Participants’ Views on Needed Research
The workshop brought together a large group of participants from many different disciplines and interests. During the course of the day, participants discussed several issues and ideas for improving the projection models for health care costs for the Medicare population. In the final session, they discussed their perceptions of some of the major research areas and priorities that surfaced in the course of the day.
Two workshop presenters—Eileen Crimmins and Darius Lakdawalla—led the discussion to stimulate the wrap-up, briefly outlining some of their ideas on areas for further research. A general discussion followed on what many participants viewed as some of the more fruitful priorities for needed research on aging-related issues with the goal of improving health care cost projections.
Eileen Crimmins (University of Southern California) opened the discussion by highlighting two areas for further research. She noted that in discussing the various factors influencing health care costs the participants did not deal with, and not much research has been done on, end-of-life costs. People die only once, and this period in life can be a very intensive and expensive one in terms of health care costs. It is important to learn how end-of-life decisions are made. Do individual characteristics affect these decisions, or are decisions basically made in hospitals by physicians and not affected by individual characteristics? She emphasized that in her
opinion the end of life is the point when costs are very high and potentially more controllable.
The second area she mentioned is based on her comparison of the United States and other countries. She questioned why the United States gets so little for what it spends. Looking at cross-national comparisons of health status and the associated costs, the United States spends more money and yet has poor health relative to other countries—worse obesity and worse hypertension at a lower age, for example. Part of the expenditure difference results from worse health, but the United States is also spending money that is not buying better health. She concluded that these two examples point to the need to directly study costs.
Darius Lakdawalla (University of Southern California) noted that when thinking about the problem of modeling, for a long time the view was that there would be only one model of the economy. But the experience of the 1970s departed radically from model forecasts and cast considerable doubt on the notion of a single, unified model that could explain all economic phenomena. Similarly, it is no longer thought that one model will unify everything one needs to know about the health care system. But it is important to push the frontiers of the modeling enterprise. Several such areas have surfaced already during the workshop, he observed.
Lakdawalla emphasized three specific areas that need to be pursued. First is to build models that push the frontier in modeling medical technology. Second is to build models that advance the frontier on pricing and in particular understanding all relevant prices, not just explicit prices in the system. Part of the problem in health care, from a modeling point of view, is that all of the relevant prices are not observed. There are many shadow prices that are extremely hard to measure when there is public involvement in the system. For example, when health insurance is publicly provided, it is hard to observe its true price to consumers. So how does one go about thinking through price responses in a mixed public–private system when, as Michael Chernew pointed out, prices are one of the most important brakes on health care spending growth and may be one of the mechanisms by which projections go from being ridiculous to being reasonable. So understanding how to build those mechanisms into the model is another first-order challenge. Third is dealing with uncertainty in a more plausible manner as decision makers are often rightfully skeptical of any given model because they think there are too many assumptions built in that are artificial.
It is hard to push all three frontiers at the same time. It is hard to build a model that, for example, includes a really sophisticated supply-side evolution of technology and sophisticated accounting for prices along with changes in aging and health and takes uncertainty seriously. That seems like a very daunting task and may be fundamentally impossible, but it certainly is possible to take different approaches and get at pieces of the problem.
That might be the best way forward at this point, at least along these three dimensions.
Participants identified several areas for enhancing the current efforts to project health care costs. Most of the discussion focused on three areas: the need for a plurality of models, data collection versus research, and the value of cross-national comparisons.
Policy Versus Research Needs
Dana Goldman opened the discussion by commenting that there is a 14-year difference in life expectancy between a black man with 0 to 8 years of education and a white man with 13 or more years of education. So the distributional consequences of policy changes, while they may not be necessary for a projection done for Congress, are of enormous interest to most researchers. How one looks at distributional effects of policy change are therefore paramount.
John Haaga (National Institute on Aging) pointed out the need for a plurality of models and mentioned the possibility of building in a number of different assumptions with the microsimulation approach. Others have mentioned scenarios as something they would like to see more of. He questioned if people could either develop different microsimulation models or do scenario modeling, neither of which has really been explored.
Goldman responded by pointing out the tensions between policy and research needs for models. Policy modeling involves forecasts or projections, and research modeling involves scenarios for the future; from a research perspective, scenarios are clearly more important. The tension suggests that more communication is needed between the policy side and the research side, so it is gratifying that people from both government agencies and the research community are participating in the workshop. It may be that some more structured forums, like this workshop, would allow comparison of policy scenarios. For example, one could give modeling groups around the country some parameters around a scenario, say disease management. The groups could go off and do their thing and then get together and discuss what the scenarios would look like.
Crimmins commented that people who are developing models for policy have no choice but basically to forecast or project based on past trends and to use relatively simple models so that they are transparent. But then the question is what factors are going to change the trends? For this, one needs to model a whole set of indicators to understand the implication of
changes in them. Focused microsimulation models might show that some things do not matter and some things do. Their results might inform more general models. There is no single model that meets all needs, as stated by both Lakdawalla and Jay Bhattacharya. A generalized model is needed that is as simple as possible, informed by underlying studies that are realistic in laying out what the alternatives are.
Regarding the issue of what the models do and what people want them to do, Marilyn Moon observed that the Centers for Medicare & Medicaid Services (CMS) and the Congressional Budget Office (CBO), for example, will always have the challenge of looking at government spending, because that is the central issue they are supposed to look at. Others propose additional models and may ask that they be included as supplements in some way, along with other key issues. But these tend to get ignored when looking at government costs. For example, a change in policy that shifts costs back onto the public may look good from the government perspective but bad from society’s perspective. Similarly, two equal expenditures might look the same in terms of government costs, but one extends life substantially and the other does not, and the two would be treated the same to some extent, but in fact they should be differentiated.
Distributional impacts are another issue. Perhaps societal and distributional impacts should be outside the big government cost models, but they should be elevated to a similar kind of status somewhere in terms of discussion. One of the things that has always been a frustration for people at CBO, for example, is that they are answering only part of the question. Therefore, the kinds of discussion at the workshop and perhaps the role of the independent research community could make it possible to think about some of those other needed models and develop some consensus around them so that they could be elevated in policy discussions.
Victor Miller (Government Accountability Office), who has worked with state governments, had two suggestions for consideration on a policy agenda. The first is that there has been no discussion of the current recession, which has both short-term and long-term impacts. More people are receiving disability income and Social Security disability income and therefore are also on Medicare. The long-term impacts of what are called short-term recessions need to be taken into consideration.
The other suggestion relates to the comprehensive impact of tax policy. He mentioned that Sibley Memorial Hospital in Washington, DC, was planning to add more beds and make all rooms singles, saying that it is a standard of care. Defining single rooms as a standard of care seems to be consumption, and consumption is financed in a number of ways: through tax-exempt financing of all hospital construction, through flexible health benefit plans, or through the tax exemption of employer insurance. Adding it all up would be a useful endeavor.
Data Collection Versus Research
Bhattacharya brought up the issue of relative investment in research funds, which are limited. If one has to choose between dollars spent on data collection and dollars spent on research and model development, it is much easier to get funding for data collection efforts. Research is under consideration at the National Institute on Aging (NIA), the Office of the Actuary (OACT), and other similar organizations. He asked which the participants thought would be more valuable to them.
Richard Suzman’s response was to say both. NIA has to maintain some programmatic balance in terms of reducing low-quality investments in data. There has to be a limit on how much it can invest in data, although if the data are very well cared for and documented, the start-up costs for using them for research are lower. However, improving a data set tends to make it go on living longer, so it uses up more resources.
NIA is also trying to leverage funds. For example, the Social Security Administration has co-funded some of NIA’s retirement modeling efforts, and there is good collaboration between the two agencies. Suzman has hopes that they will be able to generate investment across the National Institutes of Health (NIH) in this area.
Steven Heffler remarked that the priorities depend on what one wants to accomplish with the money. In most of the presentations during the course of the day, the findings were based on historical analysis of very detailed data. In most cases, there were not many answers about what would happen in the future, given all of those trends. It is the application of what one learns through the analysis of the historical data that is really relevant for projecting future spending.
The question is: Do you spend the money learning what happened in the past, or do you spend the money building tools and vehicles and models to predict what is going to happen in the future even if you do not know what has happened in the past? To the extent that some of these other models can help inform the cost projections, help to identify what some of the drivers are, and can be used in some applied way, that applied research is the most important thing. He remarked that one thing the OACT staff talks about when building models is to be careful what goes into the model because it will have to be projected. One can build the greatest model in the world to explain what happened in 2008, but if there are 10,000 variables, and they have to be projected to 2050, then it is not going to be a very good or a very reliable projection. For OACT that is always the issue, and the agency feels very strongly that, wherever the resources are put, if the idea is to eventually use models in cost projections, then they need to be used so that the information can actually be applied, even if it is in some models
that currently do not have the capabilities and the details on some of the dimensions discussed here.
Goldman responded that aggregate cost projections are certainly important, but, for example, suppose someone had asked, What is going to be the impact of health care reform on low socioeconomic status populations? That is probably not something that current models are ready to answer. That is an issue that some researchers have devoted their lives to studying and would like to know the answer, not just from a research perspective. That is where some two-way communication is needed.
In that particular example, Heffler pointed out, one is trying to understand and model the behavior of a specific group. Data are needed that are representative of that group, and the modeler has to understand what the drivers for that group are, and so forth. When the model is built and it all fits, the question will be whether all of the detail for that group was needed in order to develop the correct aggregate projection. If not, then perhaps a model can be built that is unique to that group that answers the specific question. But one has to accept the fact that it does not add up or build up to one that is broader. Heffler commented that CMS is in the same situation as CBO: the costs of certain things are going to be very difficult to estimate, whether that is a group or policy or whatever, because of the information that is available and its relevance for the total.
Jonathan Skinner asked how to predict what is going to happen out in the future. All of these ideas about better data sets and integrated data sets are not going to help. There is no way to collect the information—even if all of it available anywhere in the United States today was collected—that would really help that much in trying to make those kinds of conjectures for the future. For the shorter term, there are important issues for health care reform, for the effect on disparities, for example. It is remarkable how little is actually known about the U.S. health care system. Something is known about Medicare, but there are no good answers to such questions as: What are the principal reasons why health care costs have gone up and have doubled in the last 12 years? People are obese, for example, but that does not really answer the question.
Furthermore, very little is known about health care beyond the Medicare population and some of the population under age 65 and what is going on in various micromarkets. It is remarkable how little is known about costs in the United States except for the aggregate. It is also remarkable how little is known about the Medicare-equivalent data sets in other countries, which often are not available to researchers. There is very little idea, for example, about what doctors do in Germany that is different from the United States. Cross-country comparisons of physicians per capita and hospital beds per capita show the United States trailing the countries of the Organisation for Economic Co-operation and Development (OECD). It
turns out that in Germany, for example, chemotherapy is administered as an inpatient procedure. In the United States, chemotherapy is administered on an outpatient basis.
With regard to the differences in spending between the United States and the OECD, Mark Freeland mentioned an analysis by Victor Fuchs over a decade ago that found that, for the non-aged, the United States and the other countries spend about the same per capita; all of the differences come basically with the aged.
Freeland also observed that much has been said during the day about the health care system but not much about the production of health—in other words, all of the nonhealth care interventions that cause changes in health status. Studies have shown that these are more important than health care per se. Mentioning Goldman’s point that black men with less than 8 years of education have a much shorter life span, Freeland thought that a major reason is deaths associated with violence, drug trafficking, accidents, stabbings, HIV, and such.
Crimmins emphasized that people in the United States have poor health relative to other countries, not necessarily poor health care, and it is important to understand why. For example, there are major differences in how end-of-life care is treated across countries. That care is expensive and it would be interesting to understand the different health systems.
Michael Chernew made two comments. First, apart from getting new data, getting access to the data already available and putting them together would be a reasonable first step. Second, a one-time reduction in spending is not the same as change in a trend. This issue is very important for a lot of the modeling work. Doing research on changing trends over time requires long panels of people in different settings and getting a detailed sense of what is happening over that longitudinal context. Longer panels of data and databases allow relating the longer panel data to the policy variables of interest. Collecting those data or at least compiling them, which is even better than just collecting, would be extremely useful to begin to understand if in some settings there are changes in the slope, not just the level, of spending growth.
Value of Cross-National Comparisons
Suzman asked the participants their views on the value of cross-national comparative studies projecting health care costs. He asked if one can look at OECD countries and get actual expenditures and outcomes for health and health care. Does that give some range of confidence intervals that one can project between countries, for example, the United States and Canada?
Crimmins responded that it is invaluable to understanding the issues
discussed at the workshop to put them in the context of other countries that are like the United States. She is only beginning to understand what the differences in health are. The process ranges from health, to the treatment of a given condition, to control of the severity of the condition, and to the outcome of mortality. Understanding that process—how money is spent, what kinds of treatment are provided to people with a given condition, and who survives or does not survive for how long and at what cost—is really invaluable in understanding the potential for reducing or changing costs.
Lakdawalla agreed, adding that a theoretical perspective needs to be developed, which was heard as an undercurrent throughout the day. An example would be models of political economy. Much of what the cross-country variation is driven by—differences in institutions, political incentives, and the like—has not yet been built into the models, but it needs to be.
Suzman remarked that NIA had done just that some time back. It funded a study some years ago at OECD to learn about what works best for a few specific diseases, at what cost, and with what impact. That needs to be repeated, he said, but it has become very difficult to get funds to OECD.
Goldman observed that many researchers had given up on international studies because good microdata were not available. Now, with NIA’s investments, very good microdata are available; the irony is that now the policy variables to relate to the institutional differences are not available. Surveys such as the Health and Retirement Study (HRS) and the Survey of Health, Aging, and Retirement in Europe (SHARE) have a lot of information about health in different countries, but figuring out which country has reference pricing for pharmaceuticals or what they do on disease management is another story. Some effort would be worthwhile to harmonize the policy variables so that these comparisons can be made.
Richard Suzman (National Institute of Aging) informed the participants that NIH and NIA do not, and should not, undertake policy research but will conduct research that is valuable for policy. Secondly, although the workshop focused on Medicare costs, NIH’s main concern is with health, and increasingly with value. There has been a heavy investment in the Department of Health and Human Services in comparative effectiveness research, much of which is heavily constrained by definitions that exclude costs and cost effectiveness. But NIA certainly does work on costs and cost effectiveness.
Suzman next provided his assessment of the workshop discussions and what he got out of the day’s deliberations. Clearly, a large number of research issues surfaced over the course of the day, ranging from the macro issue of the impact of growing investment in the health care sector on the
overall economy and comparative disadvantage or advantage in terms of trade, to cross-country comparisons. Some of the impact has got to depend on what that investment is buying, for example, does it allow people to work longer. He stressed that at some point certainly one must connect these issues with retirement issues and forecasting retirement and labor force participation.
The value of international comparative studies seems incontrovertible based on the Gruber-Wise papers and volumes on comparing the impact of public pensions on how long people stay in the workforce and whether, in fact, old people staying in the workforce reduce jobs available for younger people (Gruber and Wise, 1999).
Studies exploring the implications of single variables, such as obesity, smoking, low birth rate, disability, in terms of short-term interventions or long-term costs, are going to be valuable.
There are multiple views on the value of studying prevention and interventions and their current and future cost. NIA has tried to stimulate research in this area but more is needed. There are some apparently effective interventions at a corporate level that reduce costs and seem to improve health risks within the space of a year. Also, CMS is in the middle of a demonstration project in that area.
Many issues around data, including costs, can only be studied incrementally. He suggested, for example, that it would be useful if the National Health and Nutrition Examination Survey had a simple telephone longitudinal component, and the Medicare Current Beneficiary Survey, which is a panel survey, had a panel of longer duration.
He noted that following the ending of the National Long-Term Care Survey, NIA has started a new study, the National Health and Aging Trends Study. Judith Kasper at Johns Hopkins University is the principal investigator. He is hopeful that it will have good economic data.
Regarding other areas and ideas that surfaced during the day, he observed that there is great value in finding out what are the NIA-appropriate long-term research needs of our sibling agencies. NIA is not very good at filling short-order requests but would certainly want to hear the long-range needs and see which of those would make sense for NIA to consider for research initiatives. He gave as an example a research project on one-time and serial high-cost users.
Finally, Suzman commented that when he considers all the topics presented and discussed during the course of the day, there perhaps needs to be a more integrated view of the issues. For example, he would include the issues related to disability and retirement and their interaction with Medicare as well as nursing home costs. So any follow-up to this workshop might perhaps have a slightly broader perspective that also includes long-term
care, Medicaid, and interrelationships with retirement policy, and perhaps even an international perspective.
He closed by thanking the presenters for the collection of short, succinct presentations that illuminated the field of health care costs for the Medicare population.
Connie Citro (Committee on National Statistics) agreed with Suzman that the presentations were uniformly interesting, informative, and clear, as well as often provocative. The workshop presentations and discussions showed that CBO and OACT are in a certain kind of business, namely, forecasting, that is not the same as the business of researchers, but there ought to be more opportunities like this workshop for these kinds of conversations.
She stated that she was particularly struck by the importance of education as a factor in health. Crimmins showed that there are some people with very poor health prospects, and low education appears to be related to poor health. Moreover, education has so many implications for so many aspects—physical health, cognitive and mental health, and the social health of the nation—that it could be very useful to build educational attainment projections into the actuarial models. Similarly, there is a body of evidence about the importance of family structure and social contacts, which could provide additional variables that could ultimately relate to health care costs.
One thing the research community can do is to assist modelers in the design of data collection. In the situation in which there is a lengthy battery of questions about, say, health status that one wants to ask in a large, observational survey, but the need to be parsimonious in the data collected as part of Medicare health care claims, researchers can identify the best one or two boiled-down questions that still carry a good deal of the explanatory power of the longer version. That is a very useful function the research community can perform.
Citro also observed that she heard very clearly, on the supply side, that better data are needed on how doctors, in particular, function, comparable to the detailed information that is available on individuals and families. Although there are surveys of business establishments, with information on payroll, number of employees, and the like, not much is known about how business organizations, including health care businesses, operate with regard to innovation, adoption of electronic records, making decisions about end-of-life care, and similar topics. That is actually a daunting agenda.
Even more daunting is capturing diversity in the population, but that is an area that NIA and other research funders might well consider. In addition, obtaining information on policy variations within the country, let alone between the United States and other countries, is a very daunting but needed task.
In conclusion, Citro noted that there was a great deal of synergy during the day with ideas and sparks bouncing off each other. Perhaps some of the agencies that attended, not only NIA, but also other agencies, will seek out opportunities to continue these kinds of discussions in terms of research and development, toward improving health care cost projections for the Medicare population. The passage of health care reform is really the beginning and not the end of all of the issues that will be coming up as people try to see what does and does not work.