Estimating Minimal Clinically Important Differences of Upper-Extremity Measures Early After Stroke

Participants 

Fifty-two subjects with hemiparesis poststroke were enrolled in this study (table 1). Hemiparetic subjects were participating in the VECTORS study, a single-center randomized controlled trial investigating early motor recovery of the upper extremity after stroke conducted at the Washington University School of Medicine. Subjects were recruited via the Cognitive Rehabilitation Research Group Stroke Registry from the acute neurology service of Barnes Jewish Hospital and from the rehabilitation service of the Rehabilitation Institute of St Louis. In all, 1850 patients with stroke were screened to achieve the enrolled 52 subjects. Subjects were included in the trial if they had (1) an ischemic or hemorrhagic stroke within 28 days of admission to inpatient rehabilitation; (2) persistent hemiparesis as indicated by a score of 1 to 3 on the motor arm item of the NIHSS; (3) the presence of some upper-extremity voluntary activity as indicated by the ability to move proximal and/or distal joints against gravity; (4) evidence of preserved cognitive function, as indicated by scores of 0 or 1 on the consciousness items of the NIHSS and a score of greater than 19 on the Short Blessed Memory Orientation and Concentration test14; (5) the ability to follow 2-step commands, as indicated by a score of 0 on the item 1c of the NIHSS; and (6) no upper-extremity injury or conditions that limited use prior to the stroke. Subjects were excluded from the trial if they (1) could not give informed consent, (2) had clinically significant fluctuations in mental status in the 72 hours prior to enrollment, (3) had hemispatial neglect as assessed by the Star Cancellation Test, and/or (4) were not expected to survive 1 year because of other illnesses (eg, cardiac disease, malignancy). Subject characteristics are provided in table 1; data from this cohort have been reported previously.15, 16 The Human Studies Committee of the Washington University School of Medicine approved the protocol for this study. Written informed consent was obtained from all subjects prior to testing.

Table 1. Subject Characteristics (N=52)

	Variable	Values
	Age (y)	64±14
	Admission NIHSS score	5.3±1.8
	Premorbid Barthel Index score	99.6±2.2
	Premorbid Modified Rankin Index score	0.3±0.6
	Time to day 0 evaluation (days since stroke)	9.5±4.5
	Time to day 14 evaluation (days since stroke)	25.9±10.6
	Sex
	Male	21(40)
	Female	31(60)
	Race
	White	22(42)
	African American	29(56)
	Other	1(2)
	Stroke type
	Ischemic	41(79)
	Hemorrhagic	11(21)
	Affected side
	Dominant	23(44)
	Nondominant	29(56)

NOTE. Values are mean ± SD or n (%).

Protocol 

The data reported in this study are from the first 2 VECTORS evaluations: the prerandomization baseline assessment (study day 0; average days poststroke, 9.5d) and the first posttreatment assessment (study day 14; average days poststroke, 25.9d). Consistent with other reports investigating MCIDs, data from all subjects were pooled, regardless of group assignment in the trial.3, 17, 18, 19 All evaluations were performed by trained study personnel who were blinded to group assignment. At each evaluation session, the affected (contralateral to the lesion) upper extremity was evaluated using a battery of tests, as described below. At the second evaluation session, subjects were also asked to provide a global rating of perceived changes in their affected upper extremity by comparing “how well your arm is doing” using the following 7-point Likert scale: score 1 was much better; score 2 was a little better, meaningful; score 3 was a little better, not meaningful; score 4 was about the same; score 5 was a little worse, not meaningful; score 6 was a little worse, meaningful; and score 7 was much worse.

We chose a 7-point scale19, 20 to evaluate our subjects' ratings of change instead of a larger 15-point scale3 because a portion of the sample was not literate and may have had difficulty understanding the finer gradations of the 15-point scale. The scores on this item were then used as the anchors for grouping subjects during the calculation of MCID estimates.

Upper-Extremity Evaluation 

The affected upper extremity was evaluated using a battery of tests that included measurement at the level of impairments, activity limitations, and participation restrictions. Impairment measures included in this report are grip strength and composite upper-extremity strength. Activity measures included in this report are the ARAT and the WMFT. Participation measures included in this report are the MAL and the duration of arm use measured with accelerometers.

Data Analyses 

SPSSd for Windows was used for all statistical analyses. Individual change scores were calculated for each measure by subtracting scores at study day 0 from scores at study day 14. Change scores for each measure were normally distributed as evaluated by Kolmogorov-Smirnov tests.

The magnitude of change that is considered meaningful may be influenced by whether the dominant versus nondominant upper extremity was affected by the stroke. Data from affected dominant upper extremities were therefore analyzed separately from data from affected nondominant upper extremities. To estimate the MCIDs on each measure, mean change scores for each perceived change rating were computed. Consistent with other recent reports using anchor-based methods in this population,8, 9 the mean change score for the smallest meaningful change (here a score of 2 is “a little better, meaningful”) was taken as the MCID. In addition to expressing the MCID as a change in raw scores on each measure, we also calculated the MCID as a percentage of the total scale (when possible) and as a single-population effect size,43 where the change score is divided by the SD of the study day 0 score. Expressing the MCIDs in these alternative formats allows for comparison of MCIDs across the measures evaluated.

Distribution of Subjects' Perceived Change Ratings 

Overall, the distribution of perceived change ratings was positively skewed. Twenty-nine subjects (56%) rated their upper extremity a 1 (“much better”), 12 subjects (23%) rated their upper extremity a 2 (“a little better, meaningful”), 4 subjects (8%) rated their upper extremity a 3 (“a little better, not meaningful”), and 7 subjects (13%) rated their upper extremity a 4 (“about the same”). No subject rated their upper extremity as having become worse (scores of 5–7). Figure 1 shows the distribution of perceived change ratings based on whether the dominant or the nondominant upper extremity was affected by the stroke. Dominant and nondominant affected sides were equally represented in the groups of subjects rating their upper extremities as a 1 or 2. More subjects who rated themselves a 3 had their dominant arm affected, and all subjects who rated themselves a 4 had their nondominant arm affected.

View Large Image Download to PowerPoint

Fig 1. Frequency of subjects' perceived change ratings. Subjects are grouped by whether the dominant or the nondominant upper extremity was affected by the stroke. Perceived change ratings: 1, much better; 2, a little better, meaningful; 3, a little better, not meaningful; 4, about the same; 5, a little worse, not meaningful; 6, a little worse, meaningful; 7, much worse.

Mean Change Scores for Each Rating and Estimated MCIDs 

Fig 2, Fig 3, Figure 4 show mean change scores for each measure plotted as a function of perceived change ratings and affected side. Subjects who rated their affected upper extremities as a 3 or 4 were grouped together in each graph because both ratings represent less than a meaningful change. For the impairment level measures, grip strength change scores (see fig 2A) had a generally linear relationship with the perceived change rating for both the dominant and nondominant groups. Composite upper-extremity strength change scores (see fig 2B) for the nondominant group (open symbols) had a linear relationship with the perceived change ratings. This was not the case, however, for the group affected on the dominant side (filled symbols). Subjects who rated their affected dominant upper extremity as having not changed or not meaningfully changed (score of 3 or 4) had somewhat higher change scores than those who rated their affected dominant upper extremity as having a small, meaningful change (score of 2).

View Large Image Download to PowerPoint

Fig 2. Mean change scores in impairment level measures by perceived change ratings and upper extremity affected. (A) Grip strength in the affected hand, measured in kilograms as part of the WMFT. (B) Composite upper-extremity strength on the affected side; values are expressed as ratios of the unaffected upper extremity. Error bars represent SEs. Perceived change ratings: 1, much better; 2, a little better, meaningful; 3, a little better, not meaningful; 4, about the same.

View Large Image Download to PowerPoint

Fig 3. Mean change scores in activity level measures by perceived change ratings and upper extremity affected. (A) ARAT; (B) WMFT time score; and (C) WMFT function score. Error bars represent SEs. Perceived change ratings: 1, much better; 2, a little better, meaningful; 3, a little better, not meaningful; 4, about the same.

View Large Image Download to PowerPoint

Figure 4. Mean change scores in participation level measures by perceived change ratings and upper extremity affected. (A) MAL quality of movement score. (B) Duration of affected upper-extremity use, measured by wrist accelerometers and expressed in hours. Error bars represent SEs. Perceived change ratings: 1, much better; 2, a little better, meaningful; 3, a little better, not meaningful; 4, about the same.

For the activity level measures, ARAT scores (see fig 3A) had a linear relationship with the perceived change rating for both the dominant and nondominant groups. The WMFT time change scores (see fig 3B) had a linear relationship when the dominant side was affected but had little relationship with perceived change ratings when the nondominant side was affected. For this measure, subjects who rated their nondominant affected upper extremity as being much better (score of 1) had change scores that were similar to change scores of the subjects who rated their nondominant affected upper extremity as changing less (scores 2, 3, and 4). The WMFT function change scores (see fig 3C) had a better relationship with the perceived change ratings than the WMFT time scores.

For the participation level measures, the MAL how well change scores (see fig 4A) had a linear relationship with perceived change ratings, with the dominant and nondominant groups very similar. The duration of upper-extremity use measured with accelerometers (see fig 4B) had no clear relationships with perceived ratings of change for either the dominant or the nondominant groups. Regardless of which side was affected, the largest changes in use were found in those people who considered their affected upper extremity as having not meaningfully changed or not changed at all (score of 3 or 4).

We used the mean change score for the smallest meaningful change (here a score of 2 is “a little better, meaningful”) as the estimate of the MCID in our sample of people with acute hemiparesis. Table 3 shows estimated values of MCID for each measure. For some measures, the mean change score for subjects who rated themselves a 2 was smaller than the mean change score for those who rated themselves a 3 or 4 (see figs 2B, 3B, 4B). The MCID on these measures was classified as unknown. When expressed as a percentage of the total scale, estimated MCID values ranged from 16% to 21% in the group that was affected on the dominant side, and 18% to 30% in the group that was affected on the nondominant side. When expressed as an effect size, estimated MCID values ranged from 0.48 to 2.00 and from 0.59 to 2.20 in the dominant and nondominant groups, respectively.

Table 3. Estimates of MCID for Each Measure Based on Whether the Dominant or the Nondominant Upper Extremity Is Affected

	Measures	MCID if Dominant Side Affected			MCID if Nondominant Side Affected
		Raw Value	Percentage of Scale	Effect Size	Raw Value	Percentage of Scale	Effect Size
	Impairment measures
	Grip strength (kg)	5.0	NV	0.48	6.2	NV	0.59
	Composite strength (ratio of unaffected side)	⁎	NV	NV	0.2	22†	0.79
	Activity measures
	ARAT	12	21	0.78	17	30	1.10
	WMFT time (s)	−19	16‡	0.48	⁎	NV	NV
	WMFT function	1.0	17	0.91	1.2	20	1.00
	Participation measures
	MAL quality of movement score	1.0	17	2.00	1.1	18	2.20
	Duration of use (h)	⁎	NV	NV	⁎	NV	NV

NOTE. Values are raw scores, percentages of the total scale (where appropriate), and effect sizes. Abbreviation: NV, no value.

⁎ We were unable to estimate MCID values when there was no apparent relationship between the perceived change rating and the mean change on a given measure (see figs 2B, 3B, 4B). † To calculate percentage, total scale was assumed to range from 0.0 to 1.0. ‡ To calculate percentage, total scale was assumed to range from 0 to 120 seconds.

Cautions Before Interpreting Our Estimates 

Our estimates are from a sample of people with acute hemiparesis poststroke, studied during their inpatient rehabilitation stay. It is not known how these estimates might generalize to people with more chronic hemiparesis or to people in other clinical settings. In the first month poststroke, substantial improvements in neuromotor capabilities and in function often occur,44, 45 particularly for those people selected to go to inpatient rehabilitation facilities in the United States. It may be that people with acute hemiparesis require a greater change in movement capabilities to consider the change clinically meaningful because (1) a large portion of the recovery happens during this period and (2) patients have strong expectations for recovery early after their stroke. Thus, our MCID values might be higher than MCID values at later time points poststroke, when people with hemiparesis may have a greater awareness of how smaller changes may be functionally beneficial and may have lower expectations for full recovery.

Most people (56%) in our sample considered themselves to be “much better” at the second evaluation. This created a skewed distribution of perceived change scores (see fig 1), with a smaller portion of people who considered themselves to have changed a little or not at all. Skewed distributions of perceived change scores are common in studies investigating MCID values.3, 13, 20, 46, 47 Given the substantial recovery that occurs in this early period poststroke, it is nearly impossible to obtain a nonskewed distribution of perceived change scores.8, 9 We consider that the primary effect of the skewed distribution was that this left us with a smaller number of people (n=12) from which to estimate the MCID values.

The number of subjects used here to calculate MCID values is similar to numbers in other anchor-based reports in the rehabilitation literature.9, 46, 47 Because of the small sample, the variability in change scores was large (see SE bars in Fig 2, Fig 3, Figure 4). Distributions of change scores in our dataset overlap for different perceived change ratings. This means that we had subjects with change scores greater than the calculated MCID values who considered themselves “about the same,” and we had subjects with change scores less than the calculated MCID values who considered themselves “much better.” Our MCID values should be interpreted with caution, particularly when making judgments about individual patients.

Interpreting Our Estimates 

Based on our data, it appears that change scores of 16% to 30% on these upper-extremity measures may be needed for a person to consider an improvement to be meaningful. In comparison, change scores of 5% to 10% have been found to be clinically meaningful on a number of health-related QOL measures in a variety of patient populations.3, 18, 19, 47 It is possible that our estimates are proportionally higher because health-related QOL measures are designed to assess changes that are inherently meaningful to the patient, whereas upper-extremity measures are designed to assess changes in upper-extremity motor capacity, some aspects of which may be less inherently meaningful to the patient. Thus, a greater proportional change may need to occur before the change is considered meaningful on focal upper-extremity measures versus more global QOL measures.

There are only a small number of MCID values available for comparison in the neurorehabilitation literature. We have been unable to find any estimates of MCID values for impairment level measures such as grip strength. For the activity level measures, a change of 6 points on the ARAT was selected as clinically meaningful in a population of people with chronic hemiparesis because it represented approximately 10% of a 57-point scale.48 Our data suggest that in an acute hemiparetic population, the patient-perceived MCID is 12 points for the dominant side and 17 for the nondominant side, representing 21% and 30% of the scale, respectively. On the WMFT time score, an improvement of 19 seconds on the affected dominant side (16% of the 120-second limit) was considered a meaningful change in our sample, but we were unable to obtain an estimate for the affected nondominant side. Interestingly, a recent report in people with subacute hemiparesis indicated that it was follow-up time scores on the WMFT and not change scores that predicted perceived upper-extremity recovery.38 This could imply that people may place more importance on their current level of function than on how much they have progressed when determining benefits of treatments. For the participation level measures, a change of 0.5 points (10%) on the MAL was selected as clinically meaningful in a population of people with chronic hemiparesis, again because it represented approximately 10% of the scale.48 Our data suggest that in an acute hemiparetic population, the patient-perceived MCID on the MAL is at least 1 point (17%−18% of the scale). Interestingly, our MCID values were proportionally similar to MCID values on 2 global activity measures in the stroke population, where the MCID of the Barthel Index is 1.85 on the 20-point scale (19%; acute population),9 and the MCID of the motor portion of the FIM instrument is 17 on the 105-point scale (16%; subacute population).8

We determined MCID values separately for the affected dominant and nondominant sides. Our estimated MCID values were a little smaller when the dominant side was affected than when the nondominant side was affected (see table 3). Although people typically use the dominant and nondominant upper extremities a similar amount during daily life,40 the 2 sides are used somewhat differently. The dominant upper extremity is used for more skilled manipulation (eg, holding and writing with a pen), whereas the nondominant upper extremity is used in a more supportive role (eg, holding the paper still during writing). A limited ability to use the dominant side after stroke may be more burdensome than a limited ability to use the nondominant side. It then follows that a smaller change in the ability to use the affected dominant side poststroke may be perceived more quickly and considered more meaningful than a similar size change in the ability to use the affected nondominant side. Because the distributions of our dominant and nondominant affected estimates overlap (see size of SE bars in Fig 2, Fig 3, Figure 4), our data can only provide preliminary support for this idea.

Study Limitations 

We were unable to estimate MCID values in instances in which there was no apparent relationship between the perceived change score and the amount of change. This was the case for composite upper-extremity strength when the dominant side was affected, for the WMFT time score when the nondominant side was affected, and for the duration of use when either side was affected (see Fig 2, Fig 3, Figure 4, table 3). One explanation for our inability to detect a relationship between the perceived change score and the amount of change may be our small sample size. Thus, the lack of relationship may be a result of the large variability in our sample. This variability may be masking a real relationship that could be uncovered with larger sample sizes, from which it may then be possible to obtain estimates of MCID values. We consider this a real possibility for composite upper-extremity strength measure and for the WMFT time score, where a relationship existed when one side was affected but not when the other side was affected. An alternative explanation for our inability to detect a relationship may be that change scores on these measures are not related to patient-perceived change. For example, it may have little meaning to the patient that their duration of upper-extremity use improves by 1.5 hours. If this alternative explanation is true, then change scores on these particular measures may be less useful for assessing outcomes in individual patients or in research studies.