U.S. Department of Health and Human ServicesU.S. Department of Health and Human Services
This report was prepared under contract between HHS’s Office of Social Services Policy (now the Office of Disability, Aging and Long-Term Care Policy) and the University of Wicsonin. For additional information about this subject, you can visit the DALTCP home page at http://aspe.hhs.gov/_/office_specific/daltcp.cfm or contact the office at HHS/ASPE/DALTCP, Room 424E, H.H. Humphrey Building, 200 Independence Avenue, S.W., Washington, D.C. 20201. The e-mail address is: webmaster.DALTCP@hhs.gov.
This research was begun under the small grants program sponsored by the Institute for Research on Poverty at the University of Wisconsin, Madison, and the U.S. Department of Health and Human Services. It builds on work done as part of the Channeling evaluation under contract number HHS-100-80-0157. The views are those of the authors, not the Institute or the Department.
Concern about inappropriate nursing home placement and rising long-term care costs led to a series of government-financed demonstrations to study whether substituting care at home for care in nursing homes could reduce costs and improve the quality of life for the frail elderly. Based on the evaluations of these demonstrations, we conclude that expanding public financing of community services beyond what already exists under the current system does not reduce costs. Small nursing home cost reductions are more than offset by increased costs of providing expanded community services to those who would not enter nursing homes even without the expanded services. Although they do not reduce costs, expanded community services appear to make people better off. Moreover, the expanded services do not appear to cause substantial reductions in care by family and friends. The research and policy debate should move beyond the question of whether expanded public financing of community care will reduce costs to the problems of how much community care society is willing to pay for, who should receive it, and how it can be delivered efficiently.
The authors are grateful to Don Blackman, John Capitman, Donald Goldstone, Vernon Greene, Mary Harahan, Rosalie Kane, Leslie Saber, and Bill Saunders for their helpful comments on an earlier version of this paper.
The aging of the U.S. population has increased the demand for formal long-term care services; associated with that demand are increases in public and private costs. The existing formal long-term care system, it has been argued, favors nursing home over community care for two reasons. First, the broad range of personal care, housekeeping, meals, transportation, and other services needed by an impaired person to live in the community are difficult to identify and coordinate, leading to unnecessary decisions to enter nursing homes. Second, for individuals with chronic care needs, public programs pay for nursing home care but typically do not pay for long-term care such as personal care and homemakers in the community (Morris 1971; Congressional Budget Office 1977; Mechanic 1979; Kane and Kane 1980). These arguments led to a series of demonstrations of expanded government financing for case management and community services, beginning in the 1970s. This paper reviews the results of these demonstrations--their effects on nursing home use, public and private costs and the quality of clients' lives--and assesses the implications of the results.
Although the question of the substitution of community care for nursing home care is currently receiving considerable attention, this issue is not new. Framers of the English Poor Law in 1601, the first major codification of Anglo-American social welfare laws, allowed for both indoor relief (within the almshouse) and outdoor relief (in one's home) (Woodroofe 1971). In 1821 Josiah Quincy's "Report to the Committee Who Was Referred in Consideration of the Pauper Laws of this Commonwealth" compared care in the almshouse to care in one's home (Quincy 1852). He found that only those desperately in need of care chose the almshouse, because of poor conditions there. He recommended that the government finance only care in the almshouse--reasoning that total public costs would be lower because, given the poor conditions, fewer people would use this form of relief.
Research on community care has had two strands. The first analyzed the substitution of visiting nurses for acute care in hospitals. The second analyzed the substitution of a broader range of community care, including personal care and homemaking services, for chronic care in nursing homes.
Research focused first on hospital care. The hypothesis was that by having nursing care at home, patients could leave the hospital sooner. The cost savings from the reduced length of hospital stays would more than offset the costs of the visiting nurses. Two methodologies were used to test this hypothesis. First, the hypothetical cost of nursing care at home was compared to the cost of hospital care. Studies ranged from those based on physicians' judgments about the number of hospital days that might be saved if visiting nursing care were available for a sample of their patients (e.g., Scutchfield and Freeborn 1971) to those based on retrospective matching of patients who received home health care with those who received hospital care (e.g., Bryant, Candland, and Lowenstein 1974). Second, a series of randomized experiments assigned hospital patients to two groups, one with home health care available after the hospitalization and the other without (e.g., Bakst and Marra 1955; Katz et al,1968; Stone, Patterson, and Felson 1968; Gerson and Collins 1976). There were many studies of both types, and they generally concluded that the total costs of acute care could be reduced by expanding home health benefits. (See Hammond 1979 and Hedrick and Inui 1986 for reviews of the literature.)
Gradually the focus of attention shifted to the question of whether home care (including nonmedical services such as personal care and homemakers as well as nursing) substitutes for nursing home care. The distinction between home health care as a substitute for hospital care and home care as a substitute for nursing home care was not clearly drawn, perhaps because of the inherent substitutability of care at home for both hospital and nursing home care. Nonetheless, the framework that implicitly lay behind the studies of nursing homes differed from that of the hospital studies. The nursing home studies asked whether long-term care for chronic disabilities would be less costly if the disabled person received care at home rather than moving permanently to a nursing home. The hypothesis was that costs at home would be below nursing home costs for those with low levels of disability because they do not need the level of care provided in nursing homes. As the patient's level of disability increases, the cost of care in the community increases up to a break-even point, beyond which community care costs more than nursing home care. Pollak (1973) formalized this framework, which was implicit in many of the studies.
Studies of community care as a substitute for nursing home care used both hypothetical cost comparison and experimental methodologies. Among the hypothetical comparisons, Bell (1973), Greenberg (1974), Rathbone-McCuan and Lohn (1975), Brickner and Scharer (1977), General Accounting Office (1977), Piland (1978), and Arkansas Office on Aging (1981) compared the cost of home care for a sample of impaired clients in the community to what the cost would have been in a nursing home. Anderson, Patten, and Greenberg (1980) compared actual costs for a sample of nursing home patients to actual costs for a sample in the community with similar levels of disability. The results of these hypothetical cost comparisons were consistent with expectations: Community care was less costly than nursing home care, except for those with very high levels of disability.1
The hypothetical cost comparisons demonstrated that many impaired elders, including some who reside in nursing homes, can be cared for in the community at lower cost than in a nursing home. Such comparisons together with research suggesting that from 10 to 40 percent of those in nursing homes were inappropriately placed there (Morris 1971, Williams et al. 1973, Congressional Budget Office 1977, General Accounting Office 1979) -- served as the basis for the argument that public financing of community care should be expanded to reduce unnecessary nursing home use and thereby reduce public costs and improve the quality of clients' lives.
That community care is less costly than nursing home care for many individuals does not, however, necessarily imply that expanded financing of community care will reduce aggregate costs. The hypothetical cost comparisons implicitly assumed that every individual who received community care would be in a nursing home in the absence of community care. Actual experiments with expanded financing of home care were needed to determine the effect on aggregate expenditures. Without experiments, it was impossible to determine the extent to which the expanded public financing would go to those who would not enter a nursing home even without expanded community services.
The early experiments with community care as a substitute for nursing home care tested limited expansions of services: case workers (Goldberg 1970), protective service case workers and home health aides (Blenkner, Bloom, and Nielsen 1971), monitoring visits by nurses (Katz et al. 1972), and personal care, housekeeping, and escort services (Nielsen et al. 1972). All employed random assignment to treatment and control groups, but the samples were relatively small (100 to 300). Katz et al. and Nielsen et al. found statistically significant reductions in nursing home use (although the differences found by Katz et al. were significant only for important subgroups, not overall). Blenkner, Bloom, and Nielsen reported an unexpected increase in nursing home placement, but it was not statistically significant. These early field trials also found some evidence of effects on other outcomes. Nielsen et al. reported increased contentment, Goldberg reported increased social activities, and Blenkner, Bloom, and Nielsen reported decreased stress among informal caregivers.
Because the use of home health care under Medicare and Medicaid had not grown to present levels at the time of these field trials, their results may not be useful in assessing current policies. These studies did, however, demonstrate that field tests could be successfully undertaken with rigorous evaluation designs, thus laying the foundation for community care demonstrations in the 1970's and 1980's.
In this paper, we review the demonstrations from this period that provided case-managed community care to impaired elderly populations and were funded through special waivers of certain Medicaid or Medicare regulations.2 The 16 such demonstrations are listed together with their variants when more than one model was tested:
Sources for our review are preceded by an asterisk in the reference list. When the demonstration had a final evaluation report, we generally relied on it as the primary source. In four cases, we relied on a cross-cutting evaluation by Haskins et al. (1985) which was the primary evaluation of these four demonstrations.
Although we have limited our review to the demonstrations funded through Medicare and Medicaid waivers, other studies have also have been undertaken during this period (see for example, Papsidero et al. 1979, Hughes 1981, and Groth-Juncker et al. 1983).3 Their findings are consistent with those reported for the demonstrations reviewed here. (See Hedrick and Inui 1986 for a review.) In addition, numerous states have undertaken community care initiatives (Greenberg, Schmitz, and Lakin 1983, and Health Care Financing Administration 1984). Typically the state initiatives have been designed as service programs rather than experiments, and they have therefore not been accompanied by rigorous evaluations.
Our review builds on a number of previous reviews (Applied Management Sciences 1976, LaVor and Callender 1976, Doherty, Segal, and Hicks 1978, Greenberg et al. 1980, Steiner and Needleman 1981, Stassen and Holahan 1980, Toff 1981, General Accounting Office 1982, Zawadski 1984, Palmer 1984, Hughes 1985, Kotler et al. 1985, Capitman, Haskins, and Bernstein 1986, Capitman 1986, and Hedrick and Inui 1986). The previous reviews have not systematically compared quantitative results and do not include the most recent demonstrations. The present paper attempts to fill those gaps.
Large-scale community care demonstrations began in 1973 with the Worcester Home Care demonstration and continued to completion of the South Carolina CLTC and Channeling demonstrations in 1984 (see Figure 1). Despite the 12-year time span and varied programmatic and research designs, all demonstrations shared the objective of substituting community care for nursing home care wherever appropriate. Meeting this objective was expected to reduce long-term care costs and improve the quality of clients' lives. Most of the demonstrations focused on the elderly.
 |
Case management and an expanded package of community services (typically including personal care and homemaker services) were the key program elements of these demonstrations. Although there had been some debate about whether an ongoing case management role was needed after services were arranged, only the NCHSR Day Care/Homemaker experiment did not include ongoing case management in its intervention. Because it did not provide a broad range of services (one model provided adult day care; the second, homemaker services; and the third, adult day care and homemaker services), ongoing case management to coordinate of services was not essential to the intervention. Most demonstrations employed individual case managers to be responsible for assessment of needs, design of a care plan, arrangement of services, and ongoing monitoring, but four (Triage, OPEN, South Carolina CLTC, and MSSP) used teams made up of professionals from different disciplines. The intensity of case management varied. Triage had the highest average caseload: 125 cases per case manager. Caseloads of the other demonstrations reporting this information ranged from 45 to 80 clients.
Expanded community services were funded through waivers of Medicare or Medicaid regulations4 to permit payment for services not normally covered (e.g., homemaker services), in situations not normally covered (e.g., personal care without a need for skilled nursing care), or to individuals not normally eligible (e.g., those who would be eligible for Medicaid if in a nursing home but not in the community). These waivers made it possible to pay for a broader range of community care services over a longer period to different types of people than is typically possible under Medicaid and Medicare. (The only exception to funding through waivers was for the Basic Model of the Channeling demonstration, which had only limited funds to pay for services to fill in the gaps in the existing system. These services were funded directly through demonstration contracts rather than through waivers. The Financial Model of Channeling paid for the full range of community services through waivers of Medicaid and Medicare regulations.)
The demonstrations all expanded in-home service coverage to include nonmedical services such as homemaking and personal care. The specific services (and the specific waivers enabling their coverage) varied across the demonstrations. They include new services not covered at all under Medicaid or Medicare and services already covered but for which limitations on coverage were modified (see appendix Table A.1). (Extension of coverage to additional individuals is discussed under eligibility criteria be low.)
All demonstrations except the NCHSR Day Care Model covered homemaker or personal care -- the services most needed by chronic care patients at home. Other services often covered included chore, companion, escort services, transportation, and home-delivered meals. Many demonstrations also covered one or more other community services that are believed to be important for some clients in the community -- adult day care, foster care, housing improvements, respite care, medical equipment, mental health counseling, prescription drugs, etc. Most could pay for nurses and home health aides in circumstances not normally covered by Medicare and Medicaid. The demonstrations generally did not cover acute medical care. There was one important exception. On Lok covered physician, hospital, and nursing home care, laying the foundation for the social health maintenance organizations now being tested. (Triage paid clients' deductibles and coinsurance under Medicare.)
Another program element that varied across the demonstrations is service authorization power. The extent to which case managers can authorize payment for the full package of community care determines whether they can control service delivery or must act as brokers and advocates for their clients, coordinating care paid for by other agencies. (Authorization power is not always clearly described in the evaluation reports, but Table A.2 summarizes our understanding of case managers' authorization power.) Typically the demonstrations had power to authorize only expanded services (the breadth of which varied of course), the intent being to rely on existing programs before using demonstration funds. But in several cases the demonstration's authority was extended to include Medicaid or Medicare home health care or services funded under Title XX of the Social Security Act: Washington CBC (Title XX), ACCESS (Medicaid home health), On Lok (Medicare home health), Nursing Home without Walls (Medicaid home health), and Financial Model Channeling (Medicare and Medicaid home health through a pooling of funds). As indicated, On Lok could also authorize Medicare physician, hospital, and nursing home care. Finally, ACCESS and South Carolina CLTC could withhold authorization for Medicaid nursing home payment for clients not satisfying the eligibility requirements -- that is, they were preadmission screening units for nursing homes under Medicaid. Thus in all but three demonstrations, authorization power was limited essentially to community services, and in most of these it was further limited to services whose coverage was expanded under waivers.
In expanding government financing for community services, there was some concern that total costs might increase. In an effort to control costs, seven of the demonstrations (ACCESS, Georgia AHS, Wisconsin CCO, MSSP, South Carolina CLTC, Nursing Home without Walls, and Channeling) implemented limits on the amount that could be spent on community services for each individual. These cost "caps” ranged from 60 to 85 percent of nursing home reimbursement rates (see Table A.2). Most demonstrations reported procedures to allow for temporarily high cost clients to be served above the cap. Financial Model Channeling also imposed a cap on expenditures for the caseload as a whole -- average expenditures had to be less than 60 percent of the nursing home rate.
A second cost control element, client cost sharing, was implemented by three demonstrations (ACCESS, South Carolina CLTC, and Channeling). Clients with incomes above a specified dollar amount were required to contribute to the cost of services purchased by the demonstration. Because the incomes of clients were typically quite low, the extent of cost sharing turned out to be quite small.
Ten of the demonstrations were directed toward the elderly (minimum ages ranged from 50 to 65); one had no minimum age; and the other five served the adult disabled population as well as the elderly (see Table A.3).
All of the demonstrations required that clients be eligible for an existing program, usually the program (Medicare or Medicaid) under whose waivers services were funded. (One demonstration required Title XX as well as Medicaid eligibility.) Requiring Medicaid eligibility ensures that the demonstration serves a low-income population. Restricting the program to the Medicaid-eligible does, however, exclude the "spend-down" population -- those who enter nursing homes as private pay patients but use up their assets over time and become eligible for Medicaid -- which account for about half of those whose nursing home costs are ultimately covered by Medicaid.
The demonstrations sought to serve those at risk of nursing home placement and developed specific eligibility criteria to identify them. The only exception was the first phase of the Triage demonstration, although its second phase implemented disability requirements. The other demonstrations used three different approaches. Eight (Worcester Home Care, NCHSR Day Care/ Homemaker, Washington CBC, OPEN, MSSP, New York City Home Care, Florida Pentastar, and San Diego LTC) required that a client have a service need, but they did not have specific formalized disability criteria (see Table A.3). Five (Georgia AHS, Wisconsin CCO, On Lok, Nursing Home without Walls, and Channeling) imposed specific disability requirements. Finally, ACCESS and South Carolina CLTC, as indicated, identified clients as part of the nursing home preadmission screening process. To be eligible for these demonstrations clients had to satisfy Medicaid requirements for nursing home admission.
The targeting approach determined the frailty of the populations served. Table 1 presents disability measures for the demonstrations grouped by their approach to targeting. Although not a perfect correlation, the frailty of the clients increases with the stringency of the disability requirements.5 At one extreme Triage had neither need nor disability criteria, and 54 percent of its clients turned out to have at least one disability in ADL's. At the other extreme, South Carolina CLTC, relied on preadmission screening, and 95 percent of its clients turned out to have at least one ADL disability.
| TABLE 1: Client Disability at Enrollment | ||||
|---|---|---|---|---|
| Demonstration (evaluation period) | Disabled on at Least One ADL (percentage) | Impaired on at Least One IADL (percentage) | Incontinent (percentage) | Cognitive Impairmenta |
| NO NEED/FUNCTIONING CRITERIA | ||||
| Triageb (1976-1979) | 54 | 94 | --- | 1.7 |
| SERVICE NEED CRITERIAc | ||||
| Worcester Home Care (1973-1975) | 41 | --- | --- | --- |
| NCHSR Day Care/ Homemakerd (1975-1977) | 77 | --- | --- | --- |
| Project OPENb (1980-1983) | 50 | 81 | 24 | 0.6 |
| MSSPb (1980-1983) | 61 | 80 | 47 | 1.7 |
| New York City Home Careb (1980-1983) | 78 | 100 | 38 | 2.6 |
| Florida Pentastarb (1981-1983) | 58 | 97 | 22 | 1.4 |
| San Diego LTCb (1981-1983) | 55 | 97 | 43 | 2.3 |
| FUNCTIONING CRITERIA | ||||
| Georgia AHS (1977-1980) | 60 | --- | --- | 3.1 |
| Wisconsin CCOb (1978-1980) | 62 | 97 | --- | --- |
| On Lokb (1979-1983) | 85 | 93 | 60 | 3.2 |
| Nursing Home without Walls (1980-1983) | 76 | --- | --- | --- |
| Channeling (1982-1984) | 84 | 100 | 55 | 3.5 |
| PRE-ADMISSION SCREEN | ||||
| ACCESSb (1977-1980) | 82 | 99 | 44 | 2.4 |
| South Carolina LTCb (1980-1984) | 95 | 97 | 58 | 3.5 |
|
||||
Whether an evaluation provides sound estimates of the true effects of expanded community care depends on many features of the evaluation design, the most important of which is the comparison methodology. After discussing comparison methodologies, we review other issues that affect interpretation: the sample size, the number of replications of the intervention, and the extent of data available to measure outcomes.
To measure the effect of expanded community care, it is essential to be able to contrast the experiences of the persons to whom the expanded services were available -- the treatment group -- with some measure of what the experiences of the same persons would have been if they had not had the service opportunities provided by the demonstration. Every evaluation developed a methodology intended to estimate the effects of the demonstration based on such a comparison. The strength of the comparison methodologies, however, differed substantially across the demonstrations. We have classified the methodologies into three categories based on the likelihood of biased estimates of demonstration effects.
The strongest comparison methodologies are the experiments that randomly assigned eligible applicants either to receive the demonstration services (treatment status) or to receive only those services regularly available in the community (control status). Random assignment is a powerful design, because it ensures that, for a large sample, the treatment group will be similar to the control group on both measured and unmeasured characteristics. Evaluations that use randomized designs are most likely to obtain unbiased estimates of demonstration effects. Nine demonstrations are in this first group (see Table 2).
The second group -- comprising five evaluations were quasi experiments. To represent what would have happened in the absence of the demonstration, they selected comparison groups intended to be similar to the treatment group, but did not use random assignment. Three selected comparison groups entirely from outside the demonstration catchment area; two drew comparison-group members partly from within and partly from outside the area.
A comparison group methodology that does not use randomization is generally a weaker design because the comparison group may differ from the treatment group on measured and unmeasured characteristics. Nonparticipants within the catchment area are likely to differ in unknown ways from those who choose to apply or are referred to the program; a comparison group outside the catchment area faces a different service environment. The comparison group is inevitably selected in a different way from the treatment group. Potential clients typically apply to the program or are referred by a provider and then are subjected to an eligibility determination process. The treatment group is thus self- and program-selected in ways that are only partially known. Comparison groups, in contrast, must be chosen from existing lists (e.g., Medicaid rolls) based on available measured characteristics. Selecting a group that is similar to the treatment group is particularly difficult for community care demonstrations because the central outcome of interest --nursing home use -- is extremely difficult to predict. The overwhelming finding of existing research is that measured characteristics explain little of the variation in nursing home placement rates (see Grannemann et al. 1986 for a review of the literature).
Moreover, as it turned out, even the measured characteristics of the treatment and comparison groups differed for all the quasi experiments, in some cases substantially. Thus, evaluations constrained to use a comparison-group methodology rather than a randomized experiment run a high risk of inaccurately representing what would have happened to participants in the absence of the demonstration.6
Although evidence on the bias of quasi-experimental methodologies as applied to community care does not exist, recent research on evaluation of employment and training programs is not encouraging. LaLonde (1986) and Fraker and Maynard (forthcoming) compared the actual results of an actual randomized experiment with simulations of quasi-experimental results for the same demonstration using various comparison groups and estimation methodologies. They found that the results using the quasi-experimental methodologies differed substantially from the actual results of the randomized experiment. LaLonde (1986, p. 617) concludes that "policymakers should be aware that the available nonexperimental evaluations of employment and training programs may contain large and unknown biases." A similar caution applies to nonexperimental evaluations of community care demonstrations.
The third group of studies includes two (ACCESS and Washington CBC) that were countywide interventions which compared aggregate Medicaid cost and nursing home use in the demonstration counties with the corresponding estimates for a set of comparison counties. Because many factors other than the demonstrations affect nursing home use and costs, the results of these evaluations are subject to considerable uncertainty. For example, if a certificate of need for construction of new nursing home beds had been granted in the comparison county, then the newly constructed beds are likely to cause an increase in nursing home use in the comparison county. When growth in aggregate nursing home use-in the treatment county is compared to that in the comparison county, growth in nursing home use will be lower in the treatment county. It is not clear, however, whether this should be attributed to the effect of the demonstration or to the nursing home bed construction policy in the comparison county. Because of the difficulty of distinguishing treatment effects from other factors affecting service use and costs, comparisons using aggregate data for a small number of counties seldom provide persuasive evidence about program effects.
| TABLE 2: Evaluation Methodologies | ||||
|---|---|---|---|---|
| Demonstration (evaluation period) | Comparison Methodology | Number of States | Number of Sites | Sample Size |
| Worcester Home Care (1973-1975) | Random assignment | 1 | 1 | 485 |
| NCHSR Day Care/ Homemaker (1975-1977) | Random assignment | 4 | 6 | 1,566 |
| Triage (1976-1979) | Comparison group outside area | 1 | 1 | 502 |
| Washington CBC (1976-1977 | County-level comparison | 1 | 2 | --- |
| ACCESS (1977-1980) | County-level comparison | 1 | 1 | --- |
| Georgia AHS (1977-1980) | Random assignment | 1 | 1 | 1,332 |
| Wisconsin CCO (1978-1980) | Random assignment | 1 | 1a | 417 |
| On Lok (1979-1983) | Comparison group outside area | 1 | 1 | 139b |
| Project OPEN (1980-1983) | Random assignment | 1 | 1 | 335 |
| MSSP (1980-1983) | Comparison group within and outside area | 1 | 8 | 4,200 |
| South Carolina LTC (1980-1984) | Random assignment | 1 | 1 | 1,867 |
| Nursing Home without Walls (1980-1983) | Comparison group within and outside area | 1 | 9 | 1,373 |
| New York City Home Care (1980-1983) | Comparison group outside area | 1 | 1 | 704 |
| Florida Pentastar (1981-1983) | Random assignment (plus comparison group outside area) | 1 | 5 | 1,046 |
| San Diego LTC (1981-1983) | Random assignment | 1 | 1 | 831c |
| Channeling (1982-1984) | Random assignment | 10d | 10 | 6,326 |
|
||||
In the remainder of this article, we place primary emphasis on the randomized experiments. Results of quasi experiments are reported in the tables but do not figure heavily in our assessment of the findings. Finally, the two studies using aggregate data are included only in the discussion of nursing home use, which was the primary outcome they analyzed.
In interpreting the results, keep the nature of the comparison in mind. The demonstrations compared expanded case-managed community care to the existing long-term care system. Although little information is available on how much, the existing service system paid for some community care under Medicare, Medicaid, and other government programs. Thus, the demonstrations evaluated the expansion of community care beyond what already existed, not community care versus its total absence. Moreover, some of the demonstrations were undoubtedly tested in environments where nursing home bed supply was constrained by restrictions on reimbursement rates and construction of new beds. Unfortunately, we do not know the extent to which bed supply was restricted, and it is difficult to speculate on how the effect of community care might differ in different service environments.
The sample size determines the evaluation's ability to detect effects if they exist. If samples are small, estimates of program effects are subject to greater sampling error, and the absence of a statistically significant measured effect may not correctly indicate the absence of a true effect.
The sample sizes of the community care demonstrations vary widely. The smallest used a sample of only 139 people. Four of the studies had sample sizes between 300 and 600. The largest sample size was just over 6,300. Differences in sampling error therefore varied considerably across the evaluations.
The number of replications of the intervention and the diversity of service environments in which it is tested determine the ability to generalize from the demonstration results. Limited site selection increases the risk that observed results are due to special features of the particular implementation or the environment.
The number of replications was generally quite limited. Of the nine randomized experiments, only two (NCHSR Day Care/Homemaker and Channeling) were tested in communities in more than one state. Several were tested in more than one site within a state.
The amount and quality of the information collected determines the range of effects that can be measured. Five potential sources of data are available to demonstrations of this kind: individual interviews with treatment and control (or comparison) groups,7 demonstration project records (for clients only), public program records such as Medicare and Medicaid claims, provider records, and official death records.
The demonstrations varied in the range of data sources they were able to use (see Table A.4). One was limited to a single data source (aggregate county social service department data). One relied on individual interviews, supplemented by project and death records. Seven combined individual interviews with records data from Medicare, Medicaid, or project records, but did not collect both Medicaid and Medicare data. (One of these also collected death records.) Seven used individual interviews and both Medicaid and Medicare records. (One of these also collected from service providers data on the use and cost of services not covered by Medicare, Medicaid, or the project; interviewed the primary informal caregivers of a subsample of the treatment and control groups; and obtained official death records.)
The breadth of the data collection affects primarily the service use and cost outcomes. Evaluations limited to Medicare and project records, for example, will miss effects on most nursing home use, which is paid for by Medicaid or private individuals.
Measures of quality of life are obtained from individual interviews. They are subject to potential bias due to data noncomparability.8 For several of the evaluations, program staff conducted treatment group follow-up interviews, but research staff conducted control group follow-up interviews. Because of their different orientations, the data that the two types of interviewers collect may differ, introducing measurement bias into the treatment-control comparisons.9 Among randomized experiments six evaluations (Worcester Home Care, NCHSR Day Care/Homemaker, Georgia AHS, South Carolina CLTC, San Diego LTC, and Channeling) used comparable follow-up data collection for the two groups; three (Wisconsin CCO, OPEN, and Florida Pentastar) were at risk of measurement bias because of differential data collection for the treatment and control groups.
The length of the follow-up data collection period determines the capability of detecting long-term program effects. All the evaluations collected follow-up data for at least 12 months; two followed at least a subsample for 18 months; and four followed at least a subsample for two years or 9 longer.10
All the demonstrations sought to substitute community care for nursing home care, and all evaluations examined this outcome. As indicated, the data sources determine the comprehensiveness of the nursing home use measures. Medicaid and private individuals are the major payors for nursing homes. Evaluations that relied on Medicare records alone measured only a fraction of nursing home use. When clients were required to be Medicaid eligible, Medicaid and Medicare records capture nearly all nursing home use, when clients were not required to be eligible for Medicaid, however, Medicaid and Medicare records without provider records or individual interviews do not capture use paid for by private individuals. The omission of use paid for privately is potentially important because many enter nursing homes as private pay patients, only later spending down their assets to the point of Medicaid eligibility. Indeed, about half of those in nursing homes under Medicaid enter as private pay patients. For this reason, much of the effect of demonstrations that were not limited to the Medicaid-eligible could reflect nursing home use paid for privately. In presenting the results in Table 3,11 therefore, we distinguish among studies that measure essentially all use and those with partial measures of nursing home use.
Overall effects. Of the six studies that used randomized designs and had essentially complete data on nursing home use, five tell a consistent story. For Worcester Home Care, Georgia AHS, Wisconsin CCO,12 Florida Pentastar, and Channeling, treatment group nursing home use was equal to or less than the control group use, but the differences were small ranging from zero to eight days during the year after enrollment -- and not statistically significant.13 (The Florida Pentastar evaluation did not measure nursing home days; however, the percentage who had entered a nursing home by 18 months was slightly smaller for the treatment group than for the control group -- 7.6 vs. 8.5 percent, a difference that was also not statistically significant.)
In all of these cases, the populations served turned out to be at relatively low risk of nursing home placement, precluding large reductions in nursing home use. The control group experience measures the risk of nursing home use in the absence of expanded services. The control groups spent between 26 and 46 days in nursing homes during the first year after enrollment. Even if these demonstrations had cut nursing home use by 50 percent, the number of nursing home days saved would have been modest. Moreover, actual reductions were well below 50 percent, ranging from 0 to 24 percent.
| TABLE 3: Nursing Home Days during the First Year | ||||||
|---|---|---|---|---|---|---|
| Demonstration (source) | Data Source | Treatment Mean | Control Mean | Difference in Means | Percentage Difference | |
| RANDOMIZED EXPERIMENTS, ALL USE | ||||||
| Worcester Home Carea (Sherwood, Morris, and Gutkin, 1975, pp. 25, 38) | Interviews† | 46 | 46 | 0 | 0.0 | |
| Georgia AHS (Skellie et al., 1982, pp. 171-172) | Medicaid records†
Medicare records |
22 | 29 | -7 | -24.1 | |
| South Carolina LTC (Brown et al., 1985, p. 98) | Medicaid records†
Medicare records |
90 | 130 | -40* | -30.8 | |
| Florida Pentastar (Maurer et al., 1984, p. 84) | Interviews† | --- | --- | --- | -10.6b | |
| Channelingc (Wooldridge and Schore, 1986, pp. 92-93) | Medicaid
records Medicare records Provider records |
|||||
| Basic Model | 25 | 29 | -4 | -13.8 | ||
| Financial Model | 23 | 26 | -3 | -11.5 | ||
| RANDOMIZED EXPERIMENT, MEDICAID USE ONLY | ||||||
| Wisconsin CCOd (Seidl et al., 1980, p. 206) | Medicaid records† | 25 | 33 | -8 | -24.2 | |
| RANDOMIZED EXPERIMENTS, MEDICARE USE ONLY | ||||||
| NCHSR Day Care/Homemaker (Weissert, Wan and Livieratos, 1980, pp. 44, 46, 48) | Medicare records | |||||
| Day care | 5 | 7 | -2 | -28.6 | ||
| Homemaker | 3 | 4 | -1 | -25.0 | ||
| Combined | 4 | 5 | -1 | -20.0 | ||
| Project OPENe (Haskins et al., 1985, Appendix A, p. 119) | Medicare records | 0.1 | 0.3 | -0.2 | -66.7 | |
| San Diego LTCe (Haskins et al., 1985, Appendix A. pp. 100, 102) | Medicare records | 0.5 | 0.9 | -0.4 | -44.4 | |
| QUASI EXPERIMENTS, ALL USE | ||||||
| Triage (Shealy, Hicks, and Quinn, 1979, pp. 449, 450) | Interviews/diaries | 7 | 4 | 3f | 75.0 | |
| On Loke (Haskins et al., 1985, Appendix A, p. 102) | Interviews/diaries | 20 | 117 | -97* | -82.9 | |
| MSSP (Miller et al., 1984, pp. 1-24, 1-70) | Medicaid records†
Medicare records |
39 | 45 | -6g | -13.3 | |
| QUASI EXPERIMENT, MEDICAID AND MEDICARE USE ONLY | ||||||
| Nursing Home without Wallsh (Birnbaum et al., 1984, p. IV-3) | Medicaid
records Medicare records |
|||||
| Upstate | 6 | 99 | -93* | -93.9 | ||
| New York City | 5 | 40 | -35* | -87.5 | ||
| QUASI EXPERIMENT, MEDICARE USE ONLY | ||||||
| New York City Home Caree (Haskins et al., 1985, Appendix A, p. 103) | Medicare recordsi | 0.2 | 1.1 | -0.9 | -81.8 | |
NOTE: Estimates for the full sample
wherever possible. For Project OPEN, San Diego LTC, On Lok, Nursing Home
Without Walls, and New York City Home Care, estimates are for survivors only.
† All clients were required to be eligible for Medicaid. |
||||||
The possibility of delayed reductions in nursing home use is not addressed by these data, which are for only one year. As indicated (see Table A.4), the length of follow up was generally short. Three of these five experiments had more than one year of follow-up data: Georgia AHS (two years), Florida Pentastar (18 months), and Channeling (18 months). None of the three found significant differences after the first year.
Although direct evidence concerning effects more than a year or two after enrollment was not collected in these five experiments, the likelihood of a long-run reduction is low. It would require a large increase in control group nursing home use over time or a large increase in the percentage reduction in use, and probably both. The limited time-trend evidence (not shown) suggests that neither occurred.
The South Carolina CLTC project stands in contrast to the findings of the other five randomized experiments with essentially complete nursing home use data. It reported high control group use (130 days during the first year after enrollment) and a large, statistically significant reduction (40 days). Moreover, longer follow-up data indicate that the reduction persists for at least three years.
The distinguishing feature of the South Carolina CLTC demonstration appears to have been its integration with a nursing home preadmission screen. All clients came through this screen and were eligible for nursing home admission under Medicaid. They were among the most disabled of any of the demonstrations (see Table 1) and were at greatest risk of nursing home use. By identifying clients at the nursing home door and requiring nursing home eligibility under Medicaid, the South Carolina CLTC demonstration appears to have identified the intended target population and reduced its nursing home use.14 In addition to its success at identifying a high-risk population, South Carolina CLTC's reduction was higher in relative terms as well (31 percent compared with 0-24 percent for the other five demonstrations).
The three randomized experiments that only measured nursing home use under Medicare also found treatment group use below control group use. As indicated, however, Medicare claims capture only a small fraction of all nursing home use. Although larger in relative terms than the differences for the randomized experiments that measured essentially all use, none of the differences for the randomized experiments that measured only Medicare use was statistically significant, and the magnitudes of the measured differences were small (two days or less).
The quasi experiments varied widely in their findings. Two (Triage and New York City Home Care) reported small, nonsignificant increases, one (MSSP) reported a 6-day decrease (statistical significance not reported), and two (On Lok and Nursing Home without Walls) reported large, statistically significant reductions. As indicated above, however, the inherent difficulty in choosing a comparison group similar to the treatment group on both measured and unmeasured characteristics substantially reduces the confidence that can be placed in the results from the quasi experiments.
Finally, the two evaluations that used county-level comparisons both reported reductions in nursing home use relative to the comparison counties (not shown). The Washington CBC project reported annual declines in Medicaid nursing home populations of -3.0 and -4.5 percent in the two demonstration sites compared to -0.6 in the balance of the state (Solem et al., 1979, p. 66). ACCESS (the only other demonstration to rely on preadmission screening) reported that Medicaid nursing home expenditures rose 5.7 percent in the demonstration site compared to 23.1 percent in the six comparison counties (Price and Ripp, 1980, p. VII.17). Whether these (relative) reductions in Medicaid nursing home use were due to the expanded community care or other factors cannot be determined.
The weight of the evidence is that expanding community care beyond what already is provided under existing programs reduces nursing home use, but the magnitude of the reduction is small. The only apparent exception is an expansion of community care combined with nursing home preadmission screening.
Differences across subgroups. The general failure of the demonstrations to identify client populations at high risk of nursing home placement raises the question of whether there are identifiable groups for which expanded community care substitutes for nursing home care. Several of the evaluations analyzed differences in effects across subgroups of the populations, permitting identification of groups for which reductions in nursing home use were greatest.
The evaluations differed in the subgroup variables analyzed. We were able to classify variables as measures of disability, living arrangement/availability of informal supports, Medicaid eligibility, or risk of nursing home placement. (Definitions of subgroups and results are summarized in appendix Table A.5.) Because subgroup samples are small (increasing the risk of false negative tests) and because a large number of tests were conducted (increasing the risk of false positive tests), the subgroup results must be interpreted cautiously.15
Disability and variables closely related to disability were the subgroup variables most often analyzed. Measures used to define subgroups were ability to perform activities of daily living (ADL), whether certified eligible for skilled nursing facility (SNF) or intermediate care facility (ICF) care, recommended services, continence, and cognitive impairment. Among randomized experiments that measured effects on all nursing home use, differences across disability subgroups appeared in about half the analyses. Larger reductions were generally associated with greater disability, although in two cases this relationship did not hold for the most extreme level of disability. In this regard, it is worth noting that some of the South Carolina CLTC demonstration sample were not eligible for nursing home placement. Although they were not part of the basic analysis, a separate analysis of this group found low control group nursing home use and no reduction in use. (All but one of the other evaluations found either no differences in effects across disability levels or, consistent with the randomized experiments with complete data, larger reductions for the more disabled.)
Although evidence on differences in effects with measures of living arrangement/informal support is limited, it is consistent with the hypothesis that reductions in nursing home use are greater among those with limited informal support. Among the randomized experiments that measured all nursing home use, the reductions of the South Carolina CLTC were larger for the group receiving only formal care at enrollment than for the groups receiving informal care (either alone or in combination with formal care). (Reductions were, however, smaller among the group without any formal or informal care.) Financial Model Channeling reduced nursing home use during the first six months among those who lived alone but not among those living with others. (Among the other evaluations, three -- NCHSR Day Care/Homemaker, San Diego LTC, and the upstate site of Nursing Home without Walls-- found no differences in effects with living arrangement; one, the New York City site of Nursing Home without Wall found larger reductions in nursing home use for those living alone and over age 75.)
Those in a nursing home at enrollment (a small group) had significant reductions in nursing home use under both Channeling models. It is important to note, however, that Channeling clients had to be certified as able to be discharged within three months, so these were not typical nursing home patients. Although the Channeling result is rather strong, supporting evidence from other studies is lacking among the randomized experiments. (On Lok's reported reduction in nursing home use was also greater among those in a nursing home at enrollment.)
Three randomized experiments with complete data examined differences in effects associated with Medicaid eligibility. (These subgroups are of interest because Medicaid will cover nursing home care for persons with higher incomes than that required for eligibility in the community.) Results were inconsistent: no differences for Worcester Home Care and Financial Model Channeling; increased nursing home use among the Medicaid-eligible (apparently those on nursing home waiting lists), but decreased use among those with higher incomes under Basic Model Channeling; and a larger decrease in nursing home use among those who were Medicaid-eligible in the community than among those with higher incomes under South Carolina CLTC. Contrary to expectations, nursing home reductions do not appear to be greater among those with incomes above the level required for Medicaid eligibility in the community.
Summary measures of the risk of nursing home placement defined subgroups in two randomized experiments, providing limited support for the hypothesis that nursing home reductions are larger for those predicted to be at higher risk of nursing home placement. Channeling defined risk groups based on a multiple regression model of the determinants of nursing home admission. Basic Model Channeling reduced nursing home use among the high-risk group but not the extreme-risk group; Financial Model Channeling exhibited a pattern of larger reductions with increasing risk (including the extreme-risk group). Worcester Home Care relied on interviewer judgments about risks and found no differences. (Two quasi experiments that analyzed differences with regression model predictions of risk of nursing home placement-- MSSP and Nursing Home without Walls-- found larger nursing home reductions among those at higher risk.)
Other subgroup variables were analyzed by only one evaluation or exhibited inconsistent results: whether on a nursing home waiting list, hospitalization, unmet needs, prognosis, age, gender, race, diagnosis, and loneliness.16
In summary, the evidence on differences in effects across subgroups is quite limited because of the small subgroup samples, limited number of evaluations analyzing subgroups, and lack of consistency of subgroup definitions and results. The limited evidence suggests that larger nursing home reductions may be associated with being more disabled (up to some level), having less informal support, being in a nursing home (certified for discharge), and having a higher statistically predicted probability of nursing home placement (up to some level). Although the differences in effects across subgroups suggest that high risk of nursing home placement is necessary for expanded community care to reduce nursing home use, it is by no means sufficient. Some subgroups associated with high nursing home use were not associated with large reductions in its use. Moreover, many of the subgroups for which significant reductions were observed were small, and the evidence is not sufficiently precise to define eligibility cutoffs, particularly to identify those for whom the risk of nursing home placement is too high.
Although reduction in hospital use was not a main objective of most of the demonstrations (the two exceptions being San Diego LTC and OPEN), there was some hope that community care might substitute for hospital care. This could occur by enabling earlier discharge, or by enabling patients awaiting nursing home placement but not needing hospital care to be discharged to their homes. Community care programs might increase hospital use, however, either because increased monitoring of patients’ conditions may increase admissions or because patients at home may require more hospitalizations than those in nursing homes nursing homes may be able to treat some conditions that would require hospitalization if the person lived at home.
Data problems that plagued analysis of nursing home use do not as seriously affect analysis of hospital use. Because Medicare pays for most hospital use of the elderly, evaluations that used Medicare records captured virtually all use. Wisconsin CCO was the only demonstration that did not use interview or Medicare data and hence did not measure an important component of hospital use. (Data sources for hospital use were the same as for nursing home use.)
Table 4 summarizes the results for hospital use. Among the randomized experiments that measured essentially all hospital use, control group use varied from lows of 4 days over the first year after enrollment (Worcester Home Care and Georgia AHS) to 25 days (Financial Model Channeling). Although treatment group use is typically one day smaller than control group use (in only one case, Georgia AHS, is treatment group use higher; the largest difference was three days lower for OPEN), none of the differences is statistically significant. Based on these results, hospital use appears to be unaffected or at most slightly reduced by case-managed community care. Concern that hospital use might be increased by expanded case management and community services does not appear justified. (The Wisconsin CCO randomized experiment, which relied solely on Medicaid data, found a large statistically significant reduction in hospital use -- a result which appears inconsistent with the predominant evidence of no effect. Results for the quasi experiments are not statistically significant and are without pattern.)
| TABLE 4: Hospital Use during the First Year | |||||
|---|---|---|---|---|---|
| Demonstration | Treatment Mean | Control Mean | Difference In Mean | Percentage Difference | |
| RANDOMIZED EXPERIMENTS, ALL USE | |||||
| Worcester Home Carea (Sherwood, Morris, and Gutkin, 1975, Appendix B, p. 27) | 4 | 4 | 0 | 0.0 | |
| NCHSR Day Care/Homemaker (Weissert, Wan, and Livieratos, 1980, pp. 44, 46, 48) | |||||
| Day Care | 11 | 12 | -1 | -8.3 | |
| Homemaker | 16 | 16 | 0 | 0.0 | |
| Combined | 15 | 16 | -1 | -6.3 | |
| Georgia AHS (Skellie et al., 1982, pp. 173-174) | 6 | 4 | 2 | 50.0 | |
| Project OPENb (Haskins et al., 1985, Appendix A, p. 118) | 9 | 12 | -3 | -25.0 | |
| South Carolina LTC (Blackman et al., 1985, p. III.88) | 18 | 20 | -2 | -10.0 | |
| San Diego LTCb (Haskins et al., 1985, Appendix A, p. 101) | 9 | 10 | -1 | -10.0 | |
| Channelingc (Wooldridge and Schore, 1986, p. c.16) | |||||
| Basic Model | 17 | 18 | -1 | -5.6 | |
| Financial Model | 24 | 25 | -1 | -4.0 | |
| RANDOMIZED EXPERIMENT, MEDICAID USE ONLY | |||||
| Wisconsin CCOd (Seidl et al., 1980, p. 206) | 3 | 12 | -9* | -75.0 | |
| QUASI EXPERIMENTS, ALL USE | |||||
| Triage (Shealy, Hicks, and Quinn, 1979, pp. 449-450) | 8 | 6 | 2e | 33.3 | |
| On Lokb (Haskins et al., 1985, p. 102) | 6 | 8 | -2 | -25.0 | |
| MSSP (Miller et al., 1984, p. 1.24, 1.72) | 20 | 19 | 1f | 5.3 | |
| Nursing Home without Wallsg (Birnbaum et al., 1984, p. IV.13) | |||||
| Upstate | 19 | 16 | 3 | 18.8 | |
| New York City | 18 | 16 | 2 | 12.5 | |
| New York City Home Careb,h (Haskins et al., 1985, Appendix A, p. 102) | 21 | 21 | 0 | 0.0 | |
NOTE: Estimates are for the full
sample wherever possible. For Project OPEN, San Diego LTC, On Lok, Nursing Home
without Walls, and New York City Home Care, estimates are for survivors only.
|
|||||
Analysis of effects on costs are limited by the data collected. Given the multiple providers and funding sources, cost data are dispersed throughout the provider system, and hence comprehensive cost data are difficult to collect. Among the randomized experiments, two (Worcester Home Care and' Florida Pentastar) did not collect sufficient cost information for meaningful cost analysis. The other randomized experiments, in addition to project costs, collected only Medicare costs (NCHSR Day Care/Homemaker), only Medicaid (Wisconsin CCO), Medicare and Medicaid (Georgia AHS, South Carolina CLTC, and San Diego LTC) and Medicare, Medicaid, other public and private costs (OPEN and Channeling). (See appendix Table A.6). Thus, all the evaluations except OPEN and Channeling omitted at least private costs and public costs other than Medicaid, Medicare, and project costs. To the extent that case-managed community care reduces private expenditures (e.g., for nursing homes or community care) or other public expenditures (e.g., for community care), this omission leads to overestimates of cost differences. Given the partial nature of the cost data and the difficulty of collecting cost data, the cost estimates, and particularly comparisons of them across demonstrations, must be interpreted cautiously.
| TABLE 5: Monthly Costs (1984 dollars) | |||||
|---|---|---|---|---|---|
| Demonstration | Treatment Mean | Control Mean | Difference In Mean | Percentage Difference | |
| RANDOMIZED EXPERIMENT, ALL PAYORS | |||||
| Project OPENa | 963 | 1028 | -65 | -6.3 | |
| Channeling | |||||
| Basic Model | 1412 | 1330 | 82 | 6.2 | |
| Financial Model | 1878 | 1592 | 286 | 18.0 | |
| RANDOMIZATION EXPERIMENTS, MEDICARE AND MEDICAID ONLY | |||||
| Georgia AHD | 377 | 254 | 123 | 48.4 | |
| South Carolina LTC | 691 | 676 | 15 | 2.2 | |
| San Diego LTC | 1018 | 672 | 346 | 51.5 | |
| RANDOMIZED EXPERIMENTS, MEDICARE ONLY | |||||
| NCHSR | |||||
| Day Care | 813 | 534 | 279 | 52.2 | |
| Homemaker | 1095 | 786 | 309 | 39.3 | |
| Combined | 1243 | 847 | 396 | 46.8 | |
| RANDOMIZED EXPERIMENTS, MEDICAID ONLY | |||||
| Wisconsin CCO | 515 | 508 | 7 | 1.4 | |
| QUASI EXPERIMENTS, ALL DATA | |||||
| Triage | 455 | 191 | 264 | 138.2 | |
| On Lok | 1518 | 2198 | -680 | -30.9 | |
| QUASI EXPERIMENTS, MEDICARE AND MEDICAID ONLY | |||||
| MSSP | 1154 | 606 | 548 | 90.4 | |
| Nursing Home without Walls | |||||
| Upstate | 825 | 1117 | -292 | -26.1 | |
| New York City | 1633 | 1159 | 474 | 40.9 | |
| New York City Home Care | 1215 | 713 | 502 | 70.4 | |
NOTE: Statistical significance is not
reported. (Evaluations generally did not test cost differences for statistical
significance.) For breakdowns of costs by type of service and payor, see
Appendix Table A.6.
|
|||||
Although their cost estimates are subject to question,17 OPEN reported a reduction in costs of $65 per person per month and Wisconsin CCO essentially broke even (see Table 5). They did so through reductions in hospital use, not nursing home use. Since then the service environment has changed. The advent of Medicare prospective payment using diagnosis-related groups (DRGs) has increased pressure to reduce hospital lengths of stay. While this change may have increased the need for home care, expanding public financing for such care is less likely to reduce hospital use now than at the time of these demonstrations.
South Carolina CLTC does appear to have broken even by substituting community care for nursing home care. During the first year, total Medicaid and Medicare costs increased an average of $53 per person per month, an increase of 7.7 percent. Over three years, costs were increased by $15 (2.2 percent) for the subsample followed that long.18 (Total costs probably increased somewhat more if private costs are included: because more clients remained in the community, they incurred more room and board costs themselves.)
Without substantial reductions in nursing home use, all the other randomized experiments increased costs: Basic Model Channeling ($82 per person per month), Georgia AHS ($123), Financial Model Channeling $286), NCHSR Day Care/Homemaker ($279-$396, depending on the model), and San Diego LTC ($346).
Evidence from the quasi experiments is inconsistent. Four (Triage, MSSP, the New York City site of Nursing Home without Walls , and New York City Home Care) found large increases, while two (On Lok and the upstate of Nursing Home without Walls) found large decreases. Given their inconsistency and the inherent problems with a comparison group methodology, the results from the quasi experiments do not alter our conclusion based on the randomized experiments.
Families and friends provide much of the care of the frail elderly. In 1982, of the 18 percent of the noninstitutionalized elderly who had limitations in ADL's or IADL's, only 5.5 percent relied exclusively on paid formal providers, 20.6 percent relied on both formal and informal caregivers, and 73.9 percent relied exclusively on informal caregivers (Liu, Manton, and Liu 1985, Table 1). Expanding public funding for formal services has the potential of partially substituting those paid formal services for informal care provided without cost to the government by family and friends. Two views of substitution differ in their implications for public costs. One asserts that formal services will (partially) replace informal care, perhaps with benefits to the informal caregivers and clients, but at increased public cost. The other asserts that formal services will supplement informal care -- leading to some substitution in the short run but will enable caregivers to continue caregiving longer, thereby reducing total public costs in the long run (Spivak 1984; Christianson 1986).
Little attention was given to the substitution issue by the community care evaluations. Only six estimated demonstration effects on informal caregiving, and the measures were generally limited (See appendix Table A.7).
The South Carolina CLTC project substantially increased the proportion of the sample receiving informal care at home. This increase was directly associated with the decrease in nursing home placement. Because more of the treatment group remained at home, where they relied on informal care, a higher proportion of the treatment group as a whole received informal care than the control group. (Informal care in nursing homes was not measured.)
In the absence of reductions in nursing home use, however, formal care appears to substitute to a small extent for informal help with IADL tasks. Of the demonstrations that used randomized experimental designs but did not significantly reduce nursing home use, three (Worcester Home Care, OPEN, and Basic Model Channeling) had no significant effect on informal caregiving.
Two (San Diego LTC and Financial Model Channeling) did not affect informal help with ADL tasks but decreased informal help with IADL tasks. The San Diego LTC study did not report the magnitude, but the reduction by the Financial Model of Channeling was small in magnitude and concentrated among caregivers least closely associated with clients (visiting caregivers, friends or neighbors, and relatives other than spouses or children). Channeling did not affect the amount of care by the primary caregivers, who provide most of the care. Finally, the evaluation of the New York Home Care Project found a reduction in informal help with ADL tasks, but it used a comparison group methodology to estimate effects.
Whether the small amount of substitution of formal for informal IADL care in the short run extends the capacity of informal caregivers to continue giving care, thereby increasing aggregate caregiving over the long run, remains unresolved. The small reductions in informal help with IADL by the San Diego LTC and Financial Model Channeling demonstrations did not lead to substantial reductions in nursing home use during the 18 months of evaluation follow up. Whether the modest reduction in caregiving burden would have enabled them to continue giving care in the community after 18 months is unknown.
All of the demonstrations shared to some degree the objective of improving the quality of clients' lives. Providing clients the opportunity to choose to live in their own homes rather than nursing homes was one mechanism expected to make clients better off. Providing needed services to those who would live at home even without the intervention of case management and expanded community services was a second mechanism.
Although all the demonstrations sought to improve life quality, attempts to measure it varied considerably across the evaluations, making overall assessment of effects on life quality and comparisons across projects difficult. Indicators of life quality range from narrow measures such as satisfaction with arrangements for services and number of unmet needs for care to global measures such as morale, life satisfaction, and ultimately longevity.
Unmet Needs and Satisfaction with Service Arrangements. Two demonstrations, both randomized experiments, asked about the care received. Georgia AHS asked whether sample members were getting enough help. Channeling asked about satisfaction with arrangements for housecleaning, meals, laundry, and shopping. Both found small but statistically significant benefits (see appendix Table A.8). Channeling also found large significant increases in primary informal caregivers' satisfaction with care arrangements under the Financial Model, significant increases in clients' confidence about getting help under both models (not shown), and significant reductions in reported unmet needs for care (see appendix Table A.9). Together, this evidence, although from only two demonstrations, suggests that expanding coverage of community services has the immediate effect one might expect of increasing satisfaction with the amount of help being received and reducing perceived unmet needs. (New York City Home Care, a quasi experiment that used noncomparable data, reported significant reductions in unmet needs in two areas, medical and economic/social/environmental, but not in ADL and IADL care.)
Problems with the Physical Environment. Although the community care demonstrations were not intended primarily to alter community housing arrangements, some had limited ability to pay for such things as emergency shelter and removal of architectural barriers, and case managers could encourage changes in residence and assist clients in seeking new housing. Some improvement in clients' physical environment is possible as a consequence. The three randomized experiments that examined measures of problems with the physical environment showed a pattern of reductions in problems, but it was statistically significant only for Basic Model Channeling (see appendix Table A.10). Although in the expected direction, the evidence is too limited to suggest that the demonstrations improved clients' physical environments. (New York City Home Care, a quasi experiment, also reported a significant reduction in problems with the physical environment.)
Social Interaction. Although not its central focus, case managers might be expected to encourage more social activities, and the provision of transportation might permit socializing (e.g., at senior centers or adult day care). Of the six randomized tests using one or more measures of social interaction, three (NCHSR Day Care/Homemaker Combined Model, Basic Model Channeling, and OPEN) found significant increases (see appendix Table A.11). The NCHSR Day Care/Homemaker analysis was, however, based on comparison of treatment group members receiving project services with control group members receiving no similar services, rather than on the full experimental sample, and the OPEN data were not comparably collected, undercutting the confidence that can be placed in the results of these two evaluations. Nevertheless, the studies provide limited evidence that case-managed community care leads to small increases in social interaction. (The results of the one quasi experiment that examined social interaction, New York City Home Care, are consistent with this conclusion.)
Health and Functioning. Health and functioning were expected to be improved, or their deterioration slowed, through regular monitoring to identify and respond to problems and through increased access to services (e.g., physical therapy). In addition, functioning was expected to be improved by reducing nursing home placements. Nursing homes are believed to increase functional dependence because they do not permit patients to perform some ADL (e.g., bathing) by themselves. Three randomized experiments analyzed self-rated health (see appendix Table A.12). San Diego LTC found a significant increase in self-rated health at 6 months (not shown) and Basic model Channeling found significant increases at 12 months. This provides limited support for the hypothe