Background and Significance Methods and Design
3 The Medicare Enrollment Data Base
The data we propose to evaluate are from the Medicare Enrollment Data Base (EDB), collected by HCFA. This data base potentially provides the most reliable information attainable regarding the size and age of the population of older Americans, much more so than data collected through periodic censuses [22]. According to HCFA, Medicare covers 95% of the population aged 65 and older. This percentage may be higher for people aged 75 years and older.

From HCFA we requested data for all Medicare beneficiaries born before 1916 who are not, or were not at death, Railroad Retirement Board beneficiaries, and approximately thirty million records were provided from their Medicare EDB. Records from the EDB are not created by HCFA, but are transferred from the Social Security Administration (or from the Railroad Retirement Board). Enrollment in Medicare is done through a Medicare-specific application process with the Social Security Administration, which includes an age-validation component.

Based on Kestenbaum’s [9] analysis of the oldest-old using Medicare enrollment data, it was determined that the most reliable data are from records indicating SSA and Medicare Part B insured - Railroad Retirement Board beneficiaries are not included as they were determined to degrade the data. Kestenbaum determined that records indicating enrollment in Part B, or Supplemental Medical Insurance, provided the most reliable mortality data, as opposed to records for those enrolled in both Part A, or Hospital Insurance, and Part B. This is due to the fact that Part A is generally automatically provided and at no extra expense, while Part B requires the payment of a premium. For those who are enrolled in Part B, they are billed, and enrollment is terminated automatically if there is no payment made. Part A terminations are less frequent and are not automatic. According to Kestenbaum, an unknown though probably significant proportion of those enrolled in Part B terminated for not responding are deaths. This provides an important death verification element to the data set.

Other data elements acquired were the Medicare beneficiary claim number - the number assigned to each beneficiary based on their Social Security number (SSN) of an associated primary beneficiary; full name; prior surname(s); birth date; death date (if dead); race code; sex code; state and county of residence; prior residence(s) and date(s) of residence change(s); current SSN; SSA benefit payment status and history (code and date); full Part B history (start and stop date(s), status code(s) indicating entitlement or not entitled).

Social Security numbers were included, not only because they provide a discreet identification number for each record, but because the first three digits indicate the region in the U.S. from where the number was attained. Based on the SSNs alone, regional comparisons of mortality could be made. And because the data includes information about the current residence of each record-holder, comparisons can be made between current residence and the region where the number was acquired.

Kestenbaum [10] estimated age-specific probabilities of death above age 85 by race and ethnic groups for 1991 using the enhanced (Note 2) Medicare Enrollment Data Base. These were compa[red with the official 1989-1991 decennial life tables. In addition, a matching study was conducted with the 1993 National Mortality Followback Survey. Kestenbaum noted that procedural improvements had improved the EDB, concluding that, "The enhanced Medicare enrollment file supports the reliable description of the mortality and size of the extreme aged population."

There are limitations of the EDB. According to McKinley and Frase [15], during the initial registration drive for SSNs that took place during 1936 and 1937, there were major flaws to the procedure. The bulk of the responsibility for the registration process fell on the United States Postal Service, which did not have the manpower to see to the accuracy of the process. One aspect of this was confusion regarding the assignment of the area numbers, or the first three digits of the SSN, when large employers would send their employees’ application forms not to their local postmasters, but to post offices of their corporate headquarters often located in major cities. The authors state "this seriously decreased the usefulness of the area number scheme", but we do not know to what extent.

The EDB does not include full records for those individuals who died prior to 1975. Historical information for Part B enrollment, place of residence and SSA benefit payment status data elements was not included in the EDB until 1991, therefore any changes in status on these data elements are only from 1991 on.

In July of 1994, HCFA replaced the race data, which had previously been limited to the categories white, black, other and unknown. To these categories were added Asian, Hispanic, and North American Native. The race data were only replaced for active records and not for persons who died prior to July 1994. In addition, Lauderdale and Goldberg [11] argue that the method by which HCFA revised their race data was incomplete and biased. However, analysis using data for whites and data for Blacks should be relatively unaffected by the 1994 change (Kestenbaum, personal communication w/Owens).

There was a recent initiative at HCFA to repopulate the SSN field so that 99% of the SSNs would be a person's own, not that of a beneficiary a record holder may be associated with, but examination of the data indicates closer to 1.5% of records have missing or duplicate SSNs.

Background and Significance Methods and Design

logo70.gif (2450 bytes)

Evaluation of U.S. Mortality Patterns at Old Ages
Using the Medicare Enrollment Data Base.

Allan M. Parnell and Cynthia R. Owens
1999 - 2000 Max-Planck-Gesellschaft ISSN 1435-9871