U.S. Department of Health and Human Services
This report was prepared under contract #HHS-100-80-0157 between the U.S. Department of Health and Human Services (HHS), Office of Social Services Policy (now the Office of Disability, Aging and Long-Term Care Policy) and Mathematica Policy Research, Inc. For additional information about the study, you may visit the DALTCP home page at http://aspe.hhs.gov/daltcp/home.htm or contact the office at HHS/ASPE/DALTCP, Room 424E, H.H. Humphrey Building, 200 Independence Avenue, SW, Washington, DC 20201. The e-mail address is: webmaster.DALTCP@hhs.gov. The DALTCP Project Officer was Robert Clark.
The National Long Term Care Demonstration was developed in response to rapidly increasing private and public expenditures for care of the elderly. The Channeling demonstration, as it is referred to, sought to decrease costs of care for the elderly and improve their well-being by substituting in-home community-based care for institutional care. The program's essential feature was comprehensive case management, a system for coordinating the many formal community-based services that were already available to elderly individuals. The existing service system was comprised of many different providers, most of which provided a limited range of services to address specific client needs; Channeling sought to improve on this by identifying and addressing the full range of clients' needs. The program also had access to funds to purchase additional services for clients.
The evaluation of the demonstration has been conducted and presented in various reports, each dealing with different types of expected effects. Many different outcomes were examined, with the general finding that Channeling increased the service receipt and well-being of clients but did not substantially reduce the use of nursing homes and had no effect on hospital use. This result led to questions about why the program was only partially successful and has raised somewhat broader questions about the availability and effects of case management and community services in general.
The Channeling evaluation featured a randomized design. Eligible applicants to the program in each of the ten demonstration sites were randomly assigned to the treatment group, which was offered the opportunity to participate in the Channeling program, or to the control group, which was barred from participation. Both groups received baseline assessment interviews to gather data on their initial characteristics and followup interviews 6, 12, and (for half of the sample) 18 months later to obtain data on outcomes that Channeling could be expected to affect. The randomized design ensures that comparison of these two groups provides reliable estimates of Channeling impacts, defined as the difference between treatment group members' actual outcomes and that which they would have experienced in the absence of the program.
This definition, while technically correct, could result in relatively small estimates of program impacts even if Channeling actually were very effective in alleviating clients' needs and substituting formal in-home care for institutional care. This could occur if only a fraction of the treatment group actually participated in Channeling or if a substantial proportion of the control group received case management or community services from existing agencies that closely resembled those provided or arranged for by Channeling. The fact that treatment/control differences indicate no impacts of Channeling on hospital or nursing home use may merely reflect the already rich service environments into which the demonstrations were introduced. On the other hand, such services may have little impact on institutional use in any case.
To distinguish between these two alternative explanations we first examined the type and amount of case management and services received by the control group, and compared that to what was received by treatment group members. We then estimated equations that were intended to address the following specific questions:
What are the impacts of case management and formal in-home services on outcomes of interest?
Do these impacts differ for Channeling and nonChanneling clients?
Do impacts vary with the comprehensiveness of case management?
Knowledge of the differences between treatments and controls in the quantity and nature of services received, combined with reliable estimates of the difference in effectiveness between Channeling and other providers of case management and services would enable us to disentangle the possible reasons for the lack of significant treatment/control differences on key outcomes.
Pursuing this strategy, we first described the different features of comprehensive case management, as offered by Channeling, then examined the extent to which other agencies in the demonstration sites offered a comparable level of case management. We found that in most of the demonstration sites, case management that was at least close to being as comprehensive as that provided by Channeling was already available, but that only 10-20 percent of controls received it, depending on the model and time period. That proportion increased to as much as 35 percent when case management that was ongoing (i.e., that incorporated monitoring and reassessment of service adequacy over time) but perhaps less comprehensive than Channeling was included. Finally, taking into account the case management provided by home health agencies, the proportion of controls who may be presumed to have received some case management rose to 60 percent in the basic model and 75 percent in the financial model. Thus, although not many controls received case management that was comparable to that offered by Channeling, the proportion receiving at least some case management was very high.
Comparable data for the treatment group showed that the proportion reporting receipt of ongoing case management was 30-50 percentage points higher than the control group rate, depending on the time period, model, and measure. When home health services are included in the case management measure the proportion of the treatment group receiving services exceeds the control group rate by 20-30 percentage points. These differences, while no trivial, are substantially less than 100 percent, which suggests that Channeling is more properly thought of as a test of the effectiveness of alternative types of case management, rather than a test of case management per se.
Taking an analogous approach to formal community based services, we found that 14-19 percent of the control group received skilled services (nursing or therapy) and 50-64 percent received semi-skilled services (homemaking, personal care, housekeeping, etc.) during the reference week. Receipt of skilled services differed little between treatment and control groups, but semi-skilled services were received by a significantly higher proportion of treatments than controls in both models. Service recipients in the treatment group tended to receive more hours of care than control group recipients as well, at least in the financial model. Nonetheless, half of the controls in the basic model and nearly two-thirds in the financial model were receiving some semi-skilled services. Thus, it is clear that Channeling was not being compared to a situation in which no other services were available but to a situation in which a significant quantity and quality of case management and services were already being provided.
The next step was to estimate the impacts of case management and services on outcomes of interest, for both treatment groups in both models. Using regression to control for other differences between recipients and nonrecipients of case management and services that could affect outcomes, we found that neither skilled services nor any of several measures of case management seemed to have much effect on nursing home admissions or days. However, receipt of semi-skilled services was associated with significantly lower use of nursing homes. Although these results were plausible, we were concerned that the estimated impacts for case management and services reflected the effects of unobserved differences between recipients and nonrecipients of services (not controlled for by the baseline explanatory variables in the model), rather than the true effects of these services on nursing home use. This concern was heightened when we substituted six month measures of case management and services for the concurrent 12 month measures in estimating impacts on institutional use during the 7-12 month period, and obtained estimated impacts of semi-skilled services that were of the opposite sign as obtained with concurrent measures and no longer statistically significant. We also found that higher unmet needs and more informal care were associated with receipt of semi-skilled services, which further supported the belief that the regression model failed to control fully for the differences between recipients and nonrecipients of services.
An econometric procedure, two stage least squares, was then used to eliminate the effects of these unobserved factors on our estimates. This procedure is designed to eliminate the bias caused by the unobserved factors reflected in the case management and service variables by replacing these variables in the regression with predicted values of these services. However, because it was difficult to predict with much accuracy which sample members actually received case management and services, the procedure produced estimated impacts that were anomalous in size, of the wrong sign in some cases, and nearly always insignificant because of very large standard errors. Thus, we were unable to rely on statistical procedures to resolve the ambiguity about the original regression estimates.
As a final way to resolve the discrepancy between the estimated impacts using lagged instead of concurrent values of semi-skilled services we examined the data further and found that the difference was due to a fairly large proportion of those admitted to a nursing home in months 7-12 reporting receipt of services during the 6 month reference week but not for the week prior to entering the institution. Although there were few cases involved (because of the low nursing home admission rate) the potential implications for our results were considerable: if some sample members living in the community are forced into nursing homes when they suddenly lose the formal services they are receiving, this would be an important finding. Medicare records of the individual sample members in question were inspected and found to support the interview data, thus ruling out sample member recall problems as an alternative explanation. However, the average number of hours of care received was quite small for most of these cases, making it unlikely that loss of this level of care was responsible for their admittance to a nursing home. It seemed rather more likely that other events were responsible for the sample member being admitted, and once this occurred formal community care was terminated.
Because of the ambiguity of the results we are unable to determine from these data whether case management and formal community services affect nursing home use or other outcomes. While it is tempting to conclude from the initial results that receipt of semi-skilled services reduces institutional use. The drastic change in results when lagged values of semi-skilled services were substituted for concurrent measures, the failure of the model to show that unmet needs decline in response to services, and the erratic two-stage least squares estimates all cast doubt on the interpretation of the regression estimates as evidence that semi-skilled services reduce nursing home use. It is clear from the results that case management and services were widely available to and received by controls in the Channeling sites. Whether the general lack of significant treatment/control differences observed in the Channeling evaluation is due to the widespread receipt by controls of services that are comparable in effectiveness to those delivered by Channeling or due to a lack of impacts of any case management or services on institutional use is unclear.
While this result is disappointing it is not surprising. It was clear from the outset of this analysis that differences in mean outcomes between recipients and nonrecipients of case management and formal care could not be attributed to the effects of those services unless the other factors that distinguish these two groups could be controlled for. It was also recognized that this would be difficult to do because of the many unobserved factors that influence both receipt of services and key outcomes such as institutional use. The procedures that were employed were intended to overcome these problems, but the inherent complexity of the institutionalization process and the lack of a data set collected expressly for the purpose of measuring the impacts of case management and services were difficulties too severe to be overcome by econometric models.
What we have learned from analysis of the Channeling data is that the comprehensive form of case management offered by Channeling is effective in reducing unmet needs and increasing client and caregiver satisfaction with life. This impact is an especially impressive achievement given the extensive services received by the control group. However, we also know that Channeling-like programs are not more effective than less comprehensive case management in reducing institutional use; only a small fraction of the control group received such intensive case management and yet no significant treatment/control differences in nursing home use were obtained. What we have not been able to determine is whether a less comprehensive form of case management or even services without case management are sufficient to reduce institutional use. Reliable answers to these questions will require data that is more directly focused on this question.