NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

National Research Council (US) Panel to Advance a Research Program on the Design of National Health Accounts. Accounting for Health and Health Care: Approaches to Measuring the Sources and Costs of Their Improvement. Washington (DC): National Academies Press (US); 2010.

Cover of Accounting for Health and Health Care

Accounting for Health and Health Care: Approaches to Measuring the Sources and Costs of Their Improvement.

Show details

2Medical Care Accounts and Health Accounts: Structure and Data1

In Chapter 1, we identified and distinguished two useful—and nonconflicting— types of health-related economic accounts: one measures the output of the medical care sector and the inputs that contribute toward its production, and the other tracks population health and the factors affecting it. The first account, the medical care account, includes the usual economic inputs—capital, labor, materials, and so forth—plus certain nonmarket elements such as time spent in caring for the ill and household production of medical services. In the second account, the health account, inputs include medical care (the output of the first account), plus nonmedical health-determining factors including, but not limited to, the environment, diet, and health-affecting behaviors and time spent in health-affecting activities. The two accounts are complementary because the first feeds information into the second and because their information is relevant to different types of questions.

In addition, we distinguish from these two accounts and recommend, in Chapter 6, a data system for research on the determinants of health. This is not an account, as that term is normally used in the national accounts literature, but rather a research database. The rationale for recommending the research database on the determinants of health over a full-fledged health account is presented in section 2.2.3.

2.1. LINKS TO ECONOMIC ACCOUNTS

Both the medical care account and the health account have similarities with the well-known National Income and Product Accounts (NIPAs) produced by the Bureau of Economic Analysis (BEA). The NIPAs present gross domestic product (GDP), the measure of the nation’s overall market output, as well as its major components (consumption, investment, government spending, and net exports). The NIPAs provide essential information about economic growth, trends in the mix of goods and services produced and purchased, international trade balances, and the course of business cycles.

The NIPA are used extensively in economic decision making. For example, the Federal Reserve Board and executive branch agencies monitor various indicators in the accounts for purposes of carrying out monetary and fiscal policy. Because the NIPAs are a data-integrating system, they can show shortcomings, lacunae, and inconsistencies in the outputs of other statistical agencies, so they also play a central organizing role in economic measurement. Many excellent sources of information about the NIPAs exist so in this report we do not go into further detail about their construction and attributes.2

BEA includes in the NIPAs a set of industry accounts, which distribute GDP among major industries and sectors of the economy. They show industry output and also input usage. Medical care provider industries (hospitals and so forth) are included in the BEA industry accounts. In addition, one can extract from GDP components estimates of total spending on medical care and on the major elements of medical care—for example, household and government spending on hospital care. Total spending on medical care is larger than the total output of the medical care–providing industries in the industry accounts, largely because direct purchases of pharmaceuticals and medical devices by households cannot be allocated to the conventional medical care provider industries (hospitals and so forth—see below for additional discussion).

It has long been recognized that GDP is a measure of output, not a measure of welfare. Moreover, the NIPAs are organized around market output—that is, activities in which money changes hands or that are sufficiently similar to market activities that market data can be used to make an imputation (the most important imputation is the one for the value of owner-occupied housing). Nomarket activities, such as unpaid time spent caring for ill persons and personal investments in one’s own health, are not included in GDP. To measure the costs and benefits of such activities, alternative accounts—ones that include elements outside the limits of the present NIPA system—have been proposed. The medical care account we describe is predominantly, though not exclusively, a market-output type of account. The health account is largely nonmarket in character.

2.2. MEDICAL CARE AND HEALTH ACCOUNTS CONTRASTED

Any economic account incorporates an economic framework. The two accounts we consider in this report are not exceptions. The medical care account is constructed around the relation between the output of medical care services and the inputs that are used to produce these services. The inputs are conventionally notated as KLEMS:

  • capital services (K);
  • services—the vector of all labor inputs, from surgeons to janitors (L);
  • energy (E);
  • intermediate or purchased materials, which, in medical care–providing industries, includes pharmaceuticals used in hospitals and clinics (M); and
  • purchased services (S).

That is

medical services=f(KLEMS)
2.1

where all the variables, including output, should be understood as vectors, the elements of which, however, are frequently aggregated for analysis. Labor services may include, in addition to market labor, uncompensated time spent in the care of others; and capital services include those of intangible capital, such as research and development (R&D), in addition to the usual physical plant and equipment elements. In section 2.4, we discuss the inputs to the medical care account; in section 2.5, we turn to the output.

A health account, similarly, records the relationships between an output—in this case, a measure of health, which is multidimensional—and the inputs that produce it (or, alternatively, the determinants of population health). Table 2-1 lists some of the determinants of health. Thus, in parallel with equation 2.1, there is an equation for the “production” of health:

TABLE 2-1. Inputs and Outputs of Health, Market and Nonmarket.

TABLE 2-1

Inputs and Outputs of Health, Market and Nonmarket.

health=h(medical services,time,diet,lifestyle,R&D,environment,genetic components,and so forth).
2.2

Among the inputs in equation 2.2 include medical services (the output in equation 2.1), including those originating from nonmarket sources, particularly households’ provision of care. Nonmedical inputs include time spent by people investing in their own health, consumption and lifestyle variables, R&D, and the quality of the environment. Genetic endowments also play a role, especially for explaining health differences across groups and perhaps internationally.

Clearly, identifying all the variables that determine health is a broad and multidisciplinary research area. Research on this task is under way from many sources, but the medical knowledge that would permit making a complete list is a long way off. Nevertheless, much is known. That much also remains to be understood does not preclude starting on the task of accounting for health and its determinants.

Even among the variables that clearly belong in equations 2.1 and 2.2, some ambiguities arise. For example, individuals who are ill may generate time demands on relatives for managing their financial affairs; although this is a cost of their illness that is borne by others, it is not clear that it would be put into either account.3 Individuals may also spend time investing in the health of relatives and family members by, for example, preparing more healthful meals instead of relying on processed foods or participating in programs to encourage more exercise by other family members. Although such investments are similar to the time most directly relevant to the health account, spreading the net too broadly complicates measurement problems and, in the end, some boundary must be established on the inclusion of time.

Some consumption items may contribute positively to health while others, such as tobacco and some well-liked items in the typical diet, contribute negatively in the long run. As Grossman (1972) pointed out, abstention from things like tobacco and fatty foods are like investments, in the sense that abstaining from consuming them reduces utility today for the sake of benefits in the future (see also Philipson and Posner, 2008).

The R&D that appears in the health account is not the R&D that is capitalized in the medical care account. R&D in the medical care account consists of, for example, development of new pharmaceuticals or new medical procedures;4 R&D in the health account includes research that demonstrates the effects of smoking cessation, of healthy diets, or of exercise on health. In the familiar paradigm, R&D in the medical care account augments the inputs to medical care; the additional, and different, R&D in the health account augments the nonmedical determinants of health. Maintaining this distinction empirically is not easy.

The output in equation 2.2—the level of health—includes both length and quality of life, as discussed in detail in Chapter 5. It can be implemented as quality-adjusted life expectancy (QALE) and expressed either in years or dollars (Murphy and Topel, 2006).

2.2.1. Productivity Measurement and the Two Accounts

Differences between the medical care and the health accounts can be illustrated by their potential uses in productivity analysis. We do not mean to imply that the accounts are useful only for productivity analysis, or even that this is their primary purpose. Productivity measurement has the virtue that it makes use of all the data in an economic account and therefore throws relationships into stark relief.

A full or complete productivity measure is the ratio of outputs to all inputs. Productivity growth is the ratio of changes.5 Thus, the medical care accounting relation in equation 2.1 implies

productivity change (medicalcare)=growth of medical services/growthin f(KLEMS)
2.3

Equation 2.3 is conventionally estimated as a ratio of index numbers6 and is known as multifactor productivity (MFP) growth.7

Productivity growth is often interpreted as a measure of efficiency change. Thus, equation 2.3 represents the growth in efficiency in the use of resources to produce medical care services. Significantly, Triplett and Bosworth (2004) reported that MFP growth in medical care services (more precisely, North American Industry Classification System [NAICS] sector 62) between 1987 and 2001 was negative, at a rate of about 1 percent per year.8 They attributed the improbable negative productivity growth in the sector to data inadequacies in the measurement of medical care output and also to mismeasurement of several inputs, particularly the high-tech portions of medical equipment. We discuss some of these measurement problems in this chapter and others in Chapter 4.

Measurement inadequacies in the medical care account will affect the health account estimates because the output of medical services (from the medical care account) enters the health account as an input. This, among other reasons, demonstrates why one cannot create an account for health without giving major attention to improving the medical care account.

The health account also implies a productivity measure, derived from equation 2.2:

MFP change (health)=growth of health/growth of h ()
2.4

where h (•) is short for the right-hand side of equation 2.2. In this case, productivity change can be interpreted as a measure of efficiency in the use of all of society’s resources that affect the population’s health status. Computing MFP in the production of health requires, as it does in any MFP measurement, measuring all the inputs (or as many of them as can be identified) and finding an appropriate way to summarize them, analogous to the index number formula for inputs in equation 2.3. Representing the health function may be complicated because of the heterogeneity in the units in which the input variables are measured.

Partial productivity ratios may also be calculated. For example, measures of labor productivity are common. Based on equation 2.1, labor productivity (LP) growth in the medical care sector9 can be expressed as:

LP (medical care)=(medicalservices)/L.
2.5

Similarly, one may also be interested in the productivity of the resources used in the medical care sector in the improvement in health. Indeed, that is one of the most pressing policy issues of the day. Using equations 2.2 and 2.3, this measure of productivity growth (which we designate Ω) can be expressed as:

Ω=(health)/medical care=(health)/[f (KLEMS)],
2.6

where ∂ designates partial derivatives of the variables in equation 2.2. As with any partial derivative, the value of Ω will depend on the values of the other variables in the equation. Thus, the productivity contribution of medical services to improved health will depend on the value of other health inputs such as diet and the environment.

Because the contribution of medical care to health depends on other health-determining factors, simple comparisons of changes in medical care and changes in health are seldom meaningful. That is

ΩΔ(health)/Δ(medical care).
2.7

One often hears such statements as the following: The U.S. medical care system must not be efficient or productive because the nation does not have the highest health level in the world, even though it has the highest per capita spending on health care. This statement is fallacious logic because the impact of medical care expenditure on a nation’s health cannot be established without considering nonmedical determinants of health. For example, benefits resulting from expenditures on cholesterol-reducing statin drugs may simply be offsetting greater than average obesity rates for the United States, so there may be no net gain in health relative to a society that is faring better in terms of related social determinants of health. Philipson and Posner (2008) suggest the inverse relationship between obesity and income is a consequence of a fall in the price of consuming calories and a rise in the cost of exercise (once a by-product of manual labor), which higher income groups can better afford: the medical system partly offsets the negative health impacts of obesity.10 These examples demonstrate that the relationship between a nation’s medical expenditures and its health is not a straightforward function of its per capita expenditures on medical care.

Measuring the nonmedical determinants of health, though essential, is very difficult, which is why assessing the impact of medical care on health is also very difficult. Separating the influences of medical and nonmedical determinants of health is why estimating an accounting for health would be so valuable.

2.2.2. Strategies for Going Forward

A solid case can be made for beginning with and emphasizing the medical care account. For policy purposes, the most pressing needs are to measure properly medical care expenditures and outputs, to improve measures of medical inflation, and to determine what part of increasing medical care costs are attributable to increases in medical services, as opposed to price change. In addition, accurate expenditure and output data on medical care, developed and presented by a detailed cost-of-disease metric, are essential for a “health” account. Finally, in terms of feasibility, much data on medical care goods and services already exist; the challenge is to improve these data and to array them using a more useful organizing principle.

In contrast, it will be difficult to collect data on the full range of behaviors and activities that affect population health. Nevertheless, nonmedical inputs to health matter greatly (McKeown, 1976; Mokyr, 1997); one cannot understand changes in health, and health differences between countries and population groups, without considering the nonmedical inputs. It is essential to begin to collect and maintain data on nonmedical and nonmarket inputs to health and to organize them into an analytic framework. Even if it is not currently possible to measure all the nonmedical determinants of health precisely, it is important to think through the measurement and conceptual issues.

2.2.3. Priorities: A Health Account or a Database on the Determinants of Health?

Grossman’s (1972) “production function” approach to the analysis of health and its determinants (incorporated into equation 2.2, above) has become the standard for economists’ thinking about the subject (see Bolin, Jacobson, and Lindgren, 2001). Yet attempts to implement the model empirically are few. The difficulties include the fact that health is multidimensional and—more problematic—that identifying and measuring its determinants are complex tasks.

Rosen and Cutler, in an ongoing project, are estimating disease models and combining them with economic data (see Chapter 3; Rosen and Cutler, 2007). Their research is couched within a Grossman-type framework and the related epidemiological perspective, so one can think of it as estimating equation 2.2 on a disease-by-disease basis. They have selected disease categories that account for a large portion of national health care expenditures and are working toward determining the factors—including medical care—that affect changes in mortality and morbidity. A second effort along similar lines for cancer and circulatory diseases in England is reported in Martin, Rice, and Smith (2008).

Beyond the Market: Designing Nonmarket Accounts for the United States (National Research Council, 2005) suggests constructing a health account that would provide a welfare-oriented measure as a counterpart to the market-oriented measures of the NIPA. Thus, it would be structured by analogy to the familiar national accounts that record economic activity but would be built around the functional relation and the variables in equation 2.2. Beyond the Market begins by identifying “gaps” in the existing national accounts that arise because their measures of outputs (and inputs) are incomplete—only market inputs and outputs are included. If the goal is to fill gaps in the NIPA coverage and structure, then it seems reasonable to work out an expanded accounting system that is patterned after the traditional national accounts structure.

However, assembling the data for an economic welfare account for health goes beyond the requirements for a database for research on health determinants. For example, Beyond the Market lists diet among health determinants, certainly an important consideration. The report of the World Cancer Research Fund/American Institute for Cancer Research (2007) summarizes evidence connecting dietary factors to different types of cancer—consumption of red and processed meats raises the risk of colorectal cancers, and excessive consumption of salt for stomach cancer, while consumption of fresh fruits and vegetables reduces risks of a number of digestive system cancers. A research model for cancers might be designed in which dietary data are employed in conjunction with medical care data to determine the relative impacts of diet and medical care on cancer death and incidence rates. If the objective is to estimate dietary determinants of health, then information on consumption of the foods of interest (data that are readily available) is the main requirement.

In the Grossman (1972) model, and from a welfare perspective, it is necessary to compute the net gain (Nordhaus, 2003), which is not the utility from improved health. It is the value of the increment to health, less the loss in utility from abstaining from steak and ham or from eating vegetables one does not like.11 One can conceive of a research project to measure such utility losses and, though more difficult, perhaps include them in an aggregate welfare estimation. It is much less clear, however, how utility losses should be fitted into a welfare-oriented health account patterned on a NIPAs-type economic accounting structure.

One possible parallel is with environmental accounting. Some production processes (electricity generation, for example) produce both goods and “bads” (the bad in this case is pollution). In an environmental account, one subtracts the values of the bads from the goods to get a net welfare measure. In the health case, the bads are utility losses from pursuing more healthy lifestyles. The net output (the utility of health gains from changes in diet and lifestyle less the loss of utility from giving up things that give present utility but in the long run are deleterious to health) is the relevant measure for welfare and therefore for the NIPA-analog health account.

In principle, both objectives—an economic welfare account and a database for research on health determinants—should be pursued. However, in programs to generate data, choices must be made and priorities established. A database on the determinants of health that includes a measure of health has the most immediate policy value. It seems inevitable that estimating the utility loss from health-promoting lifestyle and dietary changes will compete with resources for moving forward on this work. If so, it is our judgment that the database for research on the determinants of health should receive higher priority. First, it is more immediately useful for a wide range of purposes. Second, information on the determinants of health is necessary for a welfare-type account in any case. Estimating the net gains from changes in lifestyles can follow.

Thus, it is not premature to recommend the collection of more information about the determinants of health. It may, however, be premature to recommend that statistical agencies organize health data to accommodate a health account of the welfare-oriented NIPAs type. As was true of the development of national economic accounts in the 1930s, health accounts have not yet evolved very far, but the situation should change as more work on their conceptual underpinnings and practical needs is undertaken.

The conflict between competing data needs is reduced the larger that the proportionate contribution of medical care is to the change in health. Historically, medical care was not the main determinant of past health improvements (McKeown, 1976). However, Cutler (2004) contends that medical care has been the main factor over the past half-century, particularly if one adds in the contribution of medical research to the information that has led to changed lifestyles. Even so, he concludes from an informal calculation that the loss of utility from changed lifestyles and dietary changes costs around half the gain in utility from increased health. Thus, the net gain remains substantially less than the gross.

The arguments laid out in this section provide some of the logic underlying Recommendation 1.1 that BEA should produce an account for medical care. In addition, developing a database for research on the determinants of health, as discussed in Chapters 5 and 6, should receive high priority from a range of players. Components of this database will no doubt continue to be developed by a number of the health-oriented statistical agencies, by academic researchers, and through cooperative research ventures between the two; BEA will also have a role and may provide input about which data elements would be useful for its programs. Even though a welfare account for health of the type advocated in Beyond the Market would also be of great value and should retain the interest of researchers, for the present developing this account will, and probably should, receive lower priority from statistical agencies.

2.3. STRUCTURE AND DATA FOR A MEDICAL CARE ACCOUNT

We noted earlier that, in the NIPAs, BEA produces industry accounts that link inputs and outputs, with detail approximately at the sector level of the NAICS. BEA has recently introduced a KLEMS input structure for its accounts for all sectors, including medical care (Moyer, 2008). Thus, the current industry accounts provide a good starting place for producing a medical care production account. This and the following two sections—buttressed by detail in the annex to this chapter—suggest improvements that will create a medical care account with the characteristics needed for analyzing the sector. Chapters 3 and 4 provide additional analysis and recommendations.

2.3.1. Account Boundary

Medical Care and Social Services (NAICS Sector 62)

Our discussion of boundaries for the medical care accounts begins with the definitions in the NAICS because many of the data that will be used in the account’s construction will generally conform to it. For example, data were collected by NAICS industry definitions for the 2007 U.S. Economic Census and the Producer Price Index (PPI).

We begin with NAICS sector 62, which encompasses medical care and social services. Within this category are, at the three-digit subsector level, ambulatory care (621), hospitals (622), nursing homes (623), and social assistance (624). Lower-level industries are nested within the 3-digit levels: ambulatory care (621) contains seven 4-digit industries (offices of physicians, 6211, and medical and diagnostic labs, 6215, are examples of 4-digit industries in ambulatory care), and those 4-digit groupings are in some cases subdivided into 5- and 6-digit industries (for example, NAICS industry 62132 is offices of optometrists, under NAICS 6213, offices of other health practitioners).

The level at which a medical care account will be constructed depends on balancing several competing factors. On one hand, more detailed industry-level accounts are preferable because they give more analytically relevant information and minimize aggregation difficulties. The size of the medical care sector is large enough to make subsector (3-digit) level analysis a data priority; also, aggregation conditions favor this approach because the production functions of the separate NAICS 3-digit subsectors seem quite distinct from one another.12 On the other hand, data availability hinders developing accounts at too detailed a level. For example, data on inputs for building an account for the production of optometrists’ services (NAICS 62132) are not currently available. The NAICS 3-digit subsector is also the level at which medical care appears in BEA’s industry accounts programs. Thus, accounts corresponding to equation 2.1 should be constructed for ambulatory care (NAICS 621), hospitals (NAICS 622), and nursing homes (NAICS 623), as well as an account at the sectoral level.

The present BEA program also contains an industry account for social assistance, NAICS 624. Although parts of NAICS 624 have some connection to health or health status (for example, NAICS 62412, services for the elderly and persons with disabilities), most of subsector 624 does not. Accordingly, a more useful sectoral account for medical care would omit the data for NAICS subsector 624.

The three subsector accounts (NAICS 621, 622, and 623) could then be aggregated to form an account that includes only medical care, that is, NAICS 62 less social services. Alternatively, the NAICS 62 sector account can be estimated directly. In the 2002 Economic Census, the three subsectors accounted for the following proportions of the total receipts of NAICS 62 less social services (U.S. Census Bureau, 2002 Economic Census Geographic Area Series summary statistics, see http://factfinder.census.gov):

Ambulatory care (NAICS 621)44%$488.7 billion
Hospitals (NAICS 622)45%$500.1 billion
Nursing homes (NAICS 623)11%$127.1 billion
 Total100%$1,115.9 billion

Nursing homes are the smallest subsector; however, with $127 billion of receipts in 2002, it is surely large enough to merit constructing an economic account for it alone (BEA currently combines nursing homes with hospitals). In addition, it is a portion of medical care in which the disease unit of output may so that the producing units have closely similar production processes, so far as possible, and so that dissimilarities in production processes provide the “breaks” that separate one industry from another. Triplett (1990) provided the conceptual bases for the industry structure of the NAICS. not be entirely the appropriate one, which provides another reason for keeping nursing homes separate from the hospital subsector.

Augmenting the NAICS Medical Care Provider Subsectors with a Household Production Account for Medical Services

We noted earlier that one can obtain from the NIPA an estimate of total national spending on medical care. This total exceeds the output of the medical care providers because of (among other things) households’ direct purchases of pharmaceuticals and medical devices, as well as expenditures of government hospitals. Direct household purchases are recorded in the Personal Consumption Expenditures part of the NIPAs and obviously do not go through the medical care provider industries in NAICS 62 (only pharmaceuticals and medical devices provided in such industries as hospitals, nursing homes, and clinics are included in the inputs to NAICS 62 and its subsectors). Yet the medical care expenditures in equation 2.1 should match the total medical care inputs into the health equation (equation 2.2).

A natural way to reconcile the output from BEA’s NAICS industries to the national spending total is to construct a household production account for the provision of medical care services, that is, to add a household “industry” to the existing medical care provider industries. The household production account can then be aggregated into the medical care sector—households can be treated as supplying some part of medical care. That makes sense, because they do, and it is also consistent with putting household time spent caring for ill persons in the inputs for the medical care account, which we also recommend below.

A great amount of interest in household production accounts has accumulated recently, but probably too little attention has been paid to medical care inputs from the household. Ideally, BEA would develop medical care and household production accounts in concert. However, a full-blown household production account is not immediately on the horizon. In contrast, a medical care sector account drawn along current industry lines is practical now and is the subject of a great amount of current data development. One would not want the development of a medical care sector to wait on its lagging component. Nonetheless, a medical care account without drugs, devices, and durable medical equipment that are purchased at retail by households will not be as useful as one with a broader definition that matches the national expenditure total for medical care.

The NAICS 3-digit subsector level is the best starting place for constructing a medical care account. However, BEA should augment the existing NAICS industries with a household production account for medical care services. This account should initially include the portions of drugs and devices not in NAICS 62, along with the output of other medical providers, such as government hospitals, to which a dollar value can be readily attached. It can later be expanded to include other, nonmarket household inputs.

Recommendation 2.1: An account for medical care should encompass as its core the North American Industry Classification System (NAICS) 3-digit subsectors ambulatory care (NAICS 621), hospitals (NAICS 622), and nursing homes (NAICS 623) and be aggregated up to the level of NAICS 62 less social services (NAICS 624); alternatively, a sector aggregate might be estimated independently. To this core, a household production sector that will account for other spending on medical care, such as on pharmaceuticals and medical equipment, should be added.

In constructing a medical care account that extends beyond NAICS 62, BEA would undoubtedly also keep a NAICS 62 aggregate. Indeed, it would be hard to do a broader aggregate without first estimating NAICS 62 from the existing (or improved) industry data. BEA will likely encounter reconciliation problems that have not yet been thought through, but such challenges arise in many areas of the national accounts.

Currently, an estimate of national spending on medical care exists and is useful. It would also be useful to have a production account for the medical care sectors, which does not now exist, and to have it map into the national spending total. One need not, and should not, get in the way of the other.

2.3.2. The Medical Sector Boundary: Other Options

One could contend that the sector boundary we have drawn is too extensive. For example, part of the output of nursing homes is custodial—food, housing, laundry, and so forth. These services are not necessarily related to health, and probably the nursing home provides no more of them than other living arrangements. It might be contended that custodial or housing services should be eliminated from the nursing home industry (and from the hospital subsector, on the same logic) to produce a purer unit of medical care output.

However, a narrower boundary that excludes custodial services confronts data collection and conceptual difficulties. Collecting information about revenues from different services is feasible in industry data, if records are kept that way, so the value of custodial care might be collected, or even estimated, and removed from output. But even when the outputs can be separated, many of the inputs are joint and cannot be. For example, if the nursing home industry is inside the boundary, then all the output of a nursing home associated with all of its inputs (ranging from pharmaceutical purchases, to food, to custodial services) has to be included, unless the nursing home has records that separate its inputs by the services they provide. As an example, Dawson and colleagues (2005) estimate from National Health Service data the expenditures by hospitals in the United Kingdom on food and cleaning, in order to adjust their measures for improvements in custodial amenities; they note, however, that this leads to inevitable double counting because it is not possible to partition cleaning costs into the part of cleanliness that has medical implications and the part that augments custodial amenities. In general, the output vector in an industry account must line up with the inputs; accordingly, the boundary for an accounting of medical care outputs and inputs usually must be drawn along NAICS lines as a practical matter.13

In summary, in the medical care account, data constraints make it difficult to separate the strictly medical outputs of medical care industries from the value of nonmedical services that may be provided to patients as ancillaries to their treatments. It is clear that hospitals have extended their “hotel” functions over time, in response to patients’ demands for improved living standards. These hotel-like amenities are provided more abundantly by U.S. hospitals than by hospitals in most other countries. At this point, it is not productive to spend much time debating the extent to which these ancillary services do or do not belong in a medical care account since, as a practical matter, it is unlikely that data could be found that would permit eliminating them from the existing totals on both the output and input sides.

The panel also considered other reasons for adopting a narrower definition. Some medical procedures have less direct or obvious health impacts. Examples that might be excluded, even though medical, are certain kinds of cosmetic surgery not covered by insurance (reconstructive surgeries are typically covered). However, the panel decided that trying to draw the boundary by recourse to insurance coverage was problematic, in part because insurance coverage lines are not uniform and are often in flux. For example, fertility treatments are sometimes covered and sometimes not; other categories of health care that are not uniformly covered by insurance include over-the-counter drugs and annual physicals. For many policy analytic and research purposes, it is important to track these kinds of treatments in the data system, even if they are not productive in improving population health status.

2.4. INPUTS TO THE MEDICAL CARE ACCOUNT

For the year 2002, BEA’s industry account for NAICS 62, a major component of the ideal medical care account, indicated the breakdown of inputs to medical care shown in Table 2-2. Similar tabulations appear in the BEA industry accounts for NAICS subsectors 621, 622, and 623, described above. With the provisos that elements of household production should be included and that social services need to be removed from NAICS 62, as noted above, these data serve to introduce a discussion of data needs to support the input side of the medical care account. The discussion applies, unless otherwise stated, to inputs for the sector aggregate (NAICS 62 less social services) and to the inputs for the three individual subsectors recommended above. We focus on this industry account because of its central position in the medical care account.

TABLE 2-2. Inputs to the Medical Care Account ($ billions) (% of output).

TABLE 2-2

Inputs to the Medical Care Account ($ billions) (% of output).

Figures in Table 2-2 for NAICS sector 62 are the shares of inputs in the total receipts of the sector. Of course, expenditure on an input equals its price times the quantity of it that is purchased. Compensation of employees equals compensation per hour times the number of hours of employment in the subsector; the calculation is similar for other inputs such as purchased services. The change over time in medical sector expenditures on any input equals a period-to-period price change measure times the growth in the quantity of it purchased.

A full medical account would show each input represented by a price and a quantity change measure. For example, it would record not only total compensation, but also the hourly compensation (price of labor) and the amount of labor used as well; the same would be true for other inputs, including capital. The quantity change measures are essential for an analysis of the consumption of inputs, of productivity, and of the efficiency of the sector (as is true for the accounts for other sectors of the economy). The measurement problems associated with determining the shares of inputs to medical care are not particularly difficult (although there are some).14 A much more challenging task is separating the changes in input shares into price and quantity elements. Improving price and quantity measures for medical care inputs provides the focus of the following sections.

2.4.1. Capital

Capital is the most complicated of the inputs to medical care, which justifies an extended discussion. Its contribution to output is the flow of services provided by the industry’s stock of capital goods—physical capital (such as buildings and equipment) and also intangible capital (which includes, among other things, intellectual property).15 The flow of capital services is derived from the stock (see Schreyer, Diewert, and Harrison, 2005). The price of capital services, in concept, is the charge for the use of a capital good for a unit of time. Estimating the price and quantity of capital services is complex because it requires (except when the capital good is rented or leased, and the lease payment provides the price measure) determining the capital stock, deflators for the stock, and measuring depreciation, all of which pose major empirical difficulties. However, medical capital goods pose no difficulties that are not already familiar ones in work on the topic for other sectors of the economy. The methods, accordingly, are not reviewed here (see Schreyer, 2001).

BEA now publishes capital stock for the medical care sector, tabulated in two ways:

  1. by type: medical equipment and hospital and special care structures, and
  2. by industry: an aggregate equipment and structures capital stock for all NAICS medical 3-digit subsectors.

The National Health Expenditure Accounts (NHEAs) have long contained data on investment in medical structures and now use BEA medical equipment data.16 The NHEAs have traditionally also included investment in education in their accounts. However, the NHEAs do not estimate capital stock, and their flow-of-funds approach makes the estimation of capital services not relevant. Accordingly, the BEA capital estimates are the focus of our attention, beginning with a discussion of data needs for the capital services category.

The new BEA KLEMS input series contains the share of capital services in industry receipts for each industry, including data for NAICS 62 and its constituent subsector groupings. However, it is still true that BEA produces the capital stock series while the Bureau of Labor Statistics (BLS) estimates capital services. Even though the BEA stock data are inputs into the BLS services data, BEA does not include the quantity and price measures of capital services in its KLEMS accounts (only their product). This imposes costs on researchers because related data must be obtained from different sources, with the attendant possibilities for errors in aligning the appropriate series, as well as creating the need for dealing with any inconsistencies among them. While BEA and BLS have historically worked well together, this is a case in which coordination could be improved.

Recommendation 2.2: The Bureau of Economic Analysis (BEA) should add capital services and capital services prices to the KLEMS measures in its medical care account. BEA should also make the measures available to data users on its website and in its publications.

Several relatively unpublicized yet vital data issues arise in obtaining capital measures for medical care industries. The problems are threefold.

First, for services industries the Census Bureau collects far less information on purchased inputs, including capital inputs, than it does for manufacturing. For example, on the 2007 Economic Census form for hospitals (NAICS 621), the only input information collected was employment and payroll, although a great amount of detail was collected for output activities and receipts. In contrast, the census form for the electromedical and electrotherapeutic equipment industry (NAICS 3345) lists, in addition to payroll, 12 classes of inputs, identified by Census Bureau material codes, for which expenditure information was collected.17

Hospitals, of course, are in services; electromedical equipment is in the manufacturing sector. Differences in their coverage of inputs stems from different Census Bureau policies for collecting information in the goods-producing and services-producing sectors. The resulting data incongruity handicaps analysis of services industries, including medical care industries, which require the same kinds of information that has long been provided for goods-producing sectors.

Critics of the U.S. medical care system have frequently asserted that it overuses imaging devices, partly because once this expensive equipment is put in place, the incremental cost of images is low, and they are often used for applications that have low value. It is accordingly bizarre from a research and policy analysis standpoint that data collections in the Economic Census do not even tell us how much imaging equipment is going into the medical care sector, let alone the total stock of it that is in place.

Recommendation 2.3: The Census Bureau should take steps to meet the urgent need for detailed data on inputs and capital goods purchases—especially those of technologically advanced intermediate inputs and capital goods—by service sectors such as medical care. Such information is important for research and policy analysis. If capital goods investment and stock information cannot be generated for service industries generally, then special surveys of capital goods purchases in the medical care sector should be undertaken, which should include not only investment but also stocks of technological equipment in place.

More detailed information supporting this recommendation appears in the annex to this chapter.

A second source of problems arises because surveys of industries that produce medical capital equipment have inadequate detail. Compounding the problem, the detail that is published suffers from inconsistent product classifications in various surveys that need to be used together and from archaic classifications that no longer match the technology. Evidence of the latter are surveys in which the largest product group is labeled “other” (which means that the detail is not useful). More information on this serious data problem appears in the annex to this chapter, which provides additional support for the following recommendation.

Recommendation 2.4: The Census Bureau, Bureau of Economic Analysis, and Bureau of Labor Statistics should work jointly to harmonize the classifications and structure of their published data on electromedical equipment and to reduce, so far as practical, the share of shipments falling into the “all other” categories in the Current Industrial Reports.

Deflators for equipment are the third major problem in medical capital data. All capital goods—including medical care capital goods—require price indexes as deflators, in order to convert information on expenditures into information on quantities.

BEA generally obtains capital goods deflators from the PPI and from the International Price Index (IPI) program. The PPI indexes for medical care capital equipment have serious shortcomings, and the IPI suffers from insufficient detail—even more so than the other measures we have discussed. The lack of agreed-on product lists, discussed above and in the annex, contributes to the inadequacy of the deflators.

PPI medical capital goods deflators include items falling within the following industry codes:

  • 334510: electromedical apparatus manufacturing, which includes the following:
    • 334510-1: electromedical equipment, including
      • medical therapy equipment,
      • medical diagnostic equipment, and
      • parts and accessories;
    • 334510-3: hearing aids;
  • 339112: surgical and medical instrument manufacturing (including an index for catheters);
  • 339113: surgical appliance and supplies (including an index for artificial joints);
  • 339114: dental equipment and supplies (including separate indexes for equipment and supplies);
  • 339115: ophthalmic goods (including separate indexes for plastic, glass, and contact lenses); and
  • 334517: irradiation equipment manufacturing, which distinguishes ionizing equipment from all other types, but does not distinguish medical from nonmedical.

PPI industry codes parallel NAICS industries. Some, if not most, of the products in industries 33912-915 are intermediate products, and hearing aids (included in industry 334510) are not investment goods to the medical industries, although they should be included in medical care inputs.18

The medical equipment information presented in the PPI is quite sparse, particularly in the electromedical equipment area (334510). Aside from irradiation equipment, only two indexes exist for this area, and they are broad aggregates: medical therapeutic equipment and medical diagnostic equipment. Worse, these two cover only a part of the Current Industrial Reports (CIR) first-level aggregations, discussed in the annex, so they cannot be matched with Census Bureau data. As a result of both the scarcity of PPI indexes and the fact that they do not match Census Bureau data on shipments, BEA does not actually deflate medical equipment at the lowest level of PPI detail available, falling back instead to undesirable higher levels of aggregation.

Historically, the more difficult-to-measure commodities have had less PPI coverage, and medical equipment bears out this historical regularity. PPI medical equipment indexes number about the same as indexes for storage and primary batteries, which happen to lie just above medical equipment in the PPI commodity publication structure. Clearly, there is more inherent interest in, and need for analysis of, the medical care sector than for the production of storage batteries.

The PPI sample is drawn on a probability basis. The publication structure for the PPI largely depends on sample size. One solution to inadequate detail in PPI medical equipment indexes is for BLS to increase the sample size for medical equipment, in view of its importance for the analysis of medical care, and to increase the published detail for the PPI indexes in this category.

What should that detail be? Producing PPI indexes that do not match the Census Bureau’s detailed data on shipments of medical equipment, or expenditures on them as investment goods, will bring little advantage. Therefore, expanding PPI detail should accompany and coordinate with the interagency task force suggested in Recommendation 2.4.

Turning now to the PPI indexes themselves, a number of questions can be posed about their present state of development. The first has to do with quality change of goods and services. For many years, economists have known that quality change poses serious price index measurement problems. Among the many relevant references that could be cited are Stigler (1961), Griliches (1964), and the Advisory Commission to Study the Consumer Price Index (1996). Quality change measurement errors are most likely when goods and services experience rapid technological change, and medical equipment has certainly experienced rapid technological change.

The two PPI electromedical equipment indexes are highly aggregated, as noted above. No doubt there are machines in them that have not experienced rapid technological change. Yet the relatively slow rates of decline of these indexes (a little less than 35 percent—total—over the past 20-plus years) seem low for categories that include such machines as scanners and imaging equipment.

A scanner is essentially an imaging device coupled to a computer. Vast strides have been made in imaging technology in recent years, and computers have declined in price at the rate of 20–30 percent per year for more than 50 years (see Triplett, 2003). Moreover, the only study of scanner prices that exists (Trajtenberg, 1989) found that CT scanner prices fell at a rate more like those of computers than the relatively modest (by computer standards) declines recorded in the PPI diagnostic equipment index. More accurate measurement of high-tech medical equipment is an urgent need.

Recommendation 2.5: The Bureau of Labor Statistics, the Bureau of Economic Analysis, and the industrial production division of the Federal Reserve Board should devote resources to addressing the quality change problem in price indexes for medical care equipment, using hedonic methods or other approaches, as the agencies see fit.19 Agencies, including the National Science Foundation, that fund research on medical care issues should consider proposals for improving knowledge about the change in price of medical equipment and on methods for quantifying improvements in their technological capabilities.

2.4.2. Labor

Intuitively, it seems that measuring the labor input—for which the basic unit is labor hours—should be easier than measuring capital. Yet data for the labor input to the medical care account contain serious shortcomings, some of which are not widely understood.

Human capital is a major contributor to output in all industries, but especially in medical care. A medical care account requires incorporating human capital adjustments to its measures of labor input. Otherwise, the account has the potential to seriously understate the amount of resources going into the production of medical care.

Methods for incorporating labor quality, or human capital, have been developed for use in industry accounts (Jorgenson, Gollop, and Fraumeni, 1987), and BLS includes a labor quality adjustment in its published MFP measures. These approaches are promising for the medical care account. Dawson and colleagues (2005) propose a human capital adjustment of this kind for the United Kingdom.

However, the traditional human capital measures distinguish mainly years of schooling or levels of educational attainment. They need to be extended in order to take into account the unique sets of skills and training in modern medicine. Although we do not know how this should be done, it is clearly an area for which research is needed. Indeed, the literature attempting to create human capital measures that portray the skilled labor input into medical care is extremely thin.

Recommendation 2.6: Research should be undertaken in statistical agencies, or funded by agencies interested in medical care topics, on the treatment of human capital in a medical care account. The Bureau of Economic Analysis should include human capital measures in the labor input for its medical care account (even at their present state of development).

In BEA’s present industry accounts, nonmarket labor is not included in the labor input, because time spent in activities that are not paid is not included in GDP. However, volunteer labor in hospitals, hospices, and so forth and time spent caring for ill relatives and friends are important inputs in the production of medical care. In an experimental—or satellite—account, there is no reason to conform to all the conventions of national accounts, even if the medical care account retains the market output boundary of the NIPAs.20

BLS’s American Time Use Survey offers relevant data. A researcher from BEA has already produced a paper on this topic, using these data (Christian, 2010). He found that time spent in home and volunteer production of health-related services was surprisingly low—about 14 percent of the total labor input to medical care. However, estimates generated from the National Health Interview Survey on disability suggest that the value of unpaid time providing long-term care services may be greater than the value of similar market-provided labor (LaPlante et al., 2002). This question needs review and additional research.

Recommendation 2.7: The Bureau of Economic Analysis should develop measures of nonmarket time devoted to the care of others and incorporate them into the labor input in its medical care account.

Even though employment itself may be easy to measure, that does not mean that U.S. measures of industry employment are robust. Triplett and Bosworth (2008), looking at three separate published measures of industry employment— two from BLS and one based on Census Bureau data—show that the three present surprisingly divergent information on industry employment trends. For all industry accounts, including those for medical care, divergences on the order found by Triplett and Bosworth suggest errors in existing labor productivity growth estimates as well as in other studies based on industry accounts.

Recommendation 2.8: The Bureau of Labor Statistics (BLS) and the Census Bureau should investigate and coordinate their surveys of employment by industry and take steps to eliminate or reconcile the substantial discrepancies that now appear in these data. Absent a full reconciliation, the Bureau of Economic Analysis should use the Census Bureau employment series in its medical care account, rather than the BLS employment series, because the census data are compatible with the other inputs in the account and with the output measure, all of which come from Census Bureau information.

The report, Understanding Business Dynamics (National Research Council, 2007), provides a comprehensive set of recommendations for reconciling various business lists and data sources. It also discusses specifically how the Internal Revenue Service code needs revising to permit sharing data across statistical agencies to facilitate data improvements.

Another problem concerns the allocation of professional income between labor income and return to capital. This is an old issue in economics (see, for example, Friedman and Kuznets, 1945, and Christensen, 1969). Tax return data for partnerships and the self-employed, used extensively in the compilation of industry accounts, do not distinguish between the part of a professional’s income that is properly a return to capital, as opposed to labor—what that professional could earn if employed by someone else. Because a large amount of medical services takes place in partnerships and individual practices, this issue looms larger for medical care industries than for some others, although it is indeed a problem for industry accounts generally.

The BLS productivity program has developed methods—applying ideas from Christensen (1969), Jorgenson, Gollop, and Fraumeni (1987), and elsewhere—for separating the capital and labor components. In the BLS application, the labor compensation of the self-employed is estimated from comparable employment in the same industry; the capital return is estimated from corporate entities in the same industry. Because the resulting estimates typically exceed the total income in the noncorporate sector, the labor and capital estimates are then compressed until they just exhaust the income total in the noncorporate sector. The BEA allocation is less robust (for more information, see Triplett and Bosworth, 2004). Going forward, BEA should, for the medical care account, do research to incorporate improved methods for separating labor income from property income.

2.4.3. Energy and Materials

Recent research on productivity and the analysis of production has accounted separately for energy inputs. BEA has already done so in its accounting for intermediate inputs in its industry accounts (Moyer, 2008), and there are no obvious special considerations with respect to the medical care sector’s consumption of energy.

Material inputs to the medical care sector range over a wide variety of commodities, from writing paper and medical examination gowns to pharmaceuticals and stents. Moyer (2008) explains how BEA has recently improved its accounting for intermediate inputs, including purchased materials.

Some intermediate materials inputs to the medical sector present no unique problems. Hospitals consume paper products and cleaning supplies like other businesses. However, much recent technological change in medicine has proceeded through materials inputs that are unique to the medical care sector, including pharmaceuticals and medical devices such as stents. It is not clear how well these intermediates are represented in the current BEA medical account—the problems of coverage, of detail, of the quality of information on interindustry flows, and of matching Census Bureau receipts to PPI indexes that were discussed above in connection with high-tech capital inputs have parallels for high-tech intermediate materials. And again, problems of quality change in the measured deflators are paramount. Some good work on pharmaceuticals can be cited (see the summary in Berndt et al., 2000, and Danzon and Furukawa, 2008), but it is not nearly comprehensive. We are aware of no similar research on medical devices, despite their importance in some recent improvements in medical treatments.

Recommendation 2.9: For economic accounting purposes, work needs to be undertaken on the problems of accurately measuring prices, quantities, and the changing quality of medical devices and pharmaceuticals. Research funding agencies should consider these topics, which should also be on the research agendas of the statistical agencies.

2.4.4. Services

It has long been held that measuring the output of service-sector industries is difficult (Griliches, 1992, 1994). It is no less difficult when the services are inputs.

At the 3-digit NAICS level, purchased medical services include those from other medical industries, where they present problems similar to those of measuring medical outputs (section 2.5). In addition, purchased medical services contribute to the intraindustry or intrasector flows noted earlier. In particular, medical labs, imaging centers, blood banks, and so forth that are now (perhaps inappropriately) in the ambulatory care subsector will be removed from net output via the intervention of the intraindustry flows mechanism used by BLS and recommended above for the medical care account. But they are still inputs to the subsectors concerned, so they still represent measurement tasks. Most of them embody substantial technological elements so they pose similar quality change difficulties as those already discussed for medical capital equipment and for medical devices and pharmaceuticals. As was true of some other inputs, we know of little relevant research.

Successfully measuring the output of scanning and imaging centers (these centers provide imaging services to other medical industries) would substantially improve understanding of the role of imaging machines as capital goods. Looking at the outputs of various imaging machines might even be an alternative to direct investigation of the price movements of those machines. The PPI, however, contains an index for scanning, which does not differ greatly from the relevant index for electromedical equipment (see section 2.4.2, above).

2.4.5. Education

Medical education is treated as an investment in the NHEAs. In the usual framework for industry accounts, this investment in human capabilities is implicitly treated as a stock of human capital, the product of which is incorporated into the labor input. In other words, education augments the input of labor hours. Investment in education itself does not cumulate as a form of capital stock.21

Labor augmentation may not be the only way to think about medical education. For example, like most other forms of higher education, medical schools also produce research, and the production of research is integral to the training function. Because there is uncertainty about how to quantify any economies of scope in medical education, we support retaining for now the traditional treatment (medical education enhances the labor input in the production function for medical services). But we would also encourage imaginative research on alternatives.

2.4.6. Research and Development

Medical R&D poses questions that are similar to the discussion of medical education. Adding R&D to investment in national accounts has been proposed, and BEA has produced a satellite account that capitalizes R&D in the economy as a whole. If R&D is counted among capital assets, then the capital services provided by R&D would go into industry accounts as an input. Many issues surround methods for capitalizing R&D and for estimating capital services provided by R&D—and the literature on the topic is immense. The relevant materials for national accounts are summarized in Fraumeni and Okubo (2005).

Relative to other industries, medical R&D poses no unique issues. However, its treatment in a medical care account needs to be carefully thought through. Consider, for example, R&D that leads to development of an improved drug. If the effectiveness of the improved drug is accounted for in the medical care account (in which the drug is an input), as it should be, then also including the R&D that led to the improvement would double count. This is a kind of “stage of process” problem: pharmaceuticals are inputs to the medical care sector, so R&D on pharmaceuticals belongs in the industry account for pharmaceuticals, not in the industry account for medical care. R&D in the medical care account should include only R&D that is specific to medical care. In practice, however, it may be difficult to make the appropriate distinction.

2.4.7. Summary Comment on Inputs to the Medical Care Account

In this section, we have noted formidable difficulties in measuring inputs to the medical care sector and have suggested data improvements to surmount them. Most of the problems we have identified imply that input growth is understated in medical care industries. For example, human capital in medicine is growing more rapidly than labor hours, and quality-adjusted scanner investment is likely growing considerably more rapidly than the rate estimated using existing PPI deflators. Correcting those biases—because the corrections would raise the overall growth of inputs—would make the already notoriously negative MFP growth in medical care even more negative.

With all the changes in medical practice in recent years, we believe that negative MFP growth in medical care is implausible. This, then, leads to the presumption that output growth must also be understated, and by more than the understatement of input growth. The remainder of this chapter, and most of the following two, concerns methods for measuring the output of medical care.

2.5. MEASURING OUTPUT IN THE MEDICAL CARE ACCOUNT

2.5.1. The Concept

Many health economists agree that the unit of output for measuring medical care is an episode of treatment for a specific disease or condition. Authors who have proposed, developed, or supported the disease-based approach include Scitovsky (1967); Newhouse (1992); Berndt, Busch, and Frank (1998); Cutler et al. (1998); Berndt et al. (2000); Cutler and Berndt (2001); Shapiro, Shapiro, and Wilcox (2001); Triplett (2001); Atkinson (2005); and Dawson et al. (2005).

The consensus has extended to statistical agencies: U.S. PPI hospital price indexes, beginning in 1992 (Catron and Murphy, 1996), the Eurostat manual for measuring services in national accounts (Eurostat, 2001), new German output measures for medical care (Organisation for Economic Co-operation and Development, 2009), and the Organisation for Economic Co-operation and Development manual on measuring education and medical care output (Schreyer, 2009) all accept the view that the “product” of medical care is the treatment of a disease. Moreover, the new North American Product Classification System (NAPCS) specifies the products of medical care industries by a disease classification system (basically, the International Classification of Diseases [ICD]).

Griliches (1992) remarked that controversies arise in measuring services outputs because “it is not exactly clear what is being transacted, what is the output, and what services correspond to the payments made to their providers.” Fifteen years ago, the concept of output for medical care was in dispute, and, in the absence of research on medical care employing a coherent output concept, this report would have been forced to justify at length the proposal to treat medical care on a cost-of-disease basis.22 Now, the holdouts against this approach are few.

The rationale for the disease-based approach can be derived from the consumer side, from the production function for health, or from the point of view of the producer and of the transaction. First, consumers seek medical care to improve their health. The medical care system pursues this through treatment of ailments and diseases. Doctors’ appointments, hospital patient days, drugs— things that have typically been treated as outputs in past measurements of medical care—are, from a consumer perspective, more accurately viewed as inputs used in the production of treatments or, perhaps better, as intermediate stages to the ultimate goal of obtaining a treatment. Obviously, medical care is only one determinant of health, but that does not change the fact that consumers seek better health when deciding to visit a health care establishment.

A second approach starts from equation 2.2. The medical care system’s contribution to health is

contribution to health=(health)/(medical care),
2.8

if other variables in equation 2.2 are constant. Thus, the output of the medical care sector is its incremental contribution to health when a medical treatment or other procedure, including preventive interventions, is undertaken.

The only effective way to measure a medical intervention is through a disease-based metric. Most interventions are disease-specific (although a few might directly affect overall health). The effect of an intervention, its contribution to health, also manifests itself on a disease-specific basis. For example, one might want to determine the effect on health from treatment of a heart condition. The effect comes from either reduction in mortality from heart disease or reduction of disabling effects of the heart condition. Both the treatment and the effect (measured by QALE, for example) are disease-specific.

It was once argued that, from the producer side, a hospital or a doctor does not produce a treatment. Rather, a doctor “sells” an office visit, and a hospital “sells” a day in the hospital, because those were the units in which the patient’s bill was rendered (Gilbert, 1961). A treatment is not (or was not) the transaction unit.

This view of the economics of medical care has less applicability today. Much of the U.S. compensation system for medical care is now set up along treatment lines: the Diagnostic Related Groups (DRG) system used by Medicare for hospital payments can be linked to the ICD at the chapter level. DRG systems, which were first developed in the United States, have become international. The Australian DRG system was derived from the U.S. one, and several European countries have developed similar DRG systems, Germany being a recent convert to the system.23 Accordingly, economic data on a disease-based system, collected from providers, are increasingly available internationally, which adds to the practicality of measuring health care output on a disease-based metric.

The analytic advantages of a cost-of-disease system are its ultimate justification. Suppose, for example, that a heart attack patient is moved to a nursing facility for part of the recovery period in order to save higher hospital per-day costs; if so, hospital and total costs may fall, but nursing care costs rise. The traditional accounting (in the NHEAs) arrays spending by type of institution that receives the funds, so health policy analysts might see nursing home costs rising and conclude incorrectly that nursing care costs are a source of rising health care costs. Actually, rising nursing home costs may indicate cost saving, not cost increase.

In a cost-of-disease framework, the focus is on the total costs of circulatory disease in which all costs from whatever institution that receives payment are aggregated along disease lines. The total cost of treating a disease, not (or not mainly) how the cost is distributed among the industries/institutions, provides the appropriate focus.

As well, much other medical information, which an analyst would use in conjunction with medical care data and health accounts, is collected, tabulated, and organized on a similar principle. Scientific advances also fit into the system: research on new treatments takes place at the level of a specific disease, so it can be fitted naturally into the ICD. This is important for developing methods to deal with treatment improvements in output measures for medical care (discussed in Chapter 4). Thus, using a disease-based classification system such as the ICD for measuring medical care output makes the economic classifications line up with both the classifications that are used for other scientific work and the classification used for payments. This is a great advantage.

Cost of disease is the appropriate focus for consumer, insurance payer, and government policy perspectives. Essentially, all these are buyer perspectives, so the costs of the whole episode are what matter. Moreover, the analytic focus is on whether costs of, for example, circulatory or digestive diseases are growing more rapidly, not on the rate of increase in hospital costs compared with, say, nursing home costs.

2.5.2. From Concept to Practice

Cost-of-disease estimates require pricing the full series of elements in an episode of illness. The Organisation for Economic Co-operation and Development (2001) defines health care output as “the number of complete treatments with specified bundles of characteristics” that “capture quality change and new products.” A complete treatment is a “treatment pathway,” in the language that has come into use, a measure that would encompass all contacts of a patient with the medical system, be they hospitals, doctors’ or specialists’ offices, clinics or other facilities. The Organisation for Economic Co-operation and Development’s definition is consistent with much other recent writing of economists who have given attention to the subject. For example, Atkinson (2005, p. 113) writes: “Ideally, we should look at the whole course of treatment for an illness.”

Going to a disease-based metric poses many practical difficulties, which we discuss in this and the following two chapters. To progress, major problems must be dealt with—arbitrariness around the definition of an episode, the issues of joint costs when multiple conditions are present (comorbidity) and treated (joint production—the situations where a single intervention improves more than one health condition), dealing with chronic episodes, and medical treatments (e.g., preventive) that do not fall neatly into disease categories. But these frequently voiced reservations need to be put into context: cost-of-disease accounts have been produced for many years, since Rice (1966). The challenge is not “Can it be done?” The challenge is to improve on what has been done in the past.

Modeling the Cost of Treating an Episode of Illness

The treatment pathway model takes its rationale from the patient’s side. If it were also implemented with data from the patient, one would collect all the contacts, and all the bills, that are associated with one episode of illness—for example, a heart attack. One might collect all the claims paid by the patient’s insurance and all the out-of-pocket expenses for medical care, plus any expenses that were incurred for which the provider was not compensated. Although sampling and data collection issues arise, the task seems straightforward.

The concept of a patient pathway inherently suggests data be collected from patients, as in the Medical Expenditure Panel Survey (MEPS), for example, or by aggregating insurance claims for a particular patient. However, an aggregation of patient pathways can also be implemented from the provider side.

In the 2007 Economic Census, hospitals (NAICS 622), offices of physicians (NAICS 6211–6213), outpatient care facilities (NAICS 6214), and diagnostic laboratories (NAICS 6215) were asked to provide receipts classified by ICD chapters (e.g., circulatory system diseases, respiratory system diseases, digestive system diseases). PPI indexes for hospitals have been constructed for many years using the same classification system, and BLS has proposed new PPI indexes for doctors’ offices and medical labs that would mesh with the new Census Bureau data collections. These data are too new to have been tested and evaluated. However, they provide an option for constructing disease-based health care output measures that have not heretofore been available in the United States, nor in most other countries. With accurate measures for the different institutional units (subsectors) of medical care, the task is then to assemble these separate parts in a way that is motivated by the treatment pathway model.

Certainly, complications arise. For some illnesses, part of the cost is incurred in the hospital, part from fees paid to the doctor’s office, part from laboratory charges, and part from nursing homes or other rehabilitation units. One might successfully collect from the hospital all the costs associated with its treatment of the patient, but the hospital would not usually know about services provided by other parts of the medical system or their costs. In principle, one could also collect costs from the patient’s doctor and from labs and so forth that provided data on the patient’s case. However, linking this information with hospital data is seldom easy in the absence of some system that provides patient linkages across providers (an insurer, for example, might have such data).

Two measurement problems arise for data collected from care providers: first, data must be linked across providers when only parts of the costs of an illness arise within any one provider. We contend this is not so serious a problem, despite what has sometimes been suggested. Second, treatments with the same outcome sometimes shift between providers in a way that lowers cost but that is not measured in traditional price index collections that draw repeated prices from the same seller. The latter problem, which is a serious one, is considered in Chapter 4 because (in the U.S. payments system) it is a price measurement problem.

Applying the Episode-of-Disease Model to Provider Data

The episode-of-disease model implies a patient’s (or a purchaser’s) perspective. To implement the model, it is natural to think of collecting data from patients or from insurance claims, from which one can in principle collect all the cost information at the same time or link costs from different sources via patient identifiers. Once the full cost of a disease has been collected at the patient (or claims) level, aggregate data on costs of diseases can be generated by aggregating individual cost-of-disease estimates over individuals.

The episode-of-disease model can also be applied to the provider side. Collecting cost-of-disease data from hospitals, doctors’ offices, freestanding clinics, and so forth means, in effect, that the aggregation over patients has already been done at the provider (industry) level. Summing the cost data across industries then produces an aggregate cost-of-disease estimate that is, in principle, the same as that produced via patient or claims data.

To demonstrate, let cij be the cost of an episode of treatment for a specified disease for patient i at institution j (a hospital, say). Then ∑jcij is the total cost of the disease episode for patient i, aggregated across all the jth providers, as reported from individual or claims data. The aggregate cost of that disease is, aggregating across individuals, ∑i (∑jcij).

For data collected from providers (via the Census Bureau–BLS mode), ∑icij is the cost of disease collected from provider j (cumulating the costs of all the ith individuals who have interfaced with provider j).24 Then, the aggregate cost of that disease is, aggregating across providers, ∑j (∑icij), which is the same aggregate as the one obtained from individual claims data.

It has sometimes been said that collecting from providers (as in the new Economic Census health industry data and the PPI price indexes) is deficient because the statistical agency does not collect the full spectrum of a patient’s disease episode from any one provider. This is a misconception: aggregating cost-of-illness expenditures across providers gives the same aggregate cost-of-illness estimates, in principle, as aggregating cost-of-illness expenditures across patients.

Trade-Offs: Establishment Compared with Household Collections

As noted above, two choices exist for collecting cost-of-illness data. The data can be collected from individuals or insurers—from claims records, say. Alternatively, they can be collected from establishments that provide the services (the collection framework for Census Bureau collections of health care sector data and for BLS in the PPI).

Claims data offer much larger samples and facilitate linking costs for a disease episode across providers. Provider data offer the advantage of more precise control over the characteristics of the price and possibly greater accuracy of the totals.

Because cost-of-illness estimates are not fully developed, especially in time-series form, relatively little empirical information exists to compare the two alternatives. Thinking about the matter can benefit from considering nonmedical areas in which household and business (usually establishment) collections exist.

As one example, every month BLS publishes employment estimates from two surveys. One is a household survey, the Current Population Survey; the other is a survey of establishments, now called the Current Employment Survey (known informally as the 790 survey). The survey results are in fact reported in the same monthly BLS press release.

It is well known that the two surveys of employment yield differing estimates for monthly changes and also in some cases over longer time periods. The many “reconciliation” studies done over the years only partly account for the differences.25 Although the dominant professional opinion puts more credence in employment estimates from the establishment survey, it is very hard to confirm the empirical basis for this belief. Household and establishment surveys of employment have differing and conflicting statistical properties, and those statistical properties and biases produce differences in employment estimates that have been hard to account for.

As a second example, two estimates of national consumption also exist. BLS publishes the Consumer Expenditure Survey, which reports expenditures on consumption obtained from a household survey. BEA publishes personal consumption expenditures, a measure that is compiled mainly from business reports, such as retail trade. Again, an extensive literature exists (Triplett, 1997; National Research Council, 2002; Garner et al., 2006), and again the many attempted reconciliations have been only partly successful in explaining the differences. As with the employment example, professional opinion favors data from the business source, but in this case as well, the evidential basis for that belief is not so convincing as is the evidence that the two consumption data sources give different estimates.

Once medical economists have the luxury of comparing estimates from personal interview surveys, like MEPS, with business collections, such as the imminent Census-BLS medical care data, we anticipate that differences will provoke another reconciliation literature comparable to those that exist for employment and for consumption.26 The pattern of those other cases suggests that personal interview surveys and other household side surveys will have great value, but that their advantages over data from business samples will not be overwhelming. At the present underdeveloped state of medical data, one cannot decide whether household or interview or claims-type medical data necessarily dominate medical cost data collected from establishments.

However, if the data collection also has the objective of generating micro-data on patients, collecting information from the patient side (or from insurance company claims) is the only option. Microdata are useful for many research purposes.

Recommendation 2.10: The Bureau of Economic Analysis should construct cost-of-illness estimates for its medical care account and should consider the advantages and disadvantages of provider-side and patient-side (or claims) data for this purpose.

More detailed recommendations pertaining to this task appear in Chapter 3.

2.5.3. Measuring Spending on Nondisease-Specific Health Care Goods and Services

The treatment of diseases represents a large part of the activity of the health care sector. Roehrig and colleagues (2009) produced estimates of national health spending by medical condition using 260 categories defined in the Agency for Healthcare Research and Quality Clinical Classification Software, which groups the numerous ICD-9 codes into broader categories that are “clinically meaningful.”27 In so doing, they provide support for the idea that accounting for major diseases covers a substantial portion of medical care activities in terms of costs.

The research team reallocated NHEAs totals using data from MEPS and a crosswalk methodology developed by Thorpe, Florence, and Joski (2004). While MEPS data are known to have some shortcomings (e.g., undercount of high-cost cases, which introduces a downward bias for some conditions), the estimates provide a general idea of the distribution of expenditures using one among many possible grouping structures. The circulatory system diagnostic category, which includes heart conditions (coronary heart disease, congestive heart failure, and dysrhythmias), as well as hypertension, cerebrovascular disease, and hyperlipidemia, accounted for the highest portion of costs, 17 percent of personal health spending in the United States in 2005. The next largest seven categories—ranging from mental disorders (9 percent) to nervous system disorders (6 percent)—accounted for about half of total expenditures. By contrast, prevention, exams, and dental— which would not fit cleanly into disease-specific categories—accounted for around 6 percent of personal health spending; another 6 percent of total personal health expenditures from the NHEAs was unallocated.

The Roehrig et al. estimates are in same ballpark as those produced in earlier studies. For example, Hodgson and Cohen (1999) found that the “big five” disease areas—(1) mental, (2) nervous system, (3) circulatory, (4) respiratory, and (5) digestive—accounted for about two-thirds of personal medical care expenditures.

These findings notwithstanding, medical care is clearly not limited to treatment of specific diseases. Long-term and preventive care are examples of services that do not always fall neatly into a disease classification. Some patients in long-term care are frail in many ways: their entry into a nursing home may be triggered by a disease—for example, disability due to a stroke—but their medical conditions are not tied to a single disease or health condition.

In addition, nursing homes provide rehabilitation and convalescent care in many episodes of disease. Ideally, one would divide nursing home days between episodes that are disease-specific and those that are not. Accordingly, the output of nursing homes is perhaps best measured as a day in a nursing home, with the days classified by level-of-care codes (such as those used for Medicare and Medicaid compensation). Because of the difficulty of otherwise separating nursing home and postdisease treatment (i.e., hospital “rehab” units) into the disease classification system, it may not be desirable to shoehorn all medical care spending (and associated health effects) into a disease-based classification system.

The same reservation applies to some instances of medical management— screening, diagnosis, and prevention. Although some are directly linked to a disease (mammography is clearly a preventive cost associated with cancer), many others are not. Yet quantities and prices of goods and services in these nondisease-specific categories of medical care should be tracked over time. When lab costs appear as inputs to the other medical sectors, they usually do so as costs of a specific ailment; but when lab output is a final product (tests done on behalf of the patient, for example), it is best measured in conventional ways. Prices and quantities of medical labs, in other words, are measured in terms of the tests the labs perform.

As well, doctor visits and many medical tests are sought for reassurance that a medical condition does not exist. Those contacts with the medical system contribute to well-being. Some are probably reported in claims records and so forth under the disease that is found not to be present, and there is a heading for “symptoms, signs, and ill-defined conditions” in the ICD.

Recommendation 2.11: Although starting with medical care on a disease-by-disease basis is a realistic way to proceed in order to begin accounting for a very significant share of the medical care economy, work should also begin on estimating the costs of, and eventually the health return from, interventions other than treating specific diseases (e.g., management, preventive, diagnostic, screening) and long-term medical services.

2.5.4. A Treatment Index or an Outcomes Index?

Equation 2.8 suggests that medical care interventions are valued by their incremental contributions to health—that is, the output of each intervention is its medical outcome measure. If so, why not measure medical outcomes directly, disease-by-disease, and combine them into a weighted measure, rather than forming measures of treatments? That is, why not ignore the intervention entirely and look only at its effect on health? Indeed, Dawson and colleagues (2005) proposed exactly that. Their preferred basic measure of the output of the National Health Service is a weighted index of quality-adjusted life years (QALYs), grouped by disease classifications, in which one QALY is valued at £30,000 (see their equations 12 and 111).28

Moreover, even a treatment-based system requires medical outcome measures. As spelled out in Chapter 4, medical outcomes are needed to adjust the output measure for improvements in treatments. The issue, then, is whether a treatment index should be constructed that is adjusted by medical outcomes, or whether an output index should be constructed that is composed entirely of medical outcomes, QALY or QALE (see Chapter 5), without recourse to counting or valuing treatments.

One reason for preferring the medical outcome measure is the general knowledge that not all treatments are effective. Errors and mistakes, misprescriptions and misdiagnoses (patients still receive antibiotics for viral infections, for example), botched operations, and variance across areas in modes of treatment29 are well known. Some interventions do not make a positive incremental contribution to health. For these cases, bypassing the treatment measure would bring the output measure closer to one that truly measures the incremental contribution to health that the medical care system makes.

Similar phenomena occur in other parts of the economy and are not adjusted out of national accounts. Botched and inappropriate car repairs, for example, occur with considerable frequency; sometimes they are corrected by the original repairer so the corrections do not result in new output, but sometimes the customer seeks out a new shop, so that repairing the botched job actually increases GDP. Such “redos” are not subtracted from GDP, even though they hardly contribute to consumers’ welfare, nor is GDP adjusted for defective manufactured products that are also not infrequently produced.

But parallelism does not necessarily lead to good measurement practices. Methods for measuring medical output need to be considered on their own merits, apart from other national accounts practices, especially if the medical care account will be some form of alternative or satellite account, as seems likely.

Nevertheless, the distinction between output and welfare has a bearing on measurement principles. Particularly when there is to be a health account, in addition to a medical care account, adjusting medical care output is not the only way to handle defective and inappropriate treatments. Whether appropriate or not, treatments are still produced in the medical care sector, and they still use resources in the medical care sector. By that standard (the conventional way of looking at output), they are outputs of the medical care sector.

Determining whether or not medical sector output arising from inappropriate treatments contributes to welfare is a task for the compilers of the health account, particularly since they are more likely than national accountants to have the expertise to determine when treatments are not effective. When George Washington was bled as a treatment for pneumonia, his doctors must have thought they were contributing to his health, and the national accounts of his day, had they existed, would have recorded a treatment (or its resource use) in national output. When medical knowledge advanced enough to understand that Washington’s treatment hastened his death, was the accounting revision that the advance in medical knowledge demanded best put into the national accounts measure of medical output or into the health account? The most important thing, surely, is that the revision be made. But for both consistency and expertise reasons, it seems better to make the revision in the health account—that is, national output for 1799 remains unrevised by the new scientific knowledge, but the estimate of national welfare is revised downward.

Consider also the role of medical care output in the account for health, for which it is an input; the output of the health account is health. One never wants to measure an input by its output (nor an output by an input). It must in principle be possible that the output effect of a change in input quantity differs from the input change. If the output of the medical care sector were measured as a health outcome, and that measure then used as an input in the health account, the possibility of productivity change in the health account is largely eliminated by convention. One of the things that a health account should be designed to reveal is the productivity of the medical care sector in the production of health. To estimate that, the measure of medical sector output must not be identical to the output of the health sector.30

A third reason is also compelling: a health care output measure that is based on disease treatments (with medical outcome measures as quality adjustments) is grounded on a more precise statistic than an output measure that is based entirely on medical outcome measures. Our ability to measure health care output by treatments, that is, by a disease classification system, may not be that far along, as we emphasize in this report. Nevertheless, information on expenditure by disease, on numbers of treatments by disease, and even on health care prices by disease is further developed than are medical outcome measures. Treatment information is inherently more concrete and therefore more precisely measured information. For a medical care account to attain public confidence, it needs to be seen as transparent, at least in relation to other comparable economic measurements. Measuring medical care output by treatments is not that different from the way car repair is measured in national accounts (Triplett, 2001) and can readily be understood within the usual framework of economic statistics. In contrast, even health economics professionals raise difficulties, both conceptual and practical, with existing medical outcomes measures (Meltzer, 2001). A sound measurement principle is to minimize the use of undeveloped and potentially controversial measures, using them only when they are necessary and not when more straight-forward alternatives exist.

We are not minimizing the potential contributions of such medical outcome measures as QALY and QALE. Indeed, we believe that they should be developed as rapidly as possible (see Chapters 5 and 6). We advocate using them in the health account and also in the medical care account. Nonetheless, at this stage in their development, the time is not propitious to rely on medical outcome measures exclusively as the output measure in a medical care account.

ANNEX: PRODUCT DETAIL FOR ELECTROMEDICAL EQUIPMENT DATA

Part I. Problems with Available Data on Medical Equipment

Problems occur in estimates of capital stocks for medical equipment. Capital goods (including computer software, as well as medical equipment) are allocated among using industries by the Bureau of Economic Analysis (BEA) capital flow table, an adjunct to the input-output table that tracks flows from capital goods–producing industries, or imports, to using industries. The 1997 capital flow table shows, for example, that two-thirds of hospitals’ capital investment goes to equipment, which is about the same as for the sector as a whole. Of hospitals’ equipment expenditures, 38 percent are for medical instruments and related equipment and 29 percent are for electromedical equipment (the category that includes scanners). Other hospital investment expenditures are for structures and a range of capital equipment that is also used by other industries—computers, software, and other (nonmedical) electronic products account for a quarter of hospital equipment. Not surprisingly, nursing homes spend less, relatively, for equipment, and they spend their equipment money in different ways.

Data on medical equipment in the capital flow table are quite coarse: the table distinguishes only the two gross aggregates “medical instruments and related equipment” and “electromedical equipment.” One reason is that the table presents flows for all sectors of the economy. Most electromedical equipment, not surprisingly, flows to the medical care industries (according to the capital flow table, education is the second largest using industry). Providing more detail on medical equipment would not suit other industries in the capital flow table, even though more detail would be useful for the analysis of medical care.

The available survey data on medical care investments do not contain that much more detail. Several Census Bureau data sources present different and sometimes conflicting information, but for most of them, the useful detail is not appreciably greater than in BEA’s capital flow table. Worse, the detail present in different surveys does not match up, which greatly diminishes the usefulness of the data.

These data sources—which include the Economic Census, the Annual Surveys of Manufactures (ASM), Current Industrial Reports (CIR), and the Annual Capital Expenditures Survey (ACES)—are not well integrated and can be confusing, so we present a brief summary of their data on medical equipment.

Data for Medical Care Capital Equipment

As explained in the text, the Census Bureau, in the Economic Census, still does not collect the range of input data for services industries that it has long collected for manufacturing and other goods-producing industries. The resulting data incongruity handicaps analysis of services industries, including medical care industries, which require the same kinds of information that has long been provided for goods-producing sectors. The same flaw carries through to the Annual Services Surveys, which also are deficient in input data.31 The Census Bureau has instituted ACES to fill the gap, but this survey has been directed toward obtaining investment data for the economy as a whole, and its usefulness is greatly limited by inadequate industry and commodity detail.

Promising for our purposes is an information and communications technology (ICT) supplement to ACES that distinguishes the category “electromedical and electrotherapeutic equipment.” This ICT category in ACES includes major types of equipment used in the medical care industries and matches the electromedical equipment category in the BEA capital flow table. The ACES survey form collects 4-digit North American Industry Classification System (NAICS) industry codes.

However, the promise of ACES has not been fulfilled. ACES published for 2005 and 2006 only at the 2-digit NAICS level (NAICS 62). No detail is published, only the total for investment in electromedical and electrotherapeutic equipment, plus information by type of acquisition (capitalized, leased, and so forth). ACES thus provides only a very limited benchmark—for one aggregated type of medical equipment, at the level of the medical sector as a whole.

Other relevant Census Bureau surveys collect data on U.S. production of medical equipment, not investment. The most informative, CIR, collects data on U.S. production of the products of NAICS 33451, electromedical apparatus manufacturing, and of some other medical equipment, at considerably more detail than ACES. ASM distinguish as the main product of NAICS 33451 “diagnostic and therapeutic” equipment, presumably the same “electromedical and electro-therapeutic equipment” products that are collected in ACES. No detail beyond this aggregate is published in ASM.32

The 2007 Economic Census form for the industry “Electromedical and Electro-therapeutic Apparatus” gathered information on receipts from “electromedical equipment including diagnostic, therapeutic and patient monitoring equipment.” This is the same level of fairly gross aggregation as in ASM, and the Census Bureau form specifies that it is the same aggregate as on CIR. Thus, the Economic Census, which collects in many industries more detail than in annual collections, in this case does not approach the detail in the CIR.33 The Economic Census also collected data for NAICS 33911; this industry makes nonelectronic medical equipment.

The problem of inadequate capital data is not unique to medical care industries. However, medical care is the largest sector of the economy for which detailed data are not available on purchased inputs, including capital inputs. Moreover, medical care industries purchase a range of very highly technological equipment, which is importantly linked with technical change. Thus, unlike some other services industries in which absence of capital expenditure detail is merely an annoyance (for example, NAICS 812, Personal and Laundry Services), in medical care the data gap threatens understanding of essential aspects of recent developments in the sector.

Critics of the U.S. medical care system have frequently asserted that it over-uses imaging devices. It is accordingly bizarre from a research and policy analysis standpoint that data collections in the Economic Census do not reveal how much imaging equipment is going into the medical care sector, let alone the total stock of it that is in place.

Greater data detail on technological capital goods used by the medical care sector is essential. The model for improving medical equipment data is the data published for computers and office equipment—the second largest category of medical industry equipment investment and another notable category of technological investment products.

Some years ago, government data on computers and related equipment were as seriously undeveloped as medical equipment data are today. A multipronged effort by all three major statistical agencies (BEA, Census Bureau, Bureau of Labor Statistics [BLS]) involved

  • a new and more relevantly descriptive system of product codes;
  • improved and more detailed data on shipments and sales receipts by detailed product; and
  • improved deflators that, using hedonic price index methods, allowed for the rapid rate of technological changes characteristic of nearly all electronic goods.

The value of this extensive data development exercise was shown in the analysis of the substantial influence of information technology investment in the post-1995 U.S. productivity expansion—see, for example, Jorgenson (2001) and Jorgenson and Stiroh (2000). Without the development of a comprehensive data set on the production of—and investment in—computer and related equipment, analysis of the role of information technology in the U.S. economy would have been, if not impossible, certainly greatly handicapped.34

Medical equipment performs a similar role in sparking, facilitating, and implementing technological innovation, except it does it exclusively in the medical care industries and not economy wide. Much anecdotal information exists about medical equipment investment, but it is not quantified in the way data on other ICT equipment are for purposes of economic analysis. The lack of good information on medical equipment is one more way in which data for the analysis of medical care suffer from long-term neglect.

The first step in a data improvement project such as the one needed for medical equipment is getting agreement among the agencies on a common product classification scheme. This seemingly mundane task is necessary because otherwise data, especially from BLS and the Census Bureau but also from different Census Bureau surveys, do not fit together, and data expansions by individual programs and agencies proceed in inconsistent directions. Moreover, there is no center in the U.S. statistical system for coordinating such matters; it usually takes a special task force composed of agency representatives.35 We explore ways for improving a specific type of capital equipment in greater detail below.

Part II. Product Detail for Electromedical Equipment Data

In this part, we discuss product detail for electromedical equipment and its inadequacy for producing consistent and meaningful data. Similar reviews could be carried out for the other categories of medical equipment and indeed for investment in medical structures, which have their own unique problems. The example we provide illustrates principles that we think should be followed.

In most goods-producing industries, the Census Bureau 10-digit commodity codes provide the standard for product nomenclature. In the case of electro-medical equipment (primarily, NAICS 33451), the list is (the last four digits only are shown):

  • 1100 electromedical equipment, including diagnostic, therapeutic, and patient monitoring equipment;
  • 1103 magnetic resonance imaging equipment (MRI);
  • 1106 ultrasound scanning devices;
  • 1109 electrocardiograp;
  • 1112 electroencephalograph and electromyograph;
  • 1115 audiological equipment;
  • 1118 endoscopic equipment;
  • 1121 respiratory analysis equipment;
  • 1124 all other medical diagnostic equipment; and
  • 3100 electronic hearing aids.

The categories 1103–1124 are subdivisions of the first one. Hence, no distinction is made between uses. For example, ultrasound diagnostic equipment and ultrasound therapy equipment are in 1106.

No one seems to use the Census Bureau 10-digit list for this industry. For example, the Census Bureau’s CIR imposes use-categories as its first disaggregation:

  • medical diagnostic equipment;
  • patient monitoring equipment;
  • medical therapy equipment;
  • surgical systems; and
  • other electromedical and electrotherapeutic apparatus.

CIR’s second disaggregation is by product. But even though CIR presents much more product detail than the ASM or the Economic Census, the CIR published product detail does not map exactly into the Census Bureau 10-digit product list. For example, where do defibrillators go in the 10-digit list? They are not diagnostic, so there is not even a place for them in the “all other” grouping.

The classification used by BEA (in its investment series) more or less follows CIR’s first-level disaggregation, even though BEA does not use CIR data for medical equipment investment. BEA’s preferred classification scheme is also inconsistent with the Census Bureau 10-digit product codes.

Product codes in the PPI industry classification are also broadly consistent with CIR’s first-level disaggregation, although a separate PPI commodity code scheme disaggregates differently. In addition to measures for the industry aggregate (electromedical apparatus), both the PPI and CIR contain data at the first-level disaggregation (that is, “diagnostic equipment” and so forth). But “medical diagnostic equipment” is still far too broad: CIR is right that meaningful disaggregation would produce series such as “ultrasound scanning devices” and “EKG.”

The BEA and PPI codes suggest an incipient interagency agreement on the CIR first-level disaggregation, except for possibly the Economic Census and ASM. However, even incipient agreement between PPI and CIR does not yield detailed data on medical investment because BEA does not use the PPI for deflation at this level, and indeed it does not use CIR for any of its investment estimates. CIR contains much product detail, but the PPI is insufficiently fleshed out to match it.36

The CIR contains more detail, and more meaningful categories, than do any of the other surveys, including the PPI. At the product level, however, the CIR is problematic. For one thing, the size of CIR product categories labeled “all other” equipment makes the CIR categories less informative than they ought to be (see Table 2A-1). To take the worst case, the largest—by far—category of patient monitoring equipment is “all other patient monitoring” equipment; it accounts for 86 percent of the total. “All other” miscellaneous categories are not useful ones, and when they are large and growing they tend to hide the most vigorous technologies where they cannot be observed.

TABLE 2A-1. “All Other” as a Proportion of Shipments in Electromedical Equipment Categories, Current Industrial Reports.

TABLE 2A-1

“All Other” as a Proportion of Shipments in Electromedical Equipment Categories, Current Industrial Reports.

In some cases, the “all other” category may be large because of disclosure problems. For example, the number of producers of MRI machines is small, so if there were a line for MRI equipment in the CIR medical diagnostic equipment category, it could not be published, to avoid disclosure of individual producer information. However, it is hard to believe that the disclosure problem applies to each of the “all other” equipment categories.

Moreover, in addition to the “all other” categories within the major types of electromedical equipment (e.g., “all other patient-monitoring equipment” as part of “patient-monitoring equipment”), CIR electromedical equipment contains a whole first-level category labeled “other electromedical and electrotherapeutic apparatus.” Accordingly, almost half of the total shipments of electromedical equipment falls into “all other” classifications.

Substantial government funding and a substantial amount of respondent burden are costs of the CIR medical equipment survey. Because the administration of this program has not optimized the value of the survey, it is not producing sufficient information to justify those costs. These are old problems that urgently need attention.37

We believe that careful study by an interagency team would produce a more detailed, workable product classification scheme that could be implemented by CIR, the Economic Census, and the PPI. Disclosure difficulties are likely to arise in implementing a detailed classification scheme in production data. However, obtaining information on investment purchases of MRI and other technological equipment by hospitals and other medical sector units presents no disclosure possibilities and so obviates the difficulty in collecting information from domestic producers. In medical care analysis, the investment data—that is to say, information from the buyers—are more crucial than domestic production data (information from the sellers), although both are valuable.

Footnotes

1

Portions of this chapter are drawn from Triplett (2011).

2

The BEA website provides a wealth of information on the methodologies, content, and scope of the NIPA. For example, http://bea​.gov/national/pdf/NIPA_primer​.pdf provides a good introduction to the accounts.

3

“Burden-of-disease” studies typically do not include these costs.

4

However, see the caveat in section 2.4.7.

5

It can also be thought of as the rate of growth in the production function f(•), or its time derivative.

6

The form of the function implies an index number formula; the Tornqvist index and the Fisher index have valuable theoretical properties (Caves, Christensen, and Diewert, 1982).

7

Sometimes it is called total factor productivity (TFP). TFP and MFP are synonyms. The term MFP was introduced in a report by the Panel to Review Productivity Statistics (National Research Council, 1979) to avoid the implication that equations such as (2.3) have necessarily enumerated all the inputs—an alternative interpretation of productivity change is that it reflects inputs that have not been accounted for fully.

8

Harper et al. (2008) also report negative MFP growth in medical care.

9

Triplett and Bosworth (2004, p. 263) reported that LP in the medical care services sector had negative growth from 1987 to 1995, but after 1995 it turned positive. They attributed the sign change to data improvements in the measure of hospital output in the second period, which demonstrates how measurement issues impact the analysis of medical care.

10

Michaud et al. (2009) and Lakdawalla, Goldman, and Shang (2005) provide empirical evidence that as much as 30 percent of the growth in spending on medical care in the United States can be linked to increased rates of obesity in the population.

11

On this point see also Philipson and Posner (2008).

12

Ideal aggregation of producing units demands identical homogeneous production functions across the units—the standard reference is Fisher (1993). In practice, aggregations into industries are chosen in the NAICS

13

However, we note that if the output of custodial activities were removed from medical care as an input to the production of health, then the partial productivity measure discussed in equation 2.6, above, cannot be computed when industry data are used in the analysis, even though the MFP measure in equation 2.4 is defined. The reason is that medical care output, however defined by the researcher, enters into equation 2.4, but the resources used in medical care enter into equation 2.6, and if the resources are measured with industry data, the difficulties of separating inputs, discussed in the text, will apply.

14

On determining shares, see Yuskavage (1996), who discusses difficulties in estimating the property income share and the intermediate input share.

15

A long history of debate over measurement concepts of capital in production analysis has spilled over into national accounts measurement. The debate is now settled on the lines suggested in Jorgenson, Gollop, and Fraumeni (1987). See, in particular, the Organisation for Economic Co-operation and Development handbook on productivity measurement by Schreyer (2001), which records the consensus and makes recommendations for the measurement of capital stocks and capital services that are wholly consistent with the methods now generally accepted in production analysis. From a conceptual view, medical care presents no unique problems in the measurement of capital.

16

For a detailed discussion, see Sensenig and Donahoe (2006).

17

However, the data on output of medical equipment are not structured very well, a point we discuss in more detail below. In addition, none of the 12 input classes relates to purchased services, an input class that has been growing rapidly in all parts of the economy.

18

In addition, in the PPI product code structure (a different classification system from the industry structure), one finds PPI product code 11709-06, x-ray and electromedical equipment, which includes the following:

  • 11790-0512 irradiation equipment;
  • 11790-0514 diagnostic electromedical equipment;
  • 11790-0516 electrotherapeutic equipment;
  • 11790-0517 other electromedical equipment, excluding diagnostic and therapeutic; and
  • 11790-0524 parts and accessories.

Indexes by product code do not always agree exactly with product code indexes from the industry code PPIs, for reasons that need not be explored here. We add this note because the users may find these indexes confusing.

19

The IPI contains a category of computer and electronic equipment called “navigational, measuring, electromedical and control.” Improvements to the electromedical measures might justify participation of the FRB in the efforts to improve data on medical equipment, which we would greatly encourage.

20

BEA defines satellite accounts as follows (Bureau of Economic Analysis, 1994, p. 41):

[S]atellite accounts are frameworks designed to expand the analytical capacity of the economic accounts without overburdening them with detail or interfering with their general purpose orientation. Satellite accounts, which are meant to supplement, rather than replace, the existing accounts, organize information in an internally consistent way that suits the particular analytical focus at hand, while maintaining links to the existing accounts. In their most flexible application, they may use definitions and classifications that differ from those in the existing accounts. . . . In addition, satellite accounts typically add detail or other information, including nonmonetary information, about a particular aspect of the economy.

The United Nations System of National Accounts offers a similar description (United Nations, 1993b, pp. 45, 489):

Satellite accounts provide a framework linked to the central accounts and which enables attention to be focused on a certain field or aspect of economic and social life in the context of national accounts; common examples are satellite accounts for the environment, or tourism, or unpaid household work. . . . Satellite accounts or systems generally stress the need to expand the analytical capacity of national accounting for selected areas of social concern in a flexible manner, without overburdening or disrupting the central system.

21

Another way to put it it that, for any form of capital, it is the services of the stock that enter into industry accounts. The services of the stock of education, being those of human capital, are entered into the industry accounts through labor quality augmentation.

22

Significantly, the only study that touched on health care in the Conference on Research in Income and Wealth services volume was Murray (1992), who reported negative productivity growth in Swedish government-run hospitals. He remarked, though, that hospital output would have been measured better had it used a disease-classified output concept.

23

However, Colecchia and Schreyer (2002) show that, of nine countries that have adopted DRG systems, each differed from the others.

24

The statistical agency will usually present this number as the jth agency’s receipts from treating the disease.

25

See the paper presented to the Federal Economic Statistics Advisory Committee, 2006, at http://www​.bls.gov/bls/fesacp2120905.pdf.

26

There is already at least one published paper on MEPS-NHEA reconciliation—see Sing, Banthir, and Selden (2006). This study does not compare cost-of-illness results because the NHEA contains no information on cost of illness.

27

This research was presented to the panel at its March 14, 2008, workshop.

28

Garber and Phelps (1997) further substantiate the use of a QALY measure as an indicator of health care output.

29

Interarea differences in medical practices may be errors or may be differences of opinion about best practice. But even if the latter is true, presumably more knowledge will eventually show that some treatments that were thought to be best practice in some areas were in fact errors.

30

To avoid confusion, it is not inconsistent to make the quality adjustment for a changed treatment depend on the ratio of medical outcomes for the new and old treatments. This thorny question is addressed in Chapter 4.

31

This old data lacuna in services-producing industries is discussed more fully in Triplett and Bosworth (2004, Chapters 10 and 11).

32

Both CIR and ASM record that production of these products was considerably greater than total U.S. investment in them in 2005–2006, suggesting that net exports were high, but it is known that, for some of these products, foreign producers are important suppliers.

33

The Economic Census also collected information on other products, including irradiation equipment, scientific instruments, nonelectromedical surgical and medical apparatus, catheters, and so forth, that are also made in this industry (in which they are secondary products).

34

The Federal Reserve Board has also contributed more recently to improving the deflators.

35

The industry classification system, NAICS, is an Office of Management and Budget standard, and there is an emerging NAPCS for products in the services sector. But no similar, formal standard exists for goods-sector products.

36

We leave aside any judgments about whether CIR data are equal in quality to ASM or Economic Census data.

37

A previous problem with CIR that prevented publication of consistent aggregates has been corrected, at least at the level of electromedical equipment as a whole (formerly, there were gaps in the published aggregations that were caused by the methods adopted to prevent potential disclosure problems).

Copyright © 2010, National Academy of Sciences.
Bookshelf ID: NBK53339

Views

  • PubReader
  • Print View
  • Cite this Page
  • PDF version of this title (1.3M)
  • Disable Glossary Links

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...