor unit rate coding is severely impaired during forceful and fast muscular contractions in individuals post stroke. Journal of neurophysiology. 2013;109(12):2947. DOI 10.1152/jn.00615.2012 Source examined: Full text. Official PMC HTML.

Source note: SRC-04d4641ec1c9 Chou LW 2013

17. Kristensen OH, Stenager E, Dalgas U. Muscle Strength and Poststroke Hemiplegia: A Systematic Review of Muscle Strength Assessment and Muscle Strength Impairment. Archives of physical medicine and rehabilitation. 2017;98(2):368. DOI 10.1016/j.apmr.2016.05.023 Source examined: Abstract and publisher extracts. Full original review unavailable after public and institutional search.

Source note: SRC-25798cc91474 Kristensen OH 2017

18. Rabelo M, Nunes GS, da Costa Amante NM, de Noronha M, Fachin-Martins E. Reliability of muscle strength assessment in chronic post-stroke hemiparesis: a systematic review and meta-analysis. Topics in stroke rehabilitation. 2016;23(1):26. DOI 10.1179/1945511915y.0000000008 Source examined: Abstract. Full original unavailable.

Source note: SRC-ce4835dc1f2f Rabelo M 2016

19. Beckwée, David, Cuypers, Lotte, Lefeber, Nina, De Keersmaecker, Emma, Scheys, Ellen, Van Hees, Wout, et al. Skeletal Muscle Changes in the First Three Months of Stroke Recovery: A Systematic Review. Journal of rehabilitation medicine. 2022. DOI 10.2340/jrm.v54.573 Source examined: Full text. Europe PMC full article text.

Source note: SRC-7815ea2e7030 Beckwee 2022

20. Knight R.L. Maximal muscle power after stroke: A systematic review. Clinical Practice. 2014. DOI 10.2217/cpr.13.97 Source examined: Full text. Publisher-hosted public PDF.

Source note: SRC-fcac2bd4383a Knight R 2014

21. Mentiplay, Benjamin F, Adair, Brooke, Bower, Kelly J, Williams, Gavin, Tole, Genevieve, Clark, Ross A. Associations between lower limb strength and gait velocity following stroke: a systematic review. Brain injury. 2015. DOI 10.3109/02699052.2014.995231 Source examined: Abstract. Review summary used; original full body not inspected.

Source note: SRC-baa3e53d20b8 Mentiplay 2015

22. Bertrand AM, Fournier K, Wick Brasey MG, Kaiser ML, Frischknecht R, Diserens K. Reliability of maximal grip strength measurements and grip strength recovery following a stroke. Journal of hand therapy : official journal of the American Society of Hand Therapists. 2015;28(4):356. DOI 10.1016/j.jht.2015.04.004 Source examined: Full text. Licensed publisher HTML recovered through institution; unpaginated.

Source note: SRC-496ff02575b7 Bertrand AM 2015

23. Aguiar LT, Martins JC, Lara EM, Albuquerque JA, Teixeira-Salmela LF, Faria CD. Dynamometry for the measurement of grip, pinch, and trunk muscles strength in subjects with subacute stroke: reliability and different number of trials. Brazilian journal of physical therapy. 2016;20(5):395. DOI 10.1590/bjpt-rbf.2014.0173 Source examined: Full text. Europe PMC.

Source note: SRC-2ac74277e3b0 Aguiar LT 2016

24. Ekstrand E, Lexell J, Brogårdh C. Isometric and isokinetic muscle strength in the upper extremity can be reliably measured in persons with chronic stroke. Journal of rehabilitation medicine. 2015;47(8):706. DOI 10.2340/16501977-1990 Source examined: Full text. Official Lund repository PDF.

Source note: SRC-d358fd9550ee Ekstrand E 2015

25. Chen, Hui-Mei, Chen, Christine C, Hsueh, I-Ping, Huang, Sheau-Ling, Hsieh, Ching-Lin. Test-retest reproducibility and smallest real difference of 5 hand function tests in patients with stroke. Neurorehabilitation and neural repair. 2009. DOI 10.1177/1545968308331146 Source examined: Abstract. Article body retrieval incomplete; original abstract supports limited findings.

Source note: SRC-ac69e481195a Chen 2009

26. Lang, Catherine E, Edwards, Dorothy F, Birkenmeier, Rebecca L, Dromerick, Alexander W. Estimating minimal clinically important differences of upper-extremity measures early after stroke. Archives of physical medicine and rehabilitation. 2008. DOI 10.1016/j.apmr.2008.02.022 Source examined: Full text. PMC accepted manuscript and retrieved article body.

Source note: SRC-73bee2794332 Lang 2008

27. Lee BJ, Hong HT, Yu GH, Kim KT. Validity and Reliability of the InGrip® Load-Cell Handgrip Dynamometer Compared With the Takei Handgrip Dynamometer in Healthy Adults and Patients With Stroke. Journal of Korean medical science. 2026;41(20):e147. DOI 10.3346/jkms.2026.41.e147 Source examined: Full text. Europe PMC.

Source note: SRC-bf54b88e3479 Lee BJ 2026

28. Itoh S, Tanikawa H, Kondo H, Ozeki S, Ito T, Fujimura K, et al. Minimal Detectable Change in Muscle Strength Measurements Obtained Using a Hand-Held Dynamometer in Patients with Stroke. Japanese journal of comprehensive rehabilitation science. 2025;16:9. DOI 10.11336/jjcrs.16.9 Source examined: Full text. Europe PMC; same-day repeatability.

Source note: SRC-71a11ff482e9 Itoh S 2025

29. Eng, Janice J, Kim, C Maria, Macintyre, Donna L. Reliability of lower extremity strength measures in persons with chronic stroke. Archives of physical medicine and rehabilitation. 2002. DOI 10.1053/apmr.2002.29622 Source examined: Full text. Official PMC article HTML.

Source note: SRC-e203220d392b Eng 2002

30. Flansbjer UB, Holmbäck AM, Downham D, Lexell J. What change in isokinetic knee muscle strength can be detected in men and women with hemiparesis after stroke? Clinical rehabilitation. 2005;19(5):514. DOI 10.1191/0269215505cr854oa Source examined: Full text. Official Lund repository accepted manuscript PDF.

Source note: SRC-7fa1aa4fe61b Flansbjer UB 2005

31. Saunders, David H, Greig, Carolyn A, Young, Archie, Mead, Gillian E. Association of activity limitations and lower-limb explosive extensor power in ambulatory people with stroke. Archives of physical medicine and rehabilitation. 2008. DOI 10.1016/j.apmr.2007.09.034 Source examined: Abstract. Original full body not inspected.

Source note: SRC-2b0aa96f1b9a Saunders 2008

32. Dawes H, Smith C, Collett J, Wade D, Howells K, Ramsbottom R, et al. A pilot study to investigate explosive leg extensor power and walking performance after stroke. Journal of Sports Science and Medicine. 2005. PubMed. Source examined: Abstract and publisher text. Public JSSM original text located; complete tables not extracted.

Source note: SRC-ed9fbf567a62 Dawes H 2005

33. Fujita T. Reliability and validity of a new test for muscle power evaluation of stroke patients. Journal of Physical Therapy Science. 2011. DOI 10.1589/jpts.23.259 Source examined: Full text. Official J-STAGE PDF.

Source note: SRC-9b61ff582e22 Fujita T 2011

34. Shimose R, Shimizu S, Onodera A, Shibata K, Ichinosawa Y, Enoki I, et al. Decreased rate of leg extensor force development in independently ambulant patients with acute stroke with mild paresis. Journal of biomechanics. 2019;96:109345. DOI 10.1016/j.jbiomech.2019.109345 Source examined: Abstract. Original full text unavailable.

Source note: SRC-6c2f1423aa64 Shimose R 2019

35. Silva RSD, Cerqueira MS, Maciel DG, Silva STD, Figueiredo MCC, Cardoso DCR, et al. Rate of torque development of paretic lower limb is an excellent predictor of walking speed in chronic stroke individuals. Clinical biomechanics (Bristol, Avon). 2022;91:105527. DOI 10.1016/j.clinbiomech.2021.105527 Source examined: Abstract and publisher extracts. Original full text unavailable.

Source note: SRC-60af5a770120 Silva RSD 2022

36. Lomaglio, Melanie J, Eng, Janice J. Nonuniform weakness in the paretic knee and compensatory strength gains in the nonparetic knee occurs after stroke. Cerebrovascular diseases (Basel, Switzerland). 2008. DOI 10.1159/000165111 Source examined: Full text. Official PMC HTML.

Source note: SRC-c16f301d436f Lomaglio 2008

37. Eng, Janice J, Lomaglio, Melanie J, Macintyre, Donna L. Muscle torque preservation and physical activity in individuals with stroke. Medicine and science in sports and exercise. 2009. DOI 10.1249/mss.0b013e31819aaad1 Source examined: Full text. Official PMC HTML.

Source note: SRC-ab5b6b87ae00 Eng 2009

38. Gomes Costa RR, Ribeiro Neto F, Gonçalves CW, Carregaro RL. Accuracy and cut-off points of different models of knee extension strength analysis to identify walking performance in individuals with chronic stroke. Brazilian journal of physical therapy. 2021;25(5):610. DOI 10.1016/j.bjpt.2021.03.002 Source examined: Full text. Official PMC HTML; cutoff reporting inconsistencies.

Source note: SRC-79acc4305bc8 Gomes Costa RR 2021

39. Sunderland A, Tinson D, Bradley L, Hewer RL. Arm function after stroke. An evaluation of grip strength as a measure of recovery and a prognostic indicator. Journal of neurology, neurosurgery, and psychiatry. 1989;52(11):1267. DOI 10.1136/jnnp.52.11.1267 Source examined: Abstract. Original scan not fully inspected.

Source note: SRC-6f0f320bd038 Sunderland A 1989

40. Suzuki M, Omori Y, Sugimura S, Miyamoto M, Sugimura Y, Kirimoto H, et al. Predicting recovery of bilateral upper extremity muscle strength after stroke. Journal of rehabilitation medicine. 2011;43(10):935. DOI 10.2340/16501977-0877 Source examined: Abstract. Official PDF link located but download failed.

Source note: SRC-593cd433a7e0 Suzuki M 2011

41. Hyun CW, Han EY, Im SH, Choi JC, Kim BR, Yoon HM, et al. Hemiparetic Knee Extensor Strength and Balance Function Are Predictors of Ambulatory Function in Subacute Stroke Patients. Annals of rehabilitation medicine. 2015;39(4):577. DOI 10.5535/arm.2015.39.4.577 Source examined: Full text. Europe PMC.

Source note: SRC-7b6fc699bf94 Hyun CW 2015