| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| ||||||||||||||||||||||||||||||||
1 Gerontological Research Programme, Faculty of Medicine, and2
Department of Psychological Medicine, National University of Singapore.
3 Institute of Mental Health, Ministry of Health, Republic of Singapore.
Address correspondence to Tze-Pin Ng, MD, Gerontological Research Programme, National University of Singapore, Department of Psychological Medicine, National University Hospital, 5 Lower Kent Ridge Road, Singapore 119074. E-mail: pcmngtp{at}nus.edu.sg
 |
A |
|---|
 Top Abstract MethodsResultsDiscussionReferences |
|---|
Methods. Using data from a random population sample of noninstitutionalized Chinese, Malay, and Indian older adults 60 years old and older in Singapore (N = 1072), we modeled the dimensional structure of the 8-item IADL Scale using exploratory and confirmatory factor analyses, and assessed its convergent and divergent validity using known group differences and strengths of association.
Results. Factor analyses yielded two strong and reliable factors representing underlying physical and cognitive dimensions of IADL. The validity of the model was supported by the pattern of associations of the IADL with age, gender, education, self-reported health status, hospitalization, physical comorbidities, dementia and depression, and Mini-Mental State Examination (MMSE) scores. Notably, cognitive IADL showed a greater total effect on MMSE cognitive performance score than did physical IADL, with the effect of physical IADL on MMSE score mostly explained by cognitive IADL. Reasonably good cross-cultural validity was demonstrated among Chinese, Malays, and Indians, with strongest validity for Indians.
Conclusion. The eight-item IADL Scale has physical and cognitive domains and is cross-culturally applicable. The cognitive IADL domain taps a set of activities directly related to cognitive functioning.
Whether functional disability is a unidimensional or multidimensional construct is controversial (2,4). As proposed by a number of authors (5–7), IADL tasks that include getting to places outside the house, grocery shopping, preparing meals, doing housework or handyman chores, and doing laundry could be categorized as those relying on physical health or strength, and thus could be conceptualized as "physical IADLs." In contrast, tasks such as using the telephone, taking medications, and managing finances may be conceptualized as those requiring complex cognitive resources and thus are termed "cognitive IADLs." Disability in these two aspects (physical and cognitive) of IADL is postulated to be associated with different antecedent conditions and subsequent outcomes (8).
The validity of such a two-dimensional structure for the IADL scale (2,6) has, however, been evaluated in very few studies. The objectives of this study were to model the internal structure of the eight-item IADL scale and to evaluate its convergent and divergent validity with specific correlates of physical and cognitive disability, as well as its cross-cultural validity across ethnic groups in an Asian multiethnic population of community-living older adults.
|
METHODS |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
Questionnaire Data
IADLs were assessed by using a modification of the original Lawton and Brody IADL scale (3). The participant's level of dependence in performing eight IADLs (using the telephone, getting about, grocery shopping, preparing meals, doing housework, doing laundry, taking medicines, and managing money) within 1 or 2 days before the interview was reported. The items refer to what respondents were able to do (not to their actual performance) on other days if they did not engage in the said IADL activities within the reference days. Respondents indicated whether they: were unable to do at all [0], needed some assistance [1], or needed no help [2]. Only those receiving assistance from a person (as opposed to a device) were considered to be receiving help.
Sociodemographic data were collected for age, gender, ethnicity, marital status, education, employment status, housing type, and living arrangement. Among the ethnic groups, it is well documented that the Chinese have the highest levels of socioeconomic status and generally better health status, whereas Malays have the lowest (9–11). Housing type and size has been shown consistently in numerous studies to be a reliable surrogate indicator of socioeconomic status, with persons living in lower-end, small-sized (1–2 rooms) public housing apartments having lower average income and education than others living in higher-end housing types and sizes (9–11).
Medical conditions.-- Participants were asked to report the presence in the 12 months prior to the interview of any of a list of 16 specific medical conditions, which included coronary artery disease, heart failure, hypertension, dyslipidemia, diabetes, stroke, cancer, hip fracture, arthritis, asthma, chronic obstructive pulmonary disease (COPD), cataract, and other conditions. For every medical condition reported by the respondent, information on the frequency of hospitalization in the same 12-month period was also collected.
Self-rated health status was assessed by a single question ("In general, would you say your health is excellent, very good, good, fair, or poor?"), which is well-documented in many studies to predict health outcomes (12–14). Self-rated health status was analyzed as a dichotomized variable ("poor and fair" vs "good, very good, and excellent").
Psychiatric morbidity.-- The diagnoses of dementia and depression were made by using the Geriatric Mental State (GMS), a standardized, semistructured clinical interview designed to assess psychopathology for older adults (15). The 25- to 40-minute clinical interview is generated from a computerized algorithm, Automated Geriatric Examination for Computer Assisted Taxonomy (AGECAT) (16), nine diagnostic categories that included organic brain disorder (dementia) and depression. A diagnostic confidence level is provided for each syndrome, ranging from 0 (no symptoms) to 5 (very severely affected). Level 3 and higher represent likely cases, consistent with Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM IV) and International Classification of Diseases, 10th revision (ICD-10) diagnoses; levels 1 and 2 are considered to be subcases. The GMS is widely used internationally, and its validity has also been reported for use in developing countries (17). Psychiatric and community nurses and field interviewers underwent extensive training in the use of the GMS by certified training psychiatrists in the Institute of Mental Health, using training vignettes; completed and co-rated four to six supervised training interviews; and received further detailed supervision and close monitoring of their field interviews.
Cognitive functioning was assessed by the using Mini-Mental State Examination (MMSE) (18), which measures performance in domains that include memory, attention, language, praxis, and visuospatial ability. As a measure of global cognitive functioning, the MMSE has been shown to correlate well with other scales of cognitive functioning, such as the Cognitive Abilities Screening Instrument (CASI), the Hasegawa Dementia Screening Scale (HDSS), and the Blessed Memory-Information Concentration test (MIC Blessed), and to discriminate well between demented and nondemented participants in many studies (19). The Chinese version of the MMSE has also been demonstrated to have good test performance as a screening instrument for dementia in Shanghai (20,21) and in Singapore (22). The Malay version of the MMSE was developed by forward and back translation from the English. The item in English "no ifs, ands or buts" was replaced in the Chinese MMSE by "forty-four stone lions," and in the Malay version by "marah, merah, murah." The total score on the MMSE can range from 0 to 30, with higher values denoting better cognitive functioning. The assumption of unidimensionality of the MMSE scale supports the use of the MMSE as a summative score (23).
Statistical Analysis
We used exploratory (EFA) and confirmatory (CFA) factor analyses to assess the dimensional structure of the IADL. EFA took into account the ordered categorical measurement of the IADL data, by using an input matrix of polychoric correlations of the eight IADL items, based on a three-stage "underlying variable" approach (24–26). Each observed ordinal variable is assumed to be underlain by an unobserved continuous variable that is marginally normally distributed; it is also assumed that corresponding to any pair of observed ordinal variable is a pair of unobserved continuous variables that are jointly bivariate normally distributed. Estimation of the factor model then proceeded in three stages: (i) first-order statistics such as thresholds, means, and variances were estimated by maximum likelihood (ML); (ii) second-order statistics such as polychoric correlations were estimated by conditional ML, given the estimates from the first stage; and finally (iii) the parameters of the structural part of the factor model were estimated using a generalized least squares approach with the polychoric correlation matrix estimated in the second stage (25,26).
The correlation matrix was factor analyzed with varimax rotation to obtain a clearer resolution of the dimensional structure of IADL. Two factors were then extracted based on the hypothesized two-dimensional structure of the scale. The interpretation of the factors was guided by the factor loading, expressing the relationships between the manifest variables and the hypothetical factors. The meaning of a factor was given by the variables with large loadings of at least 0.50.
CFA was performed using an asymptotic distribution free estimation method with the bootstrapping technique recommended for data that are not normally distributed to improve the standard error, hence better precision of the parameter estimates (27). The measurement models used were assumed to be recursive or unidirectional, with neither feedback loops nor reciprocal effects. In the first-order model, both latent factors (physical and cognitive IADL) were allowed to correlate to each other ("oblique assumption"). The second-order model assumes that common variance between the two latent factors (physical and cognitive IADL) would be fully explained or represented by the common (or second-order) factor (named "IADL"; see Appendix 1).
|
Convergent and divergent validity were evaluated by analyzing known group differences and strength of anticipated associations of physical and cognitive IADL disability with pertinent psychosocial and health variables. We hypothesized that more IADL dependence would be associated with older age, female gender, non-Chinese ethnicity, and generally lower socioeconomic status (as is widely reported in the research literature); cognitive IADL disability (more than physical IADL disability) would show greater strengths of association with health states and conditions known to cause or be associated with cognitive deficits, such as low MMSE scores, dementia, and depression. Conversely, physical IADL disability (much more than cognitive IADL disability) was expected to show greater strengths of association with conditions such as arthritis and hip fracture, whereas other conditions such as diabetes and stroke are expected to be associated with both physical and IADL disability. Despite the moderately skewed distribution, it was reasonable to uphold the assumption of normality in analyses of variance. EFA was done using SAS (version 9.1; SAS Institute Inc., Cary, NC), and CFA was done using AMOS (version 5; Assessment System Corporation, St. Paul, MN). Other statistical analyses were performed using SPSS (version 13.0; SPSS Inc., Chicago, IL).
|
RESULTS |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
Table 1 shows individual IADL items, listed in descending order of the mean score of dependence, with higher mean scores denoting better functioning levels. The respondents were least frequently dependent in cognitive IADL items such as using the telephone, taking medications, and managing money, thus suggesting a scale-wise hierarchical relationship between physical and cognitive items (Table 2). All values of IADL items showed skewed distribution (skewness values ranging from –1.67 to –4.92). Item-wise, participants were most highly dependent in three tasks—doing laundry, getting to places outside the house, and grocery shopping—with prevalence rates ranging between 20% and 25%. Overall prevalence for cognitive IADL dependence (11.1%) was lower than that for physical IADL dependence (36.4%).
|
|
Two distinct factors, as hypothesized, were extracted from the EFA (Table 3), which were readily labeled "cognitive IADL," comprising three items (using the telephone, taking medications, and managing money), and "physical IADL," comprising the remaining five items (grocery shopping, getting to places outside the house, doing housework/handyman work, doing laundry, and preparing meals). The total variance explained by the two scales was 87.5%. The Cronbach's alpha reliability coefficient, which measures the internal consistency of the collection of items for each scale, was 0.91 (95% confidence interval, 0.90–0.92) for physical IADL and 0.78 (95% confidence interval, 0.76–0.80) for cognitive IADL.
|
EFA by Ethnicity
To assess whether the two-factor solution was replicable across ethnic groups, we estimated separate factor analytical models for each ethnic group. Two latent factors were generated for each ethnic group with total variances of 86.2%, 90.4%, and 87.6% in Chinese, Malays, and Indians, respectively. The first and dominant factor consistent across ethnicities was the physical IADL. The patterns of factor loadings in all ethnic groups were similar to that in the whole sample of participants. Three items (getting to places outside the house, grocery shopping, and preparing meals) were almost equally loaded on more than one factor (with factor loadings >0.50). For each of these instances, the item was assigned to the factor on which it had the highest loading. The reliability coefficients were uniformly high for the two latent factors across ethnicities, ranging from 0.74 to 0.93.
CFA, which assumed the two latent factors (physical and cognitive IADL) to be correlated, was performed with the asymptotic distribution-free method as the parameter estimation method (Table 4). The results showed that the two-factor model had relatively more satisfactory fit statistics (in particular, the relative fit statistics) than did the unidimensional model, in the full sample as well as in the subsamples. In addition, a second-order two-factor model assuming that the covariation between the two latent factors (physical and cognitive IADL) would be fully explained by their regression on the second-order latent factor (IADL) showed goodness-of-fit statistics similar to those of the first-order two-factor model. For reason of parsimony, only the results of the first-order two-factor model are shown in Table 4. This measurement model was good with the lowest factor loadings of 0.52 and 0.72 for "using the telephone" and "getting to places outside the house" items, respectively.
|
|
|
|
Whereas the associations of socioeconomic variables with IADL dependence were nonselective for either physical or cognitive IADL, differential associations of cognitive and physical IADL dependence were observed with known health correlates, mostly but not all in the expected directions (Table 7). As expected, selected chronic conditions such as hip fracture and arthritis were associated only with dependence in physical IADLs, and not with cognitive IADL dependence. Conditions such as diabetes and stroke, as expected, showed associations with both physical and cognitive IADL dependence, with relatively stronger associations with physical IADL dependence. Few conditions were associated selectively with dependence on cognitive IADLs only; these conditions included those that would not be expected (kidney failure, urinary problems). Depression, dementia, and cognitive impairment were associated with both cognitive and physical IADL dependence, but the associations were stronger with cognitive IADL dependence.
Eta-squared statistics, indicating the effect sizes for individual chronic conditions, were uniformly low (<4%), suggesting that little of the variability in both physical and cognitive IADL scores can be explained by any single condition. Higher eta-squared values were observed for self-reported health status in association with physical IADL dependence (12%) and cognitive IADL dependence (8%). It was interesting also to note that the highest eta-squared values were observed in association with cognitive IADL dependence for MMSE cognitive impairment (23%) and for dementia (19%). The corresponding eta-squared values in association with physical IADL dependence were 20% for MMSE cognitive impairment and 13% for dementia.
As the F statistics and eta-squared values in Table 7 were generated from separate analysis of variance models, we next used structural equation modeling to regress MMSE score (continuous exogenous variable) on the latent factors (physical and cognitive IADL). In this approach, both scales were included in a single model to concurrently predict cognitive impairment (as indicated by MMSE). As shown in Figure 1, the standardized regression coefficient of the cognitive IADL scale was higher than that of the physical IADL scale (0.35 vs 0.20), whereas the zero-order correlation coefficients were 0.55 and 0.49, respectively. The latent mean structure analysis testing the equality of the regression coefficients showed a significant difference (
X2 = 16, df = 1). This result indicates that cognitive IADL predicts MMSE cognitive function better than physical IADL does, as the effect of physical IADL on MMSE score was mostly explained by cognitive IADL. This finding provides considerable support for the postulate that cognitive IADL taps a set of activities more directly related to cognitive functioning than the physical IADL does.
|
|
DISCUSSION |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
Recent work by several researchers have constructed scales that combined basic ADL with IADL items (29,30) and have uncovered a three-dimensional structure in ADL/IADL items (2,6,31,32). Some authors have also argued that an assumption of unidimensionality and a hierarchical relationship between basic and instrumental ADL is debatable (2). Other researchers have contended that Katz's ADL scale and Lawton and Brody's IADL scale are separate constructs and should be modeled separately (33,34). The issue is as yet unresolved. We have taken the latter stance in our analysis. Nevertheless, it is interesting to note that, in the proposed three-dimensional ADL/IADL structure, barring one dimension that describe basic self-care, the two dimensions derived from IADL items (2,6,31,32) were remarkably similar to the "physical" and "cognitive" IADL scale in this study.
Some issues in EFA and CFA require further brief discussion. The sample size was sufficiently large to compensate for the non-normality of the data. The arbitrary use of ordered categorical response in the IADL scale using the conventional Pearson correlation matrix is problematic and could give misleading results on the underlying factor structure. A recommended approach is to use a three-stage "underlying variable" approach (24–26), assuming that each observed ordinal variable is underlain by an unobserved continuous variable that is marginally normally distributed. In situations where CFAs do not provide strong support for the factor solutions generated by EFAs, the conventional wisdom is to rely on the consistency of results from EFAs. We were able to observe good consistency in the factor solutions of EFA in split samples and ethnic subsamples, and this consistency supports the hypothesis that the IADL scale is one that has two conceptually and statistically distinct dimensions (31).
Our findings are limited by the relatively young age of the sample, hence the frequencies of "cognitive" IADL dependence and cognitive impairment were relatively low. It may be noted that although 29.6% of the sample had cognitive impairment (MMSE 
23), 11% reported one or more cognitive IADL difficulties. Because IADL difficulty is a prerequisite for the diagnosis of dementia, we examined the correspondence between cognitive impairment and cognitive IADL difficulty among the participants with dementia. The proportion with cognitive impairment (MMSE 23) was 97.5%, and the proportion with difficulty on at least one IADL task was 85% (76% for physical IADL and 58% for cognitive IADL). As the diagnosis of dementia incorporating the prerequisite for functional disability was made by GMS interview independently of the assessment of MMSE and of IADL disability, some discordance is expected. It should also be noted that, although the diagnostic validity of GMS/AGECAT for diagnosing organic disorder (dementia) is well established in developed countries, it is less discriminating in populations with low levels of education (17). Although the proportion of participants with cognitive IADL difficulty appears lower than that of participants with physical IADL difficulty, it should be noted that there is a smaller number of items in cognitive IADL (3) than in physical IADL (3). As shown in Table 7, the difference in proportions of cognitive IADL difficulty between demented and nondemented participants (58.2% vs 7.4%, 8-fold difference) was much greater than the difference in proportion of physical IADL difficulty between demented and nondemented participants (75.9% vs 33.3%, 2-fold difference). Although this finding indicates that the cognitive IADL scale had greater ability than did the physical IADL scale to discriminate dementia from nondementia cases, it should be emphasized that, in itself, cognitive IADL difficulty in the Lawton scale still lacks sufficient sensitivity to be used alone to screen for dementia.
Another limitation to the study is found in the fact that, whereas 29.6% of the sample were determined to be cognitively impaired (MMSE 23), a smaller proportion (13%) of the responses in the sample was given by caregivers. Whereas all participants with dementia (7.4% of the sample) provided informant reports, a considerable proportion of cognitively impaired participants appeared to give self-reports, thus raising concerns about the reliability of the data for these participants. The standard procedure in the survey was to have the participants themselves provide information unless the caregiver or the interviewer determined that the participant was unable to provide reliable information. It is difficult to be certain about the bias in such self-reported data and the possible bias on the results. Misclassification errors in cognitively impaired participants may tend toward under-reporting of IADL dependence and other health conditions. If this is so, it is likely to bias the associations of IADL disability with cognitive impairment toward the null, and to be nondifferential for health and other variables.
The cross-sectional nature of the data calls for some caution in interpretation of the results. Although the concurrent validity of the cognitive and physical IADL domains is strongly supported, their predictive validities require results from further longitudinal studies. In particular, future longitudinal studies should investigate whether the two subscales predict differentially the need for home-based or institutionalized care of elderly persons. On another note, the use of a simple aggregate composite index that assumes equal weighting of tasks without differentiating the contribution of specific tasks to cumulative disability is an assumption held in this analysis that may possibly limit the discriminative performance of the scales. The validity of this assumption may also be investigated in future studies. This study confirms physical and cognitive domains of the eight-item IADL that is cross-culturally applicable, and identifies cognitive IADL as a dimension that taps a set of activities directly related to cognitive functioning.
|
A |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
Tze-Pin Ng conceptualized the study, formulated the hypothesis, designed the epidemiological survey, and drafted the manuscript. Mathew Niti performed the literature review, analyzed the data, and participated in the drafting of the manuscript. Peak-Chiang Chiang was the lead investigator and trainer of the psychiatric morbidity survey and participated in the conceptualization of the study, the interpretation of the results, and drafting of the manuscript. Kua Ee Heok participated in the conceptualization of the study, the interpretation of the results, and drafting of the manuscript.
No commercial company sponsored or played any role in the design, methods, participant recruitment, data collections, analysis, or preparation of the article.
|
F |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
Received March 24, 2005
Accepted January 10, 2006
|
References |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
This article has been cited by other articles:
 |
|
 |
|
  S A M Sikkes, E S M de Lange-de Klerk, Y A L Pijnenburg, P Scheltens, and B M J Uitdehaag A systematic review of Instrumental Activities of Daily Living scales in dementia: room for improvement J. Neurol. Neurosurg. Psychiatry, January 1, 2009; 80(1): 7 - 12. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 A. Chan and S. Jatrana Gender Differences in Health among Older Singaporeans International Sociology, July 1, 2007; 22(4): 463 - 491. [Abstract] [PDF]
|
|
| ||||||||||||||||||||||||||||||||
| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
