Creation of New Race-Ethnicity Codes and SES Indicators for Medicare Beneficiaries - Chapter 4
4. Producing Requested Tabular Analyses Incorporating the SES Measure
The second objective of this project has been to produce tabulations of Medicare beneficiary utilization that includes variables representing race/ethnicity and SES. The tabulations were to further elaborate upon tabulations that were completed for the CMS task order discussed earlier (Health Disparities: Measuring Health Care Use and Access for Racial/Ethnic Populations, Contract No. 500-00-0024 Task No. 8). The elaboration consisted of the addition of the newly created SES index to the tabulations. The added SES index was added as a four level categorical variable in which the numerical level of the SES variable corresponds to the quartile of the distribution of the SES index scores. Beneficiaries included in SES level 1 contained index scores that were in the first quartile (lowest values) of the SES index distribution, and beneficiaries in SES level 4 had SES index scores in the fourth quartile (highest values). Those in SES levels 2 and 3 were in the second and third quartiles with SES index scores in the next to the lowest or next to the highest quartiles, respectively. One could think of these levels as representing low, medium low, medium high, and high SES.
4.2 Tabulation Format
We have prepared approximately 1,500 tabulations covering almost 3,000 pages for this sub-task. While the measures of utilization included in the tabulations are numerous and varied, the format of the tabulations, and of the appendices in which they are included, are in much the same format. This uniformity was intended to facilitate review and use of these tabulations in subsequent analyses.
Because the tabulations are so numerous, they have been organized into separate appendices. There are four separate appendices containing tabulations Tabulations dealing with the use of cancer screening procedures are in Appendix B. Tabulations related to secondary prevention of the complications of diabetes are contained in Appendix C. Appendix D contains tabulations involving ambulatory care sensitive conditions. The final set of tabulations in Appendix E involve hospitalizations for common discharge diagnoses and report on average length of stay in days and mean expenditures in dollars.
Each of the appendices is divided in two separately bound parts. The first part of the appendix contains weighted tabulations. The weighted tabulations represent utilization of services by the population of beneficiaries in fee-for-service Medicare during 2002, the year for which the utilization data were extracted from Medicare claims. The second part of the appendix presents the same tabulations, but they are unweighted. They are labeled as "unweighted" because they only represent the sample. The unweighted tabulations have been included so that persons using the weighted estimates can check the actual number of observations on which the estimates are based to be sure that the weighted estimates are based upon a large enough sample to be considered reliable.
The tabulations in Appendices B-E are typically presented in the same format. That format presents numbers and percentages of beneficiaries using a particular service according to race/ethnicity, SES level, gender, and age group. One exception is in Appendix B where tables may be limited to males (PSA test) or females (mammogram or Pap smear) alone. A second exception is in Appendix E, where in addition to the number and percentage of beneficiaries using the service (hospitalization), there are separate tables also presenting the mean expenditure in dollars and length of stay in days.
The numbers in the cells of the tables refer to the utilization variable. Arrayed across the top of the table is the ordinal SES measure (Total, SES level 1 (lowest), SES level 2, SES level 3, and SES level 4 (highest)). Note that the Total column only includes beneficiaries who have been assigned to one of the four SES levels. Arrayed down the left side of the table are the nominal race/ethnicity groups (Total, White, Black, Hispanic, Asian/Pacific Islander, American Indian/Alaska Native, Other, and Unknown/Missing). It should be noted that the way CMS has created the Hispanic category on the EDB is without regard to race, so it is a non-redundant category. We adopted the same approach in our algorithm for creating the improved NEWRACE variable. This means that the White, Black, Asian/Pacific Islander, American Indian/Alaska Native, and Other categories are implicitly non-Hispanic. Nested within the race/ethnicity categories are categories of gender (male, female) and age group (under 65, 65-74, 75-84, and 85 or over).
All of the tabulations present a column of numbers and a column of percentages for each SES category and the total across all four SES categories. The numbers in the columns represent the beneficiaries in the particular cell (column and row intersection) of the table who used the service being analyzed. The percentages are the result of dividing the number of beneficiaries who used the service (numerators) by the entire number of persons in the particular cell whether or not they used the service (denominators). Tables containing the denominators for the percentages in the tabulations - national and MSA, weighted and unweighted - are included in Appendix H. However, note that the denominators for the percentages in the tables in Appendix B (cancer screening) may be gender specific depending upon the utilization variable (mammogram and Pap smear include females only, PSA includes males only) and that in Appendix C for tables subsequent to Table 1 (the number of beneficiaries with diabetes), Table 1 contains the denominators for the cells because the remaining tables are based on beneficiaries identified as having diabetes.
As a courtesy to the analysts who will use the weighted tabulations, we have not suppressed any weighted estimates, regardless of how small the sample size on which they are based. However, we have appended a column containing an asterisk whenever the sample frequency in the unweighted tables on which the weighted estimate was based did not reach 50 cases, the statistical criterion that we have adopted as the minimum acceptable sample size for estimates to be considered stable or reliable. Note, that while there is only a single asterisk per cell, it applies to both estimates (percent and number, and in Appendix E to mean cost and length of stay, as well). The asterisk also appears in the unweighted tables to highlight cells containing fewer than 50 observations.
|We strongly recommend to everyone using the weighted tables that they always check the corresponding unweighted tables before deciding to use any numbers from the weighted tables. We definitely recommend that no weighted numbers based on fewer than 50 sample cases be cited or reported as part of this or any subsequent analysis. We do not consider estimates in this project based on table cells of 50 or fewer sample members to be stable or reliable, and that is why we have issued this warning.|
Within both the weighted and unweighted sets of tables, in all of the appendices, the tables are numbered, labeled (except for the word "unweighted"), and arranged in exactly the same way. The tables are numbered according to the letter of the appendix and start with "1" within each appendix. The numbering is complete with the weighted tables, and repeats with the unweighted. In other words, Table D-1 in the weighted part of Appendix D has a counterpart Table D-1 in the unweighted part.
The tables that are numbered only, e.g. B-2, D-6, E-12, etc., are tabulations for the entire Medicare fee-for-service population (if weighted) or sample (if unweighted) and are labeled as "National". Recall that we are also preparing similar tabulations for individual metropolitan statistical areas (MSAs). These tables always follow the tables for the national Medicare fee-for-service population (if weighted) or sample (if unweighted), but are numbered and titled differently. They are numbered with the same Appendix capital letter and number but also have a lower case letter after the number. These lower case letters run from "a" to "p". The designations "a" to "p" always correspond to the same 16 MSAs and that correspondence is presented in Table 4.1.
Table 4.1 Lower Case Table Letters and Corresponding MSAs Used in Appendix Tables
|Table Letter||MSA||Table Letter||MSA|
|a||Los Angeles, CA||i||San Diego, CA|
|b||Miami, FL#||j||McAllen, TX#|
|c||New York, NY#*||k||Honolulu, HI*|
|d||San Antonio, TX#||l||San Francisco, CA*|
|e||Riverside, CA#||m||Oakland, CA*|
|f||Chicago, IL#*||n||San Jose, CA*|
|g||El Paso, TX#||o||Orange County, CA*|
|h||Houston, TX#||p||Washington, DC*|
Due to their sheer volume, the tabulations included in Appendices B through E have been prepared and delivered in separately bound volumes and as separate EXCEL spreadsheet files on compact disks. Note that because some of the appendices are extremely long, we have bound Appendices B through E in two volumes each - one for the weighted tabulations and one for the unweighted. Appendix H (Denominators) has also been bound separately, but because of its small size it was not split into two volumes.
4.3 Data Sources
The tabulations in the Appendices draw on data from a number of different sources. Some of these have been mentioned in previous sections of this report. Here we intend to focus exclusively on the variables used to prepare the tabulations in the event that someone would like to repeat them.
The race/ethnicity variable is the NEWRACE variable that RTI staff developed as part of the earlier task order we have referred to several times. It should be noted that this variable was updated in the first sub-task of this task order to incorporate beneficiaries who joined Medicare between mid-2003 and October 2005, but since we were analyzing 2002 services utilization in this sub-task, we used the original NEWRACE creation based on the beneficiaries on the mid-2003 EDB.
The beneficiary gender and age group variables are based on the gender and birth date variables on the mid-2003 EDB. Age was calculated as of the end of 2002, the year for which we had claims to analyze. Age was grouped into four ordinal categories: under 65 years of age, 65 to 74, 75 to 84, and 85 years of age or over. The SES measure was discussed at length in the previous section of this report. RTI created the SES index score (and the four SES categories used in the tabulations) from block group level data representing characteristics of the residential addresses of beneficiaries extracted from Summary File 3 (SF-3) of the 2000 U.S. Census.
The largest source of data for the tabulations is Medicare claims for 2002. We abstracted the service utilization for selected health services and diagnoses for the members of the stratified probability sample of 1.96 million Medicare beneficiaries enrolled in fee-for-service Medicare for the entire calendar year of 2002. The exact data file source of the data in each table is noted at the bottom of the table.
4.4 Tabulation Contents
As we indicated earlier in this section, there are tabulations dealing with the use of cancer screening procedures, secondary prevention of the complications of diabetes, ambulatory care sensitive conditions, and hospitalizations for common discharge diagnoses based on data extracted from Medicare claims for services provided in 2002. For the tabulations on cancer screening presented in Appendix B, we examined the use of screening procedures for breast (mammography for women), cervical (Pap test for women), prostate (prostate-specific antigen or PSA test), and colorectal cancers (fecal occult blood test or FOBT, flexible sigmoidoscopy, and colonoscopy). All are covered Medicare services.
For the tabulations on secondary prevention of diabetes complications in Appendix C, we identified four services from claims filed for beneficiaries with diabetes-foot care (claims for therapeutic shoes or for a podiatry visit ), eye examination (claims for diabetics with eye examinations), physiological monitoring or testing (claims for testing services for hemoglobin A1c, lipid profiling, or micro albumin for monitoring insulin needs), and instruction in self care (claims for obtaining instruction in diabetes education and self-monitoring) that are covered in the traditional fee-for-service Medicare plan.
We used claims for 15 selected diagnoses that resulted in being admitted to a hospital or observed in an emergency room during the 2002 calendar year as the basis for the set of tabulations on 15 ACSCs in Appendix D. Hospitalization for these ambulatory care-sensitive conditions (ACSCs) is useful as an indicator of inadequate access to or poor-quality of primary care (Bindman, Grumbach, Osmond, et al., 1995). Among the 15 ACSCs we examined were five chronic conditions (chronic lung disease - asthma and chronic obstructive pulmonary disease combined, congestive heart failure, seizures, diabetes mellitus, and hypertension); eight acute conditions (cellulitis, dehydration, bacterial pneumonia, urinary tract infection, gastric or duodenal ulcer, hypoglycemia, hypokalemia, and ear, nose and throat infections); and two preventable conditions (influenza and malnutrition) (McCall, Harlow, and Dayhoff, 2001). Tabulations were also done for the sets of chronic, acute and preventable conditions and the diagnosis of any ACSC.
In Appendix E, we examined hospital utilization patterns among Medicare fee-for-service beneficiaries for six conditions during 2002. The conditions included heart disease, cerebrovascular disease (stroke), malignant neoplasms (cancers), diabetes, pneumonia, and fractures. In particular, we tabulated the number and proportion of persons with each of these specific diagnoses at discharge, the mean payment made per user, and the mean length of hospital stay in days.
As indicated above, the tabulations for each of these types of utilization are in separate appendices. Documentation of the programs used to create these variables in the tabulations is included in Appendix F that is bound in the report.