Researchers often need data for Veterans who received medical care from a commercial hospital. A common example is a cost-effectiveness analysis, which requires information on the costs of all healthcare utilization, including care obtained from non-VA providers.
There are three general methods for estimating these costs:
- VA databases on contract care
- non-VA databases
- patient self-reports
Each of these methods is discussed below. One caveat, however, is that the analyst should take care not to double-count non-VA care reported in different sources.
Non-VA Medical Care files report financial and limited clinical information on non-VA care that was paid by a VA facility. These files include care provided to Veterans at military facilities, primarily in Alaska and Hawaii. The files exclude most care provided through sharing agreements at university affiliates and services at state Veterans homes.
There is some overlap between Non-VA Care files and standard VA utilization data files. Community nursing home stays appear in both the VA extended care PTF and the MCA outpatient national data extract. A record is created in the PTF Non-VA Hospitalization file when authorization is given for inpatient non-VA medical care. Once the stay is completed and claims from the provider are submitted and paid, the record will be updated as needed to conform as closely as possible to standard PTF format. A record for the same stay will appear in the Non-VA Medical Care data only after a claim is paid. For this and other reasons, overlap between the PTF NVH and Non-VA Medical Care inpatient files is only moderate. The HERC technical report features an analysis of the overlap.
While VA data sets apply for the fiscal year when health care services are rendered, Fee Basis data sets are named for when services are paid (i.e., MED13 indicates outpatient encounters paid in FY13). This is a very important distinction. Furthermore, in Fee Basis, each record is not an encounter, but a CPT payment.
VA data, including HERC data, are currently in transition to the VA Corporate Data Warehouse (CDW), a national VA data repository. The National Data Systems (NDS) oversees the request and approval process through the Data Access Request Tracker (DART) system. For more information on data available at CDW and the current request procedures, please see the VHA Data Portal (Intranet-only: http://vaww.vhadataportal.med.va.gov).
VA Directive 1663 allows VA medical centers to contract for clinical and related services with affiliated medical schools, faculty groups, hospitals, and other providers. These contracts, also called sharing agreements, are developed under several circumstances: when a clinical service cannot be provided by one VA facility and the patient cannot be transferred to another VA facility; when VA cannot recruit a needed clinician; when it is determined not to be in the best interest of the VA to hire, such as when only a portion of a clinician’s time is needed; in order to reach market-rate pay for certain highly paid subspecialists; and when it is cost-effective to share a service or space with another entity rather than to develop stand-alone capacity for VA.
Sharing agreements vary considerably in scope and purpose. Most cover specialty services such as anesthesiology, cardiology, neurosurgery, ophthalmology, orthopedic surgery, or radiology. A few concern non-clinical services such as space rental, animal care and use, prescription drug storage, laundry, antenna leasing, or athletic facilities. Sharing agreements can be based on full-time-equivalent (FTE) employment, on specific procedures, or on relative value units (RVUs).
Sharing agreements represent an alternative to the Non-VA Medical Care program. Non-VA medical care is provided on a fee-for-service system and is subject to several coverage and payment limitations. Some common forms of contract care, such as care at community nursing homes and state veterans homes, kidney dialysis at community facilities, and long-term oxygen therapy in the patient’s home, are handled exclusively through the Non-VA Medical Care system. Sharing agreements are longer-term contracts with non-VA providers to offer selected types of care, usually to many patients. Unlike much non-VA medical care care they do not center on an individual patient and are not subject to similar coverage limitations.
Sharing agreements are developed through a multistep analysis of patient need, medical center capability, costs, and alternative options. Services are valued based on local and regional market analyses. Payments may exceed Medicare rates depending on prevailing market rates. See VA Directive 1663 for complete details on the contracting process. To find it visit https://www.va.gov/vapubs.
About half of all VA medical centers have sharing agreements, typically 2-4 per year. The total estimated value of the contracts exceeded $64,000,000 in FY2008. An annual tabulation of all contracts approved, in process, rejected, or withdrawn is maintained by the Medical Sharing branch of the Procurement and Logistics office in Washington.
Most care provided through sharing agreements does not appear in VA utilization databases. One exception is services purchased from the Department of Defense (DoD), which are recorded in the Non-VA Medical Care database. The data capture system is imperfect at present, although this may be remedied by a VISTA patch currently being developed. Another exception would be care provided at a VA facility by a contract provider. In such cases the provider uses VISTA and the encounter is recorded as usual. There is no plan to record sharing care from other sources in Non-VA Medical Care files or other utilization databases available to researchers.
A regulation change in FY2007 made it difficult for VA stations to enter into sharing agreements. Their number fell precipitously as local officials moved to purchase contract care through the Non-VA Medical Care system instead. These events may have caused an anomalous increase in non-VA medical care. Due to the considerably larger scope of Non-VA Medical Care contracts, however – more than $3 billion in FY2008 – the increase due to the FY2007 regulation change may not be noticeable. Changes in the contracting process since FY2007 have led to a substantial return to the use of sharing agreements throughout the system.
The VA Financial Management System (FMS) records spending by fiscal year, location, budget object code (BOC), and cost center. The data are stored in files at the Austin IT Center. FMS data will be useful for determining summary figures, but they cannot be tied to care for individual patients. Most BOCs pertaining to contract care fall in the range of 2560-2636. Exception include 2674, 2692, and 2693, which contain a mixture of contract and non-contract care, and 4110 and 4120, which include payments for services related to medical care, to cemeteries, and to other veteran benefits. Most contract care falls within cost centers in the 8300 series. Again there are exceptions, such as 8204, 8222, and 8272. The current policies defining BOCs and cost centers may be found by visiting https://www.va.gov/vapubs and searching by keyword.
Sharing agreements with Department of Defense facilities are recorded under a single BOC. Using the vendor files in the Non-VA Medical Care data one could locate all paid claims for sharing care provided by DoD. The resulting figure would be less than actual spending due to incomplete reporting. FMS will not suffer from incomplete reporting, however, and so the difference between FMS and Non-VA Medical Care totals for DoD care should indicate the level of underreporting in the Non-VA Medical Care files.
There are eight Non-VA Medical Care files for each fiscal year. Four record details of individual health care services, one each for inpatient stays, inpatient ancillary services and physician payments, outpatient care, and outpatient pharmacy. There are additional files for travel reimbursement data and for information on Fee Card holders, veterans who receive long-term outpatient care through the Non-VA Medical Care services. The final two files contain information on Non-VA Medical Care providers, known as vendors, one for pharmacy vendors and another for all other vendors.
Three of the clinical files contain basic characteristics about care provided, including procedure and surgery codes. Non-VA Medical Care records provide a wealth of detail on charges and payments for services. The files can be linked by patient ID to standard VA utilization files.
Non-VA Medical Care files are in the Corporate Data Warehouse (CDW) and researchers can access them through the VA Informatics and Computing Infrastructure (VINCI).
There is an alternative method for accessing summaries of Non-VA Medical Care expenditure data, one that does not require an AITC time-share account. A "Non-VA Care Cube" has been created as part of the national Financial and Clinical Data Mart. The data cube shows combined payments from the four central Non-VA Medical Care files: Inpatient, Outpatient, Inpatient Ancillary, and Pharmacy. One can view and download a series of standard reports or create a unique "view". The data cube does not present individual encounter records. Rather, it shows payment totals summed within category (Purpose of Visit) and time period (month or fiscal year). The data cube is accessed through the VISN Support Services Center (VSSC) web site. The site is accessible only through the VA intranet.
When Veterans use non-VA health services, much of this can be obtained from existing databases (namely Medicare and Medicaid). More details on these datasets are below.
There are other databases, but those that are publicly available do not include patient identifiers. A good example is HCUPnet, which provides statistics from the Healthcare Cost and Utilization Project. Among other non-VA files available to researchers are Medicare Cost Reports, American Hospital Association Survey, ZIP Code Files, and state hospital discharge reports. If you are only interested in obtaining charges for a certain inpatient procedure or diagnosis, a good resource is the Healthcare Utilization and Cost Project (HCUP) webpage. HCUP is one of the databases developed by the Agency for Healthcare Quality and Research. They have an interactive web page that allows one to enter criteria and see average facility charges; note that HCUP does not include physician payments. HCUP reports cost-adjusted charges; see Adjusting Charges from Private Hospital Data.
Researchers sometimes purchase claims data through companies, such as Optum and Truven (now IBM). Those data are not available in VA.
The Centers for Medicare & Medicaid Services (CMS) is responsible for administering Medicare. It also works in with states to administer Medicaid (in California Medicaid is referred to as Medi-Cal).
If you are a VA researcher interested in obtaining Medicare data for Veterans, please contact VIREC.
As the administering agency, CMS makes some utilization datasets available for researchers. These datasets are note specific to Veterans and they come in two flavors-Public Use Files (PUFs) and Beneficiary Encrypted Files (BEFs). As it suggests, the PUFs are at an aggregated level and have no information that can be used to link files to patients. On the other hand, BEFs provide encounter level data, but the social security number is scrambled. If you need to identify the Medicare beneficiaries, then the researcher must obtain special permission to access the research identifiable files (RIFs). Obtaining a RIF is very complicated and HCFA will only release RIFs to certain researchers for specific projects; obtaining a BEF is more complicated than getting a PUF.
The PUFs are free (see the CMS web site), RIFs and BEFs cost money and require human subjects approval to use the data.
There are many inpatient and outpatient files available for Medicare and Medicaid. For more information, contact the Research Data Assistance Center (ResDAC). They have a CMS contract to provide technical assistance. They can answer questions about access to data. They also provide limited technical assistance to help researchers identify questions that can be answered with the data.
Often we want to analyze charge data from utilization databases, like the Healthcare Cost and Utilization Project (HCUP). Yet, it is generally known that in health and medicine, charges rarely equal costs. In most cases, charges are greater than costs. However, the degree to which charges exceed costs is not completely random. Hospitals and medical centers are somewhat idiosyncratic in how they generate bills.
Hence, we want to adjust the charges for two reasons:
- to deflate charges so that they more closely reflect costs, and
- to remove hospital specific idiosyncracies.
The cost to charge ratio (CCR) is one way of making this adjustment. You can compute the CCR manually using the Medicare cost reports. Luckily, there is an easier way. CMS releases an Impact file each year. The Impact file is a hospital level dataset that provides a lot of information, including the CCR.
Different departments or units within a hospital will often have different cost to charge ratios. We strongly recommend not to use department-specific CCRs; this is subject to considerable interpretation and error within a hospital.
The cost of providing medical/surgical care in a hospital includes both a facility and a physician component. This answer explains how to estimate the hospital component.
Payments are based on the Diagnosis Related Group (DRG) and adjusted for each specific hospital. With the DRG and hospital Medicare identifier number, you can determine what Medicare would have paid for the hospital stay.
For an inpatient stay, Medicare calculates (1) hospital operating costs and (2) capital costs. Both 1 and 2 start with a standardized amount; hospital in urban areas with more than 1 million persons face one set of standardized amounts, while all other hospitals face another set of standardized amounts.
The standardized amount is first adjusted to account for labor costs, using the wage index. The standardized amount is then adjusted by the diagnostic related group (DRG) weight to reflect the patient’s case mix. In addition to the adjusted base amount, hospitals receive additional payments to reflect their share of low-income patients (known as disproportionate share) and their indirect medical education.
In addition to receiving Medicare reimbursement for hospital operating costs and capital costs, hospitals can request outlier payments and additional reimbursement based on their direct medical education. The outlier payment is for those patients who are much sicker and who stay much longer than average, and the Centers for Medicare and Medicaid Services have a description of outlier payments on their web site.
With some simplifying assumptions regarding the wage index, disproportionate share and indirect medical education, a researcher can calculate a “national average” reimbursement for a particular DRG. In fact, this would result in two averages: one for large urban hospitals and one for all other hospitals. These averages would not include any outlier payments and they would not reflect direct medical education. Approximately 28% of all US hospitals in FY04 were teaching hospitals. For teaching hospitals, DME can represent an additional 10-20%.
Researchers should also recall that these estimated Medicare payment are for the facility and exclude professional fees (i.e., the physician’s reimbursement).
The Centers for Medicare & Medicaid Services (CMS) distribute a PC program called the Prospective Payment System (PPS) PC PRICER. If you enter the hospital identifier (PPS number), the DRG, the admission and discharge date, and the billed amount (charges), and whether the stay involved a transfer, the program tells the Medicare reimbursement. The program calculates teaching payments, disproportionate share payments, geographic adjustments, and capital payments for that hospital. Since these change each year, there is a different version of the program for each year.
Unless the stay is unusually long (an outlier), the billed amount does not affect the reimbursement. Thus it is possible to estimate a reimbursement without knowing the charges. Also, the length of stay does not usually affect reimbursement, so it is not critical to know the admission or discharge dates either. It is important to indicate if the stay involved a transfer to or from the hospital.
The program requires the hospital identification number. This six-digit identifiers consists of two digits for the state and four digits for the hospital. View the "Impact File" for one source of provider identifiers.
The Pricer software yields the cost of the hospital stay from Medicare's perspective. It may not be the “economic cost,” for example, if the stay is more expensive than expected for that DRG, Medicare pays the flat (DRG based) payment, and the hospital loses money on the stay.
Information on non-VA services use may be obtained directly from the participant. Below are a couple things to keep in mind.
Categories of care
Self-report questions should focus on categories of care for which costs are different and for which patients cognitively keep them separate. For example, it is common to ask about nights of inpatient care and then number of outpatient visits. However, it is probably not a good idea to ask about nights of intensive care unit (ICU) hospitalizations. Patients may recall being in an ICU, but are unlikely to be able to separate these categories accurately.
If the analyst uses the VA utilization databases and uses self-report for non-VA care, it is highly recommended that one set of questions specifically pertain VA care and a second set of questions pertain to non-VA care. This permits validating the accuracy of self-reporting for the study sample.
In almost every case, patients are not reliable sources of cost information because they rarely see the bill(s); bills can be confusing to say the least, and because recall of these data is not accurate. Therefore, the analyst must multiple the self-report data (i.e., units) with estimated unit costs.
Researchers at the Health Economics Resource Center have been involved in a number of clinical trials where there is an interest in understanding how the intervention affects veterans' use of care, including non-VA care. They have developed a standard set of questions on self-reported non-VA utilization. These questions are often tailored or augmented for a particular study.
It is important to keep in mind that self-report involves cognitively complex question. The patient is being asked to go back in time to remember events. The accuracy of the recall can be confounded if the events occurred a long time ago or if the patient used a lot of care. There is also inherent tension between asking precise questions and making the questions more cognitively complex to the point where the respondent is confused and the data are less accurate. For example, a visit to the emergency department (ED) where the patient stayed in the hospital overnight is unambiguously an ED visit but it is unclear whether it was also an inpatient stay. Adding qualifiers, such as "admitted to the hospital," may add precision at the risk of reducing accuracy.