Appendix F
Harvard Market Variables Memorandum
HARVARD PRICE INDEX MEMORANDUM (12.21.11)
Background
Hospital referral region (HRR)-level price indices will be used for three purposes:
1. To adjust imputed values for capitated claims based on market-level differences in price; and
2. To examine price variation across markets; and
3. To use as an alternative method creating output price adjustment.
Creating the National Standard Price
The national standard price for each procedure code and type of claim (diagnosis-related group or current procedural terminology) is calculated as the national mean payment for each procedure. In order to calculate the national mean payment per procedure, claim-day records were calculated by summing payments across all records with the same Enrollee ID, service date and procedure code. Capitated claims and non-capitated claims with zero-dollar spending are excluded when calculating the national standard price.
This standard (national mean) price is then applied back to each claim-day with the same procedure code. For example, we compute the mean spending per person per day for a specific code for computed tomography
(CT) scan. We then apply that price to each person day for the same CT code (conditional on being greater than zero for that person on that day) regardless of where they live or what amount was actually paid.
Creating a Market Basket
Harvard has proposed using the following services in the market basket:
• Top 100 DRGs, in terms of total non-capitated expenditures in 2007
• Top 100 outpatient CPTs (see below), in terms of total noncapitated expenditures in 2007
• Top 200 DRGs or CPTs (in terms of total non-capitated expenditure) that are not included in the other two categories in 2007
CPT codes are used in both outpatient and inpatient professional settings. In Medicare there are three components paid for a CPT claim: a work component, a practice expense and a malpractice component. If the service is provided in a facility (which is always the case for inpatient services and may be the case for outpatient services) the practice expense component is reduced and the facility is paid separately (by DRG in the case of inpatient or CPT in the case of outpatient). We cannot identify if and how the professional and technical components are broken out, and therefore we will collapse outpatient claims with the same procedure code, enrollee, and day into 1 “claim-day” observation. This will be our unit of quantity, and it will be used to construct the market baskets and when determining the price index.
The Price Index
Each procedure in the market basket is assigned a weight equal to its proportion of total market basket spending (weights sum to 1).
Within each area, every weight is multiplied by the mean price for that procedure in the area and all procedures are summed. This produces a single value specific to each HRR.
HARVARD MARKET LEVEL VARIABLES MEMORANDUM (10.20.11)
Background
Market level variables are defined at different levels of aggregation (i.e., HRR and county). Competition is inherently defined at the HRR level.
However, the following market variables are available at the county level:
• Percent uninsured: American Community Survey (ACS) (source); available at the county level, only for 2008 and 2009 (could also use Current Population Survey [CPS] at the state level for 2005-2009)
• Commercial health maintenance organization and preferred provider organization penetration as well as Medicaid, traditional medicine (TM), and Medicare Advantage penetration: Interstudy (source)
• Physician workforce composition: area resource file (ARF) (source)
• Malpractice risk: Centers for Medicare & Medicaid Services (source); malpractice geographic cost index
• Population density: ARF (source)
Approach
We will aggregate the county variables to HRR level (which is imperfect), using a crosswalk based on percent of population in the HRR from a given county. We will then regress the fixed effects from the individual level regressions on the market level variables in order to assess relationship between market variables and geographic variation. For hospital service area (HSA) analysis we will use a similar strategy, applying HRR competition to the HSA.
Alternative
We could assign market level variables to individuals based on their county and then include those variables in the individual level regressions. This is not possible for competition measures because they are collinear with fixed effects. We would still need a second HRR level model to relate fixed effects to competition or we would need to use random effects.
Rationale
We prefer the approach suggested because it keeps all market variables together in the analysis and is more straightforward to explain. Sensitivity analysis and descriptive analyses of within HRR variation in county-level variables will reveal whether the county-level variables produce different results.
HARVARD MARKET LEVEL ANALYSIS METHODOLOGY MEMORANDUM (11.21.12)
Background
A previous memorandum (Market Level Variables Memo) detailed the construction of HRR and HSA market-level measures. The resulting files are attached to the Portal and contain estimates at the HRR (or HSA) level for each measure that is analyzed. The following explains Harvard implementation of this file and the market-level analysis.
Empirical Approach
The market-level file was merged by geographic unit to a file containing estimates of spending, quantity, input-price adjusted spending, and quality derived from regressions (i.e., a file similar to the Subcontractor’s Spreadsheet). We then used multiple linear regressions to assess the relationship between various dependent variables and market-level characteristics. Specifically, we regressed a range of market-level measures (outlined in Harvard’s Final Report) against spending, quantity, input-price adjusted spending, and certain quality measures. We employed weights according to population size.