Chapter 3: An Introduction to Medicare Data

Introduction and Goals of This Chapter

A Note on the Source and Structure of Our Claims Data

Part B Carrier Claims Data

Durable Medical Equipment (DME) Data

Outpatient Claims Data

Inpatient Claims Data

Skilled Nursing Facility (SNF) Claims Data

Home Health Claims Data

Hospice Claims Data

Commonly Retained Elements in Administrative Claims Data

Master Beneficiary Summary File

Provider Data

Example: Identification of Emergency Department (ED) Utilization

Example: Identification of Surgical Services

Chapter Summary

 

This chapter introduces the contents of the Medicare claims and enrollment files we will use for our research programming project. The purpose of this introduction is to build on the explanation of the Medicare program presented in Chapter 2 and further prepare the programmer for using Medicare administrative data to complete our example research programming project. In Chapter 2, we learned the basics of the administration of the Medicare program. We defined Medicare, discussed enrollment, eligibility, and coverage, and provided a very simple sketch of how Medicare pays for services. We also briefly discussed how this information about Medicare influences the content of the data files used throughout the remainder of this book. In this chapter, we will build on those concepts and dive into each Medicare file we will be using for our example research programming project.1 We will review the contents of each data set, including examples of information commonly pulled out of each file for research purposes. We will end with an example of how to use the datasets to identify services provided for a surgery and services provided for a visit to the emergency department.2 Through these examples, we will see that the contents of each file are not always intuitive. Rather, the contents of each file are governed by how the services are billed and paid. We will learn to be cognizant of certain quirks, like the fact that not all services performed in an inpatient hospital appear in the inpatient claims data set. Looking forward, in Chapter 4, we will build on the information presented in this chapter by planning our programming project and, in Chapter 5, we will request the data described in this chapter. In subsequent chapters, we will actually use these data to address the research questions posed by our example project described in Chapter 1. For example, we will load and transform our claims files in Chapter 5, and calculate utilization of services in Chapter 7.

The base claim detail (also known as header-level information) is basically summary information, including information that identifies the claim, the beneficiary who received care, the provider of service, the beginning and ending dates for the services billed on the claim, the diagnoses on the claim, and the total amount paid to the provider as reimbursement for the services performed. The claim line detail (sometimes generally referred to as line-level information) is a set of in-depth information about the specific services performed. For example, a line will contain a beginning and ending date of services, the patient’s diagnoses (represented by diagnosis codes), the services performed (represented by procedure codes), payment amounts for the services, and the identifier of the provider that performed each service. In a SAS dataset, the claim is spread across multiple records with one record per service rendered.

In Chapter 5, we will transform these separate files to contain all claim- and line-level information in a single record. In other words, we will take the file with multiple records per claim and construct arrays on a single record to represent each of the line-items. This maneuver is not something you would do in your own research programming but, as we will discuss, the advantage is that we can work with a structure of claims data sets that can be created using data from any of the three sources. As such, this transformation “levels the playing field” and makes the remainder of the book accessible to a broader audience. In addition, such a structure just happens to be my preferred method of working with claims data because it affords the opportunity to see the entire paid claim in one record! Because such a transformation can consume valuable resources, more advanced readers should determine for themselves the most efficient way to structure their claims data.

Carrier claims data contain claims information filed by non-institutional providers using the CMS-1500 claim form. When a Medicare beneficiary goes to the doctor for a checkup, the provider electronically submits a bill to Medicare Part B using the ANSI 837 5010 electronic format (or, using the CMS-1500 paper form). The CMS-1500 form contains details like the beneficiary and provider identifiers, dates of service, the procedures performed, and the patient’s medical diagnosis. The submitted claim goes to the provider’s regional Medicare Administrative Contractor (MAC) for adjudication, processing, and payment. The final action Medicare claims are then made available to the research community as the carrier claims data.

In my experience, users most readily associate this file with claims filed by individual physicians like internal medicine specialists or cardiologists. In other words, most users associate the carrier data set with claims filed for services provided in a doctor’s office, like a checkup. However, the carrier data set also includes claims submitted by providers that you may not readily consider, like clinical social workers, chiropractors, ambulances, nurse practitioners, and physician assistants. In addition, the file also includes claims for services performed outside of a doctor’s office, like ambulatory surgical centers, hospital outpatient departments, and hospital emergency departments.

The DME file contains information on final action claims submitted by non-institutional DME providers using the CMS-1500 claim form.6 These claims are processed by the provider’s Medicare Administrative Contractor (MAC) for adjudication and payment. The final action Medicare claims are then made available to the research community as the DME claims data. Durable medical equipment includes wheelchairs and walkers, hospital beds, blood glucose monitors and related supplies, canes and crutches, splints, prosthetics, orthotics, respiratory devices like oxygen equipment and related supplies, and dialysis equipment and supplies.

DME is provided to any beneficiary with Part B insurance as long as it is medically necessary. Beneficiaries with Medicare Part B must have the equipment prescribed for use in their home by their doctor or “treating practitioner” (e.g., a physician assistant or nurse practitioner). A long term care facility can qualify as a beneficiary’s home, but equipment is not covered if it is used in a hospital or a skilled nursing facility. For example, the hospital bed in an inpatient facility is included in the facility charge found in the Inpatient claims data.

Outpatient claims data contain information on final action claims filed by institutional outpatient providers.7 Most commonly users think of these providers as hospital outpatient departments. However, the data also includes the claims of other types of institutional outpatient providers like ambulatory care surgical centers, outpatient rehabilitation facilities, rural health clinics, and even community mental health centers.

Medicare Part B pays for many of the outpatient services a beneficiary receives in a hospital like the hospital charge for an emergency department service (as we will see below this does not include the doctor’s charge), getting stitches or a cast, lab and X-ray services, outpatient surgery, observation required to determine if a beneficiary should be admitted for inpatient care, and even the administration of drugs if a beneficiary cannot self-administer the drug. For this reason, the claims in the outpatient data are sometimes referred to as “institutional Part B” claims. Looking forward to Chapter 8, we will see this in action when we do not use the outpatient data set when we sum total Part A costs.

A beneficiary is considered an outpatient if the beneficiary’s doctor has not written the beneficiary an order to be admitted as an inpatient. These services are covered under Medicare Part B and are paid using the Outpatient Prospective Payment System (OPPS). The OPPS pays hospitals a predetermined payment rate (the rate differs by geographic region) to provide these services to Medicare beneficiaries.

Inpatient claims data contain information on final action claims submitted by long-stay and short-stay inpatient hospitals for the reimbursement of their facility costs.8 These claims are paid through Medicare Part A. Hospitals file claims for these services using the UB-92 form, also known as the CMS-1450 claim, or the electronic 837i format. It is important to make the distinction between facility costs and other costs for services provided in the hospital (again, more on this in the example below). Facility costs include things like room charges and even some drugs provided during a beneficiary’s hospital stay. As we will discuss below, a doctor’s visit or even a surgeon’s services provided in the facility are billed using the CMS-1500 form and are therefore found in the carrier data.

A beneficiary is considered an inpatient starting the day the beneficiary’s doctor formally admits the beneficiary to the hospital. The distinction between inpatient and outpatient status is very important because it influences how Medicare pays for the services received by the beneficiary. For example, as noted in Chapter 2, a beneficiary cannot be admitted to a skilled nursing facility unless the beneficiary has been discharged from a hospital within the last 30 days with an inpatient stay that lasted at least three consecutive days.

SNF claims data contain information on final action claims filed by SNF providers.9 Like inpatient claims, SNF claims are billed using the CMS-1450 form or the 837i electronic format and paid through Medicare Part A. Skilled nursing facilities may be part of a nursing home or a hospital and provide skilled nursing and rehabilitative care through specialists such as registered nurses, physical therapists, occupational therapists, speech pathologists, and audiologists.

The purpose of this care is to treat, observe, and manage the conditions of beneficiaries leaving the hospital. The goal of skilled nursing care is to improve or maintain the beneficiary’s current condition, as well as assist beneficiaries in maintaining their independence. Examples of skilled care include physical therapy and intravenous injections. Once admitted to a SNF (see the above section for more information on rules governing admission to a SNF), Medicare pays for SNF services for up to 100 days per spell of illness under specific circumstances. A spell of illness ends 60 days after discharge from a SNF, at which time a patient can be admitted to the hospital and then to a SNF for another 100 days.

SNFs are also paid on a perspective payment basis - resource use groups (RUGs). Each of the 66 RUGs has associated nursing and therapy weights that are applied to create daily base payment rates.

Home health claims data contain information on final action claims filed by Home Health Agency (HHA) providers.10 Home health services are paid from Medicare Part A and Part B. Home health agencies are entities that provide skilled professional care in a beneficiary’s home.

These agencies provide services to Medicare beneficiaries like occupational therapy, physical therapy, skilled nursing care, and even speech therapy and services like counseling that can help a beneficiary cope with the impacts of their illness on their mental health.

Medicare pays for home health services in units of 60-day episodes based on a prospective payment predetermined rate. Patients receiving five or more visits are assigned to 1 of 153 home health resource groups (HHRGs) based on clinical and functional status and service use which are adjusted for the geographical area where services are delivered.

The hospice claims data contain information on final action claims data submitted for hospice (also referred to as “end of life” or palliative care) services.11 Hospice programs provide care for Medicare beneficiaries who are terminally ill. The focus of hospice care is on the comfort of the patient and not on healing or curing an illness.

A beneficiary qualifies for hospice services if the beneficiary has Medicare Part A and is terminally ill. A beneficiary is considered terminally ill if the beneficiary’s doctor certifies that the beneficiary has less than six months to live, should the beneficiary’s illness follow a normal progression.

Hospice services include physical care provided by doctors and nurses, care provided by hospice aides, occupational therapy, physical therapy, medical equipment and supplies, drugs, and even grief counseling and respite care (care provided in a facility designed to give family members a break from giving care) for the beneficiary’s family members. All hospice services are covered as long as they are related to the beneficiary’s terminal illness (services not related to the terminal illness are covered by other Medicare benefits). Many of these services may involve pain management, and services may take place in a facility or the patient’s home.

Hospice care is insured in benefit periods, meaning that a Medicare beneficiary can get hospice care for two 90-day benefit periods and then an unlimited number of 60-day benefit periods. The doctor must certify that the beneficiary is terminally ill at the beginning of each period. Medicare beneficiaries have the right to stop hospice care at any time for any reason, and the right to change hospice providers once each benefit period.

With about 47 million beneficiaries to account for, Medicare administrative claims files can be very large. Generally, the carrier claim file is the largest Medicare claims dataset used for research purposes. The large size makes sense because the file contains claims for the most common type of service, a visit to the doctor!

One very common way of easing the use of large datasets is to limit the files to contain only the key variables needed for your project. In Chapter 4, we will start the process for planning our project at which time we will begin to form a good idea of the variables we will need for the successful completion of our sample research programming project. At a minimum, when performing research, it is standard practice to keep the following variables in the administrative claims files:12

• The Medicare beneficiary’s identifier:13 Commonly referred to as the HICAN (usually pronounced “high-can”) or HIC, this variable may not exist as its own data element in the Medicare research files we will receive. Although the data we will work with contains a ready-made, scrambled beneficiary identifier called a BENE_ID, you may have to create the HIC by concatenating the beneficiary’s Claim Account Number (CAN) and the beneficiary’s identification code (also known as the BIC). A beneficiary’s identifier can change over time if a beneficiary’s status changes (e.g., a beneficiary remarries). Therefore, it is important to have a master beneficiary identifier when looking at administrative claims and enrollment data over time.

• The provider identifier: A Medicare provider performed the services appearing on a claim, and each provider has a unique identifier that is reported on the claim record. For example, the National Provider Identifier (NPI) can be used to identify the provider that performed the services billed on the claim (known as the “performing” or “rendering” provider). The NPI is a 10-character identification number uniquely assigned to a Medicare provider. This identifier must be used by the provider for all financial transactions with Medicare. The NPI is known as an intelligence-free identification number, meaning that it does not contain any information about the provider, like specialty or Medicare region. We will see that the provider identifier for hospitals found in inpatient claims data does contain embedded intelligence.

• Codes identifying services: The services performed by the Medicare provider can be identified using procedure codes represented by Healthcare Common Procedure Coding System (HCPCS) codes. HCPCS Level I codes, also known as Current Procedural Terminology (CPT) codes, are five digit character codes describing medical services, and are a copyrighted coding schema of the American Medical Association (AMA). HCPCS Level II codes are five character alphanumeric codes defining services not described by the Level I codes.14 In other cases, revenue center codes and ICD-9-CM procedure codes (more on this below) can be used to identify services Revenue center codes correspond to cost center units (or divisions) within a hospital, like emergency departments.

• Codes identifying the beneficiary’s medical condition: The International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) is the United States’ method of assigning medical diagnoses to patients (like diabetes, COPD, or prostate cancer). As of the writing of this text, in October 2014, ICD-9-CM will be replaced by ICD-10-CM.15 ICD-9-CM can also be used to identify procedures in inpatient, skilled nursing facility, and outpatient claims.

• Dates of service: This refers to the dates of service of the procedures performed by the provider. This can include specific dates of service for specific procedures (from detail service lines), or admission and discharge dates (header information) from a stay in an institution (like a hospital).

• Payment information: The total cost of a claim includes the amount paid by Medicare, as well as deductible and coinsurance payments. We are interested in the total cost to Medicare for the services on a claim, so we will utilize only the Medicare payment variables (without using information such as deductibles and coinsurance payments).

This list is by no means exhaustive. In some cases, investigators need to look at payment information like deductibles and coinsurance, provider identifiers like Tax Identification Numbers or CMS Certification Numbers, and information on the provider’s medical specialty. In your work, you may need to utilize information in inpatient claims data such as the source of admission to the hospital, the discharge status or discharge destination, and a Diagnosis Present on Admission (POA) indicator (a field that identifies whether a diagnosis was already present upon admission to the hospital). In addition, some investigators may wish to review the Medicare Severity-Diagnosis Related Group (MS-DRG), a patient classification system comprised of a set of codes developed for Medicare. MS-DRGs use information from a beneficiary’s paid hospital claims (like age and medical diagnosis) to classify hospital services provided to patients. Finally, you may wish to review the level of hospice care provided (e.g., routine home care, inpatient respite care, and continuous home care).

As you can imagine, even this small set of data elements can be used to paint a very detailed story about the nature of a particular claim, like a visit to the doctor for a checkup. When combined with other final action claims, this information can be used to draw broader conclusions about a particular beneficiary or provider.

We have reviewed seven files containing paid claims data, but our project also requires information about the beneficiaries we are studying. When I started using Medicare claims data, the most common source of data for beneficiary demographic information was the Denominator file. The name “Denominator file” may seem odd, but it is appropriate. In many applications of beneficiary demographic information, a calculated number (say, the number of visits to a doctor’s office) is divided by a count of beneficiaries (perhaps a sample population). For example, if we calculated the total utilization of emergency department (ED) services, we may wish to get the per capita utilization rate by dividing by the total number of beneficiaries in our sample. In this equation for per capita utilization of ED services, the count of sample beneficiaries is the denominator! Information from the demographic file is often used as the source of denominator in these kinds of equations. Therefore, the file containing this beneficiary information for the full Medicare population for a given year is named the Denominator file.

In March 2010, my beloved Denominator file was integrated into something called the Master Beneficiary Summary File (MBSF).16 The MBSF contains demographic information for beneficiaries who were alive and eligible for Medicare at any time during the calendar year for which the file was created. This information includes the beneficiary’s identifier, sex, date of birth, and date of death (if applicable). In addition, it includes geographic information like the beneficiary’s state and county codes. Perhaps most importantly, it includes information pertaining to the beneficiary’s enrollment in the Medicare program, like data elements that can be used to determine the beneficiary’s enrollment status in Medicare Parts A, B, C, and D.

For our example research programming project, we assume that we have been given a list of the providers in our study population (along with their associated beneficiaries), and we do not require information on those providers above and beyond what is contained in the paid claims data. For example, the Part B carrier file contains information such as a performing provider’s NPI, provider type (e.g., group, solo practitioner, independent lab, and institutional provider), specialty, state, zip code, and specialty code.

CMS also provides descriptive data on providers through the National Plan and Provider Enumeration System (NPPES, pronounced “in-pez”) file. The NPPES file is a publicly-available dataset that can be downloaded from CMS’s website. The file contains data elements like the provider’s NPI, information on whether the provider is an individual or an organization (like an ambulatory care center), the Health Care Provider Taxonomy Code (HPTC, which is, in very simple terms, a code that describes the provider’s specialty), name, business name and address, contact information, and license number.17

Let’s say you are on a game show called Medicare Millionaire, where you have to answer research questions using Medicare claims and enrollment files. Personally, I would tune in faithfully every night, but I have a feeling that I may be one of the only viewers (after all, it is a small group of people who do this work)! Imagine that you have successfully answered several questions and now have a chance to win a large cash prize based on your answer to the following question: Which of the claims data files are used for the identification of ED utilization? If you already used your lifeline to call a friend, you would need to take your best guess. What would that guess be? Without any knowledge other than what we have learned in this chapter, my guess would be the inpatient claims data. Considering that emergency rooms are typically found in inpatient hospitals, this would be a good choice! Unfortunately, the answer is only partially true!

Complete information on emergency department-related information can be found in no less than three sets of claims data: inpatient data, outpatient data, and Part B carrier data!18 Although we will write code to calculate ED utilization for our example research programming project, discussing how to identify ED utilization at this time is enlightening. There are several factors that make identification of ED services complex. First, there is the issue of whether or not the beneficiary was admitted to the facility for an inpatient stay. Second, there is the issue of identifying what is often referred to as the “professional component” of the service (described below).

If a beneficiary is admitted to the hospital following the ED visit, the information pertaining to the ED visit can be found in the inpatient data set. These ED services are identified using revenue center codes 0450-0459 and 0981. The inpatient claims data contain diagnostic information gleaned from the emergency visit. On the other hand, if the beneficiary is seen in an ED and not subsequently admitted to the hospital, the claim for the ED visit can be found in the outpatient data! These claims can be identified using the same revenue center codes used for identification of ED visits in the inpatient data.

The professional component of a service can simply mean the portion of the service provided by a physician. For example, when a lab test is performed, the “professional component” is a provider’s (typically a pathologist or other physician) interpretation of the test and the “technical component” is the administration of the test itself. In the case of ED services, the professional component will appear in the Part B carrier data because it is billed using the CMS-1500 claim form and paid through the beneficiary’s Medicare Part B coverage.

Again, we will construct SAS code to identify ED utilization in Chapter 7. For the time being, this example illustrates the complexity of using Medicare data to identify utilization of some common services. Just in case you are not convinced of the potential pitfalls inherent in using Medicare data for research purposes, let’s discuss another example, the identification of surgical services.

Let’s say that you made it through to the bonus round of Medicare Millionaire. Maybe you are starting to sweat under the hot lights as the host lines up a really hard question. After a short commercial break to build the anticipation, the host throws a real hardball question at you: “How do you identify surgical services in Medicare claims?” Wow! Hopefully, you smile and answer the question knowingly because the previous night you brushed up for the game show by reading the following information in this chapter!

The identification of surgical services in Medicare claims data requires the use of inpatient, outpatient, and Part B carrier data sets. If you are interested in post-acute care, you can even add the home health, hospice, and SNF data sets to the list! Let’s start with tracking down all of the information related to the surgery by looking in the most intuitive location first: the inpatient data set. After all, when most folks think of surgery, they picture an operating room in an inpatient hospital setting. Surgeries performed in an inpatient setting are typically part of an inpatient stay and found in the inpatient claims file. That said, surgeries are commonly performed in the outpatient department of hospitals and these claims appear in the outpatient data set. Finally, surgeries can also be performed at Ambulatory Surgical Centers as well, and these services appear in the outpatient and Part B carrier data sets. The facility charges for the surgeries provided in any of these settings can be found in the data mentioned above. However, the story doesn’t end there! There is also the matter of the surgeon’s professional services and, lest we forget (or assume the beneficiary was asked to bite down on a stick), the anesthesiologist’s services. These services appear in the Part B carrier data set.

This chapter provided an introduction to the Medicare claims and enrollment data commonly used in research programming efforts. We learned:

• Data exist for institutional and non-institutional claims paid by Medicare Part A and Part B.

• There are seven types of paid claims (not including Part D information): Part B carrier, DME, outpatient, inpatient, SNF, home health, and hospice.

• For instructional purposes, we will request our claims and enrollment data from CMS’s data distribution contractor. Non-institutional data arrive in separate files for Part B carrier base claim, Part B carrier claim line, DME base claim, and DME claim line detail. Institutional data arrive in separate files for base claim and revenue center information (claim line detail) for each claim type. In Chapter 5, we combine these base claim and claim line (or revenue center) information into single records by claim type.

• At a minimum, the following data elements are commonly required for research programming with claims data:

∘ Information identifying the beneficiary identifier, such as a BENE_ID or HIC.

∘ Information identifying the provider of the service, such as the NPI.

∘ Information describing when the services occurred (start and end dates, or admission and discharge dates).

∘ Diagnostic codes describing the medical condition of the patient, and procedure codes or revenue center codes describing services provided to a beneficiary.

∘ Information on the amount paid by Medicare for the services on the claim.

• Enrollment data exist for Medicare beneficiaries. Medicare beneficiary enrollment data include information on beneficiary date of birth, date of death, residence, and reason for entitlement and FFS enrollment status.

• Using Medicare claims is not intuitive. Identifying all components of ED or surgical services involves the use of multiple claims files.

---