Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 143
Accounting for Health and Health Care: Approaches to Measuring the Sources and Costs of their Improvement 6 Linking Population Health to the Array of Health Inputs Throughout this report, we have described conceptual issues that must be resolved in order to move forward on the production of an experimental national health account—or at least a well-organized national health data system—and to improve the medical care components of the standard economic accounts. A fully developed health data system would coordinate (1) detailed and comprehensive data covering the nation’s expenditures on, and utilization of, medical care, organized in such a way that prices and quantities of meaningfully defined units of production and consumption can be measured; (2) data that allow the nation’s health to be tracked along a number of dimensions; (3) data for monitoring non-medical factors affecting the population’s health; and (4) research results that attribute changes in the population’s health to changes in spending (of money and time) on medical care and other health-affecting goods, services, and activities. Chapters 1 through 4 focused mainly on topic (1), which involves identifying, quantifying, and valuing the outputs of medical care—organized around the treatments of specific medical conditions—that are inputs to health. In addition, while medical care is focused mainly on treatment, some spending is targeted toward disease prevention, and many other factors determine the incidence of disease. In Chapters 2 and 3, we describe how satellite health accounts might be structured and identify the first steps in their construction—defining the units of measurement for medical care and estimating economy-wide expenditures on those units. These are tasks that the Bureau of Economic Analysis (BEA) is working on now and presumably will be for some time. Chapter 5 focused on (2) and touched on (3), by first defining measures of current population health in a way that reflects both mortality and quality of life. They can be conceptualized jointly as quality-adjusted life expectancy and
OCR for page 144
Accounting for Health and Health Care: Approaches to Measuring the Sources and Costs of their Improvement expressed either in years or dollars (Murphy and Topel, 2006). In addition, we discussed briefly how the risk factors that predict future health might bechosen and collected. In the health account, we proposed expressing health outcomes in both natural and monetary units. Monetary valuation of health is difficult: even when the provision of medical care involves prices (although often ones that do not closely reflect cost), some inputs, such as volunteer labor for the chronically ill, do not. Also, nonmedical nonmarket inputs include time invested in one’s own health (for example, exercise and sleep) or in a relative’s health, and such activities often have additional goals besides future health, which complicates evaluation. A comprehensive—and at this point admittedly futuristic—health account would also attempt to incorporate topic (4). It would not only identify, quantify, and value the flow of nonmedical health inputs, such as behavior trends (e.g., diet, risk taking, smoking, consumption of alcohol), research and development, and the quality of the environment; it would relate both these and medical inputs to current and future population health. While emphasizing the value in monitoring both inputs and outcomes, we have been largely agnostic on exactly how researchers go about the task of quantifying causal links between medical care, health-enhancing activities, and other inputs to the population’s health through disease modeling. This is a difficult area of inquiry, both conceptually and in terms of data requirements, that is being pursued in leading-edge research taking place across many institutions, primarily on a disease-by-disease basis; it is the type of work that BEA will probably never do. That said, results from this research could eventually be used to enhance the usefulness of a national health account that the statistical agencies play a role in constructing. While it is beyond the scope of this study to offer detailed recommendations on this academic research, in this chapter we review some of the ongoing work, offer some general guidance for U.S. efforts going forward, and describe how a national health account would provide a useful centralized data depository from which investigators could draw and into which results may feed. Much of this chapter is concerned with developing a data system that would allow changes in the population’s health (death and impairment) to be linked to changes in spending on medical care and other factors. While the panel is skeptical about how well, at least in the short run, outcomes can be linked to medical expenditures and other factors, we strongly recommend beginning the process of gathering data in a way that improves the ability of researchers and policy makers to draw causal inferences. Creating and pooling electronic health records (discussed more below) would seem to be a prerequisite on which to focus this line of development. 6.1. ATTRIBUTION OF HEALTH EFFECTS TO INPUTS As discussed in Chapter 2, a major policy issue motivating research on the topic of this report is how to gauge the productivity of the medical care system,
OCR for page 145
Accounting for Health and Health Care: Approaches to Measuring the Sources and Costs of their Improvement which provides one rationale for beginning there. For this purpose, it is important that investigators have data that allow them to attribute population health effects to factors that work separately or interactively with medical care. It is advisable to begin by trying to do the medical part well, but elements included in a broader boundary of health goods, services, and activities become more important when trying to determine causality for outcomes. There are many ways of linking medical care and other inputs to health. One is through a standard medical trial of certain inputs or interventions (e.g., diet counseling), looking at the effect of treatment on a primary outcome in which the investigator attempts to keep other factors determining the outcome constant by standardizing patients and treatment and using randomization. Another approach is the one discussed in Chapter 3, whereby econometric methods are applied on national data over time on multiple inputs and health outcomes, together with clinical insights, disease modeling, and common sense to figure out what is causing what. Intermediate approaches include epidemiological studies using specialized panel data sets such as Framingham or Surveillance, Epidemiology, and End Results (a registry of the National Cancer Institute). These specialized data sets become more useful when information from patient claims have been linked to them. Despite the problems with trials (e.g., cost, delay, external validity), we certainly support the paradigm but want to open up the possibility of using other approaches for developing data linking medical treatment and other inputs to health outcomes. Improved data on expenditures, prevalence, and death that are classified by disease will be useful for other research projects as well. They may be used for constructing comparisons over time or across countries, regions, or subpopulations in terms of burden. Whatever methods are pursued, a system for attribution entails more than just collecting data on the multitude of factors that affect population health. Such a data system is merely a tool to help researchers working in this field. While data on high blood pressure and other personal risk factors should be collected and presented, attribution is very difficult for a number of reasons—perhaps one can say that lower hypertension will lead to fewer deaths, but overall mortality is based on many things that occur in the past, present, and future. The problem is most severe when the objective is to attribute outcomes to services or causes; in many cases, the medical linkages are not known or well understood. For example, in the United States (and many other countries), functional limitations in the elderly will be much more important in the coming years, simply because of the demographic shift. Mobility impairments tend to be the result of multiple medical conditions, some of which are not ordinarily thought of as diseases. Arthritis (most commonly osteoarthritis, but sometimes cartilage problems, rheumatoid arthritis, or other conditions) is usually a component. Poor balance and impaired proprioception may also be important, and these may be the result of strokes or simply “aging,” meaning that there is some neurological problem but no known disease that caused it. Projecting the effects of, for example,
OCR for page 146
Accounting for Health and Health Care: Approaches to Measuring the Sources and Costs of their Improvement joint replacement may be difficult because the underlying arthritis is only one of the causes of the mobility impairment. In other words, there is not a one-to-one correspondence between the functional limitation and a disease. So the boundary and attributional issues are likely to become even greater challenges. Even when the change in a person’s health state can be attributed to one condition—say a chronic disease—defining an episode can be difficult. Furthermore, for an elderly population with multiple chronic diseases, the determination of a primary diagnosis for a hospitalization can be somewhat arbitrary. Someone may truly be admitted because of an exacerbation of heart failure, but he or she might have needed hospitalization only because of impairment from other conditions; otherwise, he or she might have been treated on an outpatient basis. Ideally, the complementary surveys collecting quality of life data (discussed in Chapter 5) could be better coordinated with common survey questions identified to get at nonmedical care inputs measured consistently over time. It would be valuable to assemble aggregate data on all the conceivable determinants of health in some researcher-accessible location. Depending on what kind of method will be used for attribution, very aggregate data may suffice (over time or across countries). The national health account program could begin accumulating data on time use (particularly in preventive activities), consumption trends, other risk factors, behavioral trends, the environment, etc. Even before these data are integrated into a health account, such a data clearinghouse would give researchers attempting to link cause and effect a starting place. Such a data system will be in a constant state of evolution. With better data or understanding of causes, data components could be expanded incrementally and coordinated, and, indeed, the determinants of a population’s health will change over time, in terms of both the set of relevant factors and the impact of each. Environmental factors are notoriously difficult to pin down, since, even if measures like ambient air and water pollutants at a given site over time are available, it is hard to determine an individual’s location at all times to derive exposure values. This is also a problem for occupational exposures. It will be a big step to develop more robust data sources on personal factors like blood pressure, total cholesterol, and the glomerular filtration rate; however, their determinants (e.g., diet, air pollution exposure) involve physical or laboratory measurements, and it would be a huge undertaking to add these to surveys. That said, the Health and Retirement Study (HRS) undertakes similar tasks that can serve as a model for learning more about experience with, and the cost of, such additions. 6.2. MEDICAL CARE EXPENDITURES AND HEALTH The idea of a data system that coordinates information about medical spending, health outcomes, and the population’s quality and length of life may be relatively new in the United States, but efforts internationally trace back further.
OCR for page 147
Accounting for Health and Health Care: Approaches to Measuring the Sources and Costs of their Improvement Statistical agencies in all developed countries—including the United States—produce at least some components of a national health account, since all calculate total medical spending, and many perform cost-of-illness analyses (Heijink et al., 2008). Several countries have gone further, measuring health trends in addition to expenditures with the objective of facilitating disease-by-disease comparisons. In this section, we do not add to the extensive discussion from Chapters 1-4 of the medical care input to health; we only identify a sampling of efforts that have been made by government agencies to begin establishing linkages between medical expenditures and health. Statistics Canada, the country’s statistical agency, has been experimenting with health measurement since the early 1990s. The agency has adopted the Health Utilities Index Mark 3 (HUI3) health assessment tool (see Chapter 5) and has incorporated it as a permanent component in its National Population Health Survey (Statistics Canada, 2007). Using these surveys, the government reports health-adjusted life expectancy at birth and at age 65, stratified by gender, province, and income group. The HUI3 has also been used by the agency to estimate trends in the health impact of various diseases. At the same time, the Canadian Institute for Health Information maintains the National Health Expenditure database, which tracks annual medical spending in the country (Canadian Institute for Health Information, 2006). Expenditure estimates are reported separately for over 40 disease categories, 5 payer sources, and for each province. These two agencies have joined together to publish national estimates of the economic burden of illness for the years 1987, 1993, and 1998 (Health Canada, 1998). For each of 20 disease categories, these studies report direct costs (hospital, physician, drugs, research, institutional, other) and indirect costs (premature mortality, long-term disability, and short-term disability), stratified by province, age, and gender. Over the decade, these studies have been enhanced by methodological refinements and by collection of more detailed data. The Australian Institute of Health and Welfare divides medical spending for the country into 176 disease categories in such a way that accounts for 94 percent of medical spending. Expenditure data are available by age, sex, and service category (though not for every year). The accounts have been linked to population health data from Australia’s Burden of Disease and Injury Study (Mathers, Vos, and Stevenson, 1999). Investigators from this study used methods similar to those of the Global Burden of Disease Study (Murray and Lopez, 2006) to estimate the health of the Australian population in 1996 and 2003. Efforts are now under way to estimate the returns to medical spending for each disease category. The motivation for much of this work is to improve the productivity of health care systems—that is, to: Produce outputs of health services with a minimum of real resources (technical efficiency) at each level of care, while also minimizing the (relative) costs of inputs (cost efficiency).
OCR for page 148
Accounting for Health and Health Care: Approaches to Measuring the Sources and Costs of their Improvement Provide a mix of care that maximizes impact on health outcomes at a minimum input cost (cost effectiveness). Set overall resources for health care consistent with achieving wider goals of social welfare and to allocate services across individuals at levels that make the best use of these resources (efficiency and equity in resource allocation). Technical or cost efficiency can be high if a given set of medical activities—or outputs—are produced with a limited amount of inputs (see Figure 6-1). However, if the impact on health status is limited, little social value is obtained from these outputs. This is why the medical community focuses more on cost-effectiveness as it provides a measure of the actual health returns to spending (Gold, 1996). There are a number of challenges to establishing health outcomes as the standard in the actual practice of system evaluation. First, for some interventions, it can take a long period of time before they have any significant effect on health. Second, there are not always demonstrated links between interventions and health. For example, available data may show an impact only on behaviors that affect health. Alternatively, the data may show no impact at all, and frequently there are simply no data on the effects of an intervention. Finally, when data do exist, the data collection and surveillance systems may not provide the level of detail necessary to measure desired changes in health. The goal, then, is to strike an appropriate balance between intermediate output measures and longer term health outcomes. The challenge is in defining those outputs and in gaining access to sufficient data to measure both prices and quantities. FIGURE 6-1 From inputs to outcomes. SOURCE: Adapted from Joumard and Häkkinen (2007, p. 12).
OCR for page 149
Accounting for Health and Health Care: Approaches to Measuring the Sources and Costs of their Improvement FIGURE 6-2 A schematic framework for population health planning. NOTES: HRQoL = health-related quality of life. QOL = quality of life. SOURCE: Kindig, Asada, and Booskie (2008). Reprinted with permission. 6.3. NONMEDICAL AND NONMARKET INPUTS TO HEALTH Chapter 2 lays out, in general terms, the structure of an accounting system that includes, on one side, data on the inputs to health and, on the other, data on the output, defined as population health. Beyond the Market (National Research Council, 2005) describes a similar structure. Figure 6-2 summarizes the basic elements of this relationship. Generating the data and specific accounting structure needed to quantify these relationships is a much more difficult task. Complicating the goal of establishing links between inputs to health and health itself (and even just figuring out which kinds of data to collect to inform the task) is that, as shown in the figure, health is a function of much more than just medical care.1 Population health, measured in terms of life expectancy or more subtle quality of life metrics (discussed in Chapter 5), is mediated largely by such factors as personal behaviors (e.g., sedentary lifestyle, smoking), environmental exposures, and public health measures that transpire outside the medical care setting. While health policy gives some attention to public health issues, it deals little with the social context of life, which can exert profound influences on health (Woolf, 2009). Dramatic disparities affect poor and minority populations, who endure poorer health and on average die younger than more affluent groups. 1 Research on this well-established relationship has its antecedents in research by McKeown (1976), Fogel (1986), National Research Council (1993), and may others, relating determinants of health and life expectancy.
OCR for page 150
Accounting for Health and Health Care: Approaches to Measuring the Sources and Costs of their Improvement While social determinants such as education, income, and race are clearly interrelated, they exert independent effects on health as well (Robert Wood Johnson Foundation, 2008). A full explanation of changes in population health requires information on multiple aspects of the social context. Furthermore, if health consequences arising from social policies originating outside the health care sector are to be monitored, data collection in less traditional settings—such as schools and community recreation centers—will eventually be needed. Solving the nation’s most pressing health care problems, then, will require a greater understanding of the full range of the factors that determine health and of their complex interrelationships. It is increasingly recognized that the most urgent public health challenges cannot be adequately addressed within a single discipline but instead require a more comprehensive approach. Kindig sets forth a schematic framework for population health planning—on which Figure 6-2 is based—that provides a preview of both data collection and the cross-disciplinary expertise needed for developing a health account (Kindig, Asada, and Booske, 2008). Perhaps the most important message conveyed by this framework is that population averages can be deceiving. Mean mortality or health-related quality of life measures mask real disparities in outcomes—disparities that can be identified only with adequate data (compiled at a sufficiently disaggregated level) on the social determinants of health. Kindig divides these determinants into five categories based on the Evans-Stoddart model (1990)—(1) medical care, (2) individual behavior, (3) social environment, (4) physical environment, and (5) genetics—to which one might add age, gender, and medical history; this top-level organizational structure may be a good place to start when specifying needs for a national health data system. While some population surveys in the United States (such as National Health and Nutrition Examination Survey [NHANES] or HRS) are a good source of nonmedical health data, ultimately data on determinants of health will need to come from multiple sources as there are many other variables—safe sex practices, occupational and geographic exposures, the physical environment, and others (many of which can be very difficult to measure)—that are not covered in any single survey. 6.3.1. Valuing Informal Care and Other Time Costs As emphasized above, a comprehensive health account requires tracking the full range of factors that affect health, even if they do not entail market transactions. Indeed, not even all medical service inputs are reimbursed. Many of the quantitatively significant influences on health—some relating to medical care and some not—can be linked to the way in which the population spends its time. A population that spends time actively, in physical exercise, for example, will be healthier than an otherwise similar sedentary one. Another example of a nonmarket cost is waiting time—time spent in physicians’ offices or hospitals waiting for care to be provided.
OCR for page 151
Accounting for Health and Health Care: Approaches to Measuring the Sources and Costs of their Improvement The most quantitatively significant of these services is care provided by relatives. The amount of quality-adjusted time that people spend caring for the ill should be positively related to the health outcomes of the ill person (although, frequently, it is negatively related to the health outcomes of the caregiver). Unlike formal home health care, unpaid caregiving is not included in the National Health Expenditure Accounts (NHEAs). As a result, NHEAs understate total resource use in the care of the ill (or the very old or very young, for that matter, which may also affect health). Furthermore, there is a bias in estimates of the growth in resource use over time, depending on whether informal care is rising or falling relative to market care. The issue is analogous to the treatment of home production in the National Income and Product Accounts. Market-purchased services (paying a laundromat, going to a restaurant) are counted as part of gross domestic product (GDP); home production (doing the laundry at home, cooking at home) is not. As a higher percentage of women have entered formal labor markets over recent decades, and more services previously provided at home are now purchased, estimates of GDP growth could in theory exceed actual increases in valued economic activity due to the displacement of home production. Almost all analysts of national income accounting—including those who produced the report Beyond the Market (National Research Council, 2005)—argue that nonmarket activities that are very close substitutes for market counterparts ought to be included in at least some version of GDP. Indeed, the same study states (Recommendation 6.3) that, ideally, estimates of the value of nonmarket medical care inputs, including time use, ought to be included in national health accounts. Time spent providing health-related services would be valued based on a replacement labor-cost approach; time spent in activities that improve or maintain one’s own health would be valued using an opportunity-cost approach (see National Research Council, 2005, pp. 127-130, for a full discussion of this rationale). These estimates may be small initially because of inadequate data on many aspects of time use, but this will challenge statistical agencies and researchers to call for improved data. Indeed, one reason for including these inputs in an account is that the value of time resources expended could be quite large relative to the value of market-provided services, particularly in such areas as elder and child care (LaPlante et al., 2002). Rules will have to be established dictating exactly what kind of time expenditures should be included. Work to quantify the amount and value of time inputs to health will be constrained for the foreseeable future by data availability, although the survey options are growing. For example, data on various components of informal care provision are available in currently conducted surveys. The HRS asks respondents to report activities of daily living—such as ambulating (walking), transferring (getting up from a chair), dressing, eating and drinking, performing personal hygiene, taking medication—or instrumental activities of daily living—such as driving, preparing meals, doing housework, shopping, managing finances and medications, and using the telephone.
OCR for page 152
Accounting for Health and Health Care: Approaches to Measuring the Sources and Costs of their Improvement Ideally, because of the disease-organized framework that we recommend, time-use data would be linkable to specific diagnoses. At this point, the American Time Use Survey (ATUS) does not offer this, although this is certainly a modification that is possible. For example, this was done in the RAND Health Insurance Experiment, in which doctors were asked to link the procedures, tests, and drugs prescribed to a diagnosis on the form.2 Additionally, the HRS could provide time-use information for older populations; however, very few data are available for children, so time spent in care of such conditions as autism would be underestimated. Other possibilities could, with further development, provide additional information: the Medical Expenditure Panel Survey (MEPS) indicates why children see doctors (e.g., their condition), and the ATUS indicates the amount of time parents spend in relevant activities. MEPS would be useful for tracking time spent in care of the chronically ill (see Box 6-1).3 6.3.2. More Boundary Issues for a Health Account When considering the nonmarket and nonmedical contributions to population health, boundary issues (and interaction with market accounts related to other areas of economic activity) become important. Due to its close proximity to market-provided medical care, a factor such as unpaid time spent caring for an ill person, discussed above, is likely to be in scope, as might be the costs and benefits of a (nonmedical) program designed to reduce the population’s intake of foods high in saturated fats. But where should the line of inclusion be drawn? For example, education improves health, but where does its production get classified in a fully integrated satellite accounting structure? And if the spending on education inputs does not get included in the account, should the health benefits (assuming they could be isolated) be included? Probably not, but there is no obviously correct answer here. Beyond the Market points out that, at least initially during early development, there will inevitably be overlap of nonmarket accounts that cover such areas as home production, education, health, and the environment. The values from these accounts would not add up to a meaningful total, but that may be necessary for the foreseeable future given the time it would take to 2 The claim form is reprinted in Newhouse and the Insurance Experiment Group (1993, p. 85); the relevant items are 20E and 21B. 3 Time allocation would be tricky, however. A time-use survey may report that a respondent spent 5 days in bed, but there will not necessarily be an International Classification of Diseases code tied to the record explaining why. At this point, the time use would be a residual that would have to be assigned in proportion to diseases, chronic categories by dollars spent, incidence, or something somewhat arbitrary. Similarly, time spent procuring medical service—for example, waiting time—cannot be linked to a claim; an analyst would have to do something like assume all visits have the same waiting time. The HRS linked to Medicare could be used for time spent in convalescence. Bed rest could be included, but this might be hard, compared with caregiving. Waiting time can be allocated using claims. Convalescence is harder. One could leave it unallocated or use regression methods, but it is not clear if there is enough data to do so. There will also be a bias issue—it is not random where one would be missing time costs. Even so, collecting the data is a first step.
OCR for page 153
Accounting for Health and Health Care: Approaches to Measuring the Sources and Costs of their Improvement BOX 6-1 Difficult Issues in Time Use Accounting A major research topic is how various time inputs into health should ultimately be counted and valued. For example, it is not clear that time spent by caregivers and time spent on one’s own health—for example, by patients waiting for service or actually in treatment or recovery—should be treated identically. Unlike unpaid time spent in caregiving activities (which has a market replacement value and should in general improve the health of the patient), time spent on some other activities cannot properly be called a “good.” For caregiving time, more is better from the point of view of the recipient (holding incidence of illness constant), so each hour spent in the activity is assigned a “wage” that adds value to the account totals (see National Research Council, 2005, Recommendation 6.4). It is intuitive, in aggregating inputs to health, to assign a positive value to a population’s time spent in health-improving activities, such as nursing an ill person or on the treadmill to improve one’s own cardiovascular health. If such nonmarket elements were left out of an economic account, either a larger share of the improved health would be incorrectly attributed to market inputs or there would be a bigger residual reflecting a health “profit” of sorts. In contrast, valuation is different for such inputs as time spent waiting in doctor’s offices or time required for recovery from a procedure, for which it would make little sense to simply multiply the quantity of these hours by the patient’s wage rate. While these time requirements are certainly inputs in the production function for health, adding them to the account as a positive would create the counterintuitive outcome whereby if a new method of treatment (or a new patient appointment system) were introduced that increased recovery time or waiting time (with no better outcomes), it would increase the total value captured in the account. One could think of these things as a downward quality change that, in principle, should be captured in the price index. At this point, we only raise these issues—they clearly need to be given more attention. In practice, none of this matters too much yet, since this dimension of an experimental account is so far off, but it is important to get the concept right early on. develop accounts in all areas for which they could productively exist. That said, in an experimental context, work on a health account should aspire as a long-term goal to include data on the value of all inputs—medical care and otherwise, market transacted and not—associated with the output (improved health). So how should a health accounting program begin the process of prioritizing which health-affecting factors to track from the practically endless list? One conceptual approach, identified earlier, is to begin with health inputs that are closest to the medical care system, specifically with treatment of diseases, and gradually move outward to the proximate determinants of disease, such as obesity, pollution, smoking and tobacco, illegal drugs, and possibly their determinants, such as eating and exercise.
OCR for page 176
Accounting for Health and Health Care: Approaches to Measuring the Sources and Costs of their Improvement use of health services; health behaviors; health outcomes; and other nonhealth and/or nonmedical determinants of health. An economic accounting structure is needed so that inputs to health (most immediately, medical care), and output (population health) are tabulated independently. Improvements in health—both quantity and quality of life—are the most critical variables on the output side of a health statistics system designed to quantify the impact of health care spending. 6.6.1. The Need for a Computerized National Data Infrastructure The United States is a long way off from being able to satisfactorily model population health and health determinants in a systematic way, but a starting point is to begin keeping better track of risk factors, health indicators, and other related data. It is encouraging that progress has been made along some fronts, but serious data gaps remain. The remainder of this chapter discusses options and makes recommendations on how to address some of the current data deficiencies in order to move toward more integrated and comprehensive “health statistics needed for the 21st century” (Rice, 2000). To develop a national data infrastructure, operational definitions must be developed, and core and optional data elements must be determined. Pieces of the data infrastructure will need to be created and pilot tested, in particular the linkages between databases. This will require a coordinated developmental effort, far beyond the current scale of activities. It will take several years, a research network, and a core commitment from leaders to make this happen. While building this infrastructure will be costly, failure to do so will be far more so—taking a substantial toll on the health (and wealth) of the population and the ability of policy makers to rein in wasteful spending. The World Health Organization outlines four priorities for data collection in support of producing national health accounts (World Health Organization, 2005): (1) to use all suitable existing data, (2) to adjust existing data to make them more suitable, (3) to improve or enrich surveys and administrative records to increase their suitability, and (4) to identify and arrange for the collection or generation of data that remain “missing.” To recognize the practical constraints to data acquisition, we add a fifth goal (which really should precede the first): (5) to develop a schema to prioritize the acquisition of data. Currently, for the United States, partial snapshots relating to the bigger attribution picture can be created. Spending on medical care can be tracked by provider and payer (NHEAs); global health and certain physiological measures are available (e.g., from such surveys as NHIS and NHANES, and HRS is promising); in addition, records on two universal health outcomes, birth and death, are collected. However, at this point, it is virtually impossible to link these things—spending on medical care to individual outcomes to population health effects—to expenditure data (in part because so many factors, in addition to medical care, affect health).
OCR for page 177
Accounting for Health and Health Care: Approaches to Measuring the Sources and Costs of their Improvement The development of a national data infrastructure designed to organize and coordinate current data and frame future data needs is critical if a national health account is to be developed. The matter appears to be a high priority for the current administration, which committed about $19 billion of the 2009 recovery package to encourage doctors and hospitals to install and use electronic health records.13 The fact that a number of private-sector technology and telecommunications companies have embarked on related projects adds to the prospect that the country may soon have an effective e-health records system. Not only will this modernization, designed to link patient health histories as well as treatment guidelines, be helpful for research and policy purposes, but also it could, if properly deployed, improve care and help curb costs by helping stem unnecessary tests, reducing errors, and coordinating treatments. Perhaps the most critical (and most difficult) piece in the development of a national health IT infrastructure will be creating a level of interoperability that approaches what exists for most other 21st century industries (see Box 6-2). While it is beyond the scope of this panel to make specific IT infrastructure recommendations, we point out that long-term comprehensive strategies and short-term incremental strategies need not be viewed as mutually exclusive. While current data may not be ideal, policy makers can learn a great deal simply by using—and developing better mechanisms to link—data sets that already exist. Thus, the policy choice is to determine how to most effectively allocate information systems resources to meet both short- and long-term goals. 6.6.2. Longitudinal Data While current surveys and administrative systems contain extensive crosssectional data, a shortage exists for large-scale longitudinal data to better track health status and health events over the life course of individuals in the population. The high expense and respondent burden associated with longitudinal surveys means that any increases in longitudinal data collection will be incremental. Other countries, having confronted this same issue, have addressed the lack of longitudinal data with dynamic microsimulation models, as described above (Wolfson, 1991, 1995; Spielauer, 2007). When microsimulation models are used, longitudinal data are still needed to set the framework, but in much smaller quantities. Recommendation 6.8: A study should be commissioned by a funding agency (National Institutes of Health or National Science Foundation) to take an inventory of other countries’ population health statistics systems, the role played by microsimulation modeling, the implications for longitudinal 13 Currently, only about 17 percent of physicians in the United States use computerized patient records, even though some very large service providers, such as Kaiser Permanente and the Mayo Clinic, do so. See http://www.nytimes.com/2009/09/10/technology/10records.html.
OCR for page 178
Accounting for Health and Health Care: Approaches to Measuring the Sources and Costs of their Improvement BOX 6-2 Interoperability and the U.S. Health Care Nonsystem The health care system in America is not a system. Rather, it is a disconnected collection of large and small medical businesses, health care professionals, treatment centers, hospitals, and all those who provide support for them. Most players have their own internal structure for gathering and sharing information, but nothing ties those isolated structures into an interoperable national system capable of making information easily shared and compared. Interoperable systems are invisible but essential, and people have come to depend on them in everyday life. Interoperable systems allow one person to speak to another using cell phones with different cellular service. ATM cards are good at virtually all banks nationwide and most banks internationally; they allow people to buy groceries and pay for gym memberships—all of this is possible because of secure interoperable systems. These systems work because the telephone and banking sectors have developed methods and standards that allow participants in their systems to easily access and exchange information while the companies operate independently and compete vigorously. If banks (which require standards of privacy and confidentiality) can use interoperable systems, health care should be able to as well. The benefits of putting such a system, which will be dispersed across many stakeholders, in place surely outweigh the costs (which include any change in the risk of confidentiality disclosure). The banking industry probably did several of these same cost/risk-benefit analyses 20 years ago, but the lasting impact of the network of point-of-service banking and portability of finance is now clear. NOTE: For more on this topic, see Walker et al. (2005). The authors estimate that interoperability and health information exchange could lead to $77 billion in savings. In addition, it would provide clinicians with full information at the point of care; it would enhance portability of information for patients; and it would give researchers a centralized data source. data collection, and the advantages and disadvantages of the different approaches. The study should be oriented toward providing guidance on how existing surveys, such as NHANES or MEPS, could be modified to optimize their analytic value. The ability to make NHANES longitudinal seems to exist now; it is just not done very often.14 Data are released biannually—and with a quite short lag. A longitudinal component would be extremely useful, but the cost would be 14 An example of an exception is the Epidemiologic Follow-Up Study, a national longitudinal study designed to investigate the relationships among clinical, nutritional, and behavioral factors assessed in the first NHANES.
OCR for page 179
Accounting for Health and Health Care: Approaches to Measuring the Sources and Costs of their Improvement high unless it was scaled back to include only a few targeted survey questions. It would also be useful to add NHANES measures into MEPS or to add claims information to NHANES. Recommendation 6.9: The National Health and Nutrition Examination Survey, which is already conducted on an ongoing basis, should contain a longitudinal component to optimize its value as an input into a health account and into cost-effectiveness research. Due to confidentiality concerns, the longitudinal component (or at least publicly accessible data products derived from it) will need to be limited to key variables that do not unduly increase disclosure risks. A trade-off exists in terms of the timing of longitudinal data collection. For example, it may be better to collect a sample that is twice as large every other year instead of annually. These options would have to be explored during the design phase. It would also be important to be able to link the data to Medicare records, which could perhaps be done in data centers. Some of this could be funded by reducing the scope of NHIS—which would just be used as a sampling frame—and then expanding the other surveys. 6.6.3. Data Sharing and Data Linkage In considering future prospects for improved health statistics to meet policy needs, it must be acknowledged that resources will not grow in parallel to the demand for data and medical services. Budgetary pressures require that current data collection and dissemination procedures be constantly assessed. These trends imply that it is time for the statistical agencies to make stronger efforts to coordinate data collection efforts across surveys and agencies—which may entail a cultural shift from within—to invest in the 21st century vision for health statistics. The ability to exploit multiple data sources will be a key ingredient to the success of a national health account. However, there are a number of barriers to linking provider data on expenditures with data on patients. For example, some modernizing of key data sources is needed to ease the task of creating a linkable system. If public and private data are to be linked, a common identifier will be needed. And trends in spending and in outcomes need to be measured at same metric. Because surveys are expensive and burdensome, cost efficiency requires linking across sources whenever possible. NHEA needs to be linkable to individual-level data contained in other health data sources (NHIS, HNANES, MCBS, MEPS, etc.).15 Ideally, one would like to be able to link data from these key 15 America’s Health Insurance Plans is now working to standardize electronic personal health records with the hope that claims and other patient records (diagnoses, procedures, medications) can be carried from one insurer to another (see http://www.insurancetech.com/showArticle.jhtml?articleID=201400212). In addition, they could be linked to the National Death Index in principle.
OCR for page 180
Accounting for Health and Health Care: Approaches to Measuring the Sources and Costs of their Improvement surveys to administrative sources (Medicare and Social Security, specifically). Expanding such efforts to the non-Medicare population is possible, but the large claims databases are managed for employers by data management firms, such as Ingenix and Medstat, so access may be an issue. Moreover, these databases are not representative of the entire private market because they typically do not include the individual and small group insured portions of the population, and of course the uninsured are not included. Data sharing and linkage could reduce duplicative and overlapping reporting systems and in turn maximize the results of separate efforts. Some promising progress has been made on this front. NCHS has developed a program that links several of the its population-based surveys (such as NHANES and NHIS) with death certificates from the National Death Index (NDI) Medicare enrollment and claims data from CMS; and Old Age, Survivor, and Disability Insurance and Supplemental Security Income benefit data from the Social Security Administration (Centers for Disease Control and Prevention, 2010). AHRQ has developed a number of linkages to MEPS as well. Some of these data are restricted; some, such as the MEPS and NHIS public use data sets, require completion of confidentiality forms; others, such as the linked household and insurance component files, can be accessed only at AHRQ or the Census Bureau research data centers. A number of federal statistical agencies and nongovernmental organizations (such as the National Bureau of Economic Research and the University of Michigan Institute for Social Research) have been designated statistical data centers, where many of these linked data sets reside. There is still a long way to go, however, before access to these data is satisfactory. For example, both Medicare claims data and NDI data can be linked to NHANES, however, access to these linked data sets is difficult, time consuming, and sometimes expensive to obtain. Furthermore, while some progress has been made linking data collected by different government agencies, linkages between private and public data have not yet been accomplished on any meaningful scale. While it is widely believed that the private sector does not want these linkages, this is not universally the case. A number of currently linked data sets are underutilized because researchers are unaware they exist or have not used them before. The statistical agencies could probably productively undertake efforts to increase researcher awareness of, and ease of access to, existing linked data sets and data linkage tools, such as the U.S. Census Bureau’s American FactFinder, available at http://factfinder.census.gov. Technical difficulties in reusing data—as with creating data set linkages—are substantially eased by good data management practices (including documentation) and timely collection and provision of “data about the data” or metadata (Piwowar et al., 2008). Data providers must create adequate, accurate metadata at the time of data set creation for efficient, cost-effective, quality data sharing downstream. Metadata requirements should be identified a priori. This requires
OCR for page 181
Accounting for Health and Health Care: Approaches to Measuring the Sources and Costs of their Improvement coordination among the data collectors (or agencies), the scientific disciplines, and the computer and information science disciplines. 6.6.4. Data Standardization, Quality, and Timeliness Standardization of data elements across surveys and other data sets is necessary for meaningful comparisons and for the development of a coordinated system. While some progress in developing standards has been made, surveys often continue to produce disparate estimates due to the lack of standardization. Indeed, the disease groupings in the vital statistics systems (ICD-10) are not even the same as those in the national expenditure surveys (MEPS and MCBS use ICD-9-CM); in the absence of standardization, mapping diseases from the two sources to the same groups is extremely difficult. NCVHS has made some progress developing uniform minimum data standards, but these efforts must be continued as needs and demands for data at the state and local levels grow (Rice, 2000). Greater standardization of definitions, methods, and data reporting would certainly be a prerequisite for developing disease-based subaccounts within the NHEA. Furthermore, standards facilitate data reuse, thereby increasing the funders’ returns on investment in the data. They make data sharing easier, saving overhead and losses of time in data loading, conversion, getting systems to work properly with the data received, and with interpretation. Development and maintenance of an electronic inventory of standardized data elements in a straightforward and accessible form can markedly increase the use of these data and therefore the return on the investment from standardization. Even if a system is well coordinated, some important data may be available only after a lengthy lag, making them out of date for policy purposes. Vital statistics data, which underlie some of the most important population health metrics, are collected on an ongoing basis. However, the process of cleaning the data and preparing them for public use means they are released only after a lag period of a year or more. Even the NHEAs have a 12-month lag until release, limiting their usefulness to policy makers. Furthermore, the lag increases substantially when NHEA-to-microdata linkages are required—as in the development of the disease-based health accounts discussed in this report. MEPS files become available 2 years after the year for which they are collected. The MCBS Access to Care file—which provides information on beneficiaries’ access to, satisfaction with, and usual source of care—and Cost and Use files—which link Medicare claims to survey-reported events to provide expenditure and source of payment data on health care services, including those not covered by Medicare—are released with a 1- and 2-year lag, respectively. There has been a lack of political commitment to investing in the collection of quality health statistics in the public and private sectors. Until this changes, it is even more important that the investments that are made are used wisely. While
OCR for page 182
Accounting for Health and Health Care: Approaches to Measuring the Sources and Costs of their Improvement many federal statistical agencies, such as NCHS, make considerable efforts to ensure data quality, the quality and reliability of much private-sector data are unknown. Because survey results are subject to sampling, reporting, processing, and nonresponse errors; data cannot be fully interpreted unless these errors are reported. While federal statistical surveys routinely report standard errors, they are often unavailable in reports emanating from facility and manpower private-sector surveys. Improvement in the quality and reliability of health statistics reported in the private sector is urgently needed. 6.6.5. Characteristics for a Minimum Data Set In moving forward on development of a national health account, it is essential to identify the ideal data needs to inform policy and then to specify a minimum data set, designed to collect information consistently across all relevant surveys to ensure broad representativeness. Key long-term goals include standardizing questions representing the same concepts across surveys, identifying and filling data gaps (such as populations underrepresented in or absent from national surveys), and adding questions to surveys to obtain comparable data across survey populations. Recommendation 6.10: A research project should be commissioned by the National Institutes of Health to identify the minimum data set of variables needed to support the infrastructure of the ideal health statistics system. Ideally, the integrated population data set would include some biological measures, in addition to general health measures. This idea is not new—HRS already compiles biological and genetic data, so a model and some data exist. It would certainly be worthwhile to explore the possibility of enhancing MEPS with biological information. The central idea is to have data on both expenditures and health outcomes in the same place and by the same categories. Table 6-1 provides an example of what that data element list might look like. As has been pointed out, a subset of nonmedical determinants of health will need to be collected routinely on surveys, and the research recommended above should provide guidance for the routine updating of surveys to generate data on key nonmedical influences on health. Funding agencies (e.g., NIH, National Science Foundation [NSF], Agency for Healthcare Research and Quality, etc.) should strongly consider supporting work designed to identify a small set of simple, low-burden questions that could be appended to surveys to learn more about such factors as physical environment and individual monitors (of stress, sleep hours, exercise, etc.) that affect health. The project should also catalogue the portions of the population that are and are not covered by the various data sources. It is unrealistic to think that all of the data elements needed for a health account could be collected and coordinated in a centralized repository imme-
OCR for page 183
Accounting for Health and Health Care: Approaches to Measuring the Sources and Costs of their Improvement TABLE 6-1 Potential Data Elements for a Minimum Data Set Domain Possible Variables Population Health Measures Mortality Generic HRQoLa Specific measures TBD Individual Characteristics Demographics Age, gender, race/ethnicity, social security number, marital status Socioeconomic status Income, education, marital status, employment/work conditions and hours, social support networks, health literacy, social environment, physical environment, personal health practices, health services Individuals: Physical and Mental Measures Anthropomorphic measures Height, weight, waist circumference, blood pressure, heart rate Chemistries Creatinine, glucose, direct low-density lipoprotein cholesterol, etc. Functioning ADL, IADL, mobility, cognitive limitations Individuals: Behaviors Smoking Smoking/amount Alcohol abuse Ethanol intake Physical activity Indicators of sedentary lifestyle Adherence to medical therapies Medicines prescribed/taken Health Services (ideally differentiating between acute and chronic) Primary care—outpatient Hospital-based clinic, Y/N?; diagnoses for the encounter (ICD); procedure codes (CPT); provider ID; dates of service (for inpatient and output); type and other characteristicsb Specialty care—outpatient Hospital-based clinic, Y/N? type? Hospital services Hospital services—provider Pharmaceuticals Prescription drugs taken chronically (probably more important than those taken short term).c Other (the current National Health Expenditure Accounts service categories, plus others) Core set of diseases linkable to utilization data ICD diagnoses or top 20-30 diseasesd Long-term care Palliative (end of life) care Hospice, Y/N, home/elsewhere Residential care (if included in the domain of medical care) Assisted living, retirement homes
OCR for page 184
Accounting for Health and Health Care: Approaches to Measuring the Sources and Costs of their Improvement Domain Possible Variables Identifiers and other medical claim form information Unique patient identifiers, ICD-9 codes for admission diagnosis, unique provider ID, dates of service, codes for identifying inpatient procedures performed, Healthcare Common Procedure Coding System for outpatient (such as lab work), service charges (amount billed), allowed amount, and paid amount.e Identifiers and other pharmaceutical claim form information Unique patient ID, Unique pharmacist ID, National Drug Code and quantity of specific medicine, prescribing physician identifier, allowed amount, paid amount Identifiers and other hospital discharge abstract information Hospital identifier, admission source, discharge status, admission and discharge dates, length of stay, type of secondary insurance, patient demographics, ICD-9-CM diagnosis and CPT codes, assigned diagnostic related group and major diagnostic category, total charges Other Determinants of Health Environment Climate, air quality, access to clean water Geographic hospital data Wages, supplies, other hospital input prices, payer mix, disproportionate share status, hospital type (academic, community), rural, urban aHRQoL indicates health-related quality of life. bFor example, a measure of how concentrated the medical care market is. cIf it is not feasible to include entries for each drug, the top 20-30 (by sales) could simply be listed. dThere are many ways to approach which diseases are most important but, again, one simple way is to rank by cost of illness. ePaid amount will be the hardest to get for commercial claims, but for many purposes it is the most important. diately. The infrastructure will have to grow incrementally, and therefore data collection will have to be prioritized. In addressing how to pick which data to collect from a potentially endless list of factors (health, social, environmental, financial) that may affect current and future health and health care spending, potential criteria for the selection of a set of key indicators include the importance of what is being measured in terms of its impact on health status and health expenditures, the policy relevance, and the susceptibility of the problem to intervention; the scientific soundness of the measure in terms of its validity, reliability, and evidence base; and the feasibility and cost of obtaining nationally comparable data for the measure.
OCR for page 185
Accounting for Health and Health Care: Approaches to Measuring the Sources and Costs of their Improvement It will be important to develop a consensus on and formalize a set of criteria by which the selection of key data for additional collection, standardization, etc., is prioritized. The development of a functional, policy-responsive, integrated national health data system is clearly a long-term proposition. That said, it is an achievable goal made all the more possible by the ongoing collaborative efforts of the national data agencies. It is the hope of the panel that our full set of recommendations provides a logical set of building blocks that—together but implemented separately—would contribute significantly to a more coherent and policy-responsive system of health statistics. 6.6.6. Privacy Protection and Confidentiality The research environment is increasingly complex, both because of rising public concerns about the privacy of individuals’ personal information and because of the complexity and lack of understanding of the legislative and regulatory frameworks guiding the conduct of research using such personal information. Since the passing of HIPAA in 1996, there has been considerable confusion over how the patient privacy rules affect quality improvement studies. From a legal standpoint, HIPAA rules do not apply equally to everyone. Public health authorities can access patient data without consent in their efforts to prevent disease outbreaks, and hospitals can use data to improve quality of care. The problem, however, lies in defining the boundaries of these activities. When is quality improvement clinical care, and when is it research? To cite one among countless examples, is a study of an intervention to estimate the reduction in rates of catheter-related bloodstream infection (Pronovost et al., 2006) simply research or is it part of care? The problem is that the line is not clearly drawn, and immediate concerns over privacy seem to supersede the longer term needs for research. At a 2003 NSF workshop on confidentiality research, Peter Madsen described this tension as the “Privacy Paradox”: The rush to ensure complete levels of privacy in the research context paradoxically results in less social benefit, rather than in more…. [T]hrough the additional concept of utility, people will recognize that while they surely have the right to privacy, they may also come to the realization that they have a duty to share information, if the common good is to be furthered. In addressing HIPAA challenges to research, a balance must be struck between the public’s right to know and to pursue improved health care and the right of individuals and institutions to protect their privacy. The Institute of Medicine recently concluded that the HIPAA privacy rule is not only inadequate in safeguarding patient privacy, but also significantly impedes secondary research (Institute of Medicine, 2009). Privacy and the safeguarding of personal information against unauthorized disclosure are fundamental individual
OCR for page 186
Accounting for Health and Health Care: Approaches to Measuring the Sources and Costs of their Improvement goods in terms of respecting personal dignity and protecting patients from discrimination. Privacy also holds societal value because it encourages individuals to participate in socially desirable activities like research. At the same time, research is an equally compelling individual and societal good that can help address some of the nation’s most pressing health problems. Indeed, research on the determinants of health can help guide national efforts to focus life-saving interventions where they can do the most good in terms of improving individual and public health (Gostin and Nass, 2009). Although informed consent is meant as a safeguard, it can also be a barrier to valuable research. The Institute of Medicine has proposed new rules that would make health research exempt from HIPAA privacy rules and emphasize data security, transparency, and accountability, regardless of the funding source. The proposed rules would add consistency in regulatory oversight and ensure protection of participants. In addition, the new system would include two alternatives to consent: (1) a system with ethical oversight to protect data privacy and security and (2) a certification system that would allow researchers to link deidentified data sets. The panel supports these recommendations.