In this paper we present a new framework for understanding the factors driving the growth in private health insurance spending. Our analysis estimates how much of the rise in spending is attributable to a rise in treated disease prevalence and spending per treated case. Our results reveal that the rise in treated disease prevalence, rather than the rise in spending per treated case, was the most important determinant of the growth in private insurance spending between 1987 and 2002. A rise in population risk factors and the introduction of new technologies underlie these trends.

Both employers and workers have identified the high and rising costs of health care as a key economic issue facing the United States. The rising cost of health insurance has been associated with a reduction in the share of U.S. workers receiving employment-based coverage.¹ High health care costs have been a major source of labor strikes during the past two years.² Recently Rick Wagoner, chairman and chief executive officer (CEO) of General Motors, cited the cost of health care as a key factor reducing the international competitiveness of U.S. business.³ In light of these factors, slowing the growth of health care spending has taken center stage in the national health policy debate.

Crafting effective policies for slowing the growth in health insurance premiums requires a clear understanding of why spending is rising. Previous analyses have focused on the sources of spending increases by tracking trends in where the dollars are spent (hospitals, drugs, physician services, and so on).⁴ Although useful from a national accounting perspective, the data provide little insight into the factors underlying the growth in health care spending. Other attempts to understand the causes of spending growth have quantified the factors responsible for the rise in spending.⁵ Most of this literature has concluded that the factors we can explicitly measure—population aging, the spread of insurance, rising income, and administrative costs—account for a small proportion of the overall rise. Instead, technology, which is captured in the residual, is thought to account for most of the growth.⁶ Researchers have paid relatively little attention to increases in certain population risk factors (for example, the rise in obesity, changing environmental factors such as air pollution and ozone levels, stress, and exposure to aeroallergens) and the growing emphasis in medicine on the early detection of chronic conditions, both of which could lead to a rise in the prevalence of treated medical conditions.

Our analysis presents an alternative framework for understanding the factors responsible for the rise in private health insurance spending. Per capita health care spending is a function of treated disease prevalence and payments per treated case. We can attribute increases in spending growth to increases in either or both of these factors.⁷ Payments per treated case are largely driven by technology, and as medical technology has grown more advanced, payments per treated case have risen. Treatment associated with heart disease and heart attacks represent one such example.

Changes in treated disease prevalence are caused by a rise in the population prevalence of disease, changes in clinical thresholds (and awareness) for treating and diagnosing disease, and new technologies that allow physicians to treat additional patients with a particular medical condition. A rise in total disease prevalence (both diagnosed and undiagnosed) is associated with changing population risk factors such as obesity. For instance, among adults ages 20–74, obesity prevalence increased from 14.5 percent (1976–1980) to 30.4 percent twenty years later (1999–2000). During the same period, total diabetes prevalence, which is clinically linked to obesity, increased 53 percent, and diagnosed (treated) diabetes prevalence increased 43 percent.⁸ Other risk factors that influence population levels of disease include stress, which has been shown to be associated with several chronic health conditions, illness, and changes in physiological functioning; and aeroallergens (such as dust mites), air pollution, and smoking (both primary and secondhand), which have been shown to be associated with pulmonary conditions, respiratory diseases, and asthma in both children and adults.⁹

Treated disease prevalence may also rise if the clinical threshold for diagnosis and the awareness, detection, and treatment of disease change over time. For example, increased awareness about and recognition of depression among both patients and clinicians has led to a rise in treatment of depression even though total disease prevalence has been constant over time.¹⁰ Treated prevalence of depression has doubled since 1987.¹¹ Finally, new technologies often allow physicians to treat more patients with a particular condition. The introduction of new pharmacological options for treating high blood pressure and cholesterol has led to substantial increases in cases treated, disproportionately so among obese patients.¹² Physicians may also be more likely to prescribe medications today at lower blood pressure and serum cholesterol thresholds, coinciding with revised guidelines defining what constitutes a “normal” blood pressure level.¹³

Our analysis is designed to identify how much of the rise in private health care spending is attributable to the rise in treated disease prevalence compared with higher spending per treated case. We also examine how changes in one risk factor—obesity—have increased treated disease prevalence over time.

Study Data And Methods

We estimated the level and change in health care spending among privately insured (those with private insurance at least six months during the year) adults ages 18–64 in 1987 and 2002. We examined spending on the top twenty medical conditions responsible for the greatest (inflation-adjusted) dollar growth in private health insurance spending.¹⁴ Data for our study are from the 1987 National Medical Expenditure Survey (NMES) and the 2002 Medical Expenditure Panel Survey (MEPS).¹⁵ The 1987 NMES surveyed 13,974 people ages 18–64 meeting our definition of “privately insured,” while the 2002 MEPS sample included 14,091 people. Both surveys are nationally representative samples of the U.S. civilian noninstitutionalized population. The surveys include detailed information on self-reported medical conditions, monthly markers of health insurance coverage, patient demographics, spending, and use of service. We adjusted the 1987 spending data from charges to payments using methods developed by the Agency for Healthcare Research and Quality (AHRQ).¹⁶

Both surveys collect data on respondents’ reports of their medical conditions for each medical event. The condition information comes from a specific question that asks respondents directly whether the visit was related to any specific health condition. These data were then subsequently professionally coded using the International Classification of Diseases, Ninth Revision (ICD-9). The ICD-9 codes were then collapsed into three-digit codes and grouped into 259 clinically relevant medical conditions using the Clinical Classification System (CCS) developed by AHRQ.¹⁷

We linked each self-reported medical encounter to one of the 259 CCS groups. Some medical events or visits may be associated with more than one condition. We addressed this issue by tabulating spending per event in the cases with more than one condition reported as well as total spending per event where one condition was reported. For example, we calculated total spending associated with heart disease (when it was the only condition reported) as well as heart disease and hypertension (when two conditions were reported). In the latter case, the ratio of the two spending totals (heart disease spending divided by heart disease and hypertension spending) was used to allocate costs when more than one condition was reported.

To estimate how much of the change in spending is linked to the rise in treated disease prevalence, we began by calculating the (inflation-adjusted) dollar change in spending for each condition between 1987 and 2002. We decomposed the change in spending into three categories: change attributable to a rise in treated medical conditions (treated disease prevalence), the rise in the cost per treated case, and population growth.¹⁸

To measure obesity’s role in increasing private insurance spending, we estimated medical care spending attributable to overweight and obese adults in 1987 and 2002. We used these estimates to calculate the increase in private spending linked to increases in obesity levels. We based these calculations on a two-part regression model estimated on the 1987 and 2002 samples, in which total annual per capita spending was the dependent variable. For controls, we used weight (underweight, normal, overweight, obese categories), age (18–29, 30–39, 40–49, 50–64), smoking (current smoker), sex, region (East, Midwest, South, West), education, race and ethnicity (black, Hispanic), marital status, and income as a percentage of the federal poverty level (under 100 percent, 100–199 percent, 200–399 percent, and 400 percent or more).

For each person in the sample, we calculated predicted (retransformed to dollars) per capita spending by multiplying predicted values from the first and second stages. We then calculated hypothetical per capita spending levels if all adults in the sample were underweight, normal weight, overweight, or obese. These predictions allowed us to net out the impact of observable characteristics included in our model on per capita spending. Standard errors and 95 percent confidence intervals were calculated using 1,000 bootstrap replications.¹⁹

We calculated attributable spending as the dollar difference between predicted per capita spending for obese and normal-weight adults multiplied by the number of obese adults (weighted using the svymean command in STATA, version 8) and divided by total annual private health care spending. We used the same procedure for overweight adults in each year. We applied this percentage to total private spending to calculate the additional dollar spending linked to the higher use of services among overweight and obese adults.

Study Results

Between 1987 and 2002, inflation-adjusted per capita private health insurance spending increased nearly 60 percent, or 3.1 percent per year. Exhibit 1 presents the twenty medical conditions accounting for the largest portion of the rise in private health care spending during this period. In 1987 these conditions accounted for 42 percent of private insurance spending; by 2002, they accounted for 53 percent. The twenty conditions also accounted for 67 percent of the growth in private health insurance spending during the period. We found similar results in a study of changes in spending levels for all age groups.²⁰ Spending on newborn and maternity care was the condition accounting for the largest increase in spending: more than 8 percent of total growth between 1987 and 2002.