Eliminating Health Disparities: Measurement and Data Needs (2004)

Jeffery J. Geppert, Sara J. Singer, Jay Buechner, Lorin Ranbom, Walter Suarez, and Wu Xu*

The purpose of this paper is to analyze variations in current methods for the collection of racial and ethnic data among states and between states and the federal government, to analyze the costs and benefits of enhancing or standardizing such data collection, and to describe selected recommended practices for the collection and use of data on race and ethnicity. States obviously have an interest in promoting the health and welfare of all their citizens and ensuring minimum health disparities among population subgroups. The ability to identify meaningful characteristics of individuals, including their race and ethnicity, for purposes of program planning, implementation, and evaluation is an important tool of public policy at the state and local government levels. Therefore, states have made significant efforts to collect detailed health data by race and ethnicity (and other cultural and socioeconomic categories) (National Conference of State Legislatures, 2002). These data are beneficial for planning program eligibility, identifying populations with particular needs, generating hypotheses about potential causes, and tracking progress. However, there are also important barriers and costs

to collecting such data, including problems with small numbers for particular subgroups, tensions between sufficient detail and meaningful aggregation, reconciling trade-offs with data collection of other important individual characteristics, and other resource constraints.

Through interviews with health agency staff and review of published sources, we examined how selected states approach the collection of racial and ethnic data. The background section describes the most common data sources used by states that collect racial and ethnic data. The methods section describes our data collection procedure for interviews and data analysis. The results section describes some of the variation among states in the collection of data on race, ethnicity, and other socioeconomic characteristics, and the benefits and costs of such data collection from the perspective of state administrators. The conclusions section provides some considerations for further research.

States collect racial, ethnic, and other socioeconomic data through multiple data sources as part of their administration of state and federal health programs. Some of the most common data sources and their characteristics are described below and summarized in Table E-1.

Some 37 states collect discharge abstracts on patients admitted to nonfederal hospitals (that is, excluding VA and Department of Defense facilities). Approximately 85 percent of the states use a Uniform Bill for Hospitals (UB-92) format for collecting discharge data (AHRQ, 1999). Because the primary purpose of the UB-92 is to pay a claim, racial, ethnic, and socioeconomic data are not included in the UB-92 core billing standards. However, 27 states (55 percent) collected data on patient race and ethnicity as a part of their inpatient data using state fields.

All 50 states collect birth and death certificate data, and virtually all states provide the National Center for Health Statistics (NCHS) with racial, ethnic, and socioeconomic data from birth and death records. The standard certificates are defined by NCHS and data collection is specified in a certain way. Most states use the standard for reporting to NCHS, although there is variation at the local level. About 12 states collect data differently from the standard.

Forty-five states have cancer registries. The National Cancer Institute (NCI), through its Surveillance, Epidemiology, and End Results (SEER) program, sponsors 11 population-based registries. The Centers for Disease Control and Prevention, through its National Program of Cancer Registries (NPCR), sponsors registries where the NCI does not. Together, the two agencies cover 100 percent of the U.S. population. Many registries began in individual hospitals in accordance with guidelines established by the American College of Surgeons (ACoS) as part of the requirements for accreditation of oncology services. The main purpose of the hospital registry was to provide physicians with the data needed to maintain quality of care through peer review and to compare performance with recognized standards. Racial, ethnic, and socioeconomic characteristics are important data to epidemiologists who investigate cancer, so the categories are quite detailed. Differences in incidence rates among different racial, ethnic, and socioeconomic groups generate hypotheses for researchers to investigate.

All 50 states participate in the state-federal Medicaid program. Enrollment data include information on each beneficiary’s race, ethnicity, and

income. The data are collected from hospitals, physicians, skilled-nursing facilities, and other practitioners participating in the Medicaid program.

Many states collect supplementary data on self-reported health status through telephone and other surveys. Health assessment surveys based on large populations give state health planners, policymakers, county governments, and community organizations more detailed pictures of the health and health care needs of various segments of the population. Typical information includes where and how people get health care as well as the number of adults and children without health insurance. Health surveys also collect information on important health conditions such as cancer, diabetes, and asthma.

To collect information for this paper, we interviewed and surveyed representatives in four states (Rhode Island, Ohio, Minnesota, and Utah) from various state agencies responsible for the collection of health data that include information on individual race or ethnicity. Participating states and individuals were selected nonrandomly, based on recommendations from officials in the National Association of Health Data Organizations. For five selected states (California, Florida, New York, Utah, and Wisconsin), we also examined various data sources to investigate the nature of the racial, ethnic, and socioeconomic status data actually collected. The survey instrument is included at the end of this appendix. We interviewed and collected written responses from state administrators responsible for the most common state data sources (i.e., state discharge data, vital statistics, cancer registries, Medicaid enrollment, and health interview surveys). The questions were divided into two sections. The first included questions about the perceived benefits of collecting, enhancing, and standardizing racial and ethnic data. The second section included questions about the perceived barriers or costs to enhanced or standardized racial and ethnic data collection. The questions are summarized below:

Questions on Costs

• What racial and ethnic data do your agency currently collect?

• How is this information collected? Self-reported? Reported by clinician or other third party?

• What organizations report this information to your agency?

• How are the data your agency collects currently used in your state?

To whom are they reported?

• Who has the authority to make a decision about the data that your agency collects? Are data elements governed by law, regulation, agency authority, other?

• What stakeholders would be opposed to changes to the data your agency currently collects?

• How long would it take your agency to implement the data requirements? What would require the most time?

• Can you quantify or estimate the costs to your agency and the organizations reporting to your agency of any of the changes necessary to implement new data collection requirements?

Questions on Benefits

• What is the intended purpose(s) of the racial and ethnic data your agency currently collects?

• What qualitative benefits to your agency, the organizations reporting to your agency, other government entities, beneficiaries of public programs, or the public (i.e., stakeholders) do you foresee as a result of more detailed or standardized data collection? Better service provision? Greater comparability in reporting? Opportunities for improved quality of care?

• Can you identify any quantitative benefits to your agency or other stakeholders that might result from more detailed or standardized data collection? Less supplementary data collection? Opportunities for economies of scale? Simplification? Improved ability to target public or private programs?

• To whom would these benefits accrue? Your agency? Other state or local agencies? Hospitals and other health care providers? Public programs or program beneficiaries?

• How long do you expect it would take for your agency or other stakeholders to achieve these benefits? Immediate? Less than 1 year? Less than five years? Longer than 5 years?

• Can you quantify or estimate the savings or other benefits to your agency or other stakeholders due to more detailed or standardized data collection?

In addition, state representatives provided information about recommended practices they have experienced or observed. Recommended practices are strategies and practices related to the collection and use of racial and ethnic data that are, in the opinion of the state data organization representatives, particularly effective.

Our emphasis was not on a systematic analysis of all state collection

practices of racial, ethnic, and socioeconomic status data, but rather a review of selected state practices in order to identify important issues and considerations for future research. This section describes the results from our interviews with state health administrators, a review of published documents, and preliminary data analyses. First, we provide examples of variation among and within states in the reporting of racial, ethnic, and socioeconomic status (SES) data. Second, we provide examples of variation among states in the actual collection of racial, ethnic, and SES data. Third, we highlight the perceived benefits and barriers or cost the collection of these data. Finally, we describe some recommended practices of states that have used such data, and of other cultural or socioeconomic data, to improve health and health delivery in their states.

States vary in the categories used to collect racial, ethnic, and socioeconomic status data for the common data sources described above. However, the variation is more extensive within states among the various types of data than among states for a particular data source. As noted earlier, race and ethnicity are not included among the UB-92 core elements for state hospital discharge data. Individual states have added either a single data element on race, ethnicity, and socioeconomic status or two separate elements. The actual categories of elements vary from state to state (Table E-2). In particular, states vary in the treatment of Hispanic ethnicity, as a field either separate from race or interacted with race. There is also variation in the categories for Asian and Pacific Islander, and in the treatment of unknown or no response.

There is even more variation within states, among discharge data and the other types of data that states collect. There is a spectrum of specificity of data collected: in general, hospital discharge data are at one extreme, with limited racial and ethnic categories; survey and registry data are at the other extreme, with multiple categories, including the ability to record more than one race per individual; and Medicaid enrollment data and vital statistics data are in the middle. For example, the California Cancer Registry has 29 race categories and 9 Hispanic ethnicity categories (Table E-3). In contrast, California Medicaid has 18 race categories. Birth and death records in California include 16 race categories and 7 Hispanic ethnicity categories.

One accompanying factor in the extensive collection of racial, ethnic, and SES data is the need for training in data collection. The more extensive the data collection, the more intensive the training required. Vital statistics and registries both have extensive training documentation for the reporting of racial, ethnic, and SES data. The categories of race and ethnicity were

central to the origins of these training systems and to their current development and use.

In addition to variation among states in the categories of data collected, states also vary in the completeness of that data. Missing data could be the result of various factors, including whether the data are formatted into a common State Inpatient Data (SID) format. The SID data use a common denominator approach to coding race and ethnicity, and variation among states in the coding may mean that some states do not have race or ethnicity reported in the Healthcare Cost and Utilization Project (HCUP) data set. States with more diverse populations might be more proficient or have more extensive training in data collection and reporting.

States also vary in their requirements for compliance by data suppliers to report racial ethnic data to the state agency maintaining the discharge

data system. States may mandate the reporting of racial, ethnic, and socioeconomic status data, or they may mandate the reporting of discharge data but not require race or ethnicity to be a part of the record.

Table E-4 shows that states that require the reporting of race or ethnicity as part of the discharge data submission have higher rates of compliance than states that collect the data voluntarily or do not require re-submission if the data are missing or invalid. States with mandated reporting showed 97 percent compliance in data collection and reporting, while states with voluntary reporting showed 83 percent compliance. The range of rates of compliance among hospitals was also narrower in the mandatory states

than in the voluntary states. The results again seem to demonstrate that reporting requirements and training matter in data collection.

The state administrators we interviewed identified five main benefits of collecting racial, ethnic, and SES data in terms of the programs administered by their agencies.

(1) Identifying population needs

States use data on race and ethnicity to measure the health of subgroups, in terms of both levels and trends, and to identify the nature and extent of potential disparities or other special needs in health among subgroups.

(2) Program planning

States use data on the race and ethnicity of residents to plan program eligibility, especially by identifying subgroups that are growing faster than the general population.

(3) Program design

States use data on race and ethnicity to develop hypotheses about the potential causes of health disparities or other special needs among subgroups and to develop special initiatives to address those problems.

(4) Program evaluation

States use data on race and ethnicity to evaluate the effectiveness of existing initiatives targeted to addressing health disparities or other special needs among subgroups, to monitor the performance of health systems, and to track overall improvement in health.

(5) Public reporting

States use data on race and ethnicity to report to the public, health care professionals, the state legislature, and federal government on the health status of the population of state and federal program participants.

(1) Small numbers

For many states, there are relatively few individuals in particular subgroups. This is often characterized as a self-sustaining problem, as the data to document the actual numbers and to justify potential additional data collection are not available. Others also argue that from a public policy perspective, small, subgroups are even more likely to have special needs that go unrecognized and unaddressed, so that it is even more important to collect the data for these subgroups. For example, particular Asian or Hispanic ethnicities can constitute small distinct communities with distinct needs and programs. Approaches to dealing with small numbers include averaging over multiple years and reporting confidence intervals on estimates.

(2) Beyond “white vs. nonwhite”

In reporting and analysis, there are often trade-offs between using detailed racial and ethnic categories and aggregating to more general categories. Detailed categories may permit more specific inferences and hypotheses, but they add to the complexity of the analysis and may suffer from small numbers that make rates difficult to interpret. General categories may simplify the presentation and analysis, but at the expense of detail necessary to reflect local circumstances or particular conditions, or to provide sufficiently actionable information. Demand for particular detailed analyses from state policy makers and other constituencies are often the deciding factor on the level and extent of aggregation, rather than considerations based on evidence or reporting standards.

(3) Information crowd-out

In data collection, careful consideration must be paid to constraints on time and resources that may require reconciling the relative importance of additional data collection on race and ethnicity with data collection of other important individual characteristics. Surveys in particular face constraints related to respondents, and administrative data face constraints related to providers and other organizations that collect the data on any given individual.

(4) Resource constraints

Any modifications to existing data collection require resources for programming, database development, and training. In order for these costs to be justified, there must be clear supporting arguments and demonstrated need.

In order to realize these benefits and overcome these barriers and costs, states have adopted various strategies to collect and use data on race,

ethnicity, and socioeconomic status. These “recommended practices” are grouped below in four general categories: data collection, training, research and analysis, and reporting.

Initiatives in data collection include the development of policies and procedures for the collection of racial, ethnic, and SES data, practices to ensure the quality of the data collected, and new initiatives to collect data on previously underreported subpopulations.

Policies and procedures. Examples of racial, ethnic, and SES data collection policies and procedures include agencywide policies covering some or all of the following: minimum standard racial and ethnic categories to be collected, format for data collection forms, language for telephone or self-administered surveys, recommended method for collecting data from subjects, standard data edit criteria, and standard for maximum missing responses. Some state agencies took the opportunity to develop such policies as part of the conversion to the revised federal guidance implemented with the 2000 census.

The Family Outcomes Project at the University of California-San Francisco and the Center for Health Statistics in the California Department of Health Services developed a document entitled “Guidelines on Race/Ethnicity Data Collection, Coding, and Reporting” to standardize practices across the department. The guide includes recommended practices for data collection, coding and tabulation.

Quality control. As we have seen, ensuring data completeness and quality is as important as well-defined policies and procedures. Methods of quality control include reabstraction and/or recontact studies for specific databases, routine or special feedback of summaries of collected data to collectors, and linkage studies across databases.

Initiatives to collect data on previously underreported subpopulations. Improvement of data collection on specific subpopulations requires special effort (Office of Minority Health, 2000). In one example, officials with the Ohio Commission on Minority Health partnered with the National Council of La Raza, state agencies, and community organizations to develop the first demographic overview of Ohio’s Latino population, from recent history and heritage to educational attainment and health status.

We have also seen that programs vary in the intensity of the training, which can be critical when the state is relying on other institutions for primary data collection. Some examples include state-sponsored training sessions or state-developed materials for specific groups of data collectors (internal or external to the agency), such as primary care physicians for reportable diseases, funeral directors for death certificates, and hospital admissions clerks for hospital discharge data and Emergency Department data. Again, such programs have been developed to support the change to the new federal guidelines.

In addition to training data collection professionals, more general cultural sensitivity training and interventions among health care providers can lead to improved data collection or measurable improvements in the health of subgroups. For example, Alabama recently trained 29 bilingual people in English-to-Spanish medical interpreter skills—including skills in medical terminology, ethics, interviewing, and Hispanic culture. The New York University Center for Immigrant Health offers training to bilingual people in medical interpretation and a pilot program in remote-simultaneous medical interpretation, where practitioners and patients wear headsets to communicate in several different languages.

Data that are used are more likely to be collected. Cross-sectional studies of disparities (statewide or local), trend analyses for racial and ethnic groups, Healthy People 2000/2010 baseline and progress studies (U.S. Department of Health and Human Services, 1990, 2000), minority health data on agency websites, and web-based dissemination systems with racial and ethnic analysis options are all examples of initiatives to provide the incentive for collecting better data on race, ethnicity, and SES.

Such studies can lead to the development of initiatives to address measured health disparities. In South Carolina, surveillance studies demonstrated that the state had one of the highest rates of prostate cancer mortality in the nation. The identification of culturally related barriers to education and screening efforts was a central strategy for reducing overall mortality. The South Carolina Office of Minority Health (OMH) gathered focus groups of African American men and women and uncovered barriers to screening. These barriers included ideas about masculinity, misconceptions about how cancer develops, fear of death, and lack of knowledge about the location and function of the prostate. As a result, the OMH developed an information campaign called “Real Men Get It Checked,” which has been used by hospitals and churches throughout the state.

Examples of efforts to overcome obstacles to using data collected for reporting purposes include agencywide policies on categories to use when presenting multiple race responses and methods to allow “bridging” from previous categories to new categories.

South Carolina’s Budget and Control Board linked data from state agencies and the private sector to create a fuller picture of various populations served in the state. The board’s Office of Research and Statistics linked Medicaid claims, child-care vouchers, education, welfare, vocational rehabilitation, mental health services, motor vehicle crashes, juvenile justice, private inpatient hospitalizations, emergency room visits and admissions, home health visits, and other services. The data system uses unique person numbers rather than personal identifiers to ensure confidentiality.

States’ efforts to collect racial, ethnic, and SES data either have evolved over time in response to changing circumstances, as in the case of hospital discharge abstracts and other sources of administrative data, or have been carefully planned in advance and in detail, as in the case of many of the state-based health assessment surveys. These data collection efforts meet a variety of public policy needs, and must be tailored to specific local conditions and program characteristics. Yet standardized data collection, analysis and reporting practices are more likely to produce accurate and useable information. We conclude with a few considerations for future research in meeting these multiple objectives.

The identification of a standard set of socioeconomic and cultural factors of importance in identifying meaningful subgroups could be implemented across data sources. In addition to race and ethnicity, other individual characteristics might include income, education, and insurance status, based on existing evidence from health services and clinical research. A “minimum” set of factors could be used in administrative data systems, and a “maximum” set for registries and surveys. Defined procedures could permit aggregation from detailed categories into more general categories in two ways: vertically and horizontally. Vertical integration would allow, for example, Hawaiian, Pacific Islander, and various Asian nationalities to be pooled into a broader category, if desired. The level of detail could be tailored to specific local circumstances. Horizontal integration would include defined procedures for combining across socioeconomic and cultural

factors. Evidence might suggest certain combinations or weights that might be relevant for particular uses, with race and ethnicity assuming more or less weight as appropriate.

In health services research and performance measurement, there is always a perceived dichotomy between the relative merits of using administrative data and data collected from other, more detailed data sources, such as medical charts or individual surveys. Administrative data sources have the advantages of a relatively low cost to collect, universal coverage of entire populations, and transparency and reproducibility of data collection methods and analysis results. The disadvantages of administrative data are the lack of detailed information on individual demographic and clinical characteristics needed for meaningful analyses. Nonadministrative data such as medical chart abstraction and surveys can overcome these limitations, but often at considerable cost and limited availability and reproducibility.

Rather than selecting one approach over another, each approach should be pursued simultaneously and the bridge between them narrowed through standardization and automation. Standardization can enhance administrative data systems by adding additional data elements that are important for analysis of health disparities and quality of care. Standardization can also facilitate the automation of clinical data. Training is an essential and often overlooked component to this process. The lessons learned and practices developed for training in vital records and registries in particular could be applied to administrative data systems more generally.

Federal standards and state incentives to adopt them have an important role in creating a common infrastructure for the collection of racial, ethnic, and socioeconomic status data. For vital statistics, where NCHS has adopted a standard and provides resources to states to implement those standards, states have adopted the standard and adhere to it in their own reporting. In contrast, when the decennial census changes to OMB-specified categories, states may not comply and the federal government has no control over what states do (due to state mandates controlling the collection). To change to the new census/OMB standards, states will have to go through their legislative committees and rule-making procedures, and the change rate will vary by state. Similarly, the current HCUP project, a state and federal partnership to collect hospital discharge data, relies on states to report what they have and in what format they have it. The “common denominator”

data set contains a mix of categories and completeness, as we have seen. The incentive for states to change to a uniform format does not exist.

When the federal government takes the leadership role in setting standards and assisting states in paying for data collection in keeping with the standards, states are motivated and generally comply. Such standards also play an important role in building local constituencies and assisting states in moving from using the data for planning purposes only to using the data for research and analysis. As they may be adaptable to local circumstances, federal standards often make it easier to build the case for local implementation and use.

Finally, any federal standard must permit individual tailoring to meet state requirements and to reflect local conditions. In some states and for some instruments, going beyond the standard set will be essential to understand the state’s racial, ethnic, and other socioeconomic diversity.

Agency for Healthcare Research and Quality 1999 Healthcare Cost and Utilization Project 1998 Data Availability Inventory. (Conducted by NAHDO and Medstat). Rockville, MD: Agency for Healthcare Research and Quality.

National Conference of State Legislatures 2002 Racial and ethnic disparities in health care. State Lawmakers’ Digest 2(4): Summer.

Office of Minority Health 2000 Assessment of State Minority Health Infrastructure and Capacity to Address Issues of Health Disparity. (Developed by COSMOS Corporation under Contract No. 282-98-00127). Washington, DC: Office of Minority Health.

U.S. Department of Health and Human Services 2000 Healthy People 2010: Understanding and ImprovingHealth. Washington, DC: U.S. Government Printing Office.

1990 Healthy People 2000. Washington, DC: U.S. Department of Health and Human Services.

National Academy of Sciences

Panel on DHHS Collection of Race and Ethnicity Data

Cost-Benefit Analysis of State-Level Data Collection

The questions below are designed to solicit information from data collection agencies in selected states about the potential costs and benefits of enhancing and standardizing race and ethnicity data collection in the

state and among states and the federal government. In answering the questions below, please think specifically about your own agency’s data collection efforts and, where indicated and to the extent possible, the data collection efforts of other entities in your state. While we are interested in race, ethnicity, and socioeconomic status data in general, our analysis will focus on the following data sources: Medicaid data, Hospital Discharge Abstract data, Birth and Death Records, Cancer Registry data, State Health Interview Survey data, and Behavioral Risk Factor Surveillance System data. Therefore, we request that you also focus your responses on the subset of these data sets about which you are aware.

Please respond briefly in writing to the questions below, and return your response to Sara Singer at singer@healthpolicy.stanford.edu and Jeff Geppert at jgeppert@nber.org. A conference call will follow.

Costs: What would be the costs of enhancing and standardizing race and ethnicity data collection in your state? Nationally?

• What race and ethnicity data do your agency currently collect? Please specify the definitions used and category choices offered.

• How is this information collected? Self-reported? Reported by clinician or other third party?

• At what level of detail (i.e., individual patient discharge, aggregate) does your agency require this information to be reported?

• What organizations (i.e., hospitals, health plans, physician organizations) report this information to your agency?

• What race and ethnicity data are voluntary? Required?

• What is the approximate level of compliance with these reporting requirements?

• Are you aware of differences in race and ethnicity data currently collected by your agency and others within your state? If so, please specify.

• How are the data your agency collects currently used in your state? To whom are they reported?

• What other entities collect race and ethnicity data in your state, if known? Do you currently cooperate with any of these entities for data collection? Reporting? Other purposes?

• Who has the authority to make a decision about the data that your agency collects? Are data elements governed by law, regulation, agency authority, other?

• What stakeholders would be opposed to changes to data your agency currently collects?

• How would changes in data collection requirements affect implementation of and compliance with data collection? Within your agency?

Within hospitals, health plans, and physician organizations reporting data to your agency?

• How would changes affect your reporting of data currently collected?

• How long do you expect it would take to agree on data standards across agencies in your state? What would require the most time? What is the greatest unknown?

• How long would it take your agency to implement the data requirements? What would require the most time?

• Can you quantify or estimate the costs to your agency and the organizations reporting to your agency of any of the changes necessary to implement new data collection requirements?

Benefits: What would be the benefits of enhancing and standardizing race and ethnicity data collection in your state? Nationally?

• What is the intended purpose(s) of the race and ethnicity data your agency currently collects?

• What qualitative benefits to your agency, the organizations reporting to your agency, other government entities, beneficiaries of public programs, or the public (i.e., stakeholders) do you foresee as a result of more detailed or standardized data collection? Better service provision? Greater comparability in reporting? Opportunities for improved quality of care?

• Can you identify any quantitative benefits to your agency or other stakeholders that might result from more detailed or standardized data collection? Less supplementary data collection? Opportunities for economies of scale? Simplification? Improved ability to target public or private programs?

• To whom would these benefits accrue? Your agency? Other state or local agencies? Hospitals and other health care providers? Public programs or program beneficiaries?

• How long do you expect it would take for your agency or other stakeholders to achieve these benefits? Immediate? Less than one year? Less than five years? Longer than five years?

• Can you quantify or estimate the savings or other benefits to your agency or other stakeholders due to more detailed or standardized data collection?

Thank you for your input.