A Pilot Study to Modify the SF-36V Physical Functioning Scale for Use With Veterans With Spinal Cord Injury

Study Design 

A 2-phase mixed qualitative and quantitative research design was used. Approval from appropriate research groups at the Tampa VHA and the institutional review board of the University of South Florida was obtained before commencing the study.

Data Collection Instrument 

Based on the results of the focus groups, a reworded pool of PF scale items was developed. Because we were concerned that the traditional PF items might unduly emphasize lower-body mobility, we added additional items from the function component of the Late Life Function and Disability Instrument, a 32-item tool measuring 3 dimensions—upper-extremity, basic lower-extremity, and advanced lower-extremity functions.13 The instructions to respondents when completing the PF scale were modified to require respondents to consider each item as if they were using their assistive devices. A group of experts in SCI reviewed the preliminary instrument and recommended 2 additional items.

The pool of 30 items was included in a written questionnaire with the other components of the SF-36V and information about activities of daily living (ADLs), level of injury, physical functioning, and demographics. The SF-36V was modified from its original published version to reflect minor wording and response scale modifications recommended in the most recent version of the SF-36 designed for non-VHA populations.14, 15 The VHA Spinal Cord Dysfunction Registry Questionnaire (SCDRQ) was used to obtain a self-reported measure of disease, impairment, and function.16 Items from the SCDRQ were designed to measure the impact of disease, including the cause of the spinal cord condition, the level of injury, areas of the body affected, loss of function, and comorbid conditions. Items from the SCDRQ designed to measure impairment included those measuring basic ADLs and instrumental ADLs (IADLs). Items related to resource utilization such as mobility aids and assistants were also included. Items from the SCDRQ used to classify either paraplegia or tetraplegia included self-reported neurologic level, description of movement loss, and area of the body affected. The resultant questionnaire was pilot tested with 20 patients from the inpatient unit at the Tampa VHA.

Data Analysis 

A series of analytic steps were conducted to evaluate the validity and reliability of the revised PF scale. To establish unidimensionality of the underlying construct of physical functioning, regardless of injury level, we conducted exploratory factor analysis using principal factor extraction followed by orthogonal and oblique rotation. Because the factor solutions correlated, we interpreted the oblique (promax) rotated factor pattern. We also provide the oblique rotated factor structure as a reference for those readers who prefer this matrix when interpreting correlated factors. Because we were concerned that the construct of physical functioning might be operationally different in respondents with less severe injuries compared with those with more severe injuries, we conducted the factor analysis separately on respondents with cervical-level injury (tetraplegia) and those with lower-level injury (paraplegia). Similar factor solutions would support that the items represent an underlying construct in both groups and that the additional analysis could be conducted on the whole sample. Dissimilar results would suggest that separate analyses should be conducted.

The subset of items identified through factor analyses were then analyzed using item response theory (IRT) techniques. We used a generalized partial credit (GPCM) IRT model and the marginal maximum likelihood estimation procedures of the Parscale software17a as recommended by Ware.18 This program is specifically designed to analyze data contained in 2 or more predefined categories. The GPCM, formulated by Muraki,19 is based on Masters’s partial credit model but relaxes the assumptions of uniform discrimination of test items. It also differentiates from Masters’s model in the basic assumptions it makes about the latent trait. The GPCM is viewed as a member of a family of latent variable models.19 We restrict our analysis here to a 1-parameter GPCM model.

Item calibration and scaling were conducted as described by Muraki.19 We examined goodness of fit for each item and for the total set of items. Based on the model specifications, respondents’ scores were placed into 10 groups with expected versus observed frequency distributions and were tested using a likelihood-ratio chi-square statistic. We also interpreted the item-category response functions (ICRF), item information curves, and test information and standard error curves for the set of items. It is beyond the scope of this study to provide an extensive overview of the ICRFs generated by a GPCM IRT model. For such a description, we refer readers to Ware.18 However, we will provide a brief description of the basic components of the ICRF to enable readers unfamiliar with the method to interpret our results. Figure 1 provides an ICRF for one of the items from the item pool calibrated with paraplegic respondents. This item. “lifting or carrying groceries,” has 3 categoric response options: “limited a little,” “limited a lot,” and “not limited at all.” The solid lines represent the model’s prediction of the probability of choosing each of the item response categories for varying levels of ability (physical functioning). The horizontal axis is physical functioning normalized with the mean equal to 0 and the standard deviation equal to 1. The vertical dashed lines represent the intersection points of the first and third item-category response lines, where respondents are equally likely to choose the 2 adjacent response categories (thresholds). These thresholds are used to define the “difficulty” of each response category.18 The midpoint of the interval between the thresholds represents the mean threshold, which we use to represent the item’s location on the PF scale. Figure 2 presents the item information curve for the same item, which for the GPCM represents the change in response with the change in ability level. Item information curves of individual items are summarized in the test information curve (fig 3).19

View Large Image Download to PowerPoint

Fig 1. ICRF: generalized partial credit model (normal metric). Example of the ICRF “lifting and carrying groceries” calibrated with data from paraplegic respondents.

View Large Image Download to PowerPoint

Fig 2. Item information curve: generalized partial credit model (normal metric). Example of the item information function “lifting and carrying groceries” calibrated with data from paraplegic respondents.

View Large Image Download to PowerPoint

Fig 3. Total information and standard error curves for respondents with (A) tetraplegia and (B) paraplegia.

The total IRT scores for each respondent were converted to 0-to-100 SF-36V scoring conventions. Finally, along with the scores from the other SF-36V scales, the SCI-specific PF scores were converted to standardized scores, and the PCS and MCS scores were calculated as described by Kazis.20 To establish reliability the Cronbach α was calculated for the scale. To further establish the validity of the SCI-specific PF scale, a series of bivariate analyses were conducted. Pearson product-moment correlation coefficients were calculated between the SCI-specific PF scale and the other scales of the SF-36V, the PCS and MCS, and the functional status and independence and ADL scales.

Data were scanned into a database using Teleform software.b SASc was used to conduct factor analysis and other descriptive and bivariate analyses. Parscale, version 4.1, was used to conduct IRT analysis.

Phase 1 

The pool of revised PF items is outlined in table 1. Three items were unaltered SF-36V PF items, 12 were reworded PF scale items, and 15 were adapted or new items. In revising items we changed the wording as little as possible. For example, in the item describing “vigorous activities,” the term “running” was replaced with “wheelchair jogging/racing.” Based on the focus groups we could not identify the best way to reword the items that include the word “walking” or “climbing” in the original PF scale. Therefore, 2 options of alternative words for walking (“wheeling” and “going”) and 2 options for “climbing” (“climbing” and “going”) were used. We relied on empiric evidence from phase 2 to choose which of these wordings to include in the final scale.

Table 1. Physical Functioning Item Pool

	Item Pool	Original SF-36 Item	Reworded SF-36 Item	New Item
	Pouring from a large pitcher			X
	Unscrewing a lid			X
	Bathing or dressing yourself	X
	Washing dishes by hand			X
	Making a bed			X
	Bending as if to pick something from the floor			X
	Holding a full glass of water			X
	Reaching overhead			X
	Opening a heavy outside door			X
	Transferring to bed			X
	Lifting or carrying groceries	X
	Shopping for groceries			X
	Getting into and out of a car			X
	Transferring to toilet			X
	Moderate activities, such as moving a table, pushing a vacuum, or bowling		X
	Getting up and down from a curb			X
	Bending or stooping	X
	Wheeling several hundred yards		X
	Wheeling more than one mile		X
	Going up several wheelchair ramps		X
	Wheeling one hundred yards		X
	Going up one wheelchair ramp		X
	Climbing several wheelchair ramps		X
	Climbing one wheelchair ramp		X
	Going several hundred yards		X
	Going more than one mile		X
	Going one hundred yards		X
	Vigorous activities such as wheelchair jogging racing, lifting heavy objectives, and participating strenuous sports		X
	Moving around one floor of home			X
	Shopping for two or more hours			X

NOTE. “X” denotes the source of the item.

Phase 2 

Questionnaires were sent to all veterans with SCI who had been seen at the Tampa VHA SCI center in the previous year (N=787). A total of 392 usable questionnaires were obtained, yielding a response rate of 49.8%. Thirty-three of the respondents were found to have diagnoses other than SCI (eg, multiple sclerosis) and were eliminated from the analysis. Of the 359 patients in the analytic data set (table 2) most were white non-Hispanic (n=287 [80%]), aged 40 to 59 years (n=171 [51%]), and male (n=330 [94%]). Slightly more respondents were categorized as being paraplegic 190 (53%) than tetraplegic 163 (45%), and there were 6 (2%) respondents for whom data were not available to accurately assign them to one of the categories. Respondents used a wide range of devices to aid mobility. Most used a manual device (n=184 [51%]), and 166 (46%) indicated use of a scooter or electric wheelchair. Relatively few respondents (n=22 [6%]) reported that they could walk without the help of assistive devices, and even fewer (n=6 [2%]) reported being bedridden.

Table 2. Respondent Demographics

	Variables	n	%⁎
	Race
	American Indian or Alaska Native	14	4
	Asian American or Pacific Islander	2	1
	White, not of Hispanic origin	287	80
	African American/black, not of Hispanic origin	36	10
	Hispanic	12	3
	Other	8	2
	Age (y)
	<40	29	9
	40–59	171	51
	60–79	116	35
	≥80	18	6
	Sex
	Male	330	94
	Female	21	6
	Injury category
	Paraplegic	190	53
	Tetraplegic	163	45
	Tools to aid mobility
	I walk without help of a special aid or tool	22	6
	I walk with help of a special aid or tool (cane/braces)	57	16
	I use a manual wheelchair	184	51
	I use a scooter or electric wheelchair	166	46
	I don’t get out of bed	6	2

⁎ Rounded to nearest percentage; n range, 334–435 (multiple responses allowed for “Tools to aid mobility” item).

Bivariate analysis of items with 2 different wordings (“going” vs “wheeling,” etc) found no significant difference in response patterns in these items. Because it was acceptable to replace the original PF items worded with both “walking” and “climbing,” the word “going” was chosen to be included in the SCI-specific PF scale. Exploratory factor analysis of the revised PF items for the severity of injury groups (paraplegic and tetraplegic) and interpretation of the eigenvalues and scree plots for each analysis supported a 2-factor solution for each group. For respondents with tetraplegia (n=121) the 2-factor solution explained 81% of the unique variance, whereas for respondents with paraplegia (n=141) 79% of the variance was explained by the factors. (Factor analysis was restricted to respondents with no missing values on any item.) The 2 factors highly correlated for respondents both with tetraplegia (r=.49) and with paraplegia (r=.57); therefore, oblique factor pattern coefficients were interpreted. The factor structure matrix for each factor is provided for readers who prefer to interpret that matrix with correlated factors. The pattern of loadings on the first and second factors in each group was substantially different in the 2 severity groups. For respondents with tetraplegia, the highest loadings in factor 1 involved items measuring upper-body activities, whereas the highest loadings in factor 1 for respondents with paraplegia related to mobility. This pattern reversed itself in the factor loadings for factor 2 in each group. Although these results suggest that the whole item pool did not represent a unidimensional construct in both groups, a close inspection of the results identified a set of 9 items with similar factor loading on factors 1 in both groups (table 3, shaded area). All of the items had a simple structure (loading on only 1 factor) based on responses of those with tetraplegia, but 3 of the 9 item loadings were complex (loading on both factors) among those with paraplegia. The presence of simple structure in the results supports the validity of the factor solution.

Table 3. Results of the Factor Analysis for Respondents With Tetraplegia and Paraplegia

	Item	Factor Loadings (Tetraplegia)				Factor Loadings (Paraplegia)
		Factor Pattern		Factor Structure		Factor Pattern		Factor Structure
		Factor 1	Factor 2	Factor 1	Factor 2	Factor 1	Factor 2	Factor 1	Factor 2
	Pouring from a large pitcher	.91	—	.87	.37	—	.78	.41	.77
	Bathing or dressing yourself	.89	—	.88	.41	—	.73	.52	.78
	Bending as if to pick something from the floor	.83	—	.85	.45	—	.53	.58	.70
	Washing dishes by hand	.83	—	.85	.38	—	.54	.58	.70
	Making a bed	.83	—	.81	.38	.41	.45	.67	.69
	Transferring to bed	.82	—	.77	.32	—	.78	.45	.78
	Unscrewing a lid	.81	—	.79	.36	—	.84	.37	.77
	Transferring to toilet	.80	—	.76	.31	—	.76	.33	.76
	Holding a full glass of water in one hand	.77	—	.75	.34	—	.64	.30	.61
	Shopping for groceries	.77	—	.84	.52	.56	—	.72	.60
	Lifting or carrying groceries	.74	—	.82	.54	.60	—	.75	.62
	Moderate activities, such as moving a table, pushing a vacuum, or bowling	.72	—	.78	.48	.49	—	.65	.56
	Reaching overhead	.71	—	.78	.49	.39	.39	.61	.61
	Bending or stooping	.70	—	.75	.44	.39	—	.53	.48
	Opening a heavy outside door	.69		.78	.53	.47	—	.71	.69
	Vigorous activities such as wheelchair jogging racing, lifting heavy objectives, and participating strenuous sports	.54	—	.61	.42	.58	—	.59	.35
	Getting up and down from a curb	.54	—	.62	.44	.26	.35	.46	.50
	Getting into and out of a car	.47	—	.61	.52	—	.56	.51	.68
	Going more than one mile	—	.90	—	.80	.89	—	.76	.28
	Going several hundred yards	—	.89	.39	.87	.93	—	.87	.42
	Going one hundred yards	—	.85	.46	.87	.81	—	.82	.47
	Going up one wheelchair ramp	—	.77	.49	.83	.63	—	.75	.48
	Going up several wheelchair ramps	—	.75	.46	.79	.82	—	.77	.39
	Moving around one floor of home	—	.53	.49	.64	.34	—	.75	.58
	Shopping for two or more hours	—	.51	.54	.65	.67	—	.73	.48

NOTE. Values were replaced by “—” where factor loadings were less than ∣.30∣ and the difference between loadings on factors 1 and 2 was greater than |.10| within neurologic level of injury. Shaded items had similar loadings on factor 1 across neurologic level of injury groups.

The 9 items loading on the first factor for respondents with tetraplegia and paraplegia were then included in separate IRT model analyses for each severity group. Based on a review of item fit statistics of these models, 1 item (“moderate activities, such as moving a table, pushing a vacuum, or bowling”) was eliminated. For the remaining 8 items both individual item and total model fit statistics supported model fit based on likelihood-ratio chi-square statistics, with all of the P values being greater than .05. Inspection of the item category response functions supported the ordinal relation of the response options for all items. The rank order of the items, as described by the mean category threshold, was very similar, with the most obvious difference being the item “reaching overhead,” which was the easiest item for the respondents in the paraplegic severity category and the fifth most difficult item for the tetraplegic respondents (table 4). Also provided in table 4 are polyserial correlations between responses to the individual items and the total score for the scale. The correlations were very high for respondents with tetraplegia and paraplegia (≥.59). The total information curves and standard error curves for the SCI-specific PF scales for respondents with tetraplegia and paraplegia (fig 3) suggest that the set of items provides substantial information in the middle two thirds of the ability scale, with a rapid decrease for more extreme abilities. Standard errors were lowest in the middle range of the ability scales. Neither model resulted in a wide range of mean category threshold values. Descriptive responses to the items showed a floor effect on many of the items (table 5), particularly for tetraplegic respondents who chose the “yes, limited a lot” option for at least half of the revised PF items.

Table 4. Item Parameters for the SCI-Specific PF Scale

		Tetraplegia			Paraplegia
	Item	Mean Threshold	Standard Error	Polyserial Correlation⁎	Mean Threshold	Standard Error	Polyserial Correlation⁎
	Vigorous activities (eg, wheelchair jogging racing, lifting heavy objectives, participating strenuous sports)	1.48	.20	.76	1.43	.14	.67
	Getting up and down from the curb	0.86	.19	.59	0.80	.14	.66
	Bending or stooping	0.84	.18	.74	0.57	.13	.61
	Opening a heavy outside door	0.64	.18	.89	0.16	.14	.84
	Making a bed	0.62	.17	.86	0.21	.14	.79
	Reaching overhead	0.49	.21	.70	−0.27	.14	.80
	Lifting or carrying groceries	0.47	.18	.84	0.14	.15	.79
	Shopping for groceries	−0.02	.21	.86	−0.23	.13	.80

NOTE. Paraplegic slope ± standard error equals .83±.05; tetraplegic slope equals .80±.07.

⁎ Polyserial correlation between item and scale score.

Table 5. Responses to Items in the SCI-Specific PF Scale

	Item	Tetraplegia⁎			Paraplegia⁎
		Yes, Limited a Lot	Yes, Limited a Little	No, Not Limited at All	Yes, Limited a Lot	Yes, Limited a Little	No, Not Limited at All
	Vigorous activities (eg, wheelchair jogging racing, lifting heavy objectives, participating strenuous sports)	119(77.3)	24(15.6)	11(7.1)	116(64.0)	50(27.6)	15(8.3)
	Getting up and down from the curb	102(66.7)	34(22.2)	17(11.1)	85(33.3)	73(41.0)	20(11.2)
	Bending or stooping	102(66.2)	36(23.4)	16(10.4)	80(44.7)	69(38.6)	30(16.8)
	Making a bed	95(62.5)	37(24.3)	20(13.2)	64(35.8)	78(43.6)	37(20.7)
	Opening a heavy outside door	95(62.9)	39(25.8)	17(11.3)	61(33.3)	85(46.5)	37(20.2)
	Lifting or carrying groceries	91(58.7)	44(28.4)	20(12.9)	56(30.8)	91(50.0)	35(19.2)
	Shopping for groceries	74(47.7)	56(36.1)	25(16.3)	50(27.2)	79(42.9)	55(29.9)
	Reaching overhead	86(55.8)	54(35.0)	14(9.0)	40(21.7)	98(53.3)	46(25.0)

NOTE. Values are n (%). All percentages rounded to nearest 0.1%.

⁎ Paraplegia n range, 178–183; tetraplegia n range, 151–155.

Finally, a series of analyses were conducted to establish reliability and validity of the PF scores based on the models developed. Internal consistency reliability (Cronbach α) calculated for the SCI-specific PF scale was .90. Pearson product-moment correlation coefficients (table 6) between the SCI-specific PF score and the raw scores from the other 7 scales on the SF-36V found that the strongest association (r=.55) was between the revised PF scale and the role–physical subscale. The weakest association (r=.18) was between the PF score and the mental health subscale. The SCI-specific PF scale correlated more highly with the PCS (r=.64) than with the MCS (r=.16). The SCI-specific PF scale score was very strongly associated with ADLs (r=.63) and even more strongly associated with IADLs (r=.70).

Table 6. Measures of Convergent and Divergent Validity

	Variable	n	Pearson Product-Moment Correlation
	SF-36 scale
	Physical health
	Role–physical	337	.55
	Bodily pain	340	.29
	General health	346	.32
	Mental health
	Vitality	346	.22
	Social functioning	346	.29
	Role–emotional	331	.32
	Mental health	341	.18
	Summary scales
	PCS	326	.64
	MCS	326	.16
	Activity of daily living
	ADLs	354	.63⁎
	IADLs	354	.70⁎

⁎ Scale scores reversed to facilitate comparisons.