Center for Studying Health System Change

Providing Insights that Contribute to Better Health Policy


Insurance Coverage & Costs Costs The Uninsured Private Coverage Employer Sponsored Individual Public Coverage Medicare Medicaid and SCHIP Access to Care Quality & Care Delivery Health Care Markets Issue Briefs Data Bulletins Research Briefs Policy Analyses Community Reports Journal Articles Other Publications Surveys Site Visits Design and Methods Data Files


Health Spending:

January/February 1999
Health Affairs

Tracking Health Care Costs:

November 1999
Issue Brief No. 23

Health Care Costs:

June 1997
Issue Brief No. 10

Tracking Health Care Costs:

January 1997
Issue Brief No. 06

Despite Fears, Costs Rise Only Modestly in 1998

Fall 1998
Data Bulletin No. 13

Tracking Health Care Costs:

September/October 1998
Health Affairs

Tracking Health Care Costs:

July/August 1997
Health Affairs

Tracking Health Care Costs

Fall 1996
Health Affairs

Tracking Health Care Costs:

November 2000
Data Bulletin No. 20

Access to Health Care:

April 1997
Issue Brief No. 08

Health System Change:

Winter 1995
Health Affairs

A Primer on Understanding Health Care Cost Trends:

The Story Behind the Numbers

Issue Brief No. 05
December 1996

irtually every study shows that rates of increase in health care costs have declined throughout the 1990s, but often the bottom-line numbers differ. There are numerous reasons for this disparity, including which data are used and how researchers analyze them. Unfortunately, there is no ideal source of data; the best sources vary widely with regard to accuracy, timeliness, and, most important, whether the information lines up with the questions that are being asked. This Issue Brief provides information about how to understand what the numbers really represent.

What Is Being Measured?

hen a government or employer survey reports that the rise in health care costs is slowing, the first questions to ask are: What is the study measuring? All health care costs? Whose health care costs? It is important to know this because in almost all cases the data presented in the study have been prepared and adjusted to answer a particular policy or research question.

For example, when analyzing Medicare data to measure progress in controlling costs, researchers often adjust for general inflation and changes in the population served. This is done to separate out changes that result from factors other than cost control efforts, such as the aging of the population. However, if researchers want data to plan financing for Medicare in the 21st century, demographic trends would be critical and, accordingly, they would not want to adjust for them.

Another way to look at health care data is by the various components -- for example, the amount spent for hospital care, on physicians, and for drugs. These data can show important trends in the quantity and price of services. They also can provide insight into what is behind the overall trends in health care costs and a basis for projecting future cost trends.

Employer surveys use premium costs as a proxy for health care costs. Most employer surveys gather data on the total premium cost -- that is, both the employers’ and employees’ share. Trends in premium costs are a rich source of information for employers, but are often a poor indicator of health care cost trends.

Back to Top

How Data Vary by Source

he three major types of data used to develop cost trends are provider costs, claims incurred by insurers, and premiums paid by employers. Data ideally should be timely, cover the entire population, reflect resources used in providing services rather than what is paid for them, and provide insight into the performance of different types of health plans. In reality, each of the major sources of health data has its strengths and weaknesses.

Cost data from providers score well in a number of dimensions. They are the most timely, in large part because providers log cost or expense information as it happens. They also cover the entire population, including the uninsured. A weakness of cost data is that hospitals are the main source of information, and there is a paucity of good data available from other providers, such as those operating home health services and clinical laboratories. This problem is likely to increase over time as inpatient services continue to decline as a percentage of health care spending.

Claims data from insurers provide complete spending information for covered services. Moreover, the data not only can show expenditures but also can separate them into price and quantity components. Because of the proprietary nature of the information, however, it is not possible to get claims data from all insurers. In addition, health maintenance organizations (HMOs) record visits by encounters not claims (if they record them at all), and encounter data generally cannot be combined with claims data. This means that claims data provide an indicator of cost trends only for a nonrepresentative and declining segment of the health care system. Another weakness of claims data is that changes in the benefit structure are difficult to capture.

Costs vs. Expenditures

The terms "costs" and "expenditures" often are used interchangeably. They are, however, conceptually different. Costs reflect the resources devoted to health care that are not available to produce other goods and services. Expenditures are what is paid for health services by purchasers or what is received by providers. The two differ when the payment (expenditure) is greater -- or less -- than the resources (costs) that go into providing the services.

Premium data from employers let researchers track trends by type of health plan. This is useful for learning about the mechanisms behind changes in cost trends. For example, are changes due to a shift in enrollment from fee-for-service plans to HMOs? Are both types of health plans experiencing a slowing of the increase in premium?

On the downside, premium data are not as timely as they appear. Premiums are established months in advance based on claims data, and generally are in effect for a year. For example, 1996 premiums may have been set in late summer 1995 based on experience since the beginning of that year. Premiums also reflect a three-year underwriting cycle in which premiums play catch up based on real claims experience. If they were set too high, premiums might remain the same for two years; if they had been set too low, premiums might rise steeply. In other words, the current premium is not likely to reflect the current underlying claim costs.

Other drawbacks of premium data are that employer data exclude people covered by Medicare and Medicaid and the uninsured and, therefore, are not representative of the nation. In addition, as with insurer data, they do not separate out the effects of any changes in the benefit structure.

Examples of Health Care Cost Databases

  1. The gold standard and best-known database on health care costs is the National Health Accounts (NHA), prepared by the Health Care Financing Administration (HCFA), Department of Health and Human Services. It uses information from providers and insurers and is produced annually. However, data for calendar year 1994 were not released until late spring 1996.

  2. Milliman & Robertson’s Health Cost Index (HCI) draws data from provider surveys on the major components of health spending: hospitals, physicians, and prescription drugs. The data, which are proprietary, are collected monthly and published quarterly, with only a three-month lag time.

  3. The Employment, Hours, and Earnings (EHE) data series published by the Bureau of Labor Statistics, U.S. Department of Labor, covers payroll data for all nonfarm establishments broken down by industry code. The series for production workers in health services establishments can be used to track health care costs. Although they cover only labor costs, the data are high quality and are published monthly, with a lag time of less than two months.

  4. Several leading accounting and benefits consulting firms conduct annual surveys of employers’ health plans. The surveys, which are designed to meet the needs of employers, include data on premiums and benefits. Many of them attempt to maintain a core panel of respondents and induce participation by providing analyses of survey data to enable respondents to compare themselves with their peers.

The graph on the facing page shows statistics from three databases over a 10-year period. The general trend, which is adjusted for inflation for all three databases, is downward but the routes taken vary. Because the downward trend is so striking, the differences among the databases do not appear to be as great as they have been in the past -- and likely will be in the future. Also, the downward trend is less dramatic than it appears since general inflation also declined -- from 4.3 percent in 1990 to 2.5 percent in 1995.

Back to Top

A Matter of Interpretation

he limitations in health care cost data are compounded by people who use them for purposes other than the ones they are intended for. For example, employer surveys are not designed to track cost trends in the economy. Instead, they are designed to let employers know what they are spending on health insurance and what benefits they are giving their employees, especially compared with other employers. They are a tool for competition in the business place, not a measure of trends in health care expenditures or of how well the nation has been able to control health care costs.

1. Calculations using data from Milliman & Robertson’s Health Insurance Trend Model and expanded to include Medicare.

2. From the Health Care Financing Administration, National Cost Estimates Unit. Reflects revisions published in May 1996.

3. From the Bureau of Labor Statistics, Employment, Hours, and Earning. Payroll calculated as the product of production workers, average hours per week, and average hourly wage.

While generically the same, the employer surveys differ for a number of reasons, including the samples used. For example, both employer surveys referenced here show a significant decline in premium costs for all kinds of plans -- fee-for-service as well as managed care -- but the numbers are not the same.

KPMG Peat Marwick and Hay-Huggins not only have different firms in their samples, they also have different methods of drawing their samples. The Peat Marwick survey is a stratified random sample; Hay-Huggins’s is drawn from its clients with some additional firms to make it representative.

There are other differences when changes in benefit structure are considered. Actuaries at Hay-Huggins, for example, performed an analysis for the Center for Studying Health System Change that looked at changes in benefit structure by plan type from 1993 to 1995. They found the benefit structure of fee-for-service plans virtually unchanged, while the benefits in the average HMO had increased somewhat. The data behind the graph do not incorporate this actuarial analysis. When benefits are adjusted for, HMO premium increases are between 0.5 and 1.0 percent lower per year.

Survey results also can differ by the way the change in premium is calculated. For example, another employer survey, which showed a lower rate of increase, based its calculations on the average premium across all plans. But since people are switching from traditional plans to managed care plans, which tend to have lower premiums, the trend in the average premium will be lower than a weighted average of the trends for each type of plan. The extent to which lower cost trends come from shifts among plan types versus by plan type has implications for the way people perceive the future of health care costs. For example, the cost outlook is more encouraging if each plan type is slowing its rate of premium increase, rather than if premiums are growing just as fast as before but more people are switching to plan types with lower premiums.

Back to Top

Reading Behind the Headline

he headline of a recent release of data from the Department of Labor (DOL) reads: "Growth in employers’ health costs slows." Does this mean that cost controls are really working? The data behind the story are based on the Employment Cost Index, which includes a component for health insurance costs borne by employers.

Questions to Ask about Cost Data on Employer-Sponsored Health Insurance

  • Do data on premiums show the entire premium or only the employer contribution?

  • How current are the data? Employer premiums do not reflect what employers are paying now, but what costs were more than a year ago.

  • Are data adjusted for changes in general inflation?

  • Are population changes accounted for (e.g., changes in the proportion of employees enrolling in a health plan or of active employees vs. retirees)?

  • How do trends in fee-for-service premiums and HMO premiums differ?

Several reasons are cited for the slowing growth in employer health insurance costs, but many of them have little to do with overall efforts to control costs. Rather, they reflect:

  • shifting costs from employers to employees -- e.g., employees pay a greater share of the premium; and

  • a drop in the proportion of employees receiving employer-sponsored health insurance either because the employer stopped offering it or because the employees -- usually people in low-paying jobs -- are less likely to enroll, especially for dependent coverage.
This is not to say that the DOL data are wrong, poorly conceived, or misleading; they are not. But data cannot be all things to all people, and much of what is presented to researchers and the public was not developed to portray overall health care cost trends.

This Issue Brief is adapted from "Tracking Health Care Costs" by Paul B. Ginsburg and Jeremy D. Pickreign, which appeared in the Fall 1996 issue of Health Affairs. The next Issue Brief will focus on data documenting the falling rates of increase in health care costs.

Back to Top


Back to Top
Site Last Updated: 9/15/2014             Privacy Policy
The Center for Studying Health System Change Ceased operation on Dec. 31, 2013.