A number of methods for examining functional status in older adults have been derived, among the oldest of which is the Katz index.1 This is an index that grades individuals on a binary scale of dependent or independent for 6 activities of daily living (ADL): bathing, dressing, toileting, transferring, continence, and feeding. The significance of this index is not only that it grossly quantifies one's level of functional independence, but it was also found to represent a hierarchy of ADL difficulty with bathing being the most difficult and eating the least difficult. Based on this hierarchy, patients could be given a functional grade from A to G based on their abilities to perform the tasks in this index. In this index a grade of A represents independence with all ADLs, and a grade of G represents dependence in all ADLs; intermediate letters are assigned in order of increasing dependence on the Katz index. Notably, the population in which this hierarchy was found is older adults the majority of whom appear to have come from the community or the acute care setting.1
More recently, disability-staging systems have been used on the basis of the FIM (Uniform Data System for Medical Rehabilitation, Amherst, New York) for the evaluation of patients in an inpatient rehabilitation facility2–4 and an ADL questionnaire for community-dwelling older adults.5 However, as of yet the hierarchy of ADL has not been well studied in the skilled nursing facility (SNF) population. Functional status has in general not been well studied in this population, which likely has either medical complexity or nursing needs that exceed their community-dwelling counterparts. As such, this study seeks to describe the relative difficulty of the 6 ADLs in the Katz index in patients in the SNF population as a method of functional status grading.
This study used the National Nursing Home Survey for the most recent year for which data were available, 2004. This is a cluster-stratified survey designed to capture data from a representative sample of more than 13 000 nursing facility residents spread throughout the United States from 1174 nursing facilities. Data collected include demographic information (eg, age, length of stay, race, and gender), medical data (admission diagnoses, medications, and pain levels), functional data (eg, ADL independence and fall data), payor source, and facility information. Complete information and the data set itself are available from the Centers for Disease Control Web site.6
This study examined the functional data for the 6 items in the Katz index.1 The Katz index was chosen, because it was designed for use in older adults and has been well studied previously in numerous populations.1,7–10 In addition, its binary responses make it straightforward to complete even for health care providers with minimal training. Continence is recorded as 2 separate questions in the National Nursing Home Survey, one for bladder and the other for bowel. As such, a single continence category was created by combining these 2 questions. There are also questions regarding bed mobility, walking on and off the unit, locomotion on and off the unit, and personal hygiene, which were not analyzed since they are not part of the Katz index and do not themselves comprise a validated functional measure. Independence levels for the ADL questions are recorded on a scale of 0 to 5, where 0 represents independence and 5 represents complete dependence. These ordinal values were converted to binary responses of independent and dependent to mirror the binary response system of the Katz index. Responses of 0 were regarded as independent in the binary scoring system, and all other responses were graded as dependent.
Subjects with complete data for all 6 items were included in this analysis. Rasch analysis was performed to determine the item difficulties. The nature of Rasch analysis is that it is sample independent.11,12 This means that it does not make any assumptions that a sample provides a replicate representation of a population, but it is still necessary that the sample consists of members of the population of interest, preferably with a sufficient range of abilities over the variable of interest. As such sample bias from overrepresentation of some groups within the population, underrepresentation of other groups, or bias due to missing data is inconsequential.
Rasch analysis is a 1-parameter item response theory model, which assumes that the probability of a respondent providing a positive answer to an item is the logistic function of the difference between that respondent's ability and the item's difficulty.13 Item difficulties are determined not only by the frequency with which an item gets a positive response but also by the ability levels of subjects who have a positive response. In this way, difficult items are those items that are generally given positive responses by those with high-ability levels and negative responses by those with low-ability levels.13 In this study, item parameters were estimated using conditional maximum likelihood estimation from person raw scores.14 Item fit parameters were also determined with computation of the Rasch model, which indicate how well individual items fit the Rasch model.
Once item difficulties were established with Rasch analysis, 2 grading systems were devised. The first grading system assigned a subject a grade based on the most difficult item that he or she could perform as the ability-focused grading system. The other grading system assigned a grade based on the least difficult item that a subject could not perform, as the disability-focused grading system. To test the validity of the item hierarchies, the proportions of individuals who could perform an ADL independently and all easier ADLs were calculated for each item. The validity of the item hierarchies was also tested by comparing the 2 different grading systems using Cohen's κ treating the 2 grading methods as distinct raters.
Data analysis was performed with the statistical package R version 2.1315 using the packages eRM16 for Rasch analysis and psy17 for other psychometric functions.
A total of 13 507 patients were included in the original survey data set. Of these patients, 13 113 had data available for all 6 ADL items (97% of the entire survey). In general, the sample represented older adults with a mean age of 80.6 years. The majority of subjects were also white (87.9%) and female (72%), though all ethnic groups were represented. Full descriptive statistics of the sample are as presented in Table 1.
The difficulty hierarchy of the 6 items of the Katz index was found to be in order of least to most difficult: eating, maintaining continence, transferring, toileting, dressing, and bathing. A person-item map showing the relative location of these items to each other is visible in Figure 1. This figure demonstrates the difficulty of each item along a latent trait, which in this case is independence with ADL. As shown in the top of the figure, there was a wide range of person subject abilities, with the largest ability group being those with a low level of functional independence. All 6 items adequately fit the Rasch model based on Mean-Square (MSQ) infit values, which ranged from 0.507 to 1.273. The internal consistency of the scale was moderately good with a Cronbach's α of .792. For all items more difficult than eating, the majority of patients who could perform each item could also perform all easier items. The item that performed least well in this regard was transferring, with only 56.7% of the patients able to do this task also able to perform all easier activities. The item that performed the best in this manner was maintaining continence; 72.4% of patients who could perform this item were also able to do all easier items. Values for item difficulty, infit, and percentage of those able to perform all easier tasks are as presented in Table 2. In this table, items are ordered by difficulty from easiest to most difficult. “Performs lower hierarchy items” gives the fraction defined by the number of subjects who could perform the task and all easier tasks independently divided by the total number of subjects able to perform the task independently. The κ coefficient was 0.73 for the 2 different grading systems, the ability-focused method and the disability-focused method.
This is the first study to look at a hierarchy of ADL difficulty in the SNF population and apply the findings to the creation of a grading system of function. This study found that such a hierarchy does exist and that the majority of patients fit the theoretical hierarchy. In addition, this study developed 2 grading systems: one that assigns a grade based on the most difficult task a patient can perform and one that assigns a grade based on the easiest task that one cannot perform. These 2 grading systems gave similar though not identical results. A number of psychometric properties of the Katz index in the nursing facility population were also determined, which indicate that this index has construct validity and internal consistency, though it could likely be further optimized to reduce redundancy. The value of the Katz index is that it has only 6 items each of which has only 2 possible responses. The information contained in this index can be obtained quickly even from brief discussions with nonrehabilitation staff involved in a patient's care. This makes it easy to use in all clinical settings, including those with a limited number of rehabilitation specialists where optimal use of a physical therapist's time is especially important. This information can also be used as a quick tool to monitor changes in functional status and to prioritize goals of therapy.
In general, the results of this study match prior results in terms of item difficulty with eating being the easiest task and bathing being the most difficult, which was expected. The hierarchy of difficulty found in this study exactly matches the item difficulty order demonstrated originally in the Katz index, which was studied in patients in the acute care setting, long-term care facilities, and at home.1 This hierarchy is also in accordance with another staging system developed in patients in the acute inpatient rehabilitation setting based on FIM that regards eating as one of the easiest activities and bathing as one of the most difficult in inpatient rehabilitation facility patients.4,18 Importantly, this study also suggests an expected pattern of functional decline or recovery. The order of loss of functional independence with ADL would be expected to follow the pattern from the most difficult ADL to the least difficult ADL, or conversely recovery would expect to show a pattern of recovery from easiest ADL to most difficult ADL. However, since this study did not examine longitudinal data it was not possible to test whether recovery or decline match the order expected based on item difficulty.
Both a grading system based on the most difficult ADL one can perform and a grading system based on the easiest ADL one cannot perform were developed in this study, and they gave similar, though not identical results. This provides support for the concept of a functional independence grade in the nursing population based on the Katz index, but some patients' grades will change depending on whether they are assigned an ability-focused or a disability-focused grade. It is unclear which grading method is superior, but the choice of whether to use an ability-focused grade or a disability-focused grade should likely be based on the decisions based on the grade. For example, in decisions regarding safety that typically demand a high level of caution, the disability-focused grading system would be the prudent. However, when thinking of how to optimize a patient's daily quality of life, it may be best to take an ability-focused approach.
Although the primary goal of this study was to examine how ADLs relate to one another with regard to difficulty in the nursing facility population, the use of Rasch analysis to accomplish this goal also provides valuable psychometric information regarding the Katz index in this population. In general, the existence of a meaningful order of item difficulties found using Rasch analysis indicates that a test has construct validity and in a sense defines the construct as the ordered hierarchy of items.19 Thus, the existence of a meaningful order of items found using Rasch analysis in this study provides evidence in support of the construct validity of the Katz index in the nursing facility population.
The fit of individual items in the Katz index was also determined by calculating item fit statistics using MSQ infit values. All 6 items appear to fit the Rasch model adequately well for inclusion in the index. MSQ fit values of 1.0 represent perfect model fit, whereas those greater than 1 represent underfit, and those less than 1 represent overfit. Underfit implies that there is more stochasticity than would be expected by the Rasch model, and overfit implies that there is an overdetermination in the index.20 Acceptable ranges of fit values are a matter of judgment and context, but in general MSQ fit values in the range 0.5 to 1.5 are the most productive for measurement,21 which indicates that none of the items degrade measurement. The items that most overfit the model were bathing, dressing, and toileting, which have infit values of 0.578, 0.652, and 0.507, respectively, which suggests that there is some redundancy in the Katz index. Bathing is a powerful discriminating item, because it is so far removed in difficulty from the other items, which would suggest that it serves value in the index and is therefore not redundant despite the low infit value. Toileting and dressing are so close in difficulty that one of these items could likely be removed without a significant decrease in the index's discriminative ability. However, the continued inclusion of all 6 items is unlikely to compromise the value of the measure obtained from the index, so the main value of removing an item would be a reduction in the data collection effort. The results of this analysis suggest a number of future lines of research, most notably the use of an adaptation of the Katz index or a similar metric using ordinal responses. The addition of ordinal responses might make an index somewhat more complex to consistently administer, but it would be likely to increase the discriminative ability, especially for items with very close difficulties on a binary index. It would be interesting to understand the hierarchy of levels of dependence in a scale based on ordinal responses.
This study had a number of limitations, which are important to recognize. The most notable limitation is that a functional measure was studied for a large group of patients without regard for underlying diagnoses. Thus, patients with strokes, fractures, heart failure, and dementia were analyzed together. Unlike acute inpatient rehabilitation facilities, there is no information about the chief disabling condition for which a subject was admitted to the nursing facility. This does not invalidate the results found here, but it does indicate that these results could be further refined using another data set. In addition, this is cross-sectional data, so it is not possible to relate any of the findings in this study to subsequent outcomes, such as mortality.
This study provides results that suggest a number of new lines of research. Chiefly, it would be important to know whether SNF patients decline or recover in function in the order that would be suggested by the ADL difficulty level. If this is the case, then this ADL hierarchy could be used to guide rehabilitation plans for patients. In addition, it would be important to know whether the disability-focused grading system or the ability-focused grading system is of greater value with regard to patient safety, burden of care, and quality of life. Finally, this study suggests some ways in which the Katz index could be optimized for efficiency of administration by reducing the number of items. This provides a starting point for the development of future functional measures in this population.
Among older patients residing in nursing facilities, we found a hierarchy of ADL difficulty, which from easiest to most difficult are as follows: eating, maintaining continence, transferring, toileting, dressing, and bathing. This hierarchy can be used as the basis for a grading system of functional independence in the nursing facility patients. In addition, the Katz index, which is composed of these 6 ADLs, demonstrates construct validity with moderately good internal consistency in this population.
1. Katz S, Ford AB, Moskowitz RW, Jackson BA, Jaffe MW. Studies of Illness in the Aged: The Index of ADL: A Standardized Measure of Biological and Psychosocial Function. JAMA. 1963; 185:(12):914–919.
2. Jette DU, Warren RL, Wirtalla C. Validity of functional independence staging in patients receiving rehabilitation in skilled nursing facilities. Arch Phys Med Rehabil. 2005; 86:(6):1095–1101.
3. Stineman MG, Ross RN, Fiedler R, Granger CV, Maislin G. Staging functional independence validity and applications. Arch Phys Med Rehabil. 2003; 84:(1):38–45.
4. Stineman MG, Ross RN, Fiedler R, Granger CV, Maislin G. Functional independence staging: conceptual foundation, face validity, and empirical derivation. Arch Phys Med Rehabil. 2003; 84:(1):29–37.
5. Schussler-Fiorenza CM, Xie D, Stineman MG. Complex versus simple disability staging: New tools for physiatry. Paper presented at: Association of Academic Physiatrists Annual Meeting, Las Vegas, NV; 2012; .
7. Aguero-Torres H, Fratiglioni L, Guo Z, Viitanen M, von Strauss E, Winblad B. Dementia is the major cause of functional dependence in the elderly: 3-year follow-up data from a population-based study. Am J Public Health. 1998; 88:(10):1452–1456.
8. Branch LG, Katz S, Kniepmann K, Papsidero JA. A prospective study of functional status among community elders. Am J Public Health. 1984; 74:(3):266–268.
9. Katz S, Branch LG, Branson MH, Papsidero JA, Beck JC, Greer DS. Active life expectancy. N Engl J Med. 1983; 309:(20):1218–1224.
10. Reijneveld SA, Spijker J, Dijkshoorn H. Katz' ADL index assessed functional performance of Turkish, Moroccan, and Dutch elderly. J Clin Epidemiol. 2007; 60:(4):382–388.
11. Mair P, Hatzinger R. New developments for extended Rasch modeling in R. Paper presented at: useR! Gaithersburg, MD; 2010; .
12. Wright BD, Stone MA. Rasch model derived from objectivity. Rasch Meas Transact. 1987; 1:(1):5–6.
13. Wright BD, Mok M. Rasch models overview. J Appl Meas. 2000; 1:(1):83–106.
14. Mair P, Hatzinger R. Extended Rasch modeling: the eRm package for the application of IRT models in R. J Stat Softw. 2007; 20:(9):1–20.
15. R: A language and environment for statistical computing. [computer program]. Version 2.13. Vienna, Austria: R Foundation for Statistical Computing; 2011; .
16. eRm: Extended Rasch Modeling. [computer program]. Version R package version 0.14-0; 2011; .
17. psy: Various procedures used in psychometry. [computer program]. Version R package version 1.0; 2009; .
18. Stineman MG, Ross RN, Granger CV, Maislin G. Predicting the achievement of 6 grades of physical independence from data routinely collected at admission to rehabilitation. Arch Phys Med Rehabil. 2003; 84:(11):1647–1656.
19. Wright B, Stone M. Validity. In:Wright B, Stone M., eds. Measurement Essentials. 2nd ed. Wilmington, DE: Wide Range; 1999; :167–172.
20. Walter S. Conceptualizing overfit or over-parameterization. Rasch Meas Transact. 2008; 22:(2):1165.
21. Wright BD, Linacre JM. Reasonable mean-square fit values. Rasch Meas Transact. 1994; 8:(3):370.
© 2013 The Section on Geriatrics of the American Physical Therapy Association.