Time-Course of Changes in Arm Impairment After Stroke: Variables Predicting Motor Recovery Over 12 Months

Subject Recruitment and Clinical Assessment 

Twenty survivors of hemiparetic stroke (64.1±10.8y) were recruited within 4 weeks after a stroke. The characteristics of the subjects are summarized in table 1.

Table 1. Characteristics of Stroke Subjects

	Subject	Age (y)	Sex	Stroke Hemisphere	Stroke Type
	S1	39	Female	Right	Hemorrhage
	S2	73	Female	Left	Hemorrhage
	S3	72	Female	Right	Hemorrhage
	S4	62	Female	Left	Hemorrhage
	S5	78	Male	Left	Hemorrhage
	S6	77	Female	Left	Hemorrhage
	S7	51	Female	Left	Ischemia
	S8	65	Male	Right	Ischemia
	S9	75	Male	Right	Hemorrhage
	S10	60	Male	Left	Hemorrhage
	S11	57	Female	Right	Hemorrhage
	S12	66	Male	Right	Hemorrhage
	S13	61	Male	Right	Ischemia
	S14	70	Male	Left	Hemorrhage
	S15	42	Male	Right	Ischemia
	S16	73	Male	Left	Hemorrhage
	S17	61	Male	Left	Ischemia
	S18	68	Female	Right	Ischemia
	S19	60	Male	Left	Hemorrhage
	S20	72	Male	Left	Ischemia
	Mean ± SD	64.1±10.8	12 Male/8 female	9 Right/11 left	13 Hemorrhage/7 ischemia

Subjects with stroke were drawn from the inpatient service of the Rehabilitation Institute of Chicago. The sample consisted of consecutive cases satisfying the inclusion criteria. Our rate of declination was approximately 10%, primarily because of travel constraints. Our population reflected the stroke population at large, and there was thus no indication of sampling bias.

Patients met the following inclusion criteria: (1) absence of aphasia or cognitive impairment, (2) normative tone and no motor or sensory deficits in the nonparetic arm, (3) absence of severe muscle wasting or of dense sensory deficits in the paretic upper limb, (4) presence of spasticity in the involved elbow muscles, (5) ability to perform even limited elbow extension and flexion at first assessment, and (6) no previous stroke history. All subjects gave informed consent to the experimental procedures, approved by the institutional review board of Northwestern University.

All subjects received intensive PT and OT from our stroke team: 1 hour each of PT and OT for 6 days a week for about 3 weeks, as acute inpatients, and 1 hour 3 times a week each of PT and OT for 2 months after discharge, followed by 1 to 2 hours a week of each for an additional 2 to 3 months thereafter.

Survivors of stroke were assessed clinically before each experiment using the 5-point MAS to assess muscle tone11 and the 66-point FMA to assess motor impairment.8 These scales are widely used clinical assessment instruments.

Experimental Procedure 

A description of the apparatus is provided elsewhere.12 Subjects were strapped to an adjustable chair with the forearm attached to a beam mounted on a torque cell, through a custom fitted fiberglass cast. Shoulder abduction was 80°. The elbow rotation axis was aligned with the axis of the torque sensor and potentiometer.

Subjects moved the forearm voluntarily from full elbow flexion to extension at maximum speed. These movements were repeated 5 times and ensemble-averaged. The experiment was repeated at 5 time points after stroke onset (ie, at 1, 2, 3, 6, and 12 months postinjury).

Elbow position (fig 1A) and torque were recorded with a precision potentiometer and torque transducer. An elbow angle of 90° was set as the neutral position and defined as 0. Position and torque signals were filtered at 230Hz to prevent aliasing and sampled at 1kHz by a 16-bit analog-to-digital converter.

View Large Image Download to PowerPoint

Fig 1. A typical movement trajectory of rapid elbow extension generated by a typical nonparetic and paretic arm at 1 month poststroke for (A) position, (B) velocity, and (C) acceleration. Abbreviations: Ap, peak acceleration; AROM, active range of motion; Vp, peak velocity.

Data Analysis 

Study participants were asked to generate an isometric MVC in elbow extension and flexion at the neutral position for 5 seconds, and the output was recorded. The process was performed 3 times, and measurements were averaged.

Angular velocity and acceleration were calculated from the first and second derivatives of the elbow angular position data, respectively (figs 1B, 1C). These data were used to quantify movement kinematic parameters: peak velocity, peak acceleration, movement speed, active ROM, and movement smoothness.12

Impaired voluntary movements are characterized by these parameters, including a loss of smoothness in movement trajectory.13, 14 In nonparetic arms, rapid voluntary movement trajectories are smooth (see fig 1A) with single-peaked, bell-shaped velocity profiles (see fig 1B). In contrast, movement trajectories of paretic limbs are rippled (see fig 1A) with multiple peaks and irregularities in both velocity (see fig 1B) and acceleration (see fig 1C).

Statistical Analysis 

We used the growth mixture model15, 16, 17, 18, 19, 20a to find the recovery patterns (class) for FMA over 1 year. In our earlier study, we have demonstrated the validity of this model applied to the kinematic and kinetic data.21 The growth mixture model assumes that the population can be divided into several latent classes (subpopulations) and that there is a unique random effects model characterizing the associations between the longitudinal responses and a set of predictors in each subpopulation. Furthermore, the growth mixture modeling allows the membership of the latent classes to be associated with a group of baseline factors through a multinomial logistic regression model. Estimation of the model parameters in the growth mixture model is based on maximizing the likelihood function through the expectation-maximization algorithm.22 Entropy was used to evaluate the performance of the model; the larger the entropy, the better the model.

In the fitted growth mixture model, the multinomial (polytomous) logistic regression19, 20 was used to characterize the association between the membership and kinematic and kinetic parameters. To predict the membership for each subject, we calculated the probability of the subject's data lying in each of the potential subclasses, and identified the FMA membership as the class with the highest predicted probability. This procedure helped us explore the association of the kinematic and kinetic parameters at 1 month poststroke and FMA class membership in the growth mixture model.

Standard t test procedures were used to test for significant changes in kinematic and kinetic parameters, grouped by limb type (paretic and nonparetic).

Results with P values less than .05 were considered significant.

Recovery of Motor Impairment 

Two classes (types) of recovery patterns were found for the FMA scores, on the basis of data collected over 1 year after stroke (fig 2). These classes were defined using the growth mixture model of the relation between elapsed time and FMA scores. Classification based on likely class membership resulted in class counts of 9 for class 1 and 11 for class 2, and the corresponding proportions of total sample size were 45% for class 1 and 55% for class 2. Figure 2A shows the observed and estimated mean FMA scores for class 1 and class 2 over time.

View Large Image Download to PowerPoint

Fig 2. Recovery patterns of motor impairment assessed by the FMA. (A) The predicted means are the solid lines; the observed means are the dashed lines. (B) The mean trends are the solid lines. The boundaries (±1 SD) are the dashed lines. Red, class 1; blue, class 2.

For class 1, which defines the relation between elapsed time and FMA scores for 1 group of subjects, the growth mixture model provides an intercept of 7.42 (P<.002), indicating that a significant value of the FMA was observed at 1 month poststroke. The slope of this (FMA-time) relation was 4.55 a month (P<.001), indicating that significant growth in FMA scores occurred in the course of the 5 measurements performed over the period of 1 year. The estimate for the quadratic term is −.24 a month (P<.001), indicating that a significant drop in growth speed of FMA scores occurred over the 5 measurements in 1 year poststroke.

For class 2, our estimate for the intercept is 53.27 (P<.001), indicating that an even higher-level FMA score was observed at 1 month poststroke. The estimates for the slope and quadratic terms here are .55 a month (P=NS) and .001 (P=NS), which are both NS at the .05 level.

Figure 2B shows the estimated FMA score and its 95% CI over time for class 1 and class 2. The figure shows that there is clear separation between classes 1 and 2, without any overlap of the CIs. Subjects in class 1 started with a low-level FMA score and then increased quickly before tapering off gradually over time, while subjects in class 2 tended to start with a high-level FMA score and then remained at approximately the same level with minimal change.

Entropy for the model was .99, indicating excellent separation. The closer entropy approaches 1, the better the model is.

A summary of this information is provided in the first few rows of table 2, which focus on FMA results.

Table 2. The Information of the Growth Mixture Model and the Logistic Regression Model

	Growth Mixture Model and Logistic Regression Model	Recovery of Function (FMA)
		Class 1	Class 2
	No. of subjects	9	11
	Percent sample size	45%	55%
	Growth mixture model
	Intercept	7.419(P<.002)	53.274(P<.001)
	Slope	4.552(P<.001)	0.546(P=.396)
	Quadratic	−0.246(P<.001)	0.001(P=.986)
	Active ROM
	Logistic regression model
	Coefficient intercept	Classes do not overlap based on active ROM
	Coefficient active ROM
	Active ROM (deg)	≤42	>42
	Peak velocity
	Logistic regression model
	Coefficient intercept	5.196(P<.001)	Reference
	Coefficient peak velocity	−0.059(P<.01)
	Peak velocity (°/s)	≤88	>88
	Movement speed
	Logistic regression model
	Coefficient intercept	−2.748(P<.001)	Reference
	Coefficient movement speed	0.1(P<.02)
	Movement speed (°/s)	≤27	>27
	Peak acceleration
	Logistic regression model
	Coefficient intercept	−5.289(P<.003)	Reference
	Coefficient peak acceleration	0.015(P<.05)
	Peak acceleration (°/s2)	≤353	>353
	Movement smoothness
	Logistic regression model
	Coefficient intercept	Classes do not overlap based on movement smoothness
	Coefficient movement smoothness
	Movement smoothness (deg)	≤31	>31
	Flexor MVC
	Logistic regression model
	Coefficient intercept	Classes do not overlap based on flexor MVC
	Coefficient flexor MVC
	Flexor MVC (Nm)	≤5	>5
	Extensor MVC
	Logistic regression model
	Coefficient intercept	Classes do not overlap based on extensor MVC
	Coefficient extensor MVC
	Extensor MVC (Nm)	≤1.8	>1.8

NOTE. For active ROM, movement smoothness, flexor MVC, and extensor MVC, no regression values are warranted because of absence of overlap of the data sets.

Relation Between Motor Recovery and Kinematics and Kinetics 

To identify the key kinematic and kinetic variables, we quantified the magnitude of several impairments during rapid voluntary elbow extension movement of the paretic upper limb. Figure 1 shows the movement trajectory, velocity, and acceleration for the paretic and the nonparetic arm in a representative survivor of stroke in the acute phase. For the paretic arm, the active ROM was 84° (42%) smaller than that of the nonparetic arm. Peak velocity and peak acceleration were approximately 78% and 74% smaller, respectively, in the paretic than in the nonparetic arm.

Figure 3 shows the group means and standard errors of major kinematic and kinetic parameters for paretic and nonparetic arms at 1 month. Movement speed was smaller, peak velocity and peak acceleration smaller, active ROM smaller, isometric muscle strength of elbow flexors lower, and isometric muscle strength of elbow extensors lower (all P<.001). Furthermore, movement of the paretic arm was more jerky, indicated by the ripples on the movement trajectory, velocity, and acceleration graphs (see fig 2). The group results show that movement smoothness was significantly smaller in the paretic than the nonparetic arm (P<.001). We used logistic regression models to predict the recovery patterns of the FMA on the basis of these key parameters, recorded at 1 month poststroke.

View Large Image Download to PowerPoint

Fig 3. Kinematic and kinetic variables that are significantly different between the paretic and nonparetic arm at 1 month poststroke. Values are group averages ± SDs. Abbreviations: Ap, peak acceleration; AROM, active range of motion; MVCEX, isometric muscle strength of elbow extensors; MVCFL, isometric muscle strength of elbow flexors; SM, movement smoothness; Va, movement speed; Vp, peak velocity. All variables are significant for comparisons between the paretic and nonparetic arms (P<.001).

This method was used to predict FMA class membership in a binary scheme, based on the key kinematic and kinetic measures at 1 month. (In such a scheme, the task is to assign an outcome measure to one or another class.) The classes are defined by the FMA time history, as described. The estimated coefficient was 5.19 (P<.001) for the intercept and −.06 (P<.01) for the peak velocity, indicating that the logit and thus the probability for class 1 membership decreases as peak velocity increases. Here class 2 serves as the reference category.

In summary, subjects with peak velocity of 88°/s or less at 1 month after stroke will be more likely to belong to class 1—for example, their FMA scores increase over time before they flatten out. Subjects with peak velocity greater than 88°/s at 1 month poststroke will be more likely to belong to class 2—for example, their FMA scores will remain the same over time with minimal change. A summary of this information is provided in table 2.

Similarly, using the logistic regression, we predicted the effects of movement speed at 1 month after stroke on FMA class membership. The estimated coefficient was −2.75 (P<.001) for the intercept and .10 (P<.02) for the movement speed, indicating that the logit and thus the probability for class 1 membership increase as movement speed increases. In this analysis, class 2 serves as the reference category. In summary, subjects with movement speed of 27°/s or less at 1 month poststroke will be more likely to belong to class 1, and subjects with movement speed greater than 27°/s will be more likely to belong to class 2.

As summarized in table 2, the results of the logistic regression analyses also indicated that subjects with peak acceleration 353°/s2 or less at 1 month poststroke will be more likely to belong to class 1, and subjects with peak acceleration greater than 353°/s2 to belong to class 2.

Our data showed that in subjects with active ROM less than 42° at 1 month poststroke, the FMA score will increase over time before it flattens out, and in subjects with active ROM greater than 42° at 1 month poststroke, the FMA score will remain the same over time with minimal change. These findings show that active ROM separates the class membership perfectly, making the logistic regression modeling unnecessary.

Similar to active ROM data, our data showed that movement smoothness can predict the class membership of the FMA without using modeling. Thus, all subjects in class 1 had a movement smoothness of −31° or less, and all subjects in class 2 had a movement smoothness greater than 31°.

Similarly, kinetic parameters at 1 month poststroke appear to predict perfectly the 2-class membership of the FMA without using the logistic regression modeling. Thus, subjects with isometric muscle strength of elbow flexors of 5Nm or less at 1 month poststroke belong to class 1, and subjects with isometric muscle strength of elbow flexors greater than 5Nm belong to class 2. Finally, all subjects in class 1 had an isometric muscle strength of elbow extensors of 1.8Nm or less, and all subjects in class 2 had an isometric muscle strength of elbow extensors greater than 1.8Nm.

Prediction of Motor Impairment Recovery 

Based on the key kinematic and kinetic measures at first visit (ie, 1mo poststroke), the logistic regression modeling was used to predict the recovery patterns of FMA scores over 1 year. Table 2 summarizes which motor impairment recovery is more likely to be predicted for subjects with different initial impaired voluntary movements, on the basis of kinematic and kinetic measures. Although all these measures were found to be good predictors of motor impairment recovery, active ROM and MVC values are the easiest ones to measure and may have the most clinical significance.

Recovery Patterns of Motor Impairment 

Earlier longitudinal studies in stroke have evaluated motor impairments using different clinical assessments. The FMA has been used most frequently, particularly for the assessment of upper extremities in stroke survivors,3, 5 because it is shown to assess motor impairments reliably and validly and detect stroke recovery.23 We used the FMA to track the recovery of motor impairment over a long period after stroke. We found 2 distinct classes of recovery patterns. Class 1 started with low values, increased over time, and then leveled off. Class 2 started with higher values but did not change significantly with time. The class 1 pattern is consistent with the mean recovery pattern observed by others3, 5, 6 overall; for example, it increased exponentially with time then remained relatively constant. However, there are a few major differences.

First, the timing of changes is different between our studies and others; our results showed improvement of impairment up to 8 to 9 months, whereas improvement was limited to a maximum 10 to 20 weeks in others' reports.3, 5 This may be partially explained by the fact that they studied mean value of the FMA across patients, which may not accurately detect and reflect the possible different contributing patterns.

Second, the duration of poststroke monitoring used by other longitudinal studies was different from that of this study. We examined the recovery over 1 year after stroke, whereas most studies have monitored the changes over a much shorter period.

Although the movement impairments are caused, at least initially, by decreased excitatory synaptic drive on motoneurons, these impairments can also be induced by changes in muscular properties secondary to neurologic damage. Therefore, the progress of recovery after stroke may not reflect a single mechanism, and could depend on which mechanism is dominant at each time point. Consequently, averaging across the total stroke population may not be an appropriate procedure to characterize adequately the motor performance recovery pattern over time. For class 2, it is noteworthy that our classification scheme does not eliminate the possibility of significant rapid improvement within the first 4 weeks, because our first evaluation is at 1 month. This possibility is therefore also theoretically consistent with earlier studies that reported rapid neurologic and functional recovery within the first few weeks after stroke.2, 5 The existence of 2 classes, 1 slow and 1 fast, is consistent with the findings of Newman,6 and demonstrates the need for prolonged assessment intervals to study adequately the natural history of stroke recovery in all subpopulations.

Clinical Significance 

A clear understanding of the contributions of different impairment mechanisms at different times poststroke is a prerequisite for the rational development of effective therapies. This understanding can be achieved, in part, by characterizing the recovery patterns of motor impairment after stroke. Our study findings are novel in that we are the first to identify 2 classes of recovery for FMA over 1 year after stroke and predict the recovery even beyond 1 year.

Our results revealed a curvilinear pattern of recovery for the FMA. This is consistent with the exponential recovery pattern reported by others,3, 5, 6 indicating that the major recovery of the arm motor impairment occurs over the first few months. Surprisingly, our results demonstrated that this significant fast improvement occurs in patients with lower FMA scores (class 1). Indeed, the comparison of class 1 and class 2 for our FMA measure shows that the percentage of improvement for the mean FMA score was approximately 250% for class 1, whereas it was less than 10% in class 2. The rate of improvement (estimated from the slope of the relations) also was 9 times larger for class 1 than for class 2 (see table 2). These findings indicate that subjects with severe motor impairment had a greater chance of recovery (relative to their initial state) than subjects with minimal or mild motor impairment, whereas the common expectation is that stroke survivors with lesser initial impairments have a higher chance of improvement.

Our results demonstrate that the kinematic and kinetic measures are good predictors of the motor recovery after stroke. Most of these physiologic measures such as active ROM and MVCs are evaluated in a routine visit by clinicians, and may therefore provide useful information to clinicians about which therapeutic interventions are likely to be most effective.

Finally, it is conceivable that naturally occurring FMA patterns, such as those of class 1 patients, who show rapid relative improvement in FMA scores, may also reflect the capacity of the injured nervous system to improve maximally in response to therapy. If so, immediate and intensive appropriate physical interventions may have substantial impact on the recovery of survivors of stroke who show class 1 features. This recovery potential remains to be validated.

Study Limitations 

Our analysis was based on a relatively small sample (N=20), which is a limitation of the study. A larger study population will be needed to confirm the potential clinical applications.