We calculate the level and growth in health care spending attributable to the fifteen most expensive medical conditions in 1987 and 2000. Growth in spending by medical condition is decomposed into changes attributable to rising cost per treated case, treated prevalence, and population growth. We find that a small number of conditions account for most of the growth in health care spending—the top five medical conditions accounted for 31 percent. For four of the conditions, a rise in treated prevalence, rather than rising treatment costs per case or population growth, accounted for most of the spending growth.

The rising cost of health care, and what to do about it, is perhaps the most challenging health policy issue facing the United States. Health care is projected to account for 15.2 percent of U.S. gross domestic product (GDP) in 2004, compared with 11.1 percent fifteen years ago.¹ During this period health care spending increased at an average annual rate of 7.5 percent per year (in nominal dollars) and 5.1 percent per year when adjusting for inflation (using the GDP deflator).² During the past three years, the cost of health insurance has increased by an average of 12.5 percent per year.³

The most common factor cited as driving rising health costs has been the explosion of new medical technologies, which can improve care but tend to cost more than older modalities of treatment.⁴ However, total cost is also a function of how many people are receiving treatment for a given condition. The rise in treated-case prevalence may reflect improvements in medical technology that allow expanded treatment of a particular condition. It could also reflect changes in the diagnosis or reporting of disease. Finally, the rise could reflect factors such as the aging of the population. Distinguishing among these scenarios—increasing cost per case and increasing population-based use of treatments—could provide an important context for understanding U.S. health care spending. In particular, it could allow us to more effectively target interventions designed to rein in the growth in health care spending.

Although several studies have examined the factors associated with the rise in health care spending, they have largely tracked overall changes in payments for hospitals, physicians, and pharmaceuticals.⁵ Linking health care spending to the treatment of specific medical conditions can establish a framework for understanding (1) changes in real health care spending by disaggregating the effects of new technologies developed for treating those conditions and (2) increases in the number of people who are treated. Such an analysis also allows a better match between underlying cost drivers and potential interventions/solutions. Moreover, a disease-based analysis affords a more natural comparison to changes in medical benefits purchased.

To address the issue of what is driving health care spending growth, we undertook a study designed to track changes in spending over time by medical condition. We examined the change in this spending as a percentage of the change in total health care spending. We then decomposed this change, by medical condition, into changes in treated prevalence, treated cost per case, and population growth.

Study Data And Methods

Data. Our analytic approach was to estimate the level and change in nominal health care spending over time by patients with the fifteen most expensive medical conditions. Data for our study are from the 1987 National Medical Expenditure Survey (NMES) and the 2000 Medical Expenditure Panel Survey, Household Component (MEPS-HC).⁶ The 1987 NMES surveyed 34,459 people, and the 2000 MEPS, 25,096 people. Both surveys are nationally representative samples of the U.S civilian noninstitutionalized population. They contain detailed information on health spending, use of services, patient demographics, insurance coverage, markers of health status, and self-reported medical conditions. We adjusted the 1987 and 2000 spending data to make them comparable using the methods developed by the Agency for Healthcare Research and Quality (AHRQ).⁷ The process adjusted the 1987 data from charges to payments, the same measure used in the 2000 data.

Both surveys collect detailed information on respondents’ reports of their medical conditions and other measures of health status. When a survey respondent reports a medical event, such as a physician office visit, he or she is asked to describe the reason for the visit. In both years the data were professionally coded from respondents’ verbatim text using the International Classification of Diseases, Ninth Revision (ICD-9). Up to four ICD-9 codes are listed per medical event. The ICD-9 codes are collapsed to three-digit codes and subsequently coded into 259 clinically relevant medical conditions using the Clinical Classification System (CCS) developed by the U.S. Department of Health and Human Services (HHS).⁸ Although the medical conditions were self-reported, previous research has found a high level of agreement between descriptions of conditions (at the CCS level) reported by patients and those provided by physicians.⁹

Study methods. Following the methods of Benjamin Druss and colleagues (2002) and of Joel Cohen and Nancy Krauss (2003), we linked diagnosis codes for each self-reported medical encounter (provider visits of any type and prescribed drugs) that prompted a patient to seek medical care.¹⁰ For each patient, we calculated total annual spending and total spending for each of the 259 CCS medical conditions reported. We compiled the fifteen conditions with the largest nominal growth in spending between 1987 and 2000. We then tabulated total annual spending by medical condition in 1987 and 2000, and the change in spending by medical condition. For each condition, we tabulated the change in spending as a percentage of the change in national health spending among the noninstitutionalized population—both are reported in nominal dollars. Since the NMES and MEPS samples include a complex stratification design, we used STATA version 8 and used the “svymean” for the means and standard errors of all spending data. This accounts for both the complex sample design and the weighting of observations.

Some medical events were associated with multiple conditions. For example, a patient may seek care to treat an existing heart condition as well as hypertension. As a result, this approach will double-count the spending associated with some medical conditions. On the other hand, simply using the principal diagnosis, perhaps through the use of a disease hierarchy, may understate spending associated with a medical condition. Recognizing this potential, we present a range of estimates. Our upper-bound estimate added up total spending for each health care event for which a given condition is reported. Since up to four medical conditions can be reported for each event, this will obviously include some double-counting. As a lower bound, we summed spending from each medical event for which only a single condition is reported. Although the total spending calculated from this approach obviously does not account for all spending associated with a given condition, it does not include any double-counting. Finally, we developed a “best guess” estimate of condition attributable spending using the following approach. We tabulated spending per event for those reporting a single medical condition (for example, heart disease and no other condition). We then tabulated spending per event for those reporting two or more medical conditions associated with the event (for example, heart disease and hypertension). We calculated the ratio of these two spending totals and used it to determine how much of the spending associated with heart disease plus other conditions should be attributed to heart disease.¹¹

Study Results

Nominal health care spending among the noninstitutionalized population increased by $314 billion—5.5 percent per year—between 1987 and 2000 (Exhibit 1). After inflation was adjusted for using the GDP deflator, total spending increased by $199 billion—about 3 percent per year.

Exhibit 2 shows the growth in nominal spending over time by medical condition. Between 43 and 61 percent of the total nominal change in spending between 1987 and 2000 is attributable to the fifteen most costly conditions. Our “best guess” estimate, adjusted for double-counting, approximates the share to be 56 percent. Most of this change is concentrated in the five most expensive conditions: heart disease, mental disorders, pulmonary disorders, cancer, and trauma, which account for approximately 31 percent of the overall change in spending between 1987 and 2000.

The data presented in Exhibit 2 reveal a substantial rise in treated prevalence in eight of the fifteen conditions experiencing the largest rise in spending. For instance, treatment of mental disorders nearly doubled, and cases involving a pulmonary disorder, such as asthma and upper and lower respiratory diseases, increased 50 percent. There also was a substantial rise in the treated prevalence of hypertension and diabetes (Exhibit 2).

We now turn to decomposing the change in spending, by medical condition, into changes traced to population growth, changes in treated disease prevalence, and a change in annual spending on the condition per person reporting the condition. Since we are primarily interested in explaining the factors associated with increased spending within each medical condition, we are not concerned with double-counting across conditions. Exhibit 3 presents the results of our decomposition.¹² For several medical conditions, the rise in treated disease prevalence was a key factor accounting for the rise in spending. It accounted for 59 percent of the increased spending on mental disorders and figured prominently in the rise in spending on cerebrovascular disease (stroke and cerebral ischemia, 60 percent), pulmonary conditions (42 percent), and diabetes (50 percent).

In eight of the top fifteen conditions, a rise in the cost per treated case, not rising numbers of cases treated, accounted for most of the growth in spending. For instance, the treated prevalence of heart disease remained constant between 1987 and 2000. Thus, a rise in the cost per treated heart disease case accounted for nearly 70 percent of the rise in medical care spending between 1987 and 2000. The rise in cost per treated hypertension case accounted for 60 percent of the overall growth in spending. The rise in spending is traced to several new prescription drugs available to treat hypertensive patients. The treated prevalence of trauma declined during the period, with a rise in cost per treated case accounting for the rise in medical care spending.

Finally, population growth has also contributed to the rise in spending by medical condition. In our tabulations, it accounted for about 19–35 percent of the increase in condition-specific spending across the top fifteen medical conditions. This shows that demographic factors, in addition to factors such as changes in medical technology, have a large impact on nominal spending changes over time.

Discussion

A small number of medical conditions were associated with much of the increase in health care spending between 1987 and 2000. The top fifteen conditions accounted for approximately half of the overall growth in spending. For some of these conditions, such as mental disorders, most of the increase was associated with increased treated prevalence. A rise in treated prevalence, in turn, might represent either an increase in epidemiological prevalence or more widespread access to care among people with a disease. This mix varies across conditions. For instance, the prevalence of mental disorders has remained relatively stable over time; however, rates of treatment have been rising.¹³ The sharp rise in treated prevalence reflects two trends: increasing recognition and diagnosis of mental disorders, particularly depression and a rapid expansion of new psychotropic medications. Given the historical underdiagnosis and treatment of disorders

such as depression, this wider use of treatments, and the associated increase in health care spending, is likely to represent benefits that outweigh the cost.¹⁴

Potential interventions. For several conditions, the rise in the epidemiological prevalence appears to be responsible for the growth in treated cases. This result highlights the importance of developing interventions designed to reverse the rise in disease prevalence. This appears to be the case for pulmonary disease, which accounted for nearly 8 percent of the rise in spending over the decade. Prevalence and death rates for asthma have been rising since 1975.¹⁵ Factors accounting for the rise in asthma and other pulmonary disorders are not well understood. They have been linked to environmental exposures (both indoor, such as dust mites and smoking, and outdoor air quality).¹⁶ In addition, diabetes accounted for up to 3 percent of the rise in health care spending, with about 50 percent of the rise traced to a rise in treated prevalence. The U.S. Centers for Disease Control and Prevention (CDC) reports a continued rise in diabetes prevalence that now exceeds eighteen million among adults alone.¹⁷ The rise in the treated prevalence of diabetes closely tracks the substantial rise in obesity in the population.¹⁸ Since effective treatments exist for both of these conditions, however, it would be a mistake to see increased spending to treat them in a completely negative light.

Value of increased spending. Increased spending per person for these top fifteen medical condins may appear at first glance to reflect a truly “wasteful” increase in health care spending. However, the technologies used to treat patients with heart disease—such as new drugs, the use of diagnostic cardiac catheterization, and angioplasty—increased sharply during this period.¹⁹ These new approaches replaced less costly (and less effective) means for treating heart disease, and heart attacks in particular. While spending per person with heart disease is going up, death rates associated with this condition continue to go down.²⁰

Health policy analysts, policymakers, employers, families, and the media pay a great deal of attention to annual increases in nominal U.S. spending for health care. In recent years the rate of increase in health spending has been greater than the growth of the overall economy and has therefore led to an increase in the share of economic output devoted to health care.²¹ This is usually viewed negatively, because an increasing share of the economy devoted to health care means a lower share devoted to other goods and services. Moreover, rising health care costs have also been shown to reduce the number of people with health insurance.²²

In light of our results, however, we believe that some of the concern about the growth in spending may be misplaced. Discussion of the magnitude of health care spending growth usually does not take into account changes in disease prevalence and demographic factors behind spending growth. Moreover, at issue is whether the higher growth in spending is purchasing larger increments in medical care benefits or whether the same improvements in health can be purchased at lower cost. However, in light of how we track trends in health care spending—by provider (such as hospital, prescribed drugs)—analysts have been largely unable to address this key issue. Our focus on tracking the level and growth in spending by medical condition allows a more natural evaluation of this important issue, because it can provide a direct comparison to changes in health benefits. Recent research has found that higher spending on treating heart attacks, low-birthweight babies, cataracts, and depression has benefits that outweigh the increased costs.²³ Inasmuch as treatments for these conditions are cost-effective, their more widespread use is likely to represent an appropriate if costly expenditure by society.

Study limitations. These findings should be considered in light of several limitations. First, use of treatments and diagnoses are based on self-reports, which may have led to underreporting of medical conditions and spending. Second, the analysis excludes health care spending among the institutionalized population. Spending on some of the medical conditions reported here may have been incurred in nursing homes, which we do not observe.

Our current approaches for tracking spending are useful, although they provide little information for policymakers or purchasers for assessing what we are buying and whether the additional

spending is worth it. Addressing this key issue requires a focus on changes in spending and benefits along the lines presented here: by medical condition.

The authors thank their colleague Benjamin Druss for comments on an earlier draft.

NOTES

1. Centers for Medicare and Medicaid Services, “Health Accounts,” 24 March 2004, www.cms.hhs.gov/statistics/nhe/default.asp (26 July 2004). These are the National Health Accounts (NHA) estimates of total health care spending.
2. Ibid.
3. Henry J. Kaiser Family Foundation and Health Research and Educational Trust, “Summary of Findings,” Employer Health Benefits: 2003 Annual Survey, September 2003,
www.kff.org/insurance/ehbs2003-1-set.cfm (28 July 2004).
4. J.P. Newhouse, “An Iconoclastic View of Care Cost Containment,” Health Affairs 12 Supplement (1993): 152–171.
5. B.C. Strunk and P.B. Ginsburg, “Tracking Health Care Costs: Trends Stabilize but Remain High in 2003,” Health Affairs, 9 June 2004, content.healthaffairs.org/cgi/content/abstract/hlthaff.w4.354 (26 July 2004).
6. Agency for Healthcare Research and Quality, “Overview of the MEPS Web Site,” www.ahrq.gov/data/mepsweb.htm#full-year (26 July 2004). Compared with the spending estimates developed by the Department of Health and Human Services (the NHA estimates), the MEPS spending estimates focus on the noninstitutionalized population and do not include the same breadth of services (for example, spending for nursing home care). As a result, MEPS produces estimates of national health care spending that are lower than those produced through the NHA approach. However, both the populations and the services included in MEPS are those typically financed through private insurance. A detailed crosswalk between the two estimates has been developed by T. Selden et al., “Reconciling Medical Expenditure Estimates from the MEPS and NHA, 1996,” Health Care Financing Review 23, no. 1 (2001): 161–178. This review found substantial agreement in the estimates for the noninstitutionalized population for services generally included in private health insurance plans. When the NHA figures are compared with MEPS (on a comparable basis, focusing on spending included in both surveys among the civilian, noninstitutionalized population), spending totals were within 6.7 percent of one another.
7. S. Zuvekas and J.W. Cohen, “A Guide to Comparing Health Care Expenditures in the 1996 MEPS to the 1987 NMES,” Inquiry 39, no. 1 (2002): 76–86. The unadjusted spending data from the 1987 were based on charges, while the MEPS spending data used payments to providers. We used the approach outlined by AHRQ to make the two surveys comparable by transforming the 1987 NMES data to payments. The unadjusted charge-based total spending in the 1987 NMES was $363.6 billion. The adjusted NMES total based on payments used in our analysis was $314.1 billion.
8. J.W. Cohen and N.A. Krauss, “Spending and Service Use among People with the Fifteen Most Costly Medical Conditions, 1997,” Health Affairs 22, no. 2 (2003): 129–138.
9. N. Krauss and B. Kass, “Comparison of Household and Medical Provider Reports of Medical Conditions” (Paper presented at Joint Statistical Meetings, Indianapolis, Indiana, August 2000).
10. B. Druss et al., “The Most Expensive Medical Conditions in America,” Health Affairs 21, no. 4 (2002): 105–111; and Cohen and Krauss, “Spending and Service Use.” We replicated the totals reported by Cohen and Krauss for 1997 in their Exhibit 1.
11.For four of the top fifteen medical conditions, this ratio was close to 1. These were cases where a substantial share of total spending was traced to events with a single medical condition. As a result, in these four cases the best guess and the upper-bound estimate are the same. Moreover, the upper- and lower-bound estimates for these four conditions were virtually identical.
12. We divide the change in spending, by condition, into the overall change in national health spending among the noninstitutionalized population. This is done by evaluating the change in spending that would be generated by the observed changes in one of these components, leaving the others constant. Algebraically, the decomposition is derived in the following way: Cost in any year is the product of cost per case in that year, treated prevalence in that year, and population in

that year. Change in expenditures is the difference in cost in 2000 and 1987. Change in expenditures is equal to the sum of change in cost per case, change in treated prevalence, and change in population. This, in turn, is equal to a more complex expression that sums three products, each involving a difference and two other multiplicands. The multiplicands in each product are as follows (each group of three multiplicands is separated by a semicolon): difference in cost per case in 2000 and 1987, treated prevalence in 1987, and population 1987; difference in treated prevalence in 2000 and 1987, cost per case 2000, and population 1987; and difference in population in 2000 and 1987, cost per case in 2000, and treated prevalence in 2000.
13. M. Olfson et al., “National Trends in Outpatient Treatment of Depression” (2002).
14. Ibid.
15. D.M. Mannino et al., “Surveillance for Asthma—United States, 1960–1995,” Morbidity and Mortality Weekly Report, Vol. 47, No. RR-05 (24 April 1998): 1–28.
16. See National Center for Environmental Health, “Asthma: General Information,” 6 May 2004, www.cdc.gov/nceh/airpollution/asthma/basics.htm (26 July 2004).
17. U.S. Centers for Disease Control and Prevention, National Diabetes Fact Sheet: General Information and National Estimates in the United States, 2003 (Atlanta: CDC, 2003).
18. See National Center for Chronic Disease Prevention and Health Promotion, “Data and Trends: Diabetes Surveillance System,” 14 May 2004, www.cdc.gov/diabetes/statistics/comp/table4dtl.htm (26 July 2004).
19. American Heart Association, Heart Disease and Stroke Statistics—2004 Update (Dallas: American Heart Association, 2003).
20. National Center for Health Statistics, Health, United States, 2003 (Hyattsville, Md.: CDC, 2003), Table 29.
21. K. Levit et al., “Health Spending Rebound Continues in 2002,” Health Affairs 23, no. 1 (2004): 147–159.
22. D.M. Cutler, “Employee Costs and the Decline in Health Insurance Coverage,” NBER Working Paper no. 9036 (Cambridge Mass.: National Bureau of Economic Research, July 2002).
23. D.M. Cutler and M. McClellan, “Is Technological Change in Medicine Worth It?” Health Affairs 20, no. 5 (2001): 11–29.


Ken Thorpe (kthorpe@sph.emory.edu) is the Robert W. Woodruff Professor and Chair, Department of Health Policy and Management, Rollins School of Public Health, at Emory University in Atlanta, Georgia. Curtis Florence is an assistant professor and Peter Joski, a research associate, in the same department.

DOI: 10.1377/hlthaff.W4.437
©2004 Project HOPE–The People-to-People Health Foundation, Inc.