Task 1: Limitations of CMS data: Medicare

It is helpful to understand the contents of the Medicare files. Refer to Course 1, Module 4, Task 1 for more information regarding the types of services which appear in the different claims files.  A useful document to reference is the “Analytic Issues in Using the Medicare Enrollment and Claims Data Linked to NCHS Surveys”, which appears in the Resources section, below.  You are also encouraged to visit the ResDAC website for more information on Medicare data.  See the Resources section for the link to the website.


Info icon Information

Much of the information for this module was based on a NCHS technical paper, Analytic Issues in Using the Medicare Enrollment and Claims Data Linked to NCHS Surveys.  The link is provided in the Resources section.


Key concepts about the limitations of Medicare data

The advantages of Medicare data are that they are population-based, not subject to recall bias, and can be linked to NCHS population health surveys to expand their analytic potential.  However because Medicare data are collected for the purpose of making healthcare payments, and not for research, there are limitations to the data that you should consider when constructing your analytic data file and conducting analyses.

Importance of the Denominator File

You will need the Denominator File for the years of claims data that you wish to examine.  The Denominator File contains basic demographic and enrollment information about each beneficiary entitled to Medicare during each calendar year and is needed to construct an analytic data file, particularly to identify Medicare beneficiaries enrolled in a Medicare Part C plan. Medicare Part C plans are also referred to as Medicare Advantage (MA) and include Health Maintenance Organizations (HMOs), Preferred Provider Organizations (PPOs), Private Fee-for-Service (PFFS) Plans, Special Needs Plans, and Medicare Medical Savings Account Plans. 


information icon Information

Although the Denominator File contains gender, race, and date of birth variables, NCHS recommends that you use gender, race-ethnicity and date of birth provided by NHANES.


Exclusion of claims paid by a source other than Medicare (e.g., Medicare Part C plans)

CMS generally does not receive claims data for Medicare beneficiaries who enroll in Medicare Part C plans (including private fee-for-service plans paid on a capitation basis).  Please note that exceptions to this do exist.  For example, all Hospice claims are processed as Medicare claims regardless of whether the beneficiary is in a Fee for Service (FFS) or a Medicare Part C plan.  During the time covered by the NHANES and CMS linked data, enrollment in Medicare Part C plans fluctuated between 1999 and 2007.

In general, if your study is based on analysis of claims data, you should exclude Medicare Part C enrollees from your beneficiary sample and your analysis should indicate that extrapolations of study results to the Medicare population at large may in fact only be representative of the Medicare FFS population. Variables on the Denominator File enable you to identify, month-by-month, whether a beneficiary was enrolled in a FFS or a Medicare Part C plan.

You may wish to determine the approximate number of respondents in your analyses after participants enrolled in Medicare Part C plans have been excluded.  Course 2 Module 8, Task 1 describes the process of determining your approximate sample size for proposed research projects using tables on the NCHS website.  The link to the table showing percentage of NHANES survey respondents who were enrolled in a Medicare Part C plan by year and survey is provided in the Resources section.  

If you are conducting a health outcome or epidemiologic study (as opposed to health care utilization or cost study) an alternative approach for dealing with Medicare Part C plan enrollees is to include them for the time period prior to entering a Medicare Part C plan and then censor them at the time they enter a Medicare Part C plan.

Services not covered

Although Medicare provides coverage for a wide range of services, there are health care services not covered by Medicare (e.g., long-term care, personal care, and case management).  Other non-covered services change over time, but during the linkage period they include routine physical exams, some cancer screening procedures, and vaccines. 

Medicare data contain little information on prescription drugs prior to 2006, except for infused drugs, some vaccines, and intravenous and some oral chemotherapy. However, beginning in 2006 prescription drug coverage for Medicare beneficiaries became available through the Medicare Part D program. Prescription drug information paid by Medicare for 2006-2007 is available on the Part D Prescription Drug Event (PDE) File. Prescription drug information for data years 1999-2005 includes:

Medicare does not pay for chemotherapeutic agents that are administered exclusively in an oral form (e.g., Tamoxifen) and prior to 2006 most outpatient prescription drugs were not covered by Medicare.

You may find more information on what is not currently covered by Medicare in the Medicare and You Handbook, produced by CMS.  See the Resources section for the link to this material.

Cost sharing

Medicare beneficiaries often have a number of cost sharing requirements (i.e. deductibles and coinsurance).  Although claims are generated for services where beneficiary cost sharing is involved, the Medicare payment amount does not necessarily represent the full cost to the beneficiary for the service. It is not possible to determine whether the beneficiary paid the cost-sharing amount “out-of-pocket” or whether the cost-sharing was paid by a third party, such as a private supplemental policy or the Veterans Administration (VA).

The total payment on the claim refers to the Medicare payment, not all payments by third parties which may have been made for the service.  It is possible that beneficiaries receive non-covered care, resulting in a total payment amount that is $0. However, this does not imply that the service was not received. 

Discrepancies in the coverage period

You should be aware that the Medicare enrollment period for linked NHANES respondents may not overlap the NHANES interview date.  For example, an NHANES respondent who is 70 years old in 1999 at his/her interview and matched at some point to the 1999-2007 Medicare Denominator File will only have Medicare data for the time period 1999 to 2007. This person is likely to have been receiving Medicare services for many years prior to 1999, however the linked claims information for these years is not available.

A similar NHANES respondent, who was 63 years old at his/her interview in 1999 will be unlikely to match to the Medicare Denominator File until reaching the age of 65 – in the year 2001. You need to determine how to address these discrepancies in coverage periods in your analyses.

On Denominator file with no claims data

There may be instances where an NHANES respondent is on the Denominator file but there are no claims data.  It is possible to be enrolled in Medicare but not utilizing Medicare services during the coverage period. In addition, there may be some record keeping inconsistencies because CMS data are collected for administrative, not research purposes.

Medicare entitlement and coverage variables

The Denominator File includes three variables indicating Medicare entitlement: original reason for entitlement, current reason for entitlement, and Medicare status code.

A beneficiary’s original reason for Medicare entitlement is found in the variable ORIG_REASON_FOR_ENTITLEMENT.  This variable is coded by CMS using information provided by the Social Security Administration and/or Railroad Retirement Board. Knowing a beneficiary’s original reason for entitlement can be useful for identifying which aged beneficiaries were formerly Medicare disabled, since their cost and utilization profiles tend differ from other aged beneficiaries, especially at ages 65-74. ORIG_REASON_FOR_ENTITLEMENT values include: Old Age and Survivors Insurance (OASI), Disability Insurance Benefits (DIB) and End Stage Renal Disease (ESRD).

A beneficiary’s current reason for Medicare entitlement is found in the variable CURR_REASON_FOR_ENTITLEMENT.  Possible values include: Old Age and Survivors Insurance (OASI), Disability Insurance Benefits (DIB) and End Stage Renal Disease (ESRD).  This variable is populated from the Medicare Enrollment Data Base (EDB).  The EDB is a master enrollment file of all people ever entitled to Medicare.  Many of the variables on the Denominator file are extracted from the EDB.  The EDB is not available to researchers.

The variable MEDICARE_STATUS_CODE specifies the most recent status of the beneficiary’s entitlement to Medicare benefits. Medicare status code is a CMS coded variable that is created from the following variables available on the EDB: Age, original reason for entitlement, current reason for entitlement, and an indicator of End Stage Renal Disease (ESRD).  Possible values include Aged without ESRD, Aged with ESRD, Disabled without ESRD, Disabled with ESRD, and ESRD only.

Impact of Prospective Payment System (PPS) on Medicare payments

Medicare’s Prospective Payment System (PPS) refers to a method of reimbursement where the Medicare payment is made based upon a predetermined, fixed amount.  An example of this is the use of diagnosis-related groups (DRGs) for payment of inpatient claims.  Medicare uses a separate PPS for different care settings, wherein the particular payment amount is based upon the classification system for that particular service.  Please note that for outpatient and home health agency (HHA) reimbursable claims, the PPS was implemented in July 2000, meaning that claims submitted for reimbursement before this date will be different than after July 2000.

You may find more information on Medicare’s Prospective Payment System (PPS) on the CMS website.  A link to this information may be found in the Resources section.

Claims data on the Medicare Provider Analysis and Review (MedPAR) File

The MedPAR file contains claims for both inpatient hospital and skilled nursing facility (SNF) stays.  During the time period covered by the NHANES and CMS linked data, the portion of claims for hospital stays on the MedPAR file fluctuated between 85% in 1999 and 82% in 2007.

You may wish to consider the portion of claims hospital or SNF stays when determining the approximate number of respondents in your analyses.  Course 2 Module 9 describes the process of determining your approximate sample size for proposed research projects using tables on the NCHS website.  Course 2 Module 8, Task 1 describes the process of using the NHANES-CMS linked data Feasibility Files to determine if there are sufficient numbers in your study population for an analysis. 

Vital status

Date of death information obtained by CMS is available on the Denominator File, but you should use caution with this death information.  CMS updates the Denominator File with death information collected through the first three months of the following calendar year.  Deaths to Medicare eligible beneficiaries occurring in the first quarter of the year will be recorded on that year’s Denominator File, but may also be recorded on the previous year’s Denominator File.  For example, a CMS recorded death occurring on 02/01/2000 will have a date recorded in variable DATE_OF_DEATH  ‘Date of Death’ on the 2000 Denominator File and may also have a death day recorded in variable DATE_OF_DEATH  ‘Date of Death’ on the 1999 Denominator File. 

While every effort is made to report correct death data, information is occasionally misreported to CMS. These cases can be identified because the beneficiary continues to be eligible for Medicare benefits in later years or they have new death information recorded in a later Denominator File. The erroneous information is not corrected by CMS in prior Denominator Files and, therefore, you should use extra caution in utilizing these data. In addition, the actual date of death information is occasionally mis-reported to CMS, since benefits are administered on a monthly basis.

Mortality information is also available from the NCHS Linked Mortality Files that were described in Course 1, Module 2 Task 1.  A link to this information may be found in the Resources section.  No attempt has been made to reconcile inconsistent death information from CMS and these other sources.

information icon Information

RDC research proposals that intend to analyze mortality outcomes should utilize death information from both the Medicare data and the NCHS Linked Mortality Files.



close window icon Close Window to return to module page.