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 12
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop 2 Allocating Nominal Expenditures on Medical Care: A Disease-Based Conceptual Approach In the Bureau of Economic Analysis’s (BEA’s) phased plan for implementing a satellite account, the first major task is to define expenditure categories and devise a method for allocating economy-wide spending on medical care into those categories. Expenditure data are needed for multiple purposes—for health program administration, for the production of price indexes, for productivity analysis of the economy’s medical care sector, for national income and product accounting, for disease treatment monitoring, and for making cross-country comparisons of health systems—and the ideal data set characteristics and organizational framework will be different for each. Currently, the U.S. health expenditure accounts, produced by the Centers for Medicare and Medicaid Services (CMS), essentially track the flow of funds based on final payments from payers (private insurance, government programs, out-of-pocket) to payees (hospitals, physicians, drug vendors, nursing homes).1 For many of the purposes raised during the workshop, the capability is needed to aggregate expenditure data into units defined along different lines—specifically, the real outputs of medical care. As Dale Jorgenson put it, the main objective is to collect data on the prices and quantities associated with the output of the sector and to cope with the enormous heterogeneity and the very rapid evolution of the character of the products, which differ both within and across providers. Workshop participants agreed that, because they serve as a building block for many kinds of health data systems, creating new ways of organizing and tracking health care expenditures is an immediate priority. This work would be useful for both 1 To get a sense of the breadth of expenditure information produced by CMS, see the data tables produced on the agency’s website (http://www.cms.hhs.gov/NationalHealthExpendData/downloads/tables.pdf).
OCR for page 13
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop the experimental health accounting and national income and product accounting purposes, as well as for price and productivity measurement. Ideally, medical sector goods and services would be defined in such a way that: (1) expenditures could be estimated each period for every good or service produced by the industry, (2) meaningful quantities and prices (nominal and real) could be tracked, and (3) quality change of the goods and services could be monitored. The first task in the accounting exercise is to allocate nominal expenditures to the various array of outputs. Assuming that patients seek medical care to treat specific conditions or diseases, the medical care output should be defined and arrayed to reflect that consumption objective. Many of the researchers present at the workshop favor an episodes-of-treatment organizing principle for doing this. 2.1. METHODS FOR ATTRIBUTING SPENDING TO TREATMENTS: THE BIG COMORBIDITY ISSUE During her presentation, Ana Aizcorbe identified several options for attributing spending across treatment episodes, or “disease buckets,” as several participants described them. One is an encounter-based method in which spending is attributed to one or to several diagnoses as reflected by data extracted from patient claims. A second, broader approach involves constructing episodes of treatment—which may include numerous encounters over a predefined period—then adding up dollars spent nationally on each of the range of diseases and conditions. A third possibility is a person-based approach, in which spending on various treatments is tracked on a person-by-person basis over a set period of time. Within these approaches, there are different techniques available for assigning the dollars spent to the treatment categories. The applicability and appropriateness of the methods varies by the accounting objective, and each has its pros and cons. Aizcorbe conceded that, at this point, it is unclear which is the best way to move forward for BEA’s specific application. BEA is working both internally and with the Cutler-Rosen team to establish what the allocations may look like under the different methods, and whether it matters for estimating expenditures and prices (see Section 2.3.). Speaking about this project, which has begun producing episodes-of-treatment cost estimates, Allison Rosen noted that spending could also be further broken down into subcategories along functional lines, such as disease prevention, diagnosis, and screening activities. This is important, since not all spending on medical care can be attributed specifically to the treatment of a disease or condition. Whichever method of allocating expenditures is used, it has to offer a solution to the comorbidity problem. Dealing with the reality that many patients utilize medical services for multiple conditions is a major issue to be confronted in health accounting. It is a problem on the expenditure side—BEA must figure out how to allocate spending for cases in which patients receive medical care
OCR for page 14
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop for more than one disease. It is also a problem on the outcome side—how can researchers determine which treatments are incrementally affecting the quantity and quality of life among populations receiving care for multiple conditions. Sherry Glied identified another dimension, distinguishing between horizontal comorbidity—multiple things happening to a patient simultaneously, which muddies the question of primary diagnosis—and vertical comorbidity—which deals with sequences of risk factors and also complicates designation of an episode of treatment. On one hand, if a patient’s cholesterol is treated and that person never goes on to develop heart disease, how is that handled? Where are those expenditures grouped? On the other hand, if a person is treated for heart disease and is made even sicker because the drugs have side effects, where do those treatments fit? Are they part of heart disease treatment, or should they be categorized elsewhere? Rosen cited cardiovascular disease as perhaps the classic example of comorbidities. She noted that the place where comorbidity issues are most marked is in the risk factors—it is rare to see one factor without at least one other—and one could consider making separate buckets for patients in this group. For example, diabetes probably needs to be separated out because of all of the other complications that it causes. Glied concluded that, in many cases, defining what is a final product of the medical care industry is going to be a tough task for BEA to handle. Encounter-Based Approach One relatively simple method for reporting the cost of illness by disease involves tracking spending that takes place at the patient encounter level. Information about the cost of specific patient encounters with the medical care system can be found in administrative data, such as claims forms (often, these have the payment the provider requests, but the payer pays something less than that); expenditures can be allocated based on diagnosis identifiers. The Altarum research (described below) used primary or first-listed diagnostic category for this purpose. Within this method, disease buckets can be allocated at varying levels of detail. For the Altarum project, 660 clinical classification categories were used based on groupings created at the Agency for Healthcare Research and Quality (AHRQ). Figures for these categories can, if desired, be aggregated into a smaller number of buckets. Aizcorbe made the point that relying on the primary diagnosis may seem like a coarse decision rule, but in fact these kinds of compromises are often needed in the production of the national accounts. Firms (or even establishments) exist that produce a range of different goods, sometimes across more than one primary industry, and their outputs have to be allocated. In such cases, BEA analysts must figure out where to allocate dollars associated with each type of output in the industry accounts. It is important to have time series data produced on a consistent basis, even if the way that the dollars are allocated is not completely accurate.
OCR for page 15
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop Rosen pointed out that there is a rich history of using encounter-type approaches in cost-of-illness estimation for the health care system, going back to Rice, Hodgson, and others (Rice, 1966; Rice, Hodgson, and Kopstein, 1985); a number of these studies have used data from the Medical Expenditure Panel Survey (MEPS). Among the pros of the approach is that calculating spending and attributing it to diseases is easier on an encounter basis than it is for some other kinds of measurement units. On the downside, the encounter approach is fairly limited in its capacity to handle comorbidities; a number of providers do not use claims (though some organizations, such as Kaiser and the Veterans Administration, do have, in effect, dummy claims that associate costs with services) or use claims that do not always provide valid disease diagnoses. Episode-Based Approach An alternative organizing principle is a medical care or treatment episode, which is a broader concept than an encounter. As Rosen explained, under this approach, claims are organized into clinically distinct episodes of care that are “adjusted for disease severity and complexity.” In the case of a heart attack, the episode involves not only a patient’s hospital stay, but also the convalescent time and the care that is given afterward over some discrete window of time. In addition, because there might be an acute myocardial infarction or an acute myocardial infarction complicated by congestive heart failure, there can be varying degrees of severity within a given disease; ideally this variation would be accounted for in the expenditure allocation. A number of commercial firms create so-called episode groupers. As described by Rosen, a major purpose of the groupers is to try to get at some of the differences between chronic long-term diseases and acute short-term diseases. For example, for the office visit of a diabetes patient, the doctor may assign an International Classification of Diseases (ICD-9) code, and the spending will be attributed as such under an encounter-based approach. However, the doctor may also give a prescription for a medication for hypertension, at least in part because of the patient’s diabetes. If that is not something that gets picked up on the pharmaceutical claim, then it may be assigned to another category. The grouper methods attempt to identify a window of time so that, for the diagnosis of diabetes, all spending for some predetermined length of time would be assigned to that disease. For another diagnosis, the appropriate time period may be different. Using the disease classification codes (such as ICD-9) to categorize patients is fairly straightforward; what needs considerable work is the question of determining the rule set for defining what that chronic episode looks like and how long it lasts. The episode unit of analysis can be distinguished from the encounter-based approach in that it consists of groups of claims that take place over an expanded, variable time window. These characteristics allow users of the approach—who
OCR for page 16
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop already include a number of commercial insurers—to take into account both the per-unit cost and the volume and mix of services. At this point, the major problem is that grouper software is proprietary, and the algorithms underlying these tools are concealed. Ideally, a statistical agency’s products should embody methodological transparency to users. Person-Based Approach The person-based approach estimates expenditures as a function of spending at a different unit of analysis—the individual who is being treated for some vector of diseases or conditions. In their broad-based health accounting work, Cutler and Rosen have been using regression techniques to assign spending across disease episode treatments at the person level. The dependent variable is cost, or total expenditures on medical care, which is regressed against a set of disease dummy variables. The expenditure period is a prespecified time window—Cutler and Rosen have been using one year. The results provide a picture of the incremental per-patient annual spending attributable to each disease category. Rosen expressed the view that the person-based regression approach is probably the best of the options for handling comorbidities. At the event or encounter level, many patient contacts with the medical profession (doctor visits) are attributed to a single specific reason, if they are coded at all. Although there can be a huge number of disease buckets (Aizcorbe noted about 700,000 if combination categories are allowed in the claims data), clinical knowledge can be used to narrow these down to essential groups of comorbidities. For example, the cardiovascular disease risk factors might include hypertension, hyperlipidemia, coronary heart disease, and the like. Given the limited sample sizes of available data sets such as MEPS, it is absolutely necessary to identify the most relevant comorbidity combinations. During open discussion, Joseph Newhouse made the point that, since no accounting system can manage 700,000 separate disease buckets, some will be collapsed into the regression’s residual category. Implicitly the magnitude of this residual is dictated by how many interactions are specified in the regression. He wondered whether, for the comorbidity issue, the magnitude of the problem was more or less the same in the episode and the regression approach, because it all turns on what is specified in the interaction terms. Another attractive conceptual feature of the person-based regression approach is that it can be readily matched to health improvements, because analyses on the health services and outcomes side tend to be monitored patient by patient. Health improvements are not typically monitored immediately before and after treatment; researchers look at how a person or group of persons fares over some period of time after receiving treatment. The costs of cases for which there are no valid claims or ICD-9 codes can still be attributed through surveys as well, which is another positive feature.
OCR for page 17
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop These features notwithstanding, the person-based regression approach does have limitations. Ralph Bradley of BEA raised several methodological issues that have to be confronted, particularly if the method were to be used in price index construction. One has to do with population sample coverage; if an analyst runs the regression using data for the entire population, it will yield one set of coefficients; if data are separated, for example, between those who are insured and uninsured, an analyst will get completely different sets of coefficients. Alan Garber (Stanford University) agreed with Bradley that the analysis will beheavily sample dependent. In order to minimize the omitted variable bias, a representative sample is required. Any causal interpretation of the disease dummy coefficients would be problematic. If—in trying to answer a question like, “What would happen if we eliminate a disease by using a particular drug or doing a particular operation?”—an analyst simply plugs results into a model to estimate overall expenditures or drug expenditures, it is likely to be wrong, because there will be omitted variables and the change will not be the same as predicted by the sample from which the data were generated. There is also the issue of how, in the regression approach, to allocate the intercepts for the base spending for the year. These criticisms aside, Garber expressed sympathy for the regression approach, in part because there are not many alternatives. The key is to be cautious about how the model’s coefficients are interpreted and applied. Rosen agreed that the approach has problems to be overcome, and they tend to be related to the tremendous amount of heterogeneity in terms of who gets what medical care. Matthew Shapiro made the point that, ideally, one would want to stratify the results. For example, the elderly will have a very different spending profile for certain diseases than the young, and one would like to be able to deal with that. With advanced age, comorbidity is much more likely to be present. Therefore, adding up simple cases—diabetes, heart attack, etc.—is not going to work very well; it might be hard to extrapolate from a 50-year-old’s noncomplex heart attack to what would happen with a 70-year-old. Fortunately, data exist with which to investigate these issues; however, the more that the analysis is driven to define activity at group levels, the greater the required sample size becomes. Shapiro added the related observation that, if database size were not an issue, one could think of comorbidities as separate diagnoses; a simple heart attack would be one diagnosis, a complex heart attack or heart attack plus diabetes another. As discussed in Section 2.3., which cost allocation method is best will differ on the basis of how it is to be used. For the creation of price indices, a person-based approach may not be as appropriate as an episode-based approach. If the goal is to broadly relate cost and health improvements or to compare costs and health improvements within a given disease on a micro level, as done in cost-effectiveness studies and decision analysis, that might be better done with a person-based regression approach.
OCR for page 18
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop 2.2. ALLOCATING PERSONAL HEALTH EXPENDITURES BY MEDICAL CONDITION: THE ALTARUM INSTITUTE PROJECT Charles Roehrig presented his and his colleagues’ work at Altarum, a nonprofit institute based in Ann Arbor, Michigan, on reallocating national health expenditure (NHE) estimates produced by CMS into medical condition categories. This work evolved from efforts by the institute to develop a model to forecast national health expenditures consistent with the NHE accounts. Their interest was primarily to further understand the drivers of health care spending growth and the prevalence of medical conditions. The project, a nine-month effort supported by the Pharmaceutical Research and Manufacturers of America, benefited from the advice of several experts on measuring health care expenditures including Linda Bilheimer, Mike Chernew, Joel Cohen, Mark Freeland, Rod Hayward, Steve Heffler, and Judy Lave—several of whom attended the workshop. Roehrig detailed through how the project allocated expenditures by medical condition. The first step involved revising the NHE revenue categories to create a more function-oriented picture. For example, hospital-owned nursing home revenues were shifted from the “hospital” category to the “nursing home” category. The “purified” service categories consisted of the following: Hospital Physician Prescription drugs Nursing home Home health Dental Other professional Other personal Durable medical equipment Nondurables For reasons that become apparent below, MEPS records also had to be mapped into the NHE categories, as shown in Table 2.1. The method for reallocating expenditures into the purified categories was based primarily on a detailed study done jointly with AHRQ and the Office of the Actuary at CMS. The results for year 2002 are shown in Table 2.2. The first column shows how the $1.3 trillion in personal health expenditures were allocated by the original NHE service types. The post-reallocation numbers, intended to provide a more functional picture, are listed in the column on the right. Of note is the large reallocation of expenditures out of the hospital (8.1 percent) and physician (15.6 percent) categories to the others (for example, 2.7 percent to home health and 4.4 percent to nursing homes). The totals under the two structures are the same but, according to Roehrig, the hospital expenditure is more closely aligned with what most of us think of as hospital services.
OCR for page 19
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop TABLE 2.1 Mapping of MEPS Event Categories into NHE Service Types MEPS Event Type Charge Type NHE Service Inpatient Separately billed doctor (SBD) Physician Inpatient Facility Hospital Outpatient SBD Physician Outpatient Facility Hospital Emergency Room SBD Physician Emergency Room Facility Hospital Office based Doctor Physician Office based Other provider Other services Home health n/a Home health Prescription drugs n/a Prescription drugs NOTE: n/a = not available. SOURCE: Workshop presentation by Charles Roehrig. TABLE 2.2 Sample Calculation of Medical Care Expenditures by Functional Category for 2002 (in billions of dollars) Service Type Baseline NHE Shifts Out Shifts In Purified NHE Hospital 488.6 39.6 0.0 449.0 Physician and clinical 337.9 52.7 0.0 285.2 Dental 73.3 0.0 0.0 73.3 Other professional 45.7 1.8 33.7 77.6 Home health 34.3 5.7 13.3 14.9 Nondurable medical products 30.9 0.0 0.0 30.9 Prescription drugs 157.9 0.0 10.1 168.0 Durable medical equipment 20.8 0.0 8.2 29.0 Nursing home 105.7 0.0 21.3 127.0 Other personal care 46.3 0.0 13.2 59.5 Total 1341.4 99.8 99.8 1314.4 SOURCE: Workshop presentation by Charles Roehrig. The second step of the allocation exercise was to calculate the distribution within each functional expenditure category by population group; the designated groups are the civilian noninstitutionalized population, various institutionalized populations, and active-duty military, because that is how the data sources break down, more or less. The researchers primarily used the MEPS-sourced data developed by Sing et al. (2006). Finally, for each functional category by subpopulation cell, expenditure totals were distributed by medical condition. Altarum used the AHRQ clinical classification system, which Cutler and Rosen have also used in their project. The civilian noninstitutionalized population accounts for the overwhelming share of
OCR for page 20
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop spending. For example (again based on analysis of the Sing et al. data), 84.2 percent of hospital spending was by the civilian noninstitutionalized population; the next highest population group—nursing home patients who had an acute episode and were admitted to the hospital for a period—accounted for a comparatively modest 6.5 percent. Similarly, about 82 percent of personal health expenditures were attributable to the civilian noninstitutionalized population, in the sense captured in MEPS, and another 14 percent to the nursing home population. Altarum relied heavily on MEPS for data on the civilian noninstitutionalized population. MEPS provides spending by person, encounter or event, type of service, and the medical condition broken down into 260 Clinical Classification Software (CCS) categories by 7 service types—in some instances, multiple conditions are present, and sometimes there are missing conditions. For care delivered to nursing home residents, the researchers used data from the National Nursing Home Survey. For nursing home residents who were admitted to a hospital, Healthcare Cost and Utilization Project data were used. Roehrig reported that they plan to use the Medicare Current Beneficiary Survey for future work. The project’s final database includes 10 years of data from 1996 through 2005, all of the years for which MEPS data are available. Altarum was unable to attain conditional distribution information for about 9 percent of personal health expenditures on items like other nondurables (e.g., tissues, things bought at the pharmacy) and durable medical equipment. Roehrig indicated that they would be able to allocate these items. “Other personal care,” a catchall category, includes such items as industrial implant services and Medicaid waiver programs aimed at keeping people in their homes and out of nursing homes that are difficult to assign to specific categories. Next, Roehrig explained how they dealt with comorbidity—patients in the MEPS data set with multiple conditions. The vast majority of expenditure data in MEPS is on individual events—inpatient episodes, outpatient visits, prescription drugs—that have only one condition assigned to them. A patient could have multiple medical issues but, for example, if he breaks a leg, the treatment record typically indicates just that primary purpose. At the event level, the issue of comorbidities is not nearly as conspicuous as it would be in a person-level or even episode-level analysis. Roehrig also noted that there are sharp differences across medical conditions—some show up much more often with comorbidities. For example, inpatient events for back problems almost always show up in MEPS with that singular condition; the same is true for cancer. However, inpatients hospitalized with diabetes or hypertension more often than not have other conditions recorded in the MEPS data. The project team considered a couple of ways of dealing with these comorbidity problems. The simplest option is an unweighted allocation—if the patient has two conditions, spending is split 50-50 between them; if there are three conditions, it is split in thirds. Roehrig termed this the proportional approach. The
OCR for page 21
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop second option is a weighted approach that takes into account the average cost of an event for each condition when is appears alone. So, for example, the estimate for diabetes is based on the average cost of all hospital events that are just for diabetes. The same is done across all conditions. If diabetes appears with a second condition, the two are weighted proportionately with their average standalone costs. This is the approach that was ultimately used for the research effort, largely on the grounds that it made intuitive sense. The exception is the nursing home population, a great majority of whom display multiple conditions. Here, the weighted approach was not feasible and the unweighted allocation was used. Roehrig then presented the study’s results. Figure 2.1 shows expenditures by diagnostic category over a 10-year period; 262 AHRQ CCS categories were grouped into ICD-9 codes. The circulatory system category accounted for the largest share, about 17 percent of personal health expenditures. The next seven codes each contributed between 6 and 9 percent of the total. Approximately 50 percent of expenditures are captured in these groupings. MEPS data were also tabulated to estimate the most costly medical conditions. (See Table 2.3.) Comparing these levels with those from 1996 allowed Altarum to estimate spending growth rates for medical conditions. Pneumonia, chronic obstructive pulmonary disease, lung cancer, stroke, and coronary heart disease were categories showing the slowest expenditure growth rates, all at 4 percent or less. This may reflect some beneficial effects of reductions insmoking over the period. FIGURE 2.1 Annual health care expenditures by diagnostic category. SOURCE: Workshop presentation by Charles Roehrig.
OCR for page 22
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop TABLE 2.3 The 15 Most Costly Medical Conditions, in Terms of Personal Health Expenditures (PHE), 2005 (in billions of dollars) Medical Condition PHE Mental disorder 142.2 Heart conditions 123.1 Trauma 100.2 Cancer 99.4 Pulmonary conditions 64.6 Hypertension 50.2 Osteoarthritis 48.0 Back problems 40.1 Kidney disease 35.9 Diabetes 35.8 Endocrine disordersa 29.2 Skin disorders 27.2 Cerebrovascular disease 26.8 Hyperlipidemia 22.8 Infectious diseases 22.5 aExcludes diabetes and hyperlipidemia. SOURCE: Workshop presentation by Charles Roehrig. 2.3. COMPARING THE METHODS Over the past year or two, the Cutler-Rosen group has been working with BEA to empirically assess differences in the various approaches to allocating medical care expenditures by disease; Rosen reported some preliminary findings. The research objective is to reconcile disease categories among the encounter-, episode-, and person-based regression approaches; to simulate costs of diseases using each; and to compare and contrast the findings. For this project, Rosen and Cutler have been using health claims data for the period 2003-2005 from Pharmetrics Inc. For 2003, the data cover just over 3 million patients and include total spending of $9.09 billion on inpatient and outpatient services, office visits, prescription drugs, skilled nursing facilities, and laboratory services. Up to four ICD-9 diagnoses are present on a given claim, although only the primary diagnosis is listed for hospital claims. Symmetry software from Ingenix was used to link medical expenditures to disease categories. In order to reconcile the three approaches to common disease categories, Rosen et al. first mapped ICD-9 codes into CCS categories. These were aggregated into 65 clinically meaningful groups that had been developed earlier based on advice from physicians. Cost categories were created primarily for diseases with known treatments that have led to health benefits and for which more detailed analyses could be done matching quality to costs. For the person-based regression approach, the authors were able to use all listed diagnoses on claims in a given year. For the encounter approach, an algo-
OCR for page 23
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop rithm was used to determine the diagnosis (usually the first listed) to which the majority of spending went, and the dollars were assigned to that category. For the episode-based approach, each episode treatment group (ETG) was allocated to the clinical group that accounted for the largest share of spending. There were a few problems with the ETG approach. For example, there was no ETG for cervical cancer; also, the transparency issue remained—the method of aggregating data into the ETGs was still essentially a black box. Under the encounter approach, about 19 percent of the spending recorded from claims had no listed diagnosis, and the dollars could not be allocated to a condition. Using the episode-based approach, only 1 percent of spending originated from claims with no ETG. Using the person-based approach, expenditures for individuals with no diagnoses accounted for only about 0.6 percent of the total. The problem of unlinkable spending was clearly most serious with the encounter-based approach. The Rosen-Cutler work demonstrates that the cost of illness can be estimated by each of the proposed methods. The total dollars that can be allocated differs, and, certainly, the fact that noncomparable data sets are being used for the different methods also has an impact on the results. Table 2.4 shows annual spending by condition estimates. The spending estimates varied significantly for some disease categories. For example, the person-based approach yielded very high annual expenditures for dementia—on the order of $9,000—relative to the encounter- and the episode-based approaches. The likely reason is that the regression used does not include all of the needed interaction terms, so the estimates essentially capture unobserved correlates of spending. Instead of getting just spending on dementia, the coefficient is picking up aspiration pneumonias, feeding tube treatment, and other things for which clinically meaningful buckets need to be created. This, Rosen emphasized, is why it is important to bring clinical insight into analysis. For some of the same disease categories, the encounter-based approach appears to underestimate expenditures. Rosen noted that this may have something to do with the way risk factors for diseases are commonly coded. For example, physicians may be more likely to code coronary heart disease than they are diabetes, hypertension, or hyperlipidemia. In contrast to the encounter-based approach, which relies entirely on physician coding on claims, one nice feature of the person-based approach is that coding can be captured over time, so more information about multiple conditions can be obtained; the approach also allows the claims data to be supplemented with surveys, injecting information from patients that can enrich the picture. Rosen also noted that they have not yet done any time series using MEPS data. Making some direct comparisons, they have found some of the same things—for example, dementia and acute renal failure, which tend to occur in patients being treated for other conditions simultaneously, end up being much higher with the person-based regression approach than the others.
OCR for page 24
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop TABLE 2.4 Annual Per-Person Cost for Selected Diseases by Method, 2003 Disease Encounter Episode Person Colon cancer $8,100 $4,458 $10,475 Lung cancer 12,082 14,213 23,895 Dementia 596 1,111 9,231 Depression and bipolar disease 616 984 1,070 Hypertension 225 522 376 Coronary atherosclerosis disease 3,415 4,342 3,303 Congestive heart failure 2,869 2,476 12,645 Cerebrovascular disease 2,563 2,818 5,759 Asthma 348 639 519 Chronic renal failure 11,105 11,433 11,964 Osteoarthritis 1,184 1,726 1,450 SOURCE: Workshop presentation by Allison Rosen. At this point, criteria for ranking the suitability and accuracy of the different methods have not been fully sorted out; Rosen noted that the right answer has a lot to do with the specific applications. She also encouraged others in the community to provide feedback on the topic. If the best unit of measurement is the episode, whether defined by an existing grouper or in some other way, it is important to proceed so that the underlying approach that would be used by government agencies is transparent. During open discussion of the different methods for allocating expenditures across conditions, Steven Cohen (AHRQ) pointed out that the Centers for Disease Control and Prevention, AHRQ, the National Pharmaceutical Research Council, and the Medicare chronic disease directors have developed a chronic disease cross-calculator for the attribution of costs across different conditions. He applauded the work by the Cutler-Rosen group and others at the workshop to help understand the nuances in terms of the different methodologies. Cohen noted that the opportunity to look at distributional aspects in the costs of chronic illness treatment is incredibly useful to his agency; having information about the concentration of expenditures and the characteristics of treatments will be particularly applicable for estimating the impact of preventing or reducing the incidence of some of these conditions—both in terms of valuing health care and monitoring potential savings. The metric might not be cost savings, but better value for the dollar. Mark Freeland (CMS) added that it was exceedingly valuable for his agency to see how different the results can be and that there is some purpose—price index construction, national benefit cost analysis, the national income and product accounts, etc.—for which each of the constructs may be the best. Likewise, Jack Triplett was encouraged by the work to improve data and
OCR for page 25
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop methods for allocating nominal medical care expenditures, noting the large amount of progress that has been made over the past 10-20 years. In the past, efforts to disaggregate the national health accounts into disease categories have not produced any time series data. When Triplett estimated a real expenditure account for mental health a decade ago, an enormous amount of work was necessary to reconcile an existing series of annual cross-section estimates because, in their construction, no attention had been paid to making new estimates compatible with those of earlier years. Now, not only is work progressing on time-series data, but also several alternative methods are being discussed; Triplett cited this as a big step forward. Shapiro endorsed BEA’s cautious approach. He suggested that the agency should probably not buy into a particular approach too early, adding that it would be useful to see whether the choice actually matters for the statistics. This means that there might need to be parallel sets of accounts going on, at least on a research basis, for some time. 2.4. DATA NEEDS FOR EXPENDITURE ACCOUNTING Workshop participants touched on data needs at many points during the day’s proceedings, and that discussion is sprinkled throughout this report. This section summarizes a few of the key data themes that emerged. Drawing from Multiple Sources Dale Jorgenson emphasized the need to consider a wide range of possible data sources on medical treatments to underlie the satellite account. The country’s health system is characterized by a lot of patient and treatment heterogeneity, and it is not easy to collect this information, especially at the level of detail needed to capture the rapid evolution of the character of medical treatments. For measuring the quantity and price of these treatments, one can look to providers, customers, and third-party payers. Service provider data are the starting point for the producer price index at the Bureau of Labor Statistics (BLS), and Jorgenson expressed the view that this type of information is going to have to play a role for the BEA work, particularly on the industry side of the accounts. Another source from which information can be collected is the customer. Jorgenson noted that this is where the distinctive features of the medical system really come into play. Unlike some other areas of the economy, it is difficult to get accurate and relevant information from the consumer (patient). Patients pay a relatively modest portion—some information can be collected from the patient about the episode and the treatment, but when dollar figures are needed, one has to go to the provider. The final source is the third-party payer, which plays a big role in the strategy laid out by BEA. A vast amount of claims data are maintained by both private and public payers.
OCR for page 26
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop Ideally, these data would be matched against the payments that are received by the providers. Jorgenson commented that he did not think it would be fruitful to look at the data issue as an either/or proposition. In the medical panel data, the MEPS survey, an attempt has been made to combine the information collected from providers with information collected from payers. This, he said, is a good practical approach and one that should be encouraged. His recommendation to BEA and BLS is that they should coordinate their efforts to get the most reliable information on prices with data from the payers (i.e., claims) and also from the providers. In thinking about measuring prices and quantities, the relevant data that can be brought to bear will have to include the sort that the producer price index already collects. Aizcorbe agreed that all data sources that could usefully supply the accounting system should be considered. She did, however, raise one deterrent to extensively using data from providers, which is how to link expenditures to patients. Using the example of depression, she questioned how the different care elements—the doctor visit, the drug purchase, and the talk therapy—could be linked together for the same patient. So far, in the satellite program, BEA has not used the raw data that BLS uses for its indexes exactly for this reason—they cannot be linked to patients. The MEPS data do link to patients, which is what BEA needs for at least a sample of the population. For most of the data approaches being considered by BEA, the treatment is what needs to be priced; this requires data at the patient level reflecting the full combination of inputs into the treatment. Jorgenson made the point that a solution will still need to be found for combining data on providers’ prices with the information collected from claims, and he suggested that kind of work be put on the table for BEA. Aizcorbe agreed, remarking that it is important for the agency to think about what it will be doing 10 years from now; it is not obvious yet how to take the data that underlie the price programs at BLS and use them directly for BEA’s purposes. Another aspect of the data coordination task involves reconciling the microdata in MEPS with the national health expenditure accounts, because they don’t add up to the same national totals. That is primarily because the scope of the populations and of the spending are not quite the same, a situation that calls for regular updating of the reconciliation work done by AHRQ and CMS. Aizcorbe reiterated the importance of working out how to coordinate and exploit multiple data sources, as well as anticipating how this strategy will play out once the research program is in full swing. Sample Size: Capturing Information on Less Common Conditions and Morbidities David Cutler pointed out that, when relying on survey data sources such as MEPS, the key challenge is that they do not include enough patients to capture
OCR for page 27
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop rare diseases or less common comorbidity combinations. For example, not many lung transplants are going to appear in MEPS, which in 2005 had a sample size of 15,000 families and 39,000 individuals. Cutler suggested that 30 million person records would be needed to cover the full range of diseases and combinations thereof. Since such a massive expansion of survey sources is impractical, a next best option may be Steven Cohen’s suggestion to oversample rare diseases.2 Several workshop participants also suggested that, ideally, any results should be stratified to account for different spending profiles by age for certain diseases. Regarding other population coverage gaps, Constance Citro made the point that the American Community Survey may offer an opportunity to cover some institutionalized populations (through its group quarter sample) omitted from the scope of MEPS. Cohen added that CMS conducts the Medicare Current Beneficiary Survey, which captures a fairly large segment of the institutionalized population. He reported that his department’s data council has been thinking broadly about where the gaps are for other individuals in long-term facilities. Aizcorbe acknowledged the value of these ideas, noting that BEA (as well as Cutler-Rosen) are looking into some of them already. A lot could be done with claims data for the insured population simply because of their enormous size and coverage. BEA participants agreed that using MEPS as the backbone of the data infrastructure, and then claims information in a supplemental role wherever gaps appear, was a reasonable strategy. Even so, treating different comorbidities as separate disease categories—which Aizcorbe agreed was a good idea and would have to be done to some (as now unknown) extent—still runs up against data inadequacies. Even with the largest data sources, a portion of spending occurs in buckets that have very few observations, and creating separate categories for comorbidities still does not always work. This is why medical expertise is needed when setting up the account structure. Data Representativeness The tradeoff between sample size and representativeness is one data issue to which workshop participants returned on several occasions. The work by the Cutler-Rosen group has relied heavily on microdata from national surveys that sample individuals, such as MEPS, supplemented with the Medicare Current Beneficiary Survey. As noted above, while the survey data are essential to the accounting project—and very useful for high-prevalence conditions, particularly the cardiovascular disease and cardiovascular disease risk factors—there are real sample size inadequacies for conditions with lower prevalence. In contrast, the insurance claims data provide a large sample, but at the expense of representativeness—no single source provides a national sample. It is easiest to 2 This then begs the question of how to find these people; MEPS uses a sampling frame built from the National Health Information Survey, which records only self-reported diseases.
OCR for page 28
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop find data on people in large firms with standard kinds of benefits. However, there are no large samples of uninsured people; for this group, something like MEPS will inevitably have to be more heavily relied on. Newhouse added that the transaction prices for the uninsured are also very complicated. The hospital trying to collect on a debt may turn it over to a collection agency and may agree to some payout period that stretches over months if not years. Another problem with existing claims data sources, which BEA is struggling to get a handle on, is that they typically track patients only as long as they are covered by a particular plan. So changes in employment could lead to discontinuities in the data. If a person switches jobs—and even if both the old and new insurance plans are in BEA’s database—it may not be possible to connect the records in a way that ensures that one is dealing with the same patient. This may not be particularly worrisome if changing jobs or plans is not highly correlated with disease incidence or conditions. However, Aizcorbe pointed out instances for which that may not be the case. People may select less expensive plans, such as health maintenance organizations (HMOs), until their situation changes; for example, if a woman becomes pregnant, who the plan provider is may suddenly become important, and she may switch to a different kind of plan. At this point, it is difficult to know exactly what the optimal balance will be in terms of how to utilize the different kinds of data. It is clear, however, that the satellite accounts will need to draw from many data sources and methods will need to be developed to coordinate them. Combining the survey sources with Medicare records can reach a significant share of the population, but it is unlikely that data will be comprehensive to the point of providing a picture for a group of the population that is completely random any time soon. Commenting that systematically missing data coverage is especially worrisome, Aizcorbe noted that statistical methods can be used in the accounts to minimize some of these problems. For example, weighting is used to correct for the fact that the annual Survey of Manufacturers disproportionately samples large firms (those with more than 5,000 employees). She also noted that, even if there were enough of every different type of patient and plan in the data, reweighting would be needed to make it align with the sampling frame of MEPS or some other national surveys. Cohen reaffirmed comments by Cutler and Aizcorbe about the need to ensure the national representation of data sources. He pointed out that, for many disease areas, expenditures are highly concentrated. Some of the uninsured, for example, are in that predicament because they have chronic diseases; as a result, data could be highly skewed in terms of the segment that is missing. He added that communication among BLS, BEA, and the Department of Health and Human Services is essential as the agencies think long term about needs and potential oversampling strategies to fill gaps in a much more efficient manner. Given the limitation of departmental resources for surveys, Cohen noted the importance of opportunities to link MEPS to the National Health Interview Survey to create ways of predicting the likelihood of an individual being uninsured in the long term; subsequently,
OCR for page 29
Strategies for a BEA Satellite Health Care Account: Summary of a Workshop high probability portions of the population could be oversampled. He added that, if resources permitted, it may be possible to look at specific chronic diseases in which there are well-developed evidence-based processes of care. Even with fixed resources, there may be ways of differentially sampling the population in order to meet both departmental objectives and to help inform policies and programs at other agencies. Cohen reported that conversations have already begun taking place between BEA and BLS about how to help meet the needs that BEA has on the spending side versus the needs that arise on the industry side. Looking down the road, the question of how big a hindrance to health accounting the lack of data representativeness will be is a major one. For research purposes, if partial pictures can usefully be explored, it is less of a problem. For the national accounts, which must be complete and national in scope, the problem is more severe and may require short-term compromises. For something like measuring quality change of treatments (discussed in Chapter 3), the satellite account methodology may have to rely on inferences based on more common diseases, at least for a while; this would seem better than no quality adjustment at all.