Hospitalization of Nursing Home Residents With Cognitive Impairments: The Influence of Organizational Features and State Policies

Abstract

Purpose: The purpose of this study was to quantify the effect of specific nursing home features and state Medicaid policies on the risk of hospitalization among cognitively impaired nursing home residents. Design and Methods: We used multilevel logistic regression to estimate the odds of hospitalization among long-stay (>90 days) nursing home residents against the odds of remaining in the nursing home over a 5-month period, controlling for covariates at the resident, nursing home, and county level. We stratified analyses by resident diagnosis of dementia. Results: Of 359,474 cognitively impaired residents, 49% had a diagnosis of dementia. Of those, 16% were hospitalized. The probability of hospitalization was negatively associated with the presence of a dementia special care unit (adjusted odds ratio [AOR] = 0.90, 95% confidence interval [CI] = 0.86–0.94) and with a high prevalence of dementia in the nursing home (AOR = 0.96, 95% CI = 0.88–1.03). Higher Medicaid payment rates were associated with reduced likelihood of hospitalization (AOR = 0.95, 95% CI = 0.90–1.00), whereas any bed-hold policy substantially increased that likelihood (AOR = 1.44, 95% CI = 1.12–1.86). We observed similar results for residents without a dementia diagnosis. Implications: Directed management of chronic conditions, as indicated by facilities' investment in special care units, reduces the risk of hospitalization, but the effect of bed-hold policies illustrates how fragmentation in the financing system impedes these efforts.

Dementia is the most common diagnosis among nursing home populations and has been reported to affect as many as 50% of residents (Magaziner et al., 2000). These residents are vulnerable to the same acute medical problems as others, but the sequelae of dementia, including communication and behavioral disorders, increase the complexity of their care. This places an additional onus on nursing staff to identify changing health conditions and act accordingly. Research has found that residents with dementia have fewer physician visits and hospitalizations than other residents but similar use of emergency departments (Burton et al., 2001). Furthermore, studies have shown that when residents with dementia are hospitalized, the admitting diagnosis is more often the result of an ambulatory care sensitive condition (Carter, 2003). Once hospitalized, these residents may be at greater risk for complications such as delirium because service delivery in this setting is often not consistent with the needs of people with dementia (Cunningham, 2006).

The complexity of providing care to people with dementia and discussions around the intensity with which medical occurrences should be treated have resulted in conflicting views on approaches to care. Given this divergence, Porell and Carter (2005) tested the variability of rates of discretionary hospitalization for nursing home residents with and without dementia. Based on the professional uncertainty thesis, which postulates that non-medical factors, such as local practice style, will have a stronger influence on treatment decisions for conditions that do not have an agreed-upon optimal treatment, they hypothesized that they would observe more variability for residents with dementia diagnoses. They did find greater interfacility variation in hospitalization rates for residents with dementia than for other residents and concluded that this reflected a stronger influence of contextual characteristics because of the higher level of uncertainty around medical treatments for people with dementia.

Although studies have documented the effect of facility characteristics on risk of hospitalization (Carter & Porell, 2003; Castle & Mor, 1996; Intrator, Castle, & Mor, 1999; Intrator, Zinn, & Mor, 2004), few have specifically examined this among residents with dementia. Those that have found that more skilled staff and the presence of specialized care each played a role (Carter & Porell, 2005; Porell & Carter, 2005). Other research has consistently shown dementia special care units to be associated with improved outcomes that extend beyond residents on those units (Castle, 2000; Castle & Fogel, 1998; Mitchell, Kiely, & Gillick, 2003; Mitchell, Teno, Roy, Kabumoto, & Mor, 2003). Research that has examined other forms of specialized care in the nursing home, such as hospice, has found that the practices associated with specialization diffuse through the nursing home and affect the care of residents who were not the direct recipients of that specialized care (Wu, Miller, Lapane, & Gozalo, 2003). Other expressions of dementia as a contextual factor, such as concentration of residents with the diagnosis, are less understood. We presume that each of these contextual factors results in a level of knowledge, either through specialization or experience, that improves the staff's ability to manage medical conditions and reduces the need for hospitalization.

Beyond the nursing home itself, the context within which the facility provides care influences resident hospitalization. Investigators have documented state-to-state variation in hospital admission rates, and, although limited, research has begun to show that this variation is related to state Medicaid funding policies. More generous Medicaid per diem reimbursement rates are associated with higher staffing and the use of nurse practitioners/physician assistants, which has an indirect effect on the ability to prevent hospital use (Grabowski, Angelelli, & Mor, 2004; Grabowski, Feng, Intrator, & Mor, 2004; Intrator et al., 2005). Recent studies have also shown a direct effect, with $10 increases in daily reimbursement associated with a reduction of 5% to 9% in the risk of hospitalization (Intrator et al., 2006; Intrator & Mor, 2004).

Bed-hold policies guarantee that Medicaid will pay some portion of the per diem rate to the nursing home in the event that a resident is hospitalized. Although the specific details of the policy (including number of days held, percent of per diem paid, and occupancy conditions) vary by state, these policies aim to promote continuity of care by enabling a resident to return to his or her original nursing home following hospital discharge. However, some research has suggested that these policies also create an incentive to hospitalize and that they account for state differences in hospitalization rates of long-stay residents (Freiman & Murtaugh, 1993; Intrator et al., 2006).

The purpose of this study was to quantify the effect of Medicaid reimbursement policies and nursing home features pertinent to dementia care on the risk of hospitalization among residents with a diagnosis of dementia and to compare this with their effect on risk of hospitalization among residents without a diagnosis of dementia but with cognitive impairment. Based on the professional uncertainty thesis and the work of Porell and Carter (2005), we hypothesized that contextual nursing home and state policy variables would have a stronger influence on risk of hospitalization for residents with dementia than for residents without dementia. We further hypothesized that residents with dementia who reside in nursing homes with a special care unit or a higher concentration of residents with dementia would have a lower risk of hospitalization than would residents with dementia who reside in nursing homes without these characteristics. To our knowledge, this is the first study to examine the effects of state Medicaid policies on hospitalization among this subgroup.

Methods

This study derives from a larger study that was conducted to test associations of various nursing home, market, and state Medicaid funding policies with the risk of hospitalization among long-stay nursing home residents (Intrator et al., 2006). We selected our sample from the same cohort of residents and developed a comparable analytic plan (described below). The original study did not test the effects of either resident diagnosis of dementia or nursing home characteristics related to dementia care.

Data Sources

We obtained data from multiple sources in order to characterize the resident, nursing home, market, and state Medicaid policies for reimbursement of nursing home care. All data were for the year 2000.

Resident-level data came from the Minimum Data Set (MDS), a comprehensive clinical tool used to collect data with more than 400 items, including cognitive and physical functioning, diagnoses, and other health conditions. Regular MDS completion was mandated as part of the Omnibus Budget Reconciliation Act of 1987; it is conducted in every federally certified nursing home on every resident at specified intervals: upon admission, quarterly, and following a significant health change (Mor, 2004; Phillips & Morris, 1997). Data for this study came from the Centers for Medicare & Medicaid Services repository and represent the 48 contiguous U.S. states (excluding the District of Columbia). We used Medicare inpatient claims and eligibility files to characterize hospital use.

Data on nursing homes came from the Online Survey Certification and Reporting (OSCAR) database, which provides information on facility organization (e.g., proprietary status and chain membership) and structure (e.g., bed size and staffing). OSCAR data are collected through annual surveys of all certified nursing homes in the United States. These data are the most comprehensive national source of facility information and have been used in several other studies (Castle, 2000; Intrator et al., 2004; Lapane & Hughes, 2004).

We defined the nursing home market at the county level. Other research has suggested that the county is an adequate approximation of the market given patterns of funding and patient referral (Banaszak-Holl, Zinn, & Mor, 1996). We took county descriptors from the Area Resource File, which contains an array of demographic, health, and social resource information gathered from sources such as the American Hospital Association, the U.S. Census Bureau, the Centers for Disease Control and Prevention, and the National Center for Health Statistics.

The information on state Medicaid policies came from a 2002 survey of state Medicaid offices conducted through the Center for Gerontology and Health Care Research at Brown University in Providence, Rhode Island. The content of the survey focused on Medicaid reimbursement policies for nursing home beds and covered the years 1999 through 2002. We included data on average Medicaid reimbursement rate and existence of a bed-hold policy in 2000 in this study (Grabowski, Feng et al., 2004).

Study Sample

For this study, we used data from long-stay nursing home residents in facilities in urban counties in the continental United States. We excluded Alaska and Hawaii because their remote location likely results in a different set of geographic and cultural influences on nursing home and hospital use. Research has shown that rural and urban nursing homes differ on a number characteristics, including resident profile, resident preferences, and approaches to medical care (Bolin, Phillips, & Hawes, 2006; Coburn, Keith, & Bolda, 2002; Gessert, Elliott, & Peden-McAlpine, 2006; Gessert, Haller, Kane, & Degenholtz, 2006; Phillips, Holan, Sherman, Williams, & Hawes, 2004). Furthermore, the relative isolation of rural nursing homes may result in a different set of responses to state policies and constraints on hospital use. Restriction to urban facilities eliminated potential confounding by features that differentially characterize urban and rural nursing homes. We excluded facilities with fewer than 50 beds for several reasons. Small facilities may have different operating incentives than larger ones and appear to cater to a select population. These nursing homes are also less likely to report a special care unit, one of the determinants of interest (Agency for Healthcare Research and Quality, 2000; Leon, Cheng, & Alvarez, 1997). We further restricted inclusion to freestanding facilities, because these likely differ from hospital-based facilities on their resident case mix, admission profile, and resource availability.

We selected all resident assessments from the second quarter of 2000. We then restricted inclusion to long-stay residents because short-stay residents have a different likelihood of hospital admission based on their clinical profile, need, and familiarity to staff. We defined long-stay residents as those who had at least one other MDS assessment prior to the 90 days immediately preceding baseline. We also excluded residents who were younger than 65 years of age, who were in a coma, who were enrolled in a health maintenance organization during the 2000 calendar year (to ensure completeness of claims data), or who could not be matched to either OSCAR data or Medicare enrollment records. Of the 1,740,074 residents in a nursing home during the second quarter of 2000, major exclusions were as follows: resided in a rural nursing home (22.4%), short-stay (15.5%), not matched to Medicare (11.7%), or health maintenance organization enrollee (6.1%), for a total of 558,608 eligible residents in 8,293 nursing homes (55.1% of certified nursing homes).

Finally, we restricted the sample to residents with at least moderate cognitive impairment to improve comparability between residents with and without dementia. Other studies have shown a consistent association between cognitive impairment and a lower risk of hospitalization. Therefore, we decided to exclude residents with no or mild cognitive impairment so that any observed differences between residents with and without dementia would more likely be related to the diagnosis itself rather than to differences in cognitive abilities. We measured cognitive impairment using the Cognitive Performance Scale, an MDS-embedded scale that has been validated against the Mini-Mental State Exam and the Test for Severe Impairment. The scale has seven levels and a score of 0 to 2 is indicative of no or mild impairment (Morris et al., 1994). The final sample consisted of 359,474 nursing home residents, of whom 174,563 had a diagnosis of dementia.

Outcome Variable—Hospitalization

We used Medicare data to identify residents who had an acute-care hospitalization or who died (whichever came first) within 150 days of the baseline MDS assessment. When there was no evidence of nursing home discharge or home health care provision, we assumed that those residents who were not hospitalized and did not die remained in the facility. We distinguished between residents who died (but were not hospitalized) and residents who remained in the facility because they are presumably different in ways that could be related to their probability of hospitalization. There were 41,832 residents (11.6%) who died but were not hospitalized during follow-up (52.3% had a diagnosis of dementia).

Determinant Variables

Although there were several resident-, facility-, and county-level variables included in the final models, the main focus of this analysis was on specific facility features related to dementia care and state policies for Medicaid nursing home funding.

Organizational Features

Whether a facility had a special care unit for the care of residents with Alzheimer's disease or other dementias came directly from OSCAR data. We considered the presence of a dementia special care unit a reflection of a facility's investment in dementia care. The creation and maintenance of a special care unit requires an initial investment for planning and modification, whereas its maintenance requires ongoing investment in aspects of staffing (Grant, Kane, Potthoff, & Ryden, 1996; Maslow, 1994) and programming (Leon, 1994).

The other facility variable of interest was an indicator of the prevalence of dementia (Alzheimer's disease or other dementia diagnosis) among long-stay residents within the facility. The higher the prevalence of dementia in a nursing home, the more experience staff have with the specific symptoms, needs, and conditions associated with the dementia. This may in turn result in greater capacity to identify and treat emerging conditions early and to better discriminate between conditions for which hospitalization is an appropriate response. We created this variable by calculating the proportion of long-stay residents in each facility who had a diagnosis of Alzheimer's disease or other dementia. For ease of interpretation, we dichotomized the variable at the median (35% or greater).

State Policies

We tested the effect of two Medicaid reimbursement policies on nursing home resident hospitalization. Data on both policies came from our survey of state Medicaid offices (Grabowski, Feng et al., 2004). For the multivariable analyses, we standardized the Medicaid payment rate so that the 48 state mean would be $100, with $10 increments.

We were also interested in the effect of a state bed-hold policy. For this study, we contrasted the presence of any bed-hold policy with no policy.

Other Covariates

Because of the complexity of the issues under study and the multiple levels that potentially affect hospitalization, we adjusted for several control variables in the multivariable analyses. We chose variables that have been shown to be associated with (a) the risk of hospitalization, either in published literature or in preliminary analyses, or (b) the provision of care that would influence the risk of hospitalization.

Resident Characteristics

The majority of resident data came from the MDS assessments. This included age at assessment, gender, race (Black vs others), completion of do-not-resuscitate and do-not-hospitalize orders, weight (severely underweight, body mass index < 18; obese, body mass index > 30), cognitive performance (Cognitive Performance Scale score), functional status (flacker score), indicator variables for specific diagnoses (diabetes, cancer, emphysema/chronic obstructive pulmonary disease, and congestive heart failure), unstable medical condition, fever, recent weight loss, and more than nine prescribed medications. We took hospice enrollment during the follow-up period from Medicare claims files.

Organizational Features

Variables taken from the OSCAR database were as follows: presence of either nurse practitioner or physician's assistant (Intrator et al., 2004, 2005), more than half full-time equivalent physician other than the medical director (Intrator et al., 2004), nurse staffing greater than the recommended 4.55 hour per resident day (Harrington et al., 2000), ratio of registered nurses to all nurses (Porell & Carter, 2005), total number of beds, proprietary status, and owned by a chain. We created other facility-level variables by aggregating MDS data for all residents. These included mean case-mix index based on all annual assessments for 2000, percent Medicaid (>85% vs others), percent Medicare (>15% vs others), and percent private pay (>35% vs others). We included occupancy (>90% vs others) as a proxy for financial stability (Weech-Maldonado, Neff, & Mor, 2003).

Market Features

We described the nursing home market at the county level and derived the variables from the Area Resource File. We controlled for market income (per capita income), demand for services (percentage of population 75 years or older), and availability of health services resources (number of hospital beds per 1,000 persons 75 years or older). We quantified competitiveness of the nursing home market using the Herfindahl Index, which is calculated by summing the square of each nursing home's bed share in the market (Zinn, 1994). Finally, we used OSCAR data to calculate a county's average number of empty nursing home beds (per facility).

Analytic Design

We used descriptive statistics to characterize the sample and a multilevel model to estimate effect sizes and standard errors. Multilevel models allow for several levels of nesting; in this case, residents nested in nursing homes, nursing homes nested in markets, and markets nested in states. This approach to modeling explicitly recognized that residents in a given nursing home are likely similar on unmeasured (and perhaps immeasurable) characteristics that may influence the likelihood of hospitalization and that the same was likely so for nursing homes in markets and markets in states. We used a binary response model to contrast persons who were hospitalized against those who were alive, and presumably remained in the nursing home, at the end of study period. In early versions of the model, we tested interactions between each diagnosis of dementia and nursing home characteristics, diagnosis of dementia and state policies, and nursing home characteristics and state policies. We found no evidence of statistical interaction for any combination. Therefore, we decided to stratify the model by resident diagnosis of dementia because those with and without a formal diagnosis may differ on other characteristics associated with the likelihood of hospitalization. We present model results as adjusted odds ratios (AORs) with 95% confidence intervals (CIs). We conducted all descriptive statistics using SAS (Research Triangle Institute, Research Triangle Park, NC) and estimated multilevel models using MLWiN (Multilevel Models Project, Institute of Education, University of London, England).

Results

Of the 316,268 nursing home residents included in this study, approximately half (48.2%) had a documented diagnosis of Alzheimer's disease or other dementia. Within 150 days of baseline measurement, 18.4% of the sample had been hospitalized and there was little difference between the two diagnostic groups (Table 1). Residents with and without a dementia diagnosis were similar on most characteristics. The majority of residents were older than age 75, female, and White. Compared to residents without dementia, residents with dementia were more likely to have a diagnosis of chronic obstructive pulmonary disease/emphysema, congestive heart failure, or cancer. They also tended to have more severe cognitive impairment (Cognitive Performance Scale ≥ 5). Do-not-resuscitate and do-not-hospitalize orders were more commonly recorded for residents with dementia.

Table 2 presents the descriptive characteristics of included nursing homes, counties, and states. Of the 8,293 nursing homes, 20.3% reported an Alzheimer's/dementia special care unit and 50.3% had a prevalence of dementia of at least 35%. The average number of nursing home beds was 129, and the majority of facilities were for-profit (74.4%) and owned by a chain (60.7%). Approximately half had occupancy rates above 90%. On average, one third of nurses were registered nurses; few nursing homes had a physician other than the medical director employed at more than half full-time equivalent, although 22.8% reported having either a nurse practitioner or physician's assistant employed or under contract.

There were 810 urban counties. The mean per capita income in these counties was $26,760, but this varied substantially. The mean percentage of the population older than 75 was 5.8%, with a range of 1.4% to 16.3%. Similarly, the number of hospital beds for every 1,000 people older than 75 years showed substantial variability; the mean was 48 beds, and the range was 0 to 417. Nearly 29% of counties had a highly competitive nursing home market. More than three quarters of states had some form of bed-hold policy, and the average Medicaid payment rate in 2000 was $103 (ranging from $67 in Oklahoma to $160 in New York).

Table 3 presents the multilevel model results. Only the estimates of effect for the facility features and state policies of interest are shown. The full list of control variables is in the footnote to the table.

In both models, the odds of a resident being hospitalized decreased in the presence of a special care unit (dementia: AOR = 0.90, 95% CI = 0.86–0.94; no dementia: AOR = 0.93, 95% CI = 0.90–0.98). A higher prevalence of dementia in the nursing home had a similar but slightly weaker effect on an individual resident's odds of hospitalization (dementia: AOR = 0.96, 95% CI = 0.88–1.03; no dementia: AOR = 0.93, 95% CI = 0.96–1.00). For each $10 increase in the Medicaid rate over the $100 average, we observed a slight decrease in the likelihood of hospitalization for all residents (dementia: AOR = 0.95, 95% CI = 0.90–1.00; no dementia: AOR = 0.95, 95% CI = 0.91–1.00). We observed the strongest effects in both diagnostic groups for bed-hold payment policies. A nursing home resident who lived in a state with any type of bed-hold policy had a 40% higher likelihood of being hospitalized than a similar resident in a state without such a policy (dementia: AOR = 1.44, 95% CI = 1.12–1.86; no dementia: AOR = 1.47, 95% CI = 1.19–1.82).

Discussion

Our findings do not support our hypothesis that contextual factors would more strongly influence the risk of hospitalization for residents with dementia than for residents without dementia. Unlike Porell and Carter (2005), we did not restrict our analysis to discretionary hospitalizations, and this may, at least partially, account for the differences in our findings. As well, we limited our comparison to residents without dementia but who were cognitively impaired. These residents may not differ enough from residents with dementia (and cognitive impairment) to elicit any measurable differences in care practices by staff. For all residents, though, we did find that facility features representing investment in and experience with dementia care were negatively associated with a resident's risk of hospitalization, whereas the effects of state Medicaid funding policies were dependent on the incentives created for nursing home practice.

Nursing homes with dementia special care units are systematically different from nursing homes without such units, not only in terms of structural characteristics, such as size and ownership, but also in terms of resident outcomes (Leon et al., 1997). Prior research findings suggest that nursing homes with special care units can be characterized by an overall less aggressive approach to care. Both feeding tubes and physical restraints are found less frequently in nursing homes with special care units than in other homes (Castle, 2000; Castle & Fogel, 1998; Castle, Fogel, & Mor, 1997; Mitchell, Kiely et al., 2003; Mitchell, Teno et al., 2003). Our finding of a lower probability of hospitalization among residents in nursing homes with special care units compared to residents in other nursing homes is consistent with this. Furthermore, our results suggest that it is not simply the higher percentage of residents with dementia diagnoses, a group with a well-reported decreased risk of hospitalization, that accounts for these observed associations between special care unit presence and care practice. Nor is this finding an artefact of special care unit concentration in states with higher Medicaid reimbursement (r =.14, p =.36). Rather, there does appear to be some fundamental underlying organizational or philosophical approach to care that may not be easily measured but is captured by the dementia special care unit indicator. Exactly why this should be the case is unclear; however, other investigators have found that all staff in nursing homes with special care units, and not just those who work on the special care unit, receive more comprehensive dementia-related training than staff in other nursing homes (Grant et al., 1996; Kane, Jordan, & Grant, 1998). Furthermore, nursing homes with special care units are more likely to engage in other innovative service activities (Morris & Emerson-Lombardo, 1994). Perhaps what these nursing homes really offer is an environment that better prepares staff to provide regular care to residents and then supports and assists staff in making care decisions, such as whether to hospitalize, when residents' needs change.

Although the presence of a dementia special care unit may represent a certain practice style, a high facility prevalence of demented residents suggests that more experience is beneficial to resident care outcomes. That this finding similarly applies to residents both with and without diagnoses of dementia suggests that reduced rates of hospitalization are not merely a result of familiarity with the sequelae of Alzheimer's disease and related disorders. It may be that the more often staff work with residents with dementia, the greater their opportunity to develop skills in communicating with and assessing the needs of cognitively impaired residents. This, in turn, may result in a more highly attuned ability to identify treatable conditions early as well as more confidence in the ability to manage certain conditions within the nursing home. Alternatively, this finding may not be a reflection of the practice-makes-perfect hypothesis but simply an example of selective referral (Gandjour & Lauterbach, 2003). That is, those nursing homes that are reputed to provide better care to cognitively impaired residents are also more likely to attract incoming residents with dementia diagnoses. Unfortunately, we are unable to parse out the direction of the relationship here.

As with other research, we found that a $10 increase in Medicaid per diem rate could decrease a resident's probability of hospitalization by 5% (Intrator & Mor, 2004). This finding suggests that higher funding enables nursing homes to provide more in-house services, such as nurse practitioners and physician assistants, and to better manage chronic conditions regardless of their expertise in dementia care. In turn, this presumably reduces the occurrence and severity of acute exacerbations and the ensuing need for hospitalization.

In theory, bed-hold policies benefit nursing home residents by ensuring that in the event of a hospitalization, return to the original facility is possible. This may be important for both continuity of care and quality of life. However, our findings, and those of others (Intrator et al., 2006), clearly illustrate that these policies inadvertently create incentives that lower the threshold for hospitalization. Even at less than the regular per diem, state bed-hold policies mean that a nursing home receives payment for a particular resident without providing care to that resident.

Although the policies we studied are very different in intent, their effects on hospitalization represent a common underlying tension in the current acute-care paradigm. Medicare's restricted coverage of physician and hospital services means that continuing management of chronic conditions becomes Medicaid's responsibility. There is no incentive, however, for state Medicaid systems to fund nursing homes to fully manage chronic conditions, including dementia, because they do not benefit from the savings that result from reduced hospitalizations; rather, only Medicare is affected, from a fiscal standpoint, by states' efforts to minimize hospitalizations. This fragmentation in the long-term-care system undermines the spirit with which Medicare was enacted and limits the ability of government programs to promote optimal long-term care for older Americans. Integrated models, such as Evercare, promote medical management within the nursing home and have been successful in preventing unnecessary, and costly, hospitalizations (Kane, Keckhafer, Flood, Bershadsky, & Siadaty, 2003).

We formally tested interactions between diagnosis of dementia and each of the contextual characteristics under study and found no evidence of differential effects. This suggests that risk of hospitalization for residents with dementia is no more influenced by external forces than is the risk for residents without dementia but with similar cognitive impairment. With respect to the effect of state Medicaid policies, residents with dementia are equivalently affected by the current fragmented system. Despite this, the role that dementia plays in modifying the effect of organizational and policy features on care practices and outcomes remains an important consideration as reforms on the long-term-care system and funding mechanisms continue to occur.

There are limitations to this study. First, we did not differentiate between necessary and unnecessary or potentially preventable hospitalizations. Although these distinctions are certainly important for developing a full understanding of patterns of hospitalization, we were unable to reliably make them given the available data. Even so, there is some suggestion that those nursing homes considered to have low rates of hospitalization have a lower percentage of potentially preventable hospitalizations than do those nursing homes considered to have a high rate of hospitalization (Carter & Porell, 2005). The large amount of data prohibited us from modeling a multinomial outcome that included death within the multilevel framework. Instead, we estimated the odds of hospitalization against the odds of remaining in the nursing home while censoring residents who died. This estimation method provides unbiased estimates of effect (i.e., the beta coefficients were relatively unaffected) but results in inflated estimates of the standard errors around the coefficients (Begg & Gray, 1984). However, because the sample was large, inflated standard errors were not a serious concern. Third, we did not test the effect of special care unit residence on hospitalization. Although the MDS does include an indicator for special care unit residence, this variable is not required on quarterly assessments and is only available for all states on the annual assessment. Testing this would have required a different sampling strategy. This may be an important question for future research. Finally, we should note that there is some debate regarding the definition of a dementia special care unit and evidence to suggest that a not-insignificant proportion of such units do not appear to provide anything special (Holmes & Teresi, 1994; Morris & Emerson-Lombardo, 1994). Regardless, the repeated finding that nursing homes with special care units perform differently than those without suggests that the special care unit is a clear marker for a distinct practice style (Castle, 2000; Grant, Potthoff, Ryden, & Kane, 1998; Mitchell, Teno et al., 2003; Morris & Emerson-Lombardo, 1994). Future research is needed to better describe how special care unit presence and overall nursing home quality are related to each other.

The aging of the baby boomers and the continued expansion of life expectancy at 65 years have raised both awareness about the increasing cost of long-term care and concerns about the quality of that care. Given the economic costs and potentially detrimental health consequences associated with the hospitalization of nursing home residents, a better understanding of this issue is important for addressing both of those concerns. We know that multiple factors influence a nursing home resident's probability of being hospitalized, many of which are not a direct measure of that resident's health status or need. We have shown that nursing home features, in particular the practice styles represented by the presence of a dementia special care unit, as well as financial incentives created by state Medicaid funding policies all impact on a cognitively impaired resident's risk of hospitalization and that their influence is not different among residents with dementia.

The ability of individual nursing homes to avoid unnecessary hospitalizations will be contingent on the identification and dissemination of those organizational features and practices that support day-to-day management. More importantly, however, may be the development of a policy environment that promotes facility adoption of best-practice procedures, such as those associated with special care unit care. This will mean better integration of payers such that both Medicare and Medicaid jointly invest in the medical management of nursing home residents and jointly benefit from a reduction in hospitalizations. Other researchers have made this call for better integration of Medicare and Medicaid (Cassel, Besdine, & Siegel, 1999; Feder & Lambrew, 1996; Leutz, Greenlick, & Capitman, 1994), but our findings reinforce its need. Ongoing chronic care management in the nursing home, which represents the intersection between long-term and acute care, requires adequate funding in conjunction with policies that discourage the shifting of responsibility between state and federal governments. Only once this type of coordinated response is put into action will reductions in hospitalization mark both improvements in nursing home care and responsible fiscal constraint.

References