This Article Abstract Submit a response to this article Services E-mail this article to a friend Alert me to new issues of the journal

D A T A W A T C H
M E D I C A R E  Q U A L I T Y
W E B  E X C L U S I V E   

07 April 2004  
  

Medicare Spending, 
The Physician Workforce, 
And Beneficiaries’ Quality Of Care

Areas with a high concentration of specialists also show higher spending
and less use of high-quality, effective care.

By Katherine Baicker and Amitabh Chandra

ABSTRACT:

The quality of care received by Medicare beneficiaries varies across areas. We find that states with higher Medicare spending have lower-quality care. This negative relationship may be driven by the use of intensive, costly care that crowds out the use of more effective care. One mechanism for this trade-off may be the mix of the provider workforce: States with more general practitioners use more effective care and have lower spending, while those with more specialists have higher costs and lower quality. Improving the quality of beneficiaries’ care could be accomplished with more effective use of existing dollars.


Recent research has found large and persistent differences across states in the quality of care that Medicare beneficiaries receive.¹ One way to measure these differences is through differences in the use of effective, high-quality care, such as the administration of beta-blockers after heart attacks, mammograms for older women, influenza vaccines, or eye exams for diabetics. These procedures are relatively inexpensive, are known to have desirable medical benefits, and are rarely contraindicated. It is therefore puzzling that the use of these procedures varies so widely between states; for example, in 2000 the use of beta-blockers within twenty-four hours of admission for patients with heart attacks and without contraindications ranged from 50 percent in Alabama to 86 percent in New Hampshire.²

In this paper, we first determine whether quality differences can be explained by differences in Medicare spending. That is, are states where there is more spending per Medicare beneficiary also more likely to provide effective care? Clearly, spending more is unlikely to cause lower-quality care but rather serves as a marker for a particular style of health care provision or use of resources. Something in the underlying infrastructure or allocation of resources may drive both higher spending and lower quality of care. We next examine whether high-spending states provide more care along other dimensions, such as multiple specialist consultations, hospitalizations, and use of intensive care units (ICUs) in the last six months of life. Prior research has shown that end-of-life care is extraordinarily costly but not correlated with the underlying sickness of the population, patient outcomes, or patient satisfaction.³ Finally, we explore potential mechanisms through which intensive care might crowd out high-quality care. We analyze the effect of the underlying physician workforce (generalists versus specialists) on both spending and quality differences across states.

Study Data And Methods

Empirical design. We first examined the relationship between the provision of high-quality care for Medicare beneficiaries and Medicare spending at the state level. We performed this analysis at the state level mainly because of the availability of external state quality measures, but also because the states serve as useful proxies for geographic variation in care.⁴

We explored the determinants of state spending and quality using generalized least squares regressions weighted by the size of the Medicare population in each state (although unweighted regressions provide similar results). These cross-section results are not sensitive to the price and demographic adjustments described below, or to the exclusion of Medicare home health spending, which was dramatically curtailed by the Balanced Budget Act (BBA) of 1997.⁵ Neither are the results affected by the choice of year of the quality measures (1998–99 or 2000–01) or by the inclusion of state-level health maintenance organization (HMO) enrollment and the rate of heart attack discharges (adjusted for the age, sex, and race of the state population).

To assess whether these findings are driven by unmeasured differences in the underlying sickness of state populations, we used a subset of four highly informative quality measures that were available for both 1995 and 1999 from Medicare claims data from the Dartmouth Atlas project: beta-blockers administered at discharge, mammography every two years for women ages 65–69, and hemoglobin (HbA1c) monitoring and annual eye exams for diabetics. We analyzed the relationship between changes in the use of these quality measures and changes in spending within each state. This technique eliminates unobserved confounders that are fixed over time (for example, fixed differences in demographic structure, patients’ severity of illness, or reliance on outpatient clinics), although unobserved state-level confounders that change over time may still affect these results. We also included changes in the level of HMO enrollment among Medicare beneficiaries and the rate of heart attack discharges.

We used similar methods to examine whether higher Medicare spending is a marker for a different pattern of spending. We analyzed the relationship between spending and end-of-life care, such as the fraction of patients admitted to the ICU and the number of days spent in the hospital. Although this care may be costly and have less observed impact on health, it may be highly valued by patients, so we also analyzed the effect of spending on patient satisfaction.

Finally, we explored one of the mechanisms that may be responsible for the trade-off between high-quality health care and costly end-of-life care: the composition of the medical workforce. We regressed spending per Medicare beneficiary and overall quality rank on the number of specialists, general practitioners, and registered nurses per capita, controlling for the total number of physicians per capita, to explore the effect of changing the composition of the medical workforce.

Data. Quality measures. This study uses the twenty-four quality measures developed by the Medicare Quality Improvement Organization (QIO) and computed at the state level by Steve Jencks for 2000–01.⁶ These measures use samples of patient discharge records for the treatment of six common medical conditions (acute myocardial infarction, breast cancer, diabetes mellitus, heart failure, pneumonia, and stroke) and capture interventions and evaluations “for which there is strong scientific evidence and professional consensus that the process of care either directly improves outcomes or is a necessary step in a chain of care that does so,” such as the prescription of warfarin for atrial fibrillation or biennial eye examination for diabetics.⁷ Detailed risk adjustment is thus not critical for these measures, as few patients are contraindicated for these procedures. Jencks and colleagues ranked states for each measure and averaged the ranks (weighting each measure equally) to compute each state’s overall quality rank.

Medicare spending and use measures. We calculated Medicare reimbursement per beneficiary at the state level using Medicare claims data from the Dartmouth Atlas projects; we included spending for fee-for-service (FFS) Medicare beneficiaries.⁸ This spending is adjusted in three ways. First, spending is adjusted for inflation using the Consumer Price Index (CPI).⁹ Second, differences in state price levels are taken into account using a state-specific cost-of-living adjustment.¹⁰ Third, spending is adjusted for the age, sex, and race of the states’ Medicare populations.

We used several other measures computed from the Medicare claims data, including the number of days Medicare beneficiaries in their last six months of life spent in a hospital and what fraction of these beneficiaries are admitted to the ICU. These measures abstract from unmeasured illness confounders by focusing on the deceased. Other research has established that such care is pervasive in areas that have a lot of beds, specialists, and health care facilities and has also demonstrated that its provision does not improve patient outcomes or satisfaction.¹¹ In some specifications we controlled for the number of acute myocardial infarction (AMI) discharges and AMI mortality in each state, adjusted for the age, sex, and race composition of the population, also computed from the claims data.

Satisfaction measures and HMO penetration rates. Patient satisfaction measures come from the 1992–1995 Medicare Current Beneficiary Survey (MCBS). We used measures of overall satisfaction, satisfaction with access to care, and satisfaction with providers’ technical proficiency. We also used these data to compute state-specific measures of Medicare HMO enrollment.

Workforce measures. Data on the number of specialists, general practitioners, and registered nurses (RNs) were obtained from the 2003 Area Resource File (ARF).¹² The ARF gathers information from the American Medical Association (AMA) Physician Masterfile and the County Hospital File and is reported at the county level. We summed county-level data into state measures. We computed per capita workforce measures for each state by dividing state physician workforce counts by population counts from the 2000 census.

Study Results

Higher spending is associated with lower quality of care as seen in Exhibits 1–3. These relationships are statistically significant: Spending is not merely uncorrelated with the quality of care provided.¹³ Exhibit 4 quantifies the relationship between an increase in spending of $1,000 per beneficiary (roughly the rise in average spending from 1995 to 1999) and the twenty-four individual quality measures, as well as end-of-life care and patient satisfaction. For convenience, we also report (from the work of Jencks and colleagues) the percentage of Medicare beneficiaries nationwide who received the indicated evaluation/intervention in 2000. The effect of increased spending on fifteen of the measures is estimated to be negative and statistically significant, and there is no statistical effect on the remaining nine. The first row demonstrates that a state spending $1,000 more per beneficiary dropped almost ten positions in overall quality ranking (p < .001). Similarly, states spending $1,000 more per Medicare beneficiary had beta-blocker usage rates at discharge that were 3.5 percentage points lower (p < .02), and mammography rates that were 2.1 percentage points lower (p < .01) than the average usage in 2000.¹⁴

We also explored the role of two other covariates. We included the fraction of Medicare beneficiaries enrolled in HMOs and discharges for heart attacks (or AMIs) adjusted for age, sex, and race. HMO enrollment may affect the cost and quality of care provided, and AMI discharges capture an important component of the overall health of the Medicare population. The results were not affected by the inclusion of these variables.

To ensure that these results are not the artifact of omitted variables, such as the possibility that some states have (unobserved) systematically different patients than others, the same relationship is illustrated in changes: Do states that increase their spending also improve their quality of care? The results in Exhibit 5 show that there is no correlation between changes in the use of high-quality care and changes in Medicare spending. The negative association between spending and quality persists; states that increased spending reduced their usage of beta-blockers, mammograms, and annual eye exams for diabetics.

Although this method eliminates all confounders that are fixed over time by differencing them away, we may not have accounted for some state-specific factors that change over time. Here, too, we included the change in the fraction of beneficiaries enrolled in an HMO (which increased substantially over this period) and the change in adjusted discharges for AMI (which would proxy for an overall change in the health of the state population). The bottom panel of Exhibit 5 shows that the inclusion of these measures does not affect the reported coefficients: There is still no correlation between quality of care and Medicare spending. Together, these results also validate the claim that the quality measures are not sensitive to risk adjustment.

Where does the money in high-spending states go, if not to highly effective care? It seems to be spent on expensive health care that has not been shown to have a positive effect on patient satisfaction or health outcomes. Exhibit 4 shows a positive relationship between Medicare spending and the percentage of Medicare beneficiaries who were admitted to the ICU (or the number of days beneficiaries spent in the hospital) during their last six months of life. Medicare patients in states that spent $1,000 more per beneficiary spent an average of 1.3 more days in the hospital (p < .01) and were 3.9 percent more likely to be admitted to an ICU (p < .005).¹⁵ These increases do not seem to be associated with higher levels of patient satisfaction. As shown at the bottom of Exhibit 4, spending is uncorrelated with several different measures of patient satisfaction.

What causes some states to be high spenders and provide lower-quality care, while others are low spenders and provide higher-quality care? One possibility is the composition of the medical workforce. Exhibits 6–11 examine this hypothesis, illustrating the relationship between the medical workforce, spending, and quality. The exhibits adjust for the total number of physicians in a state and study the effect of specialists (Exhibits 6 and 7), general practitioners (Exhibits 8 and 9), and nurses (Exhibits 10 and 11) per capita on overall quality rank and Medicare spending. We are thus examining the effect of changing the composition of the medical workforce, holding the overall size of the physician workforce constant. Together, these workforce measures can explain 42 percent of state-level variation in Medicare spending per beneficiary. These exhibits show that states where more physicians are general practitioners show greater use of high-quality care and lower cost per beneficiary. Increasing the number of general practitioners in a state by 1 per 10,000 population (while decreasing the number of specialists to hold constant the total number of physicians) is associated with a rise in that state’s quality rank of more than 10 places (p < .0005) as well as a reduction in overall spending of $684 per beneficiary (p < .0005). Conversely, states where more physicians are specialists have lower-quality care and higher cost per beneficiary. The estimated effect of increasing the fraction of specialists by 1 per 10,000 is a drop in overall quality rank of almost 9 places (p < .005) and an increase in spending of $526 per beneficiary (p < .004). The supply of nurses does not seem to affect either the use of high-quality care or total spending.

It is possible that although areas with more specialists do not provide higher-quality care along these dimensions, they may be better at the treatment of more acute conditions.¹⁶ It is also possible that areas “specialize” in different types of care: Some areas specialize in primary care, while others may specialize in the delivery of technologically aggressive care for heart attacks.¹⁷ We do not find evidence of this here: States with more specialists have neither lower mortality rates from all causes nor reduced post-AMI mortality.¹⁸

Discussion And Policy Implications

States that spend more per Medicare beneficiary are not states that provide higher quality care. In fact, additional spending is positively correlated with end-of-life care but negatively correlated with the use of effective care. While higher spending per se is unlikely to cause a drop in the use of high-quality care, it seems to be a marker for a particular pattern of care. Our analysis suggests that the mix of the physician workforce plays a critical role in the use of highly effective care. States with relatively more general practitioners have both higher rates of use of effective care and lower spending. Surprisingly, we find no relationship between nurses and the provision of high-quality care.

Given the reliance on cross-area variation in spending and quality, inferences about causal mechanisms should be made with great caution. First, ecological inferences always raise concerns about omitted variables, such as risk adjustment or legal environment. This is unlikely to be a problem in this analysis, for three reasons. First, the QIO quality measures were specifically selected to be robust to the absence of risk adjustment. Second, for incomplete risk adjustment to drive the cross-section results in Exhibit 4, it must also be the case that sicker patients medically require less of the “high-quality” care—a highly unlikely scenario. Third, the results are equally strong in the within-state panel-data analysis summarized in Exhibit 5, which controls for any persistent differences in illness in state populations, malpractice laws, or regulations.

A second concern might be that specialists locate in areas where patients are sicker and that sicker patients are more likely to be hospitalized for longer stays or admitted to the ICU. If this were true, then the positive relationship between specialists and end-of-life spending could be spurious. Here, too, several factors limit this potential bias. First, examining care that is based only on the sample of deceased people implicitly controls for the underlying sickness of the patient population. Furthermore, other researchers have found that underlying population risk does not seem to drive the presence of specialists and that outcomes are not improved by increased access to these specialists. In particular, in the area of neonatology, specialists are associated with neither higher risk nor lower mortality.¹⁹ The results on the ineffectiveness of specialists for the provision of high-quality care are thus consistent with the findings of a broader literature.

What, then, are the policy implications of the negative relationship between spending and quality? It clearly does not suggest that we mandate lower spending, because it is probably not spending per se that reduces quality. Spending captures many aspects of local health care delivery systems, such as physician practice styles, composition of the medical workforce, and capacity constraints. Therefore, naïve policies that simply target spending could have the undesirable effect of reducing the quality of care in high-spending states even more. Also, the quality measures we use do not capture the totality of health care provision. Although specialists may not drive the provision of effective care, they often provide better care in their area of specialty.²⁰ This suggests that specialists are clustered in areas where costly intensive care crowds out high-quality care and that one mechanism for this is a lesser presence of general practitioners. Encouraging greater access to general practitioners, or involving specialists in the provision of effective care, could improve the overall quality of care received by elderly Americans.

With Medicare’s mounting fiscal crisis, understanding the relationship between the variation in Medicare spending and beneficiaries’ quality of care is critical. The negative relationship we found between spending and quality and the factors that drive it are of immediate concern. Policies that improve quality of care (such as establishing national practice benchmarks for basic quality measures) need not be costly and could even improve Medicare’s financial solvency.²¹

This research was funded by National Institute on Aging (NIA) Grant no. P01 AG19783-02. Amitabh Chandra acknowledges the support of the Rockefeller Center at Dartmouth. The authors are grateful to Jonathan Skinner, Douglas Staiger, Jack Wennberg, and two astute anonymous referees for helpful comments and to Dan Gottlieb for help with the Medicare survey data. The opinions reflected in this paper are those of the authors and should not be attributed to the National Bureau of Economic Research or the NIA.

NOTES

1. S.F. Jencks et al., “Quality of Medical Care Delivered to Medicare Beneficiaries,” Journal of the American Medical Association 284, no. 13 (2000): 1670–1676; S.F. Jencks, E.D. Huff, and T. Cuerdon, “Change in the Quality of Care Delivered to Medicare Beneficiaries, 1998–1999 to 2000–2001,” Journal of the American Medical Association 289, no. 3 (2003): 305–312; E.A. McGlynn et al., “The Quality of Health Care Delivered to Adults in the United States,” New England Journal of Medicine 348, no. 26 (2003): 2635–2645; E.S. Fisher and J.S. Skinner, “Comparing the Health Care of States,” Providence Journal-Bulletin, 17 March 2001; E.S. Fisher et al., “The Implications of Regional Variations in Medicare Spending, Part 1: The Content, Quality, and Accessibility of Care,” Annals of Internal Medicine 138, no. 4 (2003): 273–287; and E.S. Fisher et al., “The Implications of Regional Variations in Medicare Spending, Part 2: Health Outcomes and Satisfaction with Care,” Annals of Internal Medicine 138, no. 4 (2003): 288–298.
2. Jencks et al., “Quality of Medical Care”; and Jencks et al., “Change in the Quality of Care.”
3. Fisher et al., “The Implications of Regional Variations in Medicare Spending, Part 1”; Fisher et al., “The Implications of Regional Variations in Medicare Spending, Part 2”; and J.E. Wennberg, E.S. Fisher, and J.S. Skinner, “Geography and the Debate over Medicare Reform,” Health Affairs, 13 February 2002, content.healthaffairs.org/cgi/content/abstract/hlthaff.w2.96 (3 March 2004).
4. An alternative geographic unit of analysis is the Hospital Referral Region (HRR) constructed by the Dartmouth Atlas of Health Care. These (smaller) regions capture the geographic level at which health care for the Medicare population is delivered. In parallel work we have performed similar analyses for the subset of quality indicators that we could construct at the HRR level. K. Baicker and A. Chandra, “The Productivity of Physician Specialization: Evidence from the Medicare Program,” American Economic Review (forthcoming). This analysis yielded similar results.
5. H.L. Komisar, “Rolling Back Medicare Home Health,” Health Care Financing Review 24, no. 2 (2002): 33–55.
6. Jencks et al., “Quality of Medical Care”; and Jencks et al., “Change in the Quality of Care.”
7. Jencks et al., “Change in the Quality of Care,” 1670.
8. J.E. Wennberg and M.M. Cooper, eds., The Dartmouth Atlas of Health Care 1999 (Chicago: American Hospital Association, 1999); J.E. Wennberg and M.M. Cooper, eds., The Dartmouth Atlas of Health Care 1996 (Chicago: American Hospital Association, 1998); and D.E. Wennberg and J.D. Birkmeyer, eds., The Dartmouth Atlas of Cardiovascular Health Care (Chicago: American Hospital Publishing, 2000).
9. Bureau of Labor Statistics, Inflation Calculator, data.bls.gov/cgi-bin/cpicalc.pl (3 March 2004); and U.S. Census Bureau, “Resident Population of the Fifty States, the District of Columbia, and Puerto Rico,” 1 April 2000, www.census.gov/population/www/cen2000/respop.html (3 March 2004).
10. H.B. Leonard and J.H. Walder, The Federal Budget and the States: Fiscal Year 1999, 15 December 2000, www.library.unt.edu/gpo/ACIR/ir/FY1999.pdf (3 March 2004).
11. Fisher et al., “The Implications of Regional Variations in Medicare Spending, Part 1”; Fisher et al., “The Implications of Regional Variations in Medicare Spending, Part 2”; and Wennberg et al., “Geography and the Debate over Medicare Reform.”
12. National Center for Health Workforce Analysis, Area Resource File, U.S. Department of Health and Human Services, 2003.
13. We focused on quality rank because of the attention given to this statistic in the original Jencks papers. However, since small differences in the overall index could result in large differences in the final ranking, we studied the relationship between spending and an index of quality usage. We constructed this index by summing the usage rate of each quality measure (weighting each measure equally). We ignored the time to PTCA and thrombolytic therapy since these measures are inversely correlated with quality. Our index is virtually identical to the Jencks rankings. A regression of our index on average spending predicts that a spending increase of $1,000 is correlated with a decline of 0.7 standard deviation in the value of the index. Exhibits 1–3 illustrate the relationship between spending and two subcomponents of the quality index, and Exhibit 4 shows the regression estimates for all subcomponents.
14. In Exhibits 1–3 it can be seen that the inclusion of Hawaii reduces the magnitude of the estimated relationship; eliminating Hawaii results in an even stronger negative relationship between quality and spending.
15. Fisher et al., “The Implications of Regional Variations in Medicare Spending, Part 1”; and Fisher et al., “The Implications of Regional Variations in Medicare Spending, Part 2” demonstrate that the correlation between end-of-life spending and average per capita Medicare spending is 0.83. A large portion of the estimated relationship is therefore driven by end-of-life spending. Since additional dollars are always spent on the margin, it makes sense to include end-of-life spending in total spending.
16. J.Z. Ayanian et al., “Specialty of Ambulatory Care Physicians and Mortality among Elderly Patients after Myocardial Infarction,” New England Journal of Medicine 347, no. 21 (2002): 1678–1686; and J.G. Jollis et al., “Outcome of Acute Myocardial Infarction According to the Specialty of the Admitting Physician,” New England Journal of Medicine 335, no. 25 (1996): 1880–1887.
17. A. Chandra and D. Staiger, “Network Externalities in Health Care: Evidence from Variation in the Treatment of Acute Myocardial Infarction” (Working Paper, Department of Economics, Dartmouth College, Hanover, New Hampshire, 2003).
18. We calculated state-specific mortality rates among patients admitted post-AMI, adjusting for age, sex, race, and ten comorbidities using the Medicare claims data.
19. D.C. Goodman et al., “Are Neonatal Intensive Care Resources Located According to Need? Regional Variation in Neonatologists, Beds, and Low Birth Weight Newborns,” Pediatrics 108, no. 2 (2001): 426–431; and D.C. Goodman et al., “The Relation between the Availability of Neonatal Intensive Care and Neonatal Mortality,” New England Journal of Medicine 346, no. 20 (2002): 1538–1544.
20. Ayanian et al., “Specialty of Ambulatory Care Physicians”; and K.B. Wells and R. Sturm, “Care for Depression in a Changing Environment,” Health Affairs 14, no. 3 (1995): 78–90.
21. G.R. Wilensky, “The Implications of Regional Variations in Medicare—What Does It Mean for Medicare?” Annals of Internal Medicine 138, no. 4 (2003): 350–351.


Katherine Baicker (kbaicker{at}dartmouth.edu) is an assistant professor of economics at Dartmouth (Hanover, New Hampshire), a senior research associate at the Center for the Evaluative Clinical Sciences at Dartmouth Medical School, and a faculty research fellow at the National Bureau of Economic Research (NBER). Amitabh Chandra (achandra{at}dartmouth.edu) is an assistant professor of economics at Dartmouth; a senior research associate at the center, a faculty research fellow at NBER; a research fellow of the IZA in Bonn, Germany; and coeditor of Economics Letters.


DOI: 10.1377/hlthaff.W4.184
©2004 Project HOPE–The People-to-People Health Foundation, Inc.