The workshop brought together participants from different disciplines and interests who heard and commented on issues and ideas for moving forward with developing a measure of medical care economic risk (MCER). The final session highlighted their perceptions of some of the major areas and priorities that surfaced during the day.
After opening remarks by Michael O’Grady, three workshop presenters—Pamela Short, Sara Collins, and James Ziliak—led off the discussion, briefly highlighting some of their ideas. A general discussion followed of participants’ views and concerns.
Michael O’Grady (NORC at the University of Chicago) observed that a number of methodological data concerns surfaced during the day’s discussions. There are also issues of practicality in developing the measure of MCER to avoid having it derailed by the desire for ideal data that are not there. He then highlighted some of the key methodological issues that the study panel will have to wrestle with.
He started with the concept paper by Meier and Wolfe, which lays out a set of decision models. The study panel will need to make some decisions about the key methodological issue of risk, a prospective measure, and burden, a retrospective measure. There are pros and cons to the two approaches, whether to take an experimental path, try to do both and compare early results, or go in one direction or the other.
The discussion about having different thresholds was interesting. Should there be different thresholds for the poor, the near poor, and the elderly or for the chronically ill? Will the study panel be comfortable coming to a consensus on that? Is it something the study panel can explain to a broad audience compellingly? The question of income versus assets is also an important issue.
He next addressed the realities and the quality of the data, getting to the question of using data from the Current Population Survey Annual Social and Economic Supplement (CPS ASEC), the Medical Expenditure Panel Survey (MEPS), or both. What are the options in terms of imputation, collecting new information, or some other way?
O’Grady said he found the discussion of TRIM3 interesting, bringing to mind that this measure has to go out to a broad audience. He observed that once the talk moves away from a survey or an imputation of a survey toward modeling, it is viewed much more as a black box.
Pamela Short repeated the question Michael Hurd asked earlier, why a measure of MCER is being developed. From the day’s discussion, she came to realize that the problem of evaluating the effects of the Affordable Care Act (ACA) was probably front and center, at least from the sponsor’s point of view. That’s why the workshop participants spent so much time talking about prospective measures and insurance and not so much about burden.
People are indeed going to be spending more on insurance because it will be required, she said. The fact that the ACA requires everyone to be insured presumes that there are benefits, so there is a need for a good and convincing way of quantifying those benefits. But a lot of people who spend more on insurance will not benefit, because they will not get sick and so will not spend a lot out-of-pocket. To add up the benefits of having insurance, there is a real conceptual problem if one only looks after the fact at out-of-pocket expenses, because that does not attribute any benefit from having insurance to people who do not use it.
The situation is similar to buying life insurance. Why buy life insurance, since one does not want to use it? The benefit of health insurance is that some people who did not have insurance are going to be freed of the risk of major out-of-pocket medical expenses, and some people who have insurance are going to get much better insurance, at an affordable price. Consequently, Short observed, this issue of showing the value of insurance or measuring the value of insurance is central in evaluating the ACA. A closely related issue is defining adequate insurance (or “minimum essential benefits” in the ACA).
There will be before-and-after comparisons of family spending, in which this measure gets used as a tool for describing winners and losers from the implementation of health care reforms. With that as the driving force, at least in the near term, Short observed that the issue of state variation is really important. The framework for reform leaves a lot to the states, and the numbers that Sara Collins showed emphasized how much variation there is among the states.
In a way, she said, the train has already left the station with the release of the Supplemental Poverty Measure (SPM). It seems to be moving in the direction of looking retrospectively at high out-of-pocket expenses relative to resources, however they are measured and whether or not assets are included with income. In terms of work for the study panel to encourage in the short run, Short thought that there is probably time to bring together the needs standards for other consumption that are in the SPM, to try to give more empirical basis to the idea that people under 200 percent of the poverty line should not spend more than 5 percent of family income on health care. The idea would be to figure out how much room there is for premiums and out-of-pocket expenses, beyond the SPM need standards, rather than arbitrarily picking a threshold of 5 percent of income, 7.5 percent of income, or whatever—numbers that mostly seem to appear out of the air. A measure grounded more in a model of real consumption needs, probably drawn heavily from the SPM, could provide an empirical basis for picking a threshold percentage.
In terms of trying to measure the adequacy of insurance, Short favored the collection and use of information about people’s insurance policies. Gary Claxton’s presentation underscored the importance of detailed information about plans and variation among plans. The distinction between nongroup insurance and employment-based insurance is important and is collected in some surveys.
There also is need for data about employers, as there are important differences in the plans offered by small employers and large employers. At one time, MEPS got copies of people’s policies and abstracted them. There was some discussion during the workshop about actuarial value, which is one way of comparing policies, as are out-of-pocket limits, deductibles, limits in scope and duration, and covered services.
On the issue of assets, Short said she is not sure why the question of counting in assets has been linked to measuring medical care economic risk. Basically, one expects elderly people to use their assets for all kinds of regular consumption.
Sara Collins highlighted a couple of items from the range of issues identified by Short. On the threshold issue, she thought it is probably important to think in terms of income. She did not necessarily agree with Short’s idea about making a stronger empirical basis for picking a percentage. There
is some historical experience of this with the Children’s Health Insurance Program; she asked why one would move away from it at this point to a different or higher income cutoff.
On the asset issue, Collins was struck by Jessica Banthin’s zero asset level for working-age people who are poor, and not much higher than that in some of the upper income categories in that group. There just isn’t much for people to draw on at lower income levels if they do have high out-of-pocket spending.
She also emphasized the importance of being able to disentangle the premiums from the out-of-pocket medical costs. She agreed with Short’s point about premiums being a required price to pay with the individual mandate and the question of whether they buy a sufficient amount of coverage. In the law, the affordability of the premiums is clear; they are on sliding income-related scales for that purpose. Separate from the affordability of the out-of-pocket costs, exposure is a gray issue as a share of income, on an actuarial value level and translating into what it means for families. The measures do different things, premium as a price of health care, and exposure to out-of-pocket costs, in terms of tracking health reform, not only as an economic burden but also as a way to enable people to get the health care that they need.
Collins remarked that Schoen’s work on the underinsured over the past several years shows that people who are underinsured exhibit somewhat similar behaviors to people who are completely without coverage, not getting the health care that they need because of costs and having lots of medical bills. This access issue is very striking in those data.
She echoed Short’s point about needing to examine differences across states. The ACA is implemented at the state level, with a lot of federal guidance. But once it goes to the states, they are in a position not only to implement their exchanges but also to enforce their new market rules. So a state-by-state measure on out-of-pocket exposure is going to be extremely important. The point about vastly different rates of health care cost growth across states also speaks to the need to track by state.
Finally, taking off on what Michael O’Grady said about policy makers, that these measures should be simple for both federal and state policy makers to understand, Collins emphasized that the simplicity of the measure should also include being able to see and understand things that are going in different ways: out-of-pocket costs going in one way and income going in the other way, and being able to disentangle those effects.
James Ziliak observed that, going back to the Wolfe and Meier presentation, he was very sympathetic to the notion that risk is a prospective concept.
At least part of the motivation he detected from the sponsor in developing a measure of MCER is to figure out how to improve well-being. So
it does not seem inconceivable to construct an ex post index, much like the SPM, which could be used to assess different programs, such as the Supplemental Nutrition Assistance Program and Temporary Assistance for Needy Families.
Clarification is needed on what is of most interest: a measure that indicates the policy’s effectiveness contemporaneously or one that indicates the risk of individuals facing a large out-of-pocket expense or medical emergency in the future. The latter is more the intellectual exercise that one associates with risk, insuring against uncertain events.
Short mentioned in her presentation that one of the problems with the SPM is that, if one assigns the uninsured to be insured, it makes them look worse off economically because the SPM records just spending. She said that what is needed is an MCER measure to show that the insured are better off.
If these measures are constructed as separate indices, there is no obvious way to capture that any one individual is better off or, in some aggregate sense, that all are better off. He said he actually likes the idea of two separate indices.
Ziliak pointed out the literature on multidimensional measures of poverty and deprivation. The United Kingdom has a measure of deprivation with something like 70 different items, and they are added up into a single index. So surely in the United States it is possible to add two measures together to come up with some index. There is recent work on multidimensional measures of well-being that could be aggregated into a single index (i.e., Alkire and Foster, 2011; Bourguignon and Chakravarty, 2003).
He commented that measuring poverty appears a lot easier, perhaps because it has been going on for a lot longer. One draws a line based on some measure of needs and then counts resources and compares one with the other. It seems straightforward.
But with the concept of MCER, the notion of the thresholds is still not well defined at this point in time, because what people need has not been well defined. But if one were to use different thresholds, he said he thought that it would be important to capture employment status. Part of the reason is that a lot of money is spent in this country on work-related injuries and illness. He noted that coverage and type of coverage seem critical in thinking about thresholds. Geographic adjustment is also important, as indicated by cross-state variation.
Finally, in terms of data, Ziliak said he leans heavily toward moving forward using the CPS ASEC as the data set of choice, in part because of the need to go forward; there is a mandate in the ACA about constructing measures reflecting state differences and medical need. The CPS ASEC has large sample sizes, and it has introduced medical out-of-pocket spending,
so there is progress in that dimension. Income data collection in the CPS is not as detailed, especially the transfers going forward. But it seems that the CPS ASEC does have most of what is needed to move forward in developing the new measure.
GENERAL DISCUSSION AND COMMENTS
Most of the discussion centered on the benefits, or lack thereof, of requiring everyone to purchase health insurance, as called for in the ACA. Many views were expressed.
Emmett Keeler (RAND) opened the discussion with the observation that one of the main purposes of the measure of MCER is to evaluate the ACA. But he wanted to make sure that it is an honest evaluation.
Although it is clear to him that giving people insurance is a good deal for them, it is not clear that forcing people to buy insurance is a good deal for them. It depends on the price and what the benefits are. Somehow this measure has to take account of that. To the extent that making people buy insurance is a burden, something in this measure needs to that show that economic reality. In the current context, if it were a good deal for uninsured people to have insurance, they would have it. Basically uninsured people are saying they would rather have the money to spend on food or shelter, than to have the money to spend on an insurance premium, and the markets do not work very well for them. There are many reasons why uninsured people do not have insurance; it is a choice that they make. Subsidies can make it a good deal for them. Somehow the measure must include an honest evaluation of the value of that insurance.
Constance Citro (National Research Council) commented that it is very important to compare apples with apples, which is why one would like to compare not only the SPM pre- and post-ACA, but whatever this measure of MCER is pre- and post-ACA. The SPM may show that some people are worse off, in the sense that the money they have available for the basics that are in that measure, but the MCER may show that on that dimension they are benefiting from the insurance. Of course it is a judgment as to what people think of the different values of the different costs—the higher out-of-pocket costs on one hand and the greater insurance against risk on the other hand. But it would be really too bad if all there was at the time that the ACA is implemented is a measure comparing the effects of health care reform on just the SPM.
Keeler commented that he was trying to think of benefits of reducing risks that do not apply to reducing burden. After people have been sick and have had to pay a lot of money for it, they are in a bad situation. But why would they want to reduce risk? He thought of two reasons: one is peace of mind—maybe buying insurance also buys peace of mind. The other reason
is that people like to buy out of unpleasant decisions. So, for example, a good thing about having insurance is that one does not have to say, my kid is sick but I am not sure that I can afford to take him to a doctor. That is a very unpleasant situation. Other than that, it seemed to him that one simply wants to look at people who have had big medical expenditures and been disadvantaged by them.
Pamela Short commented that the theory of insurance is that, with something like medical care, only a very small percentage of people will get sick, but it costs a lot of money when that happens. So the idea is to pool that risk, to average it out. Everybody pays the average, and nobody faces the extreme tail. Because the extreme tail drives consumption way down, risk-averse people would prefer the average over the risky possibility. Those expenses do not go away, but for individuals they are pooled. She certainly was not trying to say in her presentation that the study panel needed to propose a measure that would be sure to show that the ACA is a good thing. This is important is because it is a way of quantifying the value of insurance to people who do not use it. Again, she said, think about life insurance or car insurance, which people do not want to use. People spend quite a bit on insurance of different sorts, and this workshop is just trying to give medical care insurance its due.
Sarah Meier raised the question of the value of insurance. Insurance is a mechanism that results in an income transfer in the event of illness that people would not be able to afford in the first place, she said. If the next thing they would buy if their income were expanded were health care, then that is the value in the insurance. She also commented that people will be mandated to buy insurance premiums that are not necessarily affordable. The reform structure is that premium rating can vary by age categories. It also limits the subsidy structure at 400 percent of the family poverty level. So it is feasible that people who fall just above this level, who are also pre-Medicare age, will actually be paying a surprisingly large percentage of their income for the premium.
Referring to an earlier comment regarding the regression-based model, Meier agreed that it is the better approach. She pointed out that families are made up of individuals. She and Wolfe suggest in their paper classifying health risk at the individual level, so the suggestion is that these predictive variables are individual risk characteristics. Not only are they thinking about health risk as an individual aspect, but also insurance policies within a family may vary. Some family members may have a Medicaid-covered individual, and some a privately insured individual. So figuring out how to model a family’s likelihood of exceeding a certain value, when everyone in the family could contribute a different amount of money to that value, under different scenarios, and on top of that everyone in the family has perhaps a different insurance dynamic, is something they could not pull
together in the paper. Hence they put forward the idea of modeling a probability distribution and then simply applying insurance characteristics and aggregating them.
Ziliak commented that one approach is weighting by family size, taking whatever family concept is selected and putting it at the individual level. But the CPS has individual-level data on earnings and income, and the insurance status is at the individual level, not the family level. So, he said, one can go fairly far with the individual level and then weight it up to the family.
Meier said that one is not trying to assign an individual a risk score or an expected value in the next year but instead recognizing that an individual at the start of a year has an entire spectrum of outcomes that could happen. The shape of that distribution and where it is centered is going to depend on initial health status. If someone has diabetes, one can be reasonably certain that it will increase expenditures. But the actual shape that falls around the mean value is what she meant in talking about modeling—looking at that shape, seeing what insurance does to the shape. Ideally, if there is an out-of-pocket maximum, it should just slice off the tail. And aggregating that upward to the family, instead of having an expected value of expenditures for a family in the next year, her vision is that a family actually has a risk distribution. From that distribution, one could come up with some understanding of how likely they are to be placed in poverty or to experience expenditures at 10 percent of their income. Tacking on the premium would show whether the premium itself put the family in poverty, before out-of-pocket expenditures are even factored in.
A participant commented about the discussion of the value of insurance in the context of a person who does not value the insurance, who has low income, and who now is forced to pay a premium and has therefore a lower amount of money available for everything else. That must be captured, he said, but there are also the benefits to other people of that person’s being insured. People who are insured have an opportunity for more efficiently provided health care. They can actually get their relatively minor issue treated much sooner and so skip the emergency room later on, when the situation is far worse. There is also the opportunity for the health system in general to be more efficient if more people are insured. It is not just the person who gets insurance who gains from having coverage or having better insurance. In terms of explaining the overall effect of some changes in health care policy, leaving that out very much understates the benefits of changes that add or improve the quality of insurance.
Collins commented that the protectiveness of the coverage is not just about getting sick or not. It is also the first dollar, coverage for kids, getting better health screening, and improvements for people with currently bad coverage.
In Schoen’s underinsured study, she said, the drivers are not the high, out-of-pocket catastrophic costs, but increasing rates of people with high deductibles and high out-of-pocket spending because of their low first-dollar coverage on their policies.
Short remarked that at some point it might be important to look at routine recommended medical expenditures. It is not so much a matter of risk, as to see whether the allowance in the SPM for a little bit more would adequately cover routine kinds of care for people at lower income levels. She questioned if the little bit more is actually enough; there might be an argument for increasing it.
Barbara Wolfe made the point that everyone, if they have value, could buy health insurance. The problem is that one cannot buy a decent policy in the individual market; the policies are not available or are very limited. In addition to being all there is, they are quite expensive relative to policies in the group market, even the small-group market.
Her second point is that, before the ACA, many individuals with preexisting conditions could not buy coverage. They certainly could not buy coverage for their existing condition, at least for a fixed period of time. She gave as an example one of her students with a preexisting condition that could not be covered even though she was part of very large group, the state government sector, at the time.
If part of the purpose is to evaluate the ACA, then a good starting point is probably 2009 or earlier. To have a benchmark of some of the gains, then some of the work should use an earlier starting point, before any of those preexisting conditions, including ones for children, which were among the first to go into effect.
In talking about these matrixes and groups, the closer one gets to a narrowly defined group, the further one moves away from risk, because eventually, there is a cell of one. She said she thought that the study panel will have to think about how narrowly defined those cells should be. When people buy insurance for their home or their auto, they are getting a rate that depends on some kind of group; they are not getting their own individual rate.
People do face a risk. Some of that risk is just based on age, and it should not depend on an individual’s already-diagnosed preexisting condition. Some people, particularly individuals with limited access, won’t know that they have a condition because they have not received the medical care that would put them into the risk index. So it is important to think through how narrowly defined those matrixes should be.
Many other aspects of medical care, such as oral health, have not traditionally been included in most health care policies. These are components that have important potential for poverty in the future, so they are important in thinking about how to define this benefit plan.
Laura Wheaton, referring to O’Grady’s comment about staying away from the black box and developing something straightforward, asked what sort of additional level of complexity he was thinking about. Is taking some information from MEPS and then transporting that onto the CPS the kind of a layer of complexity that people might find objectionable? Would it be better just to stick with MEPS for this medical risk index? She has done statistical matching and imputation, which adds some time to the analysis. If MEPS seems to be clearly better in many ways in terms of the data elements, perhaps one should just stick with it.
The advantage of the CPS in having the large sample size in the states has been pointed out. MEPS has information that varies by state but is not the underlying sample size to support state analysis. Is there too much, then, when one imputes that onto the CPS, saying there is enough sample size in the CPS?
O’Grady responded that there are trade-offs. As the study panel members discuss this issue, they will consider the pros and cons of using MEPS and using the CPS. It may take a number of test runs. Utilization data and some other items are really strong on MEPS, but the priority may be to make state-level estimates. The study panel will consider that both have their strengths and weaknesses.
Citro agreed with the idea of going back a couple of years to the totally pre-ACA environment. The ACA is the law now, and various provisions will go forward; she is sure that the study panel is not interested in prejudging the outcome of the measurement but just how to appropriately measure, given all the complexities.
She also commented on the frustration in this measurement area. The country spends lots of money on health, including on Medicaid and Medicare, and it has been great. But getting credit for it in the economic measure of poverty has been a problem. Short made the comment that the purpose of medical care in some sense is to restore people to some state of health rather than get them to a higher level. Before Medicare and Medicaid, what could be done for people? They could have a broken bone set. They could be given digitalis for congestive heart failure. There were a few antibiotics and a few vaccines, but that was about it. There was no cholesterol medicine, no decent antidepressants, nor many other treatments that are now available. It is definitely a benefit that those treatments are here, but they cost money. As a comparative example, what food stamps provide is food to meet basic calorie requirements, and food intake requirements have not changed over the millennia. What medical care can provide has changed, and it is a benefit, but it is very hard to measure. Trying to put it into the same framework as food, clothing, utility bills, and the list is part of the frustration that lies behind developing this measure. That was a major motivation for the 1995 NRC panel to say yes, medical care must
be looked at, somehow keeping it related but separate because it is so different conceptually.
Wheaton remarked that the group has talked about the individual level and the family level, but there are also different definitions of family. Is the appropriate unit of analysis for the ACA the tax unit with dependents? That could be quite different from a broad definition of family in the SPM, which includes all related persons plus cohabiters.
O’Grady responded that the study panel could defer to the SPM methodology and use whatever it called a family. The ACA cutoff for dependent children is age 25. But even with that age cutoff, family resources are still being spent, for example, for a 27-year-old child who has aged out of family coverage. If fully uninsured at that point, the family still is on the hook for that person’s finances. In closing, Michael O’Grady thanked everybody for coming. He has expressed his thoughts, concerns and comments throughout the discussions and really got a lot out of it. He has appreciated everything, especially the presenters. They did a great job.
Connie Citro added her appreciation for a productive day and gave everyone a round of applause.