Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
4 Design for a National Health Care Data System The foregoing discussion makes it clear that the panel believes that existing data systems are not adequate to yield the range of statistical infor- mation needed to fully address the complex health care issues facing this nation. NCHS has proposed a general strategy to expand and extend its provider-based surveys to be more responsive to data needs and has taken the first steps toward geographically integrating them with the National Health Interview Survey (NfIIS). These proposed changes, however, do not satisfy the urgent needs for data to monitor the health care of the nation. The panel's appraisal has demonstrated that, as currently conceived, the center's plan does not provide the capacity to address important questions about the interrelationships between the health status of individuals and the patterns and cost of health care services they receive from a broad range of providers and service settings over time. The long-term agenda, therefore, requires consideration of further, more fundamental restructuring. Since a design is dictated by the objectives of a data system, this chap- ter first briefly states the panel's perception of the objectives of a National Health Care Data System and then presents the panel's recommended de- sign framework for achieving these objectives. The benefits and limitations of the proposal are discussed, and issues that require further research are identified in the discussion. The chapter concludes with a suggested imple- mentation schedule. 62
DESIGN FOR A NATIONAL HEALTH CARE DATA SYSTEM STATEMENT OF OBJECTIVES 63 The objectives of the National Health Care Survey proposed by NCHS should be driven by the goals and objectives of the nation's health care system, a primary goal of which is to deliver health care services effectively and efficiently to the entire U.S. population. A simple statement then of the objectives of a National Health Care Survey would be to provide statistics on a continuing basis that reflect the condition of health care in the United States, particularly with respect to the critical issues of access, quality, and cost. Such a survey should have the capability of obtaining data related to the characteristics and health status of individuals and their demand for, and use of, health care over time and across a broad range of providers and service settings that can be linked at the individual rather than the aggregate level. Data are needed to support analysis along each of the following . . dimensions: Access: there is need for information on the supply of health care providers (including information on numbers, distribution, and type) and the demand for and utilization of health care services by specific segments of the population (e.g., racial and ethnic minorities, migrant populations, the uninsured, the elderly, the disabled) and by other char- acteristics such as health care insurance coverage status. Quality: there is need for information concerning the health and func- tional status of the patient prior to and after treatment, the appropriate- ness of treatments or procedures provided, the degree to which health care providers resolve problems, and the satisfaction of the patient with the process. Cost: there is need for information on expenditures for health care services by type of provider, the cost of treatment for an episode of illness for a specific diagnosis, its distribution among provider types, and the source of payment for care provided, including how much is paid by insurance and how much is paid out-of-pocket. In addition to obtaining information from providers, information is needed about health care consumers. For example, if a consumer has had contact with several provider types for a given episode of illness, some way has to be found to obtain that information in a coordinated way. Some people may have had no contacts with the health care system at all; this group is impor- tant to identify, and information is needed on their health care needs. For example, answers are needed to the following questions concerning the health care of consumers:
64 TOWARD A NATIONAL HEALTH CARE SURVEY . What is their health status, in terms of disability, functional status, and chronic conditions? What are their unmet health care needs? What are the individual's perceptions regarding access to health care? What is the person's health insurance coverage, including no cov- erage at all? What is the person's use of health care providers over a defined period of time? For each episode of care? For what conditions? At what cost, both out-of-pocket and covered by insurance? What are the person's knowledge, attitude, and practices related to health promotion and disease prevention, such as knowledge of risk factors, attitudes toward risks, and risk behavior? What are the person's demographic characteristics? No one single survey can provide answers to all these questions. An integrated data system is needed with linkage capability at the individual level that includes a variety of approaches, including surveys of specific provider types and service settings, follow-up of patients seen for specific conditions by specific types of providers in specific settings, longitudinal surveys of the household and nursing, home populations, and possibly sur- veys of episodes of illness. In addition, there is need for improved collaboration, coordination, and integration of health care data collected by NCHS and other agencies of the Department of Health and Human Services. Care should be taken to ensure that health care surveys carried out by NCHS and the other agencies use standard definitions and classifications and do not unnecessarily duplicate information and that other needed steps are taken to facilitate integration of data sets for analysis and dissemination. DESIGN CONSIDERATIONS The panel was guided by three basic considerations in proposing a design framework for a National Health Care Data System: (1) Any design proposed has to meet existing information needs, espe- cially those currently not being met, such as those identified in Chapter 2. iSome of the methodological issues associated with accessing, linking, and integrating data sets are discussed in Wachter and Straf (1990).
DESIGN FOR A NATIONAL HEALTH CARE DATA SYSTEM 65 (2) As the nation's health care system changes, so too will the informa- tion needs that give the system direction. The design for the Na- tional Health Care Data System must therefore possess sufficient flexibility to be capable of responding in a timely manner to chang- ing information needs. This flexibility should include the capabili- ty to use special supplements to ongoing surveys to quickly provide data in support of emerging information needs and to support the conduct of ad hoc special studies. (3) The design proposed should minimize any disruption to the current NCHS survey program to facilitate implementation of any opera- tional changes for NCHS that the design might require and, as far as possible, preserve continuity between existing and new data out- puts. A FRAMEWORK FOR A NATIONAL HEALTH CARE DATA SYSTEM With the above design considerations driving its thinking, the panel has proposed a design strategy to meet the various data needs that have been identified in its deliberations. Some of the currently unmet needs relate to cost of care for patterns of illness, outcomes, and patterns of provider utili- zation during an illness, access to provider care, functional status, severity of illness prior to and after treatment, insurance coverage, information about specific population subgroups (e.g., racial and ethnic minorities, migrant populations, the uninsured, the elderly, the disabled), as well about other categories (such as providers not currently covered by NCHS provider sur- veys and rare illnesses). As stated in the preceding chapter, the event-based follow-up design proposed by NCHS, as an extension of existing NCHS provider surveys, would allow NCHS to expand its provider-based data gathering to include outcomes of hospital and private practice physician care, two important and currently unmet data needs. There are major issues, however, surrounding feasibility, operating procedures, and informed consent that need to be re- solved before the design could be implemented. Certain important data needs remain unmet with the adoption of the event-based follow-up design. Some examples are the ability to produce person-based estimates of illness, access to care, patterns of care for epi- sodes of illness, and associated costs. Still needed, therefore, is a broader strategy that addresses these unmet data needs in an effective manner. Such a strategy is presented below, not so much as one specific design but as a design framework for a National Health Care Data System within which a variety of survey approaches linked to the NHIS can be developed, ~.,
66 TOWARD A NATIONAL HEALTH CARE SURVEY and from which a broad range of health-related information needs can be met. The panel believes that, taken as a totality, its recommended course of action represents the long-term direction in which the National Health Care Data System should evolve. Although NClIS has taken the first important steps to expand on the coverage and content of the provider surveys and geographically linking the provider surveys with the National Health Interview Survey sample design, in the panel's judgment the NCHS has not gone far enough. The panel pro- poses in this chapter a comprehensive approach for the National Health Care Data System that goes beyond the expansion planned by NCHS and that links the design and operation of the National Health Interview Survey with the provider surveys. Such a strategy would provide not only new data but also a basis for linkage of data on a population-based sample of individuals with data on their health care providers, thus further enhancing the benefits. Since a central feature of the panel's proposed design framework is to link the NHIS design with the operations of the National Health Care Data System, it is useful to review some of the design features of the existing provider surveys to see what degree of integration may already exist. As shown in Table 1 in Chapter 3, the primary sampling units for the National Hospital Discharge Survey (NHDS) and the National Ambulatory Medical Care Survey (NAMCS) are subsamples of the NHIS primary -:~mple. This design implies that samples of providers for these two survey currently being chosen exclusively from lists of hospitals and medical practices that are located in some of the same areas in which NHIS interviewing is being conducted. This geographic integration, in tum, must also imply that some degree of operational coordination already exists, or should exist, among these surveys. Moreover, data collection in all these surveys is being con- ducted by the Bureau of the Census. Extension of this level of design integration to nursing homes is currently being planned for the next round of the National Nursing Home Survey (NNHS). The key elements or components of the proposed framework are to: (1) Change the origin of the provider samples from listings of provid- ers and service settings maintained by NClIS to identification of providers and service settings by respondents to the National Health Interview Survey (at least for those providers not in the inventories maintained by NCHS). (2) Sample from NHIS respondents to gather longitudinal person-based data on the health status and health care received by individuals along with the associated costs and expenditures. (3) Modify the sampling design of the National Nursing Home Survey to collect longitudinal data on health care utilization by the institu- tionalized population not presently covered by the NHIS.
DESIGN FOR A NATIONAL HEALTH CARE DATA SYSTEM 67 (4) Sample from NHIS respondents and collect longitudinal data on the process of health care, the utilization of providers, and the costs and expenditures associated with episodes of illness. NHIS-Based Sampling of Providers As discussed in Chapter 3, one of the features of the NCHS plan is to extend the types of providers to include alternative sites of health care covering hospital and surgical care, ambulatory care, and long-ter~ care. The panel is concerned that current NCHS plans for expanding the coverage of the National Health Care Survey are not as broadly based as they should be with respect to national statistics on providers of health care. The panel therefore proposes a method for extending the coverage of providers of health care that is capable of covering the full range of providers, both physicians and nonphysicians. Although well-developed lists of short-term hospitals, office-based physi- cians, and nursing homes do currently exist, a major problem with the pro- posed extension of health care providers is the lack of adequate lists from which these "other providers" can be sampled and the high cost of developing and maintaining up-to-date inventories for each category of health facility or provider in a set of first-stage sampling units. This section outlines the pan- el's proposal for a sample design that will meet in a cost-effective manner the need established by NCHS to expand its data gathering to be able to include a fuller range of providers and health care settings than currently planned. Rather than develop separate sampling frames for each and every type of "other provider" (i.e., other than short-stay hospitals, office-based physi- cians, and nursing homes), the panel proposes that appropriate national samples of "other provider" categories be identified by screening National Health Interview Survey respondents for the names and addresses of the health care providers and settings they visited in a very recent period, such as a defined two-week reference period. The essence of the proposed design is the identification by NHIS re- spondents of providers to be sampled for the expanded National Health Care Data System. The design is flexible, in that it can be applied for any type of provider, and it can be readily applied to collect data for new health care settings that emerge over the course of time. Each sampled provider is subsequently contacted and asked to provide data for the visit that led to the provider's selection and for a sample of other visits. Apart from the manner in which the providers are sampled, the survey procedures can be similar to those used in the current provider surveys. The basic sample design and data-gathering plan for the NHIS does not change, except that the questionnaire would have to be expanded to collect information to identify and locate providers that the respondents visited in the specified reference period.
68 TOWARD A NATIONAL HEALTH CARE SURVEY The primary advantage of changing the origin of a sample of "other providers" from provider lists to reports of health provider care received by NHIS respondents is the significant savings in both cost and time to acquire sampling frames for each type of "other provider." The time factor is of considerable concern, particularly in a changing environment, with newly emerging modes and numbers of health care providers. The NHIS-based approach offers rapid identification of emerging health care providers, which is just not possible with the current approach of developing and maintaining inventories for given types of providers to serve as sampling frames. Another important and attractive feature of this sample design is that the health providers in the NHIS-based sample are selected with probabili- ties directly proportional to their use or size (i.e., number of visits for health care they receive in a given period), provided that (a) the NHIS interviews are spread evenly throughout the time period and (b) the NHIS respondents are sampled with equal probability. In practice, the latter condition is not satisfied because, for instance, of the oversampling of minority populations. However, a subsample of NHIS respondents can be drawn in a manner that creates an equal probability sample of NHIS respondents for use in sam- pling providers. The selection of providers with probabilities proportional to size (PPS) enhances the potential for increased reliability of the estimates derived from the sample of visits subsequently selected to gather utilization data from each type of provider surveyed. The selection of a constant number of visits from each selected provider yields an equal probability sample of visits and balances the data-gathering workload. Thus, for example, 20 visits might be sampled from each selected provider. The choice of the number of visits to select from each provider depends on cost and variance considerations. This number may well vary among different types of pro- viders. Research is needed to determine the most appropriate number of visits to select from each provider. The panel recognizes that there are a number of potential difficulties in the implementation of this design. It relies on complete reporting of visits by NHIS respondents; significant underreporting of certain types of visits (such as those for less socially acceptable diseases) could create serious problems. Also, the NHIS respondents may not be able to provide suffi- ciently accurate inflation to enable the sampled providers to be located. Some of the named providers might be located outside the NHIS PSUs, thereby increasing the cost of subsequent data collection. The sampled providers may be reluctant to provide the requested visit data. How should providers sampled in this way be approached to maximize their coopera- tion? Methodological studies will clearly need to be conducted on such issues before this design is implemented. Since, for some types of provider visits, recalling visits could be prob-
DESIGN FOR A NATIONAL HEALTH CARE DATA SYSTEM 69 lematic, the decision to limit recall to 1-2 weeks may be necessary. The only recent comparable experience on obtaining permission to contact pro- viders from household samples comes from the National Medical Expendi- ture Survey (NMES), wherein signed respondent permission and successful contact were achieved for 81 percent of identified providers. Based on preliminary estimates, approximately 40 percent of the nonresponding pro- viders in this process were lost because of an inability of the interviewer to obtain sufficient identifying information for the provider (Harper et al., 1991~. The remainder were lost because of the interviewer's inability to obtain signed permission or because the provider refused to comply with the request for records data. There is also the issue of people making multiple visits to providers for different conditions and the problem of matching respondent reports to provider records. Further research would be needed to investigate the level of recall error and to minimize this attrition. Experi- ence would suggest that keeping attrition at an acceptable level is at least possible, given the relatively high household survey response rates achieved by the NHIS. Identifying a provider sample through a household sample chosen with- in well-defined primary sampling units presents the logistical problem that some providers will be located outside their boundaries, some perhaps some distance away. Since excluding providers outside the PSUs would destroy the validity of the provider sample, there is no recourse but to attempt to enroll them all. This then raises the issue of what percentage of sample providers will be located outside the boundaries and how best one might handle dealing with them. Recent experience from NMES suggests that this would be a relatively minor problem for the hospitalization and the ambula- tory care components. Research would be needed, however, to find creative yet practical ways to administer these surveys for providers outside the PSU, for whom field staff might have to commute some distance. Recommendation 4-1: The panel recommends that providers other than those currently covered i.e., short-term hospitals, office-based physicians, and nursing homes- be surveyed using provider samples generated from the list of providers visited by respondents to the National Health Interview Survey as identi- fied through the survey screening. The proposed NHIS-based sample design for providers is particularly useful for providers for whom no list frames exist, and especially for newly emerging health care settings. In principle, the design could be expanded to include visits reported by NHIS respondents to short-stay hospitals and office-based physicians. Currently, since provider utilization in terms of number of visits is not known in general, the NHDS uses proxy measures of size (e.g., the number of beds) to select samples. The NAMCS is not a PPS
70 TOWARD A NATIONAL HEALTH CARE SURVEY sample, whereas the NHIS-based sample is. Thus the NHIS respondent- generated samples of short-stay hospitals and office-based physicians con- ceivably could be more efficient statistically. By contrast, hospitalizations (or other visits) that lead to deaths or institutionalizations will not be repre- sented in the NHIS-based sample, with resulting biased estimation. Clearly, many factors can enter into a decision whether or not to change the origin of the samples of short-stay hospitals and office-based physicians from inventories currently maintained by NCHS to identification of such providers by NHIS respondents. As there is insufficient information on the relative feasibility, costs, and efficiency of these alternative approaches, NCHS should research the issues prior to any final decision. Recommendation 4-2: The panel recommends that NCHS ex- am~ne the feasibility and utility of selecting its samples of short- term hospitals and office-based physicians from inventories of each of these types of providers visited by respondents to the National Health Interview Survey and identified through the survey screening. NHIS-Based Surveying of Cohorts of Individuals As discussed in Chapter 2, comprehensive data on the use of health care services are urgently needed on a continuous basis if the critical issues of cost, access, and quality of care are to be meaningfully addressed. This implies continuously gathering longitudinal, person-level data reflecting patterns of care sought and received, including what services were provided, by whom, and with what outcome, as well as the associated charges by type of service and how the charges were paid. The panel believes that the National Health Care Data System can and should meet these data needs. Specifically, the panel proposes a prospec- tive design in which the NHIS will be used to identify cohorts of individu- als to be interviewed periodically over specified time periods to provide longitudinal data on their health care utilization and expenditures.2 The national health care statistics on key policy-related items generated by these cohort surveys should be published annually. The cohorts based on subsamples of NHIS respondents should be se- lected so as to permit provision of national statistics that would best address health care policy issues. In addition, oversampling should be used to sample policy-relevant cohorts such as low-income people, uninsured peo- ple, minorities, people with poor perceived health status, disabled people, 2 The analysis of longitudinal data is an active area of research with many new techniques. For an introduction of this literature see Kasprzyk et al. (1989); Cox and Cohen (1985); Duncan and Kalton (1987); Heckman and Singer (1985); Kalton et al. (1989); Kasprzyk and Jacobs (1991); Office of Management and Budget (1986); Pearson (1989); Ruggles (1991).
DESIGN FOR A NATIONAL HEALTH CARE DATA SYSTEM 71 and elderly people. Since unbiased national estimates for the general popu- lation should be generated in any event, sample cohorts representing those individuals not eligible for any of the special interest cohorts should also be selected from among the NHIS respondents and interviewed periodically as well. To ensure the completeness and accuracy of the information reported in the follow-up interviews by sample persons on the health care services they received, the cost of those services, and the sources of payment, the provid- ers of those services and the health insurers will need to be contacted to obtain the details of each provider visit experienced by at least a subsample of the persons in the cohort. Signed permission to make these contacts and to release the relevant data will need to be obtained from each person sampled. The relative feasibility of collecting such signed permission forms by mail versus face-to-face, for those interviewed by telephone, will need to be explored by NCHS. Numerous design issues will need to be addressed, for example: (1) timing in the selection of the samples of NHIS respondents relative to the initiation of data collection, (2) frequency of the periodic follow-up inter- views, (3) mode of the initial follow-up and repeat interviews, and (4) respondent motivation and sample attrition. One problem that arises in selecting the samples for follow-up from NHIS respondents is the potential for a sizable time interval between the NHIS interviews and the initiation of the follow-up interviews and the at- tendant loss of sampled individuals who have moved in the interval. To minimize recall errors, the follow-up interviews should be conducted fre- quently, say once a quarter, or even more frequently if suggested by the findings of the 1987 NMES, conducted by the Agency for Health Care Policy and Research. The cost of carrying out the follow-up interviews could be mitigated by the use of computer-assisted telephone interviews whenever possible. Retention of sampled individuals for the full set of follow-up interviews is always a challenge; appropriate incentives to mini- mize attrition should be devised. The selection of a sample of individuals from NHIS respondents to collect data on health care use and expenditures is not a new concept. An NCHS-funded study (Cox et al., 1987) evaluated alternative designs for linking the 1987 NMES with the NHIS. Included in the alternatives exam- ined were optimally allocated designs that utilized data collected in the NHIS with respect to race, poverty status, age, and health status to construct strata. The study found, for example, that the NHIS-linked, optimally allo- cated, not-self-weighting designs for these policy-relevant strata, at least, could achieve significant cost savings for estimates of average annual utili- zation and expenditures by type of provider relative to the cost of achieving the precision of an unlinked design. The optimally allocated design essen- tially uses characteristics of NHIS respondents to oversample potentially
72 TOWARD A NATIONAL HEALTH CARE SURVEY heavy users of health care services, thereby yielding greater efficiency. This design, however, was not in the final plan for the NMES that was developed. The panel recognizes that surveys for gathering information on medical expenditures are expensive. Yet it believes that it is important in today's environment to be informed on a current basis about health care needs, patterns of care received, and the associated costs. Linking the cohort samples to the NHIS and oversampling these samples in policy-relevant domains, together with greater use of telephone interviewing, might pro- duce the needed data at somewhat less cost than has been experienced to date in previous population-based medical expenditure surveys. It should be noted that the proposed design for a continuing longitudi- nal survey of health care utilization and expenditures is in many ways simi- lar to that for the NMES. When fully operational this survey would gather basic data similar to the data gathered in NMES, but at a somewhat lower cost. The proposed design will use a cohort of individuals selected from the NHIS respondents, and the screening to identify providers will be done within the context of NHIS, an ongoing survey with an existing cadre of interviewers. Basic data on health care costs and expenditures (both out-of- pocket and total expenditures) will therefore be available on a more current and resource-efficient basis than is possible now from ad hoc comprehen- sive surveys conducted at infrequent intervals. Recommendation 4-3: The panel recommends that NCHS de- velop and implement, as a component of the National Health Care Data System, a continuous, longitudinal survey of health care utilization and expenditures and their health care provid- ers, using cohorts of individuals selected from among National Health Interview Survey respondents. The panel believes that it is essential that NCHS work in coordination with the Agency for Health Care Policy and Research in implementing this recommendation, which it sees as a natural next step in the development of routine collection of health care cost data. NNHS-Based Surveying of Cohorts of Residents Currently little or no data are available for residents of nursing homes on their use and expenditures for health care services from other providers (i.e., other than the nursing home) while institutionalized. This deficiency can be addressed by collecting health care utilization and expenditures data for a sample of nursing home residents on a continuous basis in a manner comparable tO that outlined in the proposed NHIS-based cohort survey. Thus a sample of current residents in each NWHS sample facility could be selected periodically, to serve as a cohort of institutionalized persons. Longitudinal
DESIGN FOR A NATIONAL HEALTH CARE DATA SYSTEM 73 data, analogous to that collected from the NHIS-based cohorts, on the use of other health providers by members of the cohort could then be collected periodically, over the course of a year, from nursing home records and from the health care providers who treated them. The field work for such a design is facilitated by the center's plan to conduct the next NNHS in the subsample of the NHIS PSUs. The panel urges NCHS to explore the use of this method to collect health care use and expenditure data for people in other types of long-term health care institutions as well, such as mental hospitals. Recommendation 4-4: The panel recommends that NCHS de- velop and implement a survey capability to obtain longitudinal data for cohorts of residents of nursing homes, while institu- tionalized, on their use of and expenditures for health care re- ceived from providers other than the nursing home itself. NCHS should explore the possibility of obtaining this information for residents of other long-term institutions. There are other subgroups of the population, such as the homeless, migrant workers, people in jails and prisons, who are not well captured by the current NHIS sampling design, even when it is supplemented by nursing homes and possibly federal and long-ten hospitals. The implications of excluding these groups in surveys monitoring the nations's health status and the use and costs of health services are obvious, considering that these groups are at high risk of poor health. The panel regards the appropriate representation of these subgroups, while difficult, to be very important and urges NCHS to initiate research to find ways to adequately capture these groups in their surveys. NHIS-NN~S-Based Surveying of Episodes of Illness An important feature of the proposed National Health Care Data Sys- tem is the capability of collecting longitudinal data on the health status and health care utilization of a population-based sample of individuals and link- ing them with the associated information on the health care providers. The design strategy recommended in the previous sections represents a major step forward by linking information on the individuals with visits to health care providers, thus enabling important policy-related data on the condition of the health care system to be collected and reported continuously. How- ever, the design does not necessarily capture the complete dynamics of the illness and treatment process. In order to more fully portray the health care system, data are needed that reflect the illness and treatment process from onset of the illness, diagnosis, treatment, and outcome. Although there are a number of potential ways in which such longitudi- nal data on individuals can be organized (for instance, by period of observa-
74 TOWARD A NATIONAL HEALTH CARE SURVEY lion), an attractive organizational concept from many standpoints is the "episode of illness" and the patterns of care associated with such episodes. The episode model focuses on the observed patterns of care as a reflection of the care sought and received in conjunction with an episode of illness through possible treatment and outcome. Thus an alternative approach to gathering population-based statistics on health care services, wherein one follows a cohort of individuals, could be to limit the longitudinal data col- lection to those persons reporting art illness episode. This model could be used for studying patterns of utilization, cost of care, quality and access to care, and the role of the different types of providers and technologies. Since only those persons experiencing an episode of illness would be followed, this approach also has the potential to gather the requisite health care data more efficiently than from a cohort representing all persons, as presented earlier, and could conceivably replace it in the long run. The panel recognizes that despite the attractiveness of the concept and its many advantages, its application has been relatively limited because of practical difficulties. Steinwachs (1991) identifies several problem areas that have limited the use of the episode of illness concept in examining health services utilization. These include the lack of standard definitions and classifications to be used in constructing the episode of care resulting from an episode of illness, as well as the lack of standard summary mea- sures of episode characteristics, such as duration of an episode, intensity of care, level of use, cost, and outcomes.3 The application of the episode concept to chronic conditions is not clear: Is there a single long episode, or is it more properly conceptualized as a series of episodes for maintenance, care, or treatment of acute exacerbations? These issues should be resolvable, both from methodological and re- source perspectives. In particular, classifications will need to be developed to resolve such issues as: When does an episode begin (e.g., does it require limitation in activity for at least one day)? When does an episode of illness end (e.g., is this the last day of any limitation)? When does the care for the illness end, how should comorbidities be identified and included in the episode, when is the episode part of a chronic disease process and when is an acute and time limited condition, and when should outcomes be mea- sur-d (e.g., single or multiple points in time)? Clearly, an investment in methodological work is needed by NCHS and the Agency for Health Care Policy and Research to resolve these and other issues before pursuing the theoretical attractiveness of the episode concept into application, particular- ly in large-scale surveys. 3Recent literature examining definitional and operational issues associated with episode of care analysis includes, among others, Keeler (1991, 1988a, 1988b), Hornbrook (1985), Star- field et al. (1991), Frank et al. ( 1991), and Prien et al. (1991).
DESIGN FOR A NATIONAL HEALTH CARE DATA SYSTEM 75 Assuming that the definitional and related issues are resolved, the panel proposes a design that would efficiently use survey resources to develop episode information. Within the context of a population survey, those indi- viduals reporting a recent onset of an episode of illness (within two weeks) would be the target population, including mental and emotional illnesses. This group can be expected to include many minor self-limited illnesses or injuries for which no care is sought, yet there may be some individuals who are at the early stage of a serious condition and have not received care or made an appointment for care. It would be important to follow all individ- uals seeking care, plus those who have not sought care, to the end of the episode of illness or to the end of the care process, which ever comes later. This would provide unique information on patient outcomes, their relation- ship to care seeking and to the pattern of care received from one or more providers. The interpretation of the interview and follow-up data can be substantially enhanced by supplementing these data with information from providers and patient records for those episodes receiving care. The population can be viewed as consisting of people with different patterns of illness. For example, experience would range from no episodes during the specified reference period through minor episodes of relatively short duration, that require no provider care, to acute episodes that would require provider care and end in either death or recovery. Some individuals may experience more than one illness episode (possibly overlapping) during the period. The majority of persons with short-duration episodes would be part of the noninstitutionalized portion of the population. More episodes associated with chronic conditions (i.e., those extending over longer peri- ods of time) than short-term acute episodes are often experienced by per- sons institutionalized in long-term health care facilities. A sample of eligible individuals with selected episodes would be fol- lowed until the outcome of the episode occurs, or until some operationally defined stopping point is reached (e.g., six months after onset, some defined state of health or disability). Defining a stopping point for episodes such as injuries with a permanent disability is complicated, even if an arbitrary time period after onset is used. A full range of data (e.g., diagnosis, provider visits, costs and treatment strategies tied to individual provider visits, limi- tation of activities, and final outcome) would be gathered for each selected episode from health care providers and insurers. Identifying information about each reported provider, along with signed permission to abstract data from the medical and accounting records for the respondent at each identified provider, would be obtained at the conclusion of each initial and each follow-up interview. Permission forms for the providers visited at the time of the first interview may not be sufficient, since persons may go to other providers after that time. The length of the
76 TOWARD A NATIONAL HEALTH CARE SURVEY reference period must reflect the respondent's ability to recall visits and minor episodes. The episode of illness model illustrated here is linked to the NfIIS design and operation. Except for expansion to the NfIIS questionnaire to accommodate screening for illness episodes in addition to obtaining identi- fying information for visited providers from the list of providers, the exist- ing design and data gathering plan for the NHIS would not change. The process of screening is similar to that described for the other cam; --ants of the design framework described in the preceding sections and wary ~ pro- ceed in the manner described below. The NHIS respondents first would be screened to determine those who began an episode of illness during a short reference period immediately prior to the interview. Thus, a representative sample of the group of illness episodes occurring to individuals over a time period of interest would be defined, then repeat contacts would be made tO chart the course of the episode arid to determine when it was completed. Contacts would be made with the providers of care for the episode to collect data on the treatment process, outcome, and costs. A subsample of the NtIIS respondents would be screened for all pro- viders visited and illness episodes suffered during a recent reference period. Two types of illness episodes can be identified through this screening pro- cess: (1) those with one or more provider visits (other than to NNHS- eligible nursing homes) occurring during some reference period ending jUSt prior to the regular NHIS interview and (2) those that ended during the same reference period but for which no provider visits occurred during the course of the episode (e.g., colds, flu, minor injury, and illness for which one cannot afford to seek care). Only those persons experiencing a "new" illness episode that led to utilization of medical care would be eligible for follow-up. In other words, the subset of illness episodes, with a provider visit occurring during the reference period, becomes the sampling frame of illness episodes screened for further follow-up beyond the NHIS screening interview. By restricting the sample to episodes that begin in the two-week period immediately prior to interview, the sampling probabilities for the group of episodes are easily determined and the complication of selecting episodes with probabilities proportional to their length is avoided. A major limitation of this model as illustrated is that the emphasis on new episodes does not adequately address the need for information on chronic illness care, both routine maintenance and acute problems. An option for capturing information on the treatment of chronic conditions is to determine if during the past two weeks the individual experienced limitations in activ- ity related to a chronic condition. If yes, information could be sought regarding whether the limitation was more severe, the same, or less severe than previously. If the limitation is worse, this would define a "new"
DESIGN FOR A NATIONAL HEALTH CARE DATA SYSTEM 77 episode in the ongoing care of the chronic condition. These would be followed using a methodology parallel to episodes for conditions that are not identified as chronic. It is also likely that a significant number of people experiencing a chronic condition will never experience a well-defined "chronic episode," with a well-defined beginning and end, yet will interact with health care providers periodically. If that is the case, use of the episode of illness approach would be limited to episodes of acute illness. Under such circum- stances, a sample of NHIS respondents with chronic conditions would need to be identified and interviewed periodically to determine their utilization and expenditures for health care during a time period of interest, such as a calendar year. Since residents of nursing homes (or other long-term care institutions) are not included in the NHIS, selection of a subsample of current residents of the NNHS sample facilities should be considered for tracking episodes of illness that require health care services to be sought outside the nursing home. To facilitate episode screening, the NNHS sample would be subsam- pled, similar to the NHIS household sample. To properly reflect seasonal variations in illnesses, the NNHS subsample should be evenly spread over time. The episode screening of sampled nursing home residents would be carried out in similar fashion to the screening of the NHIS respondent sub- sample. The purpose of the screening is to identify any episode that led to medical care during the reference period, not including the episode that caused the sampled individual to become a patient in the nursing home. The information to be collected for residents suffering an episode of illness should be essentially the same as that collected about episodes of illness occurring to the NHIS respondent subsample. Recommendation 4-5: The panel recommends that NCHS un- dertake research in collaboration with the Agency for Health Care Policy and Research to examine the methodological issues of definitions and classifications and to determine the feasibility of using the National Health Interview Survey and the National Nursing Home Survey to generate a sample of episodes of ill- ness; the sample should be followed longitudinally to collect data on the associated medical care use for the episode from both the sample of individuals and the health care providers. Effectiveness and Efficiency In the face of rapidly rising health care costs, there is growing interest in treatment effectiveness in relation to cost. The episode of illness model
78 TOWARD A NATIONAL HEALTH CARE SURVEY offers an opportunity to collect not only data on each component of health care received and their costs, but also data on the outcomes of that care. Broadly based national statistics bearing on the overall effectiveness and efficiency of the U.S. health care system would thus become possible. The validity and cost of gathering health care data from health care consumers suggest that every effort should be made to take advantage of administrative records, such as health insurance claims files from both pri- vate and public insurers, patient medical records, and patient billing records. Medicare claims files provide a unique opportunity to study health care services received and costs for Medicare recipients. A similar set of claims files is needed for a national sample of persons younger than age 65. Recommendation 4-6: The panel recommends that NCHS con- duct research and develop procedures for data systems that enable linkage of health care outcomes to health care received and health care costs. The panel further urges NCHS to exam- ine the feasibility of collecting health insurance claims files from both private and public insurers for individuals included in the samples selected from the National Health Interview Survey and the National Nursing Home Survey to study health care utiliza- tion and costs. Relation to the NHIS Redesign As stated early in the chapter, a central feature of the panel's recom- mendations is to link the NHIS design to the operations of the National Health Care Data System. Work is under way on the 1995 redesign of the NHIS based on the 1990 decennial census, which will affect NHIS data collection for at least 10 years. The decisions made on the NHIS redesign will therefore have implications for the successful and timely implementa- tion of the panel's recommendations for an integrated design framework for collecting health care data. Recommendation 4-7: The panel recommends that NCHS take into serious consideration the recommendations in this report relating to the National Health Care Data System before reach- ing final decisions on the 1995 redesign of the National Health Interview Survey. POTENTIAL BENEFITS OF THE PROPOSED DESIGN FRAMEWORK Adopting the panel's recommended design framework for a National Health Care Data System has several potential advantages, some of which
DESIGN FOR A NATIONAL HEALTH CARE DATA SYSTEM TABLE 2 Summary of Data Produced Under the Design Framework Linked to Data Collected under Level NHIS Through Design Frameworka Person NHIS respondent Disability days Physician visits Acutelchronic conditions Limitations of activities Hospital stays Insurance coverage Special topics Access to care Preventive care 79 Episodeb First provider Cost of care visit during Episode outcome an episode Pattern of pro rider care l~cidencelprevalence Duration Content of provider care Seventy of illness Access to care Visit Any provider Diagnosis visit during Length of stay episode Disposition Treatment Cost of care Provider Any provider visit Facility profile whether or not for Health promotion practices an episode Treatment practices aCulTently unmet needs in italics. bIncludes only illness episodes leading to the utilization of health care. Minor episodes not requinug provider visits are also reported by respondents to the National Health Interview Survey. are tied to meeting unmet information needs (Table 2~. The data system when fully implemented would produce substantial new data at the person and visit level. Some of the new items are simply add-one to questionnaires for the existing provider surveys; others would be obtained from new sur- vey activities. The proposed data system in the long run would also provide a unique population-based database to facilitate assessment of the cost- effectiveness of treatment programs for specific health conditions. The panel's proposed data system would be able to meet the need for direct data linkage between the person and provider data. Information gath- ered for any treatment event, including data on providers visited or individ- ual provider visits, can be linked to the person who reported the provider and
80 TOWARD A NATIONAL HEALTH CARE SURVEY the visit. Thus, for example, data on the condition for which the treatment was provided and its outcome could be related TO the socioeconomic ch~ac- teristics of the individual. Linkage between person-level and provider-level data also provides a basis for confimning the illness diagnosis. The proposed data system would give NCHS sufficient flexibility to be able to adapt to changing data needs. One important facet of this flexibility is the ability to shift the focus of sampling as needed. For example, certain segments of the population (e.g., blacks, Hispanics, the uninsured, the eld- erly) can be oversampled by altering the NHIS subsampling process. Sam- ple sizes for specific conditions also could be controlled by altering the size of the NHIS subsample that is screened or by oversampling provider visits for rare conditions in the provider surveys.4 The flexibility tied to sampling needs is limited only by the difficulty that NCHS would have in making the design alterations and by the time it would take to implement them. Current NCHS scheduling plans to field the existing provider surveys annually would not be altered if the panel's recommended design frame- work is adopted. Finally, the timing is right to make the kinds of changes required by this design, since in recent years design integration involving the NHIS has been given a priority within NCHS and since significant redesign work is currently under way for the two surveys in which the greatest impact of change would be felt (i.e., the NHIS and the NNHS). IMPLEMENTATION STRATEGY The panel recognizes that, despite its many potential benefits, adoption of the recommended design framework raises several important issues that would require careful examination and resolution prior to making final de- cisions on the details of the design. Many of these issues are related to some of the screening and patient follow-up procedures. Perhaps the most important are: (1) to identify pro- vider visits, (2) to obtain useful information to contact identified providers, (3) to obtain permission from the patients to peruse the provider records of those whom the NHIS respondents identify, and then (4) to successfully enroll the providers and the patients in the study. Another important issue regarding the design of the proposed National Health Care Data System as recommended by the panel is the capacity to follow persons in receipt of care for a fixed length of time or until outcome, and in some instances beyond NHIS PSU boundaries. Follow-up might conceivably involve some combination of face-to-face and telephone inter- 40ther methods for sampling for rare conditions should also be investigated' including mul- tiplicity or network sampling. A description of this and related methods and an introduc- tion to related literature are included in Sudman and Kalton (1986); Kalton and Anderson (1986); Sudman et al. (1988); Nathan (1976); Sirken (1972).
DESIGN FOR A NATIONAL HEALTH CARE DATA SYSTEM 81 viewing. Both modes have already been considered for use in the NHIS and the NMES. The telephone, for example, might be used only as a backup to the face-to-face follow-up if the person moves away from the area. Addi- tional research into finding the most effective strategies for episode follow- up are necessary. As discussed earlier, although the concept of episodes of illness is at- tractive and such data would prove to be a very valuable analytical data- base, especially for medical effectiveness research, experience with its appli- cation is limited and problematic. Developing the capability to generate data on episodes of various types of illness raises important and complex method- ological issues associated with definition, classification, and measurement. Finally the potential for difficulties with the process of linking large person-level data files from the persons' visited providers, their health care utilization, the process of care during an illness episode, and the final out- come should not be underestimated. Recommendation 4-8: The panel recommends that NCHS es- tablish a research agenda to determine the feasibility of its rec- ommended course of action. If found feasible, the panel recom- mends that NCHS adopt the proposed design framework (with adjustments as warranted by the research) for a National Health Care Data System. In conclusion, the panel recognizes two different, but not mutually ex- clusive, courses of action for establishing an integrated National Health Care Data System. The first course of action, put forth by NCHS, calls for limited geographic integration of the provider surveys with NlIIS, extend- ing the coverage of the provider surveys, and adding patient follow-up to the NHDS and the NAMCS. In effect, this plan envisions a collection of slightly modified provider surveys with follow-up components as the Na- tional Health Care Survey. The panel believes this course falls short of meeting anticipated health inflation needs for the next century. It therefore proposes a different and more ambitious course of action, one that encompasses most of the features of the NCHS proposal, but also calls for a significant long-term expansion in the breadth and depth of information to be gathered through a compre- hensive National Health Care Data System. The section outlines below steps in a multiyear implementation strategy for the panel's proposed plan starting in 1992. The strategy is divided into four, at times overlapping, phases involving the initiation of the research agenda, developing and implementing the panel's design framework and recommendations for improvements in content and coverage of the informa- tion gathered, followed later by research into the feasibility of surveying episodes of illness and the associated patterns of care and outcomes.
82 Phase I: 1992-1995 TOWARD A NATIONAL HEALTH CARE SURVEY Suggested Implementation Schedule (1) NCHS begins work on the development and implementation of the panel's proposed design framework and He recommended improvements in the data content for the data system. Many aspects of the pro- posal have minimal impact on the designs of the underlying exist- ing survey vehicles. (2) NClIS initiates a research program to carefully consider the issues identified in the preceding sections of the report. NCHS should work in collaboration with the Agency for Health Care Policy and Research in researching the issues relating to the feasibility and efficacy of gathering data on episodes of illness. (3) NCHS begins the process of convincing the Department of Health and Human Services, the Office of Management and Budget, and the Congress of the importance of this effort for federal health care policy and to ensure that this effort receives adequate funding and staff resources. Phase II: 1995 NCHS begins implementation of the panel's recommendations with any modifications and adjustments as warranted by the research results to date. Research agenda continues. Phase III: 1994-1997 NCHS sets upon a one-year period to flesh out the design protocols for the panel's plan if found feasible in Phase I and with modifications as warranted by the results of the research. NCHS begins developmental work for implementation of the panel's long-range proposal for surveying illness episodes. Phase IV: 1997-2000 NCHS commences implementation of all components of the panel's proposed design framework for a comprehensive National Health Care Data System.
