VA Economic Data Sets VI: HERC Average Cost Datasets Todd Wagner, Ph.D. Ciaran Phibbs, Ph.D.
Learning Objectives At the end of this lecture, you will 1.Understand how we created the average cost data 2.Relate the methods to other cost determination methods 3.Know how to access the data 4.Be aware of the HERC data limitations 5.Have information to help choose between HERC or DSS
VA Cost Data In 1990’s, we lacked of accurate encounter- level cost data for economic and health services research HERC’s Goal: Create encounter level cost estimates for all care provided at VA facilities
Costing Methods Direct measurement Pseudo-bill Reduced list costing Clinical cost function Average cost per visit More detailedLess detailed Inpatient Med/Surg Outpatient AC Costs Inpt. Rehab, MH, LTC DSS
HERC Inpatient Costs
6 Average Cost Categories Ten categories of inpatient care 1.Medicine and surgery (med/surg) 2.Rehab 3.Blind rehab 4.Spinal cord injury 5.Psychiatry 6.Substance abuse 7.Intermediate medicine 8.Domiciliary 9.Psychosocial residential rehabilitation 10.Nursing home (risk adjusted FY98-00 only)
7 Step 1: Create Categories Utilization –Patient Treatment Files (PTF) bedsection file –Use bedsection variable to create category Costs –FY98-03 Cost Distribution Report –FY04- Summary of DSS data
8 Step 2: Merge and Reconcile Utilization and Costs Utilization and cost data come from different datasets –Sometimes there are costs for a bedsection with no utilization (must fix). –Sometimes there are utilization data but no costs (must fix).
9 After Reconciliation A few bedsections stays were reclassified Some costs were moved into other categories Most common fix: we reclassified substance abuse cost or days as psychiatry
10 Step 3: Calculate Daily Rate
11 For each category of care, we estimated –Local cost per day –National cost per day Flagged local costs where they differ >= 2 standard deviations from national costs COSTL= length of stay*local cost per day COSTN= length of stay*national cost per day Two Cost Estimates
12 Step 4: Med/surg Hospitalizations Use average daily rate? –Unpalatable assumption that the daily cost does not vary by diagnosis
13 Med/surg Hospitalizations Additional data can help us estimate costs –Length of stay –Diagnosis related group (DRG) Based on entry diagnosis, patient is assigned to one of 511 groups We use CMS relative value weights for DRG General approach was to use a cost regression
14 Cost Regression Med/surg hospitalizations We made a statistical model to estimate cost –Step 1: Build a model with inpatient discharge data (Medicare) –Dependent variable is cost adjusted charges (CAC) CAC i = 1 length of stay i + 2 DRG i + 3 icudays i + 4 age i +…+e i
15 Cost Regression: Med/surg hospitalizations Step 2: From the regression model, save the parameter estimates ( ’s) Step 3: With our new function, plug in VA data to estimate costs CAC i = 1 length of stay i + 2 DRG i + 3 icudays i + 4 age i +…+e i 1 length of stay i + 2 DRG i + 3 icudays i + 4 age i =Est. VA costs Model from previous page
Regression Model Used OLS with cost adjusted Medicare charges as the dependent variable Absolute value of difference between predicted cost and Medicare cost-adjusted charges was our criterion Also tried –GLM –Semi-log –HCUP data
Variance Attenuation
Implausible costs A handful of stays had costs <$0 Other stays had costs <$20 Facts or Statistical artifacts?
HERC Inpatient Discharge Data Rehab, MH/SA, and LTC Med/surg. Discharge Dataset
HERC Inpatient Datasets Discharge dataset can be merged to PTF main file Includes med/surg, rehab, MH/SA and LTC Exclude cases admitted before 10/1/97 FY file only includes discharges in the fiscal year Subtotals for each category of care
21 When Not to Use the HERC AC datasets When you need to distinguish between two procedures for the same diagnosis When inpatient costs will depend substantially on factors beyond DRG and LOS When you evaluate an intervention that compares close substitutes. –Example: CABG with radial artery vs. saphenous vein
Variation in Physician Time The method assumes physician time is proportional to the institutional costs. If an intervention will affect labor inputs, you will probably need to use other cost estimates (possibly DSS).
HERC Outpatient Average Cost Data
VA Outpatient Care in FY million visits million CPT codes used –top 5 used 23 million times 10,593 different codes used
HERC Outpatient Care Dataset Relatively easy file to use Merges directly to OPC/NPCD files Local cost and national cost estimates Estimated Medicare payment Flag for imputed values
Overview of Methods All VA outpatient costs – from CDR –2004 from DSS Assign Medicare RVU to all CPT codes in SE file. This includes Medicare facility payment. Estimate Medicare payment for each encounter Scale Medicare payments to VA costs, by category of care
Current Procedural Terminology (CPT) Codes 5 Digit code Represents physician services –clinic visits –surgery and other procedures Represents Ancillaries, e.g., laboratory, radiology HCPCS is Medicare version, which adds codes for durable medical equipment, etc.
Two-step Process Apply a common set of relative values to all CPT codes used by the VA –Medicare is the primary source of the relative values Second, scale relative values to actual VA costs
Medicare Fee Schedule RBRVS - resource based relative value scale Each CPT code has been assigned a value by Medicare known as the relative value unit (RVU) Payments are determined by the RVU
Medicare Fee Schedule Issues Non Reimbursable Medicare Procedures (GAP Codes) Non-Physician Providers Global Payments Invalid CPT codes
Medicare Facility Fees Facilities can bill Medicare when providing ambulatory care –Emergency rooms, hospital clinics, ambulatory surgery centers Since most VA facilities would be eligible for Medicare facility payments, we included facility payments in the HERC cost estimates –THIS ASSUMPTION HAS A BIG IMPACT
Ambulatory Payment Category Medicare pays a flat facility rate based on the Ambulatory Payment Category (APC) of the CPT code. When facility fee is charged, physician payment is less.
APCs Some services have separate Medicare Fee Schedules that include facility payments (e.g., lab) –No APC payments for these services The APCs for some CPT codes are subject to discounting, mostly surgical procedures –If subject to discounting, 100% if most expensive APC, else discounted 50%
Filling in the Rest of the Gaps California Workmen’s Compensation National survey of provider charges Match to clinically similar codes Remainder, used average for clinic stop Inpatient codes, used clinic stop average APCs, matched to similar services
Fee Sources
Scaling to VA Costs Followed same general method as for inpatient costs Mapped CDR to 11 separate groups of outpatient services (outpatient pharmacy is 12th category, no utilization data) –13 th category for unidentified clinic stops
Outpatient Care Categories Medicine Dialysis Ancillary Rehabilitation Diagnostic Prosthetics Surgery Psychiatry Substance Abuse Dental Adult Day Health
38 When to use HERC Outpatient AC dataset AC outpatient dataset is based on CPT codes Can use HERC outpatient AC data as long as study results are not sensitive to aggregation at the CPT code level
What Do We Miss Using Only the OPC Data? Outpatient pharmacy Possibly some lab data Other items? HERC is performing validation tasks over time –look for reports on the HERC web site
40 Don’t Use Data if… Your study has systematic differences within CPT code resource use –Example, add a brief smoking counseling component to physician visit Your study focuses on CPT codes for which HERC imputed values. Local cost don’t seem correct. There could be errors in the CDR allocations that HERC relied on prior to 2004.
Excluded Costs Inpatient and Outpatient Research and education (through HSR&D) Capital financing Malpractice expenses Contract provider costs Community nursing home services Headquarters Prosthetics
Access to HERC Data Form on HERC’s web page; fax a completed form to us. We provide the AAC authorization codes. Each user needs to register. Long-term issues –Publications with the HERC data –Unfunded projects –Non-VA funded applications/projects
43 Data at Austin Datasets for each fiscal year: –Discharge –Non-med/surg hospitalizations –Med/surg hospitalizations –Outpatient –Person-level (annualized roll up)
AC Datasets: FY04 Data CDR ended in FY04; we used summary of DSS NDE data We are making the AC Datasets more like the DSS datasets to facilitate comparisons Inpatient FY04 data and documentation released Outpatient FY04 coming soon
Questions on the HERC Average Cost data?
DSS or HERC Costs? “A person with one watch knows the time, a person with two is never sure. ” -Segal’s Law [adaptation]
Med/Surg Differences HERC costs tend to be larger than DSS Cost of shorter stays tend to be larger in HERC than DSS HERC costs tend to be larger than DSS for the very complex procedures (large DRG weight).
Rehab, MH and LTC costs Large site differences in psychiatry between HERC and DSS costs. LOS perfectly determines costs of HERC. LOS is a very strong predictor of DSS costs, but “other things” matter
HERC Technical Reports Yu and Berger. Comparison Between DSS National Data Extracts and HERC Average Costs: Aggregate and Person-Level Costs, FY2001 Wagner and Velez. A Comparison for Inpatient Costs from the HERC and DSS National Data Extract Datasets
Recommendations Consider the cost data you need Consider your audience Choose a primary database (HERC or DSS) Use the other for sensitivity analyses Do not mix HERC and DSS data (except HERC and DSS pharmacy)
When in Doubt No problem is so formidable that you can’t just walk away from it. - Shultz’s Law
Useful Links Medicare Public Use Files Physician Fee Schedule Payment Amount File National/Carrier National Physician Fee Schedule Relative Value File DRGs Relative Weight File Clinical Diagnostic Laboratory Fee Schedule
Links (Cont.) Hospital Outpatient Prospective Payment System: Ingenix (for GAP codes): VIREC: PBM: