Modeling Medical Technology
This chapter focuses on issues in modeling medical technology as a driver of Medicare spending. The presentations covered three topics:
The first presentation provided context by examining the implications for health care cost projections under the assumption of current law regarding payments and benefits. Given that health care costs are growing faster than the gross domestic product (GDP), the question is what factors will slow down that growth.
Innovation in medical technology, which has been estimated to account for about half of health care cost increases over the past 50 years, is a prime target for research to identify policy changes that could moderate cost growth. The second presentation focused on innovation in pharmaceuticals, the Future Elderly Model (FEM) was used to estimate the effects on rates of innovation and the consequences for health care spending and health outcomes from two policy scenarios designed to slow cost growth from prescription drugs.
The third presentation assessed the relative merits of different modeling approaches for estimating the effects of advances in medical technology as a driver of health care costs.
CURRENT LAW BRAKES ON HEALTH CARE COST GROWTH
Michael Chernew (Harvard University) addressed concerns about the assumption of current law in forecasting models, which does not square with the policy need to put brakes on runaway health care spending growth. He opened his presentation with a general statement that speeding objects clearly need something to slow them down, and the same applies to health care spending in this country. It is understood that the rapid growth in health care spending cannot continue and that if it does not slow down, there is a problem. The question is: What factors are likely to slow it down?
The Congressional Budget Office (CBO) and the Office of the Actuary (OACT) generate spending projections under the assumption of current law. Chernew emphasized that these are not forecasts in the sense that no one expects actual spending to match the projections. To treat them as forecasts does them a disservice. They are designed to show what would happen if current law did not change and to warn policy makers of the consequences of inaction. He observed that a fundamental issue is what is meant by current law. For example, is it the current benefit structure, that is, how much spending would go up under the current benefit structure? Or is it prices, the current payment rates, or, more broadly, the laws regarding payment rates?
In general equilibrium models, prices adjust and spending growth in any sector slows down because there is a budget constraint that individuals face. However, institutional features of the health care system that are embodied in current law, such as public financing of care and administratively set prices, weaken the budget constraint. As a result, although general equilibrium models may work well for forecasting in other sectors in which prices and incentives are not distorted, they may not work as well for the health care sector.
There are other questions regarding current law and financing. Even though it is understood that people are not going to spend 80 percent of GDP on health care and that the nation certainly cannot finance 80 percent of GDP for health care, making adjustments to projections to achieve a sustainable level of spending may do a disservice to policy makers by not warning them of impending danger.
The key point is that current law weakens many brakes on health care spending. The costs of care are heavily subsidized, and these subsidies weaken the budget constraints that individuals would otherwise face.
What Will Slow Medicare Spending Under Current Law?
Chernew identified two principal factors that may slow the rate of growth of health care spending in the future. The first is cost sharing—that
is, the effect of existing benefit limits on Medicare spending. The second is spillover—that is, the effect of cost containment in the private health care sector on Medicare spending. If cost growth in the private sector slows because it does not face current law, how will that influence Medicare spending? Will there be divergence, or will there not be convergence? And what will be the distributional effects?
The Medicare benefit package is incomplete. There are deductibles in Part A and coinsurance in Part B. There is cost sharing in the form of a set of copayments and the “doughnut hole” in Part D. Most people obtain supplemental coverage to shield them from the gaps in Medicare. Some of that supplemental coverage is provided by employers, some of it is individually purchased in the Medigap market, and some beneficiaries join a Medicare Advantage plan, which is able to finance coverage of many items not otherwise covered because these plans historically have been paid generously by the Medicare program. A beneficiary who is dually eligible for Medicare and Medicaid could also get around the gaps in Medicare benefits.
A series of laws exist that govern the payment to providers of the Medicare fee-for-service system, payment for health plans in the Medicare Advantage program, payment for prescription drugs in the Medicare Part D program, and eligibility for Medicaid. Under current law, spending will reflect equilibrium based on the law. As health care spending rises, participation in supplemental coverage may decline. For example, employers may drop coverage, exposing workers to a greater share of the cost of health care. Similarly, Medigap premiums may rise, which is likely to result in individuals dropping Medigap coverage. Finally, payments to Medicare Advantage plans may decline, resulting in dropped benefits.
The decline in supplemental coverage will expose individuals to gaps in coverage that are inherent in the Medicare program, forcing people to face more of the costs of health care. The question then becomes how much and when will these mechanisms slow spending growth in Medicare.
In theory, cost sharing generates income effects that slow spending growth as spending consumes more of income. This effect will be more pronounced among low-income beneficiaries who are not receiving large government subsidies. Chernew noted that he is working with Tom McGuire on a study funded by the National Institute on Aging attempting to quantify this effect, but results are not yet available.
The second major factor that will likely slow spending is spillover effects. The basic idea is that spending will slow in the non-Medicare health care sector, resulting in more conservative practice patterns and less abundant infrastructure. With the projected health care spending growth, it becomes infeasible for employers and others to continue to pay an ever-increasing share of the cost of health insurance. The question is how the slowdown in the private health care sector is going to affect Medicare.
There is a potential for positive spillover (in which slower non-Medicare health care spending reduces spending in Medicare) and cost shifting (in which providers try to recoup losses in the non-Medicare sector by increasing costs for Medicare). In the short run, predictions about the nature of spillover effects are ambiguous. Spillover models may reflect commonality in practice styles, in which slower non-Medicare health care spending also slows spending by Medicare. In contrast, cost shifting may apply in the short run, suggesting greater pressure on Medicare budgets as providers try to recoup losses in the commercial sector.
In the long run, infrastructure issues will become increasingly important. If the commercial sector becomes much less generous, the ability of hospitals and other organizations to invest in many types of infrastructure changes. This will tend to slow Medicare spending as non-Medicare spending slows.
Chernew emphasized that there is extensive evidence that positive spillover effects exist in health care spending (Baker and Shankarkumar, 1998; Baker, 1997, 1999, 2003; Chernew, DeCicca, and Town, 2008), in practice patterns (Baker and McClellan, 2001; Bundorf et al., 2004; Heidenrich et al., 2001), and in production functions. That is, as the Medicare Payment Advisory Commission has suggested (2009), if the commercial sector becomes less willing to fund increasing costs, hospitals may actually become more efficient. Regarding infrastructure spillover, there are a number of studies that indicate that the effects are going to be important (Baker and Wheeler, 1998; Chernew, 1995; Chernew, Gowrisankaran, and Fendrick, 2002; Finkelstein, 2007). Ultimately, however, there is not yet sufficient evidence to identify the magnitude of these spillover effects in the future.
Existing current law forecasts are therefore inherently speculative. OACT and CBO continue to refine models to incorporate these factors and provide a more accurate picture of what the future may hold to help guide policy makers.
PHARMACEUTICAL INNOVATION, SPENDING, AND HEALTH
Darius Lakdawalla (University of Southern California) focused on pharmaceutical innovations and the implications for health care spending and health outcomes in the population. He briefly described research undertaken by him and colleagues flowing from FEM.1
Technology is a major driver of health care spending. In the last 50 years, roughly half of the key factors behind increases in total medical spending are attributed to technology, while other factors, such as aging, income, insurance, and prices, have played smaller roles. A natural follow-on question to this is: What explains the advances in technology? The next step in the research frontier is to think about technology in the way one previously thought about spending and incorporate it into the process of modeling. Both health care spending and technology are determined by a host of underlying factors—demographic changes, economic growth, and health care policies among them. The challenge from a modeling perspective is to understand what policy choices lead to socially advantageous paths for both technology and spending. These are coevolving trends and understanding how they evolve together should help to clarify beneficial ways in which to influence both costs and technology.
Lakdawalla and colleagues analyzed this issue in the context of pharmaceutical innovation, spending, and health: What causes innovation in the pharmaceutical sector? How does government policy for pharmaceuticals affect innovation, spending, and health through its effects on pharmaceutical discovery and pharmaceutical utilization?
Innovation Creates Social Trade-Offs
Basic economic incentives have to be incorporated into a model of general innovation in the health care sector, not just pharmaceutical innovation. Innovation creates social trade-offs. Higher prices for innovations lead to more research. Innovators respond to incentives, working harder when they expect more rewards. However, higher prices strain public and private budgets and reduce the number of people who can use new inventions. This situation leads to a difficult trade-off between current and future generations. Lower prices save money and lead to higher use of today’s technological advances by today’s health care patients, while higher prices lead to more new technologies and treatments for tomorrow’s patients.
The innovation trade-off is acute for pharmaceuticals. Maintaining high prices for drugs today may lead to more new drugs tomorrow, but
high prices today leave some of today’s patients untreated. In some sense, this is a trade-off that pits current generations against future generations. In order to understand the implications for policy, it is essential to incorporate this trade-off into a model that includes innovation. What does this trade-off mean, on balance, for the right mix of incentives for innovation, development of technology, improvements in health, and containment of health care spending?
Regulatory Choices Have Global Effects
Innovation in pharmaceuticals and other medical technology spills across the entire globe, which makes the modeling of innovation challenging. Innovation is a global good, and the global nature of innovation creates linkages across markets. A new drug or therapy benefits patients around the world. A new drug discovered in Switzerland, for example, is going to benefit U.S. patients just as it benefits Swiss patients. High U.S. prices hurt today’s U.S. residents but may help future Americans and Europeans; similarly, low prices benefit today’s Americans but may have consequences around the globe for the future. A model that incorporates innovation therefore has to account for the fact that a country’s policies have effects beyond its borders—that is, U.S. policy changes are going to have global effects just by virtue of the fact that changes in the rates of innovation affect the treatment of patients around the world.
The approach taken by the research team, as Lakdawalla described, included
Determining how changes in pharmaceutical policy affect the rate of new drug launches.
Inferring from the best available medical and economic evidence the effect of new drug launches on health outcomes and medical spending.
Building a tool that forecasts the impact of pharmaceutical policy changes on health, life expectancy, medical spending, and patient well-being.
As stated earlier, the starting point was FEM. The team analyzed two types of policy changes:
Lowering the prices paid to manufacturers, in which the United States adopts government price negotiations estimated to lower
manufacturer prices by 20 percent. That is similar to what European governments do, but it is not allowed under Medicare Part D as it is currently configured.
Lowering the price paid by consumers, in which, instead of changing manufacturer prices, the United States lowers consumer copayments by 20 percent. This is more akin to Medicare Part D, which was designed to affect out-of-pocket spending but not to have direct effects on prices paid to manufacturers.
Using these two types of policies—changes in manufacturer prices, all else being equal, and changes in consumer prices, all else being equal—the research team built a model based on FEM that incorporated the effects of new drug launches and the effect of pharmaceutical policy on the rates of new drug discovery.
The team found that when pharmaceutical manufacturer prices are lowered, global longevity is reduced. Over the next 50 years, a reduction of manufacturer prices by about 20 percent would likely reduce longevity for U.S. 55- to 59-year-olds by about three-quarters of a year and for European cohorts around the same ages by about a half year. This finding is not surprising, given that innovators are assumed to respond to lower profits by making fewer inventions.
These baseline forecasts of the policy effects raise an important issue: there is going to be inevitable controversy and debate over any set of assumptions that underlie a model and any set of parameters that are used to configure a model. They reflect fundamentally the uncertainty of the whole modeling enterprise.
The research team had the best available estimates from the medical and economic literature, although controversy exists even about those estimates. It is important to take a range of possible assumptions from the literature—not just the best or what seem to be the best available ones or the ones that most people vote for—indeed, the entire range of entire plausible assumptions. That helps to show which policies are better and worse and how risky certain policies are, because, ultimately, if modelers face uncertainty, that translates into risk faced by a policy decision maker. That decision maker has to decide how to act in the face of uncertainty. The key issue is the riskiness of one course of policy action compared with its alternatives.
More important than the baseline effect of the two policy alternatives was the research team’s conclusion about the risk–reward trade-offs faced by policy makers looking at the pharmaceutical sector. On balance, lower-
ing pharmaceutical prices paid to manufacturers is probably a risky strategy. The cost is a potential decrease in life expectancy, both in the United States and around the world, which may be modest or very large and which will vary with the responsiveness of innovation to profits. The benefit is modest decreases in U.S. medical care spending.
In contrast, consumer copay reduction policies are robustly beneficial. Reducing patient out-of-pocket costs, without changing manufacturer prices, fosters innovation although it leads to modest increases in health care spending on pharmaceuticals. At the same time, there is a benefit in terms of increases in global life expectancy; the increases may range from modest to significant and are likely to vary with the responsiveness of innovation to profits. On balance, copay reductions appear at worst to risk modest costs, and at best to generate substantial and cost-effective gains in life expectancy.
Lessons for U.S. Policy and for Forecasting
From a policy point of view, Lakdawalla described some lessons learned from the analysis:
The general thinking behind Medicare Part D was reasonable. It lowered prices for many elderly Americans who were previously uninsured without lowering manufacturing prices. On average, Part D lowered prices faced by elderly patients by far less than 20 percent, suggesting that increasing the generosity of Part D may benefit society.
Extending drug insurance subsidies to the nonelderly may provide substantial benefits.
He further described some lessons for forecasting:
When thinking through any forecasting exercise, it is important to recognize that the future path of technology is highly uncertain, along with a number of other parameters that influence the modeler’s task. Policy makers must make decisions in the face of uncertainty, and modelers must account for the need to make decisions in the face of uncertainty. This requires analyzing the riskiness of different policy actions, rather than simply providing a single expected outcome associated with any given policy.
The goal is to discover policies that limit risks but cultivate large potential gains. When such policies fail to exist, modelers can at least expose key trade-offs facing policy makers when they are
making decisions in an environment of radical uncertainty. That is another frontier that modelers need to push.
Finally, Lakdawalla emphasized two areas in which modelers need to make advances. One is pushing further on modeling the evolution of technology in addition to modeling the evolution of spending. Another is incorporating uncertainty in a much more fundamental way. Uncertainty is not just a nuisance parameter to a modeler; it is a fundamental part of decision making into the future. Every decision maker faces it, so it is essential to incorporate it into projection modeling and to try to come up with ways to expose the impacts of that uncertainty, so that policy makers have the information they need to try to do the best they can in the face of it.
MEDICAL TECHNOLOGY AS A COST DRIVER
Kenneth Thorpe (Emory University) spoke about the role of technology in rising health care costs and ways to estimate the effects on expenditures globally and for specific medical conditions. He noted that the impact of technology on health care costs is well recognized. Costs continue to increase due to treatment innovations as well as advances in detection and diagnosis of existing disease. However, for many reasons, the role of technology is difficult to measure accurately. Two general approaches have been relied on to date in an effort to determine technology’s effect on medical expenditures: (1) residual analysis and (2) case studies.
Measuring the Role of Technology
Residual Analysis Approach
The traditional approach for measuring the impact of technology on health care spending has been residual analysis. In this method, demand-side factors that are easily captured are accounted for (e.g., population demographics, insurance changes, income changes, prices, and administrative costs), and the remaining cost is attributed to technology. This method is limited, however, by its tendency to overstate the impact of technology on costs, as it fails to control for less apparent variables, such as changes in patient characteristics over time and trends in rising clinical incidence and disease prevalence that could affect growth in medical spending. For example, recent findings by Thorpe and colleagues point to the rising clinical incidence and prevalence of chronic disease in recent years, specifically among Medicare beneficiaries (Thorpe et al., 2010). The residual analysis method, which adjusts only for changes in the age and sex composition of
the population over time, is insufficient to detect important trends in disease prevalence and treatment. While the residual approach may be acceptable for measuring the impact of technology on costs over the past 50 years, it lacks the sensitivity required to explain changes in recent decades.
Another change that residual analysis would have failed to detect is the decline in the disability rate over the past decade. Thorpe observed that despite increases in disease prevalence, Medicare spending per capita decreased over the past decade in part due to lower disability rates among the elderly population.
The Case Study Approach
The case study approach involves looking at each medical condition separately, examining trends in treated prevalence of the medical condition over time, and then decomposing changes in spending over time into three parts: (1) changes due to treated prevalence of the disease, (2) changes due to spending per case, and (3) interactions between the two. The rationale underlying this approach is that any changes in costs due to spending per case can be attributed to changes in technology.
Using the case study approach to examine the growth in health care spending over the past two decades, Thorpe and colleagues found that spending increases are largely due to rising rates of treated prevalence rather than spending per case treated (Thorpe et al., 2004; Thorpe, Ogden, and Galactionova, 2010). Such a finding necessitates further investigation, as increases in treated prevalence can result from a number of factors. For example, treated prevalence can increase because the true clinical incidence of the disease has risen, which is the case for diabetes. Treated prevalence can also increase due to the implementation of new clinical guidelines about when medical personnel should intervene, which happened in the case of depression, or to improvements in disease detection that are unrelated to technology. Similarly, increases in spending per case can be attributable to a combination of factors: increases in clinical disease incidence and changes in treatment patterns, in addition to technological advances.
For example, using data from the 1987 National Medical Expenditure Survey and 2001 Medical Expenditure Panel Survey Household Component, Thorpe and colleagues examined the impact of obesity on rising medical spending. They found that the combined effect of changes in obesity prevalence and changes in spending per case accounted for 27 percent of the growth in inflation-adjusted per capita health spending between 1987 and 2001. The changes in spending per case over this period were largely due to the widened spending gap between obese and normal-weight adults—a difference that increased from about 15 percent in 1987 to 37 percent in 2001 (Thorpe et al., 2004). These findings
suggest that the cost increase in treating obese adults between 1987 and 2001 was not necessarily due to new technologies but rather to more intensive treatment patterns and changes in the clinical threshold for when to treat patients.
Data from the period 1987-2006 also highlight an increase in treated disease prevalence among the Medicare population. According to Thorpe and colleagues, important changes in a handful of chronic conditions (i.e., diabetes, kidney disease, hyperlipidemia, hypertension, mental disorders, arthritis) have driven the rise in spending among Medicare beneficiaries over time. The largest cost driver among these conditions is diabetes, which accounts for nearly 8 percent of rising Medicare costs over the past decade (Thorpe et al., 2010). To determine whether these rising costs were due to a growing incidence of treated prevalence among Medicare patients—which, for diabetes alone, increased from 11.3 percent in 1987 to about 20.5 percent in 2006—Thorpe and colleagues used the spending decomposition method to determine what percentage of the total change in spending from 1987 to 2006 was due to a change in the prevalence of treated disease. They found that, for most of these condition-specific case studies (heart disease notably excepted), increases in spending were a result of rises in treated prevalence rather than rises in the cost per treated case. Again, an important question left unanswered through use of the case study method was what factors drove the change in treated prevalence, which Thorpe notes as an important avenue for future research.
Suggested Advances in Modeling
In order to improve the predictive accuracy of health care spending projection models, Thorpe suggested supplementing the more traditional GDP-based approach, which relies largely on demand-side factors, with data that are traditionally built into epidemiological models, such as projected trends in disability, obesity, and smoking. These risk factors are typically tied to higher rates of medical spending. Accounting for these risk factors, along with changing population risk factors and health, will therefore result in more comprehensive spending models with improved predictive accuracy.
Thorpe and colleagues have used this modeling approach to determine the role that rising obesity prevalence has played in changes in medical spending, both among the Medicare population and the general population. In the two recent decades studied, the results consistently show changes in obesity prevalence to account for 25 to 30 percent of the growth in spending (Thorpe and Howard, 2006; Thorpe et al., 2004). This approach can also be used to explain the decline in Medicare spending over the past decade by taking into account such factors as declining disability rates and policy changes after the implementation of the Balanced Budget Act. Thorpe con-
cluded by emphasizing that epidemiological factors are clearly an important piece of the spending puzzle and that, going forward, researchers might examine whether adding them to GDP-based models enhances the ability to more accurately predict changes in medical spending.
Many participants had comments and questions on the topics of incorporating the supply side in projection models, the diagnosis of diseases, policy options relating to pricing, and various other issues related to technologies.
Supply Side in Projection Models
Dana Goldman observed that researchers working on the RAND FEM have tried to think about technology, starting with just asking people, but then incorporating technology and pharmaceutical spending into the process of modeling. Thorpe is doing the same from a disease perspective. Thinking about projecting forward, the only part that is missing in modeling is what goes on in a doctor’s office. Goldman questioned if one of the answers is some sort of modeling on the supply side, for example, incorporating what goes on in a doctor’s office, or some other approach.
Michael Chernew responded that in some of the forecasting work that has been done the question does arise; it is hard to think of it either as a purely demand-driven or purely supply-driven notion. The paradigm that has been dominant so far, although it may not be the correct one, has been that the demand side is dominant. If people demand different types of technologies and services, such as, for example, more time with the physician, the supply side will respond to that demand.
There are obviously other models that would be based, for example, on limitation in the amount of workforce that is needed. Members of the Technical Review Panel on the Medical Trustees Report were concerned about what would happen if there are not enough doctors to meet the demand projected by the models. But the panel thought that there will be innovations on the supply side to meet the demand one way or another. The technology would change in very much needed, endogenous ways, which Lakdawalla spoke about in his presentation.
Growth in the Diagnosis of Disease
Jonathan Skinner commented on Thorpe’s work showing growth in the diagnosis of disease. When coupled with what seems to be evidence of the
decline in disability rates among the elderly, does that mean that there are more healthy people around who just happen to have more diseases? Or on the supply side, are physicians just a lot better at diagnosing? It should be noted that the diagnostic rates of magnetic resonance imaging and computerized tomography (CT) scans and other such developments are going up at double-digit rates. Understanding the issue of risk adjustment is important; it looks like doctors who do a lot of diagnostic tests and procedures also diagnose lots of disease.
Thorpe responded that it is both. This was not an issue before 1985 or earlier, because the percentage of the population that was obese, for example, was stable for 25 or 30 years. It was not accounting for much of the growth in spending at all. Since that time, the incidence of diabetes is way up; pulmonary disease is also way up, as is the price of their treatments. The detection rates of diabetes have not changed in 30 years: two-thirds of total diabetes is detected. The problem is not detection: hypertension and hyperlipidemia have exploded. These conditions would not have been treated 20 years ago, in part because statins did not exist, but also because the clinical thresholds for treating patients have changed over that time period. So true incidence increases that one can actually do something about in terms of changing the risk factors, changes in clinical thresholds for treating patients, and better detection all contribute to the growth in the diagnosis of disease.
Alan Garber (Stanford University) added that one of the problems is a technologically driven change in the diagnosis of various conditions. For example, the rise in coronary CT angiography also leads to increases in findings of conditions, and a certain percentage of those findings get worked up. That would not have happened 10 years ago. He suggested that one should look carefully at what data sets are needed that would make it possible to identify the issues and do the needed analysis. Electronic health records and baseline data sets will probably be needed.
Policy Options Relating to Pricing
Referring to Lakdawalla’s presentation, Garber remarked that he did not understand the two policy options relating to pricing. One of these options was about reducing manufacturer prices. Is it assumed that somehow there is a policy that is going to accomplish a specific reduction without saying how to get to that target? The more important question is about the copayment reduction option. Is it a policy that says to change the coinsurance rate, or is it a policy that says somehow out-of-pocket payments will change by 20 percent?
Lakdawalla responded to the first question that the mechanism for lowering prices was price negotiations. Implementation of price negotia-
tion in European countries tended to lower prices paid to manufacturers by roughly 20 percent. In some sense what they are doing is comparing any number of polices that would have the effect of lowering manufacturing prices and policies that would have the effect of lowering out-of-pocket spending. In response to the next question regarding consumer prices, Lakdawalla stated that they were thinking about consumers paying 20 percent less out-of-pocket on a unit-adjusted basis.
Garber remarked that that is an outcome, not an instrument. To model a policy change, one does not assume the results of the policy change. There are many different ways to get to a 20 percent reduction in out-of-pocket costs for consumers with different implications for welfare.
Lakdawalla responded that it depends on the way the model is built. Behind the two scenarios they had in mind motivating policy. So on one hand, on the manufacturer price side, they had price negotiation as a motivating policy. On the other, the consumer price reduction side, they had subsidies for prescription drug insurance as a motivating policy.
Identifying Cost-Reducing Technologies
Cynthia Leibson (Mayo Clinic) asked how to find technologies that are in fact cost reducing and where can research provide some insight. From the clinical perspective, one of the new buzz words is individualized medicine. What that requires is characterization of the phenotype and the genotype at the level of the individual and then following individuals forward over long periods of time to see health outcomes, both the condition of interest and the outcomes following that condition of interest, and how the primary and secondary interventions impact those survival rates. There is not much research being done or much discussion of that approach with costs other than outside clinical trial settings, which is not very satisfactory. She questioned if there might be some room for looking at where one can intervene—that is, at what point along the life-course trajectory (from obesity to diabetes to cardiovascular disease to death) for life expectancy—as well as on whom to intervene. Who are the people that these technologies would benefit not only with respect to these clinical outcomes, but also with respect to saving costs?
Chernew responded that it is a challenge to think about what is happening now or in the next 5, 10, or 15 years in terms of technology and what will lower costs. There are many technologies that might lower costs at a single point in time; the challenge has been to lower them over time, for a number of reasons that involve not only treatment for diabetes, but also all of the competing risks in order to show that a lot of money is being saved.
Research needs to go into the basic question of what will lower spending. That question is dramatically different from the question of how to set up a system to have lower spending growth in the future that is going to have several new technologies. They are going to be interacting with spending in a range of complicated ways. Both types of studies are important.
Other Related Issues
Kenneth Feingold (Office of the Assistant Secretary for Planning and Evaluation, U.S. Department of Health and Human Services) asked Lakdawalla to elaborate the part of the model that goes from prices in drug launches to life expectancy. If there is a global market, why is there still a difference between the United States and Europe in life expectancy? Does this take into account the fact that some drugs that are launched are not expected to affect longevity, since they may be copy cats of existing drugs or may alleviate pain without actually extending life.
Lakdawalla explained that there are a couple of steps in the model. The first goes from drug revenues to the number of drug launches, and that is done on a disease-specific basis. So revenues changing in a particular disease have impacts on launches in that disease. That is one reason why the United States and Europe differ in their disease profiles. The next step is from drug launches to health. The vast majority of drug launches do not have an impact on health, so the research team went to the clinical literature, looked at top-selling drugs, and made the very conservative assumption that all of the other drugs have no impact on health. So every time a drug is launched, a question is asked, what is the probability that it will be a top-selling drug? If it is, then it gets assigned the mean effect for top-selling drugs. If it is not, and that probability is about 75 or 80 percent, then it has no impact on health. Most of the time it is a draw from the air and has no impact.
Mark Freeland (Centers for Medicare & Medicaid Services) observed that there are multiple causes of innovation, especially when one looks at the continuum of cost-increasing versus cost-decreasing innovation. For example, two of the most innovative, dynamic industries in the United States, computers and agriculture, have actually experienced declining increases in price. He questioned if Lakdawalla and colleagues looked at other industries or looked at the pharmaceutical industry for other causes of innovation, including the role of the National Institutes of Health (NIH) in a lot of basic research and development and the potential that they could have to focus on cost-decreasing as opposed to cost-increasing technologies.
Lakdawalla responded that one of their collaborators in this project has been doing a lot of work specifically on this question of how NIH funding affects private rates of innovation in health care. As one might guess, it is
fairly significant as a source of private innovation through the channel of funding universities.
Regarding the other part of the question, comparing cost-reducing innovation in health care with that of other industries, as a first approximation one looks at health care to say where are all of the cost savings inherent in desktop computers and so forth. In some sense one could argue that there is some cost savings going on. If increases in drug spending, for example, substitute for hospital stays, that is one source. But on balance it is becoming more expensive, even though people are getting more in return. Some people have identified the productivity slowdown in the pharmaceutical sector and in other areas of the medical care system as one contributor to that. But there are a lot of open research questions, and it is not well understood as to why that industry is so different.