General FAQ

Why is this information important? What are the implications?

Health care spending is consuming an increasingly larger proportion of GDP every year, but there is little evidence that the amount we are spending is producing better health outcomes for populations or individual patients. Other countries spend far less per person and have better results. One way to address the question is to study which parts of the system appear to be producing “excess” levels of intervention, which are extremely costly but provide no additional benefit over other parts of the system that operate far more efficiently. Researchers estimate that up to 30% of current spending on health care is wasted. Finding that waste and eliminating it would not only help provide financial security for the Medicare program without loss of value to the people it covers, it would also help finance the expansion of coverage.

How are the methods used in this project different from other studies?

The Dartmouth Atlas Project uses a methodology, commonly known as small area analysis, which is population-based. The focus of small area analysis is on the experience of the population living in a defined geographic area or the population that uses a specific hospital. In contrast, many others studies use a “turnstile” approach, focusing on the number of procedures or hospitalizations in the hospital, without reference to the size of the population served.

Why does the Dartmouth Atlas Project focus on Medicare data? Are there similar variations in utilization and spending in the under-65 population?

The Centers for Medicare and Medicaid Services (CMS), the federal agency that collects data for every person and provider using Medicare health insurance, makes available a uniform national claims database for research purposes. There is no counterpart to this database for the commercially insured population. However, similar studies we have done using state all-payer data in Pennsylvania and Virginia, and with Blue Cross Blue Shield data in Michigan, have shown similar variations among the under-65 population.

Why don’t you have data for Medicare enrollees who are members of health maintenance organizations (HMOs)?

Health maintenance organizations receive capitated payments from Medicare – a fixed annual amount per enrollee – in exchange for the HMO providing all required services. Since HMOs do not submit individual claims to Medicare, we must exclude members of HMOs from our claims analyses.

What explains the differences in efficiency among different regions? Is it supply driven?

The supply of resources such as hospital beds and specialist physicians does drive utilization – where there are more hospital beds per capita, more people will be admitted (and readmitted more frequently) than in areas where there are fewer beds per capita. Economically, it is important for hospitals to make sure that all available beds generate as much revenue as they can, since an unoccupied bed costs nearly as much to maintain as an occupied bed. Similarly, where there are more specialist physicians per capita, there are more visits and revisits. Other reasons for the variations in efficiency are related to practice style – the way physicians in the region practice medicine (using more or fewer prescriptions or tests, for example).

What is the relationship between health care spending and quality of care?

Studies that have looked at the additional services provided in high-spending regions have shown that the higher volume of care does not produce better outcomes for patients. Patients in high-spending regions do not receive more “effective care” (services shown by randomized trials to result in better health outcomes, such as making sure that heart attack patients get proper medication). Nor do they receive more “preference-sensitive care” (elective surgical procedures which have both benefits and risks, where patients’ preferences should determine the final choice of treatment). Rather, the additional services provided to Medicare beneficiaries in higher-spending regions all fall into the category of “supply-sensitive care”: discretionary care that is provided more frequently when a population has a greater per capita supply of medical resources. In regions where there are more hospital beds per capita, patients will be more likely to be admitted to the hospital – and Medicare will spend more on hospital care. Where there are more intensive care unit beds, more patients will be cared for in the ICU – and Medicare will spend more on ICU care. The more CT scanners are available, the more CT scans patients will receive – and so on.

What do you mean more health care is not necessarily better?

The Dartmouth Atlas Project has observed, over the course of its research, that death rates in areas where there is less capacity and less utilization are not higher than death rates in areas where there is much higher capacity and utilization – that is, the additional investment in hospital and physician resources does not “pay off” in increased longevity. Studies by Dr. Elliott Fisher et al have indicated that there is higher mortality in high-resourced, high-utilization areas than in low-resourced, low-utilization areas. One explanation for this phenomenon is that the risks associated with hospitalizations and interventions – hospital-acquired infections, medication errors and the like – outweigh the benefits.

Evidence points out that more aggressive care in managing patient populations with chronic illness does not necessarily lead to longer length of life or improved quality of life. Are you insinuating that we shouldn’t do everything we can to save a life?

Ironically, research has found that in patients with chronic illnesses, more aggressive interventions result in shorter life expectancy, probably because of the risks associated with hospitalization. This indicates that the best strategy for extending the life of people with chronic illness is to focus on those activities that provide a survival benefit – better control of blood pressure for people with diabetes, for example – rather than on “heroic” end-of-life care.

Dartmouth Atlas research points out that frequent use of services is not associated with either better performance on technical measures of care or marginal improvements in survival and functional status. How can you convince people they don’t need additional care and how can you convince doctors not to recommend it?

A recent study reported that almost three-quarters of Americans say they have declined interventions that were recommended by their physicians, because they thought that it was unnecessary or the benefits did not outweigh the risks or side effects. Other studies have confirmed that informed patients want much less surgery, on average, than surgeons are inclined to perform. Making patients aware of the risks and trade-offs associated with treatment choices is one good way of reducing demand for such things as hospital admissions, redundant or unnecessary testing, and surgery when there are other options. Because physicians are reimbursed for activities, the system encourages them to do more. Paying physicians to spend more time advising patients about treatment alternatives (for example, lifestyle changes and medications, rather than bypass surgery), without penalizing them economically for doing less, is another important strategy for reducing utilization.

How does poverty impact health care spending?

Sick people require far more care than healthy people. For people who reported that they were in poor health, average annual Medicare spending was more than six times higher than for those who said they were in excellent health in 2005. Poverty also matters for health care spending: low-income people are sicker and tend to account for greater health care expenditures than those who are comparatively wealthier. However, our research has shown that regional differences in poverty and income explain almost none of the variation among regions; and, while health status does matter, it only accounts for about 18% of the variation in Medicare spending. More than 70% of the differences in spending cannot be explained away by the claim that patients in high-spending regions are poorer or sicker.

How do you determine how much care is too much?

By accurately measuring at what point more inputs do not result in better outcomes.

Whose fault is it?

Probably the most important driver of how health care resources are established and used is the current reimbursement system. Hospitals and doctors are paid for activities – hospitalizations, procedures, tests – and are economically punished for using less-invasive, less-costly strategies of care.

This research suggests savings that can be realized within the Medicare system. Don’t we need to look at the whole picture to truly realize savings?

Obviously more information about the non-Medicare population would add to our knowledge about what is going on in the system and how it could be improved. Lacking that information, however, we can say two things. The first is that, even if we redirected only Medicare into high-quality, high-efficiency patterns of resource allocation and utilization, we would realize tremendous gains in quality and reductions in spending. The second is that, in several state-based studies of all health insurance claims (both Medicare and commercial) we have determined that the variations in resources and quality in the non-Medicare populations closely resemble those in the Medicare population. So the experience of Medicare enrollees is a reliable predictor of the experience of the non-Medicare population.

However, a hospital’s ranking in terms of per capita spending may vary substantially for commercial payers based on market-negotiated (rather than CMS-set) unit prices and the greater spending on non-chronic conditions such as pregnancy. The best strategy for addressing these limitations would be for all payers and self-insured employers to work together to produce resource input and utilization data for cohorts across Medicare, Medicaid and commercially-insured patients.

May we use your data for our program or application?

Please refer to our Terms and Conditions of Data Use.

Hospital-Specific FAQ

How were hospitals selected?

We report data for acute care general hospitals, or those that provide a range of acute care services to Medicare fee-for-service patients. Hospitals were also selected by size, as determined by the number of persons “assigned” to each hospital (by linking Medicare claims by each enrollee to the hospital he or she used during the study period). The study was confined to hospitals with large enough populations to result in statistical stability and retain the confidentiality of patient information. Inpatient data for hospitals with at least 80 deaths during the study period are provided on the Web site; for Part B data, which is based on a 20% sample of deaths, a hospital had to have at least 400 total deaths (80 deaths in a 20% sample) during the study period to be included.

Why do you focus on patients who were chronically ill and in their last two years of life?

One reason is the growing concern about the way chronic illness is managed in the United States, and about the possibility that some chronically ill and dying Americans might be receiving too much care: more than they and their families actually want or benefit from. Our emphasis on this period of life is also motivated by our interest in developing measures of efficiency and performance that minimize the chance that variation in the care delivered in different regions and by different hospitals can be explained by differences in the severity of patients’ illnesses. By looking at care delivered to patients with similar illnesses during fixed intervals of time prior to death, we can say with assurance that the prognosis of all the patients in the cohort is identical – all were dead after the interval of observation. By further adjusting for difference in age, sex, race, and primary chronic illness, we believe that we have developed fair measures of the relative intensity of care provided to equally ill patients – comparisons for which differences among patients are an unlikely explanation

How do you ensure some patients were not more severely ill than others?

The study only focused on patients who died so we could be sure that patients were similarly ill across hospitals. By definition, the prognosis of all the patients in the cohort was identical – all were dead after the interval of observation. Therefore, variations cannot be explained by differences in the severity of individuals’ illnesses.

What are the medical conditions that define a patient as having a chronic illness?

To be assigned to our chronically ill cohort, a patient must have one of the following nine conditions: congestive heart failure, chronic lung disease, cancer, coronary artery disease, renal failure, peripheral vascular disease, diabetes, chronic liver disease or dementia. ICD-10-CM (2016 and later data) codes defining each condition can be found here. ICD-9-CM codes (2015 and earlier data) can be found here.

Why do the national average population and rates given in the state and regional tables not match the numbers given in the hospital tables?

In the state and regional studies, the study population was a 20% sample of resident enrollees with one or more of the nine chronic illnesses, regardless of whether they were hospitalized during the last two years of life. In the hospital-specific studies, only decedents who had one or more medical hospitalizations for one of the nine chronic illnesses were included. Medicare enrollees who were hospitalized with one or more of the nine chronic illnesses were assigned to the hospital most frequently used during the last two years of life. A 100% sample of deaths was used for the inpatient utilization rates; a 20% sample was used for Part B rates.

The rates given for inpatient sector spending do not match the rates given for inpatient facility reimbursements. What is the difference between these two measures?

Sector spending includes Part B (physician) spending that occurred at each site of care; Part B payments for physician services delivered in each type of facility (acute care hospital, skilled nursing facility, hospice, etc.) were added to the facility payments to determine overall spending in the sector. The facility reimbursement rates do not include Part B spending for physician services.

What do you mean by efficiency?

Our approach to evaluating relative efficiency is based upon the notion of benchmarking, which entails a comparison across hospitals (or regions) along the dimensions of both quality and resource use. For example, within a given market area, one could identify the most efficient hospital based upon its relative performance in terms of quality (equal or better to all others) and costs (using fewer resources than others). It is possible, however, that the most efficient hospital within a given market would still be less efficient than “benchmark” hospitals identified in other regions.

If a hospital is seen as inefficient, does this mean that it provides poor care?

Our studies do not directly measure the quality of care. Instead, they focus on what could be called overcare – hospitalizations and procedures that cost money but do not provide a corresponding benefit. (Large numbers of days in intensive care during the last six months of life, for example, neither extend life expectancy nor provide high quality of life for the patient.) Care is often described as “poor” if the process of care is poor; this study looks not at whether the thing was done right, but if whether the decision to provide the hospitalization or procedure was the correct decision to begin with.

Where can I find direct measures of hospital quality?

The quality of care can be evaluated using accepted technical process measures such as those that can now be found on the CMS’s Hospital Compare web site. We provide summary scores on five measures for treatment of heart attacks (AMI); two for congestive heart failure (CHF); and three for pneumonia, using methods developed by Jha et al. In addition, we report a composite score, which is the weighted average of the three condition-specific summary scores. These measures are available for hospitals that had at least 25 patients in the sample for each measure, as well as for HRRs and states (weighted averages of the scores for hospitals located in each region or state).

Will patients pay more out-of-pocket expenses with an inefficient hospital?

Patients and their families who choose hospitals that tend to deliver more intense care may have to pay for that extra care out of pocket. Medicare sets the overall price for physician services and pays 80% of that amount directly to the physician, leaving patients responsible for the remaining 20%, which they must pay unless they have supplemental insurance or are covered by Medicaid. Medicare also requires a 20% co-payment for durable medical equipment, such as wheelchairs and oxygen for home use. Therefore, the patient’s share of the cost of care can vary considerably depending upon which hospital is chosen.

How do we know that patients at some “outlier” hospitals are not really sicker (i.e. do they have more co-morbidities)? And if we say no, how do we prove that?

The Dartmouth Atlas uses standard statistical adjustment methods to adjust for differences in age, sex, race and the relative predominance of the nine severe chronic conditions among the populations of the hospitals we study. Even after these statistical adjustments are made, some hospitals have substantially different rates than would be expected given the level of illness and the age, sex and race composition of their populations, indicating that it is not sickness, but practice style (propensity to use more specialists and to treat patients inside the hospital) that results in such rates.

Your determination of the intensity of terminal care includes the% of patients who died during a hospitalization that included an admission to intensive care. You cite variation, yet why is this significant? Was there a higher or lower death rate associated with ICU admission?

This measure attempts to capture the relative aggressiveness of care at the end of life. Admission to intensive care is an extremely aggressive intervention that has no measurable value to a dying person. In light of the evidence that more aggressive care in managing patient populations with chronic illness does not lead to longer length of life or improved quality of life, higher scores on this measure can be viewed as an indicator of lower quality of death, and subjects that person to pain and suffering that do not extend life but diminish the quality of life.

If payers utilize this data, as your study suggests, and direct their chronic disease populations to low-cost and low-utilization hospitals, aren’t you limiting a patient’s life-saving options?

Quite the contrary. The evidence is that higher utilization does not extend life expectancy, and might be correlated with shorter life expectancy, compared to lower utilization. Therefore, sending people with chronic diseases to higher-efficiency, lower-utilization hospitals for their care could result in both lower spending and increased quality and length of life.

Medicare restricts the revenue that a hospital can make on a specific diagnosis per hospital stay. How is it possible that some hospitals can have many more ICU admissions, and more specialist visits – are these not part of DRG guidelines? In other words, is increasing the volume of certain types of services a way in which providers can “game” the system? Are there other such loopholes that Dartmouth Atlas research exposes?

The issue is not one of explicit “loopholes”. It is that two factors are important in judging efficiency: volume (the number of discharges) and price (the payment per discharge). Medicare’s diagnosis-related group (DRG) system uses few guidelines or sets of rules governing when to admit, discharge or treat patients with specific, measurable conditions. The system actually encourages gaming – to maximize revenues through providing more acute care because it pays better than preventive or primary care.

Because DRGs reimburse hospitals on a per-case (per-discharge) basis, it is possible for some hospitals to have more cases (or more discharges) during a given period of time. This would increase total payments, and would most likely be due to the greater availability of beds relative to the size of the population compared to other hospitals. Physician services are paid on a purely fee-for-service basis, so more frequent visits would result in higher payments.

An oversupply of beds makes it easier to admit and readmit (what is known as “churning”) patients in both acute care and ICU beds. Admitting physicians have discretion about whether or not to admit patients with many common conditions such as congestive heart failure, chronic pulmonary disease or cancer. In low-resource, low-utilization areas, such patients are treated outside the hospital. In high-resource, high-utilization areas, they are admitted and receive treatment as inpatients. Admission to ICU is also discretionary, and depends on physicians’ opinions about necessity and the available supply of ICU beds.

An additional opportunity for hospitals to increase revenue is “up-coding” patients in order to increase DRG payments by claiming patients are outliers – that they have more co-morbidities and complications than average.

Where can I find more information?

Comprehensive information on our hospital-specific data and methods is available in the Appendix on Methods of our 2011 report, “Trends and Variation in End-of-Life Care for Medicare Beneficiaries with Severe Chronic Illness.”

Research Methods FAQ

How does the Dartmouth Atlas Project get access to its data? Where does the data come from?

The very large claims databases come from the Centers for Medicare and Medicaid Services (CMS), the federal agency that collects data for every person and provider using Medicare health insurance. Access to this data is provided for research purposes. Other data sources include the U.S. Census, the American Hospital Association, the American Medical Association, and the National Center for Health Statistics.

Where can I access Dartmouth Atlas data?

Research Data: Research data is available for download through the Research Data Archives of the Atlas Data Website.

Geographic Crosswalks: ZIP code to HSA to HRR crosswalk files also available at the Atlas Data Website’s Supplemental Research Data page.

Coding Trends: Coding Trends data is available at the Atlas Data Website’s Supplemental Research Data page.

Other Atlas-related Data: Other Atlas-related and health care-related datasets can be found in our research data repository, Dartmouth Dataverse.

What is an HSA/HRR? How are the populations determined?

Hospital service areas (HSAs) are local health care markets for hospital care. An HSA is a collection of ZIP codes whose residents receive most of their hospitalizations from the hospitals in that area. HSAs were defined by assigning ZIP codes to the hospital area where the greatest proportion of their Medicare residents were hospitalized. Minor adjustments were made to ensure geographic contiguity. Most hospital service areas contain only one hospital. The process resulted in 3,436 HSAs.

Hospital referral regions (HRRs) represent regional health care markets for tertiary medical care. Each HRR contains at least one hospital that performs major cardiovascular procedures and neurosurgery. HRRs were defined by assigning HSAs to the region where the greatest proportion of major cardiovascular procedures were performed, with minor modifications to achieve geographic contiguity, a minimum population size of 120,000, and a high localization index. The process resulted in 306 hospital referral regions. More information on how HSAs and HRRs were defined is available in our Appendix on the Geography of Health Care in the United States.

What population does the Dartmouth Atlas Project study?

The Medicare population in an area includes those alive, age 65 to 99, and not enrolled in a risk-bearing health maintenance organization (HMO). For physician services, the population is restricted to a random sample of Medicare enrollees having Medicare Part B physician claims. For Medicare reimbursement rates, the population was restricted to a random sample belonging to both the Medicare A (inpatient) and B (physician services) programs.

How are an area’s health care resources measured and allocated?

An area’s health care resources consist of acute care hospital beds and medical personnel. As some patients seek care outside their area, these resources (beds, physicians, other hospital personnel, etc.) were allocated to HSAs in proportion to the area residents’ use of hospital services. This allocation procedure “transfers” resources from one area to another in proportion to how they are used. Areas with high migration will be allocated more resources but the allocated amount will reflect what is actually used in contrast to what exists in an area. For health policy purposes, it is necessary to be aware of this distinction since reduction in utilization in one area may require reduction in capacity of resources in an adjacent area.

Hospital beds and personnel. All short term medical and surgical hospitals, specialty and children’s hospitals are included with a few exceptions. Hospital beds included cribs, pediatric and neonatal bassinets, medical/surgical intensive care, and cardiac intensive care beds. Full-time equivalent hospital personnel are defined as the sum of full-time employees and half of the part time employees, not including medical or dental interns, residents and trainees.

To account for patients who live in one HSA but obtain medical care in another, hospital resources are allocated to HSAs in proportion to the Medicare hospital days provided by hospitals to that HSA. For example, if 60% of total Medicare inpatient days at a hospital were used by residents of the HSA where the hospital was located, then 60% of that hospital’s resources would be assigned to its HSA. If 20% of the Medicare patient days provided by that hospital were used by a neighboring HSA, 20% of the hospital’s resources would be assigned to that neighboring HSA.

Physician workforce. All physicians working at least 20 hours a week in clinical practice are included and classified according to their primary self-designated specialty.

Physicians provide services to patients residing both in and outside the HSA where their practices are located, so the physician workforce is adjusted for patient migration. Since information on the travel patterns of patients is not available, physicians are allocated in proportion to inpatient days in hospitals located in their HSAs. For example, if an HSA had four primary care physicians and if 25% of the patient days at the local hospital(s) were to residents of a neighboring HSA, then these physicians contributed one full-time equivalent primary care physician to the neighboring HSA.

What is a rate?

A rate is the number of events or amount of resources divided by the number in the population. For example, if an area with 100,000 Medicare enrollees has 810 hip fracture repairs, then the rate of hip fracture repair is 8.1 per 1,000 Medicare enrollees. For rare events, the rate is often re-scaled to reflect events per 100,000 persons.

Why are some rates suppressed?

Rates based on a count of fewer than 11 patients are not displayed for reasons of patient confidentiality. Rates with fewer than 26 expected events are reported in parentheses to indicate lack of statistical precision; for these rates, the margin of error is greater than 20%, so the estimate is considered statistically unreliable.

How are rates adjusted?

Most rates of utilization and spending are adjusted to the age, sex and race distribution of the national Medicare population using the indirect method. First, the national event rate for each age-sex-race category was computed. These rates were then applied to the HSA population to produce the expected number of events in the HSA, that is, the number of events that would have occurred in the HSA if its rate was the same as the national event rate. It is one way to standardize for different distributions of risk factors across areas. Click here for more information about indirect adjustment.

Measures of the care of the chronically ill population are adjusted for differences in age, sex, race, primary chronic illness, and the presence of more than one chronic conditions using ordinary least squares regression.

Where can I find more information?

Comprehensive information on such topics as files used, rate definitions, code specifications, physician classifications, allocation and adjustment methods, and so on is available in our Research Methods compendium. Information about our data and methods related to the care of chronic illness is available in the Appendix on Methods of our most recent report on the care of chronically ill patients during the last two years of life.