Geographic Variations in Health Care

The national volume of the Dartmouth Atlas of Health Care, published in the Spring of 1996, brought to light the often startling patterns of variation in health care throughout the nation. Research conducted to produce the Atlas revealed large differences in the rates of allocation of hospital resources, in the physician supply, and in the use of procedures such as coronary artery bypass grafting. The analysis of these differences was at the level of 306 hospital referral regions—the natural markets, defined by patient origin studies, for the use of tertiary, or referral, care among populations in the United States.

But health care is highly local, and the analysis of patterns of resource distribution and utilization among referral regions often masks important differences between the communities which, when aggregated, make up the larger region. Moreover, the task of actually addressing the problems of variation is often a local undertaking, one for which more specific—and more local—information is needed.

The 306 hospital referral regions comprise 3,436 geographically distinct hospital service areas, which are the natural markets for care that can be delivered locally—outpatient services and most acute hospital care. The regional volumes of the Dartmouth Atlas of Health Care (this book is one of nine such volumes) focus on these hospital service areas as the unit of analysis. The regional volumes make clear that there is often as much, and frequently more, variation among the hospital service areas within states and regions than among the larger units of analysis, the hospital referral regions.

The existence of variation raises a number of important issues. Foremost is the question “Which rate is right?” Which pattern of resource allocation, and which pattern of utilization, is “correct?” The study of practice variations reveals how complex this question really is. In the case of variations in rates of individual procedures, such as coronary artery bypass grafting and back surgery, the explanation is not that patients in areas with low procedure rates are going without treatment; they are, instead, being treated differently, often with more conservative medical management (Table 1). Learning which rate is right requires learning what informed patients want. The right rate must be the one that reflects the choices of patients who have been adequately informed and empowered to choose among the available options.

Table 1.

Common Conditions for Which a Number of Treatment Alternatives Are Used.

In the case of variations in the supply of health care resources, such as the numbers of hospital beds and physicians, the question “Which rate is right?” needs to be framed in another way: What is the impact on population health of variations in resource allocation? Is more better? And if not, how much could be reallocated to other, more effective uses by reducing resources and their utilization to the level of more conservative communities?

Another important issue raised by geographic variation concerns fairness. Variation studies provide good evidence that populations in regions where health care spending is low are not necessarily sicker, or have greater unmet medical need, than those in regions where spending is high. Spending is higher, not because better health is being achieved, but because the local health care systems have greater capacity, or because the price of medical care in those communities is higher. A system that rewards high spending areas by continuing to pay their higher costs is by definition economically punishing areas that have fewer resources, use them more efficiently, and are reimbursed less. Is it fair for citizens living in regions with low per capita health care spending to subsidize the greater (and more costly) use of care by people living in high resource and high utilization regions?

The nine regional Atlases provide the data and analysis for specific hospital service areas with which these and other questions can be addressed. Strategies to address the question of the appropriate levels of supply must be developed in the absence of detailed understanding of the nature of health care needs, medical care outcomes, and what patients want. One such strategy begins by examining individual communities and comparing them to others. Such comparisons lead naturally to a search for “efficiently” operated health plans or communities—those with an adequate but not excessive supply of resources.

About Benchmarking in the Atlas

Even in the absence of a detailed understanding of the nature of health care needs, medical care outcomes, and what patients want, we must establish appropriate levels of supply. One method of doing this is to examine the way resources are actually used, and to use as “benchmarks” efficiently operated health care plans or communities that appear to have an adequate but not excessive level of supply.

Benchmarking provides answers to two related questions: How much more (or less) health care capacity would the nation need, if all areas had the level of capacity of the benchmark area? And how much more (or less) health care capacity would be required in a specific area if its per capita capacity were equal to the level of the benchmark area?

Figure 1.1 illustrates the benchmarking approach to the second question by comparing the supply of acute care hospital beds per thousand residents of Boston, Massachusetts, Hartford, Connecticut, and New Haven, Connecticut, to three benchmarks. The benchmarks in this example are the highest ranked of the three areas, Boston (which had 3.7 beds per thousand residents in 1993); New Haven, the lowest ranked (2.4 beds per thousand) and the United States average of 3.3 beds per thousand. The figure shows the result of applying the New Haven benchmark to Boston: Boston’s adjusted bed supply was 54% higher than New Haven’s (3.712/ 2.404= 1.54). If the New Haven rate had prevailed in Boston, Boston would have had 1,005.5 fewer beds (the number in parentheses). This number is obtained by multiplying the population of the Boston hospital service area by its bed rate: 3.712 x 768,694 = 2,853.4. Had New Haven rates applied, the number allocated would have been 1,847.9 (2.404 x 768,694). The “excess” beds in Boston are calculated by subtraction: 2,853.4 - 1,847.9 = 1,005.5.

Figure 1.1.

Allocated Acute Care Hospital Beds in Selected Hospital Service Areas in the New England States Compared to the Boston, Massachusetts and New Haven, Connecticut Hospital Service Areas and to the U.S. Average (1993). Benchmarks are used in this volume (more...)

In Figure 1.1, Hartford, Connecticut’s, adjusted rates are demonstrated to have been 23% higher than the New Haven benchmark; the surplus is calculated as 288 acutecare beds in the Hartford hospital service area. The figure also benchmarks Boston’s level of bed supply to Hartford’s and New Haven’s. Hospital bed rates in Hartford were 20% lower than in Boston; when the Boston benchmark is applied to Hartford, 381 more beds are needed. If Boston’s rate were applied to New Haven, 506 more hospital beds would be needed. The figure also illustrates the use of the United States average as a benchmark.

Tables

Detailed information about each hospital service area in the East South Central States, including most of the variables presented in the Atlas, are presented at the end of Parts Two through Five. Part Six presents details concerning the contribution of specific hospitals to the total allocation of hospital beds and Medicare reimbursements for inpatient care in each hospital service area. It also includes information on the number of physicians who serve each hospital service area and the locations of their practices. A more extensive database is available on CD-ROM.

Strategies and Methods

Part Nine of the national volume of the Dartmouth Atlas of Health Care provides details about the methods used in the Atlas and an explanation of the distribution graphs and the measure of association, the R² statistic, used in both the national and regional Atlases. Since some hospital service areas have small populations, areas were excluded from maps and figures in the regional volumes if the standard error of their rates exceeded 10% of the national average rate; for surgical procedures, the maximum standard error was 20%. The minimum population size for inclusion thus differs among the variables, and is reported in the notes to Tables Two through Five.

The impact of sample size is greatest for the estimates of Medicare reimbursements, which are based on a 5% sample of Medicare claims. In the national volume, these estimates were based on a one-year sample (1993). To increase the precision of these estimates, the data for reimbursements presented in the regional Atlases are based on a two-year sample (1992-93); the denominators are the enrollee person-years for the same time period. The rates thus reflect the average annual rate for the two-year period, 1992-93.

About Rates in the Atlas

In order to make comparisons easier, all rates in the Atlas are expressed on a scale that results in at least one digit to the left of the decimal point (e.g., 98.4 primary care physicians per hundred thousand residents, rather than .984 per thousand). To achieve this, different denominators were used in calculating rates.

The levels of supply of hospital beds and hospital full-time equivalent employees and registered nurses are expressed as beds, employees, and registered nurses per thousand residents of the hospital service area, based on American Hospital Association data and census calculations.

Expenditures and reimbursements are expressed as dollars per capita or per Medicare enrollee, based on American Hospital Association data, Medicare claims data, and census calculations.

The numbers of physicians providing services to residents of hospital service areas are expressed as physicians per hundred thousand residents, based on American Medical Association and American Osteopathic Association data and census calculations.

The numbers of surgical and diagnostic procedures performed are expressed as procedures per thousand Medicare enrollees in the hospital service area, (or as procedures per thousand male Medicare enrollees in the area, in the case of prostate procedures) based on Medicare claims data.

Patient day rates are expressed as total inpatient days per thousand Medicare enrollees, based on Medicare claims data.

Making Fair Comparisons Between Hospital Service Areas

Some communities have greater needs for health care services and resources than others; for example, in some communities in Florida, as many as 60% of residents are over age 65. Other areas—including some with large college populations, or ski resorts—have much larger proportions of younger people. To ensure fair comparisons between areas, all rates in the Atlas have been adjusted to remove the differences that might be due to the different age and sex composition of local populations. This adjustment avoids identifying some areas as having high rates of utilization simply because of their larger proportions of elderly residents. When data were available, rates have also been adjusted for differences in race. The methods used to adjust these rates are explained in Part Nine of the national volume of the Dartmouth Atlas of Health Care.

Some areas, such as major urban centers, have higher costs of living than others. Such areas are likely to have high health care expenditures because the costs of personnel, real estate, and supplies are higher, and not necessarily because they are providing more services. Adjusting for such variation provides a more comparable measure of differences in real health care spending that is not simply due to differences in costs of living among areas. To ensure fair comparisons of health care expenditures, hospital expenditure rates and Medicare reimbursement rates were adjusted to take into account the differences between hospital service areas in costs of living.

The methods used to adjust for age, sex, race, and price of medical care are detailed in Part Nine of the national volume of the Dartmouth Atlas of Health Care.

Communicating With Us About the Atlas

Our Atlas Home Page on the World Wide Web contains Atlas information, including a summary of Dartmouth related research and electronic copies of some hard-to-find references. Please send us your comments on the Atlas, particularly suggestions on how to improve it in the future.

We are at http://www.dartmouthatlas.org.