Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 30
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop 3 Price Indexes: Calculating Real Medical Care GDP Although the term “satellite health care account” features in the title of the workshop, much of what was discussed over the course of the day had to do with medical care price deflation—the use of price indexes to estimate real changes in the levels of inputs and outputs for the sector. Matthew Shapiro, who has done seminal work on the topic, began his comments by noting that a big part of the task for the Bureau of Economic Analysis (BEA) involves parsing out nominal expenditures in a way that is meaningful and conducive to measuring prices. BEA is already in the business of developing price indexes for the purpose of calculating real levels of economic activity, on an industry by industry basis, for the national accounts; this responsibility is particularly demanding for the medical sector in which third-party payments, and the fact that transactions do not occur in textbook competitive markets, confound price measurement. Some aspects of this problem have already been dealt with on the nominal side of the accounts by allocating the actual expenditures back to the consumer, to the government, and so on, regardless of who actually pays, which is often an insurance company. Once the nominal flows for the sector have been figured out correctly, which is a daunting problem in itself, the next task in developing the new BEA health care account is to begin estimating the disease-based price indexes. Shapiro endorsed this two-stage strategy, although he noted that, for other purposes, there were different ways—in addition to the disease unit concept—that are also useful health care price measurement. For example, hospitals would want to know about prices specifically for its industry. However, to get the price indexes from the consumer standpoint, disease by disease unit pricing seems more appropriate than a traditional industry-based approach or than a global pricing of population
OCR for page 31
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop health. Even if, ultimately, the goal is to measure the price of an incremental gain in health, Shapiro argued that, for economic accounting purposes, one is driven by the logic of the disease by disease approach, which was the focus of much of the workshop. A key question, raised by Steve Heffler (Centers for Medicare and Medicaid Services), is how the methods for parsing nominal expenditures by disease (described in Chapter 2) relate to appropriate price measurement. Ana Aizcorbe responded that those working on satellite accounts—the Cutler-Rosen group or BEA—first establish a number of disease “buckets” that make sense to the medical community; these buckets become the unit of observation for which spending and health effects data are collected. The dollar total spent per patient on a particular category—for example, diabetes—becomes the price for the newly defined unit of health service. Then, nominal spending on diabetes is given a weight based on its share of total medical care spending.1 Likewise, indexes for each disease category are weighted then aggregated. Aizcorbe cautioned that BEA is still in the phase of attempting to figure out the best way to define diseases, and that developing these kinds of indexes is still a ways off. BEA has purchased some databases covering patients who are commercially insured with the intention of experimenting with different types of indexes and different ways of defining diseases. In this context, Aizcorbe described the most important problem with producer price indexes for purposes of the satellite account envisioned by BEA: they do not identify the medical care good or service that is sought by the consumer—which most think should be the treatment of a particular disease or condition. She added that health economists have developed the conceptual tools that are needed to remedy the situation, and that putting these approaches into practice is something that BEA would be working on right away. Much of the academic literature has relied on patient claims data to provide a picture of price trends for treating specific conditions. The economic good has been defined as a completed episode. For example, for a heart attack patient, this may involve time and expenditures on a series of initial treatments plus those that take place during the recovery period. At the end of that episode, data are collected to estimate all dollars spent over the entire period; this forms the basis for pricing a completed episode. To develop a comprehensive health care account with this kind of underpinning, claims data would be needed for as much of the population as possible; Medicare and Medicaid would provide large portions. However, there are some groups for which claims data will not be available—most obvious are the uninsured, who do not submit claims—so their spending would have to be measured another way. Patients from some kinds of institutions are also not typically 1 In a fully evolved price measurement program, tracking nominal dollars spent on treatments would be viewed as only a first step. As discussed in detail in Section 3.3., a fully meaningful price measure must ultimately also consider how the quality of a treatment changes over time.
OCR for page 32
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop included in these sources, so BEA will be investigating ways of obtaining spending data for them as well. Aizcorbe also noted that the timing aspect of the treatment-based unit does not lend itself seamlessly to deflation in the national accounts. For example, for delivery of a baby born in January, most of the dollars are for services provided in the previous nine months. So, when pricing the completed episode, the reporting takes place in the year following the nominal expenditure outlays. Aizcorbe stated that, ideally, the price index should line up with the time period in which the spending actually occurred. There are other areas of the national income and product accounts (NIPAs) that share this issue (payment, consumption, or return from investment takes place beyond some point); one way to handle it is to think in terms of the price per patient over some predetermined period of time. 3.1. PRICING TREATMENTS TO CAPTURE CHANGING TECHNOLOGIES, INPUT SUBSTITUTION, AND POPULATION HETEROGENEITY A disease-specific index must embed a capability to capture the substitution of medical care treatment inputs that takes place over time. Aizcorbe used the example of treatment for depression, which has transitioned from a high to a low reliance on talk therapy as less expensive alternatives—specifically antidepressant drugs—were introduced and proliferated. Tracking patients with this condition over the past few decades would have revealed some portion switching from talk therapy to drug therapy. As this has occurred, the average amount spent on treatment of depression has fallen. However, standard price indexes do not pick up this change because they track the price of talk therapy and of drug therapy independently, and therefore they do not catch the fact that people are switching from one to the other. Even if there is no innovation in prescription drugs and no price change in either approach to treating depression, the per-patient cost falls because this substitution has occurred. If the standard indexes are used to deflate nominal spending, the resulting measure ends up showing a drop in real spending or a drop in the quantity, when in fact the same number of patients are being treated for depression. For its satellite health care account, BEA proposes to take the system-wide spending over some period of time in a treatment (such as for depression), regardless of treatment mode, and divide it by the number of patients treated. The idea is to calculate a unit value that counts all of the spending and allows for substitution across treatment types for each specific condition. Patricia Danzon, who spoke about the pharmaceutical industry, cited the growing prevalence in the market of biologics—biological products made from living organisms whose uses are similar to conventional drugs—as another example of a switch in technology that BLS will need to confront. At the moment, both a pharmaceutical index and a biotechnology index exist. But, Danzon noted, if the goal is to estimate change in pharmaceutical prices accurately, then the current
OCR for page 33
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop movement toward more biologics and fewer chemical-based drugs needs to be captured. To the extent that they are measured in separate industries, a problem arises because the index will not capture the biologics that are dispensed through physicians’ offices or retail pharmacies. They are likely to be picked up by Medicare as Part B drugs, but probably for many of the other databases, they are just part of physicians’ services. As these biologics become a significant share of total pharmaceutical spending—and they will, especially at the expensive end—it will become increasingly important to make sure that they are correctly allocated to pharmaceuticals (as opposed to physician spending). Jack Triplett described other examples of how spending by traditionally defined medical care industries combine into the price vector describing a specific treatment. Under present BLS procedures for cataracts—a case that was cited several times throughout the day—if the surgeries taking place in a hospital are sampled, then one set of price indexes would be generated for that; if surgeries shift to a clinic, then another set of price indexes would be obtained for that. If people switch from the more expensive hospital surgery to a less expensive clinic surgery, and if quality does not suffer, the ideal price index (from the perspective of the patient) would be capable of capturing the decrease in price. Triplett continued, noting that one reason people have not thought much about substitution across medical care industries is that BEA data are organized at a higher level. BEA industry counts do not go down to the five-digit North American Industry Classification System (NAICS) level of aggregation, and therefore not all of the reallocation effects are readily visible.2 He offered an analogy between the medical industries and the transportation equipment–producing industries to illustrate the industry-sector problem: the historical case of the automobile industry replacing the buggy industry. If the industry were defined as producing road transportation equipment instead of individual cars and buggies (although one might still want to get the prices of those), then in principle these substitution effects could be captured. If only the carriage and automobile industries were tracked, price and productivity measurement would capture only part of the effects that are of interest. When people found that it was cheaper per mile to go by car than by horse and carriage, and they switched from the latter to the former, the full price and productivity effects would not be completely explained by the indexes for either one. However, this problem arises when the interest is in welfare comparisons, rather than just in output comparisons. There is nothing wrong with the auto and buggy measures; rather, it is that aggregating them misses some of the welfare gains to the consumer. Triplett concluded that BEA would need to rework the way the five-digit industries are aggregated into the three-digit industries to do this—it is not only an index num- 2 NAICS uses a six-digit coding system to identify particular industries and their placement in the hierarchical structure of the classification system. The first two digits of the code designate the sector, the third digit designates the subsector, the fourth digit designates the industry group, the fifth digit designates the NAICS industry, and the sixth digit designates the national industry (http://www.census.gov/eos/www/naics/).
OCR for page 34
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop ber problem. The usual way of modeling “substitution” in price index research does not adequately handle shifts of broadly defined products (like road transportation equipment or curing cataracts) between producing industries. Triplett presumed that there are probably lots of instances of something similar happening, simply because a function is going out of one sector and into another one. Steven Landefeld agreed that these transitions probably do occur with some frequency. He noted that, wherever BEA has used quality adjustment methods, the focus has been almost exclusively been on final goods and services. That is, the agency has typically assigned the adjustment into the industry producing the final good. In this case, the goal is to examine how the change in the standard (expenditure) measure for gross domestic product (GDP) resulting from use of a new deflator works through on the industry (input) side. Aizcorbe identified other characteristics of medical care that complicate the calculation of price indexes. For example, insurance plans vary in their payments for a given service, so patients in different plans effectively pay different unit prices. Ignoring bad debt and charity care, the uninsured probably pay the most for treatments and pharmaceuticals. When uninsured individuals turn 65, Medicare Part D comes into play and the drugs that they buy become cheaper. With aging, if the price that patients were paying before was high and the price that they pay once they join Medicare Part D is comparatively low, then the revenues that pharmacies or manufacturers receive for these drugs will fall. In this stylized case, nominal totals fall but the price indexes do not pick that up because they are tracking prices for, say, someone with Blue Cross/Blue Shield coverage and for someone enrolled in Medicare Part D separately. If this price index is used to deflate spending, a drop in quantity would be shown, even if the same population group—a portion of which has shifted from commercial insurance (or no insurance) to Medicare over time—is represented. Aizcorbe suggested that handling this type of heterogeneity for deflation purposes needs to be analogous to the method for handling input substitution for the treatment of diseases. BEA would try to define the price as expenditures on all types of treatments by patients with all types of coverage and divide that by the number of patients. So in the population aging example, as people start spending less on drugs, it would be reflected as a price drop, not as a drop in quantity, which is exactly what is wanted for the national accounts. In summary, the main reasons why BEA feels it needs to construct deflators differently from what is currently being provided by BLS in its Producer Price Index (PPI) program is that they want to be able to think in terms of treatment of a particular disease, not of a specific kind of treatment for the disease. Also, BEA would like the account to be capable of reflecting as a price change—and not as a quantity change—the different prices that patients under different plans pay for treatments as they shift from one plan to another. Finally, BEA would like to control for changing trends in the severity of conditions as well, to the extent that it is possible. The ideal would be to track over time a disease in which the severity of the condition is homogeneous.
OCR for page 35
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop 3.2. BEA’S STRATEGY FOR COORDINATING THE INDUSTRY INPUT ACCOUNTS WITH THE DISEASE TREATMENT–BASED EXPENDITURE CONCEPT Operating in parallel with the expenditure side of the NIPAs, on which the majority of the workshop discussion focused, is the issue about what to do on the industry side of the accounts. Once BEA begins deflating medical care spending by consumers using a new price index, the industry-side calculations must be revisited, as real spending on inputs in the production of medical care must equal real spending on final medical care goods and services. If the deflators on the spending side are wrong, it must be the case that the industry deflators are also wrong. BEA’s proposed approach to this issue is to reorganize its accounting structure for the medical care industry. The leading idea at the agency is to base this reorganization on a stylized model of health care in which patients are assumed to work through a care gatekeeper. Patients first go to their internist, pediatrician, or other primary care physician, who diagnoses problems and then sends them to different providers—the services are outsourced to labs, to professionals performing MRIs, to surgeons, and so on. Thought of in this way, the final good is provided by the primary care physician who orchestrates the medical care; everyone else in the system is simply an intermediate good. Aizcorbe explained that, for national accounting purposes, this means that spending is deflated by the disease episode–based indexes that are allowed to cross NAICS industry lines, and the intermediate goods are deflated by PPIs. The gap between the real dollar amounts on the spending side and the amounts from the intermediate goods is attributed to the value added of primary caregivers. The critical distinction from the current accounting framework is that the specialists must be viewed as providing intermediate goods. BEA’s Plan to Revise the Medical Care Industry Accounts In his presentation, Brain Moyer provided details about BEA’s plan to modify the industry side of the national accounts so that they can remain synchronized with the disease-based organizational structure proposed for the satellite program. He began by explaining that, in addition to the accounts that register the contribution of consumer spending to real GDP, there is the less familiar set of accounts that show detailed inputs and outputs used in the production process by industry, of which health care is one. Here, real measures of value added by industry are established. Because health care accounts for a large and growing portion of the nation’s economic activity—currently 16 percent of GDP—the importance of measuring its impact accurately and in a way that avoids major statistical discrepancies is self-evident. If a new measure of consumer spending is considered for the personal consumption expenditures (PCE), then that must be traceable back
OCR for page 36
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop into the industry accounts to see which detailed components are contributing to that change. Moyer began by explaining how BEA currently handles health care in the accounts (details of the BEA methodology can be found in Appendix C). BEA’s input-output accounts show detailed transactions by industry, traditionally defined; so these transactions involve employees (i.e., consumers) who purchase health care services from various health care providers—physicians, hospitals, clinics, and so on. The health insurance industry is also viewed as providing a service to consumers (who often access the plans through employers). The GDP-by-industry accounts show real value added for the health care industries. In contrast to the proposed satellite structure, these industries produce and sell final outputs to consumers. As noted above, the proposed change to BEA’s framework will involve introducing a new primary caregiving industry. The primary caregiver industry purchases its inputs from the other industries providing health care—hospitals, clinics, laboratories, pharmacies, and so forth—which are viewed as intermediate purchases in the production of medical care. Other inputs, such as specialist physicians, could also be added. The idea is not new; its real-world counterpart is a health maintenance organization (HMO). In his presentation, Moyer detailed how introduction of the primary caregiver category changes BEA’s industry account picture. Employers still make contributions to employee health care plans, and employees still purchase health insurance; however, the new primary caregiving industry sells its output directly to consumers. This framework allows BEA to reconsider the definition of a unit of medical care and to incorporate disease-based price indexes. Most importantly it will allow the industry accounts and the NIPAs to be in balance, both on the nominal side as well as on the inflation adjusted or the real side. Moyer presented a hypothetical example to illustrate how this additional industry reconciles the two sides of the accounts (Box 3.1). He suggested that the new accounting structure for the industry side is not only a mechanism to ensure that the accounts remain in balance but also has a realistic representational element. The real value added from the other providers is unchanged, which is to be expected, whereas there is an increase in the real value added for primary caregivers. This, he said, can be interpreted as resulting from the coordinating efforts that this newly defined primary caregiving industry is providing. In summary, as a result of moving to this new framework for the satellite account, the expectation is that there would be no impact on nominal consumer spending while the measure of real PCE and real GDP would increase. BEA has produced some initial estimates indicating that real consumer spending on medical care, measured in the new way, may be about 1.5 percentage points higher per year, and real GDP would increase by about 0.2 percentage points per year.
OCR for page 37
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop Discussion of the BEA Plan During open discussion, Barbara Fraumeni—who, as a recent chief economist at BEA, has considerable experience with these issues—commended BEA for moving in parallel on both the expenditure and the production sides of the accounts, so that the inputs and outputs of the system would be identified. She summarized the key elements of BEA’s plan, which she characterized as preserving the accounting structure while putting together an important disease data set that could provide the flexibility necessary to allow price indexes to potentially capture quality change effects. Fraumeni conceded that the current state of the art for measuring quality change at the disease treatment level does not do this quite yet, but she advised that this should be a high priority for the agency. Fraumeni suggested that BEA document its plans and progress through two papers, one that discusses the accounting system and proposed changes to it, and another that describes what can and cannot be done now and what the agency would like to tackle in the future—namely, the quality change issue. Aizcorbe reminded workshop participants that BEA is still very much in the early stages of conceptualizing this industry-side structure and asked the experts assembled in the room to continue the debate on the topic. Extensive discussion followed about how the physician gatekeeper model would be operationalized in the national accounts. Fraumeni described the new model, which reroutes the way that expenditures flow through the national accounts, as essentially involving a “fake billing.” The reason is that the billing does not come entirely from the primary caregiver, and it has no impact in nominal dollar value added or expenditures. Sherry Glied agreed, noting that, even though integrated systems with gatekeepers exist in the real world (e.g., HMOs), medical care often takes place outside these systems. In medical care there are a few general contractors but, in practice, many people serve as their own contractor. From a mechanical standpoint, there needs to be a placeholder there, but it is not clear how it should work. In thinking about how to organize the industry-side inputs, Glied pointed out that sometimes there is a physical representative and sometimes it is a virtual idea. The products of the organizing industry are final episodes of treatment for specific diseases, but, she concluded, it will be a tricky task to figure out what belongs in that category. Joseph Newhouse observed that insurance companies are already working to organize information at the disease level using the ETGs—even though their goals for doing so are somewhat different. They are beginning to actively analyze their businesses into what is called disease management, a new industry that is trying to manage chronic illness better. For the purpose of national accounting, Newhouse noted that the organizer or the gatekeeper is really a residual category to make the accounting entries balance. A mechanism is needed to capture the substitution that takes place, but then double counting has to be avoided, which is difficult when expenditures are fragmented then put back together again. Ralph Bradley added that, if the physician were truly an organizer, a grouper for the
OCR for page 38
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop BOX 3.1 Hypothetical Reconciliation On the spending side of the NIPAs, suppose nominal spending on medical care changes by $100. In the table below, the first column lists how the expenditure would be treated in the current framework; the next column represents the proposed framework. Currently, in the measures of personal consumption expenditures, the PPI—in this case, 1.07—is used to calculate real consumer spending. Since producer prices increased by 7 percent, the result is a $93 real change in consumer spending. Treatment of a Hypothetical Expenditure, Conventional and Satellite Structures Current Proposed Medical care: Change in nominal spending $100 $100 Price index 1.07 1.05 Change in real spending $93 = ($100/1.07) $95 = ($100/1.05) Industry accounts: Change in nominal value added $100 $100 Primary caregiver $20 $20 Other providers $80 $80 Change in real value added $93 $95 Primary caregiver $19 = $20/1.07 $21 = ($100/1.05) −($80/1.07) Other providers $74 = $80/1.07 $74 = $80/1.07 NOTE: Real value added computed through “double deflation.” In the proposed framework, BEA would use the disease-based price index, which, in this example, increases by a slower growth rate relative to the PPI, which is consistent with some of the initial work done by BEA. This leads to a higher calculated growth rate in real consumer spending. To show how this flows through to the industry accounts in this example, the intermediate inputs of other providers are assumed to be zero. In the industry ac- claims database system would not be needed, because physicians would be selecting the pathways of all the treatments and then reporting them. He also pointed out that, when reading a claims database now, there are claims for prescriptions and other medical services that are not assigned a diagnosis. This shows that the grouper fails to assign a medical purchase to a disease for each case. The session concluded with discussion of Moyer’s point that, if the deflator for medical care on the spending side of the accounts is changed, then the rate of GDP growth changes and a gap would open between that rate of real GDP growth
OCR for page 39
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop counts, the goal is to derive a measure of real value added. To do that, the outputs and inputs of an industry must be deflated. In nominal terms, the value added is equal to the output of an industry minus its inputs. Output is deflated with an appropriate price index, as are the inputs, and real intermediate inputs are subtracted from real gross output, which provides the estimates of real value added. This is the typical way of computing real value added in a national accounting framework. Turning to the industry accounts side, under the current structure, the example shows a $100 change in nominal value added—nominal value added equals the change in health care GDP. For this example, $20 of that is attributed to the primary caregiver, $80 to other providers. The assumptions have not changed under the new structure; in nominal terms, everything adds up. Moving to the real side, under the existing structure, outputs and inputs are also deflated for the primary caregiver industry. Since a value of zero is assumed for intermediate inputs, the $20 for the industry is divided by the PPI, which gives a change in real value added of $19. Following the same procedure for the other providers produces a nominal value added of $80 and a real value added of $74. The sum totals $93 again, which matches the figure for real consumer spending on the other side of the account—the two sides balance. Moving to the satellite framework, characterized by the addition of the new primary caregiving industry, the other providers’ value added is still $80; divided by the PPI, a real value added of $74 is obtained. Changing the structure does not change the value added of these other providers. The output of the primary caregiver industry is now the value added, or $20, plus the intermediate purchases it is making from the other providers, $80. Dividing this total by the new disease-based price index, hypothetically set here at 1.05, then subtracting the real value of intermediate inputs (the nominal, $80, divided by the relevant PPI, 1.07) gives $21. This total, plus $74, equals $95—the two sides of the account are in balance. SOURCE: Workshop presentation by Brian Moyer. and the measure of change in the industry accounts (if left unchanged). This gap has to be entered as a line item somewhere; the issue is then how to interpret it. Landefeld reminded participants that this idea of making an adjustment in a top-level category on the industry side of the account, to make it balance with the expenditure side, is not unprecedented. For computers, this is how any gains in productivity are categorized. The residual is attributed to the top-level industry—computer manufacturers, not the component manufacturers. Triplett also framed the industry side accounting issue in the context of pro-
OCR for page 40
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop ductivity, which, like the national accounts, offers a logical construct for organizing data. If indeed the primary care physician industry is to be where all excess multifactor productivity3 will be categorized, he urged BEA to be explicit about the method for treating the residual difference between the expenditure-side and the industry accounts under the old and the new frameworks. This level of detail underlies the accounts, but the numbers published by BEA are for the health care industry, which is the aggregation of the subindustries—the hospitals, the doctors’ offices, and the rest—for which data are supplied by the Census Bureau and BLS. Triplett also suggested that BEA will have the problem that price indexes for the subindustries will not equal those for the aggregate level—there will need to be a reallocation term. Based on the Moyer presentation, BEA now has a story to tell about the source of economic activity driving the inequality, which, in some other contexts, is called a statistical discrepancy. In discussions of the options about how the inequality could be handled in the satellite account (other than the gatekeeper productivity mechanism), David Cutler made the point that one possibility would be to simply create a line in the accounts called “total factor productivity” and not specify which industry gets it. In addition, it is sometimes important to know how productivity is affected in overlapping disease areas below the medical care industry level. As an example, if the medical profession gets better at treating heart attacks, it might suggest that diabetes treatment has improved, since one of the things these patients die of is heart disease. Cutler observed that in one sense this is right, but in another it is not. It is right in the sense of the broad medical care industry, because people with diabetes are now living longer and their quality of life is better. In other ways, it is wrong, because it is not the treatment specifically for diabetes that has led to the improvements. So, for various purposes, productivity gains need to estimated at different levels of aggregation. The key is to make sure that the pieces add up to the total; otherwise, the person who receives treatment for heart disease may be double counted in the estimated productivity improvement from that, as well as from treatments of diabetes, high cholesterol, and hypertension. Only with an accounting structure can one ensure that entries do not appear in multiple places or, if they do, that they are parsed to add up correctly. This is done by looking not only at the expenditures on and productivity of the treatment of the disease, but also at the productivity of each particular input; so it is not just the productivity of treating the heart attack, but also of the hospital and of the physician and of the pharmaceutical company. In response to Cutler, Landefeld made the point that, to allocate the adjustment across industries or subindustries begins to call for a lot more work and, of course, all that has to be transparent. He noted the similarity here with the discussion that has gone on for years about the statistical discrepancy in the accounts. 3 Multifactor productivity is defined as output generated per unit of combined inputs; inputs may include capital, labor, energy, materials, purchased business services, and so forth.
OCR for page 52
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop because the CMS data are longitudinal and clinician approved, they are worth considering for the program. Open Discussion of BLS Plans Triplett commented that, in 1992, the PPI released its then-new hospital price indexes, which were a great advance over what had been done historically. As noted above, for many years, the unit of measurement for BLS price indexes were things like the cost of a day in a hospital. The PPI’s advance was to move toward the episodes-of-treatment concept, in which a diagnosis for in-hospital treatments would be priced out initially and then followed. It is a synthetic price that is estimated by asking the hospital what it charged for a diagnosis that has the same characteristics, the same demographics, and other conditions. The improvement resulted in a price index that grew less rapidly than the older index. Now appears to be time for the next major improvements to BLS price index programs. Triplett summarized three dimensions to the upgrade, noting the encouraging development that BLS is proposing work along all these lines. First is the need to adjust for improved (or deteriorated) medical treatments. Everybody, including BLS, agreed that it would be much better if quality adjustments were made to reflect these improved treatments. The second upgrade to the system is to extend the general approach of pricing episodes of diseases to nonhospital indexes. The third is to follow and perhaps adjust indexes when a treatment moves across facilities—or industries, as the structure is set up now. Currently, the hospital is an industry, the doctor’s office is an industry, and clinics are an industry. Next, Triplett spoke in more detail about BLS quality adjustment plans. The agency’s proposal follows the usual PPI method for making quality adjustments. The PPI is a constant input, fixed technology price index, just as the cost-of-living index is a constant utility, fixed preference function index. The CPI uses consumer preferences as a way to value a change in medical treatments, but the PPI theory on this is based on production costs. So the theoretical ideal is to use the difference in production costs between the old treatment and the new treatment to make a quality adjustment in the index. Triplett expressed skepticism about these PPI plans working because this theory of the output index embodies a problematic conceptual approach. Because the PPI is in theory a fixed input, fixed technology index, it is consistent if the quality change does not involve a change in the underlying production technology; Triplett pointed out that a lot of quality changes fit this model. For example, computers have over some periods used the same technologies, but there have been improvements that make machines faster. In the medical care context, many of the trends of interest involve new technologies. The cataract surgery that moved to a sutureless procedure is a good example. It is an innova-
OCR for page 53
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop tion, not a constant technology process. This innovation not only improved the treatment, but actually reduced its cost. One could take the old technology and ask what would it cost to produce the characteristics of the new treatment in the old technology. The problem, Triplett continued, is that the outcome could not be produced using the old technology. On this topic, Triplett concluded, the procedure described by Murphy might work well for some of the limited purposes for which it has been proposed, but it will not get at the major changes that concern most people working on productivity change in medical care. Those are the big changes in medical technology for which there is not a consistent cost estimate for the new and the old technologies. Again, Triplett praised the presenters for trying to do something about this, but cautioned that there were these limitations in terms of the ability of the methods to pick up the new technology-driven quality changes. Next, Triplett commented on differences between the grouper-based and PPI approaches for handling major changes in medical treatment that involve new technologies. The BLS method of handling the case in which outcomes before and after the change were not equal would involve a linking procedure. The alternatives discussed during the workshop are unit value indexes, in which a direct comparison is implicitly made between two different treatments, and indexes using linking methods, wherein prices of the two treatments are linked and not compared directly. In the case of the direct comparisons, changes to the good or service are ignored. A generic drug is treated as equal to its branded equivalent, and its price is compared directly (the current CPI procedure). The measured price drop is too large for the typical case in which some people do not switch to the generic version. The error arising from the direct comparison depends on the magnitude and direction of the quality change since all of it has been incorrectly been called a price change. The linking method is more complicated because the price change is implicitly assigned from the things that changed (it is not true that the linking method implies the exact opposite of the direct comparison method—that is, that all quality change is ignored). The error occurs, roughly speaking, when prices are rising or falling, and some of the price change is attributed to the product quality change. So, Triplett raised the question, how are we to know what is the right way to do it? Comparing the generic and the branded drug directly might be better than ignoring the price change that occurred when the generic was introduced, even if direct comparison contained an error; but, we should strive for better methods. He suggested that the right way is to avoid constraining oneself to either making a direct comparison (with the implication that there is no quality difference) or to using a linking technique (implying little price change). Triplett argued that an explicit adjustment is the preferable method for adjusting price indexes to account for changing treatment quality. The explicit quality adjustment in the case of medical care requires information on medical outcomes,
OCR for page 54
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop which throws the problem back to the fact that little is known about them. A small number of studies are cited over and over because that is all that has been explored; and, even in those studies, none used explicit outcomes measures. One can only conclude that a lot more work will have to be done before this kind of approach can be implemented in a broad based way throughout a statistical program. Yet, Triplett concluded, it is the right course for future work. Triplett’s final point was about the issue of following a treatment across industries, providers, or facilities. This could be done under the BLS method of collecting data from providers. If outcome measures were available to allow quality adjustment, that could also be done across provider classes. Returning to the cataract example, if BLS knew that outcomes from surgery in the outpatient and inpatient treatments were the same, they could make this direct comparison. But there are so many cases for which this is not known. Without the research, one cannot be sure that reducing the number of days after a normal birth delivery yields an equivalent outcome or whether it is just an attempt to reduce costs. This implies a major research agenda; figuring out how to do the quality adjustment requires a lot of scientific and medical information. The Illustrative Case of Pharmaceuticals Patricia Danzon commented on the BLS presentations, drawing insights from one of her areas of expertise. In the case of pharmaceuticals—as with standard cross-national comparisons that have been done for other services, such as hospital days and physician visits—the practice has been to simply divide expenditures by number of units to infer difference in prices. The expenditures are hugely different, the quantities are similar, and the inference has been that it is therefore the prices that are much higher in the United States. It is these misleading results, Danzon stated, that make pharmaceuticals a good example of why accounting for quality differences is important in medical care price indexes. The issue, then, is to determine how much of the price difference observed in these statistics is really the services and the quality. Because pharmaceuticals are precisely defined—they are measured at the level of the mechanism of action, the strength, the pack, the manufacturer, etc.—Danzon has been able to calculate accurate comparisons of utilization and price differences across countries. She and her colleagues have found that a significant portion of the expenditure difference across countries is explained by variation in the drugs being used—the formulations that may have quality dimensions to them. Clearly, simply dividing expenditures by number of prescriptions can vastly overstate price differences. Danzon raised the issue of international comparison as it relates to the discussions by Triplett and Shapiro. For pharmaceuticals, the tendency to use number of prescriptions is essentially imputing all the expenditure change to a price change (the direct comparison), whereas in reality much of it is in fact attributable
OCR for page 55
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop to new drugs or new formulations and of course the generics. If a CPI tracks volume purely based on the number of prescriptions, it will be upward biased when there is change in technology that is quality improving. She noted that, to her knowledge, BLS was handling generics appropriately now, in that they are being treated as equivalent to the branded products that preceded them. An important gray area is the changes in drug formulation that occur around patent expiration that involve strategies by the branded manufacturers to extend their patent life; for example, the firms may delay the release of new formulations and, instead, introduce single isomer versions of the original drug, or combination products. In BLS’s current procedure for drugs, Danzon suspects, these new formulations or combinations are treated as new products, which may not always be appropriate and may lead to a bias in the price indices. A common manufacturer strategy is to raise the price on the old formulation that is going off patent relative to the new formulation, in order to encourage people to switch to the newer products. Thus, if a price index is being used that tracks the older standard dosage form but does not pick up the delayed-release version that is in fact becoming the norm in the market, it will overestimate the rate of price inflation, since it includes the formulation that is no longer being used very much in the market. When the market baskets are updated, this will be picked up but, in some cases, the delays are significant. Danzon reported that, in these cases, her own research indeed found a much more rapid price increase for the formulations when they were going off patent and being replaced by versions that still had some exclusivity. Danzon also raised the issue of pharmaceutical invoice rebates. When prices are sampled at the hospital level or at the pharmacy level—or, for the PPI, at the manufacturer level—electronic rebates that go directly to payers will be missed for the outpatient pharmaceuticals. These rebates reduce the price to consumers and, in turn, the revenue that the manufacturers get. This is important for the branded products (for the generics, manufacturer revenues will be correct). What may be misreported is the amount that consumers pay, because the discounts go to pharmacies. How much that rebate to the pharmacies then gets passed on to consumers in the form of price reductions of other goods and services is unknown. She commented that, it is mostly an article of faith among economists that much of the rebate that comes from the manufacturer to pharmacy benefit managers is passed back as part of the cost of the drug benefit to the employer and therefore ultimately to the consumer. The evidence that she has seen (a study by the Congressional Budget Office) is that something like 80 percent of these drug-specific rebates were passed back to the employers. If they are not passed back, it would be picked up more in the cost of health insurance—it will not show up in the pharmaceutical component. Danzon noted that these rebates are not trivial amounts. For the generics, the average rebates are on the order of 30 percent; for the branded pharmaceuticals, her best estimate is around 12 percent.
OCR for page 56
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop Similarly, mail order is an increasingly important distribution channel. It now accounts for about 18 percent of total pharmaceutical sales. The prescription is usually for a 3-month supply, whereas from a pharmacy it is a 1-month supply. So, again, if one simply counts the number of prescriptions to estimate the price per prescription, either the mail order users will be missed, or the wrong number will be calculated, because the content of every prescription is dramatically different, by a factor of three in this case. Following up on Danzon’s example, Cutler added that, for these cases in which different strengths in dosages and formulations exist and substitutions across them occur, the more aggregated index should perform better than the more detailed version, provided quality adjustment is done properly. The aggregated index will take the total growth in pharmaceutical spending at the condition level and then ask how much of that growth is the result of new drugs and what is their net value. In this respect, moving to the larger pricing buckets is actually the right way to go and immensely important. The question then becomes: Is there any way to link across those buckets? Cutler added that, no matter which way the index is constructed, the quality adjustment is needed. It may be simpler for a more aggregated index, because each exact formulation does not have to be dealt with individually. Anything that is truly a new good will raise a different problem that will be missed either way; that is a big remaining issue. 3.5. OUTCOMES AND QUALITY CHANGE At several points during the workshop, participants made the point that, to monitor quality change in medical care for purposes of price measurement, accurate data on outcomes for treatments—defined in parallel with the expenditure categories—would be needed. Mark McClellan, of the Brookings Institution and formerly of CMS and the Food and Drug Administration, spoke about measuring treatment outcome in this context. Among participants, there seemed to be complete agreement that quality adjustment of price indexes for the satellite health care accounts is extremely important and also that it is very hard to do. McClellan began by noting that many of the measurement problems faced in the construction of price indexes for health accounts are increasingly important elsewhere for people—payers, purchasers, consumer groups, providers, and policy makers—concerned with where the health care system is going. The work that is going on in parallel creates an opportunity for collaboration on doing a better job of measuring outcomes and on putting more of a focus on value in the processes of health care decision making and policy. During open discussion, Linda Bilheimer (National Center for Health Statistics) asked what policy makers are looking for in terms of measures of outcomes. McClellan responded that it depends on whom and when you ask. If it is a briefing before the Joint Economic Committee about where health care should
OCR for page 57
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop be headed in the next 5 or 10 years, then topics related to value and to accounting for productivity changes arise. If it is a Congressional Budget Committee meeting on how to get Medicare physician payment problems addressed for next year, then the topics are about the price and the policy changes that could affect nominal budget spending. As work moves forward in this area, one of the side benefits will be a better awareness among the general public and policy makers about the distinction in these kinds of questions. When people are asked today about what is wrong with health care, their response is increasing costs, and they equate that with prices going up—the premiums that they pay on their insurance plan and so forth. Even though they are individually quite satisfied with the care they are getting and perhaps reluctant to see major changes in health care policies that could directly affect their care, there is less recognition of the broader questions of how policies are affecting value in health care and what people are getting for what they spend. Among the many challenges with measuring outcomes and accounting for them in indexes is sorting out the impact of health care on health, which can be hard to isolate, especially at the patient level; so many factors influence health. Also, there are few standard quality or outcome measures established for many aspects of health care. The trend has been to start with narrow pieces of the picture—such as a look at a specific disease—and try to expand that over time as data and technical expertise get better. McClellan reminded the audience that there is still a long way to go. Next, McClellan reiterated the point made throughout the workshop about the measurement problems created by the existence of multiple chronic conditions. He noted that it is getting more and more difficult to isolate diseases that coexist in individuals. The vast majority of Medicare spending now is on people who have multiple chronic conditions. For these patients, health professionals are increasingly realizing that focusing on one particular disease and its treatment leads to real missed opportunities to improve the coordination and results of care. Accordingly, efforts are being made to cut across disease areas with prescriptions for behavioral changes and medications, compliance systems, and the like that are not easy to attribute in patients with multiple conditions or any particular disease. Even when this kind of effort is made, however, McClellan agreed that it was hard to sort out how much of a given health status effect is due to the medical treatment and how much to other factors. Health is clearly improving over time, although at different rates for different kinds of disease treatments, and these trends reflect changes not only in medical technology, but also in biomedical knowledge that affects behavior, as well as nonmedical factors that are not measured as medical care in the economy. Wellness expenditures are a growing industry, and food improvement, education, socioeconomic status, and environmental exposures certainly affect health and are important determinants as well.
OCR for page 58
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop For all of these reasons, McClellan pointed out, there has not been a lot of practical application of outcome measures to ongoing health care policy aimed at improving the value of what society is getting for its spending. Instead, much of the focus has tended to be on process-of-care measures, for which it is easier to conclude with a reasonable level of confidence, from clinical studies and expert opinion, that using certain kinds of treatment for particular conditions or combinations of conditions leads to better results. As an aside, he noted that the results have not been particularly encouraging about whether the health care system is reliably delivering quality care. Nor have the these methods been very useful as surrogates for outcome measures; they have tended to focus on specific aspects of care and do not capture most of the things that consumers or even providers need to know in their decisions about health care. On the positive side, McClellan reported that many efforts are under way to change the way information about the health care system is processed. This is where McClellan sees some parallels and some opportunities for collaboration between the kinds of people who attended the workshop and those who are working in such areas as quality improvement, payment reform, benefit redesign, and the like. Among the interested parties are provider groups that have been struggling with the traditional ways of paying for programs like Medicare based on volume and intensity, in which the final common pathway to address rising spending is to squeeze down prices; this, McClellan stated, is not working very well in terms of promoting quality and value or even long-term cost savings, whether it is health plans or the employers who use them, who want more accountability for what they are getting for their spending. There are also interested consumer groups, like Consumers Union, that are now engaged in initiatives to make health care much more like choosing appliances and cars; they want to see information about providers and health plans, just like those in Consumer Reports for these other areas. CMS’s Hospital Compare database is another example of how this work on quality measurement—and not only processes of care, but also outcomes and satisfaction measures—is progressing. The Hospital Compare site was implemented several years ago; since then it has expanded and now includes several outcome measures. There are now CMS reports on Hospital Compare for 30-day mortality from acute myocardial infarction and 30-day mortality from heart failure. McClellan expressed the hope that more will be coming soon in the area of surgical outcome measures and a range of other survival measures. McClellan reported that a final area for which CMS is beginning to expand measures available on Hospital Compare—very much related to outcomes—involves standardized patient satisfaction measures. He said that surveyed patients generally respond favorably when asked whether they received satisfactory care. However, he continued, more detailed data providing a deeper understanding of relationships with providers and doctors and nurses, of how information about the condition was communicated, and of how patients felt about particular aspects of
OCR for page 59
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop the treatments they received can all be very helpful. This turns out to be especially true for patients with chronic diseases, who often have a good idea of what is working to keep their conditions under control. Gail Wilensky commented that it is very helpful that people are being pushed to understand better, with tools like the Hospital Compare data or other data that are becoming available, that there are different ways to measure outcomes. She also noted that questions have to be framed very carefully. For example, when individuals are asked what is important to them about health care costs or health care prices, they typically think only in terms of what they are paying out-of-pocket—that is their working definition. McClellan also mentioned that other parallel efforts are under way for physician care, nursing home care, and pharmacy care, but they are not as far along as the Hospital Compare data. They all follow a similar general model, starting with some process-of-care measures, then push toward looking at outcomes, patient satisfaction, and other aspects related to outcomes. With this increasing emphasis focusing on value—and not just volume and intensity and prices in health care reform—there have been a number of efforts bringing together provider groups, payers, purchasers, and consumer groups of health care. One purpose has been to get consensus behind methods to measure the quality of outcomes and costs, and to do it more at the episode or patient level, rather than just in a particular silo (such as hospitals, physicians’ offices, etc.) of care. McClellan described the typical process as involving a number of organizations that become involved in developing the technical details of what a quality measure, whether it is process or outcome or satisfaction, might look like. Next, a process is coordinated by the National Quality Forum, a congressionally recognized nonprofit organization, to try to get a consensus endorsement behind particular measures. These processes by themselves do not do anything to get the quality measures into use in practice. According to McClellan, there have been a number of collaborative activities developed over the last few years to do that. Most of them are in the form of quality alliances or hospital quality alliances; these are instrumental in creating a consensus behind the measures of care that are used in such systems as Hospital Compare. One of these is a group called AQA, formerly Ambulatory Care Quality Alliance, which is concerned mainly about physician and ambulatory care quality measurement and is behind some of the efforts by Medicare and private payers to put more emphasis on quality and payment reporting. On the ambulatory side, an organization called Pharm Quality Alliance is working toward some similar goals. A group called the Quality Alliance Steering Committee has been charged with trying to help these groups work together, to collaborate in this effort to get more consistent and common metrics out of the nation’s pluralistic health care system. The focus of the steering committee is to ensure that measures being developed in each of these silos, as McClellan described the various areas, are not only harmonized, but also on a track focusing on overall pictures of quality and cost at the patient level, or at least at the episode level, and on ways to get
OCR for page 60
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop synergistic benefits from using data from multiple sources together. One of the challenges in developing and using these measures is that any health plan, even Medicare, does not get a complete picture of care quality at the level of providers and their treatment of diseases or other conditions, or even at the patient level. Much of the focus on these collaborative efforts has been on trying to harmonize the different measures that various health plans, Medicare, or employers are using that would facilitate an aggregated approach that provides a more complete picture of quality and cost of care. 3.6. DATA NEEDS FOR PRICE MEASUREMENT, TRACKING OUTCOMES, AND QUALITY ADJUSTMENT Mark McClellan’s presentation also touched on some overarching data issues for measuring quality change; many of these parallel points were made during the discussions on medical care expenditures. McClellan stated that the aggregated data—compiled at the level of the provider, the health plan, or the treatment of similar patients—are what matters, not the fact that a specific patient was treated by a specific doctor with certain results. If participation processes and data collection for the health care system were carried out with some consistency, it would be possible to perform complex analyses (e.g., multivariate regressions) and to produce relatively sophisticated measures in a distributed data system. Furthermore, he pointed out that if a truly electronic health care system were created, it would have much more analytic value—not necessarily in terms of data volume but through development of consistent rules and standards being applied that would enable researchers to use it much more effectively. Danzon commented that there is a huge amount of data that are already collected by the pharmaceutical companies on comparative effectiveness of new technologies versus old technologies or new drugs versus old drugs that they have to collect in order to make their case for reimbursement in many foreign countries and increasingly with health plans in the United States. Many of the data are collected as part of clinical trials. With that comes limitations, but the data would provide some evidence about new technology versus old. In an efficient system, BLS and BEA would be able to take advantage of the millions of dollars spent collecting these outcomes data as part of this exercise. Newhouse pointed out that McClellan actually did a paper 10 years ago or so on heart attacks in which clinical trial data were used to break down the components of improvements and outcomes and attributes them to changes in specific aspects of treatment. McClellan pointed out that data could continue to be used in that way; however, in terms of actual health care delivery, there is a big gap between how well the technologies could be used and how they are used in practice. In most cases, even when there is a big medical breakthrough, it does not appear in the data from one year to the next. He cited the example of beta blockers used in heart attack patients—a treat-
OCR for page 61
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop ment that, beginning in the late 1970s, was used to substantially improve survival of patients after acute myocardial infarction. In the 1980s it started to be tracked in a few limited settings, and in the 1990s it became part of a routine quality measure that was part of Hospital Compare for a while. But it took three decades to get from the time when the studies were done to when there was complete use and practice of the procedure. So these kinds of data, in conjunction with trend information on the use of different technologies, can certainly be useful. But they are not going to provide a complete picture. The only other caution he added is that the quality adjustments could be significant; the measures on outcomes are going to get better over time, but they are almost certainly going to be different from year to year—it is very hard to maintain consistency over longer time periods. Triplett pointed out that there have still been only a small number of disease-based studies in the economics literature—the heart attack study (Cutler et al., 1998), the cataract study (Shapiro et al., 2001), the mental health study (Berndt et al., 2001), and a couple of others. However, outcome measures were not used in any of these. The heart attack study used mortality, but this was not a full outcome measure—it is a lower bound, because it does not pick up morbidity effects. For the cataract and mental health studies, there were no direct outcome measures either (for cataracts, the researchers contended that they had estimated a lower bound). Triplett concluded that there is not that much low-hanging fruit from the literature to pick up. Despite agreement in the past couple of decades that outcome measures are needed to conduct cost-effectiveness studies—which everyone agreed are important, as they are being used for health care planning as well as in research—the medical literature on the topic is still not extensive, and that limits what economists can do. McClellan responded that the technical ability to do something about this, to come up with some more reasonable and more complete outcome measures, though still very incomplete in terms of everything people might care about, has gotten a lot better. There is currently clearer policy agreement that just focusing on volume and price restrictions is not going to be enough. That said, McClellan agreed that there is still a long way to go. But, as more work gets done in this area, one of the side benefits is going to be a better awareness among the general public and policy makers about the distinction between cost and price and productivity questions. Some of the movement in the direction of quality measurement is also being driven by policy and legal pressures. McClellan cited a recent settlement in which the New York attorney general required transparency and the use of nationally recognized quality standards from major health insurers who have been trying to use measures of quality and costs of care as either conditions in their contracting with providers or as factors that influence the structure of their benefits (e.g., setting lower copays and perhaps paying the physicians and hospitals according to performance based on whatever measure each health plan or each employer came up with). The point here is that these metrics need to be based on both quality and cost, and they need to have a more comprehensive and consistent picture
OCR for page 62
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop than a health plan is likely to be able to get on its own with the limited number of patients that it covers. There has been an added push for these efforts to do network or aggregated approaches to quality and cost measurement. The practical relevance of this to the work that is being done in health accounts and price indexes is still a way off. However, McClellan pointed out that some of these broader measures are in the early stages of being constructed and made available. There is a broad national public-private roadmap planning effort to move from data that are based just on claims to data that include what might be called clinically enriched electronic information, like lab results and increasingly sophisticated information from electronic records or personal health records. There is a parallel between the kind of work that is going on here and the kind of work that is going on in the health accounts area. McClellan expressed his concern that the initiatives he is involved with, as well as the overlapping health accounting programs, move forward as effectively as possible from a policy reform standpoint; this will require efficient use of data. Ideally, a virtuous cycle could be created in which, with more and better information available on outcomes and costs of care and therefore on the value of care, there will be a movement toward payment and benefit system designs that reward and support better value and clearer evidence about what actually works. McClellan concluded his remarks by suggesting that some ongoing involvement of the BEA and the National Academies on work that is happening in these quality measurement, value measurement, and quality improvement efforts would make a lot of sense. The measurement goals across the various interests are similar; the only difference is that BEA has to focus on the national accounts as opposed to the actual impact on delivery of care.