| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| ||||||||||||||||||||||||||||||||
Correspondence: Address correspondence to Nan Sook Park, PhD, School of Social Work, The University of Alabama, Box 870314, Tuscaloosa, AL 35487-0314. E-mail: npark{at}bama.ua.edu
 |
Abstract |
|---|
 TOP Abstract MethodsResultsDiscussionReferences |
|---|
Key Words: Typology • Residential care • Assisted living • Cluster analysis • Long-term care
Developing a typology is especially useful when a group is heterogeneous and when classification systems have not been established (Everitt, Landau, & Leese, 2001), which is certainly the case with RC/AL. Typology development involves a process in which the group is classified into meaningful subgroups that share similar characteristics. In fact, typologies have already been used to identify common patterns in long-term care, allowing researchers to describe and understand the configurations of settings that commonly exist for older adults (Lawton, 2001; Timko & Moos, 1991). The majority of these studies, however, are limited to classifying nursing homes, particularly dementia care units (Davis et al., 2000; Gold, Sloane, Mathew, Bledsoe, & Konanc, 1991; Grant, 1998; Grant, Kane, & Stark, 1995; Holmes & Teresi, 1994; Sloane, Lindeman, Phillips, Moritz, & Koch, 1995). A few studies have addressed RC/AL settings (e.g., Hawes, Phillips, & Rose, 2000; Timko & Moos), but none have studied the entire range of RC/AL across multiple states.
In developing and describing typologies, researchers have commonly employed the structure–process–outcome framework (Donabedian, 1978, 1980, 1988). In this framework, structure refers to the capacity of the facility to provide care, which includes physical amenities (e.g., safety features) and human resources (e.g., the staff-to-resident ratio). Process is how the facility delivers care, which encompasses the activities that occur between care providers and residents (e.g., the provision of organized activity programs). Outcome is the effect of care received. Davis and colleagues (2000) evaluated specialized dementia programs in RC/AL settings based on two structure-related variables: facility size (i.e., small vs large) and administrative relationships with other facilities (i.e., affiliation with other RC/AL facilities); they derived five types. Hawes and colleagues (2000) classified a national sample of 1,251 larger RC/AL facilities based on reported consumer preferences for two process-related characteristics (i.e., privacy and service); they identified four types. Although both of these studies were able to parsimoniously present similarities and differences across clusters, they were limited by a focus on few variables to differentiate the types of care settings.
One component of care that both studies overlooked was resident case-mix, which is likely to be closely related to the structure of care (Wunderlich, Sloan, & Davis, 1996). For instance, facilities consisting of high proportions of residents who are functionally dependent may require more resources and staff than those caring for less impaired individuals. Indeed, Timko and Moos (1991) examined the configurations of different facilities for older adults from a national sample of 235 nursing homes, RC/AL facilities, and congregate-care apartments. They found that both facility characteristics (e.g., level of care, ownership, and size) and resident case-mix (e.g., social resources, functional ability, and gender) contributed to the social climate of facilities. They contended, in essence, that resident case-mix gauges the overall demands on the staff, which in turn affects facility dynamics.
As RC/AL prospered under the recent building boom (i.e., it has accounted for more than 80% of new projects in the senior-housing industry), the heterogeneity in the field has increased (Adler, 1998). Consequently, as recognized by a national task force, it has become increasingly difficult to even agree on a definition of RC/AL (Assisted Living Workgroup, 2003). Being able to delineate the subtypes of facilities that exist should facilitate understanding, discussion, research, policy, and practice. Therefore, this article derives typologies of RC/AL by using criteria from structure, process, and resident case-mix domains. In doing so, it uses a more comprehensive set of variables than has been applied in the past. We used mixture modeling, a special case of cluster analysis developed in the latent variable framework, in order to classify facilities, and we evaluated resultant typologies based on statistical, theoretical, and practical significance.
|
Methods |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
The CS–LTC enrolled 2,078 residents in 193 facilities from representative regions across four states: Florida, Maryland, New Jersey, and North Carolina. Baseline data were collected from October 1997 to November 1998. After consultation with national experts and a review of state regulations and current research, the CS–LTC selected these states in consideration of geographical proximity (to enable onsite data collection) and variability (their RC/AL policies reflected the diversity of the field). A purposive sample of counties (sampling regions) was selected within each state based on three criteria: (a) at least 15% of each type of the state's RC/AL facilities were in the region; (b) rural and urban diversity was reflected in terms of demographic and health-service characteristics; and (c) the region represented the state.
To ensure that all facility types meeting the definition of RC/AL were included, the CS–LTC stratified facilities into three types: facilities with fewer than 16 beds, traditional "board-and-care" facilities with 16 or more beds, and "new-model" facilities with 16 or more beds. New-model facilities with 16 or more beds were defined through a pilot study as having been built after January 1, 1987, and having either two or more different private-pay monthly rates, 20% or more residents requiring assistance in transfer, 25% or more residents who are incontinent daily, or either a registered nurse or licensed practical nurse on duty at all times (Zimmerman et al., 2003). A total of 113 smaller facilities and 40 of each of the other facility types were included in the CS–LTC. Administrators of the 193 facilities were interviewed onsite to gather facility-level information. Detailed information about CS–LTC methods can be obtained elsewhere (Zimmerman et al., 2001).
Variables Used in Typology Development
The CS–LTC offered rich information with which to develop typologies. We organized the variables under study according to the domains of structure, process, and resident case-mix. Data were not available to incorporate outcomes in the analyses. Table 1 presents definitions of the facility-level variables used in these analyses.
|
We evaluated process of care by using eight scales from the Policy and Program Information Form (POLIF) of the Multiphasic Environmental Assessment Procedure (Moos & Lemke, 1996). POLIF has been used in other studies, which have found it to discriminate between different long-term care settings and the types of residents they serve (Zimmerman et al., 2001). The eight POLIF scales included in these analyses evaluated admission policies, acceptance of problem behaviors, policy choice, policy clarity, provision of privacy, resident control, overall provision of services, and availability of social and recreational activities. The range of Cronbach's alphas for the subscales was.69–.84, showing moderately high internal consistency of the items (Moos & Lemke; Zimmerman et al., 2003). We scored each POLIF scale as a range from 0%–100%, with higher percentages indicating endorsement of more items. The analyses in this article excluded two measures (provision of health services and admission policies specific to ADL functioning), which were subsets of other measures, and so were highly correlated.
We used six variables describing a range of resident characteristics and case-mix in typology development: percentage of residents on Medicaid; percentage of residents who required assistance taking care of their own appearance; percentage of residents who needed help getting in and out of bed; percentage of residents who had been diagnosed with dementia; percentage of residents who had a diagnosis of mental or psychiatric illness; and percentage of residents who had behavioral symptoms. An administrator for the whole facility reported these variables (i.e., we did not derive them from resident-level data).
Analyses
We first examined the distribution of individual variables (e.g., mean, standard deviation) and patterns of missing values. Next, we conducted mixture cluster analyses (Muthén & Muthén, 1998–2001) separately for structure (S), process (P), and resident case-mix (R). Subsequently, we used a combination of structure and process variables as the criterion variables (SP), and, lastly, we ran a combination of structure, process, and resident case-mix (SPR) for the final model. We examined the configurations of clusters (i.e., which clusters were comparable to others and which variables were driving the differences) and fit indices at each step.
Mixture modeling (i.e., mixture cluster analysis), a special case of cluster analysis, presumes that "unobserved heterogeneity" in the sample explains variability among observed variables (Muthén, 2001; Muthén & Muthén, 1998–2001). The heterogeneity is assumed to be latent or unobserved, hence to be inferred from the data. Applying this assumption to the current study, we can say that the diversity of RC/AL is due to the unobserved heterogeneity of the settings that may be explained through the chosen criterion variables from structure, process, and resident case-mix domains.
The assumptions and procedures of cluster analysis resemble exploratory factor analysis in that both approaches can be used as data-reduction techniques. The notable difference is that factor analysis takes a variable-centered approach, whereas cluster analysis takes a case-centered approach (Muthén & Muthén, 2000). Mixture cluster analysis assumes that latent variables are categorical, whereas observed variables can be either categorical or continuous; this assumption is suited to the current study. Mixture cluster analysis is also versatile in terms of handling missing data, whereas missing values are problematic in traditional cluster analysis. That is, mixture modeling using the Mplus program uses all observations available in the data and estimates the parameters through a maximum likelihood estimator (Muthén & Muthén, 1998–2001). In the present analyses, we determined the best-cluster solutions through the largest loglikelihood values using expectation-maximization algorithms, highest entropy (i.e., an index of classification quality), and lowest Bayesian Information Criterion (BIC) values (Muthén, 2001; Muthén & Muthén, 1998–2001; Schwartz, 1978), as well as high posterior probabilities (i.e., probabilities for cases to be in their respective class; Muthén & Muthén, 2000).
Once we had classified the facilities into clusters based on each domain of criterion variables (S, P, or R) and on a combination of the domains (SP or SPR), we used chi-square tests and one-way analysis of variance (ANOVA) in order to examine associations between cluster types by facility and resident characteristics. In ANOVA, we made additional tests to compare groups on a post-hoc basis. We conducted paired comparisons by using Fisher's least significant difference procedure (where F tests were significant at 
=.05).
Finally, we examined patterns of cluster membership across the resultant five typologies (e.g., whether facilities belonging to the same cluster in the SP typology stayed together in the SPR typology). We calculated kappa statistics among different combinations of the five typologies, reflecting the degree to which different observers or, in this case, methods, classify a particular case in the identical category (Stokes, Davis, & Koch, 2000). We performed mixture cluster analyses by using Mplus 2.1 (Muthén & Muthén, 1998–2001), chi-square tests and ANOVA by using SPSS 11 (SPSS, 1991–2000), and kappa statistics by using SAS 8.2 (SAS Institute, 1999–2001).
|
Results |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
|
Facilities tended to have few restrictions for admitting residents; however, some were reluctant to admit those with behavior problems. Generally, facilities scored moderately on policies involving individual freedom, institutional order, and social/recreational activities. With respect to resident case-mix, approximately 15% of residents across facilities were on Medicaid. More residents required help in taking care of themselves than they did transferring (47% vs 16%, respectively). A small proportion of residents had a history of mental illness (12%) or required attention because of behavioral symptoms (9%). Administrators reported that, on average, one third of their residents had dementia.
Typologies
We derived typologies by using five different combinations of criterion variables based on structure (S); process (P); resident case-mix (R); structure and process (SP); and structure, process, and resident case-mix (SPR). Although each approach derived distinct clusters of RC/AL depending on the criterion variables, the S, P, and SP models proved of limited utility because their results were dominated by facility size. Therefore, we present here only the R and SPR models, because they provide more unique and comprehensive information than the S, P, and SP models. Further, the kappa value between the R typology and SPR typology was 0.26, suggesting slight agreement (p <.01) and the utility of displaying both.
Cluster Solutions Based on Resident Case-Mix Characteristics (R)
Table 3 shows the means and standard deviations of the variables in the five clusters that were derived from the six facility-level resident characteristics, as well as the results of ANOVA or chi-square tests for each variable. Fit measures of mixture modeling favored a five-cluster solution in that values of loglikelihood, BIC, entropy, and average posterior probabilities indicated reasonably good fit compared with other cluster solutions (log likelihood = –4,633; BIC = 9,475; entropy = 0.87). The five clusters were distinguishable by their criterion variables (i.e., resident case-mix); accordingly, all ANOVA tests for the criterion variables were highly significant at p <.05. Four variables in the structure domain (i.e., years in business, nursing, environment, and privacy) were differentiated by this clustering as well, as were two variables in the process domain (i.e., overall admission policies and privacy). We describe each cluster briefly here:
|
Cluster R-2 (25 facilities) was characterized by high proportions of physically impaired residents (53% requiring help in transfer, compared with 4%–28% in other facility types). Although it was not significant, dementia was common (61%), but the rate for behavioral symptoms was low (6%).
Cluster R-3 (8 facilities) contained a very high proportion of residents with behavioral symptoms (81%, compared with less than 10% in all other clusters), many of whom had dementia (64%). Also, 28% of residents in this cluster required help in transfer. There were more licensed nursing hours in this cluster than in any other (5.2 hours/resident/week vs 0.70–2.05 hours/resident/week).
Cluster R-4 (59 facilities) housed residents who were significantly less impaired in physical function (12% and 4% required help in self care and transfer, respectively, vs 44%–80% and 11%–53%, respectively, in other clusters). Facilities in this cluster had the strictest admission policies (60% vs 76%–85% in other facilities), tended to be older, and provided more privacy (nonsignificant).
Cluster R-5 (72 facilities) tended to be mid range across most variables.
Cluster Solutions Based on Structure, Process, and Resident Case-Mix Characteristics (SPR)
A six-cluster solution was the best fitting model for the SPR analyses (log likelihood = –16,231; BIC = 33,416; entropy = 0.95). Table 4 shows this cluster solution, including means and standard deviations of variables by cluster type and results of test statistics and least significant difference multiple comparisons. We describe each cluster briefly here:
|
Cluster SPR-2 (25 facilities) was characterized by higher percentages of individuals on Medicaid (68% vs 3%–49% in other clusters) and with a history of mental and psychiatric illness (19%, which was higher than all clusters except Cluster SPR-4).
Cluster SPR-3 (57 facilities) had a higher proportion of residents with functional, cognitive, and behavioral impairments than any other cluster. Privacy and service provision were moderately high.
Cluster SPR-4 (54 facilities) tended to include large facilities (mean bed capacity = 66 vs 9–16 for other clusters). These facilities had the highest environmental quality score (21 vs 13–16 for other clusters) and scored highest in terms of policy choice, privacy, services, and social activities.
Cluster SPR-5 (7 facilities) had the highest aide turnover rates (600% vs 37%–125% in other clusters), reported the highest proportion of residents with mental illness, and had many residents on Medicaid.
Cluster SPR-6 (32 facilities) was not statistically different than other clusters in any one variable, but it did have the fewest licensed nursing hours, lowest aide turnover, least policy clarity, least resident control, and lowest Medicaid case-mix.
|
Discussion |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
Facility size was an overriding factor in three of the five subtypes (S, P, and SP). Much has been written about the importance of facility size in the field of RC/AL, with key concepts including the homelike-ness of "mom and pop" styles of care and the dangers of impersonalization and institutionalization in larger models of care (Kane & Wilson, 2001; Morgan, Gruber-Baldini, Eckert, & Zimmerman, 2004). Thus, in order to better understand the RC/AL field, we endeavored to identify criterion variables other than facility size. The typology based on the resident case-mix (R) variables was successful in this regard, identifying five distinctly different facility clusters involving a large variety of variables. Variables that significantly differentiated the clusters included facility age, nursing care, environmental quality, privacy, and resident case-mix (i.e., Medicaid, self-care, transfer, dementia, mental/psychiatric illness, and behavioral symptoms). Clearly, the analytic process was successful in differentiating types of facilities based on the types of residents they serve. Furthermore, there is some indication that the structure and process of care within clusters was consistent with resident needs and/or the challenges they presented, indicating that RC/AL facilities were organized around both resident need and market conditions. For example, in Cluster R-3, high cognitive impairment and behavioral symptoms were associated with more nursing care; in Cluster R-4, low resident impairment was associated with high privacy (structure and process) and stricter admission policies (process).
Two very similar clusters were identified using both the R and the SPR strategies: high Medicaid, high mental/psychiatric illness; and high resident ADL impairment. Unfortunately, findings of the present study also indicate that high Medicaid, high mental/psychiatric illness was paired with a poor physical environment and low privacy, raising issues about a link between public costs and quality of care. Although scholars argue that higher quality is not necessarily associated with higher costs (Davis, 1991; Mukamel & Spector, 2000), the current findings suggest that this may not be the case with regard to environmental quality. Thus, older adults who are economically disadvantaged may be forced to live in poorer physical environments with limited options. However, these data do not report the actual experiences and outcomes of care in such a cluster. Although one study did report a higher risk of hospitalization in facilities with poorer environments (Zimmerman, Gruber-Baldini, Hebel, Sloane, & Magaziner, 2002), further research is necessary to characterize the linkage between environmental quality and other outcomes.
Another notable finding about the SPR model is that the large facilities formed a cluster that carried with it better scores on physical environment quality, privacy, availability of services, and policy choice. This cluster parallels the privacy/service typology identified by Hawes and colleagues (2003), but because it is but one of six clusters, the scope of analysis used here is more comprehensive. The aggregation of large facilities into such a cluster is in line with the notion that large facilities have more resources, activities, and services than smaller facilities (Morgan, Eckert, & Lyon, 1995; Weihl, 1981). The question of whether larger size involves a tradeoff, such as more impersonalization, warrants consideration, but cannot be answered with these data.
The typologies developed in this study expand on what has been done by other researchers, most of which relates to nursing home care. Gold and colleagues (1991) derived a typology of 55 nursing homes through unstructured narratives. By using eight criterion variables related to physical environment, staff/resident interaction, and staff and administrator attitudes toward residents, they identified eight clusters with an indication of good or bad type. Although this qualitatively driven typology suggests differences related to quality, the study is limited in generalizability to nursing homes for memory-impaired older adults. Grant (1998) developed an empirical typology of 390 units in 123 nursing homes through cluster analysis, a similar procedure to that used in the present study. By using seven care attributes including physical environment, resident activity participation, and staff training, he obtained six clusters with a varying combination of care attributes. Grant's typology offered a useful way of classifying dementia care in nursing homes, but, like the typology identified by Gold and colleagues, used a narrow range of variables and was limited to nursing homes. Timko and Moos (1991) developed a typology through cluster analysis using a sample of 235 nursing homes, residential care facilities, and congregate apartments. Six distinct types of facilities emerged from the seven physical and social environment variables. Their study is one of the few that included RC/AL, but it was not specific to RC/AL. Thus, although all of these typology studies provide insights into understanding the configuration of long-term-care facilities using quantitative or qualitative methods, the present study is unique with respect to its focus on a broad range of RC/AL facilities and criterion variables and its use of resident case-mix and an innovative analytic method using a latent variable approach.
Nevertheless, it is important to note that the present study has several limitations. First, although we based clustering variables on theoretical and empirical evidence, the procedure of clustering facilities into several groups may be criticized as being data driven. Indeed, one caution about cluster analysis procedures is that the analytic procedures tend to be structure imposing rather than structure seeking (Aldenderfer & Blashfield, 1984). Replication of these results by applying the same procedures to a different data set of RC/AL facilities would strengthen findings (Milligan & Cooper, 1987). A second limitation involves both sampling and policy issues. This study was based in four states, which may limit generalizability, because state regulations affect RC/AL philosophy, admission and retention criteria, and service provision (Mollica, 2001). However, although the facilities under study were not necessarily representative of all facilities across the country, they certainly were representative of the four study states and, most likely, of common facility types nationally. Third, the resident case-mix data were reported by facility administrators, rather than compiled by a detailed listing of all facility residents. Thus, an unknown possibility of reporting bias may have affected the results. Despite these limitations, however, this typology is a helpful first step in both elucidating different models of care and providing clear criteria by which to differentiate them. Furthermore, this analytic strategy can be replicated in other states for purposes of verification or to add to what is known.
In closing, cluster solutions must be examined in terms of their practicality and relevance (Everitt et al., 2001; Muthén & Muthén, 2000). From a face-validity perspective, many of these clusters parallel distinct subtypes discussed informally by research staff involved in data collection. Some facilities with higher resident ADL and cognitive impairment may be skimming off of the nursing home market and so would be required to provide more intense services. Others, such as those with a mixed resident population, may require that the facilities contract with health professionals to meet their occasional needs for health care. The clusters do not, however, recognize the heterogeneity that exists within the clusters, related to variables that were not under study. That point notwithstanding, the two models identified both the clustering of financing (Medicaid) and mental illness, and of high ADL impairment. These variables are relevant to state regulations regarding the allocation of public dollars, as well as to the level of resident disability likely to be served by RC/AL across the nation. Additional tests of relevance would include a study of the relationship between facility type and outcomes such as resident quality of life, health status, and health care utilization. Thus, although the findings reported here have immediate usefulness in describing and differentiating the range of RC/AL facilities (and perhaps in guiding care provision and helping families and professionals match potential residents with facilities), the ultimate value of the typologies requires their replication in other regions and their application to longitudinal studies of resident outcomes. Finally, continued study is necessary to test the stability and consistency of the cluster solutions within and across states that maintain different regulatory systems and that may well be characterized through the configuration of the typologies.
|
Footnotes |
|---|
1 School of Social Work, The University of Alabama, Tuscaloosa.
2 Cecil G. Sheps Center for Health Services Research, University of North Carolina at Chapel Hill.
3 School of Social Work, University of North Carolina at Chapel Hill.
4 Department of Family Medicine, University of North Carolina at Chapel Hill.
5 Department of Epidemiology and Preventive Medicine, University of Maryland, Baltimore County.
6 Department of Sociology and Anthropology, University of Maryland, Baltimore County.
Decision Editor: Linda S. Noelker, PhD
Received for publication March 21, 2005. Accepted for publication October 24, 2005.
|
References |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
This article has been cited by other articles:
 |
|
 |
|
  C. S. Schwalbe, R. J. Macy, S. H. Day, and M. W. Fraser Classifying Offenders: An Application of Latent Class Analysis to Needs Assessment in Juvenile Justice Youth Violence and Juvenile Justice, July 1, 2008; 6(3): 279 - 294. [Abstract] [PDF]
|
|
|
|
 |
|
 D. L. White, L. Newton-Curtis, and K. S. Lyons Development and Initial Testing of a Measure of Person-Directed Care Gerontologist, July 1, 2008; 48(suppl_1): 114 - 123. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. C. Stearns, J. Park, S. Zimmerman, A. L. Gruber-Baldini, T. R. Konrad, and P. D. Sloane Determinants and Effects of Nurse Staffing Intensity and Skill Mix in Residential Care/Assisted Living Settings Gerontologist, October 1, 2007; 47(5): 662 - 671. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||
| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
