George Carcagno, Mathematica Policy Research
Judith Wooldridge, Mathematica Policy Research
Thomas Grannemann, Ph.D., Mathematica Policy Research
Peter Kemper, Ph.D., National Center for Health Services Research
GEORGE CARCAGNO: Judith Wooldridge will be chairing the meeting. Judith was responsible for analyses of nursing home and hospital use, and also was responsible for the design and maintenance of the data base. Tom Grannemann, was one of the researchers on the project, and had responsibility for much of the analysis of formal community care costs and utilization. Also on the panel is Peter Kemper, who was the co-principal investigator of the demonstration, and was, I think, our leading intellectual light here. We owe him a lot.
JUDITH WOOLDRIDGE: As you all have heard, the Channeling evaluation collected data on an elderly, very fragile population who had applied for community service care in place of nursing home care. They were a screened sample who had to pass certain eligibility criteria. As you are probably aware by now, this sample is not nationally representative as a result. There is a lot of things about the way the sample was selected for the evaluation and decisions that were made in the data base with respect to the evaluations that you will need to know to make use of these data to get the maximum out of them.
Let me just reiterate a few points about the data base. The sample came from ten sites. We had two models: the basic model which provided case management, and the financial control model which also provided services. It was a randomized design so that we have a treatment group and a control group. There were over 6,000 members of the sample, so that 6,341 people were randomized into the research sample, and that is the maximum sample size that you are going to encounter in any of the data files.
I would also like to mention at this point that follow-up from the point when people were randomized was a maximum of 18 months. This is not a very long term data set. It is long term care over a short time period.
In producing public use files, we wanted to allow replication of the analysis that we conducted. This meant that all the variables used in all of our analyses are present in the files. That means if you read any of our reports and you want to use a particular variable that appears in the particular report, you can find that variable on the public use files. When we produced these files we did have to maintain the confidentiality precautions that were taken to protect our respondents. We did not do a great deal to these files, however. What we did was to delete obvious identifiers, such as names, addresses, Medicare and Medicaid numbers. We also deleted information about whether there was a legal guardian.
Other than that, we modified a couple of variables, the age variable and the ethnicity variable because you got into some pretty small cells in some sites, and we wanted to make sure that you could not identify anybody in our sample. We also deleted information about providers in the community, hospital and nursing home that we collected in the follow-up interview. That information is not in the data base. That is the extent of modifications to the data from the files that we used in the analysis to the files that are available for public use.
Regarding the sources of data for Channeling, we interviewed the sample members themselves, and they were screened to check whether they were eligible to enter the program. They were assessed within a very short time period of their screening, and there were then follow-ups at 6 month intervals so long as we could trace them up to a maximum of 18 months. Part of the sample has follow-ups for 12 months, and part for 18 months; that is, those people who came into the sample sufficiently early, in the first half of the intake period, have follow-ups for 18 months.
We also interviewed caregivers of sample members. The primary caregivers at the baseline, at 6 months and 12 months. Other data sources included the client tracking file that each of the ten Channeling projects set up in conjunction with Mathematica Policy Research (MPR). With them, we designed forms that could be used to track the clients and all the information on all of the clients who came into the program throughout the intake period are available in that file.
There was another source of information--provider records, which we used to collect information on formal community services. In addition, we used provider records' extracting processes to pick up some hospital and nursing home data, although that was not the principal source of information for those types of data.
We have Medicare claims data and Medicaid claims data from each of the ten states that the sites were located in. Those claims cover all the services provided by Medicare and Medicaid in the appropriate jurisdictions.
The financial control model sites had information on service use by their clients that had to be supplied to Health Care Financing Administration (HCFA), and we used that information for the prime community service estimates of service use. In addition, we have one other source of data which was death records. The population was very fragile and had a very high mortality rate. Whereas, we did find out that some people had died when we tried to do follow-ups. Wecollected death records from each of the ten states to insure that we knew whether or not people were alive at the end of the 18 month period. Of course, we got death dates, which allowed us to know when they died in relation to their randomization date.
There are 14 data files which constitute the Channeling data base for the public use file data set. Those files are documented in 11 separate volumes.
In addition to the detailed documentation there is a short introductory volume which explains what data files are available and, in general, describes what is in each data set.
There are three types of data files in the data base. There is what we call source files, and by that, we really mean files that are based upon an interview or an instrument. There is a file based upon the screening instrument; there is one based on the sample in the baseline; and there is a file for each of the sample member follow-ups, the caregiver baseline, the caregiver follow-ups and the client tracking form. Those we call source files.
We also have four files that we used based upon the sample member follow-up for informal care analysis; there is a formal community service file; a file with hospital, nursing home and other medical service data; and a file on the quality of life. All of those are analysis files. Finally, there is a status file, which does not contain much substantive information as it is just a file to tell you what is in the other files.
The key features of the data base is that this is a system of files, and individuals showing up in one file will generally show up in another file, as well, subject to sample limitations of particular files. You can link any individual across all of these files by means of an ID number which is, in fact, the first variable in the file.
It is important to note that there are different samples in each file. I will talk a little bit about the samples in a moment just to give you a notion of why they are different. Within a given file there may be sub-samples, because we always included the maximum number we used and then if there were sub-samples used for the analysis, we have identified those with special sample flags.
All of these files contain some basic information, like the ID numbers, the site, the treatment, the status of the respondent and the model. All but two of the files, exceptions being the status file and the client tracking file, contain a set of variables that we used throughout the final analysis. We used the same set of control variables across all the substantive areas of the analysis and we included those in each of the files except the two I mentioned.
There are three caregiver-based files which contain an equivalent set of standard control variables based upon caregivers.
All of the files which are based upon instruments contain all of the variables in the instruments and some constructive variables, certain subjects, the fact that we have deleted certain operational variables arid, obviously, confidential variables.
The analysis files contain all the dependent variables we used on our analysis so that you can identify those and use them if you wish to. Let me just talk quickly about the types of samples.
As I mentioned, there were 6,341 people in the research sample. All the files contain subsets of that 6,341, with one exception which Is the client tracking file which does not include any of the control group. It only includes clients, and it includes clients who were not in the research sample. People who entered the program before we began randomization; there were a few of those, and quite a number of people who entered the program in the year during, which the program was running its steady state who were not included in the research sample. That is a special and different sample in that one file.
Other than that, we have a variety of samples used for different purposes. To the extent that we could, we maximized sample sizes. We also tried to make them equivalent to cross-analyses as much as possible without throwing away data points. There is some special aspects to some of them. The screen file contains 6,326 sample members, 15 less than were in the research sample.
The baseline file is a subset of that group. It is all of those people who were screened who had a baseline. The follow-up file is a subset of that. It is all those who were screened who had a baseline, a follow-up and so forth. The sample sizes do get smaller and for some special purposes the sample sizes can get very small, indeed, which I will talk about as we go through the individual files.
The purpose of the screening instrument was to assess people to determine if they were eligible to participate. For that reason it has information on functional status, ADL and IADL measures, fragility of the support system, unmet needs, and so forth. It is a fairly brief instrument but it is available for everybody who is in the research sample. The screen file includes all the variables that are on the screening instrument. Let me just indicate to you that the documentation includes the instruments where it is an instrument-based file.
The sample baseline file was administered as soon as possible after individuals had been screened, found eligible and randomized. There are two things to know about the baseline file. First of all, that there are two different versions of the baseline we developed; one for use in the community setting and one for use in the institutional setting. We have quite a few people who were getting ready to leave an institution and go back into the community, so we developed two instruments. You do not have to worry about that in using the file. We have taken care of that for you as much as possible by making a set of joint variables which are available from both sources. There are a few variables that were available from one or the other sources, and the documentation does indicate that.
The only thing you need to be aware of is that the baseline instrument was administered differently to the treatment group and the control group. The treatment group had their baselines administered by the case workers or the assessment group, whereas the control group had their baselines administered by MPR interviewers. All of these people who administered the interviews had some training. It was felt for clinical reasons that it is not good to have the client group baselined by their case manager in the project.
The baseline is an extremely rich source of information about functioning, health status in the period before enrollment, informal caregiving, financial resources, demographic factors, and unmet needs. The sample size for that file is 5,626. That is a large chunk of the group that was screened, though some people refused to be baselined after they became sample members who were randomized into the sample. That is the main reason for the discrepancy between the numbers in the screened sample and the baseline sample.
The client tracking status change file is a special file that was developed for administrative purposes. The sites used the client tracking forms to keep track of where the individual clients were in the process, and then they filled out status change forms every time the person moved between statuses. People were allowed to become inactive, although we later modified that and they only allowed to move between active and terminated. They could be reactivated. The file does not have a lot of information on anything except changes in status and elapsed time between screen, baseline and between baseline. MPR used those forms as they came in to monitor the state of the program. It is very much a process file.
There are three sample member follow-up files. There is one at 6 months, at 12 months and at 18 months. They contain the same information; that is, the questionnaires were identical so that you have the opportunity to look across three points in time and look at the same information for a given individual. The purpose was to find out what kind of outcomes were developing for the sample over those three time periods. There is a lot of information in there. We have information on insurance coverage, health status, housing conditions, expenditures, transfer payments, services, in-home service use, formal community service use, hospital use, nursing home use, well-being of the sample member, income, assets and functioning. The three files each contain 782 variables. Almost as many as the baseline file and covering very much the same kinds of information that was in the baseline.
This file contains all of the variables in the instruments subject, with of course, operational variables and confidential variables taken out. Variables that we constructed for the purposes of particular analyses from these follow-up files are to be found in the analysis files rather than in the follow-up files.
The next file is the status file. The information in here that you might need to use is all the sample flags. It tells you for a given individual which of all of the many samples we developed and the sample member falls into. It also includes information obtained from the death records and information that we collected on Medicare and Medicaid entitlement from HCFA and from the Medical Assistant Bureau in each state. In addition it has information on when instruments were fielded.
The next two sets of files related to the primary caregiver of sample members. We did not initially have a plan to interview caregivers, and so we did not actually start interviewing caregivers until some months after the intake period began. In March 1982 is when we first started taking sample members in and we took sample members into the research sample for the following 15 months. It was not until November 1982, some 8 months later, that we started the caregiver baseline survey. The follow-up started 6 months after the baseline. That means that we do not have the opportunity to have as large a sample for caregivers as we do for sample members. Information was collected on only l,919 caregivers who were baselined. There is another reason for that, which is that not all sample members indicated that they had a primary caregiver. We only went through an interview if they said they had one.
The caregiver instruments, the baseline includes information on the services they provide the elderly sample member, the financial contributions they make, economic and family behavior, and psychological and social well-being of the caregiver. You will recall that psychological and social well-being of the sample member is available from the sample member baseline and follow-up instruments. The caregiver follow-up occurred at 6 months and 12 months, and the purpose of these instruments was to assess the impact and collect information on the impact of the program on the caregivers themselves. The questions, therefore, focused on care provided 6 months and 12 months after randomization. Care provided by people other than the primary caregiver who was being interviewed, other people who provide care to the individual. The financial contributions, information about institutionalization and formal service use, and one important use of the caregiver follow-up was that in most cases where the sample member had died when we went out to do a follow-up, if there was a caregiver, we asked that caregiver information about service in the period prior to their death. In this way we were able to pick up a lot of extra information that we would otherwise not have had on formal community service use and nursing home use.
Of the remaining four files, that I just want to talk about fairly briefly, the first one is the formal community service file. This file includes within it a number of samples, and it is probably the most complex of the files in that respect. We have samples in there used for some purposes and larger than samples we have for other purposes. For example, based upon the sample of people with follow-up instruments, we have a follow-up sample with a lot of information on community services. We also have provided a records extract that we collected on formal community service use, but only for 20 percent of the sample, so that we have a very small sample of people, relatively, with fully detailed information on service use collected from providers. We have a much larger data set available on service use than the sample member was able to tell us about.
Anybody who has any questions on those samples in the formal community service file, you should raise those with us. I do not want to go into too much detail now. I do not want to scare you off, but just to say that there is a set of, for the most part, nested samples. You are not going to get all the data on all the sample members that are in that file, although there are about 5,600 sample members in the file. They do not have all the data. It is a very detailed data set. It drew on the provider record extracts. We also did some surveys of individuals who were privately contracted by the household to provide services to the sample members. We have information in this file from the financial control model sites on service use provided by the projects. In addition, we have all of the information on community service use under Medicare and Medicaid that we drew from claims data. The file includes information on use of all major community services and expenditures for those services by funding source. I should mention that this file also contains information on case management, housing and transfers, which used slightly different samples than the other formal community service analysis, but the individual for whom there are full data are flagged in the file.
There are two things that I would like to mention about this file at this time. One is that the file is organized based upon the 6 month follow-up period. For example, we have data on community service use and expenditures over the period 6 months from randomization, that is from randomization to the first follow-up. Then we have information for the next 6 month period, from the beginning of the 7th month to the end of the 12th month. What the file provides is information on service use within that 6 month period. There are also some data on a snapshot period that occurred just before the follow-up instrument. That is the information that is available for the larger sample and the data for the whole 6 month period is available for a smaller sample.
The other thing I should mention is that expenditures data has been standardized and regardless of when the actual expenditure occurred, we standardized them to a February 1984 expenditures rate using price index information. From your perspective you do not need to know when the sample member received those services because across sites and everything, they have all been standardized.
I would like to just skip over the informal care analysis file and talk about the hospital, nursing home and other medical service analysis file. It is has some things in common with the community service file. For example, the data are organized in 6 month blocks the same way that the formal community services are arranged. It has many of the same data sources. It is based upon Medicare and Medicaid claims augmented by provider record extracts for those individuals where we felt that for one reason or another Medicare or Medicaid was not going to provide a complete source of information. Without going into it, I should explain that we reviewed everybody's Medicaid eligibility and how long they have been eligible to determine whether or not we felt that Medicaid was going to be a complete source of nursing home use, for example, that being a primary payer of the nursing home.
We assumed that if we had Medicare information on hospital use we had pretty complete data. We also drew on information in the sample member follow-ups and we also, when necessary, used the caregiver follow-ups for those sample members who were deceased in order to prompt provider record extracts or find out about service use in some other fashion. This file has information on hospital use, nursing home use and other medical services, such as physician services, and so forth, anything that was available to use from the Medicare and Medicaid files on other medical service use.
There are no stay data as such on this file. If you want to know when an individual went into a nursing home and when they came out, and when they went into a hospital and when they came out, and so forth, changes of that kind you can not tell from these files. What you do know from these files is how many times a person was admitted to a hospital or a nursing home, and how many days they spent in a hospital or a nursing home. Also, there are binary variables indicating whether or not a person did have any use, in addition to how many times they had an admission or a discharge.
Let me go on to the informal care analysis file. This file is based upon the sample member follow-up interviews that provides information on informal service provision by caregivers. Rather than coming from a caregiver follow-up, it comes from the sample member follow-up. There was quite an extensive data collection on informal services in the follow-ups, and all of the analysis variables that we used, rather than putting them in the follow-up file, they appear separately in this informal care file.
Anyone who wants to look up both formal and informal services would have to look across and merge files. As far as we were able to, we did try and make the variables very similar, compatible between formal and informal. We asked the same types of questions whenever we could, and we have coded up the variables in the same kinds of ways.
Finally, there is a file that was used to analyze the quality of life of the sample members. Again, this is based on the sample member follow-up interviews, and it includes information on satisfaction with care, social and psychological well-being, and functioning of the sample member at 6, 12 and 18 month intervals.
QUESTION: Are there any data on the use of home care devices?
JUDITH WOOLDRIDGE: Tom can correct me if I am wrong, but I believe the only information we have would be on durable medical equipment from the Medicare home health service claims and if there was anything equivalent, from Medicaid.
THOMAS GRANNEMANN: You should have an indicator of whether durable medical equipment was used and also the Medicaid and Medicare reimbursements for those. No indication of exactly what type of equipment or any detail on when precisely it was used except in these 6 month periods.
PETER KEMPER: There are also some interview questions. I doubt they are at the level of detail that you want, but surrounding the disability measures. For example, when they asked about incontinence, I believe there is a response about help with a device. There is also an interview question on equipment such as whether the person received any special equipment such as grab bars, that type of thing. Probably the best thing to do is once you have got that introduction, to actually go and look at the report that Tom worked on about home care, and look to see what variables were analyzed and then go look at the instruments to see what is there.
QUESTION: I am particularly interested in who used hospitals, who used nursing homes, and whether you have any measures of acuity, I know you have measures of chronicity and disability, but are there any measures of acuity from a medical point of view? What do you know about their diagnoses? Do you have anything on multiple hospitalizations, or if people used hospitals and nursing homes?
JUDITH WOOLDRIDGE: For each of the individuals in the hospital and nursing home use file there are measures of whether or not they used a hospital, how many times they were admitted in a 6 month period, and how many days they were in the hospital. There was also information available from the baseline on service use prior to enrollment and we do not have any diagnostic information on these individuals. What we do have is measures of functioning, both at baseline, and if you are interested in looking at functioning at the nearest 6 month time point, you could use the information from a follow-up interview to associate it with hospital use in a subsequent period or previous period, depending upon what the question was. You can tell, indeed, whether somebody was a multiple user from this file. It will tell you how many times a person was admitted over three 6 month periods.
QUESTION: Do you have any mental health measures?
JUDITH WOOLDRIDGE: This is an area where we do not have a great deal of information.
There is a measure in the baseline of people's mental status, what we call the portable mental status questionnaire. I do not know if that is a generally used term, but it had ten questions and the individuals were asked the questions, the question is how many they got right out of ten, you estimated them into the status on that basis. That, as far as I recall, is the only information we have, excepting that the Medicare and Medicaid claims files may in some cases have had information on personal counseling.
PETER KEMPER: There is a response on whether somebody received in-home mental health counseling and almost no one did. Also on the screen there are one or two questions on behavior problems and orientation of some clients. There is some measure on the caregiver interview of whether the person needs supervision or not. That is for a limited sub-sample who received that, for whom there was a caregiver interview.
One person you could talk to about the mental health measures is Korbin Liu from the Urban Institute who is doing some research in the area related to cognitive impairment and Alzheimer's, and is using the Channeling data.
QUESTION: I was wondering about the degree to which formal care substitutes for informal care over time. Are we loading up the system with the formal side on top of informal care?
Also, you have the 6,341 cases across ten sites in two different groups. Might the ten sites differ, like on a variable such as ethnicity, for example? Would it be useful to do any kind of comparisons across sites, saying that environments, indeed, are different. Might the population profiles be different except for a question of fragility?
GEORGE CARCAGNO: You can, indeed, look at issues of formal and informal care, and the possibility of substitution effects, that was one of the analyses that was carried out as part of our research. What I suggest you do is, start with our report on informal care and the substitution issues, and decide whether there are questions beyond that you want to address.
With respect to the comparisons across sites, there is a little supplementary report we did that has characteristics data by site. We did not do any impact analysis by sites. We did do some formal statistical tests to establish that, in fact, we could pool the data within models.
First of all, the story does not change a great deal across sites, so that we were not in a situation where when you aggregate up the results you have offsetting effects at different sites. We also tried to look at the sites and explain what differences did exist in the outcomes. That just proved to be a very difficult thing to do, because there was so much going on. There are so many differences. We had hypotheses, for example, about what you might expect to see in rural sites, and then you would find that would be true in Kentucky but not in Maine. In Maine, there was something else going on. It gets to be very difficult to go down and try to do any careful analysis or an analysis they have a lot of confidence in at the site.
QUESTION: I was actually interested in the ethnicity. That was the one I was most concerned about, not urban and rural.
JUDITH WOOLDRIDGE: You could look at ethnicity if you wanted to across sites. It does vary across sites, as you would expect. There is a big variation.
PETER KEMPER: I think those three questions illustrate an important point. This is a very rich but complex data set, and one that requires a commitment to use. If you write for the documentation, I do not know if it was absolutely clear or not, but you do not get one stack, you get twelve stacks. Finding your way into the data takes a commitment.
Each of these three areas, the site difference, the substitution of paid formal care for informal care provided by family and friends, and subgroup differences were the subject of a report that was part of the evaluation. There is not only this documentation, but there is a series of technical reports that we and others on the team did as part of the evaluation. We invested a fair amount of energy in documenting what we did and giving a full explanation. I think for people who have a particular area of interest, you should get that report and the executive summary of the evaluation.
QUESTION: One of the things that bothers me about channeling is that it was oriented toward a frail population. Is there another report or other studies that have been towards less frail populations? Has there been cross-analysis comparing the results at all?
PETER KEMPER: I guess, there are a couple of levels to respond. There have been other experiments like Channeling that served less frail populations. Quite a few of them, actually. There are reports available of those and some reviews. I have recently completed a review. If you will write to me, I will be glad to send you a copy of it.
The second level is what use is the Channeling data if given that we know it is made up of people who came to a program to get community services. They are not the same kind of people who would buy long term care insurance. It is probably the only data set that has such complete longitudinal data on service use at home, nursing homes, hospital use and expenditures.
THOMAS GRANNEMANN: I do not know if you or other people have talked about doing this crosswalk between this and the NLTCS data, where the NLTCS, at least in the screening, represents a broader population and this is a group of people who are users. They would have to see the Channeling data as representing a person, once he is in need of long term care, this is what the pattern would look like. The population is not going to be exactly representative of what would be under an insurance plan. It would provide a good indication of use patterns conditional on disability. You will have a lot of information on ADL and IADL that can tell you at a certain level of disability what are use patterns. We have got a separate analysis that breaks it down by ADL and IADL impairment. That can give you a good measure for any subgroups of data sample, what their use pattern is likely to be.
QUESTION: What can insurers learn about the case management results from channeling, both the positive and the negative?
JUDITH WOOLRIDGE: In our analysis we were interested to know whether case management would, indeed, save funds by keeping people out of expensive nursing homes and even out of hospitals to some extent. Our findings were not all that good on that point in the sense that the program cost money and did not save money, really. The actual nursing home use was extremely small, barely significant. We think that there was a slight effect, but it was one that we had to talk about to decide whether we thought there was an effect. The program did not keep people out of hospitals or cut hospital costs at all. I think if you start with our final report on the costs of the program and read it, that is a good place for you to start.
QUESTION: Do you have any suggestions as to what could have made it more effective?
GEORGE CARCAGNO: Let me just respond to your earlier question and then I will respond to that one. In the information on cost of case management for both of the two models, that information is available. There is also a qualitative report on the planning and operational experience of the demonstration that provides some descriptive information about the case management that was delivered and what kinds of people were involved as case managers and so on. That is another source of information about the case management that you might want to take a look at.
There are a lot of ways you can do case management, including having control over medical and nursing home expenditures, as well as just community care. This was a particular brand of case management and, of course, there is a demonstration that is going on now looking at social health maintenance organizations (HMO) which have much more control over the whole range of services. We still await the evidence on that.
As you look at the control group utilization of nursing homes, there was much less nursing home utilization than we had expected, despite the fact that we set out to identify people who we thought were at risk. I do not think we know much more today about how to identify those people than we did then. I think if we started out today to identify this population, we would probably do just about the same thing we did. The population made much less use of nursing homes than we had expected. It is very difficult to say that if we had done the case management differently we would have had different results.
I also find it difficult to know how to target better and come up with a different population. The problems in this kind of program are that it would be very difficult to identify people who if you didn't do anything would go into nursing homes and not get into your program, or people who, even though you did not give them services, would still stay in the community on their own. That is a problem, and I do not think we know how to do it presently.
PETER KEMPER: It seems to me that in the Channeling evaluation, the question it addressed is different from the question you would address as an insurance carrier. To me, your question is, if we had somebody entitled to a benefit, what is the role of the case manager in dealing with that client and in controlling costs? The results of Channeling may be somewhat relevant to that, but the real implications for you have to be drawn out from the experience, the whole set of information on operational experience that came from Channeling and the many other instances of case management that have been tried are very useful. If you wanted to think about what is involved in case management and what the functions are and what kinds of staffing is required to provide that case management, I think the implementation report that George and others did would be very useful to look at in order to understand that.
Certainly you would want to know what it costs to provide a case management service, at least one version or another. There is a report on the costs which actually compares Channeling to some other demonstrations and gives you some idea of what the costs and the caseloads are. Jerry Eggert has been heavily involved in case management. I would certainly talk to him about their experience with the costs of frequent screening of clients for eligibility because one issue is how often do you screen. I think there is some implications to be derived from the Channeling experience on that side of the demonstration.
Also, there were some findings in Tom's community service report that have to do with the effects of case management on the mix of services, with some suggestion that case management resulted in the substitution of lower cost services for higher cost services. In the Channeling context that was more than offset by the increase in services because of the expanded benefit. Given that somebody is entitled to a benefit, then you want to ask does a case manager substitute lower cost for higher cost services. There is some evidence on that you might want to look at. Also, the question of reimbursement rate. There was some experience in the demonstration in negotiating for rates which, once somebody is entitled, you control the rate that is paid. There is some evidence that Channeling was able to get a lower rate than people who were not in Channeling. That may have attributed to buying power or competitive bidding. There is some information there, again, in the implementation report that may be of interest.
THOMAS GRANNEMANN: One of the results of Channeling was to show the people to some extent greater satisfied with service delivery in a case managed system and even showed an impact on some of the quality of life measures. From an insurer's point of view that might mean that is a service people may be willing to pay something for, even if it is not something that saves costs. The other thing to keep in mind is, when we looked at impacts that compared Channeling to the existing case management system. In fact, many of the people in the control group got case management of some form, although not nearly as extensive as Channeling did, through arrangements with home health agencies who do some case management and many other social service agencies that are out there providing services. We do not purport to this is Channeling's impact or what case management does, but really it is an increment in what a couple of case management approaches does.
PETER KEMPER: We probably do not need Channeling to tell you who pays first is an important issue.
THOMAS GRANNEMANN: Indeed, also focusing on the nursing home costs a little bit from the targeting perspective, there was some evidence, although not strong evidence, that people who are already in a nursing home or are waitlisted for a nursing home would be the group most likely to show a reduction in nursing home use. There is also some evidence that to the extent that those nursing home costs were reduced, it is the private pay dollars more than the Medicare or Medicaid dollars that are reduced.
QUESTION: Are there any quality care measurements, such as would people get better nursing home care under case management than not?
PETER KEMPER: There is a self-report data on satisfaction with care arrangements. Uniformly, whether it is the respondent or the impaired elderly person or the informal caregiver satisfaction was increased for the Channeling group.
QUESTION: I am a nurse educator in gerontology at the College of New Rochelle. I am interested in how you selected the population group of frail elderly. Perhaps, in talking about cost-effectiveness, why you did not just select a group that possibly was just 65+ who were in need of comprehensive case management, because the elderly themselves, even when they are not frail, need to have a comprehensive kind of health care management.
GEORGE CARCAGNO: As Peter had mentioned earlier, Channeling was not the first demonstration to look at the issues that we addressed. Some of the demonstrations that had come before had a much broader approach to the population it served. Some attempted to target on people who were at risk of institutionalization. Others did not have as stringent a set of targeting criteria. It was largely on the basis of that prior work that we came to the conclusion that if you are interested in designing a demonstration that has some shot at reducing the costs of long term care that you had to target on a frail elderly population that was at risk of institutionalization. The view was that if you provided services to a much broader group of people, you are basically adding services to the system without getting a corresponding or greater reduction in nursing home use.
There is the social HMO demonstration that is going on now. From time to time people talked kind of wishfully about looking at the Medicare/Medicaid data for the people who were in Channeling to see if anything has changed over the longer term. I do not know if that is going to happen.
THOMAS GRANNEMANN: There is a congressionally mandated study on Alzheimer's, but that is still ahead of us.
QUESTION: The obvious thing is, in terms of preadmission to a nursing home list, take people who have actually signed up to enter nursing homes, or who are in for the first 90 days, and then try to see whether or not in the services that we provide along with case management would have an effect in reducing their use of nursing homes. We have not followed up with the second round of demonstrations that what we have learned from Channeling, and applied them to the next round to see what happens if you found people were actually about to enter. I know in Rochester it is a 6 month wait for people to get into a nursing home, You sign up and there is a waiting list, depending on what your payment status is. It could be anywhere from 6 months to a year and a half before you can get in. That to me is a real key group because they have already decided to go in and they may look the same as a lot in the community who have not decided to go in.
Maybe we are not looking at the right measures to predict who will go in. Maybe looking at the ADL and IADL is not enough, but we need some measures of acuity, which we really have not done.
GEORGE CARCAGNO: Regarding the measurement technology we have available to try and identify this at risk group, we do not have well tested measures. We took our best shot it may be that the measuring technology is flawed and that we need to do something more there. I do not know if there is any work going on in trying to move that along in a serious way. I am sure some people are working on it.
In South Carolina, [with preadmission screening] they got much higher incidence of nursing home use in the control group and with that they broke even. They did not save money. My personal view is that thinking about community care as a way to save money may not be the way to think about it. Figuring it in terms of taking care of people where they prefer to be taken care of; recognizing that it costs money and seeing that people seem to be better off and they prefer it-maybe it is worth doing. Hanging our hat on costs savings as a justification may be very tough to carry through.
QUESTION: I realize it is hard to sell community care as a cost saver, I am not interested particularly in what is going to happen today. I am interested in whether I have to build another 40-50 percent of my nursing home capacity in the next 10-15 years. The issue is whether it is worth the expenditure to expand the community care system, build these new nursing homes, or a combination of both. In New York we have taken a deliberate policy decision not to build nursing homes. At this point what we would like to do is design a system that provides community care at least at the same level of expenditure, hopefully less than that. The cost analysis debate is always nursing home versus home health today. The real issue is how are we going to expand the nursing home, in what proportion are we going to expand the capacity, because we know we have this increasing population. People fall to think about the future sufficiently.
THOMAS GRANNEMANN: On the issue of predicting who will 90 into a nursing home, we did a follow-up study to the Channeling evaluation that, among other things, looked at using the Channeling set to try to determine the predictors of nursing home use. We are able to identify a number of significant predictors of nursing home use. The problem is, however, that the percent of the variation explained is not very great. You can identify a lot of things that would predict nursing home use, but you can not be very good at predicting who will go in and who will not. There are a lot of unobserved things that determine who goes in. Maybe people rapidly deteriorated in conditions that we were not able to observe when we did our baseline, so it therefore was not a predictor for our point of view. There may be other factors that are not observed that the methodology for measuring ADL and other things; it is just not very precise. You can, with the data set, identify predictors of nursing home use, but those predictors are not sufficient to have a high predictive power to explain who goes in and who does not.
PETER KEMPER: This was billed as a data conference and we probably should make sure that those of you who have questions about the data, specifically, or perhaps are thinking about an analysis and would like to get a reaction about whether the Channeling data are appropriate for that, that you get a chance to ask those kinds of questions. Maybe just by showing you that stack of documentation we have discouraged you altogether. It is doable. One positive thing about that stack of paper is that it is documented and Judith deserves the credit for that. That is not true for all data sets.
JUDITH WOOLDRIDGE: I would just like to tell you a little bit about that documentation that I did not do earlier. If you are encouraged rather than discouraged, each of those files that I talked about earlier is documented quite thoroughly, I believe. There is an introduction to each of the reports which just gives you a little background. There is a narrative description of the file, the kind of variables you could expect to find in there, the sample, and so forth. The guts of each volume of documentation is a file layout that tells you what variables are present on the file. Then there is an annotated instrument for those files which are instrument-based, and the instrument is annotated with the variable name annotated in the file. That makes it easier to find the variable when you want to use it.
Every single variable that we constructed for use in the analysis is documented, and the algorithms we used and the kinds of sample flags are documented, how we selected the samples, and so forth. There are descriptive statistics in the documentation, and the first thing that you would want to do if you did get one of the files is to run a couple of those, make sure you come up with the same numbers shown In the descriptive statistics. Of course, there are the physical specifications for how to read the tapes. There is some conventions that are common across all the files. Every variable in the file has a name, and the name indicates the source of the variable. If it is from an instrument, the question number is part of the name. If it is a variable that is constructed from various sources or across questions, it has a kind of a little name that is supposed to mean something, constrained by the number of letters available.
There is a report on survey data collection design and procedures, it is one of our final reports prepared by Barbara Phillips and others, which, if you are thinking of using any of the instruments, is very useful. I would particularly recommend that if you are thinking of using the caregiver files, because the caregiver files have some curious things in them in the way the sample was developed which you will be able to understand better if you have read the fielding procedures manual.
PETER KEMPER: If you were particularly interested in the hospital and nursing home use data coming from Medicare and Medicaid claims, that is documented in the hospital and nursing home use report.
QUESTION: I have a data related question. I know that there is a difference between sites.
My question relates to variations between sites. I know in Florida they have about 23 nursing home beds per thousand over 65, wherein Rensselaer, one of your sites, where they have 60-70, among a 3 to 1 difference in the nursing home bed supply. My question is, does not it indicate that there is a different risk factor of entry in a nursing home if you have a big supply difference? It is the same thing on the hospital side, there is a big difference in the hospital beds per thousand. I am sure the AAPCC, I guess, in Miami is almost 200 percent of the national average, whereas in other places, like in Rochester, it is going to be 70 percent of the national average. The probability of getting into the hospital is much less in Rochester but more in the nursing home than it is in Florida, just vice versa. How could that be accounted for in the analysis or in the data?
THOMAS GRANNEMANN: In the analysis, we have controlled for site differences, although that was done simply with site dummies, so you can control for differences. In doing analysis with the files, you are able to control for differences across sites. If your question relates how well can you learn something about how variations in nursing home supply affects outcome, the short answer is, we only have ten sites, so you do not have very many observations on which to see what effect supply has. We did, however, look at that issue by replacing the set of site dummies we used with variables representing nursing home waitlists, patterned in the levels of use of community-based services in the sites, and did not find any significant difference where nursing home waits were high and low. Because there are only ten sites, we did not have very much power to detect such a difference.
JUDITH WOOLDRIDGE: Just as an aside, we were interested in the extent to which hospital and nursing home use might substitute for one another because of accessibility. We used that as a dependent variable and we did not have any difference between the treatment group and the control group when we looked at either the hospital or nursing home use variables.
QUESTION: In the sites that you studied, did you collect any information on supported housing that is available, such as board and care alternatives?
PETER KEMPER: As part of the interview, we collected self-report information on where the person lived. That included whether or not they were in supportive housing. That is available. There were also provider records collected from personal care homes. I do not know exactly what from that got on the public use file. Craig Thornton looked at those housing arrangements at MPR. In the report there are data on how many people were in supportive housing, how many people were in housing with congregate meals, and some information on costs. In general, the community data were of two sorts. One were data that we obtained from published sources that were available at the local level. In some cases we relied on state data. The second category was data that we collected in implementation interviews that research staff did on the sites. Those are documented in the process reports. I do not think there is anything particular on supportive housing in that.
There may be something in the interviewer training manuals, because they did do some inquiries in order to help the interviewers figure out whether someone was in a personal care home, I believe that the survey staff collected information on what personal care homes were in the community, at least what the main ones are. It's not like a data file or anything like that. It would be in memoranda.
JUDITH WOOLDRIDGE: The information on use is in the formal community services file on supportive housing and personal care homes.
QUESTION: Can you describe the selection procedure for the provider extracts? The 20 percent sample and the 10 percent sample, and cross validity checks between self-reported service use, the provider extracts and the Medicaid data that you have?
THOMAS GRANNEMANN: The 20 percent sample is simply a random sub-sample, for which we have the most information about community-based service use. In that sample you have got information from individual interviews, information from Medicare and Medicaid records, and extracts from the provider. Medicare and Medicaid records provide comprehensive information on the use of community-based services for those covered by those two payment sources, but obviously we omit a very important piece of it. The individual interview data was collected in a little bit different way, because the individual interview focused on a 2 week period that does not correspond directly to the period of observation for the Medicare and Medicaid records. We asked people what the caregiver did from the individual interviews where they said what this caregiver did, and they checked off a list of items that they did. We were able to identify from what the individual reported that the individual did and we were able to classify people, as whether they provided medical care, personal care or housekeeping type of services. That did not necessarily relate back to what we find in the provider records extracts where we have the provider's identification of what that caregiver was. For the provider extracts, obviously, if people did not identify a caregiver they will not get picked up in there, so there is a little bit of slosh back and forth there. In general, the impacts that you estimate across the data sources were consistent.
QUESTION: In extracting the records, was it only service use, or did you extract any kinds of clinical information?
GEORGE CARCAGNO: We did not get any clinical information. It was service use and basically, cost and billing records.
QUESTION: Is the bill in code?
GEORGE CARCAGNO: It is not a code. Those records were extracted according to our procedures, not a special system.
Keep in mind that most of the providers we went to, were filling in what was not available on Medicare and Medicaid, so a lot of them were community service providers rather than medical providers. Nursing homes were covered as well as community service providers of various kinds.
JUDITH WOOLDRIDGE: In case there is any confusion, let me add that there are two kinds of provider record extracts. There was a 20 percent sample of the full research sample for which we did provider record extracts for community services. We also used provider record extracting of hospitals and nursing homes for those instances where we felt, that the Medicare or Medicaid files were not going to provide us with the information we needed for eligibility reasons. For those, which we call provider record extracts by exception for hospital and nursing homes, that was not the particular sample. It was purely based upon the individual case. If the individual had a hospital or a nursing home use and we did not think we were going to get this information from the claims, then we went out and did a provider records extract.
GEORGE CARCAGNO: What that means is that for nursing home use, basically, for the whole sample you would have all payors for nursing home use and use under all payors, both public and private. For the community services, you would have Medicare and Medicaid payors, and for the financial control model of Channeling services that those payors covered, which are clearly not all home care. For this 20 percent sample, it drops to 10 percent from the 7-12 month period, and it was not continued after 12 months, you had the other payors and the other home care services. Provider records extracts differ between the home care services and the nursing home and hospital services.
QUESTION: Describe your procedures for missing data and whether those were constant across all your files and across the variables, and how that affects merging files.
As one example, if you had for members of your sample omitted data on particular items and functional status, you could estimate either by regression of everything else that you have what that would be, you could plug in a sample mean, you could just simply throw the person out for that particular variable, obviously with different implications as to how you can merge variables.
JUDITH WOOLDRIDGE: We did various things. For our standard control variables that we used in all of our analyses and all the substantive areas, we were concerned not to be throwing out cases because they were missing important standard control variables. For that reason we sometimes substituted screen data for baseline data and vice versa when we were trying to create those baseline variables. We did in a very few instances use procedures like means, I believe. That was a very infrequent procedure. Generally speaking, we only used data that was there. For particular analyses, people may have done one or two extra things. By and large, if an individual did not have the data that we needed they are coded missing and they are dropped from the analysis. As a result you will see that different analysis samples which are basically drawn from a kind of specific sample will vary in sample size a little bit because one of the dependent variables was missing. Obviously, we never substituted any information if the dependent variable was missing.
PETER KEMPER: I think it is important to distinguish between source variables and constructed variables. With respect to source variables, we adopted a principle that we wanted to provide the raw data. The source variables, I think with the exception of confidentiality procedures are as they were reported.
There were range checks done as part of the initial data cleaning procedures, and you can look at those to see if there is anything that you would not want to have had done. There is nothing you can do about those, and I do not think those were invasive procedures by and large.
Beyond that, the source data are there. If you do not like the constructed variables, those are documented. You can go back to the source variables and do what you want to do for the missing data.
For the Medicare and Medicaid claims data, the things that were derived from records data we did not include source data. Those come in hundreds of thousands of individual claims. Those have been organized as one record per person and laid out over the time period. That all hospital days have been aggregated and so on. All that work has been done for you.
JUDITH WOOLDRIDGE: We did not do imputations. By and large we made the assumption that I there were data it was complete and if there were no data there it meant zero rather than it was missing.
There were just a very few exceptions where in doing consistency checks in the secondary data we just felt that we had such a garbled claim that we simply could not use it, but that was rare.
QUESTION: Why was your sample so frail and so old when your screening criteria is so much younger?
JUDITH WOOLDRIDGE: This evaluation was intended to provide community services to a group at risk of nursing home entry to see whether case managed community services could be a cost-effective alternative to nursing home entry.
We jointly, the government and ourselves, had extensive discussions about who was likely to enter a nursing home. Since, clearly, a demonstration like this is very expensive, you want to target your group as much as possible to maximize the possibility of identifying any differences that the program provides. Statistically, you want to have as many people as possible who are likely to enter a nursing home.
I can add at this point that a very low proportion of our sample, in point of fact, went into a nursing home, more than the average for people over 65 and slightly more than average for the people of the age group, but by no means a very large proportion.
QUESTION: Was the 6 and 12 month interview for the caregivers 6 and 12 months after their initial interview, or the 6 and 12 months of the elderly persons?
JUDITH WOOLDRIDGE: Both, as a matter of fact. The caregiver baselines were administered as soon as possible after sample baselines. The 6 month and 12 month anniversary dates were the same for the sample members and their caregivers.
For sample members who joined the sample in the first 6 months, regardless of their caregivers, there was no caregiver baseline. We only did caregiver baseline for sample members who were baselined after November 1982.
QUESTION: In other words, the people who started out with a caregiver did not have their caregiver interviewed?
JUDITH WOOLDRIDGE: The early sample did not have caregiver interviews at all. The last half of the sample did and, in all cases, the caregiver and sample member follow-ups relates to the same time period.
The questions on the caregiver follow-up relate to the same general four areas I just summarized for the baseline.
There are some peculiarities about the caregiver samples used for analysis which I do not want to go into, except to say that it is possible for there to be more than one caregiver for a given sample member in the follow-up file. You could have more than one primary caregiver which is, by definition, wrong, but that was what we actually did.
Mostly, it was just one caregiver but, occasionally, there are two. Both of these caregivers can be related to the particular sample member by means of the sample member's ID which appears on both the records of the caregiver and of the sample member.
QUESTION: Why was Channeling undertaken? What do we learn from the results?
GEORGE CARCAGNO: The demonstration was motivated by the concerns that, I think, were spoken to by Steve Grossman and Bob Helms, namely, that long term care costs are a significant expenditure of public funds. The government pays about half of the national costs of nursing home care.
When you look at the data about where we were headed, it is quite clear from those demographic data that the elderly population is going to become an increasingly large proportion of the nation's population, and that the oldest old are increasing at an even faster rate, and they are the people who are at greatest risk of nursing home admission.
You look from where we are now where there is a lot of concern about the costs of nursing home care falls primarily on the Medicaid program. The state and federal governments have developed various stratagem to limit costs. Some of them, I think, are pretty obviously short term things. You can not hold down the nursing home bed supply forever. You have got to deal with the problem at some point.
What motivated the demonstration was that set of facts and projections looking ahead and wanting to test whether this particular approach to community care could prove to be a cost-effective way of delivering services to people in need of long term care in the community rather than in institutions.
When you look at the results, essentially, we found it cost more to deliver Channeling services, and it seemed to be primarily because there is relatively low use of nursing homes by the population that participated in the project than had been expected.
There was a substantial increase in the use of formal community services, particularly in the financial control model, reduction in unmet needs, some improvement in measures of well-being, but there was not any of the savings that had been hoped for in terms of reducing nursing home use because nursing homes were not used very much by anyone in the sample, and we know that by looking at what happened to the control group.
When you look at that, I think it is very important, in terms of trying to draw some conclusions from it, to keep in mind what we were testing. We were not testing something compared to nothing. We were testing a particular intervention called Channeling with its own variant of case management that had been developed.
There are other ways you can do case management. For example, case managers could have control over medical and nursing home expenditures, as well as community-based care. That is an approach that is being taken by the social HMO demonstration.
In Channeling, we just have case managers who dealt only with community services. You have a particular variant of case management that was being tested.
What is especially important is that what we were comparing Channeling to was not the absence of case management, but the case management and community care programs that were already out there in the communities that the control group had access to.
We were testing an addition or an increment to the existing system. What do you draw from that? I think one thing you do not conclude is that we should abandon all support of case management or community services because that was not what we found.
On the other hand, I think it would be very difficult, using the Channeling results, to argue for a substantial increase in the public funds available for community care except to keep pace with growth and population, or to argue that community care is a cost-effective intervention or way to save money. The results simply will not support either of those kinds of positions.
On the other hand, there are other ways of approaching case management that could be tested that could be more effective. We do not know. The socialization models are being evaluated. We may learn something from that.
If we had selected a less frail population, would the results have been different? I am sure they would have been different, but Channeling was not the first such demonstration that was attempted. It was, in fact, preceded by a number of community care demonstrations that had a wide variety of target populations that participated.
Largely on the basis of that previous work, the people who were involved in designing the demonstration came to the conclusion that if we were interested in trying to see if this intervention could be cost-effective, the only way you could do that was to try and identify people who, if you did not provide such services, would go into a nursing home, rather than taking a more preventive approach or trying to get people much earlier in the process of functional decline.
That was based on what had come before us, and what you find is that if you have a population that is much healthier and less frail than Channeling's, you can provide community services, and you provide them to a lot of people, you can probably improve well-being, but you would not see any of the offsetting cost reductions, certainly in the time periods that we were looking at in this demonstration.
If you could convince somebody to do a demonstration like Channeling or find some means of tracking people over a longer period of time so you could evaluate what happens over a 5-10 year period to these more preventably oriented programs. I can not speculate what the results would be but, certainly, it would be a different demonstration than the one we did.
QUESTION: I presume you were not looking at health outcomes, then. You were looking only at satisfaction levels? Were there any other health-related outcomes that you focused on?
GEORGE CARCAGNO: We looked at use of health-related services.
JUDITH WOOLDRIDGE: Just a self-reported status variable. The excellent, fair, poor type of question that you see in the National Health Interview Survey (NHIS).
THOMAS GRANNEMANN: We had the ADL measures.
QUESTION: You had ten different sites, and in each site you have a control group as well as the treatment group. Did you find that Channeling was cost-effective in any of the ten sites? Has anybody done site-specific analysis?
GEORGE CARCAGNO: We looked at site-specific results, and the story was essentially the same for all of the sites. There is always the possibility or concern that you have got something wonderful happening over here and something terrible happening over there, and you put them together and it appears that nothing is happening.
We looked at the impacts at individual sites and, as I said, the story that we were looking at there is essentially unchanged.
We also looked at subgroups of the population.
THOMAS GRANNEMANN: I think the subgroup analyses tend to, in general across the board, confirm the findings of the demonstration overall. We did not find any subgroups for which impacts were dramatically different for the program, say, of money or had bigger differences in impacts by a wide variety of subgroups we looked at.
In particular, to follow-up on the question on whether a less impaired group could have been identified earlier for which things might have come out differently, I think this subgroup analysis shows no evidence that a less impaired group would have a higher impact.
Of course, you can not say what happened outside the sample we looked at, but there was certainly nothing in trends along impairment levels of the groups we looked at that would suggest that. In fact, to the extent we do have some subgroup evidence of differences, the group that we show the biggest potential for reducing nursing home use, and again, that is not a large reduction, is the group of people who are already waitlisted or are in a nursing home at the time of randomization. That result tends to suggest that, rather than focusing on a broader, less impaired group, the important thing to do here is focus on a group, not necessarily more impaired, although that group probably is more impaired, but a group that has already made the decision to go into the nursing home. That operational thing that you do not get when you simply check off characteristics on a screen might be the thing that would help you better predict and target towards a group that was better. That kind of thing could have been done through a preadmission screening program, something where you get people at the point where they apply for a nursing home as opposed to Channeling which got its people, in general, from the community; anybody who was looking for help for services from within the community.
To the extent there are any results in the subgroup, I think it is just going toward a more impaired, more nursing home oriented group rather than toward a broader group.
QUESTION: I was surprised to see that you tried to evaluate your project after 18 months, and I had the idea, knowing the experience in the Netherlands where I am from, that you need the first 1-2 years to got the people together who are providing the care, so that they know each other. Then you can evaluate your project, is it for scientific reasons you wanted to evaluate after 18 months, or is it because lack of money?
GEORGE CARCAGNO: There is a very long report that talks about the planning and operational experience of the projects.
We used, by and large, established local agencies as the host agency to run the projects, so that we were not starting up with new agencies in the community. They were local agencies already well known to the local service providers.
There was a planning period of about 18 months that preceded the start up of the program, during which time local staff were hired, and they, in turn, were making contacts with the local service providers, explaining what the program would be, both with an eye to getting agencies who would refer people to their program, and also with an eye to identifying providers who would provide the community services to their clients.
People were enrolled over about a 15 month period for the research sample, and then we followed people for a 12-18 month period. The elapsed time on the calendar was just over 2 years.
By identifying a target population that seemed to be at imminent risk of institutionalization, the necessity to take a longer period of time seemed to be reduced.
I would have felt better to have had a longer period of follow-up, and that was a decision about what resources were available, and also how long we could continue doing data collection.
QUESTION: Do you have a sense of what might have happened if you had been able to continue data collection longer?
GEORGE CARCAGNO: Given the mortality rate, there would not be too many people left to analyze. One-third of the sample died within 12 months.
QUESTION: I have tracked nursing home data in Connecticut, and that is higher than the proportion of people who enter a nursing home.
JUDITH WOOLDRIDGE: That is right. We had a low proportion enter a nursing home. I think it is important to say that these were people with an acute precipitating event, by and large. They had a very high hospital use for the 2 months before they came into the program. Between 50 and 60 percent of them had been in a hospital.
Another thing I would like to mention is not only was the cumulative mortality rate going up very rapidly over time, but nursing home use was also increasing. We were not beginning to see any increase in effects over time. The difference between the treatment group and control group seemed to be pretty constant at 6, 12, and 18 months. That was something that we looked at to see whether, if we had gone, did we have any hint if we had gone on longer, there was likely to have been a bigger effect, and there really was not.
QUESTION: Why did you amalgamate the Asians and the Indians into one ethnic group?
JUDITH WOOLDRIDGE: For reasons of confidentiality.
QUESTION: We have several students who were quite anxious to look at the experience of the Asian elders.
JUDITH WOOLDRIDGE: There were not very many, so they would not really be able to do much of an analysis with them. It would be possible to identify some of those individuals, we felt, particularly at some of the smaller sites.
QUESTION: Do you have some advice about the print quality of the documentation. It is exceedingly poor, and I can not read the column numbers, for instance. I presume that you must have given a good quality print to National Technical Information Service (NTIS).
JUDITH WOOLDRIDGE: By and large, we did. One file documentation is really quite poor, though, and we have just redone some of the pages for that. That was the hospital and nursing home file. Some of the constructed variable pages were extremely pale. Everything else we gave was quite legible when we gave it to them.
QUESTION: Do you have some advice about merging longitudinally? If we wanted to merge the base 6, 12, and 18 months, we were going to have an absolutely gigantic file.
JUDITH WOOLDRIDGE: We have 14 public use files because we did not want to produce one gigantic file with all that on it. We knew that people would have particular interests. We produced individual analysis files in preparing our final report which were quite small, and sometimes we worked with more than one file in particular substantive areas just to keep things simple.
Extract only the variables that you think you are going to use from the files, keep the numbers of variables down.
QUESTION: Did I understand you to say that, in terms of the formal community services, the information is available for only the clients of Channeling, or they are available for both the control group and treatment group?
JUDITH WOOLDRIDGE: They are available for both the control and treatment groups. There is one data source that was only available for clients, and that is the financial control data from the financial control sites.
QUESTION: I am just wondering if I can get some idea of the magnitude of the relative costs between the control group and the treatment group. That is, you said it was not cost-effective.
THOMAS GRANNEMANN: The costs were not close in the sense we were quite confident of the result. The results of no cost savings are very robust. There are some differences. When we looked at cost by source of payment, as a government funded project, we focused primarily on Medicare and Medicaid costs as a primary objective, and it was quite clear that those costs went up.
In some of the subgroup analysis we found that nursing home expenditures from private sources, and in some cases, in community services went down. That was more than made up for by the Channeling expenditures on services.
We are quite confident to say the government is not going to save money by implementing a program like Channeling.
GEORGE CARCAGNO: You recall the model differences we talked about earlier, the basic case management model and financial control model, where the latter involved substantial expansion of community services.
Overall the basic model treatment group cost 6 percent more than the control group. In the financial control model it was about 18 percent.
If you look at government costs, they increased by about 14 percent under the basic model and 28 percent under the financial control model.
QUESTION: Do those costs include the costs of the overhead of running the project in any way?
GEORGE CARCAGNO: It is everything. This is for the entire treatment group and the entire control group. It is averaged over everybody, including the people who died at various stages or dropped out of the program.
We are talking about control group means of $1,300 in the basic and about $1,600 in the financial control compared to $1,400 in the basic treatment group and about $1,900 in the financial control group.
QUESTION: I want to pursue a little bit more the observation that, after a year, 30 percent of the clients entering the Channeling program had died and, depending on which program, lost 33-39 percent at 18 months.
We have in Connecticut an 8 year longitudinal set of all nursing home patients. We know what proportion of patients die at various given times, by year of admission and their length of stay.
Obviously, the chance of dying increases the longer you stay. The number 29 percent after a year sticks in my mind. You speak of the Channeling population as having a precipitating event that brings them into Channeling. What could be a more precipitating event than whatever brings you into a nursing home.
Is it possible that there is an underlying progression here of whatever is going to happen to elderly in this group that is going to happen to them whether they are in a nursing home or whether they are in Channeling?
With respect to something like mortality that maybe the type of service that is not going to make much difference. We have the Connecticut data for the most recent 3 years by ADL level and continence level, so it would be interesting to compare the functional level of the Channeling population with the Connecticut nursing home population. We do not have any of the social support data that you have. It would be interesting to note whether they are functionally similar populations.
GEORGE CARCAGNO: We did examine the characteristics of the Channeling sample against people in nursing homes from the National Nursing Home Survey (NNHS), and there are a lot of similarities.
One thing to keep in mind when you look at aggregate nursing home data is that you get a lot of short stayers in nursing homes, which will tend to have some effect on whatever aggregate means you are looking at for that population.
Your suggestions to look at that subset of nursing home population that bear similar characteristics to Channeling's population is a good one.
There are lots of people in the community who look like the people that are in the nursing homes, and sometimes it is a precipitating event that puts people into nursing homes.
In many cases it is a long, slow, worsening of condition and, finally, usually family members are unable to cope. It is not necessarily a particular thing, but an accumulation of many factors and a cumulated stress and burden on the caregiver.
THOMAS GRANNEMANN: There is a lot of information from the analysis that we have done that is available in the reports.
Information on the ADL's of Channeling clients at baseline or at follow-up is as hard to get as simply opening up the report and looking at it.
QUESTION: What is the cost of the tapes?
AUDIENCE: It is approximately $500.
JUDITH WOOLDRIDGE: Just as a matter of interest, did you request one file or did you just request one tape, the tape with all the files?
AUDIENCE: All the files.
QUESTION: Do your numbers parallel impairment assessments in terms of the NHIS?
THOMAS GRANNEMANN: The ADL scales that are used? If you are talking about impairment measures, those do parallel.
QUESTION: What happens now in the ten sites that participated? Has the project stopped and all the elderly people are at home now without community services?
Did you ever have problems saying, I allocate this older person to the control group and that older person to the treatment group, because sometimes you think it is better that this person goes to the treatment group because of human values.
GEORGE CARCAGNO: Actually, those are quite closely related questions.
There was a great deal of concern about those issues. The project had what is called an Institutional Review Board (IRB) that was concerned with the measures that we took in terms of protecting the people who were involved in the demonstration.
With respect to ending the project, one of the things the IRB did was make us think at the start of the project about how we were going to end the project, and that was very useful and important to us.
When people applied to the project, the whole randomization procedure was explained to them, and the fact that the project was of limited duration was also explained to them.
The sites were required to develop operational plans they submitted to the federal government about how they were going to conduct the operation in their sites. With this, they also had to submit a termination plan, which the first one, we were now looking at something that was happening 3-4 years down the road.
As time passed, those termination plans were refined. At one point during the demonstration we stopped enrolling new people, and we allowed the caseload to get smaller.
In eight of the sites the program essentially continued under other auspices. The funding sources shifted from being federal demonstration money to some combination of regular Medicaid program or 2176 waiver services.
In the two sites that essentially closed down, the transition termination care plans were done, and the cases were transferred to other agencies.
With respect to the ethical issues involved in random assignment, I am not terribly troubled by that. We are basically looking at a situation here, the control group gets access to whatever is available in the community that, in the absence of this demonstration, they would be able to get. The treatment group is getting a treatment that, in this particular intervention, was relatively benign and not particularly risky.
We were not likely to have adverse consequences, but you could make up stories about trying to keep people in the community might involve greater risk than if they went into a nursing home, and these kinds of risks were part of the informed consent process. We brought those things in and explained to people what they were facing.
The people who are not getting the treatment are not being denied access to anything they would otherwise have available to them. The control basically has access to whatever was there if the project had not come along.
We had people from ten sites who were in the business of delivering services to people who felt these kinds of issues very keenly. They were convinced of the appropriateness of the research design and the randomization procedures, and felt comfortable with it. All of them were not at first, but they were in the end.
QUESTION: How was it determined that people who were let into the treatment group or control group were eligible for nursing home level of care by either Medicare or Medicaid? Were they screened?
GEORGE CARCAGNO: Not in any formal sense. in several sites they do that informally applying their state level of care procedures. They did in New York and Florida, and very high proportions of the caseload met that criteria.
QUESTION: Did not only 15 percent of your controls go to nursing homes? Is this evidence of bad targeting which is what goofed up the demonstration from a cost point of view?
THOMAS GRANNEMANN: We did a fairly extensive analysis, actually, an additional piece of analysis beyond the original design of the Channeling evaluation to look at the targeting issue to try to decide whether there was a subgroup within that we had somehow missed along the way. We developed several different ways of targeting and tried to test that to see if we would get any different results.
Part of that analysis involved looking at an analysis of what predicts nursing home use. We found that using the information that was available from the baseline instrument and the screen, we were able to identify a number of factors that predicted nursing home use and some that were fairly strong, in particular, in the group that was already in a nursing home or were waitlisted was a very big predictor.
We identified a number of predictors of nursing home use. Even using multiple regression analysis to take account of all the information that was available from the baseline screen, we were not able to explain a large part of the variance in nursing home use.
QUESTION: Not on an individual basis?
THOMAS GRANNEMANN: Not on an individual basis. I think you put your finger on an important issue. We were not able to do that with this type of a screen and, as I think I suggested earlier, what that may suggest is that you can not identify people very well from these kind of criteria by simply giving them an interview and checking off their characteristics.
It may need to be operational types of things, such as pre-admission screen processes that select people who are on their way in who are there because of all of these unobservable things that cause them to go into a nursing home. We are not at the point of being able to look at somebody out in the community and say, you are going to go in and you are not, and be able to target that way, even with all the information we have got.
QUESTION: How many got in through the criterion collapse of informal support versus those that got in simply because they had unmet needs?
THOMAS GRANNEMANN: I do not have any idea.
QUESTION: Does that suggest to you that maybe we ought to think more about caregiver crisis than unmet need in terms of risk of nursing home admission?
THOMAS GRANNEMAN: One of the things we did look at in our targeting study was the role of the caregiver in nursing home placements, and there was some evidence there that caregivers do make a difference in terms of who goes into a nursing home. Particularly whether there is a caregiver available, whether there is a close family relationship to the individual, has a close tie and is available to provide informal services.
That is saying caregivers are important in determining nursing home placement, but we were not, however, able to identify differences in the impact of Channeling, depending on the caregiver's role. Even though caregivers play an important role in nursing home placement, caregivers are not the deciding factor in whether Channeling is effective or not.
QUESTION: Did the results indicate the extent to which informal care was shifted to paid services by the presence of the project?
THOMAS GRANNEMANN: The basic analysis that we did was looking at informal care and formal care, and we find that formal care increased by a fair amount substantially in the financial control site where extra funding was available for that.
One worry was, of course, that while that was simply going to be a one-for-one substitution, those services were being provided on an informal basis.
We did not find any evidence that that was a significant worry. There was a modest 5 percent reduction in informal care, even in the financial control model where we had a 40 percent increase in the community-based services that were used.
GEORGE CARCAGNO: A lot of people had grave concerns that there would be substitution. The reductions in informal care that Tom referred to were really observed only in the financial control model and, essentially, in people who were visiting caregivers rather than those caregivers who were living with the elderly sample members.
You saw that people in situations where they were in the same household, there was no reduction in the amount of informal care that was given. There was a reduction in the care provided by visiting caregivers, and most of that reduction was from friends and neighbors rather than family members.
It was people on the periphery of the informal care network rather than the primary caregivers where we observed those reductions.
That is an important finding in terms of the kind of fears people have about substitution.
QUESTION: How can we get copies of your reports?
JUDITH WOOLDRIDGE: With respect to our reports, there is an Executive Summary of the Final Report. It does tell you what the full conclusions of the project were. In the back of that report is a list of all the technical reports available about the project. That includes technical reports on substantive areas and it also includes things like instruments. If you are interested in getting copies of the interview forms themselves, they are documented, you can get those in there. Of course all the data base documentation I think is listed in the back there as well. Most of these things are available either from MPR, NTIS, or the Department of Health and Human Services (DHHS).
There is also a document called the National Long Term Care Channeling Demonstration Abstracts of Reports [http://aspe.hhs.gov/daltcp/reports/chanrpts.htm], which gives a one paragraph description of each report. This is also available from DHHS.
There is a methodological report which describes the kind of special analyses we did to address issues. We were concerned about what to do with potential design problems, potential problems of attrition in the data and so forth.
GEORGE CARCAGNO: There is a very extensive data set here that covers the formal community services that people used and their costs. There are other formal services such as nursing homes, hospitals, physician services, those are also in the data base. Also units of service, amounts of service, their costs, and who paid. Were they paid by government programs, which ones, or by private individuals or private insurers.
There are data both in terms of the amounts of formal care and of informal care, that were provided. That includes personal care as well as financial support and we also in an interview with the informal caregivers themselves collected information about their satisfaction with the services that the elderly sample member was receiving, their stress and burden, and the amount of care that they provided.
There is also information on several quality of life measures and unmet needs, also data on physical functioning. Of course there is data on mortality.
I think I pretty much covered physical functioning. There is a great deal of data, all of which were used in the evaluation itself, so that you can look at the final research reports. There is the Executive Summary, there is a summary report of the research findings, and there are a number of technical reports on each of the specific areas. There is a separate report on the impacts of formal caregiving, a separate report on formal community service, use and costs, a separate report on the benefits and costs of Channeling, a separate methodological report, and so on.
There are a large number of reports, all of which are listed in the end of the Executive Summary.
QUESTION: What advice do you have for other researchers on conceptualizing and measuring quality of care?
GEORGE CARCAGNO: There were several questions that addressed the issue, but they were all based on the self-report of the elderly sample member and/or their primary caregiver. We did not set out to attempt some direct measurement of the quality of care.
If we were going to set out to do what we did, where quality of care is one component of it and given the resource constraints that we faced, would we do the same thing again? We would, and I do not think we would make any major changes to those questions. I certainly think that if you got hold of the instruments, both the follow-up instrument of the elderly sample members and the caregiver instruments, the questions in there on quality of care are ones that, I think, you could use.
THOMAS GRANNEMANN: One issue in measuring quality of care for this group is the use of proxies as respondents. We were forced to use proxies as respondents whenever the sample member could not respond. If you are doing this kind of work with this type of a very impaired population, how do you deal with situations where you are forced to use proxy response? I think that is a problem you probably have to solve for yourself, because I do not think we have the perfect answer to that because we did use proxies when necessary.
QUESTION: I have a question about the reliability of functional status data. How did you deal with that problem?
GEORGE CARCAGNO: We trained all the interviewers. At the baseline, the treatment group was interviewed by the staff of the Channeling project themselves and the control group was interviewed by MPR staff. Then all of the follow-up interviews were conducted by MPR interviewers, both treatments and controls.
From a research perspective, we would have preferred to do all the baseline assessments ourselves to remove any questions about comparability of data, but there were compelling clinical reasons why it made sense for the sites to do the baseline assessment.
We were convinced that they needed to do it themselves in order to gain the clinical insights that they needed to develop their plans and so on, and that it was not feasible to even think about having two baselines, one that we did and one that the sites did.
We did a small sample of validation interviews where we went out ourselves and reinterviewed a small sample of the people in the treatment group that had been assessed by the Channeling staff.
That turned out to be a better idea in theory than in practice, and one of the reasons was that it was very difficult to compare those two samples because there was passage of time, it was 1-2 weeks later, particularly if you were dealing with people who were coming out of the hospital.
QUESTION: You could not go together?
GEORGE CARCAGNO: There are potential logistic problems, but I think there were some clinical issues in terms of the burden and so on.
The validation sample did not turn out to be a terribly useful way to look at these issues. We did do some separate analyses of the comparability and there is a report on the comparability of the baseline data that you can take a look at.
What we found was that on a small number of variables there was not comparability at the baseline between data collected by the site staff and our own staff. What we did in those cases was to use the similar or same variables from the screening instrument, which was comparable for everybody, because that was administered to everybody at the point of application.
In terms of differences across sites, we trained all the interviewers, so they had the same training, as the site staff did on the baseline.
We conducted fairly rigorous quality control reviews and monitored the quality of the data that we were getting. To the best of my recollection, there were not large differences in the functioning measures across sites, although there was one site that had a relatively less disabled population than all the others. I think that involved measuring something that was going on there rather than a difference in methodology.
QUESTION: It seems to me there is an awful lot of interest among private insurance companies in doing something with alternate care, as well as among legislators on the Hill, in the general concept of case management. I have seen a number of volumes that have come out on Channeling. I do not recall seeing a particular document on case management. If there is not one, is one planned?
GEORGE CARCAGNO: There are several ways you can get at the issue of case management. With respect to the costs of case management, some of those data are in the final report, and more detailed information in the formal community service use and cost report. With respect to more qualitative descriptions of the kind of case management that took place in Channeling and what the case managers are like, what kind of training they got, and what their caseloads were and so on. There is a report which is actually a two volume report, called The Planning and Operational Experience of the Channeling Projects [http://aspe.hhs.gov/daltcp/reports/proceses.htm], which has a couple of chapters that talk directly about case management and the processes that were carried out, and there are also some chapters on the organizational structure of the Channeling agencies and how they were organized and conducted themselves.
QUESTION: What did you learn from case management and Channeling? What is the best way to do it that would be applicable to private insurance companies?
GEORGE CARCAGNO: In the planning and operational experience report, we did try to provide some consensus views about how to do case management.
The focus there was not necessarily looking at private insurance companies providing a case management benefit as part of a package of long term care, community and other kind of benefits, but was more from the perspective of case management, probably more nearly in a public program context. A lot of that is very useful and applicable to the kinds of case management that a private insurer might consider making available.
Case management is an ill defined concept, Channeling was a particular variant of case management that had a particular range of authority and limits on that and so on. There are other ways you could do it.
In Channeling, for example, the case management staff involved some combination of people with social work backgrounds and people with nursing backgrounds.
Some people say you should have only nurses, and other people say you should have only social workers. Just as with most, with those bitter debates, we concluded the truth was somewhere in between. The Channeling projects I think bore that out because those that started out that did not have nurses found that they needed nurses fairly quickly and arranged to have nurses on the staff.
I think by and large, the case managers as a group in Channeling were more social service oriented than having nurse backgrounds but there were always nurses available, either as consultants or supervisors, or as other case managers. In cases where having that kind of expertise was important, those resources were available.
We tried to define case management in a way that was comparable within the demonstration so that when we saw that there was substantial case management use among the control group. We set out to identify whether people were getting what we called the comprehensive case management that was available from Channeling or case management as, again, everybody does case management.
Visiting health agencies do case management--to some extent as intensively as Channeling--only on a small proportion of their total caseload, whereas Channeling attempted to do it for all of the caseload.
Depending on which model of Channeling you look at, 10-20 percent of the control group received case management similar to Channeling.
QUESTION: I have been having a discussion with my boss about costs and ways to compare costs between community-based care and institutional care. I wonder if you can narrow down what can and can not be done with Channeling to address that question.
THOMAS GRANNEMAN: The Channeling data set is probably the most comprehensive of the data sets that we have talked about at this Conference in terms of measuring costs.
We have records data on all the important costs. We have provider records extracts for all payers. We have got interview data that allows us to make amputations for costs of housing and expenses of maintaining people in the community. If you are thinking about doing an analysis that is broader in terms of cost of maintaining a person in the community versus the cost of maintaining people in an institution, it is got the pieces of information there.
Your question addresses how you go about comparing that, because what Channeling did was not to randomize people, put some people in a nursing home and put some people in the community, and see how much they cost.
The Channeling intervention was really an intervention that added community services and then let people voluntarily decide with these additional services can you make it in the community; does it make sense for you to stay there. Then we looked to see if there was an impact.
The question is, can you look at the data in a way that allows you to get around this problem? In part, the answer is no; there are inherent differences between the two groups. People who are in the community have self selected themselves into the community, and people in nursing homes have self selected themselves into the nursing home. Even though we can control statistically for some of the differences between those with baseline characteristics, we know there is a lot that is unexplained. The Channeling data and indeed other data sets do not have all the characteristics you need to have a good prediction. As a result of that, they do not give you a result similar to what you would get if you randomly assigned people to nursing homes.
One thing we did do in a follow-up study to Channeling was try to look at what were the costs for the groups of people, treatments and controls, in the community and in institutions. We did an analysis that compared for those who went in, what were the costs. In this case we looked at a 1 year period and found the costs were higher, of course, for those people who went in a nursing home and that those cost patterns, the differential between community and nursing homes, differed between the two groups.
What we are able to calculate from that is what diversion rate would be necessary to make a program cost-effective. Even though you do not know exactly what the costs would have been, it gives you some idea of the bounds on which you can put limits on those costs.
I think in that sense, Channeling can put some bounds on the question you are addressing but can not really answer the question precisely.
QUESTION: I am wondering if it is feasible to arrive at a consensus of whether the nursing status is home nursing, intermediate nursing, rehabilitative nursing, personal care, and whether there is a possibility of defining those categories and incorporating such categories into the various data bases so that one can analyze not merely whether a person moves from community into institutional, but the various levels of institutional care that a person moves through.
JUDITH WOOLDRIDGE: Theoretically, I think the answer to your question is yes. Right now I am working on another project to do with long term care assessment and management, a program the Commonwealth of Pennsylvania is running. They have in that state, in seven of their counties, introduced a screening process for anybody who is going to go into a nursing home. That is to say, anybody who is Medicaid and anybody else who wishes to apply to be screened for their nursing home eligibility.
In that process they go at least some of the way in the kind of distinction you are making. This program that they are running is intended to divert people from nursing homes wherever possible. What they do is, they have quite an extensive assessment instrument which asks a lot of questions about functioning and informal supports. On the basis of those issues, they classify individuals as needing skilled care on a long term basis or on a short term basis, intermediate care similarly.
They also classify people as needing community care. They actually have a program so that people can be classified as needing that specific program and they have other levels of care, for example, people who need personal care or domiciliary care programs. They actually have gone quite a long way in making that kind of an assessment.
My guess is that given the quite extensive detail in the Channeling data base on informal supports and functioning at different levels, different points in time, that you could do some such thing and make such an assessment even post hoc at this stage.
QUESTION: I was wondering if you could comment on the data on informal supports.
I heard from other sessions there apparently was a formal service substitution for that. Would it be a data problem or an effect of Channeling itself?
The reason I am asking is because we intend to work with insurance companies and they constantly ask: What is the cost of that going to be if they are going to start paying for home care?
GEORGE CARCAGNO: I think for the purpose of looking at substitution effects, one great advantage of the Channeling data is that you have imbedded in it data on a behavioral response to providing services that are paid for by somebody else to this population.
It is not simply a cross-section. We have a report and data on the impacts of Channeling on informal caregiving.
What those data show is that there were small reductions in the amount of informal care that was provided. Those reductions occurred primarily from reduced levels of care given by visiting caregivers rather than caregivers who lived in the same household and those reductions occurred primarily among friends and neighbors rather than family members.
In the financial control model we observed these reductions in the amount of informal care. They were relatively small and were concentrated among people who seemed to be at the fringes of the informal care network.
One of the jobs of the case managers was to provide support to that informal care network, and different sites did that in different ways.
Some sites had training sessions that they gave for caregivers and support groups, although not a lot of people actually came to those things.
There was an attempt to provide that kind of support, and it manifested itself in support group kinds of activity in one or two sites.
The case managers had as one of their responsibilities trying to keep the caregivers going. It is difficult for me to say to what extent that accounted for the outcomes that we observed. It certainly did not hurt.
QUESTION: I think that the last thing you said is important. Formal care may not be a substitution for informal care but complementary.
GEORGE CARCAGNO: That is right. Although, as I recall, you could think through several different models of how this could work. Maybe in the short term, there is some substitution but it turns out that because of that substitution, caregivers stay at the job longer than they otherwise would.
We did not really observe that kind of thing going on, although we thought it might happen if there was substitution in the short run. The results were pretty much the same over time.
QUESTION: Could you help me with some evaluative judgments about your screening tools?
There is a lot of interest on the Hill on screening, how you select people for long term care services. I am asking it in that context as well as in the context of a program like the 2176 waivers.
GEORGE CARCAGNO: When you talk about screening I think there may be two useful distinctions to keep in mind. One is the instrumentation itself and the other is who it is you administer it to.
The Channeling screening instrument was designed to determine who was eligible for Channeling and it was largely administered by telephone to over 11,000 people.
One of the things you have to face up to, with respect to the whole screening process, is you basically want some levels of screening to take place, because you have resource constraints. We certainly did. We started out with that telephone screen.
In effect, there really was another level of screening that took place at the point of assessment, although very, very few people were turned away from the program at that point based on the in-person assessment, but in principle there was a second level of screening that could take place. Anybody that was turned away because of the assessment results was still in the research sample because it was after randomization.
It is my opinion that, even though we designed the screening instrument many years ago, the technology of measuring and identifying people at risk has not advanced a great deal and that something like the screening instrument would be used today if we were to do it over.
We screened a fairly broad group of people who voluntarily came forward and applied to Channeling, were referred by either a hospital at the point of their being discharged, by a community service agency, a social service agency, or community care agency.
It has been suggested that one might have a much better success rate in terms of identifying people at risk if you administered this kind of screen to people who were in the pre-nursing home admissions process in terms of Medicaid. That is what they did in the South Carolina demonstration, and in fact, identified a much higher proportion of people who went into nursing homes as a result.
Programmatically, if the way you got community care in the future, say, was to go through a preadmission screening process for nursing homes and I that is the gateway to publicly financed community care, over time you might observe a shift in who, in fact, was coming forward to apply for nursing homes. It is the people who really want to get community care who go and apply to the nursing home to get community care. You may not end up with the same population that the people in South Carolina got, you may end up with a population that looks more like what we had in Channeling, overtime.
THOMAS GRANNEMANN: I did a follow-up study to the Channeling evaluation that produced what will be a report to Congress on identifying people at risk of institutionalization and really addressing the targeting issue. My understanding is that this report will be released shortly. It is in the Office of the Secretary getting a final approval right now.
It may be best to focus on the group of people who are already in a nursing home or on their way in. These preadmission screening places may be the place to do this.
We looked at a broad range of predictors based on ADL, IADL and individual characteristics. We were not able to develop a checklist or screening device that allows you to predict who will go in and who will not, and therefore for whom community-based care would be a cost-effective alternative to nursing home use. it is very difficult to make a community-based service program cost-effective simply by giving them a screen that picks them out.
The solution may be along the lines of these process things focusing on people going into nursing homes rather than a simple checklist. I do not think we have a perfect screen that identifies people from these programs that can be cost-effective.
On the other hand, the difficulty of predicting who goes in I think, from the insurance standpoint, may be something of a plus rather than a minus. Whereas the government would like to identify people that will go in to provide services to them, insurance companies may like to see that these things cannot easily be predicted and therefore are more insurable. The selection bias problem will be as great as it would have been if the thing had been predictable.