Geographic Correlation Between Large-Firm Commercial Spending and Medicare Spending

Published Online: February 05, 2010
Michael E. Chernew, PhD; Lindsay M. Sabik, BA; Amitabh Chandra, PhD; Teresa B. Gibson, PhD; and Joseph P. Newhouse, PhD

Objective: To investigate the correlation between geographic variation in inpatient days, total spending, and spending growth in traditional Medicare versus the large-firm commercial sector.

Study Design: Retrospective descriptive analysis.

Methods: Medicare spending data at the hospital referral region (HRR) level were obtained from the Dartmouth Atlas. Commercial claims data from large employers were obtained from Thomson Reuters MarketScan Database for 1996-2006 and aggregated to the HRR level. County-level data on inpatient days per capita and market characteristics were obtained from the Area Resource File. We computed correlations between Medicare and commercial spending and spending growth, as well as Medicare and non-Medicare inpatient days, and examined traits of high- and low-spending HRRs in both sectors.

Results: We found a positive correlation between inpatient days per capita across counties, but a small inverse correlation between measures of commercial and Medicare spending across HRRs. Spending growth was weakly positively correlated across HRRs. Markets in the upper third of commercial spending had more concentrated hospital markets than markets in the lower third of commercial spending. The reverse was true for Medicare spending.

Conclusions: The positive correlation in utilization and lack of correlation in spending implies an inverse correlation in prices. This is consistent with evidence that the differences appear to be, at least partially, related to aspects of the market structure. If private markets are to work better to reduce cost, stronger efforts are needed to reduce provider market concentration and promote competitive pricing for healthcare services.

(Am J Manag Care. 2010;16(2):131-138)

This retrospective descriptive analysis investigated the correlation between geographic variation in inpatient days, total spending, and spending growth for traditional Medicare versus the large-firm commercial sector.


  • Commercial and Medicare spending were not highly correlated, although there was a positive correlation in hospital utilization.
  • Competition (or lack thereof) influences commercial spending differently than Medicare spending.
  • If private markets are to work better to reduce cost, stronger efforts are needed to reduce provider market concentration and promote competitive pricing for healthcare services.


Considerable research has documented variation in healthcare spending across geographic areas.1,2 For example, Martin et al3 reported striking geographic variation in health spending across the United States, with nearly a twofold difference in personal spending between the highest- and lowest-spending states in 2004; per capita personal healthcare spending in Massachusetts was $6683 while in Utah it was only $3792. Notably, in the Medicare program, areas with higher spending have not been found to have better healthcare. For example, Baicker and Chandra4 found that spending and quality of care were inversely related for a national sample of Medicare beneficiaries.

This variation reflects substantial differences in practice patterns. For example, considerable geographic variation in the frequency of discretionary procedures such as hip, knee, and spine surgeries for Medicare beneficiaries has been reported.5 Likewise, several studies have identified marked variation in the treatment of patients with acute myocardial infarctions (eg, use of noninvasive vs invasive management strategies).6-8 Fisher et al9 studied the quantity of care delivered to the chronically ill and found that the frequencies of hospitalization, diagnostic testing, and physician visits varied by geography and the healthcare system used by patients. The geographic variation was not explained by regional differences in illness levels or patient preference, suggesting that market factors such as local physician opinion and supply of medical resources may play a prominent role in defining regional practice patterns.

Research on variations in practice patterns has been very influential, but most of it has focused on Medicare spending.1,10,11 Although Medicare is a large and important program, currently covering about 45 million beneficiaries, the majority of individuals in the United States are insured through commercial plans. Policy conclusions stemming from Medicare-based research often implicitly assume that per capita spending by Medicare and commercial insurers is strongly related.

Indeed, several factors suggest Medicare and commercial spending should be positively correlated across markets. For example, physicians are likely to have similar practice styles across age groups for the same disease.12,13 In addition, to the extent that prices reflect common costs such as wages, prices should be positively correlated across different populations. Some existing empirical evidence at the hospital level suggests a positive correlation in utilization of inpatient care.14

Yet there also are reasons why Medicare spending and spending growth may differ from commercial spending across areas. First, the prevalences of disease or conditions may differ, leading to differences in services delivered. For example, commercial payers pay for childbirth, which is not relevant for the over-65 Medicare population. In contrast, services such as home care are much less frequent in the commercial population than in the Medicare population. Even common diseases that afflict both populations (eg, heart disease) may be treated differently in an over-65 Medicare patient as opposed to an under-65 patient who is commercially insured because of differences in comorbidities and frailty. Further, benefit packages differ. Prior to 2006, Medicare did not cover orally administered drugs, whereas the vast majority of large firms did.

Second, reimbursement methods differ. For example, traditional Medicare relies on an administered price system with few administrative controls on use. In contrast, commercial insurers negotiate rates with providers. For this reason we would expect the effects of hospital and provider competition to vary between Medicare and traditional insurers. Finally, although long-term spending growth per capita across the commercial and Medicare sectors has been similar, in any given year the picture may differ. If Medicare tightens reimbursement, hospitals in competitive markets may seek higher rates from commercial carriers to cover joint costs.15

This study documents geographic variation in healthcare spending by large firms (a subset of commercial spending) and compares large firm and Medicare spending across hospital referral regions (HRRs). For a number of reasons, such as differences in plan type and benefit generosity, data on commercial spending and Medicare spending are not strictly comparable.

It is also important to recognize that the factors associated with high levels of spending may not be the same factors that are associated with a high rate of spending growth.16,17 For example, researchers examining the impact of health maintenance organizations (HMOs) and high-deductible health plans on spending have noted that these plans may lead to “one time” savings that reduce spending at a point in time, but may not substantially alter the trajectory of spending.18-21


We used several data sources for our analysis. Medicare spending data for beneficiaries who are at least age 65 years came from the Dartmouth Atlas, which provided per capita age-, sex-, and race-adjusted Part A and Part B reimbursements for each of 306 HRRs in the United States. However, we used only Medicare spending on hospital and physician services to improve comparability with services commonly used by the commercially insured population. We excluded spending on home health services, durable medical equipment, and skilled nursing facilities. We adjusted all values for inflation using the All Items Consumer Price Index and expressed results as 2005 dollars.

Data on commercial spending came from the Thomson Reuters (Medstat) MarketScan Commercial Claims and Encounters Database, which collects administrative data for large firms. We used this database to measure mean spending per member per month on medical services (inpatient and outpatient spending) in each HRR. Our Thomson Reuters data include all plan types (indemnity, preferred provider organization, point of service, and HMO), although we excluded spending data from capitated plans, which may not provide data on all encounters delivered under capitation (capitated plans are similarly excluded from the Medicare data). There remained some variation in plan type within our commercial data. We elected to retain all available noncapitated data to reduce issues of selection if workers nonrandomly choose certain plan types within firms and to maintain sample size. Similarly, like much of the geographic-variation literature, we did not adjust for benefit design (eg, plan generosity in commercial plans, presence of supplemental coverage in Medicare). These plan and benefit differences would have affected the analysis only to the degree that they vary systematically across markets for Medicare differently than for our commercial sample.

We omitted spending on prescription drugs for comparability because this spending was not included in the Medicare spending measures we used. We included beneficiaries age 0 to 64 years who were not eligible for Medicare. Beneficiaries were assigned to HRRs based on their zip code of residence.

We created 2 samples of firms from the Thomson Reuters data. The first contained data from firms contributing at least 5 years of data between 1996 and 2006. The second contained firms contributing all years of data between 1996 and 2006, so that we could compare the same firms in 1996 and 2006.

The Thomson Reuters data have been widely used, but were imperfect for our task. Despite the Thomson Reuters data set containing between 16.9 million and 22.9 million observations per year, some HRRs in the smaller, 11-year subset of firms we used had fewer than 1500 member months for a given year. Therefore, estimates of mean spending for an HRR may have been materially affected by outliers. Moreover, the sample used in this study consisted of fewer than 60 large firms, and spending by employees of large firms may differ from that of employees of small firms. For example, benefit packages tend to be more generous in large firms.22 Despite these imperfections, these data are among the best available for our purpose. Nonetheless, we recognize that the results may not generalize to other commercial populations, let alone the remainder of the under-65 population.

To provide a rough insight regarding the contribution of price and utilization to the aggregate spending correlations, we analyzed inpatient days per capita for Medicare and non-Medicare beneficiaries in 2004 from the 2006 Area Resource File (ARF). These data are at the county level and are based on the location of the hospital rather than the patient. Because inpatient days are a measure of utilization, they exclude price effects and spending on outpatient services, but they do include Medicaid beneficiaries, the uninsured, and commercially insured individuals from small firms in addition to the large-firm commercially insured population.

The ARF also provides data about area population and market infrastructure, including hospital and physician supply (hospital beds per capita, primary care physicians and specialists per capita). Our measures of physician supply followed those of Starfield et al,23 who defined primary care physicians as all active patient care physicians in general practice, general family medicine, and general internal medicine. However, we do not include pediatrics in primary care. Other active patient care physicians were classified as specialists. Hospital infrastructure was measured as total hospital beds per capita.

PDF is available on the last page.


Recommended Reading

No Result Found