Psychometric properties of the Late-Life Function and Disability Instrument: a systematic review

Background The choice of measure for use as a primary outcome in geriatric research is contingent upon the construct of interest and evidence for its psychometric properties. The Late-Life Function and Disability Instrument (LLFDI) has been widely used to assess functional limitations and disability in studies with older adults. The primary aim of this systematic review was to evaluate the current available evidence for the psychometric properties of the LLFDI. Methods Published studies of any design reporting results based on administration of the original version of the LLFDI in community-dwelling older adults were identified after searches of 9 electronic databases. Data related to construct validity (convergent/divergent and known-groups validity), test-retest reliability and sensitivity to change were extracted. Effect sizes were calculated for within-group changes and summarized graphically. Results Seventy-one studies including 17,301 older adults met inclusion criteria. Data supporting the convergent/divergent and known-groups validity for both the Function and Disability components were extracted from 30 and 18 studies, respectively. High test-retest reliability was found for the Function component, while results for the Disability component were more variable. Sensitivity to change of the LLFDI was confirmed based on findings from 25 studies. The basic lower extremity subscale and overall summary score of the Function component and limitation dimension of the Disability component were associated with the strongest relative effect sizes. Conclusions There is extensive evidence to support the construct validity and sensitivity to change of the LLFDI among various clinical populations of community-dwelling older adults. Further work is needed on predictive validity and values for clinically important change. Findings from this review can be used to guide the selection of the most appropriate LLFDI subscale for use an outcome measure in geriatric research and practice.


Background
Accurate assessment of physical functional limitations and disability is critical for improving access to health care services for older adults, and for evaluating the effectiveness of interventions designed to slow or prevent the progression of late-life disability [1,2]. Detecting meaningful changes in function and disability in older adults can be challenging, particularly if the outcome tool is not designed to accurately assess or reflect the purported change. The choice of outcome measure for use as a primary outcome in studies with older adults should be guided by the construct being measured and evidence for its psychometric properties [3].
Patient-reported measures (PROs) of function and disability are commonly used in studies of older adults because of their low cost and convenience. However, many existing measures were not designed for evaluative purposes and do not offer a comprehensive assessment of function or disability based on an explicit theoretical framework [4]. The Late-Life Function and Disability Instrument (LLFDI) was developed to overcome some of these limitations [5,6]. Unlike many other PROs, the LLFDI comprehensively assesses discrete functional tasks and operationalizes disability in important life roles beyond the narrow construct of activities of daily living.
The conceptual underpinnings for the LLFDI was Nagi's disablement model [7] and also draws from the World Health Organization's International Classification of Functioning, Disability, and Health (ICF) [8]. The LLFDI assesses both functional limitations (inability to perform discrete physical tasks) and disability (inability to participate in major life tasks and social roles). The Function component evaluates difficulty in performing 32 physical tasks and is comprised of an overall scale of function and three subscales: basic lower extremity, advanced lower extremity and upper extremity. The Disability component evaluates limitations in and frequency of taking part in 16 major life activities. The frequency dimension is comprised of social and personal role subscales plus an overall scale; the limitation dimension includes instrumental and management role subscales plus an overall scale. Raw scores are transformed to scaled scores (0-100) based on a Rasch model with higher scores indicating better levels of functioning.
Since its development in 2002, the LLFDI has been frequently used as an outcome measure in geriatric research. While the original LLFDI development papers [5,6] provide preliminary support for its validity and reliability, there is no synthesis of research on its psychometric properties. The objectives of this systematic review are to characterize the use of the LLFDI in published studies of community dwelling older adults and to evaluate the current available evidence on its psychometric properties.

Methods
We conducted a systematic review of studies reporting results of the administration of the LLFDI in communitydwelling older adults. The methodology is based on PRISMA guidelines [9] for systematic reviews.

Search strategy
Searches were performed by one investigator (MB) in consultation with a librarian. Study identification began with electronic searching of the ISI Web of Science for studies citing the two original LLFDI development papers [5,6]. We also searched the following electronic databases from inception until January 28th 2013: PubMed, Web of Science, CINAHL, PsychInfo, Google Scholar, JSTOR, ScienceDirect, WileyInterscience, and EMBASE. Key search terms were "Late Life Function and Disability Instrument", "LLFDI" and "Late life FDI". Finally, reference lists from relevant studies were hand-searched to ensure all possible studies were identified.

Inclusion criteria
Two investigators (MB and CS) independently screened abstracts of retrieved papers with disagreements resolved by discussion. Full texts of relevant studies were then independently assessed by two reviewers (MB and CS) with disagreements resolved by consultation with a third reviewer (AJ). Inclusion criteria comprised: Types of studies: Any study design reporting results based on administration of the original version of the LLFDI. Types of participants: Studies including communitydwelling (non-institutionalized) older adults (mean age > 60 years).
Studies not published in English and conference abstracts were excluded.

Data extraction
Two investigators (CS and MP) independently extracted data into a standardized form. The data extraction form was pilot tested prior to its use to ensure clarity and consistency. A third investigator (MB) reviewed and verified the extracted data for each study.
Data on background characteristics (participants, study purpose, sample size, design, scales reported) were extracted for each study. Thereafter, where available, data related to construct validity (convergent/divergent and known-groups), reliability (test-retest), and sensitivity to change (between-group results and within-group analyses) were extracted.

Data synthesis
Data related to each psychometric property were summarized in tables. By convention, we interpreted a correlation coefficient of <0.3 as weak, 0.3 to 0.7 as moderate and >0.7 as strong. To facilitate synthesis of the sensitivity to change findings, where possible, we calculated Cohen's effect sizes [10] (mean change/SD baseline ) for within-group analyses. Graphs were created to visually depict the effect size results by scale. Values of 0.20, 0.50, and 0.80 have been used to represent small, moderate and large effect sizes, respectively [10].

Convergent/divergent validity
Data related to convergent/divergent validity of the LLFDI, that is, the degree to which LLFDI components and subscales correlated with measures of conceptually related (convergent) or unrelated (divergent) constructs, were extracted from 30 studies [12,13,15,17,25,[27][28][29]32,33,[36][37][38]42,44,45,47,49,51,52,56,[61][62][63]65,66,68,71,72,74]. We hypothesized that moderate to strong correlations would be seen for variables theoretically related to function and disability (i.e., health status, function, mobility, balance and physical activity measures) while weak to moderate correlations would be observed for those variables less related theoretically to function and disability (e.g., biochemical markers). The correlation coefficients reported in the text below represent the range of coefficients observed between the various scales of the LLFDI and the related measure of interest. Detailed results for each individual study (correlation coefficients and statistical significance for each subscale) are outlined in Table S2 of Additional file 1.
In general, evidence for convergent validity was strongest for the overall function scale followed by the two lower-extremity sub-scales. The upper extremity subscale showed the lowest associations with other measures of function; however the latter primarily consisted of lower-extremity tasks. Evidence for divergent validity was shown by the weaker to moderate correlations found between the LLFDI Function component and less theoretically related constructs (neighbourhood walkability scores, Acylcarnitine factor scores, Vitamin D metabolites, B12, folate, Tangible Social Support Scale, age, BMI, income, education) [17,49,63,72,74].  [65]. Moderate to strong correlations were also seen between LLFDI Disability and single-concept mobility scales such as the PASE (r = 0.54 to 0.56) [44] and mGES (r = 0.32 to 0.63) [56].
In general, the limitation dimension showed greater associations with the self-report and performance-based measures than the frequency dimension. Evidence for divergent validity was shown by the generally weak correlations between the LLFDI Disability component and less theoretically related constructs (neighbourhood walkability scores, Vitamin D metabolites, B12, folate, coping strategies, pain, body fat percentage, BMI) [17,27,37,65,72].

Known-groups validity
Data related to know-groups validity of the LLFDI, that is, the degree to which scores of the Disability and Function components distinguished between groups known to differ, were extracted from 18 studies [5,6,27,29,30,[36][37][38]40,47,48,51,52,61,68,69,72,73] and are shown in Table 1. Discrimination between groups was considered if comparisons of the LLFDI between different subgroups of an independent measure or external parameter achieved statistical significance.

Disability component
The Disability component of the LLFDI discriminated between groups based on gender [29], race [73], level of function and mobility limitation [5,68], depression [38], anxiety [61], cane use [52], gait speed [47] and walking exertion [36]. Unlike the Function component, the Disability component did not discriminate between groups based on residence status [29], urinary incontinence [37] or fall status [51]. Evidence for known-groups validity was strongest for the limitation dimension and associated instrumental role domain compared to the frequency dimension and associated domains.

Reliability
Only three studies [5,6,52] included information related to the test-retest reliability of the LLFDI. Short-term stability of the English version of the LLFDI was only examined in the original development papers.

Discussion
Since its conception in 2002, the LLFDI has been used as an outcome measure in over 70 studies including more than 17,000 community-dwelling older adults. Evidence for its psychometric properties has been demonstrated across a wide range of older clinical populations and contexts. The choice of LLFDI sub-scale for use in individual studies should depend on the construct of interest and evidence for relevant psychometric properties in the most applicable population. Results of this review can be used by researchers to guide future decisions regarding the use of the LLFDI as an outcome measure for clinical research in community-dwelling older adults.
The construct validity of both the Function and Disability components of the LLFDI was well-supported by the evidence found in this review. We noted moderate to strong convergent validity between the Function component and well-validated self-report and performance-based measures of function such as the PF-10 and SPPB. In addition, while there is no accepted gold-standard measure of disability, the Disability component was moderately associated with general health status measures such as the LHS and RAND-36 as well as with many commonly used self-report and performance-based measures of function. The LLFDI also showed strong known-groups validity with both components discriminating between groups based on various functional, demographic and medical categories. Our review did not reveal any studies evaluating the use of LLFDI measures of Function or Disability for predicting institutionalization or mortality, highlighting the need for further research on the predictive validity of the LLFDI.
Only three studies [5,6,52] investigated the test-retest reliability of the LLFDI and two were the original development papers. While very high reliability scores (ICCs 0.91-0.98) were reported for all Function scales, a wider range of reliability was reported within the Disability component (ICCs 0.44-0.82). In general, the Disability limitation and frequency dimensions showed moderate to high test-retest reliability with the limitation dimension and instrumental role domain showing the best reproducibility. The management role domain had the lowest reliability, likely due to the limited 4-item pool of this scale. Larger studies on test-retest reliability of the LLFDI would be helpful, especially in light of the lower reproducibility reported for the Disability component.
PROs are often thought to have limited capacity for detecting change given their breadth of measurement and vulnerability to external influences [1,80,81]. In this review, sensitivity to change of the LLFDI was confirmed based on findings from 25 studies. Most scales demonstrated small to moderate effect sizes in positive trials and in cohort studies in which the participants underwent a change in health status. In particular, we noted larger effect sizes for the basic lower extremity scale and summary score of the Function component as well as for the limitation dimension of the Disability component as compared to the other LLFDI scales. These results should be considered when selecting the most appropriate scale for use in clinical trials and longitudinal studies with community dwelling older adults. Only one study [75] attempted to define a clinically meaningful difference for the LLFDI, however this study included only men was based on a testosterone intervention. There remains a need for further work to determine the increments of change on the LLFDI that are clinically meaningful.
Our findings are subject to several limitations. A quality assessment was beyond the scope of this review and very few studies were designed specifically to measure psychometric properties of the LLFDI. We were unable to perform any formal meta-analysis due to the heterogeneity in study outcomes, clinical populations and design. While every attempt was made to identify relevant studies, it is possible that some studies were missed. Finally, our results are only applicable to the original version of the LLFDI administered in community-dwelling older adults. An abbreviated version of the instrument [50] has been developed as well as a computer adaptive version [82] and the psychometric properties of these instruments should be considered separately.

Conclusions
In summary, we have conducted a systematic review of the use of the LLFDI and evidence for its psychometric properties based on 71 published studies. While we have shown extensive data supporting the instrument's construct validity and sensitivity to change among various clinical populations of community-dwelling older adults, further work is needed to determine the LLFDI's predictive validity and values for clinically meaningful change. Results from this review can be used to inform the selection of the most appropriate LLFDI component and subscale for use as an outcome measure in geriatric research.

Additional file
Additional file 1: Outlines data extraction results for each study in Tables S1, S2 and S3 as per below. Table S1. Characteristics of studies reporting results based on the administration of the Late Life Function and Disability Instrument.  Competing interests AMJ has stock holdings in CREcare, LLC, a small business created to disseminate outcome instruments such as the LLFDI.
Authors' contributions MKB was responsible for the conception, design and coordination of the study, data acquisition and interpretation, and drafting and revising the manuscript. CTS contributed to the acquisition of data and revision of the manuscript. MMP contributed to the acquisition of data and revision of the manuscript. JFB contributed to the general supervision of the study, conception, design, interpretation of data and revision of the manuscript. AMJ contributed to the general supervision of the study, conception, design, interpretation of data and revision of the manuscript. All authors approved the final manuscript.