| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| ||||||||||||||||||||||||||||||||
Department of Kinesiology and Community Health, University of Illinois at Urbana-Champaign.
Address correspondence to Edward McAuley, PhD, Department of Kinesiology, University of Illinois, 336 Freer Hall, Urbana, IL 61801. E-mail: emcauley{at}uiuc.edu
 |
Abstract |
|---|
 Top Abstract MethodsResultsDiscussionReferences |
|---|
Methods. Older black (n = 81) and white (n = 168) women completed the LL-FDI, several measures of physical function, and physical activity measures, and had their body mass index assessed at baseline of an ongoing prospective study. Confirmatory factor analyses (CFA) and correlational analyses were used to examine factorial and construct validity of the measure.
Results. The CFA, using an iterative model modification technique, resulted in an acceptable 15-item solution for the function component and an 8-item solution for the disability component. This abbreviated instrument demonstrated high correlations with the original scales. Construct validity for the LL-FDI was supported. Participants who demonstrated better physical function, reported being more active, and had lower body mass index reported less disability and less difficulty with function on the LL-FDI.
Conclusions. The LL-FDI appears to be an effective instrument for assessing function and disability in older women, and the abbreviated version reported here may prove useful in certain circumstances due to its brevity. However, continued determination of the construct validity of the complete and abbreviated scales is recommended.
Physical disablement in older adults is a process encompassing functional limitation and disability (4,9). Guralnik and Ferrucci (10) used the classification scheme and operational definitions of Nagi (11) and Verbrugge and Jette (9), respectively, to distinguish between these two important components of the disablement process. Functional limitations are concerned with the individual's reduced capacity to carry out an array of activities that are relevant to effective community living, such as walking, climbing, reaching, lifting, and handling everyday objects. Assessment of functional limitation has typically been achieved by using either objective performance tests (e.g., gait speed, chair rises, stair climbing) or by self-report measures (10). Disability, in contrast, reflects individual "limitations in performance of socially defined roles and tasks within a sociocultural and physical environment" (11). Thus, being restricted in one's ability to walk, lift, or climb (i.e., functional limitations) leads to pronounced difficulty in attending to activities required for one's employment, personal care, recreation, etc. (i.e., disability). Disability, therefore, focuses on behavioral repertoires rather than the performance of discrete tasks.
The necessity for sound operational definitions of functional limitations and disability and, accordingly, the importance of accurate measurement have been voiced repeatedly (e.g., 10–14). A major methodological advance in this area has been the recent development of the Late-Life Function and Disability Instrument (LL-FDI). The LL-FDI is comprised of two components. The function component assesses advanced lower extremity function, basic lower extremity function, and upper extremity function. The disability component assesses the frequency of performing social and personal role activities and the limitation in capability of performing instrumental and management role activities (12).
The LL-FDI was developed using a sample of 150 adults aged 60 years and older. The instrument is composed of a comprehensive battery of items which appear to avoid problems associated with many short-form measures, such as ceiling and floor effects, by spreading items across a broad range of activities (12). For both the function and disability components of the LL-FDI, final scale solutions were arrived at by exploratory factor analysis (EFA), to identify latent factors, and one-parameter Rasch rating scale analyses (15), to estimate calibrations along a common scale. Interclass correlations indicated good test–retest reliability, and coefficient alpha reflected acceptable-to-excellent internal consistencies (
=.63–.92). Preliminary validity was determined by examining the ability of LL-FDI scores to discriminate between groups of known functional limitation. These latter groups were determined by scores on the physical function subscale (PF-10) of the Medical Outcomes Study Short Form-36 (SF-36; 16), and participants were classified as having no, slight, moderate, or severe functional limitation. Overall, these preliminary validity data suggest that the development of the LL-FDI can be considered an important contribution to the assessment of function and disability in older adults.
The objective of the present study was to further examine the psychometric properties of the LL-FDI in a sample of white and black older women. To do so, we tested the veracity of the proposed three-factor solution of the functional limitation component and the two-factor structure of the frequency and limitations aspects of the disability component. Assuming a robust and replicable factor structure, our next objective was to test the construct validity of the LL-FDI. As physical activity has been repeatedly associated with improved function and reduced incidence of disability (5), we expected that higher levels of physical activity would be inversely related to disability and function. Additionally, we expected to witness higher levels of physical function (e.g., gait speed, stair climbing) in those individuals who reported less difficulty with lower extremity function and less disability. Finally, as obesity has been consistently linked to ineffective functioning and higher rates of disability (e.g., 17), we expected body composition (i.e., greater body mass index [BMI]) to be associated with more difficulty in functioning and reported disability. In summary, we tested the construct validity by correlating the LL-FDI scales with measures of physical function, body composition, and self-reported physical activity.
|
METHODS |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
Recruitment
Recruitment efforts included advertisements in local newspapers, features on local television news programs, and public service announcements during local radio broadcasts. In addition, letters and flyers advertising the study were mailed to senior centers, fitness centers, adult health care facilities, clergy at predominantly black churches, and black-owned local businesses within the community. Initially, 298 individuals expressed interest in participation with 50 individuals being declared ineligible or declining further participation following a telephone screening interview.
Measures
Demographic and health information.--
Each participant was asked to provide current demographic information and details of her medical history including current medications. Baseline demographic and health status information can be found in Table 1.
|
Physical activity participation.-- Physical activity was assessed with two measures, the Physical Activity Scale for the Elderly (PASE; 19) and the Community Healthy Activities Model Program for Seniors physical activity measure (CHAMPS; 20). The PASE is a 10-item instrument specifically designed to assess physical activity in large samples of older persons over a 1-week time period. The PASE assesses frequency and duration of participation in leisure activities (e.g., walking outside the home; light, moderate, and strenuous sport; and recreation), along with participation in housework (light and heavy), lawn work and yard care, home repair, outdoor gardening, and caring for others. The total PASE score was computed by multiplying the amount of time spent in each activity (hours per week) or participation (yes or no) in an activity by empirically derived item weights (19) and summing over all activities. Scores from the PASE have been reported to be a valid measure of physical activity participation in elderly persons (21,22).
The CHAMPS is a 41-item questionnaire which can be administered by interview or by a paper and pencil and assesses the average weekly participation in physical activities over the past month. From these data, frequency of participation in moderate intensity physical activities and frequency of participation in physical activities of all intensities is determined. Using a compendium of energy values for each activity (20), the caloric expenditure per week from engagement in moderate intensity physical activity and from physical activities of all intensities can be calculated. Stewart and colleagues (20) provide good construct validity for the use of the CHAMPS with older adults.
Objective physical function measures.-- We used five commonly used measures of physical function. To assess gait speed we assessed normal walking speed over a 7-meter pathway. In addition, we assessed participants' walking speed while negotiating a 32-cm tall obstacle placed across the center of the 7-meter pathway. Participants were also timed on their ascent and descent of a flight of 15 stairs. Finally, participants completed the timed 8-foot Up-and-Go test (23), which assessed the participants' ability to rise from a seated position, walk a distance of 8 feet as quickly as possible, and return to a seated position.
Body composition.-- We used BMI as an indicator of overall body composition. BMI was calculated by dividing weight (in kilograms) by height (in meters squared).
Procedures
All data were collected at baseline of the ongoing 24-month Activity, Gait, and Efficacy (AGE) prospective trial. Upon completion of the initial telephone screening interview, participants were scheduled for assessment in our laboratories. Prior to testing, participants completed an approved Institutional Review Board informed consent and a questionnaire assessing basic demographic information, physical activity (PASE), and general medication and health information. The LL-FDI and CHAMPS data were collected via interview in a dedicated research participant reception area. Upon completion of the interview, participants were escorted to the Gait and Balance Laboratory for the assessment of height and weight and for physical function testing.
Entry to this laboratory is gained by descending a flight of 15 stairs. Participants were instructed to walk down the stairs at their normal speed and to use the handrail if desired. Timing was initiated as soon as the participant lifted a foot to begin her descent following a start command given by one of the experimenters. The trial was terminated when the participant's trail foot made contact with the floor at the base of the stairs. A similar approach was used to measure stair ascent at the end of testing as participants left the basement laboratory. Two research assistants using handheld stopwatches timed the ascent and descent trials, with final values averaged across the two experimenters. Upon entry into the laboratory, height and weight were assessed.
Next, participants completed the two walking tasks, that is, walking 7 meters with no obstacle in their path and walking over a wooden obstacle 32 cm in height (10 cm wide) placed in the center of a marked 7-meter path. Participants began each trial standing at one end of a hardwood floor walkway. Timing of each trial started when the participant initiated her first step and terminated when her lead foot crossed the line at the end of the pathway. Participants were instructed to perform these tasks at their normal walking pace and were given the option of walking around the obstacle if they did not feel comfortable walking over it. Finally, participants completed the 8-foot Up-and-Go Test, in which they began by sitting with their back straight in a chair, hands on their thighs, and feet flat on the floor, with one foot slightly in front of the other. On the start command, participants were required to get up from the chair and walk as quickly as possible around a cone 8 feet from the chair and return to a sitting position in the chair. For each of these tasks, two experimenters used handheld stopwatches to time the participants. Each condition was repeated twice, and the data were averaged across the two experimenters and across the two trials. All participants were given ample rest and recovery time between each of the tasks.
Data Analysis
Confirmatory factor analysis.--
The fit of measurement models for the components of the LL-FDI was examined using confirmatory factor analysis (CFA) with maximum likelihood estimation in LISREL 8.0 (24). Briefly, researchers using CFA postulate an a priori measurement model linking observed variables with latent factors, and then test that model for its ability to fit the data. The a priori measurement model specifies the pattern of fixed and freed parameters in matrices containing factor loadings, factor variances and covariances, and item uniquenesses. Maximum likelihood was selected to estimate the freed parameters in matrices, because it is the standard estimation technique (25) and has resulted in accurate absolute and relative fit indices with ordered categorical data of varying degrees of kurtosis (26).
Model fit.-- The fit of the measurement model for the data was based on the chi-square statistic, standardized root mean square residual (SRMR), root mean square error of approximation (RMSEA), Non-Normed Fit Index (NNFI), and Comparative Fit Index (CFI). The chi-square statistic assesses perfect fit of the model to the data (27). The SRMR is the average of the standardized residuals between the specified and obtained variance–covariance matrices. The SRMR should be less than.08 to indicate good model fit (28) The RMSEA value represents closeness of fit and should approximate or be less than 0.05 to demonstrate close fit of the model (29). The 90% confidence interval (CI) around the RMSEA point estimate should contain 0.05 to indicate the possibility of close model–data fit (29). Both the NNFI and CFI are incremental fit indices, and test the proportionate improvement in fit by comparing the target model to a baseline model with no correlations among observed variables (30,31).Values approximating 0.90 (NNFI) and 0.95 (CFI) are indicative of acceptable and good model–data fit, respectively (28,30,31).
Model modification.-- In the event that poorly fitting models emerged from the initial series of analyses, an a priori decision was made that these models would be subjected to a specification search and the model modified accordingly. All model modifications were conducted through an iterative process that involved removing a single item and then rerunning the analysis. Items were removed based on large standardized residuals (i.e., greater than ±2.00) and modification indices in the uniqueness matrix in combination with substantive arguments relative to item content such as redundancy, salience, and ambiguity (27). Large standardized residuals and modification indices in the uniqueness matrix identified pairs of items in which the covariance was either overpredicted or underpredicted by the model. One of the two items was removed based on redundant content, salience, or content that was ambiguous for determining its placement within the model. The CFA was then rerun to determine whether the modification resulted in an improved fit. This process was continued until a reasonable model was generated as indicated by absolute and relative fit indices. Importantly, modifications were only made in those cases in which it was substantively appropriate to do so (27).
|
RESULTS |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
2 = 190.24, df = 103, SRMR =.07, RMSEA [90% CI] = 0.06 [0.05–0.07], NNFI = 0.75, CFI = 0.79) and was subsequently subjected to a specification search, as previously described. This search resulted in a final model consisting of 8 items, with 4 items loading on each of the personal and social role factors. The final model provided an improved and reasonable fit for the data (2 = 29.68, df = 19, SRMR =.05, RMSEA [90% CI] = 0.05 [0.00–0.08], NNFI = 0.90, CFI = 0.93). Although the internal consistency of the social role subscale was acceptable ( =.67), the personal role subscale was somewhat problematic ( =.38). However, the composite reliabilities (i.e., variance captured by items versus variance associated with measurement error) for the social and personal role subscales were.63 and.44, respectively, which approximated the threshold value of.50 specified by Fornell and Larcker (32).
In addition, we calculated the internal consistency and composite reliability for the personal role subscale using all of the original items with marginal improvement ( =.47, composite reliability =.54). The items and factor loadings for the original model and the final model are presented in Table 2. The standardized correlation between the personal and social role factors in the final model was 0.21.
|
2 = 614.95, df = 103, SRMR =.08, RMSEA [90% CI] = 0.16 [0.14–0.17], NNFI = 0.71, CFI = 0.76). We initially attempted the aforementioned iterative approach of model modification, but could not identify a reasonable final model that was theoretically appropriate and sufficient in breadth of content. Accordingly, we reran the analyses, using only those 8 items that made up the final model from our analyses of the frequency aspect of the LL-FDI Disability component. Testing this model was justified, as it represented a scale that was similar in item-content representation and structure with the frequency component of the LL-FDI and was consistent with the original definitions of the instrumental and management subscales of the limitation component of the LL-FDI. Hence, the model consisted of 8 items, with 4 items reflecting limitations in capabilities to perform social tasks (e.g., travel, visit friends) and 4 items reflecting limitations personal tasks (e.g., taking care of the household, errands, and personal needs). For the sake of clarity, we retain these labels for all discussion of the limitation component of the LL-FDI. This model provided a reasonable fit for the data (2 = 56.72, df = 19, SRMR =.05, RMSEA [90% CI] = 0.08 [0.06–0.11], NNFI = 0.93, CFI = 0.96). Internal consistencies for both scales were acceptable with =.77 (personal role) and.83 (social role). The estimates of composite reliability for the personal and social roles subscales were.85 and.78, respectively. The items and factor loadings for the original and new, abbreviated models are presented in Table 3. The standardized correlation between the personal and social role factors was 0.70.
|
2 = 1,686.63, df = 461, SRMR =.08, RMSEA [90% CI] = 0.11 [0.10–0.11], NNFI = 0.72, CFI = 0.74). Using the iterative approach previously described, the model was re-estimated. The final model consisted of 15 items, with 5 items each for advanced lower body extremity function, basic lower body extremity function, and upper extremity function factors. This final model provided an improved and reasonable fit for the data (2 = 137.63, df = 87, SRMR =.05, RMSEA [90% CI] = 0.04 [0.03–0.06], NNFI = 0.95, CFI = 0.96). Internal consistencies were good for the lower extremity function scales, =.85 (advanced) and.76 (basic), but weaker for the upper extremity function scale ( =.58). The composite reliabilities were.86,.78, and.63 for the advanced, basic, and upper extremity subscales, respectively. The items and factor loadings for the final model are presented in Table 4. The standardized correlations between advanced lower body extremity function and basic lower body extremity function (r =.65) and upper extremity function (r =.39), as well as upper and basic lower extremity function (r =.50) were all significant.
|
|
|
Correlations Between Abbreviated Components of the LL-FDI and Measures of Physical Function, Body Composition, and Physical Activity
To further determine the construct validity of the abbreviated version of the LL-FDI, we computed Pearson product–moment correlations between the function and disability scales and the physical activity (PASE and CHAMPS), physical function, and body composition measures. The correlations are provided in Tables 7 and 8. We also calculated correlations between all measures and the full scales. These relations were in the same direction and of similar magnitude as the abbreviated version of the LL-FDI. A table of these correlations is available on request.
|
|
Body composition.-- As expected, the strongest correlations between body composition and function and disability were evidenced by the lower extremity function–BMI relationships (see Table 7). Individuals with a higher BMI reported substantial difficulty in performing advanced lower extremity tasks (r = –.43, p <.0001) and significant difficulty in performing basic lower extremity tasks (r = –.28, p <.001). Additionally, greater limitations in carrying out personal roles (r = –.25, p <.001) and social roles (r = –.18, p <.005) were associated with higher BMI.
Physical activity.-- To examine the proposition that higher levels of physical activity would be associated with better function and less disability, we correlated the LL-FDI scores with the total PASE score and the frequency of moderate intensity exercise and kilocalories expended in moderate intensity exercise, as measured by the CHAMPS. All correlations were significant, although relatively modest, with more physically active women reporting less difficulty in function and more frequent participation in social role activities, as well as fewer limitations in performance of social and personal role activities. All correlations are shown in Table 8.
|
DISCUSSION |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
Initial attempts to fit the hypothesized factor structures to the battery of items from the original measure met with little success. Subsequent iterative procedures testing the fit of a more parsimonious item structure were able to provide both an acceptable fit and to confirm the underlying structure in each of the scales, as proposed by Haley and colleagues (14) and Jette and colleagues (13). We first considered the three-factor structure of the function scale designed to assess advanced lower body extremity function, basic lower body extremity function, and upper body extremity function. The original 32-item function scale was reduced to a 15-item scale (5 items per subscale), which demonstrated good model fit and high correlations with the original scale. Internal consistency and composite reliabilities were acceptable. From a construct validity perspective, the advanced and basic lower extremity function scales demonstrated moderate associations with all measures of physical function. Given that these measures rely heavily on walking quickly, climbing stairs, etc., these correlations support the convergent validity of the function scales. More importantly perhaps, these correlations support Jette's (12) perspective that objective physical function measures, often used to assess disability and function, are not necessarily the conceptual equivalent of such constructs.
Further evidence for the construct validity of the function scales emanates from correlations with BMI and physical activity. Stronger correlations were found between BMI and limitations in advanced lower extremity tasks than between BMI and basic lower extremity function, as one would expect. For example, having greater BMI should make it more difficult to walk up and down three flights of stairs or run half a mile than, than, for example, walk around the floor of one's home. Moreover, lack of association between BMI and upper extremity function (e.g., using utensils or pouring from a large pitcher) attests to the discriminant validity of the scale. Finally, function scores were associated with physical activity participation, supporting work by Miller and colleagues (5). Once again, convergent validity was evidenced by significant correlations between physical activity and advanced and basic lower extremity function. In sum, we believe that the abbreviated function scale of the LL-FDI demonstrates acceptable factorial and construct validity.
As with the function component, analysis of all items in the frequency and limitation scales of the disability component of the LL-FDI failed to produce a good fit of the model to the data. These 16 items were reduced iteratively to produce an 8-item scale that fit the data well. With respect to the frequency of performance subscales, the items loading on the social role and personal role domains came from the same pool of items that loaded on the original scale domains. However, although composite reliability was acceptable, the poor internal consistency of the personal role domain is both surprising and troublesome, and further examination of this scale and its composition may be warranted. Examination of relationships between physical activity, physical function, BMI, and the frequency aspect of the disability component of the LL-FDI frequency subscales demonstrated a differential pattern of relationships that would suggest support for its construct validity. For example, greater BMI was associated with less frequent performance of activities in the personal domain (e.g., taking care of personal errands), but was unrelated to the social domain (e.g., going out with others to public places). However, being more physically active and performing better on the physical function measures were associated with more frequent participation in social activities, but less so in personal activities. Perhaps body fatness plays an important role in the frequency of doing personal tasks, whereas being active and having better functioning are important for doing social tasks that require a greater amount of movement (e.g., going out to public places or traveling out of town). Whether these latter relationships are affected by the suspect reliability of the personal role domain scale needs further evaluation.
Our analysis of the limitation of capabilities scale indicated good internal consistency and composite reliability for both the social role domain and the personal role domain. With respect to construct validity, better performance on the physical function tasks and lower BMI were associated with less disability relative to both social and personal roles, although these relationships were slightly stronger for the latter. Once again, these relationships suggest that physical function measures are not the conceptual equivalent of disability.
We note that the factor structure for the 8 items in the limitations component in this study are not entirely equivalent to that of the original item structure. However, the items of the newly formatted social and personal roles factors match up quite well with the management and instrumental roles definitions, respectively. The items on the social roles scale (e.g., Invite family and friends into home) tap the original conception of management as a domain that encompasses "organization or management of social tasks" but that might involve mobility or physical activity. Similarly, the items on the personal roles scale (e.g., Take care of local errands) tap the original conception of Instrumental roles as a domain that encompasses "activities at home and in the community." Additionally, as Jette and colleagues (13) point out, the nature of the instrument application will drive which domain scores or combination of scores will be used in any research endeavor. Use of total domain scores offers greater stability and precision (13), and, therefore, assignment of items to one domain or another may be irrelevant depending on the research question. If one's interest is purely in the degree to which participants have limitations in capabilities, then total scores, regardless of domain loading, are preferable.
Interestingly, there were only negligible differences between black and white women on perceptions of function and disability, as measured by the LL-FDI. In contrast, there were significant and moderately sized differences between the two groups in frequency per week of moderate intensity exercise activities and the objective physical function measures. In all cases, white women outperformed black women. Once again, these findings support the position that objective and perceptual measures of function and disability are not necessarily isomorphic constructs.
It should be noted that it was never the original intent of this study to arrive at a shorter version of the LL-FDI. However, given the relatively robust correlations evidenced between the original scale and the current version, the latter may prove useful for those researchers interested in brevity while maintaining instrument integrity. Nonetheless, one must consider why the original model did not present a good fit to these data; there are several possible explanations for this.
One explanation involves the theoretical and mechanical bases underlying the analytic methods for evaluating the factor structure. In the present study, we used CFA, which is a theoretically driven approach for testing how well a model is able to explain the covariances among a set of items. In contrast, the original development of the LL-FDI used EFA, a data-driven approach for identifying (rather than confirming) a model that explains the covariances among a set of items. Thus, there is a fundamental mechanical difference between analytic methods such that in CFA the items are typically forced to load on only one factor, whereas in EFA the items freely load on all factors but typically demonstrate a dominant loading on only one factor. It is also quite possible that the difference in goodness of fit between the two models may be due to the fact that we sampled only women whereas the original developmental study samples both men and women. Additionally, it is not known whether the factor structure may be differentially affected by race. To address these types of issues, multisample simultaneous CFAs would be required, however, our sample size precluded such an approach. Finally, the degree and severity of functional limitation and/or disability in the original and present samples may have differentially affected the factor structure.
Use of the complete battery of items may be preferable for those interested, for example, in examining intervention effects of specific aspects of function and disability. As Jette (12) notes, the practicality of short-form measures such as the abbreviated version of the LL-FDI has a presumed cost, lack of precision, rendering them limited in the applicability of all items to all participants across all situations. Jette's (12) recommendation to use the full battery of items using computer-adaptive testing may ultimately be the most practical way to assess function and disability without sacrificing precision. However, access to computer-adaptive testing will not be universal or always practical. Economics and time constraints may preclude its use, thereby making the abbreviated version of the LL-FDI more attractive. In closing, we note that construct validity is an ongoing process, and continued evaluation of the psychometric properties of scores from the full and abbreviated versions of the LL-FDI is warranted.
|
Acknowledgments |
|---|
|
Footnotes |
|---|
Received February 27, 2004
Accepted April 7, 2004
|
References |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
This article has been cited by other articles:
 |
|
 |
|
  A. M. Jette, S. M. Haley, P. Ni, S. Olarsch, and R. Moed Creating a Computer Adaptive Test Version of the Late-Life Function and Disability Instrument J. Gerontol. A Biol. Sci. Med. Sci., November 1, 2008; 63(11): 1246 - 1256. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 K. S. Morris, E. McAuley, and R. W. Motl Self-efficacy and environmental correlates of physical activity among older women and women with multiple sclerosis Health Educ. Res., August 1, 2008; 23(4): 744 - 752. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||
| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
