Perspective

Geographic Variation in Medicare Drug Spending

Yuting Zhang, Ph.D., Katherine Baicker, Ph.D., and Joseph P. Newhouse, Ph.D.

N Engl J Med 2010; 363:405-409July 29, 2010

Article

The widespread geographic variation in Medicare spending has garnered a great deal of attention in the health care reform debate, both as a marker of inefficient resource use and as a window into potential strategies for improving the quality and value of U.S. health care. Analyses to date, however, have focused solely on inpatient and outpatient spending (Parts A and B) and have lacked information on pharmaceutical spending (Part D). With drug spending accounting for a rising share of total health care spending, these data limitations hinder interpretation of the variation in spending: do Medicare patients who spend more on pharmaceuticals to control their chronic conditions have fewer physician visits, reducing total variation — or do more physician visits lead to more prescriptions, amplifying variation? Without information on total spending, it is impossible to know whether variation in spending on medical services is masking substitution of one type of care for another.

We have brought data on the Medicare drug benefit (Part D) to bear on these questions, to provide a more complete picture of the patterns of utilization and spending in different regions. We obtained both pharmacy and medical data on a 5% random sample of Medicare beneficiaries who were enrolled in stand-alone Part D plans in 2007. We restricted our study sample to beneficiaries who were between 65 and 99 years of age in 2007, who were alive on December 31, 2007, and who were enrolled for the full year in Medicare Parts A and B and in a stand-alone Part D plan. Thus, these data, like those in the Dartmouth Atlas,1 exclude individuals enrolled in Medicare Advantage (Part C) plans, because medical spending data for these beneficiaries are not available. Unlike the Dartmouth data, however, these data also exclude individuals with employer-provided retiree drug insurance, since they do not purchase stand-alone Part D plans. The resulting sample consisted of 533,170 beneficiaries. We assigned each beneficiary to 1 of the 306 Dartmouth hospital-referral regions on the basis of the beneficiary's ZIP Code of residence.

We adjusted medical spending for local price-level differences using factor prices given to us by the Medicare Payment Advisory Commission.2 We did not adjust drug spending for price differences, because the variation in drug prices among hospital-referral regions is negligible (see tableVariation in Annual Total Medicare Spending, Nondrug Medical Spending, and Pharmacy Spending and Counts of Monthly Prescriptions Filled per Person in Different Hospital-Referral Regions. for more details).3 We adjusted both types of spending for individual-level demographics (age, sex, and race), eligibility for Medicaid, a beneficiary's qualifying or not qualifying for a federal low-income subsidy for the Part D plan, and characteristics of the Part D insurance plan such as whether generic drugs were covered during the standard coverage gap (“doughnut hole”). We also adjusted both types of spending for the health characteristics of the beneficiary — another difference from many of the Dartmouth analyses. In the case of medical spending, the health measures we used were the Centers for Medicare and Medicaid Services (CMS) Hierarchical Condition Category (HCC) scores; in the case of pharmaceutical spending, we used the analogous CMS-RxHCC scores4 (see table for details). After making these adjustments, we calculated the average medical and pharmacy spending in each hospital-referral region, holding constant the variables just described and thus isolating variation among regions that is not driven by differences in prices, population characteristics, or health. The table, Figure 1Figure 1Quintiles of Adjusted Annual Drug Spending per Beneficiary, According to Hospital-Referral Region., and Figure 2Figure 2Quintiles of Adjusted Annual Total Medical Spending per Beneficiary, According to Hospital-Referral Region. show the variation among hospital-referral regions in medical spending, drug spending, and total spending after removal of the variation attributable to the aforementioned factors. It yields four major findings.

First, pharmaceutical spending represents a substantial share of total Medicare spending — more than 20%, on average — and it also varies substantially among hospital-referral regions, with the highest-spending region spending 60% more per beneficiary on pharmaceuticals than the lowest. This variation is driven by variation in both the drugs prescribed and the number of prescriptions filled each month, with physicians in higher-spending areas prescribing both more drugs and more expensive drugs.

Second, however, medical spending varies across hospital-referral regions even more than drug spending does. The ratio of spending at the 75th percentile for medical care exclusive of drug spending to that at the 25th percentile is 1.16, with a coefficient of variation of 0.12, whereas the corresponding ratio for pharmaceutical spending is only 1.11, with a coefficient of variation of 0.08.

Third, pharmaceutical spending and medical (nondrug) spending are only weakly correlated across hospital-referral regions (r=0.10, P=0.07). In other words, the areas where medical spending is the highest do not have systematically higher-than-average or lower-than-average drug spending. This weak correlation is consistent with drugs' being a substitute for medical care for some patients and a complement to medical care for others. This result is sensitive to the adjustment for health status: if we do not adjust for patient-level health status, we obtain a much higher correlation, of 0.60, which is consistent with sicker patients' consuming more of both types of care on average and highlighting the importance of taking patient characteristics into account when making these comparisons. It is, however, possible that in higher-spending regions, physicians are coding patients as sicker than similar patients are being coded in lower-spending regions, in which case the true correlation would be higher than the 0.10 that we estimated.5

Fourth, the variation in total Medicare spending is only slightly lower once pharmaceutical spending is accounted for; the ratio of the 75th percentile for total spending to the 25th percentile is 1.13, with a coefficient of variation of 0.10.

Thus, areas with high medical spending do not have offsetting lower pharmaceutical spending; in fact, if the coding practices in different regions are not too dissimilar, the substantial variation in pharmaceutical spending does not seem to be strongly associated with variations in medical spending at all. Spending on pharmaceuticals itself is variable and thus warrants scrutiny similar to that given to medical spending, in order to glean lessons about optimal prescribing, insurance characteristics, and resource allocation. Our findings reinforce the importance of understanding the drivers of geographic variation, since increases in medical spending or pharmaceutical spending do not appear to be associated with offsetting savings in the other realm. Using this more complete measure of spending reveals that area-level variation in total spending is not driven primarily by patient characteristics. These data may offer us an opportunity to gain insight into the underlying causes of the intensity of use of health care resources and the potential for public policy actions to improve the value of the health care delivered in the United States.

Disclosure forms provided by the authors are available with the full text of this article at NEJM.org.

This article (10.1056/NEJMp1004872) was published on June 9, 2010, at NEJM.org.

Source Information

From the Department of Health Policy and Management, Graduate School of Public Health, University of Pittsburgh, Pittsburgh (Y.Z.); the Department of Health Policy and Management, Harvard School of Public Health (K.B., J.P.N.), and the Department of Health Care Policy, Harvard Medical School (J.P.N) — both in Boston; and the Harvard Kennedy School of Government, Cambridge, MA (J.P.N.).

References

References

1. 1

   The Dartmouth atlas of health care 1999. Chicago: American Hospital Association Press, 1999.
   

2. 2

   Report to the Congress: measuring regional variation in service use. Washington, DC: Medicare Payment Advisory Commission, 2009. (Accessed June 7, 2010, at http://www.medpac.gov/documents/Dec09_RegionalVariation_report.pdf.)
   

3. 3

   MaCurdy T, Gibbs J, Theobald N, DeLeire T, Kautz T, O'Brien-Strain M. Geographic variation in drug prices and spending in the Part D program. Baltimore: Centers for Medicare & Medicaid Services, 2009. (Accessed June 11, 2010, at http://www.cms.gov/Reports/Downloads/MaCurdy_RxGeoPrice_TechReport_2009.pdf.)
   

4. 4

   Hierarchical Condition Category (CMS-HCC) model software and Part D risk adjustment software. Baltimore: Centers for Medicare & Medicaid Services, 2010. (Accessed June 7, 2010, at http://www.cms.hhs.gov/MedicareAdvtgSpecRateStats/06_Risk_adjustment.asp.)
   

5. 5

   Song Y, Skinner J, Bynum J, Sutherland J, Wennberg JE, Fisher ES. Regional variations in diagnostic practices. N Engl J Med 2010;363:45-53
   Full Text | Web of Science | Medline

Citing Articles (10)

Citing Articles

1. 1

   Donohue, Julie M., Morden, Nancy E., Gellad, Walid F., Bynum, Julie P., Zhou, Weiping, Hanlon, Joseph T., Skinner, Jonathan, . (2012) Sources of Regional Variation in Medicare Part D Drug Spending. New England Journal of Medicine 366:6, 530-538
   Full Text

2. 2

   Brian K. Kit, Cynthia L. Ogden, Katherine M. Flegal. (2012) Prescription Medication Use Among Normal Weight, Overweight, and Obese Adults, United States, 2005–2008. Annals of Epidemiology 22:2, 112-119
   CrossRef

3. 3

   B. Bongue, M. L. Laroche, S. Gutton, A. Colvez, R. Guéguen, J. J. Moulin, L. Merle. (2011) Potentially inappropriate drug prescription in the elderly in France: a population-based study from the French National Insurance Healthcare system. European Journal of Clinical Pharmacology 67:12, 1291-1299
   CrossRef

4. 4

   J.C. Naveiro-Rilo, L. Flores-Zurutuza, D. Diez-Juárez, F. González-Álvarez, A. Romero-Blanco, A. Pérez-Laorden. (2011) Factores asociados a la variabilidad en la utilización de recursos en la consulta de los médicos de atención primaria. SEMERGEN - Medicina de Familia 37:10, 540-548
   CrossRef

5. 5

   Adam Licurse, Aaron S. Kesselheim. (2011) Conflicts of Interest and the Future of Medicine: The United States, France, and Japan. Journal of Bioethical Inquiry 8:4, 383-386
   CrossRef

6. 6

   THOMAS G. MCGUIRE, JOSEPH P. NEWHOUSE, ANNA D. SINAIKO. (2011) An Economic History of Medicare Part C. Milbank Quarterly 89:2, 289-332
   CrossRef

7. 7

   Jonathan Skinner. 2011. Causes and Consequences of Regional Variations in Health Care. , 45-93.
   CrossRef

8. 8

   Zhang, Yuting, Baicker, Katherine, Newhouse, Joseph P., . (2010) Geographic Variation in the Quality of Prescribing. New England Journal of Medicine 363:21, 1985-1988
   Full Text

9. 9

   Richard Sullivan, Arnie D Purushotham. (2010) The Goldilocks' problem of cancer medicines. The Lancet Oncology 11:11, 1017-1018
   CrossRef

10. 10

    Joan B. Soriano, Marc Miravitlles, Luis Borderías, Enric Duran-Tauleria, Francisco García Río, Jaime Martínez, Teodoro Montemayor, Luis Muñoz, Luis Piñeiro, Guadalupe Sánchez, Joan Serra, Juan José Soler-Cataluña, Antoni Torres, Jose Luis Viejo, Víctor Sobradillo-Peña, Julio Ancochea. (2010) Geographical Variations in the Prevalence of COPD in Spain: Relationship to Smoking, Death Rates and other Determining Factors. Archivos de Bronconeumología (English Edition) 46:10, 522-530
    CrossRef