Patient and Caregiver Characteristics and Nursing Home Placement in Patients With Dementia FREE

Cognitive impairment, especially severe impairment or dementia, is one of the primary indications for nursing home placement. As many as 90% of patients with dementia reportedly become institutionalized before death.¹ However, most patients with dementia continue to live in the community until family caregivers are no longer able to care for them.² The decision to place a patient with dementia in long-term care is complex and is based on patient and caregiver characteristics and the sociocultural context of patient and caregiver. Yet most studies that have assessed the predictors of nursing home placement have focused primarily on the characteristics of either the patient, such as dementia severity or difficult behaviors,^1,3- 8 or the caregiver, such as subjective burden or health status,^9- 10 and few studies have comprehensively investigated how both caregiver and patient characteristics influence nursing home placement. In addition, most studies that have investigated predictors of nursing home placement of patients with dementia have been small (no more than several hundred patients) and have not included different ethnic groups. More research is needed to understand factors contributing to the complex decisions about nursing home placement in an ethnically diverse cohort.

Information about predictors of nursing home placement could shape interventions to delay placement. Delaying long-term care by 1 month for every person aged 65 years or older in the United States could save $60 billion annually in formal care.¹¹ Thus, interventions aimed at delaying or preventing long-term care placement could reduce health-care costs tremendously and possibly improve the quality of life of patients by maintaining their independence.¹²

As part of a prospective study of 5788 ethnically diverse patients who had dementia and lived in the community, we developed and validated a prognostic model to predict which patient and caregiver characteristics were associated with nursing home placement throughout a 3-year period. We hypothesized that including patient and caregiver characteristics would better predict risk of nursing home placement among patients with advanced dementia.

Patients were enrolled in the Medicare Alzheimer's Disease Demonstration and Evaluation (MADDE), a randomized trial for patients with dementia and for their caregivers. The intervention consisted of expanded in-home and community-based services and case management compared with usual care.¹³ Beginning in December 1989, patients were enrolled throughout a 2-year period at 8 sites thought to be broadly reflective of the demographic makeup of the United States (Champaign-Urbana, Ill; Cincinnati, Ohio; Memphis, Tenn; Miami, Fla; Minneapolis, Minn; Parkersburg, WVa; Portland, Ore; and Rochester, NY) and then observed for 36 months, with the last assessment occurring in December 1994. Patients were recruited by physician referral and self-referral. Eligibility was established from a screening questionnaire and the physician referral form. Criteria for enrollment consisted of a physician-certified diagnosis of irreversible dementia, enrollment in or eligibility for Medicare Parts A and B, residence in one of the demonstration project site's catchment areas, living in the community, and presence of a caregiver. A total of 8095 patients were screened, with a total of 5788 patients with dementia enrolled in MADDE at study initiation. Patients from the intervention and control arm were combined for this analysis because the intervention had no effect on patients' outcomes, such as nursing home placement, the use of community services, and Medicare claims records, or on caregiver outcomes, such as burden and depression.^14- 15 The Human Subjects Committee at the University of California, San Francisco, approved the recruitment, informed consent, and data collection procedure, and all patients, their proxies, and caregivers provided informed consent.

At baseline, a trained interviewer administered an in-home assessment to patients and their primary caregiver. Information on patient age, sex, marital status, education, ethnicity, living situation, and income was collected. The patient's cognitive status was assessed at enrollment with the 30-point Mini-Mental State Examination (MMSE).¹⁶ The patient's functional status was measured with the Katz activities of daily living (ADL) scale for bathing, eating, dressing, transferring, and toileting.¹⁷ Dependency in ADLs was defined as the patient having at least some difficulty with each item, as reported by the patient or caregiver. We recorded medical comorbidities such as a history of hypertension, diabetes, or coronary heart disease from information obtained from the patient's physician at enrollment. Caregivers were also asked whether the patient ever had incontinence of bladder or bowel and whether the patient typically experienced behaviors categorized as psychotic symptoms (visual hallucinations or suspiciousness), aggressive behaviors (episodes of combativeness or anger), being a danger to oneself to others (including wandering), and waking up the caregiver at night.

At baseline, investigators interviewed caregivers in person and recorded caregiver age, sex, education, marital status, relationship to the patient, and income. Caregiver burden was assessed with the Zarit scale,¹⁸ with specific items assessing caregiver sense of control, anger, stress, and time constraints (scores ranged from 0 to 32, with higher scores indicating a greater burden). We categorized patients into quartiles of caregiver burden. We collected information on the number of hours that the caregiver took care of the patient weekly, categorized according to the median number of hours (<90 hours or ≥90 hours). The 15-item Geriatric Depression Scale (GDS) was administered to assess caregiver depression, with scores ranging from 0 to 15; we used the recommended cutoff of at least 6 symptoms to indicate presence of depression.¹⁹ Caregivers were asked whether, compared with other people their age, they perceived their health as excellent, good, fair, or poor. Caregiver functional dependence was measured with the Katz ADL scale and with the Lawton instrumental activities of daily living (IADL) scale.²⁰ We characterized caregivers as independent in all ADLs and IADLs or dependent in any ADL or IADL.

Our primary outcome measure was time to nursing home placement, assessed by interviews at 6, 12, 18, 24, and 36 months after study initiation and by review of Medicare records. Nursing home placement was defined as patient admission to long-term care for at least 30 days during the 36 months of follow-up. Deaths among the study population were recorded throughout the 3 years of follow-up. We had 100% follow-up for mortality and nursing home placement outcomes.

We randomly divided the sample into a development cohort (two thirds of the cohort, n = 3859) and a validation cohort (n = 1929). We first calculated the Kaplan-Meier estimates of the proportion of patients in the development cohort entering nursing homes by 12, 24, and 36 months. We censored patients after 3 years of follow-up or at death. With bivariate models, we then determined which patient and caregiver characteristics in the development cohort were associated with time to nursing home placement by using log-rank tests.

We used Cox proportional hazards models to determine which variables were independently associated with time to nursing home placement by 36 months. Our modeling strategy had 2 stages. In the first stage, we developed separate multivariate models for patient and caregiver characteristics, including variables significantly associated with nursing home placement (P<.10) in the bivariate models. Next, we developed a model combining patient and caregiver characteristics, including variables that were statistically significant (P<.10) in either the patient or caregiver multivariate models. These variables were entered with a stepwise forward approach. A backward elimination approach produced similar results. We compared the fit of the combined model with the models including patient or caregiver characteristics separately.

To facilitate interpretation of the multivariate model, we calculated the predicted risk of nursing home placement for each patient according to the coefficients of patient and caregiver characteristics that were retained in the final model. We then divided patients into risk quartiles and calculated rates of nursing home placement at 1, 2, and 3 years for each quartile. To assess the predictive accuracy of this model, we used the coefficients from the development cohort to determine the predicted rate of nursing home placement for each patient in the validation cohort. To examine model calibration, we then compared the predicted rates of nursing home placement with actual rates for these patients. To examine model discrimination, we compared the C statistic in the validation cohort to the C statistic in the development cohort. All analyses were conducted with SAS software version 8.02 (SAS Institute Inc, Cary, NC).

The mean age of the patients with dementia in the development cohort (n = 3859) and the validation cohort (n = 1929) was nearly 79 years, and 60% were women (Table 1). Baseline characteristics of the patients with dementia were similar for the 2 cohorts (P>.05 for all comparisons). Most of the participants had at least 1 ADL dependency, and their mean (SD) MMSE score was 14.2 (8.8), with the majority of patients having MMSE scores consistent with moderate to severe dementia. About 86% (n = 4929) of patients had at least 1 difficult behavior, and 37% (n = 2124) had at least 3 difficult behaviors. The mean age of the caregivers was nearly 64 years, and 72% were women (Table 2). Most of the caregivers (63%) were the wife, daughter, or daughter-in-law of the patients, and one third of caregivers were dependent in either ADL or IADL function. The caregivers had a mean (SD) depression score of 4.4 (3.4), and almost one third had at least 6 depressive symptoms consistent with a diagnosis of depression. Baseline caregiver characteristics were similar for the 2 cohorts (P>.05 for all comparisons).

In the development cohort (n = 3859), the Kaplan-Meier estimates of nursing home placement throughout 1, 2, and 3 years of follow-up were 22%, 40%, and 52%, respectively. In the bivariate analyses, patients who were at least 80 years of age, living alone, and white and had worse functional status, a lower MMSE score, and any difficult behavior were more likely to be placed in a nursing home (Table 3). Caregiver characteristics that were associated with patient admission to a nursing home included older age, being married, caregiving for at least 90 hours a week, being functionally dependent, having at least 6 depressive symptoms, and having a higher burden score (Table 4). However, if the caregiver had an annual income of less than $10 000 or was a daughter or daughter-in-law, patients were less likely to be placed in long-term care. During the study, 619 patients died in the first year, 472 in the second, and 387 in the third.

In the multivariate model combining patient and caregiver characteristics, those of black or Hispanic ethnicity were less likely to be placed in a nursing home throughout the 3 years (Table 5). On the other hand, patients who lived alone and had greater ADL dependencies, lower MMSE scores (≤20 points), and at least 1 difficult behavior were more likely to go to a nursing home. Caregivers who were 65 years or older (compared with those <65 years of age) and who had higher Zarit Burden scores were more likely to have their family member go to a nursing home. Results of the models revealed that the multivariate model combining patient and caregiver characteristics was better able to predict time to nursing home placement compared with the model including patient characteristics only (P<.001) or caregiver characteristics only (P<.001).

We assigned patients to quartiles of risk according to the multivariate model. In the development cohort, patients in the first, second, third, and fourth quartile had a 25%, 42%, 64%, and 91% rate of nursing home placement at 3 years, respectively (Table 6). In the validation cohort, the respective rates were 21%, 50%, 64%, and 89%. The C statistic for 3-year nursing home placement was 0.66 in the development cohort and 0.63 in the validation cohort.

To our knowledge, this study is the largest conducted to determine the comprehensive predictors of nursing home placement for patients with dementia. Overall, the 3-year Kaplan-Meier estimate of approximately 50% nursing home placement is similar to the rate observed in other studies of patients with dementia.^1,6 In our cohort of 5788 ethnically diverse patients with advanced dementia, we found that patient and caregiver characteristics were independently predictive of nursing home use and that consideration of patient and caregiver characteristics together predicts long-term care placement better than consideration of either patient or caregiver characteristics alone. In addition, we identified several important variables, some of which might be amenable to intervention, that were associated with nursing home placement. We also observed a striking association between ethnicity and risk of nursing home placement. The strengths of our study were the large sample size, the inclusion of different ethnic groups, careful measurement of nursing home placement, validation of our prediction model, and ability to analyze comprehensive predictors of patient and caregiver together.

Our nursing home placement model successfully stratified patients into groups with markedly different rates of nursing home placement. For example, in the validation cohort, rates of nursing home placement at 3 years ranged from 21% in the quartile at lowest risk to 89% in the quartile at highest risk. Further, the model was robust, with only a slight loss of discrimination between the validation and calibration samples. However, our model predicting the decision to pursue nursing home placement was less discriminating than models that have been used to predict more biologically mediated outcomes such as mortality.²¹ Nursing home placement for patients with dementia is affected by much more than the patient's disease severity. It is a decision that also reflects social circumstances, preferences, and values. Without assessing each of these characteristics in detail, models predicting social decisions cannot be as accurate as predictions of biological outcomes. Despite these intrinsic limitations, our prediction model was able to differentiate between patients with diverse risks for nursing home placement from a policy and clinical perspective.

Furthermore, we found that patient and caregiver characteristics are important predictors of time to nursing home placement, demonstrating the importance of considering the patient and caregiver as a unit. This finding suggests that there may be interaction between patient and caregiver risk factors in which they are mutually reinforcing. By using the prediction models, we were able to classify the patient-caregiver units into quartiles of risk and identify large numbers of patients at relatively low risk and extremely high risk of nursing home placement. In fact, among the patients assigned to the highest quartile of risk, nursing home placement is almost certain if the patient lives long enough. If certain high-risk patient-caregiver dyads could be identified, they could be a focus of services or interventions aimed at delaying long-term care. By concentrating on the 7 patient and caregiver characteristics that were associated with the risk of nursing home placement, physicians could have a better estimate of which patient-caregiver dyad is likely to be seeking long-term care and which is not.

To our knowledge, our study is the first to describe large and striking ethnic differences in rates of nursing home placement for patients with dementia, but this association is consistent with several studies that have reported less use of nursing homes by elderly black individuals with all ranges of cognitive status compared with elderly white individuals.^22- 24 Studies of discharge planning have also reported less pursuit of nursing home care among hospitalized elderly black patients compared with their white counterparts.^25- 26 In addition, access to care and knowledge of the health care system could also explain this disparity. However, in our cohort, all the patients and caregivers were part of a demonstration project and most likely had reasonable access to information and care, even in the usual care arm. It is most likely that the influence of ethnicity on predicting nursing home placement is mediated by a variety of social, cultural, and economic factors.

Some of the patient and caregiver variables that were associated with risk of nursing home placement, such as patient's living status, level of functional and cognitive impairment, and caregiver age, have been described previously^3,5,27 and are not readily amenable to interventions. Recently, some pharmacological treatments for Alzheimer disease have been associated with less risk of nursing home placement,^28- 29 and trials of Alzheimer disease agents will increasingly use long-term care placement as an outcome. Furthermore, certain variables that we, as well as others, have observed, such as presence of difficult behaviors^3,5,30 and subjective caregiver burden, may be targets for intervention.

Difficult behaviors are common in dementia, especially in advanced stages, and may be treated with pharmacological and nonpharmacological interventions.³¹ However, often these behaviors are undertreated by physicians, and little education and few behavioral strategies are offered to caregivers. Our finding that caregiver burden is associated with increased risk of nursing home placement for patients with dementia is important because caregiver stress and depression may be prevented or treated.³² A comprehensive support and counseling intervention for spouse-caregivers of patients with Alzheimer disease reduced time to nursing home placement by nearly 1 year compared with those not receiving the intervention.^33- 34 Interestingly, one of the key components of that intervention was to teach the caregivers behavioral management techniques to reduce difficult patient behaviors. Unfortunately, many physicians fail to recognize caregiver depression and stress, and services to reduce caregiver burden are not always accessible. These services most likely would be cost-effective if they could prevent or delay the formal costs of long-term care.

The economic implications of preventing or delaying nursing home placement are substantial. In 1990, the formal cost savings of preventing nursing home placement was estimated to be $30 000 annually per patient with Alzheimer disease,² and Leon et al³⁵ estimated that a 1-month delay in the institutionalization of people with Alzheimer disease could save as much as $1.12 billion annually in the United States.³⁵ This finding suggests that delaying nursing home placement even slightly could produce large health-system cost savings. However, the health-system cost savings associated with delaying nursing home placement needs to be balanced on the societal level with the cost to caregivers of providing unpaid informal care. Our findings that minority ethnic groups are more likely to care for family members with advanced dementia at home rather than place them in a nursing home suggests that these groups are more likely to absorb the costs of informal caregiving. A recent study found that female sex and nonwhite ethnicity were also predictors of reduced employment for the caregivers of frail elderly people,³⁶ suggesting that lower rates of nursing home use by some caregivers may also be associated with economic hardship. The informal demands and the costs associated with caregiving will increasingly become a matter of public concern.

Our study was limited in several ways. We did not conduct clinical evaluations on the patients with dementia but relied on a physician diagnosis of dementia. Thus, we cannot say with certainty what diagnoses the patients had and whether predictors of placement vary by dementia classification. The generalizability of our findings is unclear because we studied patients with fairly advanced dementia who had, along with their caregiver, agreed to be part of a demonstration project, although the 3-year rate of nursing home placement was comparable to that found in previous studies. As a condition for participation in the project, all patients had to be enrolled in or eligible for Medicare Parts A and B, which eliminated consideration of patients who had dementia and were younger, but because of the strong association of dementia with age, most US patients with dementia are enrolled in the Medicare program. Finally, deciding which characteristics were classified as patient or caregiver was by necessity somewhat arbitrary and probably resulted in an overestimate of the importance of patient characteristics and an underestimate of caregiver characteristics. For example, although we classified ethnicity as a patient characteristic, it may be more representative of cultural attitudes toward caregiving.

Combining information about patients with dementia and about their caregivers can better estimate the risk of long-term care placement. Interventions directed at delaying placement, such as reduction of caregiver burden or difficult patient behaviors, could lead to greater patient independence and formal cost savings. Our findings document the need to consider the patient and caregiver as a unit in studies of interventions or nursing home placement.