National Invitational Conference on Long-Term Care Data Bases Conference Package (continued)

I. The 1982-1984 National Long Term Care Survey

THE 1982 AND 1984 NATIONAL LONG TERM CARE SURVEYS: THEIR STRUCTURE AND ANALYTIC USES

Kenneth G. Manton, Ph.D., Demographic Studies, Duke University, Durham, NC

Korbin Liu, Sc.D., Health Policy Center, The Urban Institute, Washington, DC

I. INTRODUCTION

With the rapid increase in the U.S. elderly (65+) and oldest-old (85+) populations, considerable concern has emerged over the amount of future acute and long-term care (LTC) services that will be required by that population, and of the nature of the mixture of federal, state and private programs necessary to respond to that need. One of the areas of service needs with the projected greatest rate of growth is that for LTC services. The National Nursing Home Surveys (NNHS) conducted by the National Center for Health Statistics in 1963, 1969, 1973, 1977 and, most recently, 1985 (with a follow-up in 1987) have provided considerable information on the institutional component of LTC services. More recently, because of the rapid growth of the elderly and oldest-old populations, considerable interest has emerged in home LTC options, both because of concern about the economics of institutional care and because of humanitarian concern about the level of dependency and quality of life in many LTC institutions. Until the advent of the National Long Term Care Survey (NLTCS) there was no major nationally representative survey with specially designed instrumentation that dealt explicitly both with the health and functional problems of the community dwelling disabled elderly, the home LTC options (both formal and informal) available to meet those problems, and the ability to substitute, for a specific target population, home and institutional care. The 1982 NLTCS filled this gap in our knowledge and provided considerable information on which both to plan the nature of required services and to develop private insurance products to pay for such services. The 1984 NLTCS provided a basis upon which to examine changes in the home LTC populations and to examine the trajectory of service needs at the individual level.

The 1982 and 1984 NLTCS are detailed household surveys of persons aged 65 and over who manifest some chronic (i.e., 90 days +) Activity of Daily Living (ADL) or Instrumental Activity of Daily Living (IADL) impairment. The sample for the surveys was drawn using a two-stage procedure. In 1982, 36,000 names were drawn from the Health Insurance Master file. These persons were then screened by either telephone or personal visit to see if they manifested an ADL or IADL impairment of 90 days duration (or which was anticipated to last at least 90 days). When the screen identified a person living in the community with a chronic impairment, a detailed household interview was conducted which gathered information on medical status (diag- noses), functional status (presence of ADL, IADL or other functional impairments and equipment or caregivers utilized by the person to deal with his impairments), income and assets, health service use, use of federal services, housing and living arrangements. Of particular note in the survey were detailed questions on the number and type of informal caregivers. Institutionalized persons were not interviewed in 1982.

In 1984, a different sampling procedure was utilized. First, all persons who reported chronic disability on the screener or who were screener-noninterviewed due to institutionalization and who survived to 1984 were interviewed regardless of their 1984 functional status. Second, from the original 25,541 persons who did not report functional impairments in 1982 (and who were not institutionalized), a random sample of 47% (~12,100 persons) was drawn and subjected to the same screening procedure as in 1982. Another difference from 1982 was that 5,000 persons who became 65 between 1982 and 1984 were screened so that, in addition to having a longitudinally followed sample in 1984, the full cross-section of persons aged 65 and over in 1984 could be evaluated. In addition, persons who were in institutions in 1984 were interviewed with a specially designed instrument containing a number of questions on institutional use in the interim period and the sources of payment for those services. The interview instrument used for the community population was nearly identical in 1984 to that used in 1982. A final major difference between the 1982 and 1984 surveys was that a "next of kin" interview was conducted for persons who died between 1982 and 1984. This interview collected extensive data on the medical service use and expenditures surrounding death.

These surveys conducted in 1982 and 1984 cannot be fully exploited without considering their linkage to another important data source--Medicare Part A bill files from 1980 to 1985 on Medicare reimbursed hospitalization, home health services and skilled nursing facility use. These files contain bills for individual service episodes and provide a continuous history of the exact date of service use and the amounts reimbursed by Medicare for those services. Each bill in this interval is linked to the corresponding sample person who participated in either the 1982 or 1984 survey.

The dual cross-sectional and longitudinal nature of the 1982 and 1984 NLTCS and the linked Medicare service use files allow us to analyze a broad range of questions. First, they provide an impressive array of data on the community dwelling chronically disabled elderly, a population group at high risk of both extensive acute and LTC service needs. These data can help us estimate the need for LTC services, the actuarial basis of, and markets for, LTC insurance products, the role of "spend-down" for Medicaid qualification for LTC benefits, and the impact of informal caregivers on meeting the national need for LTC services.

In addition to describing the social, economic, functional and health status characteristics of a large (~5 million persons) population group at high risk for significant Medicaid and Medicare services (and for the development of private insurance options to provide parts of those services), the data files provide considerable information on the pattern of utilization and outcome of Medicare Part A (and potentially Medicare Part B) services. That is, the continuous time Medicare service history of individuals whose detailed health and functional characteristics have been determined from the surveys is available. This linkage can allow questions to be examined such as the substitution of home health and skilled nursing facility (SNF) services for acute hospitalization after the introduction of the prospective payment system (PPS) in order to assess how the reduction of the rate of hospitalization and the shortening of hospital LOS affected the nature of the use of these other service options. This can be used to evaluate the impact of such Medicare changes and to design changes, as necessary, in the provision of hospital, home health and SNF services by Medicare.

A third major area where these data can be important is in the study of changes, both for the individual and for the aggregate, in terms of health and functional status. Because of limitations on the availability of longitudinal data, the design of service and insurance options has been constrained. The availability of large amounts of nationally representative data on long term (two-year changes) in health, functional, economic, and social status is an important and unique feature of this data set.

A fourth major use of this data set is to help provide national estimates of LTC service needs by combining national distributions of functional limitations from the survey with very detailed data from select populations in a wide variety of LTC demonstration projects and waivered programs. That is, detailed data on Medicaid and private payment for LTC services are available in the demonstration projects along with data on the effects of those services (and modifications of those services) on a wide variety of social and health outcomes. The problem is to extrapolate those findings from the multiple, local select populations in the demonstrations to the national population. Because the instrumentation of the NLTCS has many measures in common with many of the demonstration projects, there is a lot of information on which to base the extrapolation.

In this introduction we have very briefly reviewed the rationale, structure, content and some potential areas of application of the 1982-1984 NLTCS and linked Medicare files. In subsequent sections we will explore specific technical issues concerning the quality of the survey data and its analysis in more detail.

II. TEMPORAL ORGANIZATION OF THE 1982-1984 NLTCS

We briefly described in the introduction how the samples were constructed for the cross-sectional and longitudinal components of the 1982-1984 NLTCS. A more systematic review of this can be made from Figure 1.

In Figure 1 we present a time line for the 1982-1984 (and a proposed 1988) NLTCS which identifies the dates of the surveys, the dates for which Medicare Part A data are to be collected and the survey instruments applied at each date. The 1988 survey is currently in a tentative planning phase. All other elements of the survey and service use data collection are in-place except for the proposed collection of death certificates for decedents over the period 1982 to 1989.

We can see that the sample was originally "frozen" as of April 1, 1982 and contained 35,789 persons. Survey work for round 1 began in June and continued to October 1982 and produced 6,088 responses from the 6,393 persons identified as chronically disabled. In addition to the survey of disabled persons a separate survey of 1,925 caregivers (1,626 continuing caregivers and 299 caregivers who discontinued care) who were identified as having provided care to a subsample of the 6,393 persons who screened into the survey. In addition to the 6,393 community dwelling elderly disabled, 1,992 persons were found to be in institutions, either before April 1 (N=1708) or who became institutionalized between April 1 and the screening date (N=284). Thus, though no interviews in 1982 were conducted of institutionalized persons, and we cannot identify Medicaid and private pay institutionalized persons from the Medicare files, we can identify the total set (N=1992) of institutionalized persons from the screen.

On April 1, 1984, the sample components of the 1984 survey were fixed and field work again conducted between June and October 1984. At this time three survey instruments were applied to nearly 10,000 persons. One instrument was essentially the same questionnaire as was applied to the 1982 community dwelling, disabled elderly population. A second instrument was the institutional questionnaire which allowed us to examine the retrospective reports of the institutional histories of all persons institutionalized on April 1. These reports covered all facets of institutionalization (Medicaid and private pay as well as Medicare). The third type of survey was the "next of kin" questionnaire on health services received during the terminal phase of the illness for deceased persons who were reported as disabled in 1982 and who died in the two-year intervening period. Medicare Part A data cover all service use from January 1, 1980, to currently October 1986.

To get a better understanding of the size of the sample components and their change in sample status between 1982 and 1984, examine Figure 2.

We see several different types of numbers in the figure. First, above each block is a single number which represents the number of persons in that state at that time. Thus, there were 25,541 persons (of 31,934 who were not institutionalized and who responded to at least the telephone screen) who were determined to be non-disabled, community dwellers in 1982. In 1984 there were 14,130 such persons--a number much smaller than the 25,541 because only 47.4% of the 1982 non-disabled group was screened.

Under each block is a set of four numbers. For 1982 these describe the number of persons in that state who ended up in one of the four receiving states in 1984. Thus, 9,220 persons who were non-disabled, aged 65+ and sampled (of the 47.4% of people who were non-disabled in 1982) turned out to be non-disabled in 1984. Of this group, 1,562 became disabled and were interviewed in 1984, 348 persons became institutionalized in 1984, and 970 died.

The corresponding numbers for 1984 tell us where persons in those states come from. Thus, of the 6,182 persons receiving the detailed survey in 1984, 1,562 were not detailed in 1982, 4,114 were people who were disabled in 1982 (but who were not necessarily disabled in 1984--thus long term improvements in health and functional status can be tracked over the two-year period), 53 persons were interviewed in 1984 in the community who were institutionalized in 1982, and 453 persons became disabled and were interviewed from the sample of 4,916 persons drawn from those aged 63 or 64 in 1982. The deceased block shows that a total of 3,219 persons died from the four sample components over the two years.

One issue that arises in evaluating the 1982-1984 NLTCS is that, in order to increase its precision, a two-stage sample capture procedure was used to identify community dwelling disabled persons. Thus, it does not provide detailed survey data on various groups. For example, though it divides the total sample in 1982 into the set of all community disabled persons, the set of all institutionalized persons, and the set of all non-disabled, non- institutionalized persons, it provides no detailed data on non-institutional, non-disabled persons. The characteristics of such persons are described in the recent 1984-1986 Supplement on Aging (SOA)-Longitudinal Supplement on Aging (LSOA) of the National Health Interview Survey. The SOA-LSOA provides far less information on the disabled extremely elderly (only 876 persons over age 85 were identified in 1984) and both smaller numbers (~1,500) and less detailed information on the functionally impaired and the formal and informal care services they receive. Thus, each of these surveys complement one another but provide very different samplings of target populations and very different (and specially tailored) instruments. Likewise, the 1985 NNHS is being followed-up in 1987. That survey gives us much larger numbers and more specialized information than the NLTCS. However, it does not contain a true admission cohort. Thus, the three surveys may be reviewed as complementary in terms of sample coverage and instrumentation. This is illustrated in the coordinated time lines of the three surveys presented in Figure 3.

The three sets of longitudinal surveys provide a powerful, and cost effective base battery of surveys to monitor the health and functional status of the elderly population and its consumption of acute and LTC services.

III. RESPONSE RATES AND PATTERNS OF PROXY RESPONSES

Two useful measures of the quality of data in surveys are the rate of non-response and the patterns of response of proxies. This is especially true for the NLTCS because (a) it had large numbers of extreme elderly persons for whom obtaining survey responses is known to be difficult, and (b) the survey had a longitudinal dimension meaning persons have to be tracked over time.

To evaluate these issues we provide two basic types of data. The first are the non-response rates for various sample stages in both 1982 and 1984. There are two types of non-responses to be considered. The first type are the so-called "C" type "non"-responses. Actually, this is a slight misnomer in that these people did not respond because they did not qualify for the sample. The major reasons for not qualifying were (a) death, (b) institutionalization (in 1982 only), and (c) movement out of the sample area. Thus, this type of failure to respond does not represent what we typically view as non-response. The type C non-responses are described in Table 1.

The second type of non-response was labelled "A" type non-response. These represent non- responses due to either failure to locate or contact persons, refusals, or failures of the proxy to be able to respond. The frequency of non-responses is described in Table 2.

We can see that the frequency of non-response was very low producing response rates that are extremely high, both for the screening and detailed interview stages, in both 1982 and 1984. The response rates average about 96%. Thus, neither the longitudinal nature of the survey nor the high proportion of the extreme elderly seems to have caused problems in the level of response to the survey.

The second aspect of the response question is the pattern of proxy respondents. This is indicated in Table 3 where, for both 1982 and 1984, and for different levels of disability, we provide the number of responses (a) totally by sample persons, (b) totally by proxy, and (c) for combined sample person and proxy respondents.

As would be expected the proportion of proxy responses increases as the reported disability level of the individual increase. Furthermore, we see that about 500 persons in 1982 and 1984 had a proxy respondent due to senility. Indeed, the diagnosis of senility was derived from the proxy when the person was found incapable of responding due to cognitive impairment. Of the roughly 6,000 interviews in both years, about two-thirds were totally from sample persons. The pattern of proxy response (i.e., its increase with disability), the small number of non-responses due to proxy failure (Table 2), and the large proportion of non-proxy responses provide an indication of the appropriateness of the use of proxy responses in the survey.

IV. SAMPLE DESIGN AND ITS EFFECT ON THE ANALYSIS OF THE SURVEY

An important factor in the analysis of any survey, but one that frequently generates confusion, is the appropriate use of sample weights in the analysis. This is because sample weights play different roles in different stages of the analysis and because there are several different methodologies for dealing with the effects of sample design in analysis. The issues become more complex in the current study because of its longitudinal nature.

The first set of issues involves the role of weights in various stages of analysis. One stage of analysis has to do with the testing of statistical hypotheses using the survey data. The basic problem is that the samples are not simple random samples but are probability samples, i.e., different populations are drawn with a pre-specified probability to increase the precision of estimates for certain rare populations. Furthermore, in some sample designs, the samples are drawn from spatially designated clusters to reduce costs. Since persons in each cluster will tend to share certain social-economic and residential characteristics this means that their responses will tend to be correlated, i.e., each person cannot be viewed as providing an independent response.

In the NLTCS these problems are minimized because the sample design is relatively simple. In 1982 the population was only stratified on age, sex and race. In 1984 there was the additional complication that only 47.4% of the non-disabled community dwelling persons were screened-adding an additional weighting factor.

The problem in analysis is that stratification and sample clustering (clustering has little effect in this design) affect the estimate of error variance which is used in our test statistics to determine if a particular hypothesis should be accepted or rejected. The analytic problem is to determine how the sample design affects the variance of our parameter estimates. There are two analytic approaches to this problem. The first is to use some model of randomization to increase the error variance to provide a conservative adjustment to our test statistics. There are several analytic computer programs extant that do this for continuous variables in simple regression models. However, given the simple nature of the study design certain simple calculations can be used to adjust variance estimates. This was illustrated by the Census Bureau for the 1982 cross-sectional sample. A table of adjusted factors is provided in Table 4 below.

In Table 4 are the parameters for two regression equations. Both were obtained by regressing the estimate on the variance of the estimate for each of two groups, i.e., "blacks or persons receiving Medicaid" and "all others." What one does is take the number of persons having a particular characteristic in 1982 and multiply the square of that number by parameter a, and the number itself by parameter b, and add the two products. The square root of this number is the standard error of the estimate. To illustrate, in the 1982 survey there were estimated to be 1,190,764 aged persons requiring personal help in bathing. The formula described above is, symbolically

Standard error of x = ax2 + bx * f

If, as for the example, f = 1.0, then the calculation is

Standard error of x = (-.00004027)(1,190,764)2 + (1025)(1,190,764) * (1.0)

or, 34,109.

Thus, the one standard deviation (68%) confidence interval, is ±34,109 or 1,156,655 to 1,224,873. The 95% confidence interval would be ±2*(34,109). For the confidence interval of differences one uses

Standard error of difference = s2x + s2y - 2r * (sx * sy)

where sx and sy are the standard errors of the two estimates to be compared and r is the correlation coefficient (which can often assumed to be zero). Alternatively, for the 1982 tables, the standard errors of both numbers and percentages were calculated. These are presented in Table 5.

The numbers in these tables need to be multiplied by the appropriate "f" values in Table 4. We do not yet have similar tables for the 1984 survey. However, knowing the size of the various sub-samples in 1982 we can present the coefficient of variation for different of the 1982 subsamples (i.e. 10,250 is the number of non-disabled persons planned to be screened in 1984; N=6,089 was approximately the number of persons interviewed in 1982; 1,712 was the number of persons institutionalized before April 1, 1982, and 856 and 428 are half and a quarter of that number). These numbers are presented in Table 6.

An alternative approach to adjusting error variance estimates for sample design effects is based upon the realization that many of these design effects may be of substantive interest. Thus, an alternative approach is to explicitly model the design factors as part of one's analysis so that design effects are explicitly represented. Such an approach has the advantage of helping us better understand the mechanisms generating the phenomenon but the disadvantage of requiring that the correct model be developed. Though it may seem tedious and difficult to search for the "correct" model, rather than using a "general" model of randomization, it should be realized that only by producing the correct model can one really generalize the parameter estimates made beyond the particular sample, i.e., either to the general population or in forecasts of future needs. Thus, in many more situations than normally realized, the search for a model based adjustment for complex sample design effects is a necessity.

A second analytic stage where sample weights are used is in "post" weighting, i.e., where one wishes to recombine parameter estimates for sub-groups to produce the parameter estimates for the total population that was sampled. This usually involves re-weighting the data to reflect the inverse of the probability of selection. This is actually a purely algebraic procedure that is independent of the methods to calculate the effects of sample weights on statistical inferences.

V. EXPLOITATION OF THE LONGITUDINAL NATURE OF THE SURVEYS AND LINKED MEDICARE SERVICE DATA

In this section we discuss how the cross-temporal nature of the file can be exploited in several types of analyses of transition. The longitudinal nature of the file can be exploited in several ways. The first, and most basic, is simply to analyze changes in characteristics between 1982 and 1984. This is illustrated in Table 7 for disability and institutional status.

Down the left hand side of the table we present the percentage distribution of persons (weighted counts) by their status in 1982. Across the table we provide the status (including death) of persons in 1984. We see that the largest proportion of persons remain in the same state that they were in 1982, or, for the most disabled, they experienced high death rates. Interestingly, there are also sizeable numbers of persons who had long term (2+ years) functional improvements. For persons with 5 to 6 ADL's in 1982 nearly a quarter improved status by 1984. Given the high mortality rate for persons with this level of impairment (~36%) and only a moderate level of institutionalization, this suggests that the functional impairment is driven by an acute morbid condition that often produce death, but also often result in improved LTC functional status.

This table can also be stratified by other variables. For example, in Table 8, we have decomposed changes in disability by age.

In Table 8 we see that there are decided shifts with age to higher levels of disability and that, at advanced ages (85+) there is a lower proportion of persons who improve and a higher proportion who become more disabled.

The above discrete state, discrete time description of transitions can become unreliable as one attempts to stratify those transitions by more than one additional variable (e.g., stratify them by age and marital status and the cell sizes become small). To deal with this problem it is necessary to use some type of regression model where the transitions are made functions of a set of covariates. Usually, in order to conduct such modeling, some assumptions have to be made about the form of the dependency of the transitions on the covariates (e.g., in the Cox regression model it is assumed that each covariate has a proportional impact on the transitions).

Such models simply describe the probability of a change in state, as the states are described in the survey, over the two-year period. Obviously, this does not tell us how to exploit the rich data on the use of different types of Medicare Part A services over the two-year period. These data can be exploited to different degrees by models with different degrees of sophistication.

One approach is to simply aggregate different types of service use over specified periods of time. The problem with this approach is that it can produce severely misleading results due to the fact that events like death--or even other types of service use--represent constraints on the amounts of specific types of services that an individual consumes in a given period of time. Thus, a person who was in a nursing home for a year and used 20 days of home health care in a two-year period, used home health care twice as fast when eligible as a person who, in a two-year period used no other type of service.

To account for these differences in exposures, one must calculate life table type measures of ex- posure for each of the types of services. In these life tables one can deal with the constraints that death, or consumption of other types of services, represents. An illustration of these types of life tables is presented in Figure 4.

What we have done is to plot the proportion of persons who stay a given number of days in hospitals in 1982 and in 1984 for (a) community disabled persons, (b) persons in institutions at the time of the survey, (c) non-disabled, non-institutionalized persons, and (d) a residual group of non-respondents. To read the table note that the vertical axis describes the proportion of all hospital admissions in the specified year who were still in the hospital after X days. The horizontal axis represents the hospital LOS in days. We see that, in both 1982 and 1984, all groups experienced a decline in LOS and that non-disabled persons had the shortest stays, community disabled the next shortest stays and institutionalized persons the third longest stays, i.e., the hospital LOS was correlated with severity of the chronic conditions the person had. In calculating such tables one has to be careful to post-weight observations, and to deal with "exposure" weights based upon when the survey was conducted in the year, or misleading results may be produced.

These life tables can be used to describe how different types of services are used in a specified period of time after adjusting for exposure differences. To combine that longitudinal information with data from the surveys on chronic functional and health problems can be done by certain multivariate procedures.

VI. FILE STRUCTURE AND ACCESS

The public use form of the NLTCS is found on two separate tape files. The first is a rectangular tape file which contains, on a person based record, all survey data for persons in the longitudinal two cross-sectional samples (N=25,401). A rectangular file was created in order to facilitate the processing of the tape by persons with limited programming resources. That is, the original version of the file released from the Census Bureau has six record types (sample person, caregiver/children, and household members, separate by survey year). Each of these files has different numbers of cases. In order to use the data from the caregiver file with data on the individual, one had to write a program linking across the two record types. In the current file all record types are already linked together in one large record. Clearly we have attempted to trade off between a less compact form of data storage (there will be many more blank fields in such a rectangularized file) for greater ease in processing.

The second type of file is that containing the Medicare Part A bill files. This file is not person based since there can be up to 120 bills for a given person. Rather this file is bill-based and retains all the data (e.g., exact beginning and end dates) for each service episode. Certain standard edits have been performed on this file. Within each individual the bills are sorted by transaction type, then admission date. Thus, to use the bill data with the survey record one must perform the linkage operation using special survey identifiers that are found on both files.

The two files are available separately in EBCDIC form on 6250 bpi tape with standard labels from NTIS. The documentation includes instructions to the interviewers, copies of all survey instruments, instructions on sample weight calculations, and detailed codebooks on the rectangular survey and bill files. In addition there is a technical write-up on the creation of the file and its logical structure.

VII. SUMMARY

The 1982 and 1984 NLTCS are large nationally representative surveys of a target population at high risk for high levels of both acute and LTC services. In addition the survey files are linked to Medicare Part A records. These surveys and linked files represent an extremely rich data source describing all the health and functional transitions of individuals in this population, detailed characteristics on the person at the time of the survey and detailed data on informal caregivers. These files can provide extremely valuable information on the service needs of the LTC population, changes in those needs and associated acute care needs.