Washington, D.C.
The Subcommittee on Populations of the National Committee on Vital and Health Statistics met on February 11 and 12, 2002, at the Hubert H. Humphrey Building in Washington, D.C. The meeting was open to the public. Present:
The Subcommittee on Populations held hearings February 11 and 12, 2002 seeking to better understand the contributions and limitations of the various federal datasets in providing data on the health disparities of racial and ethnic groups. During the two days, the Subcommittee heard 17 testimonies from designers and users of seven datasets and discussed policy perspectives and issues of socioeconomic status and multiple-race data use.
Measurement of Health Disparities in Racial and Ethnic Groups in Federal Surveys
Medical Expenditure Panel Survey
Mr. Machlin said the Medical Expenditure Panel Survey (MEPS) was a large, comprehensive survey providing a rich set of analytic variables usable in conjunction with race/ethnicity. The purpose of the household survey component was to aid in estimating annual health care use and expenditures, provide distributional estimates, track changes and trends in insurance coverage and employment, and provide information characteristics and changes over time, as well as selected quality indicators. The core interview, consisting of demographics and health status measures, tried to enumerate all health care events of sampled persons, as well as charges and payments for events. MEPS’ sampling frame was a subsample of the National Health Interview Survey (NHIS) and similarly oversamples Blacks and Hispanics. Current sample sizes were large enough to make estimates for major race/ethnic groups. Linking to NHIS, one could expand the variables in an analysis. The race/ethnicity revisions, Office of Management and Budget (OMB) Directive 15, separated Asian-Pacific Islanders into a minimum of two groups, so Native Hawaiian and other Pacific Islanders could be separately identified. It also allowed respondents to identify themselves as multiple-race categories. MEPS could be used to do in-depth multi-variate analyses. Mr. Machlin noted analytic and contextual variables of interest for analysis of racial/ethnic disparities.
Medical Expenditure Panel Survey User
Dr. Lillie-Blanton evaluated her experience as a single user of MEPS, calling it an important dataset that allowed looking at race and contextual variables as well as outcomes. She assessed MEPS as a good data source in terms of data quality and understanding health care disparities. Sample sizes were more than adequate for bi- and multi-variate analyses. Self-reporting provided fairly good estimates for four of the five major population groups; the analysis and research agenda was good-to-fair; and MEPS rated fair-to-poor in terms of cultivating researchers who could analyze the data in terms of how racial-ethnic groups varied in health care use. In evaluating MEPS usefulness, Dr. Lillie-Blanton emphasized the broader context of its linkages to other data sources produced by the Department. And she noted that, while she was to some extent critical, she viewed MEPS as one of the most comprehensive data sources available in HHS to answer questions about health, health care use, and health expenditures. She said its weaknesses weren’t specific to MEPS, but reflected Department-wide limitations.
Consumer Assessment of Health Plans
Consumer Assessment of Health Plans, User
Dr. Sangl said the Consumer Assessment of Health Plans Survey (CAHPS) developed surveys measuring consumers' reports and ratings of care across all fee-for-service and managed care systems and reports (available to purchasers, plans, and providers) on results of the surveys. CAHPS is a consortium and products are collaborations of Healthcare Research and Quality (AHRQ), Centers for Medicare and Medicaid Services (CMS), Harvard, RAND, and RTI, with technical assistance provided to users through the Survey Users Network (SUN). Sponsors conduct CAHPS. Medicare conducts three surveys: managed care, fee-for-service, and disenrollment. Medicaid state and SCHIP programs as well as commercial sponsors conduct surveys. Some 90 million Americans are covered under plans for which CAHPS data is collected. Surveys are translated into Spanish, Russian, Vietnamese, Mandarin, Korean, and Cambodian. Two questions classified individuals by race and ethnicity. Dr. Sangl discussed confounding variables to consider with CAHPS data and data limitations. She noted the National CAHPS Benchmarking Database (NCBD) was available to researchers and had adult survey data for 793 plans and child survey data for 148 plans. Selected CAHPS items will be included in the MEPS survey, making other types of analyses possible and the survey will go below the plan level in the next phase of CAHPS.
Consumer Assessment of Health Plans, User
Mr. Moser explained that the Medicare managed care (MMC) version of CAHPS was created to obtain information from enrollees about their plans, providers and self-reported overall health status; health conditions; and health care system utilization over the last six months. He said CAHPS surveys provided a unique opportunity to investigate certain racial/ethnic groups. Non-White groups accounted for about 14 percent of Medicaid managed care enrollees. Only 200-300 individuals responded in the smaller racial/ethnic groups in any year, but combining three survey years built up the sample size, so some cross-group comparisons could be done. Mr. Moser noted numerous differences and similarities between the MMC subgroups. Asians self-reported the best health and lowest utilization levels. Whites reported good health, but were above average care utilizers. American Indians/Alaska Natives, followed by Blacks or African Americans had the poorest health. Hispanic Latinos had average self-reporting and utilization. Female MMX enrollees reported worse health, however they also reported lower levels of the five serious health conditions. Smokers were in worse health than non-smokers.
Medicare Current Beneficiary Survey
Mr. Waldo said the principle goal of the Medicare Current Beneficiary Survey (MCBS) is understanding how Medicare serves its target population. Sample persons (SPs) drawn from the Medicare master enrollment list maintained by the Centers for Medicare & Medicaid Services each January remain in MCBS for three, interviewed 3 times per year. Information is collected on: demographics, social/economic characteristics, health status and functioning, use of health services, financing of health services, and interactions with Medicare programs. SPs self-identify race and ethnicity. A decision was made early to over-represent on the basis of age, rather than race/ethnicity, in order to capture populations in need of particular services. The MCBS employs OMB protocols for racial identification; in 1998 the survey switched to allowing SPs to identify with more than one race. Mr. Waldo said a lot could be done using the MCBS to study race/ethnicity disparities. He expressed confidence in running analyses that involved a single dimension or looked at common events among the target population. He cautioned against trying to look at rare events or break the population into too many cells.
Medicare Current Beneficiary Survey User
Dr. DaVanzo presented a study the Lewin Group did that demonstrated MCBS could be utilized to identify differences. Using a 1997 MCBS supplement on information needs, Lewin looked at a combination of needs and preventive health behaviors. The survey gauged beneficiaries’ understanding of the Medicare program; Lewin looked at that understanding in relation to the subgroups and preventive health behavior. The three largest groups were collapsed and broken down by gender, race, and knowledge. Dr. DaVanzo said the dataset was extremely useful for looking at something this broad with only two dimensions. She said that identifying the combined action of ethnicity, behavior, and knowledge, they’d found out enough about the beneficiaries to educate them. Dr. DaVanzo said a lot of analysis could be done in the three broad groups, but MCBS wasn’t appropriate for studying specific subgroups or making precise population estimates.
Policy Perspectives
Dr. Clancy noted there were many non-clinical determinants of health outcomes; half to three-quarters of the disparities couldn’t be explained by income and health insurance coverage. She drew upon a framework John Eisenberg developed looking at “voltage drops” between a population-at-large, a healthcare system presumably here to serve, and everyone getting quality of care. Noting the Commonwealth Fund report considered federal data collection to include data collection on claims for services and efforts where the federal government joined forces with the private sector to assess and improve quality of care, Dr. Clancy pointed out that data needed for surveillance and monitoring might not be equivalent to data needed to improve healthcare. She remarked that data on race/ethnicity wasn’t consistently available from healthcare providers and what was available often wasn't complete, compliant with OMB or easily linked to data on the other characteristics. Dr. Clancy said her wish list was to mandate collection of these data from all who receive federal funds consistent with the OMB standard. Dr. Clancy emphasized two needs: exploring strategies for rapid release of selected data for improvement purposes and thinking carefully about a strategy for reporting these data. She suggested the Committee might need to rethink their approach to data collection. They knew all improvement was local, but most the data was nationally representative and wouldn’t address many of the ethnic populations’ needs.
Socio-economic Status
Dr. O’Campo demonstrated there were numerous individual and socio-economic status levels to choose among. She recommended using more than one, noting going beyond education and income was a move toward understanding issues of racial inequality in health. She encouraged going beyond the individual level and routinely including considerations of area-based measures of socio-economic status. Area-based measures helped capture complexities of socio-economic status, more completely characterizing population. Neighborhood SES measures increased understanding and the magnitude of neighborhood measures was larger than individual-level SES: effect modification and confounding was present when area-based measures were added. Dr. O’Campo emphasized that other areas impacted on health and that, if surveys captured that data, it could be brought into the equation and contribute to an understanding of racial inequalities. The choice of measure used depended on the health outcome studied. She recommended using multiple measures at the individual and area level and collecting socio-economic status for more than one period in the lifetime to account for socio-economic status across the races and comprehend racial inequalities in health.
National Survey of Family Growth
Dr. Abma said the National Survey of Family Growth (NSFG) was a periodic survey of the fertility of US women. Its framework, the Approximate Determinants of Fertility, contained intermediate variables most closely affecting live birth. The race/ethnic measure was one of the social factors that were more distal determinants of fertility. Five rounds of data collection starting in the early '70s formed a time series for women of all marital statuses dating back to 1982. The data was relied upon as a dependable source of national estimates to monitor trends and inform policy on differentials and issues. NCHS couldn’t produce breakdowns of the smaller race/ethnic subcategory; for some outcomes, the Hispanic subsample also became thin. NSFG’s strength, in terms of broad race/ethnic categories, was the ability to monitor trends and differentials across a wide range of outcomes in terms of women's reproductive health, pregnancy and childbearing. NSFG also included a fairly rich array of potentially explanatory variables. Modeling of causal hypotheses was possible. Explanatory analyses were another strength. Healthy People 2000 and 2010 used NSFG to supply data for ten objectives. NSFG showed characteristics of women using Title X family planning clinics and others who might be in need of such services. NCFG also provided information for CDC's HIV prevention program. The next survey, Cycle 6, would also interview males, adding the other half of the information missed for years.
Behavioral Risk Factor Surveillance Survey
Dr. Mariolis said the Behavioral Risk Factor Surveillance Survey (BRFSS) was a telephone survey (one eligible adult was randomly chosen from each household) primarily used to track the prevalence of behaviors related to chronic diseases and preventive health practices among the civilian, non-institutionalized population 18 years and older. BRFSS was a joint venture of the CDC and health departments in the 50 states, District of Columbia, Guam, Puerto Rico, and the Virgin Islands. The states held control and ownership of the data. Questions changed yearly. Every state was required to ask each core question and could choose standardized sets of questions on specific topics called modules. States were free to ask their own additional questions. The aggregate data file was available to the public. There were a large variety of topics, even in the core. None were gone into in depth. There were some 184,450 interviews in 2000; 204,000 were expected in 2001. Dr. Mariolis said race as conceptualized in standard race questions, is especially problematic to everyone. He recommended always presenting results by race/ethnicity with Hispanic as a category, not by race alone, and never breaking Hispanics out by race. Dr. Mariolis said there was minimal direct impact from allowing multi-racial choices. Choosing other race was only to a small extent an alternative to multi-race choices. People who choose other race were predominantly not multi-racial or didn't give a multi-racial response to race.
National Health Interview Survey
Ms. Lucas said the Census Bureau annually conducted NHIS, one of the nation’s largest health surveys, for NCHS. The multi-stage probability sample (100,000 persons in 40,000 households with Blacks and Hispanics in high-density areas oversampled) has a stratified cluster design drawn to be nationally representative of the non-institutionalized civilian population. The household-based interview is conducted face-to-face with all household members available. One basic module administered to all family members and modules for a randomly selected adult and child collect data on: activity limitations, injuries, conditions, health behaviors, access to health care, utilization, health insurance, demographics, income and assets, and family composition. Topical modules varied yearly, adding flexibility to address emerging public health topics. Ms. Lucas presented examples of bivariate analyses of Health Interview Survey (HIS) data and identified issues to keep in mind when using HIS data to assess racial and ethnic disparities in health. Ms. Lucas noted OMB Directive 15 had implications, not only for how health outcomes were measured and for whom data was gathered, but also for how trends in data systems were maintained, overall changes in health outcomes were monitored, and the assessment of whether observed population changes were the result of changes in classification of groups or actual behavior changes/successful program intervention. Ms. Lucas provided a snapshot of how multiple race groups fit into the overall NHIS, uses of NHIS data to examine race and ethnicity reporting, and issues related to measurement of race and ethnicity. Ms. Lucas noted the need to more fully acknowledge the fluidity of racial and ethnic identities, which might change the fundamental concept of race. She said future directions for NHIS include examining over sampling of Asian population subgroups, considering targeted over sampling to study smaller groups, cognitive work at NCHS to examine commitment to a racial identity, and experiences with discrimination in seeking and receiving health care.
National Health Interview Survey User
Dr. Hummer used HIS extensively. Because this dataset had been around a long time, he said cautious comparisons could be made across time. Given its large sample size; HIS had tremendous strength for analyses looking at relatively rare health outcomes and behaviors. One could look at several race and ethnic groups as well as health outcomes and behaviors by age and sex. A household-based survey allowed for linkage between individuals in a household and looking at patterns of health and health behavior present. And HIS provided one of the largest and best datasets for looking at mortality patterns throughout the nation. He said the most critical thing about HIS was, despite over sampling, sample sizes for most minorities remained limited for many purposes. Addressing health disparities across a wide range of race/ethnic groups became even more difficult. He noted that a relative lack of other social and cultural variables in HIS, even in the special modules, seriously impeded understanding what lay beneath disparities in question. He recommended forming expert groups to help put topical surveys together and matching them to future years. He also noted that growth of racial and ethnic minority populations was largely due to immigration, but little attention had been given to immigration issues. Dr. Hummer emphasized that a redesign of HIS came in 2005; they were at a point where they could do something.
National Health and Nutrition Examination Survey
Dr. Curtin said the National Health and Nutrition Examination Survey (NHANES) had conducted a series of national surveys for 30 years. NHANES began with a screener interview for over sampling. A household interview was directly related to HIS and sample design issues noted with HIS were even more apparent in a six-year survey with only 81 primary sampling units (PSUs) and 30,000 examined people. However, Dr. Curtin noted this was a highly screened population. He said NHANES’ mobile exam center was the key to its being a rich, invigorating data source. NHANES III and HANES were designed to get at: Mexican Americans, non-Hispanic Blacks, and non-Hispanic Whites. He noted it took ten years to plan a protocol, get it in the field and collect six-years of data, and do data clean up--And most researchers didn't want to wait that long. Designed to be a “representative sample” on an annual basis, Dr. Curtin noted severe limits in that schedule and the likely alternative of more flow bases and release every two years. NCHS anticipated a public use file for HANES 1999-2000 would be an Internet data release in July. An issue with the release involved OMB guidelines on race and ethnicity and a severe confidentiality concern. Major strengths included the ability to control selection to oversample minority populations, do stratification in screening, and that at each multi-stage the sample selected at random. Dr. Curtin cautioned that, without a large sample, randomization could cause a problem. He discussed foibles of analysis, emphasizing the small sample size, considerations in tradeoff between bias and variance, and being extremely careful with the design-based estimation in terms of influential values, sample rates, and dealing with degrees for the variance estimations. Noting that with these design limitations, it was difficult to get at specific sub-domains, Dr. Curtin discussed a plan for a more flexible data collection approach called Defined Population or Community HANES.
National Health and Nutrition Examination Survey User
Dr. Sempos said assessment and monitoring of minority health status was always a key feature of NHANES. NHANES produced estimates for Mexican Americans, non-Hispanic Blacks and non-Hispanic Whites (85-90 percent of the U.S. population) to assist federal agencies in developing and monitoring public health policy. Two types of data were collected: self-reported aspects of health (including race and ethnicity) and measurable physical attributes of individuals. HANES’ strength was measurement of those physical characteristics. HANES assessed national mean levels and distributions of health status indicators. It also got national prevalence estimates with certain, often unhealthy, characteristics and health status indicators. One could produce national trends in health status indicators using successive NHANES. It also could document potential areas of unmet medical need associated with health disparities. Dr. Sempos emphasized that NHANES described national levels and trends in health-based status indicators by age, sex, race and ethnicity extremely well. What it didn't do well, except in selected cases, was disease diagnosis. And it wasn’t good at producing subnational estimates. Most importantly, it didn’t explain how health disparities came about or could be reduced. Dr. Sempos concurred that community HANES was an outstanding way to supplement NHANES’ data deficiencies. He urged the Subcommittee to encourage federal agencies to identify communities and populations to examine on a regular cycle basis and for NCHS to expand development of cohort studies as an ongoing NHANES product. Noting the vital statistics data obtainable on the Internet as CDC Wonder was only available for Black, White and Other and that data for the Commonwealth of Puerto Rico and other trust territories and protectorates weren’t necessarily included, he suggested the Subcommittee recommend that CDC include that data.
Policy Discussion
Dr. Kington said the data sources they’d discussed were important to NIH for two reasons. They conveyed the magnitude of the problem of disparities in health and health care, helping set priorities for a research agenda. And they gave insights into fundamental causes of these differences, ultimately pointing the way toward likely places for intervention. Noting the single most important problem echoed that morning was how to obtain data on smaller subgroups, he said the idea that all data needs could be solved by over sampling reflected poor understanding of how diversity played out in the nation. Dr. Kington proposed two parallel tracks of data collection; a core tract that got at the large racial and ethnic population subgroups and another in which data were collected in a similar way, allowing comparisons with the national data, on a mosaic of smaller groups. Noting comments about the need for more complex data on a range of different dimensions (especially social and behavioral factors) for racial and ethnic minorities, he recommended a more rational planned system of cycling through supplements already used. Noting serious problems with funding, he encouraged demonstrating in concrete terms the value of good data on health across a wide array of racial and ethnic groups was in solving public health problems. And he suggested major efficiency gains could be made in how data was collected. He encouraged the Committee to go beyond simple questions and explore perceptions of race, in addition to self identification, thinking about race as an exposure variable, and adding a dynamic dimension to race and ethnicity that led toward understanding race within the context of a particular social setting.
Multiple Race Data Use
Dr. Smith said the discussion of multiple-race measures brought up new technical specifics but he emphasized reoccurring basic themes: (1) race and ethnicity were social constructs; (2) how they were conceptualized, defined and measured depended on social/legal conventions and scientific and policy purposes for which the information was to be used; (3) race and ethnicity were complex variables; (4) the information collected on race and ethnicity depended to a notable degree on the way they were measured. Dr. Smith noted race and ethnicity were difficult variables to measure, given their changing social definitions and the complexity of people's ancestry. Multiple-race questions improved measurement by providing responses that captured people's ancestry more accurately and by measuring a socially recognized and growing segment of the population. But the new multiple race standard also caused problems. The census item was ill suited for comparisons with other data sources using the traditional, one-race standard. Multiple-race questions added even more small groups. And multiple-race measurement was sensitive to the precise way in which items were worded and administered. Dr. Smith emphasized that measurement challenges weren’t reasons for avoiding collecting multi-race identifications, merely indicators that race and ethnicity must be measured carefully.
Remarks by James Scanlon, Director, Division of Data Policy, DHHS
Mr. Scanlon noted NCVHS was one of the most highly regarded and productive federal advisory committees. Timely, reliable statistics were an essential element of HHS and its partners’ (state and non-governmental entities and providers in health and human services) missions and programs. With a budget of over $400 billion and some 300 programs, he said HHS could be considered the largest health insurance plan, funder of biomedical and behavioral research, and prevention agency in the U.S. Equivalent to the fourth biggest government in the world, if among the Fortune 500 companies HHS would rank number one.
Over the years, Mr. Scanlon said HHS had a major role in looking at how to improve the area of data and statistics and many of the efforts to reduce and eliminate disparities had their locus in HHS. They had an array of data systems, both in Health and Human Services, that were regarded as the best in the U.S., and helped set the standard for the world. However, population and subpopulation data presented challenges that weren’t because of a lack of thinking or strategies. They faced a difficult area and resources that didn’t always match what was needed.
Within HHS, Mr. Scanlon said they tried to sustain one voice and one framework for speaking on data issues. They’d established the HHS Data Council as their internal group to coordinate activities, identify and address needs in a collective fashion across HHS. The Council’s task force and working group on race/ethnicity data recently reviewed virtually every recommendation made about improving race and ethnicity data over decades. The report (available on the HHS Data Council Web site) offered recommendations and an overall strategy that depend on standardized high-quality measurement. Mr. Scanlon noted HHS was probably the first cabinet department to adopt an inclusion policy. About four years ago, HHS adopted a policy to include standard race and ethnicity data in all HHS data systems. He noted it was easier to do this in data systems HHS sponsored; where HHS collected information from third parties, it was reliant on what those hospitals, providers and others collected. An updated directory containing information about the race and ethnicity detail available in the major data systems in HHS will be available next month on the Web site.
Recently, Congress directed HHS to support a study at the National Academy of Sciences to look at the adequacy of race and ethnicity data in HHS and other public data systems as well as the private sector. The academy is putting the panel together that will look at these issues, provide a projected cost analysis of what it would take to fill the gaps they identify and propose recommendations.
Work also began recently within HHS on a one-stop Internet-based gateway to data and statistics that is in its testing phases. In concept, the HHS gateway to data and statistics will be one click off the HHS home page. Shortcuts will link sophisticated users to major data systems; a search engine and data finder will help others. Initially limited to federally sponsored or supported Web pages in health and human services data, the range may be expanded later. The gateway also will include links to HHS data policy Web sites, statistical policy documents and activities in the federal government. There will also be links to the referenced literature NLM maintains and to research in progress that NIH maintains, including grants awarded.
Mr. Scanlon noted the attempt to pull all this together. Again, the data sources, publications and Web sites could only be as useful as the original data collection. For three decades, there had been standard race and ethnicity categories. Mr. Scanlon said their challenge wasn’t the categories, but having sufficient information on them and their subcategories. He said the Subcommittee was pleased to have so many ready to share ideas about where they stood with the OMB categories and ways to proceed. Even though the new OMB revision was issued in late 1997, it gave the agencies until 2003 to implement the new standards. Obviously, implementing changes in a standard used so long wasn’t a simple thing. There were questions relating to comparability with previous data, with time series, and with how to bridge previous information with the new. Mr. Scanlon welcomed everyone’s thoughts and recommendations.
Dr. Mays said the goals of that day’s hearing was to understand, through the illustration of data, the contributions and limitations of the various federal datasets in providing data on the health disparities of racial and ethnic groups. The Subcommittee had prepared questions they’d asked designers and users of the datasets, inviting them to share their insights today. One question asked how their survey data could advance knowledge of disparities in health and health care. Dr. Mays noted the Subcommittee was especially interested in differences in disease rates among the racial/ethnic groups and identification of populations that might receive unequal prevention screening or treatment services. Members also wanted to know if variables beyond race and ethnicity (e.g., confounding) were needed to document health and health care disparities among racial and ethnic groups and if analyses had been conducted in survey data. Another question to explore was how ethnic identity could best be measured. The Subcommittee asked whether it was feasible to link the various datasets to other contextual data that might help in further understanding causes and consequences of disparities. Members also expressed interest in identifying the cost of health disparities and whether these costs to society and budgets could be documented. During the day, Dr. Mays said other questions would be raised, ranging from whether surveys were translated and conducted in various languages to did they deal with the issue of under counting or over sampling ethnic and racial groups.
Measurement of Health Disparities in Racial and Ethnic Groups in Federal Surveys
Medical Expenditure Panel Survey
Mr. Machlin said MEPS was a large, comprehensive survey providing a rich set of analytic variables that could be used in conjunction with race/ethnicity. MEPS were a family of surveys with many components; he spoke exclusively about the household survey component. NCHS was a co-sponsor of MEPS and the survey’s sampling frame was a subsample of NHIS. Linking to NHIS, one could expand the number of variables in an analysis. Mr. Machlin noted NHIS variables were for the prior year, not MEPS’ sample year. MEPS was a panel survey utilizing five in-person household interviews gathering data over two years. Each family was interviewed; one respondent answered for all household members. The survey put out person- and family-level data; event and condition files and other units of analysis could be linked. The first panel followed people in that sample 1996-1997. The second panel, begun in 1997, followed people through 1998. An overlapping panel design kicked in beginning 1997: combining in any calendar year information from two panels (e.g., second year for the 1996 panel, first year for the 1997 panel). A huge survey, in broad strokes the household component’s purpose was to aid in estimating annual health care use and expenditures, provide distributional estimates, track changes and trends in insurance coverage and employment, and provide information on those characteristics and changes over time, as well as selected quality indicators. The core interview consists of demographics and various health status measures. It tries to enumerate all health care events of the sampled persons, as well as charges and payments for events.
Mr. Machlin presented an overview of the race/ethnicity revisions. The OMB Directive 15 separated Asian-Pacific Islanders into a minimum of two groups, so Native Hawaiian and other Pacific Islanders could be separately identified. It allowed respondents to identify themselves as multiple-race categories. MEPS were technically in compliance as of 2001. MEPS followed NCHS' lead in their revisions. The old ethnicity questions had a show card asking whether any of the groups represented responders’ origins. The new question asked whether they consider themselves Hispanic or Latino, and where their ancestors came from. The old question had a show card asking which race group best identified the person. The new question asked what race or races one considered oneself to be and offered more choices that stemmed from breaking down in detail the Asian and Pacific Islander groups.
As a subsample of the NHIS, MEPS similarly oversamples Blacks and Hispanics. About 24.3 percent of the 1998 unweighted MEPS sample were Hispanic. When the survey is weighed up to represent the US civilian non-institutionalized population, only 11.7 percent is Hispanic. Similarly, 14.9 percent of the sample is Black, but only 12.6 percent of the country is Black. Hispanics are oversampled at a higher rate than Blacks and disproportionately represented in the sample. In 1997, other policy-relevant subgroups (e.g., children with activity limitations) were also targeted as oversamples. Mr. Machlin pointed out that low income was a category and there was potential for the sample size of minority groups that were more likely to be low income to swell. From 1996 to 1999, MEPS samples included 3,000-5,000 Blacks and 4,600-7,5000 Hispanics. The Mexican American sample was the largest Hispanic subgroup. There were 500-700 Puerto Ricans and a couple hundred Cubans. Asians and Pacific Islanders counted for 500-800 cases. American Indians for 100-200 cases. No figures are put out for a gross national estimate unless there are at least 100 cases in a group. From 2002 forward, there will be a larger overall survey sample of 15,000 households, increasing sample sizes. A targeted oversample of Asians and Pacific Islanders is estimated to result in about 9,000 Hispanics, 5,800 Black, non-Hispanics, and about nearly 1,000 Asian-Pacific Islanders. Current sample sizes were large enough to make estimates for the major race/ethnic groups as a whole but Mr. Machlin noted one ran into cell size problems breaking down some smaller groups by many detailed variables.
Mr. Machlin presented examples of variables collected that break down by race/ethnicity. Hispanic males and females had the highest rate of uninsured (about a third) followed by Blacks (23 percent). Hispanics had the highest proportion (17 percent) of children without a usual source of care in 1996, followed by Blacks (13 percent), and Whites (about half that for Blacks). Mr. Machlin noted the average amount spent for medical care by race/ethnicity in 1996, a function of whether somebody was insured, their health status, utilization, and other factors. The highest was the average-per-person spent on Whites ($2,200 in 1996); the lowest group was Asian-Pacific Islanders. The percent of all medical expenditures by race/ethnicity paid out-of-pocket (often used as a measure of burden) for Whites and Asians was about 1-in-5. For Blacks it was about 10 percent; for Hispanics, 15 percent. About half of Blacks’ medical expenditures were for inpatient services, Asian-Pacific Islanders were one-quarter, White and Hispanic groups fell in between.
MEPS could be used to do in-depth multi-variate analyses. Mr. Machlin gave three examples. Weigers and Taylor looked at Hispanic and different subgroups with respect to insurance coverage. Weinick and Zuvekas did an analysis of the likelihood of an ambulatory care visit or usual source of care, comparing Whites to Hispanics, and Whites to Blacks, and found less-than-half the difference between race/ethnicity groups was due to differences in insurance and income. Monheit and Vistness’s paper looked at trends between 1987 and 1996 in insurance coverage for different race/ethnic groups and found that Hispanics tended to have different patterns.
Mr. Machlin noted other analytic and contextual variables along the lines Dr. Mays mentioned. In 1996-2001 there is information on whether the interview was conducted in English or Spanish. When access problems were reported, respondents were asked if it was due to a language barrier with the provider. For 2002 and beyond, MEPS asked the language spoken at home, whether the sample person was comfortable conversing in English, whether he or she was foreign-born, and how long the individual had been in the U.S. Other variables that could potentially be of interest for analysis of disparities included: income, poverty status, various access to care measures, provider characteristics, and quality indicators such as the receipt of preventive services, and care for selected high-priority chronic conditions. Mr. Machlin also noted there was interest in geographic type analyses, detailed beyond region. He noted that the smaller geographic unit got into areas of confidentiality, particularly when linking; a data center was set up at AHRQ so people could work onsite to link to confidential data. He also remarked that, for some years, MEPS had enabled the user to link to the Area Resource File and other secondary databases.
Medical Expenditure Panel Survey User
Dr. Lillie-Blanton began with two caveats: she brought a single user’s perspective on MEPS’s value and use, but as a member of the agency’s Advisory Council, she wasn’t totally unbiased. She said the key issue was whether MEPS was useful for informing an understanding of racial/ethnic differences in health needs and healthcare use, and whether it was helpful in shaping policy/provider/patient interventions to eliminate disparities in needed care. However, she added that it was important to view MEPS in the context of a single data source for AHRQ and one of many of HSS’s data sources. Evaluating usefulness, one had to look beyond a single dataset at the broader context of its linkages to other data sources produced by the Department: NCHS, CMS, and SAMHSA. Dr. Lillie-Blanton mentioned four indicators she said were important in answering the Subcommittee’s questions: (1) the quality of the data source (which she recognized as the focus of the day’s hearings), (2) the research agenda for analysis of that data, (3) whether researchers had interest and skill in using the data (because that shaped and influenced the agenda and what could be learned), and (4) issues of translation and dissemination--the extent information was put in the hands of those who could ultimately improve health care use and health outcomes.
Dr. Lillie-Blanton discussed her experience as a single user involved in an analysis of MEPS that looked at site of medical care and whether racial and ethnic differences persist. Two key questions were studied: whether these differences in where medical care is obtained persist and if they were due to varying patterns of insurance coverage among Whites, African Americans, and Latinos. MEPS was used to look at differences between children and adults age 18-64. The study focused only on those with a regular site of care (office-based provider or hospital-based provider).
Dr. Lillie-Blanton said MEPS sample sizes were more than adequate for both bi- and multi-variate analyses. The study looked at the role of insurance and whether it made a difference in the usual source of care. Noting they knew that socioeconomic position and other variables affected where people got care, Dr. Lillie-Blanton reported that with MEPS they could control insurance and broader socioeconomic factors. An important asset of the dataset was it allowed one to look broadly and contextually at factors that influenced health and health care use. Usual source or site of care was defined, in this case, to include the emergency department, because of the perception that it was likely to be a minority population’s usual source of care. In fact, she reported the proportion was minuscule. African Americans and Latinos, regardless of insurance, were more likely to have a hospital-based provider as their site of care in 1996. Dr. Lillie-Blanton said they needed to look at incentives and disincentives for where people got care and why minority populations were more likely to use a hospital-based provider as a usual source of care. She noted the need for more research to assess any consequences (differences in the patient-provider relationship) for having a hospital-based provider as the usual source of care.
She said the study demonstrated how MEPS could raise questions to pursue further. She emphasized that the most important thing about this dataset was it allowed one to look in-depth and determine whether race still related to where people got their care, understanding that the proportion of the population with insurance coverage (even if it was public coverage) had increased over the last 20 years. MEPS were an important dataset that allowed looking at race and contextual variables as well as outcome.
Dr. Lillie-Blanton assessed MEPS as a good data source in terms of data quality and understanding health care disparities. She said the self-reporting provided fairly good estimates for at least four of the five major population groups, with limitations for Asian populations and serious limitations for American Indian/Alaska Natives. There was room for improvement on some contextual variables and on issues of fairness and trust that affected health care.
Dr. Lillie-Blanton gauged the analysis and research agenda as good-to-fair. With the analysis she presented and other research, she said the data elements were there to ask and answer complex questions. MEPS allowed them to look over time and make comparisons to previous data sources. However, she noted there was room for growth and improvement in terms of linkage to other datasets and collaboration with other information sources. She emphasized that the NHIS sampling frame allowed them a broader, more in-depth look at health. But she noted there were other links and work to be done in terms of CMS, SAMHSA, and other agency data sources.
She expressed concern that the agency, in using MEPS, still presented data that wasn’t stratified by contextual variables known to influence the ability to recognize racial differences. They knew that both insurance and socioeconomic status affected comparisons. Noting they didn’t present data that wasn’t age-adjusted, she stressed that continuing to compare population groups by race without a measure of socioeconomic position didn't allow an understanding of whether they were looking at a difference by race or a function of another measure.
Dr. Lillie-Blanton noted the 2001 Health U.S. utilized the health care variables in NHIS data that had a control for poverty status for each of the measures of health and health care. She said MEPS also had that capacity and needed to move that way. Comparisons by race were important; they needed to know whether there were racial differences. But Dr. Lillie-Blanton said the question posed to the nation was why--If the answer was a function of race and other factors, agencies doing the beginning analysis had to help point in those directions.
Dr. Lillie-Blanton noted MEPS fair-to-poor in terms of cultivating researchers who could analyze the data and answer questions about the extent to which racial-ethnic groups varied in health care use. Noting that only five out of 51 publications on the AHRQ Web site looked at racial and ethnic differences in the dataset, she said a better job could be done of encouraging researchers in general to use that dataset to provide answers everyone needed in terms of health and health care use. Dr. Lillie-Blanton emphasized that, while to some extent she had been critical, she viewed MEPS as one of the most comprehensive data sources available in HHS to answer questions about health, health care use, and health expenditures. Its weaknesses weren’t specific to MEPS, but reflected Department-wide limitations. Dr. Lillie-Blanton recommended two documents to the Committee: the Commonwealth Fund report that looked at HHS data and a Medical Care Research and Review supplement Kaiser Foundation helped fund. Within the supplement were articles by Weinick and Monheit that used MEPS and other data sources, including the National Survey of American Families and a Kaiser Foundation survey. Dr. Lillie-Blanton said both sources were useful in answering whether the data sources were important in informing policy, practice, and provider behavior.
Questions and Answers: MEPS
Mr. Handler said a sore point was the multi-race question included in the 2000 Census. Some 2.5 million people identified as American Indian only, but 4.1 million people identified one race only plus American Indian/Alaska Native. He asked what the presenters did with this multi-racial data. Dr. Lillie-Blanton said she hadn't begun any detailed analyses using that category, but they had produced a report on urban Indian health using the new Census data and presented in separate columns persons who identified by single racial category and those who identified by a single category or in combination. She agreed it was important not to lose individuals who identified themselves with that race/ethnicity combination. Noting that group was small for African Americans (while with Native Americans it almost doubled), she suggested doing only one analysis among African Americans that included anyone who identified that as part of their origin. She said it was likely most would have experienced being a separate race; with other racial/ethnic groups that assumption probably wasn’t possible. She acknowledged that these were complicated issues, but she said she fully supported people's rights to not be defined by a single racial category. Mr. Machlin said he didn't know how public use files would be structured with respect to the multiple race issue. Ideally, individual researchers would have all the details and options and use their judgment. But confidentiality issues were part of what drove agency decisions about publishing. If a rare combination of races could potentially lead to identification of a respondent, details would be suppressed in a public use file. Mr. Handler noted that, before the Census, activists told different racial and ethnic communities not to identify with more than one race because their data would be lost--He said he didn't want them to be right.
Dr. Newacheck said MEPS was valuable for looking at things like usual source of care and insurance coverage, and the only place to get expenditure data in a population-based survey sample. An expenditure estimate tended to have high standard errors relative to a point estimate for usual source of care; the amount of variation person to person skewed distribution. But looking at racial and ethnic minority groups, particularly in subpopulations, they ran into problems quickly. If they looked at all children, they could get reasonable estimates of broad racial and ethnic categories of Black and White, Other, but when they looked for adolescents it fell apart--even with oversamples and two years. He asked what the trade-off would be in shifting more non-Hispanic Whites to one of these minority groups. He suggested different analytic techniques could be used to deal with the high variance problem. They could take out outliers from the top or use smoothing techniques, taking the logs of all expenditures and not reporting the actual expenditures. He emphasized that this was a big issue for the user community. He saw three or four analogies trying to use MEPS and had to throw half away.
Mr. Machlin suggested combining data across panels for some of the smaller groups. It wouldn’t be a panacea; the data were highly skewed doing analyses of expenditures for some small subgroups. A certain percentage of the population had no expenditures, a fair number had modest expenditures--And then someone very sick, had a lot of hospitalizations and a million dollars in expenses, and the average shot to the 80th percentile. Analysts needed to be aware of that and make judgments about what they could and couldn’t do. Every situation differed. Trying to up sample size, combining years, using transfer information techniques, and logging were all useful. But one outlier, even in a big group, could throw MEPS off.
Dr. Newacheck said the user community needed guidance about how to deal with these unexplainable results. He suggested an agency publication that dealt with these issues of small subgroup analysis and race/ethnicity, and provided guidance on different analytic approaches could bring consistency across users. He encouraged the agency to create a more uniform framework and a way of thinking about how to deal with this problem.
Consumer Assessment of Health Plans
Dr. Sangl said CAHPS relied on the importance of the consumer perspective. One of its goals was to develop a set of surveys to measure consumers' reports and ratings of care across all fee-for-service and managed care systems. CAHPS was to develop reports to consumers on results of the surveys, evaluate the process and outcome of application of the reports. It also was to make these products available to purchasers, plans, and providers. The CAHPS survey was a consortium. Products are developed through collaboration of AHRQ, CMS, Harvard, RAND, and RTI. Technical assistance is provided to the users through SUN to AHRQ and Westat.
Dr. Sangl noted the survey was set up differently. There are adult and child cores (both 46 items). Supplemental items (e.g., language, interpreter access) have been developed and tested. A chronic conditions supplemental item set is used mostly in the Medicare population. CAHPS has been translated into Spanish, Russian, Vietnamese, Mandarin, Korean, and Cambodian.
Unlike many surveys, sponsors (not AHRQ) conduct the CAHPS survey. The Medicare program conducts three surveys for: the managed care population, fee-for-service, and disenrollment. Medicaid state and SCHIP programs conduct Medicaid surveys. Depending on the year, up to 35 states survey Medicaid populations. About half-a-dozen states conduct the State Children's Health Insurance Program. Dr. Sangl reported there were also commercial sponsors. The Office of Personnel Management requires all plans that contract with the Federal Employees Health Insurance Program to survey annually. NCQA requires plans seeking accreditation to conduct the survey. DOD is another major sponsor. In total, about 90 million Americans are covered under plans for which CAHPS data is collected.
Race/ethnicity is self-reported from five categories: White, Black or African American, Asian, native Hawaiian or Pacific Islander, and American Indian or Alaska native. A second question is on Hispanic or Latino origin or descent. Respondents can choose multiple categories. CAHPS’ set of five core composites had 17 items. Getting needed care, getting care quickly, and communication each had four items. Helpfulness and courtesy of office staff had two items. Customer service had three. In addition there are four overall ratings: personal doctor, specialist, health care in general, and health plan. Dr. Sangl noted that sponsors could independently choose supplemental items and some added their own. Supplemental items include items on the availability of the interpreter, language spoken at home, and if one is able to access an interpreter. Additional measures can be constructed from the other items in the core set such as percentage: having a personal doctor, seeing a specialist who felt they needed one, and having an office visit in the last year. Dr. Sangl pointed out that one could also look at differences in gaining access to care and in patient experience as well as more traditional measures of access. Did certain racial groups have more trouble communicating with their provider or more difficulty getting referrals to specialists?
Dr. Sangl noted confounding variables to consider when looking at CAHPS data, and CAHPS’ data limitations. Most the time, respondents had to have at least one visit to provide some of the ratings; one didn’t rate a provider if there hasn’t been an annual office visit. While it might be possible to oversample, often the plan’s racial/ethnic composition was unknown. Education and health status tend to be related to assessment of care; those with lower education or better health status rate plans higher. (Both are often used as case mix adjustors in CAHPS data.) Potential response tendencies are related to racial/ethnic group membership (e.g., anecdotal evidence suggests Asians’ overall ratings didn't give the highest scores.) Income might also affect assessments. New Jersey added income as a supplemental item to a CAHPS survey of their commercial population and those with lower income rated higher. AHRQ will study cultural comparability in the next phase of CAHPS.
Noting that plan characteristics could be linked with the CAHPS data, Dr. Sangl discussed research on both commercial and Medicare data found 20-point plan-to-plan variations between Whites and Blacks or Whites and Hispanics having an annual office visit or a personal doctor. Dr. Sangl said it would be useful to study what those plans were doing to minimize disparities and what occurred in plans with huge disparities. Dr. Sangl noted HEDIS data was available on about 350 plans that report to NCQA. One could get aggregate clinical quality data and compare it with the consumer assessments of available plans.
Reiterating that CAHPS was formed with a network of sponsors, Dr. Sangl said they’d created NCBD, which was available to researchers, had adult survey data available for 793 plans and child survey data for 148 plans. There were 57 sponsors, over 350,000 adult responses and 44,000 children responses (with an adult respondent for the child). Information was available at www.cahp-sun.org.
Dr. Sangl noted other types of analyses were now possible because selected CAHPS items would be included in the MEPS survey. One could contrast usual use expenditure measures with consumer assessment measures linked with the national representative sample. She said they would go below the plan level in the next phase of CAHPS to include: work on group practice and individual provider level CAHPS as well as different or special populations; children with special health care needs; the Experience of Care and Health Outcomes (ECHO), which is targeted towards behavioral health populations; and an early design phase of a CAHPS for nursing home residents.
Consumer Assessment of Health Plans, User
Mr. Moser said the MMC version of CAHPS was created to obtain information from enrollees about their plans, providers and self-reported overall health status, health conditions, and health care system utilization in the last six months. Surveys were conducted annually since 1997. Mr. Moser noted CAHPS surveys provided a unique opportunity to investigate certain racial/ethnic groups (e.g., American Indians/Alaska Natives, Native Hawaiians/other Pacific Islanders) not picked up in other studies. As Dr. Sangl had explained, two multiple-choice questions were used to classify individuals by race and ethnicity. Mr. Moser said about 98 percent of respondents identified by one race. About 92 percent gave a usable response about Hispanic or Latino origin or descent. He noted 90 percent of Hispanic Latinos chose a single race; so little was lost with the question. Only 200-300 individuals responded in the smaller racial/ethnic groups in any year. Combining three survey years (1997-1999) together built up the sample size, so some cross-group comparisons could be done.
Non-White groups accounted for about 14 percent of Medicaid managed care enrollees. Female enrollees outnumbered males for every group except American Indians and Alaska Natives. The greatest disparity is for Native Hawaiians and Pacific Islanders; almost two-thirds are female. The youngest age group are American Indians/Alaska Natives: 15 percent of enrollees are in the youngest group and 12 percent in the oldest. Native Hawaiians and Pacific Islanders are the oldest group. Whites have the highest high school graduation rates; Asians have the highest college rates. About half the Blacks, American Indians, and Hispanic Latinos complete high school.
Blacks, Hispanic/Latinos, and American Indians/Alaska Natives report worse health currently. Native Hawaiians and Pacific Islanders report the greatest improvement in health compared with the previous 12 months. American Indians/Alaska Natives report the worst changes. Heart disease is a fairly serious condition for all groups. Diabetes is a particular concern to Blacks, American Indians/Alaska Natives, and Hispanic/ Latinos. It is also the most prevalent condition for Asians. COPD and stroke are less serious than other serious conditions, but he noted racial/ethnic differences.
Mr. Moser said Asians and Native Pacific/Pacific Islander MMC enrollees made the fewest visits to doctors' offices. Blacks and American Indians/Alaska Natives had the highest frequencies (five or more doctor visits). Whites were most likely to visit doctors and specialists and use prescription medicines. American Indians/Alaska Natives had high (and Asians low) rates of hospital and emergency room usage. Asians generally had the lowest use of medical services. American Indians/Alaska Natives had the highest rates of use of most medical facilities and services (e.g., special medical equipment, special therapy use, and home health care use).
Current smoking is most prevalent among American Indians/Alaska Natives and least prevalent among Asians, particularly female Asians. Asians and Hispanic/Latinos are most likely to never have smoked. Females are less likely to be current or former smokers. Former smokers are more likely to have heart disease than nonsmokers and current smokers. (One interpretation is some smokers who get heart disease are frightened into quitting.) Whites of all smoking behaviors tend to be more likely to have heart disease. Differences between racial/ethnic groups and between males and females being advised by a doctor to quit smoking weren’t statistically significant. Mr. Moser noted concern that only half of all smokers were advised by a doctor to quit in the last six months and those only received cessation counseling on half their doctor visits. Asians and Native Hawaiians/Pacific Islanders had the highest quit rates. Blacks and American Indians/Alaska Natives were least successful. Male smokers were generally more successful in quitting, but differences were small and not statistically significant. Overall, about 75 percent of smokers indicate success in quitting.
Mr. Moser noted differences as well as similarities between the MMC subgroups. Asians tended to have the best self-reported overall health and the lowest health care utilization levels. Whites reported good health, but are above average health care utilizers. American Indians/Alaska Natives, followed by Blacks or African Americans have the poorest health. Hispanic Latinos tend to be average in terms of self-reported health status and utilization of health care services. Females MMX enrollees reported worse health, however they also reported lower levels of the five serious health conditions. Smokers were in worse health than non-smokers.
Questions and Answers: CAHPS
Noting maps seemed to indicate the states implemented different portions of the survey or dealt with different populations, Dr. Lengerich asked how the survey was set up and drawn. Mr. Moser explained that CAHPS was in the public domain. CMS administered the managed care, fee-for-service version used by Medicaid programs in most states. Participants noted the MEPS data collection on selected CAHPS items would be a nationally representative sample; others might not be. Dr. Sangl suggested thinking about CAHPS as populations: the Medicare population, the Medicaid population the states could choose to administer, and the commercial population. She pointed out considerations if one wanted to pool. Medicare had a similar benefit level and one could combine across it; one might choose not to combine for Medicaid because some states added optional benefits. Commercial could be all over the map.
Dr. Coleman-Miller asked if it was clear that everyone could read and there hadn’t been language issues with MMC CAHPS. Mr. Darby said English and Spanish versions were used with the Medicare population. California Health Care Foundation sponsored translation of CAHPS into Spanish, Russian, Vietnamese, Mandarin, Korean, and Cambodian. Mr. Moser noted proxy respondents could translate if language was an issue. He clarified that MMC was a mail survey with telephone follow-up.
Dr. Coleman-Miller noted many patients who said they didn’t smoke chewed and had the same issues. Respondents reported their doctor hadn’t advised them not to smoke--but were they asked if they’d seen the doctor and said they smoked? The cultural dimension was to answer what you wanted to hear. Dr. Coleman-Miller said they had to ask questions that covered that.
Dr. Coleman-Miller asked what role the Census and OMB had in developing the Hispanic and non-Hispanic terms. Mr. Moser explained that they’d adopted the OMB regulations and wording as prescribed, because it would be used and sponsored by a government agency.
Ms. Coltin said she was curious, particularly in the MMC CAHPS, about linking survey data with other CMS data to look at whether there were differences in response rates by different racial and ethnic groups. They knew the overall rate was high for the Medicare population, but not how it varied or if a more representative membership response in certain health plans might affect results. Mr. Moser said he didn’t know the response rates, but the data could be linked to other datasets. In other work they knew people’s health plans and linked with data on the plans (e.g., benefit coverage, co-payments, co-insurance). They hadn't tried to link to any datasets that had race/ethnicity data, but potentially they could link to those and socioeconomic variables.
Ms. Coltin suggested that geocoding data on the commercial population could be a way to look at the impact of race/ethnicity on the representativeness of the samples. Dr. Sangl said there was a proposal in the draft Medicaid for the states to provide the race/ethnicity of enrollees, so one could work backwards and know the composition of the plans.
Ms. Amy Heller, Project Officer, Managed Care Task Survey (MCTS) said that for the past three years they’d had better than an 80 percent response rate that varied slightly by race and ethnicity. CMS didn't offer the plan in Chinese, so they had no response from the Chinese Health Plan. She noted a lot of work was supplementing all that Mr. Moser and others were doing. MCTS was looking at non-response analysis to see if it varied by race and ethnicity. And Harvard was doing a study of regional differences, because that was a case mix adjustor for them. MCTS was also moving towards small area estimation versus regional estimation.
Dr. Coleman-Miller noted these categories weren’t used in many countries and a foreign born person would know different meanings. Mr. Moser said there wasn’t any data on whether respondents were foreign born or how long they’d been in the US. Dr. Newacheck noted they’d heard a lot about response rates and potential for bias. There was also the issue of which plans participated in the national data pool, at least in the commercial population. He asked which of the three CAHPS survey samples would be useful for racial and ethnical analyses and which they wouldn’t want because of potential biases. Mr. Darby said it varied state-by-state with Medicaid and one had to look at each commercial sponsor. Response rates were probably lowest in Medicaid, but there were notable exceptions. A number of samples probably couldn’t be analyzed on racial and ethnical subgroups. The MEPS survey included about 15 CAHPS items that should be safe to do the analyses. Dr. Sangl reported two parallel analyses, commercial and Medicare. Medicare had over an 85 percent response rate; NVCD 45-50 percent--But the results were similar. Dr. Sangl expressed confidence with both, but said if one looked at one alone, particularly Medicaid, one might need to be more cautious.
Ms. Elizabeth Arias, Mortality Statistics Branch, NCHS, asked if respondents stated country of Hispanic origin or were grouped as Hispanic or Latino. She asked if the instructions defined what it meant to be Hispanic or Latino, as the Census form did (Hispanic meant born in Cuba, Mexico, Puerto Rico, South America, or Spanish-speaking Caribbean). Mr. Darby said they weren’t asked about country of origin and the mail survey had no special instructions.
Ms. Arias pointed out that for diabetes and other conditions they’d noted, there were differences between Hispanic subgroups. Similarly, smoking prevalence was different between Mexican Americans and Cubans. She asked if AHRQ might consider changing the question format in future surveys. Mr. Darby said they’d consider that. Because AHRQ gave the survey to sponsors to use, they’d tried to keep the core (currently 46 items) as short as possible. Probably, they’d develop a supplemental item and recommend people include it.
Mr. Handler asked if they’d ever asked why smokers quit; if they knew why, techniques could be applied to help others. Mr. Darby clarified that the smoking items weren’t part of the CAHPS core. They were added to the MMC version and the version NCQA used with the HEDIS supplement. He also pointed out that the data Mr. Moser provided didn’t cover other items in CAHPS, and he reiterated that the core idea was to measure quality of health care from the consumer's perspective. Some of the items on access, communication with the doctor, being treated with dignity and respect were items in the core CAHPS.
Mr. Darby said it would take more evidence to get the expanded race and ethnicity questions on the core. Differences in the way people reported care would be strong evidence suggesting an item should be added. Noting there probably was evidence from multiple sources, Dr. Mays said it was good to hear somebody only needed to write a letter, sending all the published evidence. Recalling Dr. Coleman-Miller's comment about racial/ethnic differences in the question itself, she asked if any methodological work was underway to highlight cultural differences in perceptions on the questions. Mr. Moser said he wasn’t aware of any. Dr. Coleman-Miller asked if the questions went through a cultural filter to confirm that they were asked correctly. Ms. Coltin noted that, as Mr. Darby had pointed out, these items were part of the NCQA HEDIS supplement, not the core text. They were adopted from the Behavioral Risk Factor Surveillance Survey and worded exactly the same. Considerable research had been done on the performance of items in that survey. Mr. Darby added that the point about designing items for different cultures was excellent. He remarked that survey researchers were probably the worst people to guess what respondents might be thinking when the questionnaire was administered. CAHPS had conducted some 450 cognitive interviews (in English and Spanish) on this questionnaire and continued to do more. CAHPS was also funding grantees for five more years to continue development work, including testing cultural comparability. Equivalence of language was an issue and they had a standard protocol for translation. And there were the cultural differences in how people viewed this; Dr. Sangl had pointed out the response tendencies. It appeared that Asians might be harsher raters. It was important to know what the respondent was thinking when asked the questions; AHRQ would study the area carefully. Mr. Darby clarified that there were translations in Mandarin, Vietnamese, Russian, and Cambodian. The next round of CAHPS would be translated into other languages. They were considering more than one Spanish version.
Medicare Current Beneficiary Survey
Mr. Waldo noted MCBS is a survey of a nationally representative sample of people enrolled in Medicare. Conducted by Westat, Inc. for CMS, MCBS’ principle goal is understanding how Medicare serves its target population. Field work began in 1991. The MCBS is a list sample. SPs are drawn from SSA’s master enrollment list each year in January. The panel to be interviewed, based on January eligibility, is selected by April. Interviewers go into the field beginning September. People who age into Medicare during the year aren’t captured. The youngest of the elderly population interviewed will be 66 years old. The sample is drawn to over-represent enrollees under age 65 (eligible based on disability) and those 80 and older (believed frailest and most at-risk for high healthcare expenses). Respondents are clustered in some 100 geographic PSUs to minimize field costs. PSUs are drawn to produce nationally representative results. SPs remain in MCBS for three years, plus the initial and closing interviews. Three interviews are conducted each year using computer assister personal interview (CAPI). Information is collected on: demographics, social/economic characteristics, health status and functioning, use of health services, financing of health services (the original genesis of MCBS), and interactions with Medicare programs. SPs self-identify race and ethnicity during initial interviews. MCBS doesn’t rely on race and ethnicity information contained in the master enrollment list, which is based on information identified at the time individuals applied for social security. A decision was made early to over-represent on the basis of age, rather than race/ethnicity, in order to capture populations in need of particular services.
Prior to 1998, CMS followed OMB protocols for racial identification, asking questions in MCBS identifying race and Hispanic origin. Respondents were shown a card and asked to identify themselves as: American Indian, Asian or Pacific Islander, Black or African-American, White, or “Other” race--specifying that race. Beginning with the 1998 survey, the question’s wording remained the same, but all respondents (not just those in initial interviews) were offered new choices in compliance with OMB Directive 15: American Indian or Alaskan Native, Asian, Black or African-American, Native Hawaiian or other Pacific Islander, White, or “Other” race--and asked to specify.
Mr. Waldo noted there was a substantial amount of stability in racial identification between 1997 and 1998 except for those who self-identified as American Indians. Some 113 people who self-identified with a single race category in 1997 identified themselves as more than one race in 1998. American Indian respondents sorted themselves out differently and were most likely to identify with more than one race. A substantial number of people who in 1997 identified as American Indian changed their designations in 1998; most self-identified as American Indian and White or as American Indian and Black or African-American. A single question was initially asked about ethnicity: “Are you or is the sample person of Hispanic origin? Yes or no.” In 1998, “Hispanic” was changed to “Latino.” Nobody changed his or her response in 1998.
Mr. Waldo said a lot could be done using the MCBS to study race/ethnicity disparities. He said could be confident running analyses that involved a single dimension or looking at common events among the target population. He cautioned against trying to look at rare events or break the population into too many cells. Mr. Waldo said one could look at the number of male beneficiaries who had blood test or digital exam for prostate cancer or the proportion of beneficiaries with a Pap smear or mammogram among White Non-Hispanic, Black Non-Hispanic and Hispanic populations with a fair amount of confidence, but there wasn’t sufficient sample size for the other racial identifications. It was possible to look at racial and ethnic differences in common conditions such as hypertension, diabetes, and osteoporosis. They could also look at things of interest to program administrators: position user rates or per capita Medicare expenditures by residents and by race, community-based expenditures and facility care for the three largest racial groups. And they could look at characteristics of the population duly eligible for Medicare and Medicaid and observe that racial and ethnic minorities represented a disproportionate share. But Mr. Waldo noted difficulties in trying to look at race/ethnic differences with other categories attached (e.g., looking at the number of sample persons who had ever been told by a doctor that they had hypertension, drawing from the 2000 Access to Care file, considering age and racial identify). White Non-Hispanic and Black Non-Hispanic, for most age groups, had a robust enough sample size. The Hispanic population was about as robust, except for the population under age 44. Only the Black African American and White populations had sufficient sample size to allow an age break in a discussion of hypertension. There weren’t any race/ethnic categories with enough sample size to enable a break by age of the number of beneficiaries told by their physician that they had lung cancer. For that kind of analysis, he said one had to collapse the age or race categories to get to a decent sample size. Trying to look at functional limitations by age and race, aside from the 65-74-year-old population, there weren't many cells populated enough to allow robust calculation and for racial identifications other than White, Black, or Hispanic there were almost no cells with enough population for this kind of analysis.
Medicare Current Beneficiary Survey User
Dr. DaVanzo presented a study the Lewin Group did that demonstrated MCBS could be utilized to identify differences. Using a 1997 supplement on information needs, Lewin looked at a combination of needs and preventive health behaviors. HCFA had been engaged in the Educate Beneficiaries Initiative and the survey gauged beneficiaries’ understanding of the Medicare program, providing baseline for HCFA. The Group looked at that understanding in relation to the subgroups and preventive health behavior (i.e., the flu shot, which was most prevalent in the data). The percentage of population receiving the shot varied by subgroup. Subjects self-reported how much they knew (a lot, a little, almost nothing) about six situations. Actual knowledge was measured with true or false questions, including “Does Medicare pay for a flu shot?” Many HCFA materials were written at reading levels beyond beneficiaries’ ability to read, and HCFA also was interested in rural and low-literacy. The study confirmed that many beneficiaries lacked an understanding of Medicare and the biggest knowledge gap was about managed care.
The focus centered on the group that knew least and that HCFA most needed to educate. The three largest groups (White non-Hispanic, Black non-Hispanic, and Hispanic) were collapsed and broken down by gender, race, and knowledge. Other groups were too small. Dr. DaVanzo said the dataset was extremely useful for looking at something this broad with only two dimensions. The data indicated that African-Americans in 1997 weren’t receiving flu shots for a variety of reasons. African-American beneficiaries were 21 percent less likely to receive a flu shot than White non-Hispanics. If you knew that Medicare covered the flu shot, you were 32 percent more likely to receive it.
Once the different factors were teased out, some materials were created specifically for African-Americans. Questions asked, if you didn't get a flu shot, why not? A variety of choices, all tied to knowledge, included: “My doctor didn't tell me to,” “I might get sick,” “I might get side effects,” “I didn't think I needed it.” Materials created for African-American beneficiaries explained about the flu shot and that it didn't make anyone sick. Dr. DaVanzo emphasized that knowing what the thrust of the materials should be came from identifying the combined action of ethnicity, behavior, and knowledge. In order to educate beneficiaries, they’d had to find out more about them. Dr. DaVanzo noted there was detailed information on service utilization and pharmaceutical useful in using MCBS to measure broad-level disparities. The depth of the information was tremendous and included supplements about HMO membership, insurance status, and functionality. Dr. DaVanzo said a lot of analysis could be done in the three broad groups, but MCBS wasn’t appropriate for studying specific subgroups or making precise population estimates.
Questions and Answers: MCBS
Noting MCBS was a sample for about 30-35 million Americans, Dr. Breen questioned if the sample size shouldn’t be bigger than 18,000. She asked about pooling several years of data to gain information on smaller populations and if there were plans for over sampling. And she expressed concern that people with Medicare had paid into social security or were dependents of people who had been employed; the worst-off population wasn’t represented. Mr. Waldo replied that, because the Medicare population essentially was those eligible for social security, over time that proportion that were race/ethnic minorities could be expected to expand. CMS didn’t plan to oversample for race/ethnic groups. That was a budget issue. The survey ran on a large shoestring--but a shoestring, nonetheless. (The total research budget in FY02 was $55 million, of which $12 million ran MCBS. The FY03 proposal called for a budget of $24 million, with $12 million running MCBS.) Mr. Waldo said pooling across years was quite possible and facilitated by the structure of the files. He noted one had to be a bit wary, as beneficiaries stayed in for three years. Combining all observations for 1997 and 1998, many respondents would be the same and the total would be far less than 40,000 observations. He suggested that taking a longer view and assuming little temporal shift could get them as far back as 1991.
Dr. Newacheck expressed surprise that MCBS didn't oversample. NHIS, Medical Expenditure Panels and most large health surveys oversampled. At least by race, it would be a relatively modest incremental cost: the administrative files had the data people indicated. And they didn't necessarily have to add additional minority populations; they could reduce Whites and substitute more minority groups, as MEPS did. MCBS was a big public expenditure and race/ethnic minority issues and disparities were high on the agenda. As Mr. Waldo had pointed out, MCBS didn't have the data for some important, although not common, health indicators. Mr. Waldo said CMS had the capability to do that. Some 1,100 respondents in the calendar year 2000 access file self-identified as Hispanic origin. The administrative files showed 400 people identified as Hispanic. He said they could oversample those who appeared on the administrative records as Hispanic, but it wasn’t clear whether that oversample represented the population identified as Hispanic or Latino, or how a correction could be made later. Another caveat about over sampling was that each time an oversample population was created the power of the survey diminished. A goal of MCBS was to be able to look at rare events in the population. They’d already limited that extent; over sampling additional populations without increasing the overall sample size would further diminish ability to detect rare events.
Dr. Heurtin-Roberts asked Dr. DaVanzo if the percentage differences of knowledge by race/ethnicity had been tested for statistical significance. Dr. DaVanzo said there were almost 800 Hispanics in the sample. At .05, they’d stayed relatively high on Hispanic and African-American. Ms. Coltin asked if the self-reported race/ethnicity data from the MCBS sample had been compared with the data on the master enrollment list and about patterns of agreement or disagreement. Mr. Waldo reiterated on what seemed to be a wide difference between the administrative and self-reported Hispanic population because the way the SSA reported out the data Hispanic was mutually exclusive of other races. A substantial percentage of the Medicare population apparently sorted themselves first as White, then Hispanic--That wasn't captured in the administrative files. Other than that, the populations sorted themselves into broad groups. The biggest difference was in the Hispanic or Latino origin.
Dr. Pokras said she understood about wishing samples were larger, but she pointed out that $12 million already was a substantial sum. She suggested they might learn from the next day’s presentations how they might oversample within the existing budget. Noting HCFA had paid for a questionnaire so people could re-self-identify and improve the data, Dr. Pokras commented on the gap between 1,100 respondents of Hispanic origin and the administrative file’s 400. She asked if CMS succeeded in translating the questionnaires. Mr. Waldo said the questionnaire was written and conducted in English and Spanish. The respondent’s language preference was identified at the first interview. CMS relied on intermediation and the interviewer’s services for other racial groups, which could be a big problem (e.g., they couldn't find enough interviewers in Massachusetts who spoke Russian and Portuguese).
Dr. Coleman-Miller questioned whether they could define what staying healthy looked like and asked how many mixed-population people the Lewin Group had working on disparity. Noting that over sampling created significant disparity and special problems when used with the minority community, Dr. Coleman-Miller expressed an inclination to disregard those special problems so they’d have something in the ballpark. Dr. DaVanzo said about 20 percent of the Lewin Group’s 100-120 person professional staff were African-American and Latino. They had PhDs or were at the master's level and conducted focus groups and collected primary data. When the Group developed a protocol they always tried to have a mixture of contributors on the team. She said the Group hadn't formulated the MCBS, but choices that evolved in later iterations came out of open ends in the interviews. Interviewers transcribed responses to many questions in an “Other” category and often the categories of responses that respondents chose from was greater than what was finally tabbed up. Analysts often rolled responses together. Questions in the MCBS came from two different genres. Some were incorporated off the shelf to make MCBS responses as comparable as possible with other surveys. Agency staff, in consultation, formulated questions related specifically to the Medicare program. Focus groups and cognitive testing of the instrument ensured that the questions went through a set of typical respondents. Dr. Coleman-Miller encouraged the Lewin Group to consider not using old questions. The disparity was huge and they weren’t getting the answers needed. With statistics up above 200 percent difference, they needed questions more relevant to communities.
Dr. Frank Wall, George Washington University School of Public Health, expressed his frustration in listening to the presentations. He said he’d been in the field a long time, working with Asian-American/Pacific Islanders, Native Americans, and immigrant refugees--None of them were found in these statistics. He said where they were was a policy question with implications for everyone in research, and they needed to critically think how to answer the question. This was an expensive survey and, as an advocate and researcher, he couldn’t use it.
Dr. Reuben Warren, Agency for Toxic Substance and Disease Registry, asked if CMSS separated the percentage of the Medicare population that was Medicaid eligible. Mr. Waldo said it wasn’t as straightforward as one might hope (there were a number of classes of Medicaid eligibility among the Medicare population), but they could support both. The various laws starting with the BBA in 1997 and the BBRA in 1998 created groups of beneficiaries. Some Medicare beneficiaries were eligible to have all of their co-payments and premiums paid for by the Medicaid program. Others were only eligible to have their premiums paid; still others were eligible to have a portion of home health co-payment aid. The states had the option of extending full Medicaid coverage to any classes. Until recently, the administrative records didn’t indicate each enrollee’s set of benefits. So MCBS had both administratively and self-identified Medicaid eligibility. A fair amount of time was spent sorting out what that meant.
Considering that both minority populations and their health status as well as Medicaid eligibility and opportunity were increasing over time, Dr. Warren remarked that looking at Medicaid vs. Medicare eligible might be a way to look in the window of race/ethnicity and get at questions of economic status. Mr. Waldo said the problem doing that with the Medicare population was that so many people became Medicaid eligible because they'd been in a nursing home. People of all different SES classes were spending down to Medicaid eligibility. He suggested a more efficacious way would be to use companies like Claritas that developed SES scores for zip codes, associating that with the respondent’s zip code and forming an environmental socio-economic score. Dr. Waldo said CMSS asked beneficiaries in the MCBS for family income and got a lot of item non-response and vagaries. He remarked that people eligible for Medicare talked about incontinence, but questions about income were considered very personal.
Dr. Ruth Perow, Executive Director, Summit Health Institute for Research and Education and co-author of the Commonwealth report, agreed that MCBS was a useful window on the experiences of beneficiaries with the system. It provided information about their use, attitudes, satisfaction, and perspective about the treatment they received (e.g., flu shots, Pap smears, and blood tests for prostate cancer). In order to complete the picture, Dr. Perow advised it would be important to get information from providers about services, treatment, and outcomes. CMS collected data from providers (e.g., sygmonoscopies, and heart treatments, and amputations) that would elucidate the disparities. Mr. Waldo agreed it would be good to figure out how to get information from providers. He noted CMS experimented last year with pharmacies sending in summaries of patient medications to match up with beneficiaries. He added that, from a practical perspective, it would be some time before CMS could do more with other types of providers. But, in a related area, there was a new effort to reach providers, educate them about CMS’s programs, and find out what worked. Mr. Waldo added that MCBS wasn’t the only tool CMS had available to look at issues relating to beneficiaries and their interactions with its programs. A series of peer review organizations, one in each state, contracted with the agency. Many were working on issues specifically about race/ethnic differences in the access and use of services and outcomes.
Dr. Gibbons understood about the administrative dataset being different, but he said he didn’t know of any perfect dataset or foolproof analysis. Mr. Waldo had clearly said what they had wasn’t very good for minorities. Given constraints on budgets and analysis, Dr. Gibbons asked if over sampling wouldn't be better. Recalling that Mr. Waldo commented that over sampling diminished the power of the survey and that a reason for the survey was to detect rare events, Dr. Gibbons asked what populations they’d pick up rare events in if MCBS couldn’t sample minority populations adequately. Dr. Gibbons remarked that, as an African-American, he appreciated that CMS made materials for African-Americans about flu shots. But he noted the data indicated more than half of every population hadn’t received a flu shot. They had a system-wide problem.
Dr. Gibbons said he was waiting to hear Dr. O'Campo's thoughts, but his impression was that a correct understanding of SES couldn’t be conferred by a single fact about income. SES was a merging of three concepts: income, education, and occupational prestige. A significant amount of data showed it wasn't family income but the relative distribution of income. Mr. Waldo said anytime one oversampled, the observation ended up weighted and couldn’t be used one for one. Rare events could be detected with a large sample size; what they couldn’t do was differentiate how that rare event was experienced qualitatively for an African-American, Hispanic--or maybe even the majority population. Mr. Waldo suggested one might assume (or hope) that racial or ethnic origin had no effect on the rare event. Dr. Gibbons said they already knew that wasn’t the case. Mr. Waldo agreed there was no question about the impact on health; he was speaking about rare health events. He said MCBS wouldn’t be the tool to define whether there were race/ethnic components to the rare event. That would be a more clinical study where the sample frame was rare medical events. Dr. Coleman-Miller said they had all the data on many already done. There were many rare events--some 200 percent higher in the racial issues. They could start where they already knew there were racial events.
Dr. Mays said she hoped everyone heard today that people had begun to identify several surveys. She said she’d heard a call for change and questions about what's possible. It might be that what they had wasn’t best, given where they were headed for Healthy People 2010. Dr. Mays said she hoped the survey people realized they were hearing input as to analytic direction as well as individual perspectives. They were all trying to get to the same place in terms of Healthy People 2010 agendas. Many of the targets were based on using these surveys, which was one reason they’d come together today. What they might not have heard was that, sometimes, the surveys presented problems even in terms of the targets.
Policy Perspectives
Dr. Clancy said people in health services research knew there were many non-clinical determinants of health outcomes. Patient, practitioner and hospital or setting characteristics; patients' preferences; and reimbursement all had an impact on health outcomes. And healthcare was only one input to quality of life: personal behavior, education, genetics, public health, and economics also contributed. Although she limited her remarks to heathcare, Dr. Clancy said a big challenge faced collectively was determining how much of the observed disparities in health was amenable to improvements in healthcare services. Based on existing federal data collection, she noted they already knew quite a bit and much of it was fairly depressing.
An analysis of data from NCHS indicated the leading cause of excess death among African-American women was cardiovascular disease, which for African-American men was preceded by trauma, HIV and drugs. AHRQ’s hospital cost and utilization project showed that people who lived in a zip code where the average income was under $25,000 were over three times as likely to be admitted to the hospital for an episode of pneumonia that was, in theory, preventable by appropriate vaccines. Dr. Clancy said race/ethnicity, socio-economic position, and many other factors interacted--Physicians saw this from the beginning of their training. She said for some time the question had been to what extent the healthcare system owned part of this problem or did clinicians and providers just find themselves trying to deal with what was presented.
Dr. Clancy noted studies clearly demonstrating significant disparities in care associated with race. The study using the predictors of referral for cardiac catheterization published in the New England Journal in 1999 by Kevin Schuman and colleagues showed physicians were less likely to recommend invasive diagnostic and therapeutic procedures for older African-American woman. A year later, another study in the same journal by an AHRQ’s investigator looking at reperfusion therapy and Medicare beneficiaries who’d just had a heart attack again demonstrated that Black women were significantly less likely to receive treatment. Dr. Clancy remarked that healthcare wasn’t doing so great overall if 59 percent of eligible patients receiving an evidence-based lifesaving treatment was the Everest of clinicians’ ambitions.
Dr. Clancy drew upon a framework John Eisenberg developed looking at voltage drops to quality of care. Possible drop offs between a population-at-large, a healthcare system presumably there to serve, and everyone getting quality of care included: (1) available insurance (Do you have the option?), (2) enrolled in insurance (Can you afford it?), (3) providers and services covered, (4) informed choices available, (5) primary care accessible, (6) referral services. Dr. Clancy clarified that she didn’t know this was the only framework, but it helped her think about all the possible steps in a cascade between where they'd like to be and where they were now. Even if income and health insurance coverage were equalized, differences in access and use of health services wouldn’t be eliminated. One-half to three-quarters of the disparities couldn’t be explained by these factors. It was difficult to identify a single factor that would resolve racial/ethnic disparities. Dr Clancy noted this wasn’t an academic proposition for AHRQ. Congress charged them to begin producing in fiscal year 2003 an annual report on prevailing disparities in healthcare delivery as it related to racial and socioeconomic factors in priority populations. In about a year, AHRQ had to produce a public report on people living in rural and inner-city areas, low-income and minority groups, women, children, the elderly, and individuals with special healthcare needs.
The first issue Dr. Clancy raised for the Subcommittee was what everyone meant by federal data collection. On one level, Dr. Clancy said many people had a clear connotation that the Committee’s mandate and jurisdiction was about surveys and clearly defined efforts. But she noted the Commonwealth Fund supported report put it all on one map that included data collection on claims for services and efforts where the federal government joined forces with the private sector to assess and improve quality of care.
She also pointed out that data needed for surveillance and monitoring might not be equivalent to data needed to improve healthcare. In theory, they were the same—and, to the extent they actually could be the same, there might be enormous efficiencies in synergy of purpose. But she emphasized that measuring quality of care required that a great deal of thought be given to criteria that: (1) measured something important, (2) was feasible enough to collect the data, and (3) was under the control of providers or healthcare systems. Noting that data on race/ethnicity wasn’t consistently available from healthcare providers, Dr. Clancy said AHRQ had funded an exploratory analysis looking at integrated delivery systems and how they collected these data. She noted it took the group several months and multiple conference calls to ensure they were defining terms equivalently. They’d found that some delivery systems actually had a blanket policy of not collecting these data. She acknowledged that getting to the why behind that policy was tricky, but cautioned they wouldn’t make headway until they understood more. She also reported that the data that was available often wasn't comparable (compliant with OMB)--They needed to move forward with compliance with OMB standards. Another challenge was that the data collected on race/ethnicity was often incomplete and not easily linked to data on the other characteristics they’d just discussed: insurance coverage, employment, income, education, and (of particular importance) local context. To illustrate her point, Dr. Clancy pointed out that knowing, in some neighborhoods, pharmacies didn't stock certain pharmaceuticals would be important in trying to address disparities in control of pain. Health statistics could indicate disparities, but couldn’t tell why they were there or how to address them. She also noted improvements in quality weren't possible without relatively short turnaround feedback. They had to experiment with ways to make some data in the federal data collection process more rapidly available.
Dr. Clancy said AHRQ was producing a report on the quality of healthcare. IOM had suggested that one of the important prime domains was equity. That report will include quality measures, stratified to the extent possible by race/ethnicity and socio-economic position, as well as consumer and patient assessments of healthcare quality. It also anticipated difficulties with adequate sample size from most federal datasets. She emphasized what they didn't know and needed to find out: why disparities occurred, what local circumstances ameliorated or increased disparities, and how to collect these important data respectfully and effectively.
Dr. Clancy said her wish list was to mandate collection of these data from all who receive federal funds consistent with the OMB standard. To do this effectively, the fields would have to be reviewed and updated regularly. Getting at the intersection of these different characteristics required either mandating collection of family income and education data or developing a mechanism to link them with other data sources. Dr. Clancy noted this raised concerns about privacy and confidentiality, but she said they had to develop and implement a strategy for identifying relevant local characteristics.
Dr. Clancy suggested that the Medicare Health Outcomes Survey (MHOS) mandated for those involved in risk or Medicare Plus Choice plans offered an opportunity to learn from data collection ongoing for a couple years. She said the survey, which will report on a two-year change score in the SF36, had huge amounts of sociodemographic data and important lessons could be learned from how that process worked and problems encountered collecting the data. Survey vendors had made important improvements to increase response rate and get the right answers. Dr. Clancy emphasized two needs: exploring strategies for rapid release of selected data for improvement purposes, and thinking carefully about a strategy for reporting these data. She said the question with MHOS was, given its wealth of sociodemographic data, how to present the results--adjust for the differences (which some would say might exacerbate the pathology) or report in a stratified fashion? To the extent it was believed that many of these characteristics exerted influence outside the control of healthcare systems and providers, Dr. Clancy advised adjusting for those differences to support fair reporting. To the extent that some of these differences reflected likely differences in the experiences individuals have with healthcare systems, the results should be stratified. She suggested they might want to try both strategies. Dr. Clancy stressed that it was important for the Committee to consider how they got there and she suggested they might need to rethink their approach to data collection. They knew all improvement was local, but most of the data was nationally representative. She said it was clear from the distribution of different populations across the country that nationally representative data wouldn’t address many of the ethnic populations’ needs. She questioned if there were ways to explore partnership opportunities with the state and communities to get at this issue.
She also said they needed to learn more from the users of these data and from other sectors. She observed that the first line in the in the D.C. area real estate ads and regular reports in the newspaper tracked the performance of the different elementary schools. There was considerable interest in dealing with socio-economic position and other community characteristics and opportunities for dialogues about how different communities were interpreting the data and using them to improve.
Dr. Clancy cautioned about the cost of not considering disparities associated with race, ethnicity and socio-economic position. Not acting ran the risk of undermining quality measurement efforts. They wouldn't necessarily be targeting the costs and considerable investment of energy and resources required by data collection to those in greatest need. Dr. Clancy said they’d also deprive themselves of important scientific knowledge. She noted that African-American woman present with more advanced breast cancer. Some syntheses of existing studies suggested that access to mammography and follow up of abnormal mammogram was part of the problem, but that there were other unknown factors. She said they could learn from measuring which were the important scientific areas for biomedical research. If they didn't move forward in collecting these data, Dr. Clancy cautioned the end result would be that resources would be misallocated for quality improvement.
Discussion
Mr. Handler suggested using racial data obtained from state birth/death records to measure health disparities. Noting Dr. Clancy had said all organizations receiving federal funds should use the OMB standard, he pointed out that (except for California) the states hadn’t decided to use the multi-race identifier on their birth/death records. There was no valid way to project outward past the population data the Census Bureau collected in 2000 and no valid way to calculate birth/death rates for the year 2000, because the denominator was based on multiple race identifiers. This was a growing problem and only a year away. Dr. Clancy noted the number of people using the multiple race category in the Census varied by different racial groups. It was a small proportion, but an important problem. The other question was whether it varied by communities. There wasn’t an easy fix, but it was worth exploring.
Dr. Heurtin-Roberts had been wondering if the 100 PSUs drawn to produce nationally representative results in MCBS had been adjusted over time. Demographics of the population changed and would change more in the future. But hearing Dr. Clancy’s thoughts, she questioned if a nationally representative sample was what they wanted. Only going with nationally representative samples would artificially deflate the impact of populations in some areas (e.g., Texas and California). Perhaps they needed to rethink how to collect this information so it was useful for policy.
Dr. Clancy said MEPS had done considerable work in developing techniques to identify people in lower socio-economic groups for over sampling. She said she was reasonably confident that enough individuals from those strata would be enrolled to make nationally representative estimates. But she acknowledged that, for some issues, nationally represented might not be the Holy Grail. Dr. Clancy agreed the question about PSUs and how they'd changed was worth exploring.
Dr. John Park said even in Montgomery County (Maryland), where they had an affluent jurisdiction, there were huge disparities between Whites and African-Americans and between Whites and Latinos. Interest groups had been formed, but Dr. Park said he was with a local government and knew resources were limited because they didn't have the locally specific data to support what they really needed to find out in order to deal with racial disparities. Dr. Clancy said she participated in a steering committee for Latino health initiatives in Montgomery County. An issue they all struggled with was the rapid increase in the county’s Latino population over the last decade. It's a diverse population and data was extremely limited. What county programs, hospitals, or the state collected weren’t necessarily consistent; none of it was consistent with OMB. They tried to push forward on a nuts-and-bolts level, adding fields to the hospital discharge.
Dr. Gibbons said the mandate from Congress to collect race/ethnicity data came up all the time at CMS. The managed care plans that worked with CMS had a strong perception that they were afraid to collect the data because of legal consequences. He cautioned that, if they didn’t collect this data, they’d lose more than the ability to perform the science. Given current demographic trends, projections in the Medicare population and shifts in minority populations, not getting accurate data, but continuing to do things that didn’t work, threatened the accurate functioning of the healthcare system. Dr. Gibbons said he was part of a minority and wasn’t against focusing on the small groups, but he felt they needed to think more about moving population parameters forward. A high-risk strategy approach might not get them there.
Dr. Clancy concurred with Dr. Gibbons’ observation about misperceptions of the legal collection of these data. She said a fair amount of work had been done about that, but more education and dissemination was needed. She recalled a frank conversation with leaders of managed care plans a few years ago about how to collect these data; nobody had a clear sense of how to do it respectfully and efficiently. Many health plans had collected baseline health data at one point. That turned out to be a really costly idea and hard to justify in terms of return on investment. Most people didn't want assessment of race/ethnicity linked too closely to enrollment, even if it was okay legally. It might create a strong perception of redlining. So when in the process did they collect it?
Dr. Pokras reported that the Office of Minority Health worked closely with ARHQ, CMS, Academic Medicine and Managed Care Forum and other agencies and organizations outside the federal government to investigate these issues. OMB had also funded a project at the National Health Law Project to review state laws and regulations addressing whether it was legal to collect race/ethnic data. In expressing their reluctance to collect data, a number of people questioned its legality. Four states did prohibit collection of race/ethnic data by health plans or insurers at the time of enrollment, but allowed collection any other time. As they’d heard, Dr. Perow's study reviewed the federal laws and regulations and didn't find anything that prohibited health plans and insurers from collecting the data.
OMB also worked with pending regulations. The State Children's Health Insurance Program was published last summer as an interim final rule. Also last summer, a notice of proposed rulemaking was published for Medicaid managed care. Dr. Pokras reported that the anti-affirmative action movement in California, concerned about the state’s racial privacy initiative and that collection of race/ethnic data might continue patterns of differential treatment, sought to prohibit collecting the information. Californians voted on the November ballot. There was exclusion for medical research subjects and patients, but Dr. Pokras said it wasn’t well defined. He noted there were various advocacy groups, including the ACLU, which was advertising for someone to work full-time gathering information about this initiative.
Mr. Michael Katz updated the Subcommittee on CMS’s activities related to population race/ethnicity data. CMS held a conference on race/ethnicity in October 1999. Mr. Katz said the numbers were bad in 2000 when CMS published Healthcare Financing Review. But by December 2001 the numbers had doubled from 1997. Over 70,000 American Indians/Alaskan Natives were enrolled in Medicare, 600,000 Asian-Americans/Pacific Islanders, and 900,000 Hispanic/Latino populations. Mr. Katz said CDC was working with SSA to get language data into the mix. SSA had language data in two parts of its organization, but CMS only received one part. The part received had a language preference for 10 languages. The client data file had over 27 languages.
Socio-Economic Status
Dr. O’Campo began with the central concept that socio-economic status captured stratification of society into classes of individuals with differential access to a variety of economic, political, health, housing and other resources that resulted in different life chances. Those life chances impact health directly and result in the gradient demonstrated in hundreds of studies. Socio-economic status is measured in a variety of ways; in health, most of them focus on individual measures. Dr. O’Campo emphasized the importance of complete categorization and measurement of socio-economic status in order to understand racial inequalities in health. And she said that measures included in surveys should help bring an understanding of whether and why some people had more material resources, power, and prestige, whether and how socio-economic status had an impact, how it varied by race/ethnicity, and how it helped explain racial inequalities in health.
Sociologists considered two groups of measures in thinking about socio-economic status at the individual level. Social stratification measures ordered individuals hierarchically according to education or income and hierarchical gradation described different life chances groups had. Relational measures divided society into classes that usually had opposing interests. Dr. O’Campo depicted classes as being related because one benefited at the cost of others--thus inequalities and gradations were created.
Dr. O’Campo summarized hierarchical measures. Gradation measures described economic gradation, income (can fluctuate year to year), and wealth (a stable, more accurate gauge of total available resources). Power-based measures included occupational hierarchies. Prestige-based measures included education. Education and income were used most. Dr. O’Campo showed an example (data with intimate partner violence) measured by annual household income, with gradation for both genders. The rate of domestic violence was inversely proportional to income. Another example (no physician contact in the past year) showed similar measures and patterns.
Dr. O’Campo noted Americans were less familiar with relational or social class measures of health, which could add an important dimension to understanding socio-economic status and health. She defined the concept as identifying social class as groups of individuals with opposing interests, differentiated by their ability to have control over ownership assets. Eric Wright developed measures of organizational assets. Questions used to gauge peoples’ status in relation to organizational assets ranked them as manager, supervisor, or neither. Managers had input in policymaking and the ability to hire/fire individuals. Supervisors had the ability to hire/fire, but often didn’t have policymaking powers. Others had neither supervisory nor sanctioning power. Dr. O’Campo said another way of looking at ownership assets was to categorize managers according to the number of levels beneath them. A chair of a department with at least two hierarchical levels beneath her or him could be considered a manager. First-line supervisors had authority over at least several workers, but only non-supervisory workers beneath them.
Dr. O’Campo showed standardized mortality rates for European population compared to the professional class that had the lowest mortality rates. The number of excess deaths swelled, as one slid down the social class line. Another example, the National Longitudinal Mortality Survey, categorized people according to professional/managerial vs. the rest of the people. The lowest rate of inequality was between professionals and non-professionals within the government; the highest level was in the manufacturing and service area. Professionals had lower mortality than non-professionals overall. An example from the Baltimore Epidemiologic Survey represented differences in major depressive disorder in particular communities by income, with the population divided into higher, mid, and lower tertile. The lowest tertile had the highest levels of major depressive disorder. Dr. O’Campo said, typically, when looking at just education or income, knowing a person's social class position indicated something about risk, at least for mental health disorders.
Noting socio-economic status was a complex issue that couldn’t be completely measured by focusing on data at the individual level, Dr. O’Campo identified area-based measures in three categories. Occupational structure characterized such information as the proportion of professionals vs. non-professionals in a given area. Educational structure characterized such data as the proportion with college degrees. Levels of income and wealth contributed to economic structure.
Dr. O’Campo emphasized that including measures of both socio-economic status and area more completely characterized population. Neighborhood SES measures increased understanding and the magnitude of neighborhood measures was larger than individual-level SES. Effect modification and confounding was present when area-based measures were added. Dr. O’Campo discussed using unemployment levels in looking at intimate partner violence during the childbearing year (pregnancy up to about six months post partum). When information on the area of residence was added to analyses of data based on individual information, odds ratios were found for the risk of intimate partner violence. Education, and income were used to measure individual levels. Three indicators of area-level economic position were: ratio of homeowners to renters, unemployment rate, and level of income. Despite being a low-income population (most received Medicaid) the women lived in all quartiles of the city. Dr. O’Campo emphasized the importance of keeping in mind when characterizing socio-economic status that low income didn’t necessarily mean living in a low-income community. When all these variables were in the model, Dr. O’Campo said they found that individual-level income was no longer important (perhaps because the sample was uniformly low income). Higher education put one at higher risk, which Dr. O’Campo noted was consistent with theories related to intimate partner violence. Even with education and income in the model, area-level measures were significant. Once the neighborhood measures were added to the sample and adjusted for neighborhood environment factors, White woman were found to be nine times more likely to be at risk of intimate partner violence than Black women. Effect modification by race was evident when information on area was added.
Dr. O’Campo said stratification measures (income and education) were more common in the U.S., but relational measures might capture important issues related to race and ethnic discrimination. Area-based measures helped capture complexities of socio-economic status. Dr. O’Campo said SES was most informative when measured more than one way and shouldn’t be limited to the individual level. She noted the best measures for women might be different than the best measures for men. Household socio-economic status might be a better predictor of outcomes than a woman's own socio-economic status, especially for education. Similar issues might apply for men. And it might be that household socio-economic status was the best indicator overall.
Data was often used to make comparisons within groups (e.g., White women to White men), but Dr. O’Campo remarked that everyone should be earning what White men did. She noted quite a bit of disparity across genders and ethnic groups. For every dollar that a White man earned, an Asian man earned 92 cents, a Black man 66 cents, a Hispanic man 62 cents. White women earned 52 cents for every dollar that a White man earns; Hispanics earned 37 cents. Over 50 percent of Hispanic women, 40 percent of Black women, and 30 percent of White women earned poverty-level wages.
Noting education was the socio-economic status factor used most in epidemiology, Dr. O’Campo indicated there were still limitations when trying to understand racial disparities in health. Adjusting for education, didn’t always account for disparities in income. And differences become starker when considering education, net worth, or wealth differentials. The net worth for Blacks who had an elementary school education was $2,500. The net worth for comparable Whites was $25,000. Adjusting samples for education missed these differences. Dr. O’Campo emphasized that education and income, alone, weren’t enough when thinking about socio-economic status. Wealth had to be considered. Because of the complexities of socio-economic status, Dr. O’Campo encouraged accounting for as much of the difference as possible. She emphasized the importance of thinking about how measures of socio-economic status differed according to race/ethnic group, the meaning of the variables, and expected association to health. In theory, adjustments were for something independent of race, but patterns of educational attainment, occupation, or income distribution were, in part, determined by race. And even when an adjustment was made, it might not be proper. A Hispanic and a White person might both earn $40,000 a year, but the social and health costs may have been greater for the Hispanic.
Dr. O’Campo said the choice of measure used depended on the health outcome studied. Looking at pregnancy outcomes, one might want something proximate to the time period of pregnancy (e.g., income). Looking at chronic diseases over lifetime, one might choose something stable like education. With an older population, wealth measures might be more important. Surveys tended to ask about a person's status at that moment, but she pointed out that wasn't the whole story. Dr. O’Campo said the realization that socio-economic status was an effect over the whole life was an important idea gaining support. She noted Dr. Barker had said cardiovascular disease in seniors was, in part, determined by health status at a younger age. When thinking about measuring socio-economic status (even if the interest was in health issues related to adulthood and old age), she suggested they might want to collect information about socio-economic status of parents in childhood or around birth. She noted this was important when considering racial inequalities of health. The socio-economic status trajectories for different race/ethnic groups differed. Rather than just collecting data at a particular point in time, for the health outcome interest and socio-economic status relevant then, they could collect socio-economic status for more than one period in the lifetime.
Dr. O’Campo reiterated that there were numerous individual and socio-economic status levels to choose among. She recommended using more than one, noting going beyond education and income was a move toward understanding issues of racial inequality in health. She encouraged going beyond the individual level and routinely including considerations of area-based measures of socio-economic status. Although she’d shown data on economic aspects, Dr. O’Campo emphasized other areas impacted on health and that, if surveys captured that data, it could be brought into the equation and contribute to an understanding of racial inequalities in health. She advised that more information and basic research were needed to understand various socio-economic status issues related to race/ethnicity and identify best measures of socio-economic status that lead to understanding women's health issues. Dr. O’Campo recommended using multiple measures at the individual and area level to account for socio-economic status across the races and comprehend racial inequalities in health.
Questions and Answers: SES
Mr. Handler doubted that racial inequalities could be totally explained by income; things happened simultaneously. He suggested that lung cancer rates might be higher for people at the lower income level because a higher proportion of that population smoked. And accident deaths might show a similar relationship because people at the lower socio-economic status engaged in riskier behavior. Dr. O’Campo said she couldn’t say exactly what proportion of the lung cancer data was attributable to smoking when they’d teased away the contribution of health behaviors from the gradient in socio-economic status. Studies indicated health behaviors didn't fully account for the gradient. And gradient remained even after accounting for health behaviors. Dr. Pokras said the IOM committee that provided recommendations and guidance to AHRQ for the Health Disparities in the Healthcare report had posed the options of an index that combined information from various socio-economic status measures or focusing on multi-co-linearity when including multiple measures in a multi-varied analysis. Dr. O’Campo remarked that it depended on the activity’s purpose: if the intent was to adjust for socio-economic status, they could “put in the kitchen sink.” But Dr. Pokras cautioned that an index might be an issue, especially if statistical power was a concern. Adjustment served a different purpose than identifying reasons for a particular adverse health outcome. Using an index or multiple indicators when adjusting or accounting, even if there was multiple co-linearity, shouldn't cause a problem. But if the focus was on how social class related to health outcome, the magnitude of the effect of education, and a common area level--than including an index would mix up all the issues and one couldn’t tease apart which issue, indicator, or aspect of socio-economic status was most important. If there was multiple co-linearity, estimates would be off and that question couldn’t be investigated. But an index would be fine for adjustment.
Noting income, education, and prestige based on occupational status were Western indicators of position and status, Dr. Heurtin-Roberts asked about exploration of other factors: e.g., a person with a low-level occupation might be a community or religious leader and have prestige. Dr. O’Campo agreed that there were potentially many ways to capture issues related to prestige and power that might differ across race/ethnic groups. It was an area that needed more exploration qualitatively and quantitatively. Dr. Lerg asked what area-based measure information should be collected on a survey, rather than linked and brought in through other pools. Dr. O’Campo noted a community psychologist would advise bringing in a lot of perceptions data. A number of studies asked about perceptions of high crime, social cohesion, and capital items in one’s neighborhood. But she didn't know if any particular socio-economic status area-based measure that could be asked about directly in a survey. Dr. Mays asked if any work looked at how different groups dealt with money and power relational, gender-dynamic issues. Dr. O’Campo said she didn’t know of any studies that tried to determine control over money and how that related to health. Dr. Ken Shue, Center to Reduce Cancer Health Disparities, said it was important to put the SES problem in perspective. They had rates on Blacks, Whites, and Asians, but no national annual rates on the poor by any division of socio-economic status. Cancer Institute tried to look at Census data once every 10 years and collect cancer incidence over time. But cancer statistics, which were a basis for disparities, didn't get at these issues.
Mr. Handler suggested another problem with using Census data to measure socio-economic status was migration. People couldn’t be pigeon holed: one out of five moved from county to county. Dr. O’Campo referenced other data that showed people moved to similar neighborhoods. There shouldn't be too much misclassification.
Dr. Coleman-Miller asked whether Dr. O’Campo had looked at the data that corrected for socio-economic status, co-morbidity, and gender: Peter Bach’s study of lung cancer validated what Dr. O’Campo said. It corrected for socio-economic status factors and found it still didn’t matter. Dr. Coleman-Miller asked if there was a way to feather that out. Dr. O’Campo said the most appropriate measure of socio-economic status depended on the health outcome. She said Dr. Coleman-Miller’s example was a good illustration of why having complete characterization of socio-economic status, even if it didn't matter, was good. You'd accounted for it and knew to look elsewhere.
Remarking that people thought they knew what everyone meant by socio-economic status, but often meant very different things, Dr. Gibbons asked Dr. O’Campo to define socio-economic status and how she distinguished that from social class. Dr. O’Campo agreed that it was important to distinguish between class and status measures and said she chose “socio-economic position” to describe both.
National Survey of Family Growth
Dr. Abma said NSFG was a periodic survey of the fertility of US women. A framework developed by demographers in the '80s had been used and refined to model fertility and from inception it guided the survey’s content development. At its core, this framework called the Approximate Determinants of Fertility contained intermediate variables most closely affecting live birth. Variables related to sexual intercourse included timing, frequency, and time spent in cohabitation and unions. Conception variables included contraceptive method used, sterilization, and infertility. Pregnancy outcome variables included miscarriage, stillbirth, and induced abortion. Dr. Abma said the race/ethnic measure was one of the social factors that were more distal determinants of fertility. Measures included the standard socio-economic status measures already discussed, religion, labor market participation, access to healthcare, and community environment.
Dr. Abma noted the data was relied upon as a dependable source of national estimates to monitor trends and inform policy on differentials and issues such as unintended pregnancy in childbearing, adoption, and relinquishment of birth for adoption, teenage pregnancy and sexual activity, breast feeding, infertility and impaired fecundity, family planning service use, and risk factors for HIV and STDs. There had been five rounds of data collection starting in the early '70s. Together they formed a time series for women of all marital statuses dating back to 1982. Some married women dated back to the beginning. Cycle 6 would begin in March 2002 and conclude in December.
The sample design had always been a nationally represented sample of women from the civilian, non-institutionalized population of reproductive age (approximately 15-44). NSFG had a probability sample, multi-stage, stratified cluster design. Cycles 1-3 were independent. Cycle 4, (1988) and Cycle 5 (1995) were linked to NHI. Cycle 6 was a return to an independent sample. In order to produce reliable statistics with these sample sizes, NSFG included an oversample of Black women. Starting in Cycle 5, an oversample of Hispanic women was also included. The 1995 data included 198 PSUs. The response rate was 79 percent and approximately equal for the 6,500 White, 2,500 Black, and 1,500 Hispanic respondents.
Dr. Abma said the question in the upcoming cycle conformed to the OMB Directive 15. The first pair of items identified Hispanic origin; the third asked respondents to self-identify and choose any groups that best described them. The fourth question, unique to some surveys, asked those choosing multiple races to pick the one best describing them. Interviewers identified by observation and recorded the race of respondent who didn’t make a selection.
Dr. Abma reported NCHS ended up with little in-house imputation of race/ethnic identification. But she noted that, even with Cycle 5’s larger sample size, only 163 women reported multiple races--NCHS couldn’t produce breakdowns of the smaller race/ethnic subcategory. For some outcomes, the Hispanic subsample also became thin. Dr. Abma said the strength NSFG brought to the data users, at least in terms of the broad race/ethnic categories, was the ability to monitor trends and differentials across a wide range of outcomes in terms of women's reproductive health, pregnancy and childbearing. NSFG also included a fairly rich array of potentially explanatory variables. Modeling of causal hypotheses was possible with the data and had been used extensively to understand processes. She said explanatory analyses were another of NSFG’s strengths. Healthy People 2000 and 2010 used NSFG to supply data for ten objectives, mostly in the family planning chapter. The NSFG was funded by and supplied data for the Office of Population Affairs.
NCFG showed characteristics of women using Title X family planning clinics and others who might be in need of such services. NCFG also provided information for CDC's HIV prevention program by showing estimates of sexual risk behaviors and other purposes including complementing the vital birth statistics collected by NCHS. There are three different sources for the measure of live births, fetal loss, and induced abortion needed to arrive at a pregnancy estimate. NSFG supplies the fetal loss data, which contain useful, descriptive findings about disparities between non-Hispanic White, non-Hispanic Black, and Hispanic women in terms of total pregnancy rates.
Non-Hispanic White report 67 percent live births. Non-Hispanic Black women report 50 percent live births and a larger proportion of induced abortions. Hispanic women’s’ report resembles non-Hispanic White women, though the non-Hispanic White teen pregnancy rate is less than half that for Hispanic and non-Hispanic Black teens.
The probability that cohabitation would result in a marriage was higher at each duration of the cohabitation for White females. The transition probabilities for non-Hispanic Black females are lower at each duration. For Black females there is a positive association between family income in the past year and probability that the cohabitation will end up as a marriage.
A study published in Pediatrics last year took the data and found race had independent effects after controlling for the socio-economic measures on the probability of a baby having been breast fed. Ever having been breast fed predicted infant mortality as strongly as low birth weight. Only 1/4 of non-Hispanic Black babies had ever been breast fed. The policy implication was raising the level of breast feeding among Black women would reduce the gap in infant mortality.
NSFG was a source of data on unintended pregnancies. Some 27 percent of births among Non-Hispanic Whites between 1990 and 1994 were unintended compared to 51 percent among non-Hispanic Blacks, 30 percent among Hispanic women. About 34 percent of births among Black females with less than a high school degree were unwanted; 11 percent among those with a bachelor's. One measure of the community socio-economic status, the poverty rate, was strongly and regularly associated with the percentage of births reported unwanted. Only 4 percent of births were unwanted for those living in low-poverty neighborhoods; those living in high-poverty areas reported 10 percent of births were unwanted.
Dr. Abma noted that contextual data were collected for characteristics of the community at three points in time, enhancing the ability to match the time an event happened with simultaneous community characteristics.
Cycle 6 was the next survey planned. Males would be interviewed in addition to females. The female survey would stay essentially the same. NCHS would add the other half of the information missed for years. They’d be able to shed light on race/ethnic differentials in such things as further involvement in activities with their children, how well and completely males reported fertility, partnerships and how many partnerships there’d been, male contraceptive behavior, the role played in preventing unintended pregnancies, male infertility, and STD/HIV risk and transmission.
Behavioral Risk Factor Surveillance Survey
Dr. Mariolis said BRFSS was a telephone survey primarily used to track the prevalence of behaviors related to chronic diseases and preventive health practices among the civilian, non-institutionalized population 18 years and older in each state. BRFSS was a joint venture of the CDC and health departments in the 50 states, District of Columbia, Guam, Puerto Rico, and the Virgin Islands. The states held control and ownership of the data.
CDC coordinated development of an annual set of core questions asked by each state. Questions changed yearly and every state was required to ask each core question. States could pick and choose standardized sets of questions on specific topics called modules. And every state was free to ask its own additional questions. CDC strove to set and monitor standards for sample design, data collection procedures, and technical support. Sampling consisted of all NXS-type residential listings. The design had to be RDD in each state. Data collection guidelines specified 15 callbacks to unresolved numbers distributed over weekdays, weeknights, and weekends.
One eligible adult was randomly chosen from each household. There were no proxy interviews. States were responsible for data collection. The data were collected and initially edited in the States with a program provided by CDC. Further editing was conducted at CDC and the data was weighted and returned to the states along with several reports. The states and CDC then generated their own reports and studies. The aggregate data file was available to the public. Dr. Mariolis noted there were a large variety of topics, even in the core. None were gone into in depth, but many topics were covered. Nearly 160,000 interviews were completed in 1999 and another 184,450 in 2000. About 204,000 were expected in 2001. The questionnaire was available on the Web site.
Dr. Mariolis said he believed race was a combination of biology and culture and that it was problematic to everyone. Hispanic origin was an ethnic category. The questions made a hard, sharp distinction between race and ethnicity. Dr. Mariolis said many people didn’t make that distinction and that influenced their answer; Hispanics disproportionately answering with “Other” was an indicator of that difficulty.
Respondents were asked if they were Hispanic or Latino. 2001 was the first year people could answer the second question with more than one race. Those that did were asked which group best represented their race. Other was an acceptable category. In 2000, when people were still only allowed one answer, 31 percent of the Hispanics coded “Other.” Dr. Mariolis considered “Other” a missing value; no one knew what it meant other than it wasn’t a standard race category. The Census Bureau would code much of the “Other” as White, because it counted 89 percent Whites. That wasn’t what people said. Dr. Mariolis said he’d hoped multiple answers to the race question would bring down the percent of “Other” for Hispanics. But while in 2000, about 30 percent of Hispanics chose “Other”, 41 percent of Hispanics in 2001 picked “Other.” People who originally gave multi answers and then chose a single race coded according to the single race chosen. Almost 3 percent of Hispanics (but only 1.34 percent of non-Hispanics) gave a multi-racial response. Multi-racial responses disproportionately came from American Indian/Alaskan Native, Native Hawaiians and a bit from the Other category. Distribution was similar for Hispanics, except Black Hispanics also tended to be high on the percentage giving a multi-racial response. Dr. Mariolis recommended always presenting results by race/ethnicity with Hispanic as a category, never by race alone. He said to never break Hispanics out by race.
Dr. Mariolis said there was minimal direct impact from allowing multi-racial choices. Choosing other race was only to a small extent an alternative to multi-race choices. People who choose other race were predominantly not multi-racial or didn't give a multi-racial response to race.
Questions and Answers: NSFG/BRFSS
Dr. Abma clarified that the fathers weren’t linked to the females. Some might have newborns; others wouldn’t. Males were also asked their self-identified race. Dr. Coleman-Miller asked if there was an opportunity with the fertility question to ask about wanted babies: the socio-economic status for minority women still had a higher infant mortality rate than for the majority female and talking about unwanted babies was the opposite of asset mapping. Dr. Abma acknowledged that was a good point. A conglomeration of questions and responses resulted in the measure of unintendedness. Whether a contraceptive method was used when the pregnancy occurred determined a routing through the queries that ended up with a timing question. The measure never used “intended” or “wanted.” Based on the disparity issue, Dr. Coleman-Miller said looking at wanted pregnancies that weren‘t live births would be of great concern, especially in relation to unwanted ones. Noting 16 and 17-year-olds came to clinics saying they couldn't get pregnant, he said fertility disparity between 15-44 year olds was a consideration. Dr. Abma said articles on the NCHS Web site mined that data in depth and profiled women with impaired fecundity and infertility. There were race/ethnic differences regarding sterilizing operations.
Remarking on the behavior list survey, Dr. Coleman-Miller pointed out that race was also problematic for the Black population and that the question of whether it was biologic, rather than cultural, and something one chose was argumentative and difficult to deal with at this point. In looking at the impact of socio-economic status and early childhood development on health across the life span, Dr. Barbara Kremgold, Center for the Advancement of Health, expressed concern over 40 percent poverty rates for Black and Hispanic children. Noting the data on the probability of marriage was strongly related to income and suggested that jobs for fathers were important in the probability rates for marriage, she asked if the data had been shown jointly to the part of HHS that dealt with welfare reform and an interagency committee with the Department of Labor. She suggested discussing the jobs programs for minority males in this context. Dr. Abma said a paper was presented at the AEI seminar on the effects of welfare reform and publications were listed on the Web site.
Dr. Heurtin-Roberts said she was puzzled by Dr. Mariolis' comment about never breaking out Hispanics by race. The fact that Hispanics kept identifying themselves as “Other” was datum. Instead of saying they wouldn’t present it, she said they ought to rethink how they questioned and considered Hispanics. Dr. Mariolis disagreed, noting the context of a state health department report looking at prevalence of smoking by race or race/ethnicity. They were constrained by the categories in the census data; when they didn't match, people questioned the data. The racial distribution of Hispanics in the BRFSS wouldn’t match what was in the state based on Census data. And mixing Hispanics with non-Hispanics and presenting straight race would also have an effect. That was the context and there were situations where one wanted to look at it. Dr. Heurtin-Roberts reiterated concern that people self-identified in conflict with the surveys. If the surveys weren’t measuring the reality of self-identification, they needed adjustment. Dr. Mariolis agreed. He said he was dealing with the data CDC had now.
Dr. Carter-Foster clarified there were Healthy People 2010 objectives related to infertility and differences by race/ethnicity. She noted Thursday morning’s session of the Secretary's Advisory Committee on Genetic Testing would include a panel of geneticists, anthropologists, and biologists discussing whether race/ethnicity data should be collected for genetic testing and screening. She said the hypothesis about the impact of multi-racial categories on the selection of Hispanics using other race had been well researched by the Labor and Commerce Departments. Research and references for the pilot study were available on the Census Bureau’s Web site, with a hot link indexed under race. Dr. Carter-Foster said they’d hoped asking about origin first would reduce the percentage of Hispanics selecting other race, but it hadn't had much impact.
Dr. Park asked if someone couldn’t give local governments local-level data on the behavior risk factor survey. He noted the youth BRFSS was done separately and metro-wide. Dr. Mariolis said the BRFSS data could be downloaded with a high-speed Internet connection from the CDC Web site. The dataset contained county information with more than 50 respondents. He noted an upcoming surveillance MMWR article looked at BRFSS by metro area.
Dr. Mays asked about data on factors that disrupted cohabitation. They knew cohabitation differed by gender: women more than men believed living together lead to marriage. And they knew that, for some ethnic groups and ages, disruptions often involved financial problems or prison. And there were sensitive issues around pregnancy--e.g., the difference between wanted and unwanted pregnancies and what that meant in terms of intervention. In looking at health disparities, Dr. Mays asked the testifiers to talk a bit about acculturation or any other variables that might help them intervene. Dr. Abma explained that the concept of unintended pregnancy was a way to measure unmet need for family planning services. It evaluated the difference between the total fertility women intended and thought ideal vs. total fertility received, what accounted for the difference between them, and percentages of population experiencing barriers to family planning services. Noting the concept of how best to measure the intention status of pregnancies evolved from this, Dr. Abma said Cycle 5 had measures that allowed them to look at separations followed by reunions. Cycle 5 wouldn’t get at in-depth answers, but was a springboard for ethnographic or further studies to investigate patterns.
With the focus on health disparities, Dr. Mays said the question was, what did they want to do and what did they want to fix? They had a factor about racial differences and cohabitation; groups didn't equally make it to marriage. They had a message. It was up to somebody else to figure out why. She said she was trying to figure out the best source. Did they link the dataset with others giving more contextual information or require cultural competence training so data analysts could push as far as possible? Dr. Abma noted NSFG linked NHIS with Cycle 5 data. It accessed a vast variety of health measures, though perhaps not these specific areas, but it would increase the options that could be added to a model to help explain the outcome. The contextual measures refer to a file linked to respondent information containing hundreds of neighborhood characteristics measured at three points in time, including sheer composition of the area with regard to race/ethnicity, family planning, service providers-per-capita, socio-economic status, female/male, and unemployment. Dr. Abma said the survey was valuable for causal modeling, though she didn’t know if it applied to this issue.
Dr. Heurtin-Roberts suggested the question was where were the partners. From what they knew about mortality rates among African-American men, a number died from disease and violence before cohabitation could lead to marriage. She said she didn’t know if the dataset could be linked to others, but she said it illustrated how limited the data was in its interpretation without other contextual information. Dr. Abma replied that NSFG had information that shed light on reasons for separations within cohabitation and marriage. In the upcoming cycle, men and women were asked to self-identify as married, cohabiting, or partner absent. A detailed follow-up question asked where they were. She said they might find other reasons prohibit continuation of union. Dr. Mays agreed that, to some extent, the numbers and pathology were there. But she noted they also needed perspective on what the resiliency was and what to build on in terms of the translational research Dr. Clancy advocated. And then, how did they move it without having people come in and do the analysis.
Mr. Handler noted that common law was a legitimate form of marriage in Puerto Rico, and New York City recognized common law as a form of marriage so long as it began in a jurisdiction where it was legitimate and the couple stayed together seven years. He asked if the survey was structured to recognize common law marriage begun elsewhere. Dr. Abma said the structure identified it and enabled a cohabitation history; they could measure the duration of the union and whether it met that criterion--But common law marriage didn’t classify as a marital status.
Dr. Gibbons commented on something he said would get more problematic: race and ethnicity weren’t the same. He noted the idea behind a paper written by Thomas Levese at Johns Hopkins: Why Should We Continue Measuring Race, But Do a Better Job: if health services research was done along other variables the way it commonly was with race, it would be considered shoddy and never published. Dr. Gibbons stressed that one had to go deeper than controlling research for race: the connotation of race in America was largely political, in the broadest sense, and that broke down as Hispanics and other minorities diversified society. Dr. Gibbons contended that an African-American understood the self-qualifier as a race question; a Hispanic heard an ethnic question. Ethnicity came from different places. Dr. Gibbons advised the Subcommittee to remember that race or ethnicity data didn’t give the same thing and to figure out what it really sought.
Dr. Mays reflected that it was clear that the surveys they’d heard from were critical to the determination of their target for Healthy People 2010, resource allocation, and what it was that the public often heard first about a slice of life. To that extent, Dr. Mays said it behooved the Subcommittee to pay attention to the power of these surveys and whether they addressed what Healthy People 2010 was designed to achieve or the mission of the offices that receive guidance from these surveys. They’d just gone through a round of guidance concerning these questions on race/ethnicity. They knew they’d made progress and that their colleagues working on these surveys were pushing ahead. She said their last question was: What did they need to know, whom did they need to know it for, and how soon did they need to do it? She noted a mid-point review was around the corner; 2010 would be upon them. They were getting presentations on data from 1996-1998 and a bit in 2000. They’d be at 2004-2205 before they could mine the most recent data and see changes.
Noting the presenters left the Committee with much to think about and many challenges, Dr. Mays said she wanted to share some challenges with them. The Subcommittee had its piece to do, but presenters belonged to organizations, wrote articles, and stood in positions where their voices were stronger than some members. Dr. Mays said she hoped this dialogue would continue as they gained input from organizations, individuals and the literature. She thanked the individual users for sharing their thoughts about priorities for this data. She said this was their tax dollars at work--They had some ownership of this data.
National Health Interview Survey
Ms. Lucas said the Census Bureau conducted HIS, one of the largest health surveys in the United States, annually for NCHS. The multi-stage probability sample was drawn to be nationally representative of the non-institutionalized civilian population. The sample has a stratified cluster design and each year included about 40,000 households with 100,000 persons. The HIS sample is redesigned following the decennial census (1995-2004). The Bureau implements the next sample design beginning 2005. The most recent redesign oversampled Black and Hispanic households in high-density areas. The 1985-1994 design oversampled only Black households.
The household-based interview is conducted face-to-face via CAPI, with all household members available. The field representative can access a full Spanish translation via a toggle key on a question-by-question basis. The questionnaire was completely redesigned in 1997. The survey was divided into two major sections: one basic module administered to all family members (plus basic modules for one randomly selected adult and child); topical modules vary year to year. The basic module collects data on: activity limitations, injuries, conditions, health behaviors, access to health care, utilization, health insurance, demographics, income and assets, and family composition. The topical modules are analogous to supplements from the earlier HIS. The topical modules added flexibility to address new public health topics as the need for data arose (e.g., the 1998 Prevention module provided data for assessing Healthy People 2000 Objectives; Prevention 2001 set baseline measures for the 2010 health objectives for the nation.
HIS also contains a number of non-health measures of interest to analysts, including: household composition (family size and relationships), demographic information (including Hispanic origin and single/multiple race data), socioeconomic status, geography, proxy measures of acculturation (including length of residency in the United States and nativity, and contextual data at the census tract and block group levels.
Ms. Lucas presented examples of bivariate analyses of HIS data that could be used to assess racial and ethnic disparities in health. The data illustrated that non-Hispanic Black children were most likely to have had an asthma attack and that Hispanic children were more likely to be uninsured, face unmet medical needs and delayed care, and lack a usual source of care. She noted that, typically, one year of HIS data could be used to assess health characteristics for only the largest racial and ethnic groups. But data from more than one survey year could be combined to increase sample sizes and create estimates for smaller population groups. Other examples illustrated: (1) respondent assessed health status for Asian, Pacific Islander, and non-Hispanic White population groups, (2) Japanese respondents were more likely to report their health as excellent than other API groups, Vietnamese respondents were most likely to rate fair or poor, (3) the proportions of U.S. and foreign Black and White persons living in areas with varying concentrations of the Black population. Ms. Lucas mentioned other types of analyses done using HIS data. HIS linked to both National Death Index (NDI) and NSFG, maximizing the analytic potential of HIS data. Multivariate analyses of HIS data also examined whether family structure and characteristics predicted child health status and whether socioeconomic and demographic factors were associated with differential health status between U.S. and foreign-born Black and White persons.
Ms. Lucas identified issues one needed to keep in mind when using HIS data to assess racial and ethnic disparities in health: (1) Usually a single year of data could only be used to assess health measures for the largest groups, (2) smaller subpopulation group analyses required two or more years of data--in some cases (e.g., the American Indian population), even several years of data might not produce viable estimates, (3) confidentiality requirements restrict the amount of information available on public use data files--some detailed/disaggregated data must be accessed through the Research Data Center (RDC), (4) HIS currently has limited ability to oversample smaller racial/ethnic populations, especially those in less heavily concentrated areas, (5) HIS doesn’t assess cultural competency or collect detailed information on languages spoken in the home or English proficiency and measures of acculturation were limited, (6) there were no immediate plans to translate the instrument beyond Spanish.
Noting that measurement of race and ethnicity was central to assessing racial and ethnic disparities in health, Ms. Lucas commented on OMB’s new standards for federal race and ethnicity data collection. She noted these standards, designating new population groups, had implications for how health outcomes were measured and for whom data must be gathered. They also had implications for how trends in data systems were maintained, which monitored overall changes in health outcomes over time, and most importantly, the assessment of whether observed population changes were the result of changes in classification of groups or actual behavior changes/successful program intervention.
Ms. Lucas summarized the new standards, which: (1) split the Asian Pacific Islander group into two groups; Asian, and Native Hawaiian or Other Pacific Islander, (2) required that items on Hispanic origin be asked prior to and separately from the race question, and (3) allowed respondents to report more than one race. Ms. Lucas said the most important changes that could be expected in data systems from the new standards were: (1) changes in tabulation and presentation of data (e.g., NCHS's publications will begin to show data for new population groups), (2) changes in trend data—monitoring health outcomes for new groups will create breaks in the data and force development of ways to make old and new data comparable, (3) changes in the interpretation of health data for racial/ethnic groups.
Ms. Lucas provided a snapshot of how multiple race groups fit into the overall NHIS for 1997-1999. About 1.4 percent of the sample reported more than one race, consistent with estimates from other sources that put the number at about two percent or less of the population. The largest multiple race group was American Indian and Alaska Native and White, followed by Asian Pacific Islander and White, and Black and White. The age distributions for the four single race groups and the American Indian, Alaska Native and White group were fairly similar. But over 60 percent of the Asian Pacific Islander and White group and almost 80 percent of the Black and White group were under the age of 17.
HIS respondents reporting more than one race were asked to indicate which group best described their primary race. Anyone who didn’t select a primary group was termed multiple race. In 1997 and 1998, a majority of respondents selected a single race group. A large majority of persons reporting as American Indian/Alaska Native and White selected White. Almost 50 percent of those who identified as Asian Pacific Islander and White selected Asian Pacific Islander; about 40 percent selected White. About half in the Black and White group selected Black; about one-quarter selected White. Persons who identified as Black and White were the least likely of all the multiracial groups to identify with a primary race group and most likely to not select any group at all to identify themselves. Ms. Lucas said questionnaire items like primary racial identity could support data systems that had to place multiracial persons in a single race category, bridging them to maintain trends in their data.
Ms. Lucas identified issues related to measurement of race and ethnicity in NHIS. She noted that maintaining trends in health data by race required a bridging method and that, with HIS, the question allowing multiracial persons to self-allocate a primary racial identity might work best. However, she cautioned that self-identity and group size changes would make allocation to a single race group increasingly difficult. As these groups grew, increasing awareness of multiracial heritage and an increasing desire to report it might require new trends. Ms. Lucas also noted issues related to interpretation of race and ethnicity data under the new standards. Native Hawaiian, other Pacific Islander and multiracial groups with distinct characteristics and patterns of illness and disease had to be studied further.
Ms. Lucas noted the need to more fully acknowledge the fluidity of racial and ethnic identities, which might change the fundamental concept of race. She suggested they consider whether primary racial identity had substantive meaning for multiple race persons and the role it might play in understanding health behaviors and outcome. Considering health characteristics of single and multiple race groups had made the already complex relationship between race and health even more multifarious.
Other uses of NHIS data to examine race and ethnicity reporting Ms. Lucas mentioned included linked file analyses that look at the consistency of race reporting in the linked NHIS-NSFG and use of HIS data to develop a bridging method for vital statistics data, which doesn’t have an item allowing selection of a primary racial identity. Multivariate analyses included a mortality profile of multiple race persons in the U.S based on their racial classification in NDI and a demographic and health profile of multiple race persons in the U.S. using HIS data 1997-2000.
Ms. Lucas said future directions for NHIS include examining over sampling of Asian population subgroups, considering targeted over sampling to study smaller groups (e.g., American Indians/Alaska Natives and Native Hawaiians/other Pacific Islander) for whom health data is needed, cognitive work at NCHS to examine commitment to a racial identity (primary race vs. multiple race) experiences with discrimination in seeking and receiving health care.
National Health Interview Survey User
Dr. Hummer used the HIS extensively, beginning with the 1986-1994 period when it linked with NDI, allowing for follow-up mortality analysis. Many of his comments stemmed from his perspective on how HIS was collected then. Because this dataset had been around a long time, depending on the questions and health outcomes, cautious comparisons could be made across time. Dr. Hummer noted another strength was its large sample size. On average, 40,000 households, a hundred thousand individuals per year, provided one of the largest health datasets in the United States. HIS had tremendous strength for analyses looking at relatively rare health outcomes and behaviors. One could look at several race and ethnic groups as well as health outcomes and behaviors by age and sex. Given the large sample size and number of variables available, breakdowns could be done (e.g., splitting out native-born versus Mexican-born individuals.
Another strength was a household-based survey that allowed for linkage between individuals in a household and looking at patterns of health and health behavior present. Dr. Hummer noted the data was available in separate files, so it was easily accessible and usable by people with minimal computer proficiency. Dr. Hummer reported response rates were better than most, if not all, surveys he’d worked with: up in the 90 percents over time, attesting to the survey’s longstanding running and interviewers’ professionalism and ability to build confidence. Questions were basically the same each year and, as Dr. Lucas pointed out, pooling across years sometimes was necessary in analyzing health disparities. Yearly shifts in topical questions provided a flexible way get at different health problems, based on perceived need--But Dr. Hummer noted what he’d seen had less to do with race/ethnicity or immigration than it did with specific diseases and behaviors.
Dr. Hummer noted HIS allowed good basic measures of health, parent and self-reports. Many objectives of Healthy People 2000-2010 could be tracked through the use of these data. Another strength was the over sampling for Blacks and Hispanic persons allowed for subgroup analyses. Weights allowed for inflating the population. Dr. Hummer emphasized the importance of the links to the mortality data for adults through NDI and the field’s demand for mortality datasets outside of vital statistics. Pooling 1986 through 1994 provided nine years of data for core surveys and solid estimates of minority group mortality. One could also look at migration variables related to race and ethnicity, although there were issues about matches made to NDI. The National Longitudinal Mortality Survey, based on the CPS surveys, was limited in content and lacked health-related datasets; HIS provided one of the largest and best datasets for looking at mortality patterns throughout the United States.
Dr. Hummer advocated matching topical surveys to future years. Hopefully, one could look at special topics and follow individuals over time. He noted few datasets could analyze the types of data one could here and get the number of deaths that they had from 1986-1994 surveys (e.g., religious involvement, health behaviors like cigarette smoking and alcohol use, health status variables to look at self-reported health and follow up mortality, and a number of interesting variables related to race and ethnic health disparities).
Dr. Hummer said the most critical thing about HIS was that, despite over sampling, sample sizes for most minority groups remained very limited for many purposes. There was a real need to recognize diversity, even within umbrella groups. Puerto Ricans were different than Cubans, who were far different from the Mexican-origin population. HIS couldn’t help with understanding much about health patterns among Latin American, Asian subpopulation, Native American, and other racial and ethnic groups. Dr. Hummer noted the percentage of non-Hispanic White households had gone down, but Whites still outnumber non-Hispanic Black and Mexican origin populations by about 4-to-1 in the dataset. He acknowledged limitations of costs, but questioned why 60,000 Whites were necessary when, at most, there were 14,965 persons of Mexican origin. He noted many people were critical of losing sample size power in the dataset. In one year there were fewer than 2,000 Puerto Ricans, about 1,000 Cubans, less than 600 Native Americans, and about 600 each of Chinese, Filipino and Asian Indian subgroups. Other smaller subpopulations weren’t identified. On a single year basis, there was limited use of the dataset for looking at health behavior and status among many racial and ethnic groups. Dr. Hummer noted that was a tremendous limitation, given the field sense in social demography that diversity in race and ethnicity went far beyond umbrella subgroups.
Commenting that sample sizes often were subdivided by age, sex, health outcome and other variables, Dr. Hummer observed that addressing health disparities across a wide range of race/ethnic groups became even more difficult. He questioned whether they could learn much about the health disparities of multiple race groups among subpopulations. One couldn’t do much more than cross tabulations of health outcome by race and ethnic groups standardized for age when considering the majority of race and ethnic groups with these data. But Dr. Hummer emphasized that the Census was only two years ago and a redesign of HIS came in 2005. They were at a point where they could do something.
Dr. Hummer said socioeconomic variables in the dataset had been enhanced, but were still largely basic indicators (e.g., education, income, occupation). In terms of gauging race and ethnic disparities in adolescent, young-adult, middle-aged, and older adult health, there were few indicators contributing to an accumulation of experiences across a lifetime. A big drawback was that this was a cross-sectional survey; but he pointed out that their big boon was the topical questionnaires they could use to get at the accumulation of disadvantages and advantages, at least retrospectively.
Dr. Hummer noted it was also difficult for researchers, but not impossible. People linked with neighborhood characteristics gleaned from the data that came from a census-based sampling frame. Wells and Horme published a piece in AJPH a few years ago describing how to use very small areas to get at neighborhood characteristics. And Felicia LeClaire had done in-house work at NCHS looking at census tracts and measures to get at health disparities. Dr. Hummer emphasized that these tract- and geographic-level data were important in understanding racial and ethnic health disparities, and that larger representations of racial and ethnic minority groups were needed on the surveys.
Again, Dr. Hummer reinforced Dr. Lucas comments, remarking on the lack of other social and cultural variables in HIS, even in the special modules. Noting literature that strongly argued that racial and ethnic health disparities were social, historical and cultural in origin, Dr. Hummer said the relative lack of these variables seriously impeded understanding what lay beneath the disparities in question. For some groups, HIS provided a real sense of the disparities. A number of works in the public health epidemiology and demography literatures tried to get at these disparities using HIS data, but Dr. Hummer said he doubted they’d gain a sense of these key mechanisms with HIS as it was currently constructed. He and his colleagues had worked with the question HIS included on religious involvement in the 1987 supplement, showing that, among African Americans, being religiously involved as proxied by attendance was associated with lower follow-up mortality risks. He suggested that example of religious involvement might lead to a better understanding of why health disparities weren't larger, but variables tapping social networks, family supports, and community involvement were largely unavailable.
Given the size and yearly construction of this dataset, Dr. Hummer acknowledged they couldn’t get at everything. He noted that data that had smaller sample sizes was a way to do this. He also pointed out that they had a number of special topic modules that had treated other topics in depth. He remarked that variables related to stress hadn’t been typically available in HIS and recommended forming expert groups to help put topical things together.
Recalling they’d heard there had been relatively little attention to immigration issues and that growth of racial and ethnic minority populations was largely due to immigration, Dr. Hummer asserted that serious thought was needed because there was more to patterns of migration and health than simply nativity and one question on duration.
Questions and Answers: NHIS
Noting they’d spent a lot of time talking about sample size issues and statistical power, Dr. Newacheck turned the discussion to the quality of the data collected, asking if any validity studies looked at the quality of the data they got from non-English and non-Spanish-speaking households in the HIS. Ms. Lucas said she wasn’t aware of any. But she added that, with the redesigned HIS being done on CAPI, many times the interviewer, while still in the household, would note some particular aspect of the interview (e.g., no one in the household speaks English or Spanish). Many times, an English-speaking member of the household did a fair amount of translation. Ms. Lucas said they hadn’t adequately explored how much the quality of the information was affected.
Dr. Newacheck agreed this was an important issue as they moved toward smaller subpopulations. He said he’d been a long-term user of HIS, but almost changed his mind after watching an interviewer struggle to translate concepts like limitation of activity to someone who wasn’t a native English speaker. They’d made this a national priority. It was important for NCHS and other data collection agencies to consider the quality of the data they got from these households and ways to improve it.
Dr. Hummer said he had a paper coming out that looked at the non-U.S. born population in HIS and mortality follow-up on the self-reported health question. The correlation of self-reported health with follow-up mortality was much weaker among the foreign born. Something about that variable was either due to the language the interview was conducted in or how it was interpreted.
Mr. Handler asked if results from HIS had been matched with NDI Plus, which included data on cause of death that might be related to life style in the home. Dr. Hummer said he wasn’t sure about the distinction between NDI versus NDI-Plus, but he’d analyzed cause of death with the links that the dataset had. To some extent, one could look at differences among racial and ethnic populations by cause of death.
Dr. Lengerich asked if the special topic modules Dr. Hummer mentioned existed or were planned for NHIS. Ms. Lucas said a sample redesign would be implemented in 2005 and NCHS was changing the computer system that the interview was conducted on from CASIS software that Census uses to BLES, which is a different CAPI system for administering the survey. Making that turnover, Ms. Lucas said NCHS was reevaluating parts of the questionnaire and items they’d like to add. The demographers, sociologists, and epidemiologists in the Illness and Disability Statistics Branch that did the primary analysis were interested in seeing modules oriented toward the people and communities they lived in and their experiences. A lot would depend on funding, timing, and the questionnaire’s length.
Going back to the issue of how people from different cultures and ethnicities interpreted questions, even when the translation was good, Ms. Greenberg noted the World Health Organization (WHO) reported huge differences in people rating their health good, excellent or poor that didn’t appear to correlate with equivalent differences in actual health status. She asked who had been looking at objective measures to correlate what people said with what they actually could do and describing scenarios for people to characterize as indicative of good or poor health. Ms. Lucas said these had been suggested, but she didn’t know if they’d be implemented for the next redesign. Their experience with the practical application of the Spanish translation done for the 1998 HIS was that, even with the translation, the field representative had to translate on the side to find a word that was appropriate in that setting.
Dr. Breen remarked that the California HIS, which was modeled on and designed to be comparable with NHIS, might help work through some of these problems. It was administered in six languages and tried to capture information at the county level. She suggested NCI might use the California data and population, which was so important because of its naturally occurring diversity, in its study of the cognitive understanding of these questions in different languages.
Noting that according to an IOM report on data systems, NHIS was the premiere source of data on health, health status and health services use in the country, partly because of its size and longevity, Dr. Breen said it needed and deserved high priority in terms of getting these data right. As they’d heard yesterday, NHIS was used to pick a subset for MEPS, the only survey providing financial information on health service use.
Dr. Breen said one problem she had with NHIS was that she couldn’t look at Asians, Asian Americans or Pacific Islanders, Native Americans, or Alaska Natives with this survey in any meaningful way. She explained that, in screening for early cancer detection, she looked at the whole population. This was a general population survey specific to the age groups for which the screening was recommended. But in trying to analyze even Blacks and Hispanics age specific for screening, in some of the older surveys she hadn’t been able to look at those populations (between 20,000-40,000) by income and education and get reasonable confidence intervals. The confidence interval size around the point estimates ranged from 20-50 percentage points. Dr. Breen emphasized that even this basic data wasn’t being delivered with the survey. She advised the Subcommittee to look into correcting that.
Dr. Breen also said NCI did an atlas of cancer mortality a few years ago and found areas in the country with high and persistent (over 50 years) rates of cervical cancer mortality. They’d used the BIRFIS and NHIS tried to look to see if PAP smear screening rates were lower in those areas, but were completely unable to look at rural distressed areas in this country. As others noted, the PSUs weren’t there.
Ms. Lucas confirmed that the issue of obtaining health measures, SES, or other demographic measures for high-income, high-education, Black and Hispanic persons came up in NHIS’ analyses. She suggested it was partially a function of how the sample was drawn: over sampling was done in high-density, urban areas, where the composition of those groups might be different. She said they needed to look more carefully at the distribution of people by their economic status in different levels of geography, because over sampling occurred where they were most likely to find people--That was the cheapest, most effective way to do it. Doing that made sense in theory, but had an impact on the kinds of people in the sample and what one could do with the data.
Ms. Lucas noted one issue was that the sample redesign was done on a five-year track when the decennial census data became available. Right after the 1990 Census, they’d begun work on the design implemented in 1995 through 2004--By 2004, they were working with a design based on 1990 Census data. A series of population changes had happened since 1990 and had to be dealt with. Ms. Lucas said it was hoped that the American Community Survey, which was intended to provide information between Census years, would supplement surveys like the NHIS that had to do a sample design and in the future they wouldn’t have to continue this ten-year redesign span.
Ms. Lucas said the Spanish translation wasn’t yet posted on the Web site because it was available as a toggle on the computer so the interviewer could switch back and forth between English and Spanish. But it probably could be put up there.
Dr. Coleman-Miller found that morning’s view of surveys difficult: they were extremely expensive and didn’t give the information health disparity experts needed. Acknowledging Dr. Hummer’s frustration, Dr. Coleman-Miller asked if the death certificate in different states could be changed to give more information for the link process. Noting Dr. Hummer had said the reason this survey was so difficult for a minority population was because it would get so big with all those considerations, Dr. Coleman-Miller questioned how to measure “big” in coming up with absolutely minimal statistics for the minority population. Methodologists in the GAO had written texts analyzing the methodology behind surveys that might be helpful and that the government gave RFPs to not-for-profit agencies that dealt with minority populations. Dr. Coleman-Miller wondered if an RFP had ever required a not-for-profit to a particular survey dimension in order to make that RFP successful. Listening to people talk about the surveys, Dr. Coleman-Miller also wondered whether there wasn’t interrelated information.
Dr. Hummer said Dr. Coleman-Miller’s questions were fantastic, but as someone who wasn’t involved in data collection, he couldn’t address most of them. But as a user, he said he didn’t need this survey to be 60 percent White. Cutting Whites back might be a simple answer. He said his biggest concern was that they got to the people they needed to find. The biggest problem he faced with HIS and links to NDI was that the “death matches” weren’t as good for most minority groups.
National Health and Nutrition Examination Survey
Dr. Curtin said the National Health and Nutrition Examination Survey NHANES used to be a health examination survey and went back over 40 years. A nutrition component was added in the early seventies and a series of national surveys had been done for 30 years. The middle one, Hispanic HANES, was a special effort to get at estimates for the Hispanic population. The last data released by NCHS was for the NHANES III survey, which finished off in 1994. The current NHANES survey will probably release data for 1999-2000.
Unlike a typical interview survey, NHANES starts with a screener interview for over sampling. A household interview was directly related to the HIS and used some of the same questions. Dr. Curtin said the sample design issues already noted with HIS were even more apparent in a six-year survey, which only had 81 PSUs and 30,000 examined people. However, he noted this was a highly screened population. He noted everything Dr. Lucas had said about race and ethnicity and basic questionnaire format applied to NHANES, but that the real key to what made NHANES a rich, invigorating data source was the mobile exam center. A series of four trailers were set up, usually in a K-Mart parking lot. People were brought through for a series of medical examinations including a cardiovascular fitness test, bone dense geometry, blood draws, and other evaluations to estimate the number and prevalence of people with disease and risk factors, monitor trends and prevalence, analyze risk factors, study relationships in diet health, explore public health issues, and establish baseline information.
NHANES III and HANES were designed to get at three population groups: Mexican Americans, non-Hispanic Blacks, and non-Hispanic Whites. NHANES III was about 30 percent Mexican Americans, 30 percent Black and 40 percent non-Hispanic White.
It took ten years to plan a protocol, get it in the field, collect six years of data, and do data clean up--most researchers didn't want to wait that long. The survey was designed to be a “representative sample” on an annual basis, but Dr. Curtin noted there were severe limits and he said a likely alternative was to do more flow bases and release it every two years. NCHS anticipated that a public use file for HANES 1999-2000 would be an Internet data release in July. Multiple files rather than a single data file with five or six thousand variables would be released. Issues with the data release involve the new OMB guidelines on race and ethnicity and a severe confidentiality concern. As the sample size diminished in a two-year data release, new foibles emerged around confidentiality that further restricted release of the data.
A major strength in the survey design was the ability to control the selection to oversample minority populations and do stratification in screening. Another strength was that every stage of the multi-stage sample was selected at random. Dr. Curtin cautioned that, without a large sample, randomization could cause a problem. Over sampling Mexican Americans and African Americans drove the sample in a certain geographic area and left the remainder “largely undisturbed,” causing a problem when dealing with a variable related to another area (e.g., rural health). When one was only selecting 15 out of 3,000 counties a year, there was a great tendency for selection bias.
NHANES was designed around a design effect (a measure of the inflation in the sampling error due to clustering, over sampling, and differential rating, usually greater than 1--HIS design effects range 1.2-1.5) of 1.5, but Dr. Curtin noted it actually rose to greater than 2. In order to get an absolute 10 percent difference (e.g., up to 30 percent versus 20 percent, with 95 percent significance level, 90 percent power) one needed a sample size of about 420. Even though the NHANES sample was a very small sample relative to other national samples, Dr. Curtin noted it was highly stratified and selected for Mexican American, non-Hispanic Black and a residual of White from 1999. Another selection strata for low income Whites was added in 2000. There were specific age groups by sex and the survey was targeted to the very young. NHANES was done under contract to Westat and was linked to the HIS PSUs for 1999 through 2001. An independent design had been drawn for 2002 that no longer linked at the HIS PSU level (designed on 1990 census information). HANES was limited to only 15 PSUs per year because of the size of the field staff and the operational considerations of the two MECs. More than one person per household was sampled and the surveys were screened for race and ethnicity.
Dr. Curtin reported start-up problems in 1999 and some PSUs selected early had historically low response rates. He said response rates for 2001 looked better and tallies would improve as they continued to combine the surveys. But he noted it created a problem because a lot of outreach was done to increase response rates and it was easy to determine where the 15 PSUs were. Anyone who had a data tape with pseudo-PSUs on it and knew their characteristics could probably identify the geographic unit. The Disclosure Review Board resisted releasing any data on the 400 people in such samples, and so no pseudo-PSUs for HANES 1999-2000 will be released. NCHS was working on a jack-knife estimate of variance, using 52 replicates related to the sample design that give approximate standard errors for what one would get under the PSU design. NCHS would probably present a scientific paper at the July Data Users Conference. In addition, because the sample size was too small and, because of the rule of disclosure avoidance, there couldn’t be more than three people in a cross-tabulation, Dr. Curtin said the type of socioeconomic status variables in terms of education, income and race/ethnicity would probably have to be limited.
Dr. Curtin reported a problem dealing with the Disclosure Review Board about households and family links. If an adult and child were selected into the sample and the dataset was on the Internet, the parent could find their informational unit and identify the child’s sensitive interview on sexual history and blood tests for STDs and herpes. Another problem stemmed from field start-up costs for 1999 that resulted in 12 mostly White-and-Black PSUs that underrepresented Blacks. Although they were weighted properly, they still underrepresented. The sampling practices were changed in 2000.
He said NCHS was also dealing with the issue of how to handle the OMB guidelines. Because the survey used HIS questions and asked ethnicity first, Mexican Americans tended not to report a race and some 20-25 percent of the HANES sample had unknown race associated with it. If you use a census type coding scheme that they used in their public law file, which would assign a no known, but Mexican American to another single race, then 1,892 would be put into a race category and the survey would severely underestimate the number of Whites and Blacks. In addition, the 441 multiple race would probably decrease; under the current coding scheme, White-and-Mexican-American entries were coded as White and other single race and then as multiple race. NCHS was still working out coding rules for release of this data.
Dr. Curtin discussed foibles of analysis with this data, emphasizing the small sample size and the considerations in tradeoff between bias and variance. The small sample numbers and randomization issues could lead to large sampling errors and outliers. And limited geography could result in unforeseen selection bias. It was important to compare external datasets with results of the NHANES survey, looking for bias or sampling error. He noted a move towards using standard normals versus T tests in estimating sampling errors and that T tests gave larger confidence intervals and reduced effective sample size (e.g., in the HANES survey there were at least a hundred Mexican Americans in the PSU; but only six in the NHANES 1999 and five in 2000; several thousand sample persons over 15 PSUs might be clustered in a few PSUs). Dr. Curtin emphasized being extremely careful with the design-based estimation in terms of influential values and sample rates and in dealing with degrees for the variance estimations.
Noting that given these limitations of NHANES design, it was difficult to get at specific sub-domains, Dr. Curtin discussed a plan for a more flexible data collection approach called Defined Population or Community HANES that moved from community to community in a large recreational vehicle or 18-wheeler to target the population defined in terms of their race and ethnic status or a health outcome.
National Health and Nutrition Examination Survey User
Dr. Sempos explained that HIS consisted of a series of periodic cross-sectional studies that extend back to the first survey in 1960. Assessment and monitoring of minority health status was always a key feature of NHANES. Presently, NHANES produced estimates for Mexican Americans, non-Hispanic Blacks and non-Hispanic Whites, covering some 85-90 percent of the U.S. population. The principal reason for the survey was to assist federal agencies in the development and monitoring of public health policy. The goal of the NHANES was to produce national estimates; establish a series of health status indicators used to assess and monitor national health status; develop, monitor and help modify federal regulatory policy; and examine the association between health status indicators and certain selected conditions and sometimes (to a much lesser degree) diseases.
Two types of data were collected as part of the health surveys: (1) self-reported aspects of health including race and ethnicity, income, nutrition, health behaviors, medical history and, most importantly, (2) physical attributes of individuals that could be measured such as height, weight, infectious disease exposure, chronic disease risk factors (e.g., blood pressure, cholesterol and obesity), environmental exposures, and the presence and absence of a few selected sets of diseases (e.g., arthritis, periodontal disease, diabetes, coronary heart disease, and gene frequency data. Dr. Sempos said measurement of those physical characteristics was the strength of the HANES program and what prompted virtually every federal agency in the Public Health Service to become involved in the design and support of the survey.
Dr. Sempos said the HANES program assessed national mean levels and distributions of health status indicators. It also got national prevalence estimates or percents of the U.S. population with certain often unhealthy characteristics and health status indicators. Using successive NHANES, one could produce national trends in health status indicators. Since 1984 the surveys documented health disparities between African Americans, Mexican Americans and non-Hispanic White Americans. It also could document potential areas of unmet medical need associated with health disparities. Dr. Sempos emphasized that, from the beginning, the goal of the NHANES program had been to describe current levels and trends in health-based status indicators by age, sex, race and ethnicity. He said it described those national levels extremely well. What NHANES didn't do well, except in selected cases, was disease diagnosis. And it wasn’t good at producing subnational, state or local estimates. Most importantly, it didn’t explain how health disparities came about or how they could be reduced.
Dr. Sempos reflected that, with improvements in health, especially declining cardiovascular diseases since the late 1960s, increased life expectancy, and realization and, to a certain extent, acceptance of racial and ethnic diversity, there had been an increasing interest in finding and eliminating illness and disease in racial and ethnic minorities and in medically underserved populations of the U.S. There also had been a desire to understand causes and solutions necessary for eliminating racial and ethnic disparities. These, along with advancements in computer technology, resulted in an increased interest in a desire for health status data at the state and local level and for coverage of Americans not specifically oversampled and studied in the NHANES survey.
In order to reduce data deficiencies, Dr. Sempos urged the Subcommittee to look at the report chaired by Anthony DeAngelo and Dr. Carter-Pokras that set out in detail data deficiencies in the U.S. data collection system, including NHANES. Dr. Sempos concurred with Drs. Curtin and Kington that the community HANES, was an outstanding way to supplement data deficiencies of NHANES. He hoped the Subcommittee would encourage federal agencies to set out a schedule for different communities in cooperation with the Indian Health Service and other federal agencies to try to identify communities and populations that would be examined on a regular cycle basis. He also recommended increasing efforts to make NHANES surveys into cohort or follow-up studies. (Starting with the 1971-1975 NHANES I study, there had been direct follow-up of all participants examined.) But although it was an excellent cohort study, the NHANES I survey didn’t include at baseline many risk factor measurements in disease exposure pertinent to developing public health policy today. Dr. Sempos encouraged the Subcommittee to recommend that the NCHS expand development of follow-up studies as a regular ongoing product of the NHANES program.
Dr. Sempos mentioned another aspect of federal data collection could be improved. He noted the vital statistics data (America’s birth and death data) obtainable on the Internet as CDC Wonder was only available for Black, White and other and, although the capacity existed, data for the Commonwealth of Puerto Rico and other trust territories and protectorates wasn’t necessarily included in the CDC Wonder data. He suggested recommending that CDC work out any problems and include that data.
Dr. Sempos concluded that the NHANES survey did what it was designed to do extremely well and what it wasn't designed to do extremely poorly. He said, hopefully, NHANES wouldn’t be changed dramatically because it served a need well, but auxiliary methods would be developed to supplement its deficiencies.
Questions and Answers: NHANES
Noting national data was only available from NHANES, Mr. Handler asked how much it cost to run. Dr. Curtin gave a ballpark estimate of $35 million just for the data collection contract and staffing at NCHS and NIH levels. Mr. Handler said increasing the sample size might double that number. Dr. Curtin explained that there were two mobile exam teams. They could either add one MEC team, boosting the cost 50 percent to increase the sample size by half, or they could double the teams, the cost and the size. He said the cost was why the Community HANES survey, which was more limited in type and scope of data collected, but cheaper and could go more places more often was interesting.
Dr. Lengerich said he hadn't realized the prevalence of the hemochromatosis gene had been mapped through NHANES and released. He asked about plans, considerations and confidentiality issues around NHANES and genetic issues. Dr. Curtin explained that when NHANES III was collected nothing in the informed consent document (done at the beginning of the survey) stated samples might be used for genetic testing (the standard at the time). The institutional review board decided White blood cells could be spun off excess sera, immortalized, and DNA samples taken, but only in an anonymous manner through a prescribed protocol procedure whereby protocol and funding went through a lengthy process. The old NHANES III samples had a process for doing that and other ongoing studies used that baseline material. The informed consent process had been changed and genetic testing was now part of the informed consent process. They might not have to do it anonymously, but there were severe data disclosure issues involved in any public release of such data. There were ongoing discussions about how best to use this. Dr. Curtin said he and his staff were going to discuss the problem the next day with the Center for Environmental Health.
Ms. Lucas clarified that HIS budget was $12 to $18 million a year for the core questionnaire with additional monies for topical module supplements. Asked what NHANES and HIS did to help minority researchers access and use these data, she mentioned the university visitation program and that people requested help from HIS to use the data for doctoral dissertations and master's theses. Years ago, NCHS had a minority research grant program. She noted the California HIS May conference dealt with public health issues related to the African American population and that a HIS representative would demonstrate using the dataset to examine public health issues for African Americans.
Ms. Heurtin-Roberts asked how they could rectify that, while NHANES could document health disparities, it couldn't elucidate the causes of or processes leading to them. She asked whether it was feasible to incorporate social, cultural, or economic contexts--or were they better off linking NHANES with supplemented data? Dr. Sempos said that was a difficult, if not impossible, question to answer. To a certain extent, social and cultural factors that determine health were on a plane above physical characteristics. It had been difficult, listening to some presenters yesterday, to try to include variables to model and tease out social and economic factors leading to disease causation. Another problem was that HANES surveys were cross-sectional, so cultural status and health status indicators were measured at the same time. With follow-up surveys, one might elucidate why disparities came about--cohort studies might better identify health ramifications of disparities and the need to eliminate them.
Dr. Curtin commented on content and respondent burden. People wouldn’t sit long enough to answer all the questions they’d like to ask. Content, like geographic and state estimates versus racial and ethnic disparities, traded off in survey design. NCHS had an ongoing process for gathering information from people, setting the agenda and survey content, but Dr. Curtin said it came down to a tradeoff with what public health researchers felt was important. It wasn’t up to them to determine content; they had to react to what was of interest to users and the people.
Dr. Sempos pointed out that a great deal of socioeconomic status data was included in the collection with NHANES. But when one looked at those socioeconomic status indicators at a national level, cross-cutting trends occurring regionally and locally canceled out: e.g., cardiovascular disease mortality overall declined for Black Americans and others, but in Mississippi cardiovascular disease mortality for Black men was rising. Even when documenting unmet medical needs nationally, the aggregate look sometimes obscured problems at state and local levels.
Dr. Mays asked if a user who wanted social support and had funding from a foundation or NIH could buy survey time. Dr. Curtin explained that a good portion of the CDC NCHS budget came from NIH and others through a mechanism that supported various aspects of health through reimbursable agreements. The money usually had been appropriated because Congress determined a need, targeted the money into somebody's budget, and turned it over to CDC NCHS to study the problem. Academic and public health communities had ways to establish that certain variables were important to the study of health. Dr. Curtin noted something else would probably have to come out of the survey, unless a three-or-four minute bank of questions could be slid in. There would be a whole review process.
Dr. Kington noted a precedent for an external group (a combination drug company and foundation) paying for an additional question justified on scientific grounds. Data would be released to the public. Dr. Curtin reiterated Dr. Kington’s observation that the additional question had to be scientifically valid and of interest to the general public. Dr. Newacheck commented that the Robert Wood Johnson Foundation paid for an access supplement to NHIS and the Gerber Foundation paid to include an early childhood survey in SLATE. Dr. Sempos noted outsiders could propose studies to NCHS for using the excess sera from NHANES III. The procedure for anyone in the public to propose content changes was posted on the NHANES web site.
Noting that the Ethical Legal Social Responsibilities Issues component of NIH was reflecting on sampling issues, Dr. Mays asked for NHANES thoughts about samples collected from the targeted communities they’d just talked about in terms of health disparities. Dr. Curtin explained the institutional review board process at NCHS looked at the cultural diversity and sensitivity relative to the sub-domain of interest. But he said nothing should be sent to the IRB unless NCHS and CDC agreed it was ethical and there weren’t problems. As Chair of the IRB, his position had been that people shouldn’t submit anything they weren't willing to put into the field. IRB shouldn’t break new ground. His personal preference was to follow the research community and its guidelines.
Dr. Breen said NCI was interested in supporting community HANES and he asked if it was still alive. Dr. Curtin said the issue was what could be done. If they did a community NHANES, they had to take the cross-sectional HANES out of the field for a while. He wondered if both might be done together or if there could be a mixture of the two. Three types of studies were under consideration: repeated cross-sectional, longitudinal follow-up, and community HANES. Pointing out that 25 percent of the NHANES budget came from external sources, Dr. Breen suggested that someone choosing to support a particular study could become a driving factor. He noted for 10 or 15 years there’d been interest in community HANES as a means of getting at special population groups. It was considered when there was a health initiative along the U.S.-Mexican border and for Manhattan following September 11. Dr. Curtin said the issue was alive and well, and would probably be discussed in future planning for NHANES cycles.
Ms. Heurtin-Roberts expressed concern that the surveys they’d discussed could document but couldn’t explain health disparities. If they couldn't explain them, how could they address them? She remarked that a survey wasn’t the only appropriate mechanism, but she thought they’d go further if they had the contextual data and did more than describe. Other means could get contextual data, but the broad scale data only a survey could provide was necessary to move things forward. Dr. Mays noted the Subcommittee would have to consider issues about putting everything on the back of these surveys at once and comment on the different perspectives of what the surveys could do.
Ms. Greenberg noted reports of survey work on questionnaire development that tailored questions to the individual a level beyond skip patterns, with different responses triggering different sets of questions. They couldn’t ask most the different types of variables, but perhaps they could ask certain ones to people who responded a particular way. She asked if either HIS or NHANES was looking at this approach. Dr. Curtin said it would have to be worked out before it could be implemented on a national basis. Standard in-the-box thinking about a dataset was that the same questions were asked of the same people and then variables on a data tape were properly weighted and analyzed. Using different flexible means to collect information and summarizing it in a consistent manner could alleviate measurement error problems. And adaptive sampling techniques (a variation of the NCI friends-and-family situation) were being developed. Many people were looking at how to sample rare, elusive populations. There was a lot of survey work in translation of instruments in the different concepts. And there was concern in the survey community about “interviewer imputation” where the field was doing something totally different from the protocol. In the type of situation Ms. Greenberg mentioned, Dr. Curtin said he would be concerned that the interviewer might go through that instrument differently than they should under protocol.
Policy Discussion
Dr. Kington noted NIH as a whole and each institute and center had developed a research plan to address racial and ethnic disparities and that the new National Center for Minority Health and Health Disparities had primary responsibility for pulling together the plan’s final version and monitoring progress towards meeting its objectives.
He said the data sources they’d discussed today were important to NIH for two reasons: (1) they conveyed the magnitude of the problem of disparities in health and health care and helped in setting priorities for a research agenda; and (2) despite a more policy-oriented purpose, they gave insights into fundamental causes of these differences and, ultimately, could point the way toward likely places for intervention.
Dr. Kington said it was time for a paradigm shift in thinking about obtaining data on the increasingly diverse population. He said the single most important problem echoed that morning was how to obtain data on smaller subgroups. He said the idea that all data needs could be solved by over sampling reflected poor understanding of how diversity played out in the country--He didn’t know that over sampling even served larger subgroups (e.g., African Americans) well. Clearly, there was an increase and interest in data on smaller, often geographically concentrated, subgroups. Ms. Lucas had alluded to how it was too costly to screen the entire country for a small highly geographically concentrated population. Dr. Kington suggested bringing this fundamental idea found within community HANES to another level. He proposed two parallel tracks of data collection; a core tract that got at the large racial and ethnic population subgroups (e.g., non-Hispanic Whites, African Americans and Mexican Americans) and another tract in which data were collected in a similar way, allowing comparisons with the national data, on a mosaic of smaller groups; Native American tribes, Asian, and smaller Hispanic subgroups.
Dr. Kington recommended that the American Community Survey might provide information that allowed running both tracts. But he cautioned that the idea of one or even five surveys that served their needs and really described the country’s complexity was ridiculous. The country wasn’t demographically composed that way now or in the foreseeable future. He urged the Committee to move away from simplistic thinking about data sources and toward a parallel tract model that got at both large groups and smaller heterogeneous ones highly concentrated in specific regions.
Noting a number of comments about the need for more complex data on a range of different dimensions (especially social and behavioral factors) for racial and ethnic minorities, Dr. Kington recommended moving to a more rational planned system of cycling through supplements already used with different surveys in various ways. NCHS and NIH recently had a workshop aimed at developing a series of core, expanded measures looking at how to measure economic status. For example, evidence suggested that often one's net worth was more predictive of one’s true financial resources and health status than income in any year). The workshop pulled together leading economists and epidemiologists to consider: (1) core measures to guide data collection for major health surveys and a series of supplemental datasets, and (2) how data collections might periodically be expanded along various dimensions to achieve, over time, a more complex picture of what was going on. Dr. Kington suggested holding similar discussions on community contacts, stress, social support and other issues on the list. Many people were thinking about these things, but no one was putting it altogether in a comprehensive, coherent plan.
Dr. Kington noted there were serious problems with funding. Having “been on the Hill,” he acknowledge that trying to make data collection “sexy” was hard to do. But he encouraged everyone to think about the real value of these datasets and do a better job of demonstrating in concrete terms how extraordinarily important good data on health across a wide array of racial and ethnic groups was to solving public health problems. Noting the cost of collecting data rose dramatically and that a number of big surveys had large cost overruns, he expressed confidence that they could all do a better job in trying to get farther with the money they had. He suggested major efficiency gains could be made in how data was collected. Response rates in the broader community were dropping and expanding dissemination of information on how to do this in a better way could lead to efficiency gains across data collection efforts.
Dr. Kington noted that they’d had a lot of discussion about the OMB directive, which was an important change. However he pointed out that people forget one word in that directive: "minimum." Everyone knew what that meant, but that hadn't affected what anyone did, because everyone automatically went and collected the minimum--The HIS question on primary racial identity was an example. Dr. Kington focused on other ways to expand this notion of race and ethnicity beyond the OMB core dataset. He noted growing interest in asking people not only what they thought they were, but what others thought they were, because discrimination in the health care systems manifested itself based on someone else's judgment. He encouraged asking this in the multi-racial categories.
Noting a friend remarked, “Sometimes I feel Blacker than other times,” Dr. Kington suggested the need to explore an expanded and dynamic notion of race as an exposure variable. If you spent your entire life in a segregated community, never having contact with Whites, you might have a very different experience of race and very different implications for what race meant for your health status, than if you constantly shifted between a situation in which you were in the majority and another in which you were in a small minority. Dr. Kington said he could relate this to his own day-to-day life and that it pointed toward thinking creatively about narrow dimensions of race and ethnicity. Saying race was a social construct wasn’t good enough. He encouraged the Committee to go beyond simple questions and explore perceptions of race, in addition to self identification, and think about race as an exposure variable, adding a dynamic dimension to race and ethnicity that led toward understanding race within the context of a particular social setting. Dr. Kington noted there were interesting scientific opportunities ahead and that meetings like this were extraordinarily important in moving them to realize what they would need in 20 years and how to lay the foundation for those scientific resources, both in terms of human resources and scientific data and information, that allowed them to get to a different place.
Discussion
Noting the front page of that day's Washington Post reported a terrorist threat “sitting out there,” Mr. Handler asked how the data system would provide information long-term on the health status of the population if there was another event--biological, chemical or nuclear. Dr. Kington noted an overall anti-terrorism task force was a subgroup of the National Council on Science and Technology and six working groups were being formed. Dr. Kington co-chaired the group looking at behavioral, social, educational, and scientific needs to deal with terrorism, and he said these were the questions efforts like theirs were addressing. Getting better quickly was an issue reflected in the President's budget. Mr. Hitchcock reported that Dr. Claire Broome from CDC would speak to the HHS Data Council the next day about what had been learned since September 11th in terms of the public health data. Dr. Kington noted RAND Corporation was retained by the Department of Defense to look at these issues and planning. Dr. Mays added that the Committee would also take up this issue at the February 26-27 meeting.
Asked about efficiency gains he’d remarked were possible in the surveys, Dr. Kington said he was less concerned about groups like HIS and HANES because they’d done this a long time and knew how to do it well and relatively efficiently. But a lot of data collection efforts underway (NIH funded a considerable number of large surveys on a regular basis) weren’t represented at this meeting. He said he believed there were ways to translate knowledge gained from large surveys done for years to the broader scientific community. One possibility was to develop regional clusters of survey research methodologists and teams so there was a reasonably efficient way of translating gains and efficiencies across the country. Grantees and federal agencies could tap into those coalitions as mechanisms for conducting research. Dr. Kington said it got harder and costlier to do these surveys and they needed to think creatively. He expressed concern they weren’t making sure that gains in methods (particularly in terms of retention and recruitment) could be disseminated broadly. Noting one of the most important determinants of response rates was the interviewer’s skill, he suggested professionally incorporating this cadre in a more standard way so researchers gained access to experienced household interviewers. Dr. Kington urged them to think seriously about how to do a better job with resources. Telephone surveys were increasingly difficult with the large number of cell phones; the whole notion of a household survey was lost when everyone had their own phone. They needed a mechanism for figuring out how implications of ways the world was changing affected their surveys and studies and reaching economies of scale in broadly disseminating that knowledge.
Dr. Newacheck asked what NIH was doing to move away from a sole focus on individual level characteristics (e.g., determinants of health in NHIS) and add contextual information so the interviewer might collect information on neighborhood conditions. Dr. Kington noted awareness in the medical and public health community of the increasing evidence suggesting community characteristics measured in various ways were predictive of health outcomes above and beyond individual characteristics. The problem was that the ecometrics wasn’t near the level of the science in psychometrics and statistical characteristics of measures at the individual level. Dr. Kington said another issue was major statistical problems had been ignored in the analyses. Clearly, people weren't randomly distributed in communities with certain characteristics, but statistically that was a difficult problem and economists and demographers had done more to statistically tease it out than the epidemiological community. Dr. Kington noted workshops aimed at pushing ahead methodologic work on measures and taking the fiscal method another level to deal with the simultaneity bias issue.
Multiple Race Data Use
Dr. Smith said the discussion of multiple-race measures brought up new technical specifics but he emphasized reoccurring basic themes: (1) race and ethnicity are social constructs; (2) how they were conceptualized, defined and measured depended on social/legal conventions and scientific and policy purposes for which the information was to be used; (3) race and ethnicity were complex variables; (4) the information collected on race and ethnicity depended to a notable degree on the way they were measured.
He noted multiple-race measures differed from traditional, one-race measures. First, multiple- race measures recorded more members of all racial groups as secondary and more remote racial identifications were drawn in (e.g., Whites were 75 percent of the population on one-race identifiers, 77 percent in multiple-race identifiers; Blacks moved from 12 to 13 percent; the number of Native Hawaiians or American Indians approximately doubled). Second, multiple-race questions allowed expression of secondary or tertiary identifications, while single race questions forced choosing between them. Third, except for the Census, many individual multiple-race categories (e.g., a person who was Black and Asian or Asian, White and American Indian) were too small for useful analysis. Dr. Smith noted data could be pooled across surveys to take more precise measurements of racial and ethnic groups that otherwise were too small to be usefully examined. Alternative classifications of racial groups (e.g., one-race Blacks versus all Blacks) could be used to determine that results were robust and racial differences substantiated. Similarly, one-race identifiers could be compared with multi-race identifiers to see if race effects dissipated in moving from single to mixed, adding further evidence racial differences were real. Fourth, while multiple-race measures capture a richer, more accurate picture of people's ancestry, they didn’t necessarily get full, complete information. People omitted racial ancestries when their background became too complex. Even moderately complex racial and ethnic backgrounds were often not fully known or reported by informants, even spouses. And ancestry in lower status racial and ethnic groups often went under reported.
While the census found that 1.9 percent of adults or 2.4 of the population as a whole reported multiple race, Dr. Smith said 5.5 percent of adults in the 2000 General Social Survey (GSS) indicated more than one race. He said the GSS figures were higher because: all racial classifications were self-reports and because the GSS format facilitated multiple-race dimensions more than the Census’s mark-all-that-apply approach. Although the Census and GSS found only a relatively small number of multi-racial groups, Dr. Smith noted the number already was larger than many delineated races and multiple-race identifications would grow over time. First, there was more intermarriage across racial and ethnic lines, and declining social barriers to intermarriage, and the changing immigrant mix would further increase the number of people with multiple racial ancestry. Second, as people become more exposed to censuses, administrative records and other data sources using multiple-racial questions, they’d become used to reporting their full racial and ethnic background. Third, as the new multiple-racial standard gained wider social acceptance and the convention of identifying with only one race eroded, more multiple racial would be mentioned.
Noting one of the challenges of switching to the new multiple-race format the Census and OMB adopted was achieving comparability across data sources, Dr. Smith predicted the problem would be minimized as federal and non-federal surveys followed the pattern of Census–led changes and adopted an equivalent standard. He noted this wouldn’t help with comparisons to historical records such as past Censuses and birth and death certificates in the vital statistics system. The challenge was to collect data that captured both the accuracy and richness provided by multi-racial data, while allowing straightforward comparison to data collected under the old standard. The census form (a self-completion form asking “mark all that apply”) gives no indication of primary identification. Drawing upon research indicating that people mentioned their primary racial and ethnic identifications first, GSS adapted its multi-racial question to record multiple racial selections as first, second and third mentions. It then explicitly asked for primary identification.
Dr. Smith reiterated that race and ethnicity were difficult variables to measure, given their changing social definitions and the complexity of people's ancestry. Multiple-race questions improved measurement by providing responses that captured people's ancestry more accurately and by measuring a socially recognized and growing segment of the population. But the new multiple race standard also caused problems. The census item was: (1) ill-suited for comparisons with other data sources using the traditional, one-race standard, (2) multiple-race questions added even more small groups, (3) and multiple-race measurement was sensitive to the precise way in which items were worded and administered. Dr. Smith emphasized that measurement challenges weren’t reasons for avoiding collecting multi-race identifications, merely indicators that race and ethnicity must be measured carefully.
Questions and Answers: Multiple Race
Mr. Handler said the problem with the multi-racial identifier on the 2000 census was it couldn't be projected outward for the Indian population because births and deaths were collected the old way. If one race was reported as American Indian, the count at the national level for 2000 was 2.5 million people; a multi-racial identifier with one race American Indian produced 4.1 million people. He noted variability and proportions changed state to state and no denominator existed to calculate birth and death rates when the 2000 data became available for the American Indian population. Mr. Handler said Indian Health Service always had a problem including American Indians in survey data because they were 3/10ths of 1 percent of the national population. Dr. Smith noted there were considerably different levels of reporting depending on whether one asked a race or an ethnicity question: many more people said they were ethnically American Indian. American Indian was the race or group named most as a second mention for both Whites and Blacks. Dr. Smith remarked there was also the issue of whether it was important to capture tribal distinctions, which fractured a small group into dozens of legally or socially recognized groups.
Ms. Lucas reported there had been significant resistance to asking about the existence of a primary identity and that HIS’s ability to keep it on the survey was up in the air. Having revised the standards to allow reporting more than one race on the census, OMB considered it almost a violation of the standards to turn around and ask people to pick one. Ms. Lucas noted a fair amount of literature written by multi-racial interest groups indicated circumstances under which multi-racial people had a primary identity, but the prevailing viewpoint was that surveys wouldn’t be able to ask this on a long-term basis. She pointed out that most users of the data felt the identifier was an almost essential tool in teasing out what happening with multi-racial reporting.
Dr. Mays remarked that they’d begun to identify the real challenge and what the Subcommittee will do in thinking through these issues. The group would meet on February 26, review what had been presented and decide about some of the themes, needs, and challenges.
Discussion
Dr. Newacheck proposed the Subcommittee think about the framework for their report, how they’d structure it and cull through all the information. Noting that with everything in “this big global pot,” they lost a lot of information about context, Ms. Heurtin-Roberts suggested that Dr. Kington’s idea about parallel tracks with one focused on smaller, more regional studies might also get them the better contextual data Dr. Newacheck requested. Dr. Mays agreed with both suggestions.
Dr. Lengerich said sampling more or differently to capture more populations was a common theme. Another was what data should be collected to identify individuals. A third was considering contextual variables for the different surveys. Training and education in statistical methods were subpieces. Ms. Coltin said she’d heard another was tradeoffs; in a zero sum game, were they surveying too many non-Hispanic Whites or not enough? Dr. Lengerich agreed tradeoffs were important and might be the bottom line. Noting they’d discussed the Department’s new effort to access the data from the data system, Dr. Carter-Pokras suggested they consider how they shared the instruments used to collect the information because significant money had gone into them. She also pointed out that doing what they could to access the instruments in other languages in which they were translated could save a lot of money. And she said a doctorate student at George Mason University wanted to convey how useful the surveys were for her dissertation. Dr. Mays said that in putting this hearing together she’d become concerned about the paucity of people with expertise in using these datasets and the background, technical skills, and commitment to continue. She noted a training need that she said was alluded to in some of the questions raised today.
Ms. Lucas commented on the fair amount of diversity within the White population that shouldn't be ignored from a health standpoint including a population of foreign-born persons who self-identified as White and had health characteristics quite different from the total U.S. White population. One could tend to think of "White" and "Black" as generic, all-encompassing terms, but Ms. Lucas emphasized the diversity they contained shouldn't be ignored in making decisions about changing sampling in order to look at other population groups.
Mr. Hitchcock teased out these thoughts about training, suggesting they should also encourage more collaborative efforts between folks like Ms. Lucas and Dr. Kington, applying the skills “out there” and other researchers to get a better handle on what’s happening. He said yesterday, listening to the National Survey on Family Growth, he’d thought about pulling in the welfare side of the Department to talk about the next steps.
Mr. Hitchcock noted one percent of the Department's discretionary budget went toward data collection. Noting if they made recommendations to expand, it would cost even more, Mr. Handler said to keep in mind possibilities of cutting back on certain things. He suggested there might be outside funding sources. In Montana, the Blackfoot tribe had said the 1990 census didn't count them properly and paid for the Census Bureau to conduct another survey. Noting the presence of a representative from Montgomery County, Maryland, Mr. Handler suggested the county might want a HIS conducted and be willing to pay towards that. Mr. John Park said if the county had the resources, they would even pay to get the information, but the county felt it had given all available resources and shouldn't have to pay more. Mr. Park said hidden below all these racial and ethnic issues were the issues of migrant workers and refugees. Refugee characteristics changed with world events, but the refugees of one country, e.g., El Salvador, were very different from people actually living in El Salvador. And the disparities in the major groups of people in certain ethnicities also differed. Korean by birth, Mr. Park noted that people who immigrated from Korea during the Korean War, were different from Koreans coming now. He questioned that issues with migrants and refugees were considered enough.
Dr. Coleman-Miller asked the Committee to invite students from historically Black colleges to come and watch and learn as the members worked to switch this paradigm, as Dr. Kington suggested. Ms. Monica Lathan, American Public Health Association, applauded the idea of a collaboration or regionalization and standardization of data collection so that they could obtain a more comprehensive dataset that could be used in multi-organizations and multifaceted ways. She encouraged the idea of training people to become more skilled in interviewing and more receptive to the communities. And she said the most important thing Dr. Kington mentioned was “What you think you are” and “What others think you are.” She said her heritage included American Indian and White, but she presented as an African American woman. Nobody saw all the other things that made up her culture and heritage.
Mr. Jim Dawes, Asian and Pacific Islander American Health Forum, reported that at the Intercultural Cancer Council Conference an official from NCHS indicated that, in order to calculate national mortality rates, they would create a standard denominator for the multi-race people. He supported using other studies to learn about and measure health disparities, and he recommended that Healthy People 2010 accept those as valid data upon which to measure and create goals on health disparities.
Mr. Walt Faye expressed the D.C. Department of Health’s concern as it moved into HIPAA compliance that the standards for privacy of individually identifiable health information would be a constraint to collecting the data needed in making decisions.
Mr. Katz remarked on the need to routinely collect language preference data (a proxy for breakdowns of the subpopulation), noting reliability and collection were low. And he said linkage had to be made between disparities and culturally and linguistically appropriate services. Verbatim translations, still used by many government organizations, didn't work yet. Translations had to be culturally appropriate. A Horizons Project at CMS was totally dedicated to looking at culturally and linguistically appropriate services, yet even some of their recommendations couldn't make it through the bureaucracy because of the low percentages of population.
Mr. Katz concurred with Dr. Kington: the data had to support breaking down populations into smaller parts of the country. Some 70 percent of Asians were in ten American cities. Efforts had to be data-driven in order to evaluate the effects interventions could have. A broader look was needed that included language and accounted for how class affected disparities.
Thanking everyone for his or her comments, Dr. Mays said she would take up Dr. Coleman-Miller's charge. She noted that part of the Subcommittee’s job was to work in partnership with each of them, giving back some of what had been asked for and inviting help in this team effort. Acknowledging everyone’s contributions and expressing hope for ongoing communication and collaboration in this team effort as they continued to hold hearings, Dr. Mays concluded the meeting at 12:35 p.m.
I hereby certify that, to the best of my knowledge, the foregoing summary of minutes is accurate and complete.
Vickie Mays
/s/ March 24, 2003
Chair Date