The first session of the workshop was designed to provide background on the concept of measuring medical care economic risk (MCER): What is the concept? Why is it needed? What are the criteria for a useful measure? How does one measure adverse medical event risk? How does one measure subsequent economic hardship? What are the pros and cons of prospective and retrospective measures?
The session focused on a paper prepared for the workshop on key issues in the conceptualization and measurement of MCER by Sarah Meier and Barbara Wolfe (see Part III of this volume). In their presentation, Meier and Wolfe discussed the need for a measure of MCER, issues in developing a measure, criteria for designing it, and their approach to developing the measure. Three discussants then offered remarks, followed by floor discussion and comments.
MEASURING MCER: WHY, WHAT, AND HOW
Barbara Wolfe (University of Wisconsin–Madison) began by noting the importance of measuring MCER in the measurement of poverty. One reason to seriously consider a separate MCER is to increase the accuracy of the measurement of poverty. MCER is a separate concept or component of poverty—and a complicated one. It is essential to measure those who are deprived in terms of needing medical care in order to accurately capture those who are poor in a broad sense.
Another reason to include MCER, she said, is its increasing importance over the years. Spending on medical care as a percentage of gross domes-
tic product has grown substantially since poverty measurement began. In 1965, it was about 5 percent of U.S. gross domestic product (Congressional Budget Office, 2008:3). In 2010 this percentage grew to about 17.6 percent, and it is projected to grow to about 20 percent by 2020 (Centers for Medicare & Medicaid Services, Office of the Actuary, 2011:1). Therefore, in terms of thinking about individuals’ well-being, capturing MCER has become increasingly important.
A third reason Wolfe mentioned to measure MCER is that it is sensitive to public policies that influence medical care coverage, such as Medicaid, Medicare, and provisions of the 2010 Patient Protection and Affordable Care Act (ACA). MCER is a way to capture the success or failure of programs in terms of their effects on both risk and targeting to those with low incomes. It will be ever more important, she said, as the nation moves into the era of expanding coverage under the ACA.
Difference Between MCER and Medical Out-of-Pocket Expenditures
Medical out-of-pocket expenditures are essentially contributions to health insurance premiums plus out-of-pocket expenditures for expenses not covered by insurance. Medical out-of-pocket is an ex post concept: it refers to expenditures—that is, actual utilization rather than risk of need. It is closer to what one thinks about in terms of traditional poverty measurement.
When people need medical care but do not have coverage or have limited coverage and limited income, they tend to underutilize care. When only actual medical out-of-pocket is included in a measure, underutilization is missed. If access to care is increased, leading to more coverage, people get recommended tests and treatments, and in fact their health is improving. In this situation, if the measure captures only medical out-of-pocket, then one may well think that an individual’s health is declining because of the additional tests and treatments, believing that they are more vulnerable because their medical out-of-pocket has gone up, even though their health has improved.
Therefore, an out-of-pocket expenditure measure that does not adjust for underutilization is not very satisfactory, either in terms of thinking about people’s real underlying needs or the success of programs. It gives the wrong picture for both. Most important, or most essential, it does not capture risk. It does not fit with the whole concept of health insurance, yet it is tied to measuring premiums and types of coverage. So if one thinks about the concept of risk, medical out-of-pocket is not a consistent or reliable measure.
Key Issues in Developing an MCER Index
Wolfe next discussed a number of major questions that need to be addressed in developing an MCER index, with the authors’ position on each one:
- Should the index be an ex ante (prospective) or ex post (retrospective) measure? Wolfe emphasized that risk is inherently a prospective concept.
- Should the core unit be the individual or the family? Medical care economic risk is an individual concept, even though one’s insurance coverage may be at either the individual or the family level. One needs to start at the individual level and then aggregate up to the family level.
- How should underutilization or overutilization be treated? If people have particularly limited or generous coverage, should adjustments be made for both under- and overutilization? It is important to adjust particularly for underutilization, but both need to be kept in mind.
- Should the index be tied to a specific benefit package? Should the measure take a benefit package as is, whether it is a generous package or a very limited one? Or should the measure be based on a specific benefit package and a specified set of benefits in terms of medical risk? Wolfe said their answer is to work with a specific package, although the dimensions of the package have to be explicitly defined, including the extent of its generosity.
- Can such a measure be developed in a relatively short period of time? It is not particularly interesting to have an MCER index that measures risk from 5 years ago. If it takes that long, it is old information. Eventually there will be a trend, but it will not tell much about the success of policies today. Therefore, it is important to have a measure that is relatively easy to develop in a short period of time. The issue of data requirements comes up in this context.
- How complex should the measure be? One can get a more accurate medical care risk index for an individual if more information is obtained and used. But the cost of using more information is that developing the index is likely to take longer. So there is clearly a trade-off between complexity and pragmatic aspects in terms of speed and ease of creating the index.
- How should resources be captured, both in terms of the detail of coverage of individuals and their income? One aspect of this question is that people frequently use their assets, not just their current income flow, to cover medical care expenses. The issue is not so
much the concept of including assets, but, with available data sets, how to possibly move forward and include something about assets. This may be an issue that is too complex to handle.
- How should extreme risks—that is, “upper tail” expenditures— be captured? Risk includes the risks at the upper tail and some individuals facing very high, catastrophic risks. Should that be the focus, at what level should that be set, and how should it be incorporated across members in a family?
Wolfe closed her remarks by summarizing the treatment of medical care needs in poverty measurement. In the original official poverty measure, there is an implicit inclusion of some medical out-of-pocket expenditures, but it does not capture variability—it is a very simple measure. The Supplemental Poverty Measure (SPM) subtracts medical out-of-pocket expenditures from the calculation of family-level resources, but it does not include the value of insurance benefit(s) in resources and does not incorporate variability in medical care needs in the thresholds. The 1995 National Research Council (NRC) report also recommended that appropriate agencies should work to develop one or more “medical care risk” indexes that measure the economic risk to families and individuals of having no or inadequate health insurance coverage. However, the report recommended that such indexes should be kept separate from the measure of economic poverty (1995:225).
Why a Separate Index?
Sarah Meier (University of Wisconsin-Madison) next discussed the need for developing a separate index, reviewed the literature on medical care risk indexes created by a number of people, reviewed various design considerations in developing an MCER index, and suggested a framework for developing this index. She also pointed out some of the limitations and challenges that must be addressed.
She noted that the incorporation of medical care need and resource considerations was problematic in the development of the SPM for two reasons. The first is the nonfungible nature of medical benefits, and the second is difficulties in defining and calculating medical need. Specifically, there is a limited ability to predict future expenditures at the individual level, resulting in misclassification. Perhaps with enough data, expenditures at a group level could be predicted accurately. But what specific expenditures an individual will have in a future period is something that cannot be predicted with any great accuracy. Even if one comes up with a reasonable prediction of expenditures that an individual will experience, variations in this predicted measure might necessitate a number of thresholds.
Developing two separate measures may resolve the problem that medical care insurance benefits are not fungible—that is, they cannot be used for other necessary expenses, such as housing and food. But this approach really pushes many of the difficulties presented by the measurement of medical care need into the domain of the second measure.
Review of Existing Measurement Strategies
Before identifying strategies for moving forward, Meier briefly reviewed some of the work done to date toward development of a measure of MCER. Two analyses warrant particular emphasis: the first is a 1995 analysis by Short and Banthin, which estimates underinsurance among privately insured adults under age 65. It focuses on economic circumstances in the case of a catastrophic event, in which individuals are assigned to a risk group on the basis of expected expenditures and a catastrophic event is defined for each risk group. An individual is underinsured if the catastrophic event exceeds 10 percent of income.
A second analysis, by Banthin and Bernard (2006), also examines insurance adequacy. It covers the broader population, including the underinsured among both publicly and privately insured. However, this analysis focuses on actual medical expenditures over 10 and 20 percent of family income, an ex post concept. So it omits the risk aspect that we are interested in talking about today.
Another measurement strategy is the empirical model developed by Handel (2010). Meier observed that because a major objective of the framework she and Wolfe suggest is to identify an empirical strategy that enables more robust treatment of medical care risk and its implications, they relied on Handel’s model to develop an MCER-relevant strategy to model and quantify risk.
Briefly, Handel’s method takes a base sample of individuals and applies their claims information into a risk adjustment software model— specifically, the Johns Hopkins Adjusted Clinical Groups (ACG) Case-Mix System model. Based on claims experience—essentially prior diagnosis information—and an individual’s demographic characteristics (age and gender), this software comes up with a risk score that is an indicator of the relative risk of individuals if one compares their scores.
Individuals are assigned to a risk cell for each claim type (four categories, including pharmaceutical claims, mental illness–related claims, physician claims, and hospital claims). Each cell includes similarly risky individuals as determined by the Johns Hopkins ACG software. Taking each claim type separately and the risk cells within the claim type, expenditure distributions are fit to the risk cell/claim type combinations, using actual claims experience.
Each individual is assigned a joint claims distribution based on his or her risk profile (risk cell membership for each claim type) and the respective estimated distribution. This joint claims distribution can be mapped to a distribution of out-of-pocket expenditures, which applies the individual’s insurance characteristics. Family-level distributions of out-of-pocket expenditures can be formed by aggregating individual distributions and coverage characteristics.
Steps and Criteria for an MCER
Meier outlined three steps for implementation of the suggested framework for an MCER measure:
- Baseline measurement of medical expenditure risk at the individual level.
- Adjustment of individual expenditure risk for risk protection (insurance), followed by aggregation of individual risk measures to form a family-level measure of medical care expenditure risk.
- Measurement of family economic resources, which preferably would include an annuitized value of financial assets. This process would conclude with an examination of the relative affordability of a family’s premium costs and its medical expenditure risk, given its economic baseline.
Prior to detailing the specifics of each step (see Meier and Wolfe in Part III of this volume), Meier highlighted some of the important criteria for the design of a measure of MCER that have been specified in prior literature. As stated previously, the 1995 NRC report recommended a prospective measure of medical expenditure risk, as well as a family-level measure, using the official poverty measure or SPM definition of family.
After the panel’s report was published, a paper by Doyle (1997) outlined additional important criteria for a well-designed MCER index:
- The index must reflect risk;
- It must reflect resources and medical need (insurance adequacy, subsidized care, and affordability);
- It must be quantifiable;
- It requires a well-defined accounting period; and
- It must be defined by available data.
Meier went on to identify seven design considerations as particularly relevant to the framework she and Wolfe suggest.
Individual Health Risk Classification—basically, the selection of risk factors that would be used to assess risk. This would entail identifying a number of factors that would be predictive of higher, or perhaps lower, next period expenditures. Characteristics that are highly predictive of these expenditures need to be selected. Standard characteristics often selected are age and gender but could also include chronic conditions, disability, and others with high predictive capacity. The problem is that the best predictive model may be extremely complex and include variables that one might not have. Certain considerations need to be introduced when coming up with this classification. These include data limitations, such as availability of a relevant variable and number of observations per risk cell. Also, feasibility considerations, such as complexity, timeliness, and cost, would be important.
The Definition of Appropriate Medical Care Coverage—Expenditure risk should be defined under a standard minimum basket of medical care services. A starting point for this could be the benefits standard introduced under the ACA. Once the decision is made to select a minimum basket of medical care services, the next step is to adjust for over- or underutilization observed in the base data set.
Selection of a Meaningful Risk Measure—The issue here is how to move from a range of possible outcomes that an individual could have in the next period to a singular measure of economic impact. Meier suggested two potential measures: (1) a measure that reflects a probability of expenditures exceeding an affordability threshold or (2) a measure that would be based on expected expenditures per family unit, in which expected expenditures are conditional on the risk characteristics of unit members.
Modeling Expenditures—Going hand in hand with the selection of a risk measure, the developers of an index will need to come up with the best approach for modeling expenditures. Two approaches might be considered. The first is the formation of mutually exclusive risk cells and then moving to fit loss distributions, and the second is a regression-based method.
Assessing Risk Protection (insurance)—In addition, developers will need to determine how to best assess the risk protection afforded by insurance. Meier suggested individual-level assessment followed by family-level aggregation. At a minimum, this assessment should include information on deductibles and an out-of-pocket maximum, but ideally it might also include coinsurance and copayments or some measure of actuarial value.
Measuring Family Resources—Family resources can be measured using an income definition consistent with the official poverty measure, or the SPM, plus consideration of assets. To account for assets, an annuitized value could be constructed whereby a family is projected to receive the value of an annual flow of income from its financial assets based on the life expectancy of adults in the family, using existing life tables. This annuitized
value would then be added to income and compared with unprotected expenditure risk, which is the risk remaining after the effects of insurance are factored in.
The Definition of Affordability—Affordability is another very difficult concept. It can be defined as a family’s risk of exceeding an affordability threshold. In this respect, the threshold could be defined as a percentage of family income, which includes the annuitized value of assets. The definition of an affordability threshold should probably vary by the family’s relative resource level—that is, its resources relative to the level of income required to cover basic needs under the SPM and the official poverty measure. At least as a starting point, the affordability thresholds outlined in the ACA could be considered.
One of the major questions that will go hand in hand with the design of a measure of MCER is whether any risk, no matter how small, of a catastrophic outcome places a family at economic risk. Basically, every family is going to be at some small amount of risk of experiencing a catastrophic outcome, and there are some difficulties with determining how to treat that.
Meier emphasized that it is important to reach consensus on the conceptualization and measurement of medical care expenditure risk at an early stage of development. In particular, agreement is required on a minimum benefit standard, a concrete definition of affordability (that is, what percentage of income will be the threshold and whether annuitized assets are to be included), whether there are going to be adjustments for underutilization, and data considerations, such as the collection or construction of new variables.
Meier reiterated that MCER is an increasingly important component of poverty. Risk is prospective and individual but can be aggregated to the family or household level. The approach she and Wolfe favor is that the MCER measure be developed as a separate index. However, she noted, there is potential for incorporation into a single measure in the future. She reiterated that an MCER index is distinct from medical out-of-pocket expenditures because it captures the risk component, which is critically important.
She said that although she and Wolfe have outlined a basic framework for the development of the index, its definition and construction is a very complex process. There are many normative considerations in the design of an index. A well-formed measure requires attention to numerous methodological details. Several areas will require particular focus in future work.
Pamela Short (Pennsylvania State University), session discussant, highlighted her thoughts on key concepts for measuring MCER. In her opinion, adhering to these concepts could make the job of measuring that risk easier, rather than harder.
She observed that there are basically two key reasons for measuring MCER separately from the general measure of poverty. First, the purpose of most medical care spending is to get people back to where they were in terms of well-being before a health loss. In keeping with this view, the SPM views medical care strictly as getting in the way of basic consumption. The implicit presumption is that medical care spending does not benefit consumers. The second reason for measuring MCER separately is the large random component involved in medical care spending. There is also a predictable component, which is reflected in adjustments in the SPM. One of the distinguishing characteristics of medical care is that the random component is very large and, for people who actually do get sick, it leaves them in a deprived state compared with other consumption needs.
The stated goal is to measure MCER. Risk is unpredictable, random variation in expenses. In their presentation, Meier and Wolfe have certainly defined risk in these terms, she said. Risk is not to be confused with predictable differences in average expenses. For example, older and sicker people will spend more on average than younger, healthier people. That is an issue in terms of well-being that can be realized with any given amount of resources, but it is not the risk the workshop is concerned with today. Also, risk should not be confused with actual expenditures. Conceptually, risk is an unknown that cannot be measured retrospectively. For example, two uninsured people are both equally at risk, even if only one of them gets sick. So risk has to be measured prospectively.
Another point is that premiums should not be confused with out-of-pocket medical expenses. The key here is that insurance turns random, variable medical expenses into regular, predictable insurance expenses. Insurance premiums are actually a lot more like spending on food, shelter, clothing, and utilities (the basic needs in the SPM) than out-of-pocket medical care spending. So, when one thinks about medical expenses in these terms, perhaps the ACA is a game changer for incorporating medical expenses into poverty measurement. If all are required to buy insurance, then the key question is how much insurance is enough to protect consumption related to basic needs other than medical care.
Where one goes with this depends a little on the planned use of the MCER index. If the index is to be used as a measure of needed income or how well needs are met by a family’s income, then a family with insurance does not need lots of income to cover uninsured, out-of-pocket expenses.
However, families that are uninsured are at risk for serious deprivation without lots of income to cover the cost of an expensive illness.
If the purpose of the index is to grapple with identifying people who are underinsured and, conversely, to ask the question, “How much insurance is enough?” then the correct way of conceptualizing the question is in terms of the amount of insurance needed to avoid deprivation in the face of a random catastrophic illness or accident. Thinking in those terms, the medical risk component does intertwine with the nonmedical component of the SPM, if one is going to be internally consistent in the measuring systems. People who are adequately insured should have enough insurance to guarantee a minimum level of nonmedical consumption in any state of health—as established by the nonmedical poverty index, the SPM.
The nonmedical consumption threshold, therefore, should include the premiums needed to buy adequate insurance. After 2014, those premiums will vary by age but not by health status. In addition, to be totally consistent, nonmedical resources should not include actual spending under Medicaid and Medicare, but the actuarial value of those programs on average (comparable to a premium).
Short explained that the SPM currently reduces resources by out-of-pocket premiums and out-of-pocket medical expenses. But a possible modification would reduce resources by the out-of-pocket premiums needed to buy an adequate amount of insurance. The right amount of insurance would probably vary by people’s income and circumstances (e.g., household size and age composition). Then, at least conceptually, if people had enough insurance as measured against nonmedical consumption needs, their out-of-pocket medical expenses could be ignored. By construction, they should be able to handle any expenses not covered by the minimally adequate insurance plan.
Thus, one gets to a kind of two-part index or two-part measurement of poverty, she observed. This two-part approach decomposes family medical expenses into predictable family spending on insurance and a random out-of-pocket component, but with a need standard for the first component that reduces and limits the second component. The nonmedical poverty index, the new SPM, would incorporate enough income to buy enough insurance. The second index—on which this workshop is focusing today— would quantify the risk of being poor, as defined by the first index, because of inadequate insurance for the out-of-pocket component. As Meier and Wolfe suggest, Short concluded, that could be the probability of falling into poverty—the probability that high out-of-pocket medical expenses would put a family below the poverty threshold.
Gary Burtless (Brookings Institution), the second session discussant, began by stating that the perspective he brings to this discussion is that of
someone who is interested in knowing how one might distinguish between the poor and the nonpoor in the data sets available.
Suppose that most necessities can be purchased either with dollars or with something almost indistinguishable from dollars, for example, food stamps. If everything can be bought with dollars, the 1995 NRC panel proposed a complicated but in principle straightforward way to derive poverty thresholds to identify the specific number of dollars U.S. families need to cover necessities. The panel could not extend this conceptual method to cover the dollars needed to buy necessary medical care, yet they thought that was a necessity also.
Medical care can be purchased with dollars, Burtless observed, but the price of some common kinds of care is so extraordinarily high that few people actually buy it with ordinary dollars. They rely on their health insurance to reimburse the cost of essential care, especially if it is expensive. What about the people who lack health insurance? Or those who must obtain a lot of care that is not covered by their insurance? Or those who are completely uninsured but are in robust good health and are very unlikely to suffer any serious health catastrophe over the next year?
The 1995 NRC panel’s solution to this problem was to suggest that family resources be calculated by subtracting the amount of money a family spent on health insurance premiums and out-of-pocket medical expenses— or the predicted amount of money it was expected to spend—from the other resources it had to pay for necessities. If family resources minus out-of-pocket medical expenses were too little to cover the cost of these nonmedical necessities in the thresholds, then the family was to be classified as poor.
Burtless said that many social scientists were dissatisfied with this solution. Based on their paper, he thought that Meier and Wolfe also appear dissatisfied with it. A problem they mention is that actual out-of-pocket spending may not reflect the necessary or recommended level of care (or spending) given family members’ health conditions. In other words, families in strained financial circumstances may forgo necessary or recommended care because they cannot afford it. If the NRC panel’s recommendation is followed, given the health condition of the members of the family, too few dollars will be subtracted from the family’s resources to reflect the necessary care it ought to receive if it is to be classified as nonpoor.
He further stated that he is troubled by a recommendation that would disproportionately lift the poverty rate of groups in the population that have been the direct beneficiaries of so much technical progress and taxpayer assistance to pay for the care—the aged, the disabled, the nondisabled adults and children who benefit under costly public health insurance programs. It seems paradoxical to him that so many public resources are provided to health-challenged and low-income people, yet the proposed
poverty measure recommended by the NRC panel leads to the inference that these populations are relatively poorer than they were in the past when remedies for their infirmities were less successful and insurance options available to them were much less generous.
Burtless observed that officials in the Census Bureau or the Agency for Healthcare Research and Quality charged with collecting and organizing data will be looking for clear and specific guidance on how to take this medical risk recommendation and on how to distinguish people who are poor from those who are nonpoor. He remarked that he is not sure they will find all of the specific guidance that they might be looking for in the Meier and Wolfe paper. The authors have offered judicious views on a lot of the issues, but readers looking for a clear blueprint will not find it.
A given person facing individual-level medical spending risks can be described by a number of characteristics that may help predict the health spending risks he or she will face. Before any actual spending is observed, all one has is a set of indicators about the person to help in predicting the probability distribution of spending amounts over the next year. With good information about the person’s current private insurance or potential eligibility for public health insurance in the event of a health crisis, it may be possible to devise a probability distribution of the person’s net spending— on insurance premiums and copayments—after reimbursement is received.
The closer one comes to assessing the risks facing a particular person, the wider the predictable inequality of risks across people. The broader and more inclusive the group over which risks are assessed, the less the inequality in risks across people. As the risk measure is estimated over progressively narrower groups, the variation in the probability distribution of their medical spending risk widens.
“Adequate insurance” is the answer to the problem of paying for necessities that have a very wide distribution of required but unanticipatable cost. How does one assess insurance adequacy? The authors offer very helpful comments but perhaps not enough concrete guidance.
The concept of adequate insurance is clouded still further by the fact that public insurance provides complete coverage to specified kinds of care when eligibility is linked specifically to household income. Imagine a 30-year-old woman who is uninsured and has a modest income. If this person experiences a mild health setback, she is going to have to pay for the entire cost out-of-pocket if she is uninsured.
However, if her crisis is more costly, if the health episode is much worse, she may qualify for free or heavily subsidized public health insurance. It will reimburse most of her medical expenses, if not all of the necessary ones. And this is particularly true if the medical episode causes her to lose some or all of her income, bringing her income below the income eligibility limit for Medicaid. In states where the income eligibility limit is
above the poverty threshold, there is in fact a public program, Medicaid, that is supposed to ensure that a person with big medical bills will not be forced into income poverty as a result of the inability to pay critical health bills.
Burtless indicated that in this case he is not sure he understands how Meier and Wolfe’s suggestions ought to guide the Census Bureau or any other poverty definer. Is this woman at risk of being poor? Or is the public program out there in fact ensuring that she will not become poor as a result of these emergency medical expenses?
He next listed a set of questions, answers to which would help influence the classification of who is poor and who is not poor:
First, what is the authors’ preferred method of distinguishing poor from nonpoor people, taking account of their varying medical risks?
Second, what is the authors’ preferred method of measuring medical risk? In particular, how specifically would they go about gathering needed data and assembling the data into a workable index? How narrowly, for example, would they define the population cells over which risk is measured?
Using their preferred method of assessing medical risk, or some other method that they might be willing to accept as a second-best alternative, how would they then use data from the Medical Care Expenditure Panel Survey, the Current Population Survey, the Survey of Income and Program Participation, or some other data set to place given survey respondents into the risk cells that they recommend, and then determine whether that person is a member of a poor family or a nonpoor family? What he is looking for is concrete and specific guidance about how to actually implement their preferred methods, not just a discussion of general principles.
If Meier and Wolfe think their preferred strategy will be rejected as unfeasible or politically unpopular or excessively costly, what do they think is a second-best strategy? And what is a third-best strategy if even second-best is unpopular or unfeasible?
When he finished reading the 1995 NRC panel report, he said, he did not fully agree with its recommendations with regard to handling health insurance and required health care expenses. Nonetheless, he had a concrete idea of how he would proceed and how the Census Bureau would go about estimating poverty under that definition. In conclusion, he reiterated that he wanted to know the bottom line for poverty measurement of Meier and Wolfe’s proposals.
Richard Bavier (U.S. Office of Management and Budget, retired), the third session discussant, began by directing attention to two poorly supported assumptions, which the 1995 NRC panel needed to justify in order to leave health care out of the poverty thresholds in the first place and which now entangle efforts to design a complementary measure of MCER.
The first assumption was criticized by John Cogan, a member of the 1995 panel, who dissented from the report (National Research Council, 1995:Appendix A). He criticized the panel’s assumption that all medical out-of-pocket spending is necessary, citing research that health care is an economic good for which spending varies with income and price. In the body of the report, the panel acknowledged that some health spending may be discretionary but suggested that medical spending by families with limited economic resources is probably all nondiscretionary. That was an assertion rather than an argument, Bavier said. He questioned whether research in the field of health economics over the past 15 years would support this assertion. For example, he asked, are Medigap premiums to obtain first-dollar coverage nondiscretionary?
Second, even household medical expenditures that are nondiscretionary still may not reduce current income available for food, clothing, shelter, and a little more—although the 1995 panel, and now the SPM, assume that they do. Some health care spending, especially involving the unusually high costs the panel was concerned about, may be funded through liquidation of assets or borrowing rather than out of current income. It seems reasonable to expect that, if they could, families would employ assets or borrow to cover high health care costs before they would allow consumption of basics to fall below necessary levels. The NRC panel correctly judged that including wealth in an operational poverty measure would be impractical. But then the panel allowed medical out-of-pocket expenditures financed by wealth to reduce current income, tested against the new poverty threshold. Aside from the inconsistency involved, this introduced an upward bias in the NRC poverty measure.
He reiterated that these two assumptions underlying the SPM are a barrier to designing a useful measure of medical care economic risk. Meier and Wolfe suggest a cell-based approach to modeling medical risk. They also propose that at least the annuitized value of financial assets should be deemed available to meet necessary medical spending needs. The authors do not present their MCER ideas as a finished system, but their approach is reasonable, and they point in the right direction, Bavier said.
However, with the best practical cell-based array of expected medical out-of-pocket spending, Meier and Wolfe’s approach would be challenging to operationalize, and large variation in actual medical spending of families within cells will remain. The incidence of disease will vary within a cell, as will severity, and so will treatment intensity and duration. So if one adopts a suggestion to measure both SPM and MCER with the same data set, one is likely to be faced with the following situation. A family will be classified as poor in the SPM after its actual medical care out-of-pocket spending, which happens to be high for families within its MCER cell, is subtracted from current income. The same family will be found not to be at MCER
risk because its expected medical out-of-pocket spending, or its assigned point on the loss distribution within its cell, is affordable when measured against current income plus annuitized financial assets.
Bavier suggested that it may be possible to design a useful MCER that is consistent with the SPM assumptions about medical out-of-pocket spending, although he doubted it. A better solution would be to abandon the NRC approach to medical needs, include medical out-of-pocket in poverty thresholds, and move ahead with MCER based on the other purposes it could serve.
The two weakest points in the 1995 NRC report are the handling of medical care needs and the rationale behind its poverty threshold concept. Both would be corrected by recognizing that needs standards underlying major federal assistance programs for food, housing, medical, and other needs are elected government’s judgment of what, in Adam Smith’s words, the custom of the country renders it indecent for creditable people even of the lowest order to be without. That is where to start building a poverty threshold that the public and policy makers could understand and support.
Finally, he said, Medicare, Medicaid, and the Children’s Health Insurance Program (CHIP) constitute a growing share of all transfers. In 1968, federal and state health care spending represented 29 percent of means-tested assistance. By 2004, health care represented 55 percent, or $323 billion. An important goal in a new poverty measure is to reflect the major poverty reduction effects of government spending for noncash transfers and tax credits. However, unlike with food or housing transfers that are counted as income, when it comes to health care transfers, the NRC proposal for the measure of poverty does not measure the full effects.
Coverage expansions, that one assumes occur among the uninsured, can affect SPM poverty. However, increases in Medicaid or Medicare utilization and advances in treatments for current beneficiaries may not increase what the NRC panel termed discretionary income at all. Among Medicaid beneficiaries, per capita costs increased 30 percent in real dollars from 1975 to 2008. Among children, the largest eligibility group, the increase was 28 percent. It is unclear how much, if any, of this additional spending could reduce SPM poverty. In fact, when it comes to Medicare, any increase in copayments from increased utilization or new treatments may actually increase the medical care out-of-pocket spending subtracted from income and increase SPM poverty.
The sheer magnitude of public spending on health care means that whatever the conceptual and measurement arguments arrayed against counting public health care coverage as income when measuring poverty, not counting Medicare, Medicaid, and CHIP spending fully is a losing argument with many legislators, at least on the right side of the political spectrum.
FLOOR DISCUSSION AND COMMENTS
Several participants expressed their views and had questions on the various issues flowing from the presentation by Meier and Wolfe on the conceptual framework of MCER. The authors also responded to some of the comments made by the session discussants.
Wolfe opened the discussion by stating that one of the traditional issues with an absolute poverty line has been that it does not reflect changing standards of living. It keeps the same standard and looks at how people are doing relative to an absolute standard set long ago. Even if a new standard is set, it will still be an absolute poverty line. In contrast, the MCER index by its very nature reflects changing standards of medical care and therefore spending. That is an important distinction and probably another argument for separating the two measures.
In response to Burtless’s concern that the paper does not provide a blueprint, Wolfe responded that they were not trying to develop a blueprint. She thought, however, that they came relatively close. They were raising more general questions rather than suggesting to Census Bureau or Agency for Healthcare Research and Quality officials how to go forward.
Also, certain questions they thought were not in their jurisdiction. One example is affordability level. That is really a political question, a value judgment, and not something to which one brings technical expertise.
Thesia Garner (Bureau of Labor Statistics) said she was interested in including health insurance premiums in producing thresholds, as Short mentioned. The health insurance premium payment is intended to reduce risk, and there is also one’s expected out-of-pocket expenditures during a certain period of time to be taken into account. She thought that BLS staff would want to try to put just the insurance part of the premium into the thresholds; it’s a good approach to start with. The issue, though, in the SPM is to come up with an appropriate insurance premium for Medicaid; they do include the insurance premiums for Medicare and perhaps some for CHIP; she is not sure.
Short responded that she was suggesting that for the health care expenses that are paid by public programs, one would want to include something like average spending per person. That might be for separate cells but is very similar to a premium. A premium then just adds on the administrative costs of the insurance. So to the extent that one thinks of turning the random expenses into premium equivalents, that lets one move from individual data to averages, and that could be a big help in terms of the data requirements. She added that sometimes it is not clear why this is done on the threshold side, but in other places it is done on the resource side.
She further observed that the ACA eventually may help with regard to data, but not in the next couple of years. But if one evaluates insurance
based on its actuarial value, then one can get away from individual expenditures and applying a lot of complicated deductibles and coinsurance and so on. If a strong argument is made that everybody should have some guaranteed cap on their out-of-pocket expenses—and the ACA certainly moves in that direction—then that is a critical piece of information.
She noted that there had not been much discussion of Medicare. Here, the gap in Part D would have to close totally to get to an out-of-pocket cap for prescription drugs. Also, one of the main reasons for buying Medigap insurance is that Medicare is still an open-ended liability, as there is no cap on Part B, which covers services by doctors and other providers (Part A covers hospital services).
Meier had a couple of comments. First, there is a very important difference between measuring whether someone is at risk and how much they are at risk and measuring how much it would cost to protect them against risk. The index that she and Wolfe propose would be a measure of whether someone is at risk and how much they are at risk. If someone does have insurance that protects against risk, then the question becomes how that insurance narrows the spectrum of potential outcomes that this person could have.
Second, they do address Medicaid in their paper. It is a well-known problem that not everyone who is eligible for Medicaid applies for it. In their approach, an individual who has not taken up Medicaid is at risk at that moment. However, one can assess the person’s eligibility for a program that protects against risk.
Both these issues need to be addressed, she said, but it should be understood that they are two very different things. Moreover, it is not up to Wolfe and her to spell out the vision of moving forward—it is up to the study panel, which needs to be clear on what it wants to measure. And to take that a step further, the ACA is designed with the mandate that everyone should be insured. And the question for the study panel is whether it wants to develop a measure that assumes that is the case and that public policy is directed at ensuring that individuals have enough money to purchase the insurance. Or does it want to design a measure that actually reflects people’s current experience? They are two different things.
Meier noted that her presentation did not mention that the discussion of risk cells in the paper does not advocate an expected value approach, because expected value, instead of examining the spectrum of potential outcomes, gives one singular measure that is a poor representation of risk in the catastrophic context. That measure is not one they advocate, she said. Rather, they have advocated looking at a family’s probability of falling at the catastrophic end of their potential spectrum of outcomes.
Kyle Caswell (Census Bureau) observed that the conceptual framework first tries to identify some sort of baseline level of risk. It then tries to pos-
sibly make an adjustment for insurance status and then to compare that outcome with some measure of economic resources. But in this cost model, baseline risk would be a function of claim types. Utilization of health care services is certainly influenced by insurance status, so the uninsured, for example, will have fewer claims. The baseline level of risk for the uninsured, before one compares that with a comparable risk of the insured person, is therefore going to be somewhat of a muddied measure at the baseline. He wanted to know if Meier and Wolfe had thought about that, or how they addressed it in their paper.
Wolfe responded that one approach is just to estimate these risk levels for people who are insured. In fact, if there were sufficient observations, one could do it for people who have insurance similar to what one has in mind as the appropriate benefit package. That way would avoid including those with so much coverage that they are using care that may not have much of an effect, as well as dealing with the underinsured or the uninsured problem.
John Czajka (Mathematica Policy Research) observed that one of the issues the study panel has to deal with is the intersection between income poverty and this risk index. One can certainly imagine people who are poor by the SPM income measure who would not be considered to have a risk of getting worse. He asked: Is that the right way to think about these two indices?
Short responded that yes, that is what she was thinking. The first part, the SPM-like piece that is really about certain spending and certain income, does not include risk. And the second part is really a measure of risk—if one sets the definition of a catastrophe at the poverty threshold from the first index, then the question really becomes what is the probability of uninsured expenses that would cause a person to fall below that threshold or fall down relative to that standard.
She also responded to Meier’s comments, stating that she might have misunderstood what Short meant in speaking about the cost equivalent of someone being at risk, because they are complementary. If someone is at risk of being poor—and that could be a yes/no indicator or a how much at risk indicator—one way of quantifying that is to determine how much money it would take to eliminate that risk. And that is kind of the way she thinks about it in an insurance framework.
Kenneth Finegold (Office of the Assistant Secretary for Planning and Evaluation, HHS) mentioned one point that comes up in the context of what spending is necessary: the Dartmouth Atlas–inspired theme that there is unjustified variation in spending geographically, suggesting that overall a substantial portion of medical spending in the country is not necessary and in fact could be avoided (see http://www.Darthmouthatlas.org). And some places have found a way to avoid it. The ACA provisions try to adopt some of those methods.
Another point he mentioned is that people think a lot about the year 2014 because the ACA is going to be implemented and lots of things are going to be changed, some of which have been mentioned. One potential value to all this is that it provides a way to measure the impact on people who have insurance now and will remain insured, but their risk for out-of-pocket expenses may be reduced. There are measures of insurance, which will be available after the ACA, so one will be able to say what the impact is on insurance. However, there is the danger that some people may actually show up as poorer because their insurance is covering more and the premium costs are higher.
His last point is that, because of the concern about 2014, there is a lot of concern about 2013 and having a baseline to understand the impact of 2014. And therefore it worries him to hear the suggestion to wait until the ACA is in effect, and then things will be standardized and there will be new data. It will be very important to have all these measures as of 2013, he said, so that when the data on 2014 are available one can say this has changed or has not changed or may have changed in an unexpected direction.
Willard Manning (University of Chicago) stated that he can very well build a model that allows him to produce an expected amount from a data set by adding up across components or across family members to get an expected expenditure under a variety of controls. But one of the issues that comes up in the Handel approach is that one is talking about either variances in the simple cases or whole distributions. There are a number of issues to think about. Services are correlated; family members are correlated. Some families have members who are all very healthy, but some will have one member who is more ill than the others, maybe not by much, especially with the elderly. How does one deal with situations in which, when building up expenditures in the Handel approach, one ignores all of those covariances? And one knows that all of those covariances are positive and are actually increasing the dispersion, rather than making it smaller.
Meier responded that, on some level, she has just deferred to the methods in the Handel paper. The specific model that was employed was supposed to take care of these points, although she thinks that family members are treated as being fully independent in the model, which could be problematic. That said, not at the level of the statistical modeling, but at the level of the actual measure, Meier would want expenditures to be modeled as if people were acting on their own. That might sound strange, she said, but one would not want parents to be forgoing services because there was only so much money in the family and they decided to devote their medical care money toward the children in the family, for example. She stated that she knows that in the data indicate decisions in which families are operating as a unit and are allocating their consumption in certain ways. But ideally
the measure should be capturing this idealized consumption that people would have if they were not facing constraints.
Wolfe added that this method takes into account the way the insurance policies are designed. This can be a problem in terms of the data that are available, but in terms of the method, it does take account of family members’ utilization. It takes the deductible, which may be a single person’s deductible or a family maximum, so all those parameters are taken into account when the family is combined. The characteristics of the insurance plan are not the issue, because if one knows them then one presumably knows them at the family level, and when one combines the risk of the individual family members, then one applies the policy as it applies to the family, if it is a family policy. There may be a data limitation in terms of what is known about the characteristics of insurance policies. But if one had those insurance policy characteristics, then one could aggregate the risk at the family level and then apply the insurance policy.
David Betson (University of Notre Dame) complimented Meier and Wolfe on their paper. As a member of the study panel, he cautioned that it is not for the panel to set what is an adequate benefits package. That is not necessarily the members’ area of expertise, although some of us certainly could weigh in on those issues, he said. But it is very important to know, as things roll out, what would be available from the data that reflects what is being done in society. In actuarial work, one often hears this kind of referral to typical benefits offered, typical large employer benefits, and typical small employer benefits. So the ACA will help very much because its benefits standards will reduce some of that variation and make data collection easier. But the study panel should not make value judgments, he said.
Michael Hurd (RAND) had two comments, one very specific to the paper, another one much more general. The first has to do with the adjustment for out-of-pocket medical expenses. It really affects the elderly the most, and that brings up the issue of assets. The elderly have quite a few assets. The paper recommends annuitizing assets for an income flow, but in fact people do not annuitize. The assets are meant partly to be precautionary assets. Medical expenditures are partly episodic and partly chronic, so people should be allowed to (and in fact do) spend episodically out of their assets. So an annuitization would distort the availability of those economic resources to buffer against risk.
His second comment was much more general. He questioned why the study panel would want to develop a measure of economic ex ante risk. That is a complex undertaking, requiring the joint modeling of economic resources and risk for health care spending and producing ex ante probability distributions that are not just a variance but the very high end because of the skewed distribution of outcomes. These have to be jointly modeled with economic resources because spending varies with economic resources.
People who are more wealthy spend more, and people who are less wealthy spend less, and it happens throughout the wealth distribution. Therefore, it is necessary to model all of those things.
With an ex post measure of outcomes, when comparing outcomes in spending with economic resources, if available economic resources fall below some level, however that is done, it is a very straightforward situation to explain to people. No modeling is needed. The data will allow one to integrate over all the insurance packages and economic positions that people occupy and take account of the covariances between their economic position and their health status and family situations.
Developing the ex ante risk index would be like developing a poverty index to predict that next year X numbers or a fraction of people are going to be in poverty. That is not of interest. Of interest is saying that X many people were in poverty this year and being able to explain it. He suggested that the same should happen with the MCER index.