Geographical Variation in Health Spending Across the US Among Privately Insured Individuals and Enrollees in Medicaid and Medicare

This cross-sectional study evaluates geographic variation in US health spending among privately insured individuals and enrollees in Medicaid and Medicare.


Construction of Spending Per Beneficiary and Inpatient Days Per Beneficiary
There are modest differences in how we measure spending per beneficiary across each payor. In our privately insured sample, we sum the allowed amounts, which capture the total of the insurer payments and co-pays, across all medical claims annually by member. For the Medicare sample, we also sum across the spending on each claim. In practice, the Dartmouth Atlas measures are limited to what the federal government pays, and thus do not capture copays and deductibles. However, because Part B deductibles are a constant fraction of payments (e.g., a 20 percent coinsurance payment), relative variations across regions are likely to be largely unaffected. Finally, in the Medicaid program, a large portion of Medicaid enrollees are in managed care plans. As a result, our Medicaid spending measure is the sum of both the allowed amounts on medical claims for individuals in Medicaid fee-for-service and the capitated payments that the state Medicaid agency pays for individuals on managed care plans. Because Medicaid imposes virtually no (or very small) copayments, there is no distinction between government payments and allowed amounts.
We sum spending by member and payor annually and weight our hospital referral region (HRR)-level spending by the number of months each individual was enrolled. We then adjust our spending measure, by payor, for age (for private insurance 18-24 and 10-year intervals from age 25-64, for Medicare 5-year intervals from age 65-84 and 85+, and for Medicaid 5-year intervals from age 0-89 and 90+) and sex using indirect standardization.
To measure inpatient days per beneficiary by payor, we rely on the admission date and discharge dates recorded in the inpatient claims. We then sum inpatient days by payor and construct measures of inpatient days per beneficiary per payor, per HRR. We adjust inpatient days per beneficiary by age and sex using the same indirect standardization as above.

Construction of Hospital Concentration Measure and Hospital Negotiated Price Index
To calculate hospital market concentration, each hospital's market is defined as the area within a 30-minute travel time from the hospital, which is measured using data from HERE, a mapping software application. A Herfindahl-Hirschman Index (HHI), which measures market concentration, is calculated for each hospital's market using the sum of hospital beds in the market as the size of the market and a hospital's count of beds as its measure of market share. We average the hospital-level HHIs to the HHR-level. HHIs range from 0 to 10,000 with a HHI of zero representing a 'perfectly competitive' market of many small firms and a HHI of 10,000 representing a monopoly market. To calculate the hospital negotiated price index, private insurance claims for all inpatient care provided to covered 18-to-64-year-olds in American Hospital Association (AHA)-registered facilities from 2017 are retained from the HCCI database. Data are then limited to general medical/surgical hospitals with at least 50 cases. Hospital payments are regressed on hospital fixed effects, a vector of patient characteristics, and a vector of patient diagnoses (DRG) fixed effects. The vector of hospital fixed effects is recovered, and the hospital price index is calculated using the sample means of the patient characteristics and the DRG indicators. Further details on the price index can be found in a previous study. 1

Linking Data
A county-to-HRR crosswalk maintained by the United States (US) Census (MABLE -14) is used to convert county-level data (where necessary) to HRR-level data. 2 For counties not available in this crosswalk, a county-to-zip code crosswalk maintained by the US Department of Housing and Urban Development and a zip-to-HRR crosswalk maintained by the Dartmouth Atlas are used. 3,4

Selection of Correlates
A list of the chosen correlates, along with their sources, is given below.

Sensitivity Analyses
We conducted robustness checks of our spending and quantity correlations between Medicare and the privately insured in areas where the Medicaid data are low quality and across the country including those areas. In areas where the Medicaid data are low quality, the correlation between HRR-level Medicare and private insurance spending is 0.052 (p = 0.69), and the correlation between HRR-level Medicare and private insurance inpatient bed-days is 0.476 (p < 0.01). Across the country, including areas where the Medicaid data are low quality, the correlation between HRR-level Medicare and private insurance spending is 0.034 (p = 0.55), and the correlation between HRR-level Medicare and private insurance inpatient bed days is 0.476 (p < 0.01).
The results of several additional sensitivity analyses are presented in the eTables and eFigures.