Factors Affecting Health Status
This chapter examines some of the factors affecting health status that are driving health care spending among the Medicare population. The workshop presentations covered three such factors, which research shows are important to consider in projecting future Medicare costs:
health and health care cost consequences of obesity among the Medicare population;
the role of socioeconomic status and health-related behavior in driving medical care spending; and
the role of chronic diseases and disability in health care costs.
HEALTH AND HEALTH CARE COST CONSEQUENCES OF OBESITY
Justin Trogdon (RTI International) described the current costs of obesity in terms of health outcomes and spending among the Medicare population, presenting cross-sectional, lifetime, and recent trend estimates. He also reviewed different types of strategies that modelers have used to estimate costs and forecast the future, both for the prevalence and health consequences of obesity and how obesity impacts projections of Medicare spending.
Health Consequences of Obesity
Obesity has the attention of researchers and policy makers. It is associated with increased risk for many chronic conditions, such as hypertension, high cholesterol, cardiovascular disease, and cancer, among others. Obesity impacts nearly every major system in the body. It is, in itself, an outcome of several different behavioral and policy decisions; it is also an input into chronic disease.
Improved treatment for many of the conditions associated with obesity means that, in some sense, an obese person in 2010 is “healthier” than an obese person in 1950 or probably even 1980. That is good from a morbidity and health status point of view, but such improvements have been achieved often at increased health care cost. Statins to reduce cholesterol and other drugs to lower blood pressure, for example, are not cheap. Also, improved treatment may or may not lead to major changes in health outcomes such as mortality.
Years of Life Lost
Does obesity shorten life expectancy? Trogdon summarized research he and colleagues conducted in which they calculated years of life lost associated with obesity using life tables by weight categories1 and smoking status. They found that overweight and moderate obesity (obese I) will not shorten a person’s life. Severe obesity (obese II and III) will take years off one’s life. At age 65, a typical age at which people enter the Medicare program, being an obese II person (about 40 pounds overweight) is associated with 3 years of life lost for whites, while being an obese III person (a little over 100 pounds overweight) is associated with 4 to 6 years of life lost across gender and race (Finkelstein et al., 2009a). These findings indicate that although rates of chronic conditions, such as diabetes and hypertension, are higher among obese people, they do not necessarily translate into a shorter life span.
Health Care Costs of Obesity
How much does obesity increase health care costs at a given point in time? To answer this question, Trogdon reported on the findings from a recently published update of national estimates of annual medical spending attributable to obesity (Finkelstein et al., 2009b). Comparing a Medicare beneficiary who is obese to one who is not obese and controlling for other
differences between those two groups, Trogdon and colleagues estimated that obesity increases per capita Medicare expenditures by about $1,723 per year. Those dollars, for the most part, go to treating all of the chronic diseases that obesity is associated with and not just direct treatment for obesity. This estimate means that the annual medical burden of obesity is nearly 8.5 percent of total annual Medicare expenditures. If the 8.5 percent estimate, generated using data from the Medical Expenditure Panel Survey (MEPS), is applied to all of the national health expenditure accounts, assuming the institutionalized population has a similar share of medical spending going toward obesity, the Medicare costs associated with obesity could be as high as $85 billion per year in 2008 dollars.
Another way to look at the cost of obesity is to ask the question: How do medical care costs associated with obesity vary over a lifetime? This question points to the importance of preventing obesity for the Medicare population. Using the life tables described above along with MEPS data, Trogdon and colleagues estimated survival-adjusted lifetime obesity cost estimates (Finkelstein et al., 2008). One of the major conclusions as a result of that work is that, compared with the private insurance market, Medicare has potentially a greater incentive to prevent obesity because costs attributable to it are near their peak around the age of entry into the Medicare program (ages 60 to 65). What that means for the private insurance market is that, often, the major costs of obesity are not likely to be its problem. The likelihood that a potential cohort of employees would still be on the company’s health insurance rolls after 5 years might be relatively low.
Thorpe and colleagues (2004) examined recent trends in the health care costs of obesity. To answer the question of how much of the increase in medical care spending over the last 10 to 20 years is due to obesity, they estimated medical costs and obesity-attributable health care costs in 1987 and 2001. They found that obesity-attributable costs increased per capita medical care spending by about $300, accounting for about 27 percent of the increase in per capita medical care spending between 1987 and 2001. This percentage has been relatively stable over the past 5 or 6 years.
Forecasting the Future of Obesity Prevalence
Currently, obesity is important to the Medicare program from a cost perspective, both on an annual basis and as each cohort of Medicare beneficiaries ages through the system. Obesity has contributed greatly to increases in health care costs.
The current prevalence of obesity in the adult population is about 30 percent. What is going to happen to the prevalence of obesity moving forward in time? This is a much trickier question. Unlike predictions of health care spending, obesity prevalence has a natural limit—100 percent. Models
that merely project recent trends into the future will predict that everyone will be obese, and it is just a matter of when. That does not seem to be the most likely occurrence, but one has to think about when and how those trends would turn around.
Several recent attempts have been made to project obesity prevalence, but none is specific to Medicare. For example, in California, obesity rates are projected to increase from 24 to 35 percent of the adult population between 2010 and 2020 (van Meijgaard et al., 2009). At the national level, there are other estimates that have been published over the last 2 to 3 years. The predicted prevalence of obesity in 2020 is estimated at around 42 percent (Ruhm, 2007) and 44 or 45 percent (Wang et al., 2008). That amounts to about a 10 to 15 percentage point increase in obesity prevalence, which is an additional 50 percent increase over the current level.
Thorpe, in a recently released report (2009), also projected the estimated prevalence of obesity for the period 2008 to 2018. His midpoint estimates of the prevalence rates for the 10-year period are also around 42 to 44 percent of the adult population. Projecting the obesity-attributable costs over a 10-year time horizon using extrapolation, he found that obesity-attributable health care spending could range from $864 per capita in Colorado to $1,906 in Oklahoma. It should be noted that these are total and not Medicare-specific costs. Total obesity-attributable health care spending in the United States was projected to increase from $79 billion in 2008 to $344 billion in 2018. If one is willing to extrapolate past trends forward, these estimates suggest that the United States may be in for a much higher obesity prevalence and increased costs in the future.
Factors Influencing Future Obesity Rates
Most forecasts of obesity, both for prevalence and costs, are extrapolation of past trends. Even when a microsimulation model is used, there is still an assumption of past trends continuing on. Whether that is a reasonable assumption will depend on policy and technological changes in the food supply and medical care systems or both, all of which will influence the future of obesity over the next several years.
Food system changes include policies on food prices, taxes, subsidizing certain products, labeling requirements, and nutrient rules such as transfat bans. The question is: If there are structural changes in some of these underlying policies, how would that impact obesity prevalence rates? A simple extrapolation is not going to answer that question.
Changes to the medical care system could change the prevalence rates of obesity. However, such changes are not necessarily going to slow the growth in obesity prevalence—some changes might actually accelerate them. Some examples are technologies to treat obesity with surgery and lap bands, sev-
eral prescription drugs specifically for weight control that are being tested for federal approval, and even some of the technological advances that would treat obesity comorbidities. These could have behavioral impacts on incentives for people to control their weight. For example, if it is relatively cheap for a person to be treated for hypertension and cholesterol by just taking a pill, there is less of an incentive to be concerned about weight and diet. These behavioral impacts will have implications, especially for Medicare, when all of the obesity-attributable costs start coming in.
Trogdon concluded his presentation by stating that, based on current knowledge, it is likely that there will be continued increases in obesity over the next 10 or 20 years.
SOCIOECONOMIC STATUS AND HEALTH-RELATED BEHAVIOR AS FACTORS IN MEDICAL CARE SPENDING
Eileen Crimmins (University of Southern California) opened her presentation by observing that socioeconomic status (SES) is a fundamental cause of health differences in the population. The United States is a society of haves and have-nots. Large differences exist between these two groups, and the SES distribution of the population relates to people’s health status. Socioeconomic differences in health exist all over the world; they tend to be larger in the United States than in other countries. They are omnipresent over geography, and they also have been present over time.
Some differentials by SES in health outcomes have been relatively stable over time. People with lower SES—low education or low income or low occupation—have worse health by almost all health indicators. They have more diseases, physiological risk indicators, disability, and physical and cognitive functioning problems. Socioeconomic status in and of itself is a fundamental cause of health problems that works through many mechanisms to affect health. It can affect health outcomes through health-related behaviors, knowledge and skills obtained through education, and the ability to use income and wealth to purchase things that affect health. People with higher incomes are more likely to have access to care, a regular provider of medical care, and health insurance coverage. Social-psychological differences, differences in depression and stress, and health care access affect health. Thus, health outcomes differ by SES, and these differences affect differences in health care costs.
In models of use of health care services, the inclusion of such health indicators as disability and diseases tends to eliminate, or greatly reduce, the effect of SES variables. Cost is yet another issue because costs are affected by geographic location and the characteristics of the environment in which a person lives, not just the characteristics of the individual. To a large extent, most of the differences in costs for people with different SES come from either observed health differences or the different places they get care.
Incorporating SES in Projection Models
Is there some way to consider SES in order to make better cost projections? For example, are changing education levels in the population, or a changing set of differentials within the population, something that needs to be incorporated in models in order to make better projections?
Crimmins presented research findings to show the significance of SES differences in health outcomes. Using data from the Health and Retirement Study (HRS), Banks and colleagues (2006) looked at the prevalence of a set of diseases by three levels of education—low, medium, and high—among non-Hispanic whites ages 55-64. For heart problems, hypertension, stroke, diabetes, chronic lung disease, heart attack, and all of these conditions combined, lower education status was associated with higher prevalence.
This association holds for all kinds of measured risks. Using data from the National Health and Nutrition Examination Survey (NHANES) for the period 2001-2006, Karlamangla and colleagues (no date) looked at SES differences in metabolic syndrome and 10-year global chronic heart disease risk and found a much higher prevalence of poor scores among low SES people. In terms of risk factors, the data from HRS on the percentage of people ages 50 and older who were obese, current smokers, and heavy drinkers by education level showed that those at the highest education level had the lowest number of risk factors and those at the lowest education level had the highest number.
SES Differences in Health Outcomes and Age
There is no question that higher rates of ill health are found among people with low SES; however, these differences vary by age. The age at onset of the deterioration in health varies by SES; problems arise earlier among those with low SES. The maximal point of difference is at older working ages; at very old ages, they disappear or are reduced, at least partly because of mortality.
One of the more important points that comes out of RAND’s Future Elderly Model (FEM) is that people who survive to old age are different from those who do not survive. People with relatively high SES survive longer and people in better health survive longer; people with low SES and those with poor health do not survive as long. That is the key to thinking about how one needs to incorporate changes over time in a model that projects health care costs. For purposes of modeling, one has to think about the timing aspect and a life-cycle effect aspect, rather than just looking at prevalence and modeling it forward.
Health events are age related; for most health problems, at the age of Medicare eligibility, low SES people are going to have more health prob-
lems but a shorter expected length of life. Crimmins observed that life expectancy at age 60 for people with low, medium, and high SES shows a difference of about 5 years. In order to understand health differentials and their ability to change overall costs, one has to figure out how long people in different SES groups live, how many years of that life are spent unhealthy by a variety of definitions, and the cost of an unhealthy year. Costs need to be determined to understand how technology and policies will change the age at onset of health conditions, the length of survival with conditions, and the overall length of life.
SES and Healthy Life Expectancy
Data from various sources show that social and economic differences in health and mortality result in more years of ill health, fewer years of healthy life, and lower life expectancy overall, for people with low SES status. These differences arise from a process of earlier onset of health problems and higher mortality. The effect of this process of health deterioration on differential population health depends on where in the process of health deterioration the change occurs. Increasing the average length of life can have relatively little change on the distribution of population health. If healthy and unhealthy life are both increased at the same time, population health may not change much at all. At the same time, the length of an individual’s healthy life may increase.
Crimmins explained that changes in population health characteristics and the life-cycle characteristics of individuals can be different, and they tend to get mixed up when people think about improving population health. For example, reducing deaths from heart disease may increase the prevalence of heart disease in the population, as well as its costs. The prevalence of disease in the population can increase because of success in lengthening the life span of people with disease.
To improve the health of the population, what needs to be done is to delay the age of onset of conditions and reduce the time with health problems. This has not happened much yet; instead, the time with health conditions has been increased. That is one of the reasons it is important to think about years with conditions and years in good health.
Trends in Health
Both the incidence and the prevalence of disease in the population have increased. In most cases, the prevalence of disease has increased because of the decline in mortality, with little or no change in the incidence or rate of disease onset. Diabetes may be a different case because the rate of onset has increased. Changes in every disease need to be looked at differently.
The prevalence of diagnosed risk factors, such as hypertension and high cholesterol, has increased. Yet disability has declined in the older population; physical and cognitive functioning and ability to work have improved. Although for some people who have disease, the progression to either becoming disabled or dying has been delayed, the underlying diseases have not been eliminated.
Rising education has been a major force for improvement in health over time, primarily in the area of disability. A number of recent papers have essentially attributed at least 50 percent of the decline in disability to change in the education composition of the population (see, for example, Freedman and Martin, 1999; Schoeni, Freedman, and Martin, 2008). This means that health processes have not changed in the population, but that the composition of the population has changed, with more people in the better educated group. Over the long run, that has been an important factor in increasing life expectancy. It is not clear, however, that this factor will continue to operate in the same way into the future, because in recent years the increase in education at older ages (60-69) is starting to slow down in the younger population (ages 50-59).
Crimmins and a colleague found that in the 10-year period, 1997-2007, the number of people unable to work and those limited in their ability to work at age 60 have declined (Reynolds and Crimmins, 2009). Rising education has been a force for improvement in disability. In general, SES differences have not changed much over long periods of time, but over time the more educated population comprises a greater percentage of the total population. There is also some evidence of widening of SES differentials in mortality in recent years, which could be a short run or long run trend (Jemal et al., 2008). That is one reason why it is important to understand changes in how SES is linked to health outcomes, because there is now a wide difference between the lowest and the highest SES groups. If the lowest group were to change to be like the highest group, there would be a substantial increase in the number of people who would need to be covered by Medicare, increasing the health care costs for the total population.
In summary, Crimmins emphasized that an important national aim is to reduce health differentials. Reducing mortality differentials and reducing differentials in age at onset could have different effects for population health. A lot more detail on the processes of health change is needed to better understand what is underlying the observed differences in the population prevalence of health problems.
Microsimulation can be used to address these processes. That does not mean one has to incorporate a simulation of changes in health status into major national projection models. Yet to understand the role of a given factor for health status, microsimulation of all of the processes involved is
needed, and the more detailed the simulations, the more one can understand the processes.
Finally, Crimmins observed that some things are known about cohort change in SES, but this research relies on cross-sectional, time-related data rather than cohort data. Clearly, there is need for more information on lifetime health circumstances to understand changes in health outcomes. Today many diseases have a life span of 20, 30, or 40 years, with long spans of treatment. The onset of risk factors and treatment can start very early in life. For example, the implications of being treated for hypertension or high cholesterol for 30 or 40 years, in terms of mortality and cardiac events, are not understood. In order to better understand the future implications of cohort characteristics and experiences, it is necessary to have more lifetime models of health rather than models that are based only on recent cross-sectional data.
DISABILITY, CHRONIC DISEASE, AND MEDICARE SPENDING
Jay Bhattacharya (Stanford University) opened his presentation with two general observations. The first purpose of forecasting models of health care expenditures is to alert Congress and other policy makers about problems in the outlying years. A second and related purpose is to answer counterfactual questions about what will happen if various events (such as the development of new medical technologies) should occur. Both purposes, but especially the second, require that the forecasting apparatus adopt an underlying theoretical idea about the primary drivers of health care spending. In his presentation, Bhattacharya proposed the development of chronic disease and the competing risks phenomenon as the theoretical ideas driving health care expenditures. A forecasting apparatus centered on these ideas is well positioned to answer counterfactual questions about the effect of changes in health status on future health expenditures.
Bhattacharya mentioned a working paper by White (2006, later published in 2008) that noted a slowdown in the growth of Medicare expenditures between 1997 and 2005 relative to previous years. The paper attributed this slowdown to a new prospective payment system for hospitals and postacute care providers and to limits on the growth of payments to physicians. If these reforms could be maintained and extended, then the future financing of Medicare would not be so bleak. However, Medicare expenditures grew at more than 8 percent, compared with 4.4 percent growth in overall health care expenditures nationwide, despite the continuation of payment reforms. When looking forward into the future, it is therefore important to understand the underlying processes that drive health care expenditures.
Ken Manton and his colleagues at Duke University, in a series of papers, have shown that disability rates among the elderly have been declining since the 1980s and that disability is an important driver of health care costs. In their analysis of data from the National Long-Term Care Survey, Manton, XiLiang, and Lamb (2006) found that, in 1982, 5.7 percent of the elderly population was unable to perform instrumental activities of daily living (IADLs), whereas in 2004, this proportion was only 2.4 percent. With the exception of the prevalence of severe disability (inability to perform 3+ activities of daily living, ADLs), a similar and even more dramatic decline was observed for ADLs (Manton et al., 2006). These findings show a reversal of the trends of the 1970s, during which disability prevalence was increasing, and the decline accelerated in the 1980s and 1990s. Combined with increasing life expectancy, these declines yield a compression of morbidity. If these trends toward declining disability among the elderly continue, then Medicare expenditures could be substantially lower than is currently expected. But will these trends continue?
Bhattacharya argued that there is good reason to believe that the trend toward decreasing disability will not continue. He and his colleagues have found that disability is increasing in the under-65 population (Lakdawalla, Bhattacharya, and Goldman, 2004). Their analysis of data from the National Health Interview Survey (NHIS) on disability prevalence for 1982 to 1996 replicated the findings of Manton and colleagues of declines in disability among the elderly. At the same time, they found that younger populations, ages 50-59, 40-49, and 30-39, were experiencing substantial increases in disability.
What caused the change in disability prevalence among older people? Was it chronic disease prevention? Or was it better management of chronic disease, such as the availability of breakthrough technologies and assistive devices? Or was it more educated people? Which of these factors was more important?
He explained that chronic disease is directly relevant to policy. The chronically ill are more likely to become disabled. The policy choice for focusing resources is between reducing the prevalence of chronic illness or, once people are chronically ill, preventing them from developing disabilities. Understanding which of these approaches has played an important role in past improvements of disability trends may therefore inform what could be expected in the future.
A lifetime perspective is essential to understand the implications for medical care expenditures. For example, a decline in the prevalence of chronic disease would reduce the prevalence of disability and lead to declines in associated medical expenditures per year. But longer life may lead to greater expenditures. The costs are higher for prevention, which is more expensive in part because one does not know who is going to get a disease.
Chronic disease management, in contrast, leads to a decline in disability prevalence among the chronically ill, but incurs higher expenditures on assistive technologies.
Disability, Survival, and Medical Expenditures
Bhattacharya next described the relationship between disability, survival, and medical care expenditures (Bhattacharya, Garber, and MaCurdy, 2010). He and his colleagues analyzed data collected annually from 1992 to 2003 in the Medicare Current Beneficiary Survey for people ages 65 and older with and without disabilities. They linked these data to Medicare administrative records for comprehensive measures of all medical care expenditures except prescription drugs. They found that survival of a person with disabilities is affected by the age (65, 75, or 85) at onset of disability. Medical care expenditures of elderly people with a disability are considerably more than those without disabilities, thus raising lifetime Medicare expenditures. Yet the disabled elderly have higher mortality rates, which would lower lifetime Medicare expenditures. The timing of disability onset therefore has a major effect on survival as well as Medicare expenditures.
Disability and Chronic Disease
Disability prevalence can be decomposed into two parts: one part attributable to the chronically ill population and a second part attributable to the nonchronically ill population. Changes in disability prevalence among the chronically ill can be decomposed further into two parts: changes in disability prevalence among the chronically ill and changes in the prevalence of chronic disease (Aranovich et al., 2009). Bhattacharya cautioned, however, that disease-by-disease decomposition may double count people with multiple chronic conditions, leading to an overestimate of the importance of chronic conditions in explaining disability trends. He argued that his research team’s estimates adjust for this double counting for the most common chronic diseases.
In their analysis of data from NHIS, Aronovich and colleagues considered the most common chronic conditions afflicting elderly populations: arthritis, chronic obstructed pulmonary disease, diabetes, hypertension, heart disease, stroke, and obesity. They found that based on data from NHIS, except for overweight and obesity (which increased sharply), chronic disease prevalence rates stayed mostly about the same or improved between 1982 and 1996 and hence did not contribute substantially to the decline in elderly disability over that period. For example, in 1999 there were fewer people with arthritis per 10,000 elderly individuals than there were in 1982. Similarly, prevalence rates for hypertension and heart disease were lower
in 1999 than in 1982. By contrast, the rise in obesity prevalence over that period, if not countered by some other factor, would have led to a rise in disability in the elderly population.
Unlike overall chronic disease prevalence, disability prevalence among the chronically ill elderly improved substantially between 1982 and 1999. This decline more than countered the increase in disability due to increases in obesity prevalence and led to the overall decline in disability observed in the elderly population. Advances in medical technology played an important role in managing and reducing disability among the elderly. For example, new pharmaceutical products that control the progress of arthritis, better pain relievers, and joint replacement surgery helped reduce disability. Likewise, more intensive medical and surgical management of heart disease, reduced smoking rates, newer portable supplemental oxygen tanks, and specialized pulmonary rehabilitation centers may have contributed to declines in disability.
How much of the overall disability trends is attributable to the prevalence of chronic disease and how much is attributable to disability prevalence conditional on chronic disease? The analysis of Bhattacharya and colleagues suggests that disability declines among the elderly are mostly not due to improvements in primary prevention of chronic disease, but rather to preventing disability among the chronically ill. Much of the decline in disability among the chronically ill involves IADLs. Such declines, which often involve the purchase of expensive assistive devices, can result in higher Medicare expenditures.
Disability and Chronic Disease in Younger Populations
Younger populations tell a different story. Bhattacharya and colleagues used the same methods they used for the elderly for decomposition of disability trends among the younger population (Bhattacharya, Chowdhry, and Lakdawalla, 2008). They found that, between 1984 and 1996, disability prevalence among people under age 65 had increased, in sharp contrast to the decline in disability prevalence among the elderly over this period. About half of this increase in disability was attributable to prevalence of chronic diseases, much of which was in turn attributable to obesity. The remainder was attributable to an increasing rate of disability among the chronically ill, including people with hypertension or chronic obstructive pulmonary disease. Among the nonchronically ill, disability rates actually fell. The main implication of this work is that younger populations are not becoming healthier. Disability prevention efforts, if they are to be successful, should focus on reductions in obesity prevalence and limiting disability among chronically ill populations.
Predicting Future Medicare Expenditure
What will be the health care status of the population 30 years from now, and how is medical technology going to affect it? What effect is that going to have on medical care expenditures?
Bhattacharya turned to projections from RAND FEM, commenting that the model is ideally suited to answer questions like this. FEM is theoretically oriented toward chronic disease and health care costs and has been used to look at three prevention interventions in this context—smoking cessation, obesity control, and diabetes prevention—to project cost savings to Medicare. It also includes information on disability. The researchers found that disability declined sharply among the elderly between 1982 and 1999, similar to the findings noted above. Prevention of disability among the chronically ill played an important role in the decline; primary prevention of chronic disease was less important.
Among the younger population, disability increased over the same period. Higher prevalence of obesity and higher rates of disability among the chronically ill contributed to the increase. Consequently, future Medicare expenditures may not decline by much, even if future disability rates decline.
Bhattacharya concluded that disability is a major driver of health care costs, but eliminating it is not necessarily a major way to improve future health care expenditures for the Medicare population. Also, primary disease prevention is not a major cost saver in future health care expenditure projections. Preventing disability may nonetheless be the right thing to do, as it will allow people to live in a nondisabled state for a longer time and improve their quality of life.
Several participants expressed their views on the various issues flowing from the presentations. Most of the discussion was broadly on measuring socioeconomic status in modeling, projecting costs of medical treatment, and data for improved health care cost estimates.
Measuring Socioeconomic Status in Modeling
Referring to the discussion by Crimmins about socioeconomic status, which focused mainly on education and her statement that it did not matter much whether one measured SES by education, income, or occupation, Joseph Newhouse (Harvard University) interpreted that to suggest that the measures were treated as causal. He had a two-part question: First, what is known about causality? Second, from the point
of view of modeling the future, and assuming that there will be changes in the distribution of the population by education as well as by income and occupation, would it matter which of those measures are causal, or are they all causal?
Crimmins responded that she views education as a fundamental cause that determines income. Income is a lot more complex as a variable because the causal relationship is much more likely to be a two-way street. As one gets sick and leaves the labor force, or one does not work as long, one’s pension will be reduced. Particularly at older ages, there is a lot of reverse causation in the income and health relationship. There certainly is some reverse causation using education in terms of people who become ill before the period when educational attainment ends, which tends to be in the twenties. These people have less educational attainment, but the effects on health tend to be small and not to lead to the diseases and conditions of old age.
The differences among population groups are always there, but they look slightly different depending on the SES measure used. Current occupation is a pretty useless measure for older people because most of them do not have one, and a lot of women never had one, although that is changing. The relationship with health is easier to understand if SES is indexed by education. Certainly going forward with a time path, one knows the educational attainment of the older population for the next 50 years, so it is a reasonably stable variable; in contrast, one does not know about income and how that is going to change over time.
Richard Suzman (National Institute on Aging) observed that not enough attention has been given in the presentations or discussions to people’s work patterns. Given the trend of being healthier and living longer, people are going to have to work longer. He therefore thought that combining Medicare projections with retirement modeling in both the United States and cross-nationally might be useful.
He mentioned that there was a lot of talk at the workshop about the short-term and long-term advantages of prevention coupled with costs. Essentially, longer life is not free; it has to be financed in some way, but there are relatively few data sets that look at the downstream impact of prevention or major medical investments over the rest of the life course. That is an important area to consider.
Bhattacharya commented that issues relating to work are important, especially in the context of disability and changes in disability trends in the younger population because disability has effects both on health care expenditures and on financing. So if a larger share of the younger population is disabled and therefore less able to work and retires earlier, the financing models are going to be off in addition to the expenditure models.
Projecting Cost Estimates of Medical Treatment
Referring to Bhattacharya’s discussion of predicting future medical expenditures, Michael Chernew (Harvard University) wanted to know, when forecasting medical spending and looking at cost effects, if the cost estimate of, say, treating a hypertensive patient or a disabled patient in 2020 is like a life table, using the cost of treating that hypertensive patient today, or if there is some growth rate beyond regular inflation to get to that point. If so, how does one inflate the cost of treating a current hypertensive to treat the hypertensive in 2020? To project the number of people with disabilities for a short time period, say 20 years, one can use the number with disabilities who are age 20 now and project out. But for longer term time periods, how does one project out the number of people with disabilities in 2050 or 2070?
Bhattacharya responded that in FEM the researchers assume that there is existing technology for everything. So they do not change anything other than the probability of transiting into obesity, for example.
In response to Chernew’s second comment, Bhattacharya explained that the method used in FEM is to look at the whole population—the transitional probabilities from age X to age X + 1 are fixed—and then age people forward. So if there are higher rates of disability among 30-year-olds today, that means there are going to be higher rates of disability for the entering cohorts at age 65, 35 years from now. But the transition probabilities from age 30 to 31, 30 to 32, etc., are just as in a life table based on today’s estimates.
Data for Improved Health Care Cost Estimates
Dana Goldman (University of Southern California) observed that the work of Crimmins and other research suggest that early determinants matter for future morbidity and mortality. That highlights the need for longitudinal panels. In addition, better cost estimates are needed, because trying to get self-reported cost information is almost impossible and leads to the need for linked data. HRS has linkage with Medicare records, but it is difficult to get those data. Although the Medicare Beneficiary Survey is available, it does not ask any of the SES questions that go back in years. It is very hard to link the household component of MEPS, and it does not include the institutionalized population. So the question is, What is needed in terms of data to improve these health care cost forecasts?
Liming Cai (National Center for Health Statistics) responded that the National Center for Health Statistics (NCHS) has provided several unique data sets that can be used for forecasting purposes. NCHS has linked data
in the national surveys, NHIS, and NHANES, to the National Death Index (NDI), Social Security, Medicare, and Medicaid down to the end of this year. So one has the health measures, the socioeconomic and demographic measures, and all of the other survey measures available in a particular panel, and these panels are linked down the road through the mortality records, the claims records of the Centers for Medicare & Medicaid Services (CMS), the Social Security earnings record, and the Medicaid claims records. The impact of trends in health, by demographic and socioeconomic factors for the entire set of entitlement programs, can thus be estimated.
These useful data sets are currently available at NCHS, but the user has to submit a research request to an NCHS research data center to use them. There are research data centers across the United States; the user does not have to go to NCHS in Hyattsville, Maryland, to do the research.
Crimmins countered that she has used the NHANES extensively, but many of the important variables are not there. Neither cognition nor depression are measured in NHANES. Early life is not measured at all. NCHS has relatively poor measures of lifetime experiences. So there is intensive information from the National Death Index and Medicare, but the independent variables are lacking. The answer to Goldman’s question is therefore a composite of data sets, because no existing data set is perfect. Lifetime information is needed, but several early life measures are missing from current data sets.
Cai responded that there are certainly some topics missing from the surveys. At the same time, for some important research topics, such as obesity, no matter what disability status a person has, lifetime health care spending is probably the same (Lubitz et al., 2003). Cai and his colleagues looked at obesity status at around age 45, using the first NHANES followup survey linked to Medicare and the NDI, and obtained their lifetime Medicare expenditures. While more obese 45-year-olds will die before reaching age 65, their lifetime spending from age 65 on for Medicare is still significantly higher than normal-weight 45-year-olds who survive to age 65 and beyond.
Crimmins remarked that some of the emphasis on obesity makes her nervous. The link between obesity and socioeconomic status was extraordinarily strong in the past, and so some of what is being interpreted as an effect of obesity could be an effect of low SES. Without a comprehensive model that includes both obesity and SES, there is a risk of misallocating the effect.
Cai further pointed out that measures of SES, such as education, are not available in census population projections from 2002 to 2050. So although education is important to understand the relationship, when projecting out 50 years, that variable is not available for a projected population.
Bhattacharya noted that a theoretical idea is key to forecasting. If the
idea is an extrapolation, then one can make do with expenditure cross-sections. If the idea is changes in disability, in obesity, in educational status, then one needs some sort of longer panel. There is a fundamental trade-off in that the longer the panel, the less representative it is of the population as a whole. So ideally one would want a long panel refreshed routinely to make it look more like the population at large.
Todd Caldis (Centers for Medicare & Medicaid Services Office of the Actuary) pointed out that the long-term models of both the Office of the Actuary and the Congressional Budget Office already include crude adjustments for the level of population health risks. In principle, it would be feasible to incorporate into those models more sophisticated measures.