Dimensionality and Construct Validity of the Fugl-Meyer Assessment of the Upper Extremity

Analysis 

Participants 

A secondary analysis was performed on data pooled from 2 studies. The first dataset consisted of 100 persons enrolled in a randomized clinical trial of therapeutic exercise.67 FMA-UE data were collected before the exercise intervention by therapists trained in standard administration of the assessment. Persons in this study met the following inclusion criteria: (1) stroke within 30 to 50 days, (2) ability to ambulate 7.5m (25ft) independently, (3) mild to moderate stroke deficits defined by a total FMA score of 27 to 90 (upper and lower extremities) and Orpington Prognostic Scale (OPS) score between 2.0 and 5.2, (4) palpable wrist extension on the involved side, and (5) Folstein Mini-Mental State Examination score of greater than 16. The second dataset consisted of 459 persons enrolled in the Kansas City Stroke Study.68 Participants were included if they met the following criteria: (1) stroke onset within 0 to 14 days before enrollment, (2) 18 years of age or older, (3) ischemic stroke as diagnosed by physician, (4) lived in the community before stroke onset, and (5) able to participate in baseline testing (eg, testing included the FMA-UE, Barthel Index,69 the physical function index of the Medical Outcomes Study 36-Item Short-Form Health Survey,70 Modified Rankin Scale71). Staff trained in the administration of the FMA-UE evaluated patients. Pooling of these data resulted in a 559-person dataset. Because lesion location may effect neural reorganization and clinically observed patterns of UE recovery,72 subjects with subcortical (ie, brainstem or cerebellum) stroke were excluded from the pooled dataset. (Note, no further information regarding anatomic area of lesion was available.) This resulted in a final dataset of 512 participants. Characteristics of the 512-person sample are presented in table 1.

Table 1. Characteristics of the 512 Person Sample

	Characteristics	Values
	Age (y)	69.8±11.12
	Sex
	Males	242(47.3)
	Females	270(52.7)
	Race
	White	411(80.3)
	African American	85(16.6)
	Other	16(3.1)
	Stroke type
	Ischemic	474(92.6)
	Hemorrhagic	38(7.4)
	Stroke location
	Right cortical hemisphere	250(48.8)
	Left cortical hemisphere	262(51.2)
	Days since stroke	16.88±31.23
	Range of days since stroke
	Minimum	0
	Maximum	145
	Stroke severity
	Minor (OPS score <3.2)	191(37.3)
	Moderate (OPS score range, 3.2−5.2)	269(52.5)
	Severe (OPS score >5.2)	52(10.2)

NOTE. Values are mean ± standard deviation, n (%), or n.

Fit Statistics 

Item-fit statistics and correlations from the initial analysis of the entire 33-item FMA-UE are presented in table 2. Two items (biceps reflex, triceps reflex) show infit values that are beyond the acceptable ranges described by Wright and Linacre60 (MnSq ≥1.7, standardized z≥2.0). These items also showed abnormally high outfit statistics and relatively low item correlations (.36, .26). Infit and outfit statistics are within acceptable ranges for all other items.

Table 2. Item-fit Statistics and Correlations for the 33-Item FMA for the Upper Extremity

	FMA-UE Item Number and Description	Infit MnSq	Outfit MnSq	Item Correlation
	Item 1: Biceps reflex	3.51	9.90	.36
	Item 2: Triceps reflex	5.20	9.90	.26
	Item 3: Scapular elevation	0.93	0.87	.83
	Item 4: Scapular retraction	0.89	0.96	.82
	Item 5: Shoulder abduction	0.79	0.77	.84
	Item 6: Shoulder external rotation	0.68	0.63	.85
	Item 7: Elbow flexion	0.67	0.81	.85
	Item 8: Forearm supination	0.70	0.73	.84
	Item 9: Shoulder adduction with internal rotation	0.65	0.49	.87
	Item 10: Elbow extension	0.57	0.53	.87
	Item 11: Forearm pronation	0.60	0.64	.86
	Item 12: Hand to lumbar spine	0.72	0.62	.85
	Item 13: Shoulder flexion to 90°, elbow extended	0.78	0.52	.84
	Item 14: Pronation-supination, elbow at 90°	0.59	0.48	.86
	Item 15: Shoulder abduction to 90°, elbow extended	0.72	0.53	.83
	Item 16: Shoulder flexion to 180°, elbow extended	0.70	0.57	.80
	Item 17: Pronation-supination, elbow extended	0.57	0.44	.83
	Item 18: Normal reflex activity	1.34	1.10	.57
	Item 19: Wrist stable, elbow at 90°	1.01	0.71	.80
	Item 20: Wrist flexion-extension, elbow at 90°	0.58	0.52	.86
	Item 21: Wrist stable, elbow extended	0.96	0.69	.78
	Item 22: Wrist flexion-extension, elbow extended	0.65	0.57	.82
	Item 23: Wrist circumduction	0.79	0.92	.77
	Item 24: Finger mass flexion	0.79	0.70	.85
	Item 25: Finger mass extension	0.70	0.58	.86
	Item 26: Hook grasp	1.29	1.08	.72
	Item 27: Lateral prehension	0.96	1.21	.76
	Item 28: Palmar prehension	0.95	0.83	.81
	Item 29: Cylindrical grasp	0.88	0.76	.84
	Item 30: Spherical grasp	0.96	1.02	.77
	Item 31: Movement without tremor	1.19	0.98	.80
	Item 32: Movement without dysmetria	1.37	1.21	.76
	Item 33: Movement with normal speed	1.10	0.93	.76

Principal Components Analysis 

Eigenvalues and factor loadings 

The PCA retained 4 factors having eigenvalues greater than 1.0 (Kaiser rule65). A single component explained 68% of the variance in the data. Three other factors explained an additional 5%, 4%, and 3% of the variance. Thirty of 33 items loaded strongly onto the first principal component (factor loading values range, .75–.91). The 3 reflex items (biceps reflex, triceps reflex, normal reflex activity) had poor to moderate loadings (r=.14, r=.24, r=.53, respectively) on the first factor.

Poor factor loadings onto the first principal component and high infit statistics suggest that the reflex items do not fit with the intended measurement construct of the FMA-UE; therefore, these items were removed from the subsequent Rasch analysis. Fit statistics, a large first eigenvalue, and strong PCA factor loadings support the unidimensionality of the remaining items.

Item-difficulty measures, fit statistics, and item correlations of the revised 30-item FMA-UE (reflex items removed) are presented in table 3. Fit statistics reveal that all 30 items fit the unidimensional assumptions of the Rasch model. The revised 30-item FMA-UE showed good internal consistency. The person reliability index, analogous to coefficient α, was .96. The instrument was very effective in discriminating UE ability across the sample dividing the sample into 7 statistically significant strata (number of strata = (4[Gp = 4.67] + 1)/3, where GP is person separation).73 Note, although these strata are statistically defined, their clinical application requires further exploration.

Table 3. Item Difficulty Measures, Fit Statistics, and Correlations for the Modified 30-Item FMA for the Upper Extremity

	FMA-UE Item	Measure	Error	Infit MnSq	Outfit MnSq	Item Correlation
	Wrist circumduction	1.67	0.10	0.90	0.96	.77
	Hook grasp	1.33	0.10	1.43	1.28	.73
	Shoulder flexion to 180°, elbow extended	1.26	0.10	0.84	0.80	.80
	Spherical grasp	1.20	0.10	1.08	1.17	.78
	Lateral prehension	1.08	0.10	1.10	1.30	.78
	Wrist flexion-extension, elbow extended	1.06	0.10	0.77	0.78	.82
	Pronation-supination, elbow extended	1.00	0.10	0.70	0.63	.83
	Wrist stable, elbow extended	0.95	0.10	1.13	0.96	.78
	Movement with normal speed	0.89	0.10	1.25	1.13	.77
	Forearm supination	0.65	0.10	0.82	0.88	.85
	Shoulder abduction to 90°, elbow extended	0.28	0.10	0.86	0.71	.85
	Movement without dysmetria	0.27	0.10	1.62	1.52	.77
	Shoulder external rotation	0.24	0.10	0.83	0.82	.86
	Wrist stable, elbow at 90°	0.18	0.10	1.20	0.96	.82
	Wrist flexion-extension, elbow at 90°	0.12	0.10	0.69	0.66	.88
	Palmar prehension	0.06	0.10	1.10	1.01	.83
	Scapular retraction	0.03	0.10	1.08	1.29	.84
	Pronation-supination, elbow at 90°	−0.17	0.11	0.72	0.67	.88
	Shoulder flexion to 90°, elbow extended	−0.21	0.11	0.94	0.71	.86
	Hand to lumbar spine	−0.40	0.11	0.90	0.85	.87
	Shoulder abduction	−0.56	0.11	0.99	1.02	.87
	Elbow extension	−0.64	0.11	0.70	0.70	.90
	Forearm pronation	−0.87	0.11	0.76	0.81	.89
	Movement without tremor	−0.91	0.11	1.51	1.36	.83
	Cylindrical grasp	−1.10	0.12	1.09	0.94	.87
	Finger mass extension	−1.25	0.12	0.89	0.76	.89
	Scapular elevation	−1.40	0.12	1.25	1.31	.86
	Finger mass flexion	−1.44	0.12	1.04	1.03	.89
	Shoulder adduction with internal rotation	−1.56	0.12	0.88	0.74	.90
	Elbow flexion	−1.76	0.13	0.92	1.18	.89

NOTE. Items ordered in decreasing level of difficulty. High positive measures represent “difficult” items while high negative measures represent “easy” items.

Item-Difficulty Hierarchy 

Table 3 also presents items in order of decreasing challenge. FMA-UE items at the bottom of the left column indicate the least challenging items; those at the top represent most challenging items. Elbow flexion (−1.76±0.13 logits) and shoulder adduction with internal rotation (−1.56±0.12 logits) were the easiest items for this sample to perform, and wrist circumduction (1.67±0.10 logits) and hook grasp (1.33±0.10 logits) were the most difficult items for this sample to perform. Five of the 9 flexor synergy and extensor synergy items group toward the easy end of the hierarchy, whereas some flexor and extensor synergy items span the item-difficulty hierarchy. For example, elbow flexion is the easiest item in the hierarchy (−1.76±0.13 logits), scapular elevation is slightly more difficult (−1.40±0.12 logits), and forearm supination is moderately difficult (0.65±0.10 logits).

Movements requiring combinations of shoulder flexion and elbow extension are much more difficult than movements demanding no shoulder motion with the elbow positioned at 90°. For example, pronation supination with the elbow at 90°, wrist flexion extension with the elbow at 90°, and wrist stable with the elbow at 90° are among the moderately difficult items. Shoulder flexion to 180° with the elbow extended, wrist flexion extension with the elbow extended, and pronation supination with the elbow extended are among the most difficult items.

In addition, although the majority of the hand items group at the difficult end of the hierarchy, some of these items also span the item-difficulty hierarchy. For example, hook grasp (1.33±0.10 logits), spherical grasp (1.20±0.10 logits), and lateral prehension (1.08±0.10 logits) are among the most difficult items, whereas finger mass flexion (−1.44±0.12 logits) is among the easiest items and palmar prehension (0.06±0.10 logits) is calibrated near the middle of the scale. Surprisingly, finger mass extension (−1.25±0.12 logits) calibrates as an easy item for this sample to perform (see Discussion).

Keyforms 

Although the previously described results show the overall hierarchical pattern for the sample under study, a critical question is the consistency of this pattern across subjects. Similar to item-fit statistics are person fit statistics. Ninety-eight percent of the sample showed acceptable infit statistics (MnSq <1.7, standardized z<2.0), suggesting that individual subjects overwhelmingly showed similar responses to the item-difficulty hierarchy. Figure 1 shows the item-difficulty hierarchy relative to the scoring pattern of 2 subjects with different UE abilities (as defined by the aggregate FMA-UE raw score). Items are listed in terms of decreasing difficulty level in the right column of each panel. The 3-point rating scales for each item are presented in the left column of each panel. As item difficulty increases, the rating scale stair-steps to the right. Circled numbers represent each participant’s actual FMA-UE item rating (eg, raw data). A solid vertical line of best fit is drawn through the circled ratings. This line crosses the bottom scale to give person-ability measure (logits). The dotted vertical lines represent the 95% confidence interval around this ability measure.

View Large Image Download to PowerPoint

Fig 1. Keyform recovery maps for 2 subjects with stroke.

A person of low ability (FMA-UE score = 20/60) is represented in panel A, and higher ability (FMA-UE score = 53/60) in panel B. The individual of low ability (A) has a tendency to receive ratings of 1 (partial performance) on the easy items (elbow flexion, shoulder adduction with internal rotation, finger mass flexion) and receives ratings of 0 (unable to perform) on the most difficult items (shoulder flexion to 90° with elbow extended and wrist circumduction). The individual of higher ability (B) has a tendency to receive ratings of 2 (faultless performance) on the easy and moderately difficult items. However, this person receives ratings of 1 (partial performance) on the most difficult items (shoulder flexion to 180° with elbow extended, hook grasp and wrist circumduction). In general, although these participants are of different abilities, their scoring pattern is the same; both score higher on easier items and lower on harder items. The pattern retains its structure when measuring subjects across the ability range.

There are exceptions to this pattern as evidenced by the rating denoted with a triangle. The person of lower ability represented in panel A receives a higher than expected rating of 2 on a difficult item (hook grasp). Overall, 12 (2%) of 512 people showed erratic scores based on infit statistics. Among these 12 subjects, only 2 to 5 items were statistically erratic.

Dimensionality 

The initial finding from both the Rasch analysis and the more traditional PCA revealed that the reflex items empirically appear to be disconnected from the remainder of the FMA-UE. According to Fugl-Meyer’s suggested stepwise course of recovery, it would be expected that persons with little to no UE motor ability would be the only individuals without UE reflexes. The high infit and outfit statistics indicate that at least some subjects of high ability fail to show UE reflexes. Multiple statistical findings (ie, low PCA loadings, low correlations and high fit statistics) suggest that the reflex items are contributing little explanatory variance and support our decision to remove these items from the instrument.

The 3 reflex items differ from the rest of the FMA-UE items in 2 ways. First, the items evaluate an involuntary response rather than volitional movement. Specifically, the behavior of the phasic stretch reflex is assessed through tendon taps of the biceps (items 1 and 18), triceps (items 2 and 18), or finger flexors (item 18). Second, the rating scales for the reflex items differ from the rest of the assessment. For example, items 1 and 2 use a 2-point scale (0, no reflex activity elicited; 2, reflex activity elicited), and item 18 uses a uniquely defined 3-point scale (0, at least 2 of 3 phasic reflexes are hyperactive; 1, one reflex is hyperactive or 2 reflexes are lively; 2, no more than 1 reflex is lively and none are hyperactive). These procedural and scoring differences from that of the majority of the assessment may cause the items to misfit and correlate as a separate factor.

Although these statistical findings may be a product of test administration procedures (testing involuntary movements vs testing voluntary movements) or the rating scale structure, they also may indicate that the FMA-UE reflex items measure a motor control construct different from that represented by the other assessment items. Reflexes are not simply “hardwired” precursors to voluntary movement as suggested by the reflex-hierarchical conceptual model. Instead, reflexes provide task-, phase-, and context-dependent modulation of volitional movement.51 This suggests that the assessment of a tendon tap in a resting patient yields little information about volitional motor ability. Given that the FMA-UE is an assessment of voluntary movement ability, our results suggest the 3 reflex items diminish the construct validity of the instrument. As suggested by Gladstone et al,39 the reflex items may confound interpretation of the FMA-UE total score.

Item-Difficulty Hierarchy 

Fugl-Meyer attributed the assessment structure and item choice to the earlier works of Twitchell49 and Brunnstrom.50 Twitchell described an “orderly progression of phenomena” characterizing the course of poststroke UE motor recovery. Initial UE flaccidity was followed by emergence of gross “flexor synergy” (defined as concurrent scapular elevation, retraction, humeral abduction, external rotation, elbow flexion, and forearm supination) and “extensor synergy” (defined as concurrent humeral adduction, internal rotation, elbow extension, and forearm pronation), movement patterns, gradual separation of synergy movement patterns, and finally distal fine-motor coordination. Twitchell attributed this recovery sequence to neurophysiologic mechanisms (ie, the cortex reassuming its natural inhibition of primitive proprioceptive reflexes).49 Brunnstrom, furthering Twitchell’s work, documented specific movements that typified each stage of motor recovery thereby mapping the recovery progression. Fugl-Meyer appears to have, at least in part, applied a numeric rating scale to Brunnstrom’s “Hemiplegia Classification and Progress Record”50 to create the FMA-UE.1

The Rasch item-difficulty hierarchy does not follow the expected “stepwise” item arrangement proposed by the FMA-UE item order. Our results suggest that the FMA-UE items are arranged according to the difficulty of motion (ie, arm movements are easier or more challenging based on the inherent task-specific demands of the movement). Elbow flexion may be easier than wrist circumduction because the movement is inherently less complex (ie, requires less agonist-antagonist coordination). Shoulder flexion to 180° with the elbow extended may be difficult because of the influence of gravity on a long lever arm.

Our task-specific interpretation of the item hierarchy may affect the interpretation of the FMA-UE score. Chae et al44 suggested that a low FMA-UE aggregate score indicated recovery of only proximal UE motor function, whereas a higher FMA-UE aggregate score indicated recovery of both proximal and distal UE motor function. This interpretation is consistent with Fugl-Meyer’s proposed stepwise item order in which movements assumed to recover early in the process appear first on the assessment, whereas those to recover later in the process appear further along the assessment. In contrast, our results indicate that a person with a low FMA-UE score is likely to exhibit some proximal arm motions characteristic of the gross flexor or extensor synergy pattern (eg, the easy items: elbow flexion, shoulder adduction with internal rotation) but is unlikely to exhibit other proximal synergy-based movements (eg, the “more difficult” items: scapular retraction, shoulder external rotation). A person with a low FMA-UE score is likely to exhibit some easy distal, nonsynergistic, hand movements (eg, finger mass flexion, cylindrical grasp). Similarly, our item hierarchy, if assumed to show the pattern of poststroke UE recovery, shows that recovery does not proceed in a strict proximal-to-distal sequence as was traditionally endorsed by some therapists74 and suggested by some therapy textbooks.75, 76 Instead, UE motor behavior during recovery may be a dynamic interaction of neural factors with the task-specific difficulty of a movement.

Our results do not entirely contradict the original clinical observations of Twitchell49 and Brunnstrom.50 Indeed, the results provide at least partial support for their classic works. For example, both clinicians observed flexion synergy and extension synergy movements to be among the first motions to return after stroke. Our results show that 3 of the 6 flexor synergy movements and 2 of the 3 extension synergy movements are among the least challenging items of the FMA-UE (elbow flexion, shoulder abduction and scapular elevation and forearm pronation and shoulder adduction with internal rotation). Furthermore, both clinicians described movements deviating from synergy occurring later in recovery. The Rasch hierarchy also partially supports this observation. Items requiring wrist motion are easier when the shoulder is neutral and the elbow is bent than when the shoulder is flexed and the elbow is extended. For example, forearm pronation supination, elbow at 90°, and wrist flexion extension, elbow at 90°, are moderately challenging items, whereas pronation supination, elbow extended, and wrist flexion extension, elbow extended, are among the hardest items.

Our results show that although the traditional FMA-UE structure was not entirely consistent with contemporary motor control conceptual models, the majority of the items have strong item-level measurement properties. Thirty of 33 items were consistent with a unidimensional UE motor ability construct. These items represent a wide range of UE motor behaviors representing the hierarchy of UE function. This hierarchy remained consistent across subjects with various levels of poststroke UE motor impairment. Our results might support and explain why the FMA-UE has historically performed well in standard psychometric analyses and studies.

One surprising finding in the present study was that the finger extension item calibrated as an easy item. Studies have shown finger extension to be highly impaired after stroke.16, 77 Our results may be explained 2 ways. First, the vast majority of participants in our sample (90%) sustained mild or moderate stroke as defined by the OPS score. It is possible that our sample was not severely impaired and therefore did not exhibit impaired finger extension. Second, the FMA-UE defines finger mass extension as “gross release of the mass flexion grasp.”1 Finger extension, as defined by the FMA-UE, may actually reflect the natural tenodesis mechanism of grossly extending the fingers by flexing the wrist. It should be noted that some researchers have more rigorously evaluated finger extension by controlling for wrist position and defining finger extension as return to metacarpophalangeal joint neutral.16, 77 One weakness of the FMA-UE is that standard administration guidelines have only recently been developed.78 It is feasible that if the assessment were administered in another way (ie, controlling wrist configuration), the item would have a different difficulty level.

Study Limitations 

There are a number of statistical and demographic limitations to this study that could limit the generalizability of our results. Traditional PCA was designed for continuous rather than ordinal data. Flora and Curran79 suggested using factor analysis approaches designed specifically for multivariate categorical data when analyzing ordinal data (eg, FMA-UE ratings). Although these approaches are not yet common in the health care literature, they should be considered for future studies. Additionally, the generalizability of the findings is confined to sample characteristics of the study. Although our participants represented a range of stroke severity, the majority, 90%, of the sample experienced mild or moderate stroke. Furthermore, only 7.4% of the sample experienced hemorrhagic stroke, and 0% of the sample experienced noncortical stroke. To generalize these findings across the stroke population, the study needs to be replicated with subjects representing a broader range of stroke severity, type, and location. Finally, the most critical limitation of the present study is that we used cross-sectional data to challenge Fugl-Meyer’s “recovery” paradigm. We are currently replicating this study by using longitudinal data.