[This Transcript is Unedited]
Hubert Humphrey Building, Room 800
200 Independence Avenue, SW
Washington, DC
DR. MAYS: We're going to get started this morning, because as you can see, we have one of those exciting, but packed days. So we are going to try and keep us as close to schedule as possible.
What I would like to do is to welcome everyone who is here, and also remind people that we have people out on the Internet. So I also want to welcome them, and thank them for joining us for the session today. The only implication of that is that we also have to speak into the mike so that as we transmit, they can participate in terms of hearing us.
What I would like to do is to start by having a set of introductions so that you know who we are, and as well as finding out who is here with us. And then we will open with some remarks about what our task is today.
I am Vickie Mays. I'm at UCLA. I'm the chair of the Populations Subcommittee. I will be the person being a bit of a task master here today. But I want to welcome all of you.
MR. SCANLON: Good morning. I'm Jim Scanlon from HHS's Office of Planning and Evaluation. I'm the executive staff director for the National Committee on Vital and Health Statistics.
DR. QUEEN: I'm Susan Queen from the Health Research and Services Administration, and I'm lead staff to the subcommittee.
DR. LENGERICH: Gene Lengerich from Penn State University, a member of the subcommittee.
MR. HANDLER: I'm Aaron Handler, the chief of the Demographics/Statistics Branch, Indian Health Service, and I'm a staff member the Subcommittee on Populations.
DR. GREEN: I'm Nancy Green. I'm an economist at the National Cancer Institute. I work in the Division of Cancer Control and Population Sciences, the applied research program there. And I'm standing in for Brenda Edwards, who is a member of the subcommittee.
DR. HUERTON-ROBERTS: Suzanne Huerton-Roberts. I'm from the National Cancer Institute. I'm a medical anthropologist. I'm health disparities coordinator for the Behavior Research Program, and a new member of the subcommittee.
DR. TELEFERO(?): Greg Telefero. I'm with the Agency for Healthcare Research and Quality. I'm co-presenting on the MEPS.
MR. MACHLIN: I'm Steve Machlin. I'm also at AHRQ in the Center for Cost and Financing Studies. I'm a statistician, and I'll be giving a presentation about race/ethnicity data in the Medical Expenditure Panel Survey.
DR. LILLIE-BLANTON: I'm Marsha Lillie-Blanton. I'm with the Henry J. Kaiser Family Foundation, and I am here as the user of MEPS.
MS. COLTIN: I'm Kathryn Coltin. I'm from Harvard Pilgrim Health Care in Boston, and I'm a member of the subcommittee.
MR. HITCHCOCK: I'm Dale Hitchcock, assistant secretary for planning and evaluation, and staff to the subcommittee.
DR. NEWACHECK: Paul Newacheck, University of California, member of the committee.
[Additional introductions were made.]
DR. MAYS: Well, welcome. The National Committee on Vital and Health Statistics, particularly the Populations Subcommittee has had a long history of interest in issues of race and ethnicity, and health data and health statistics. We have a variety of information that appears on our Web site that talks a little bit about the background of this committee, and tells of its involvement in these issues.
I have asked Jim Scanlon to give us a bit of an introduction to the background in terms of the activities that have gone on in this area before I begin to talk about the goal of the hearings today.
MR. SCANLON: Thank you, Vickie.
On behalf of HHS, I want to welcome everyone here this morning. I think the National Committee on Vital and Health Statistics, those of you who are familiar with it, is regarded as probably one of our most highly regarded and most productive advisory committees of any HHS, and probably of any in any federal agency altogether.
I don't think I have to convince anyone around the table or in the audience that timely and reliable statistics are an essential element of virtually all of the missions of all of our programs in HHS, as well as those of our partners in health and human services in states and health care providers and non-governmental entities as well in health and human services.
In HHS, we are probably, with a budget of over $400 billion a year, over 300 programs, the largest health insurance plan in the US, the largest funder of biomedical and behavioral research in the US, the nation's prevention agency, and so on. We are probably the fourth biggest government in the world, and if we were listed on the Fortune 500 companies, we would be 1.
So obviously, we play a major role in data and statistics. In fact, it would be hard to know what the impact of our programs was, or even what the compass was in terms of where we are and where we are heading without good and reliable data.
Many of the efforts to reduce and eliminate disparities over the past decade, even beyond that, have had their locus here in HHS. And we have had a number of significant events, and we have been working at improving the data in this area, but I think you will all agree that we still need to go further.
In general, in the area of data and statistics, we have been looking at over the years, how we could improve. And again, we have an impressive array of data systems, both in Health and Human Services. And I think many of them are regarded literally as sort of the best in the US, and probably set the standard for the world.
But I think we really have challenges when it comes to population and subpopulation data. And it is not for lack of thinking about these issues, or even thinking of strategies. It's just a very difficult area, and resources are not always what we need.
Let me tell you a little bit about kind of the accomplishments of the past, and then where we are now, and perhaps where we are heading. Within HHS, we try to have as much as we can, kind of one voice and one framework for speaking on data issues. And we established the HHS Data Council as our internal group to kind of coordinate various activities, to identify our needs in a collective fashion, and to try to address them in a collective fashion across HHS as well.
And under the HHS Data Council we have a very active task force and working group on race/ethnicity data. Olivia Carter-Pokras co-chairs the group. And not so long ago, Olivia, I recall that the work group did a very comprehensive review of virtually all of the recommendations that had been made previously about improving race and ethnicity data. And these probably go back to the eighties and even before.
What were the needs identified? What were the recommendations? And where did we stand on many of the recommendations? That task force report is available on our HHS Data Council Web site.
And then there were a series of recommendations as well, and in general I think there is kind of an overall strategy emerging about how to go about this area, though of course all of this depends on sort of standardized measurement, and high quality measurement to begin with.
In addition, I think HHS probably was the first cabinet department to actually specifically adopt an inclusion policy. About four years ago HHS actually adopted a policy that we would include standard race and ethnicity data in all of our HHS data systems. Now again, you know it's a little easier to do this in data systems that we actually sponsor, as opposed to those in which we collect information from third parties, in which case we are pretty much reliant on what those hospitals and providers and others collect.
We are currently in the process of updating our HHS directory of data systems. And here we have a description of virtually all of our major data systems in HHS, and a fair amount of information about the race and ethnicity detail that is available in those systems. So over the next month we will probably be updating that. That is available on our HHS Data Council Web site as well.
More recently, Congress has directed HHS to support a study at the National Academy of Sciences that will look at the adequacy of race and ethnicity data in our HHS data systems. But an equal focus is to look at the adequacy of race data in basically private sector, and in other public data systems as well.
And the Academy study, I talked to them on Friday, they are in the process of putting the expert panel together. But this will be probably an 18 month, 24 month expert panel review at the National Academy of Sciences to look at these issues. They are to come up with some recommendations. They are also supposed to give kind of a projected cost analysis of what it would take to fill many of the gaps that they would identify.
And then even more recently, just to bring you up-to-date, within HHS we have begun work on kind of a one stop Internet-based gateway to data and statistics. It's in the testing phases within HHS now. But basically, I think you all know again, we have very many data systems. And we have a lot of data on the Internet, as well as publications and so on.
But I guess a sophisticated user has no problem getting access to this data, because you usually know where exactly you are looking. You know the name of the data system. But for most other folks, it is really difficult to navigate. And unless you know exactly what you are looking for, you can become lost in the agencies and the data systems.
So the concept here is a one click off the HHS home page that would bring you to the HHS gateway to data and statistics. And here for the sophisticated users, there are shortcuts to the major data systems. For those that need a little more help, and don't know all the agency names and the acronyms used, there is a nice search engine that we developed, a data finder that would help you. You can put in key words. You can choose predetermined categories and so on. It will take you to where major data sources in that area are.
We are limiting this first wave to federally sponsored or federally supported Web pages in health and human services data. We are including state and other governmental pages as well. And we may expand this as we go on later.
The gateway also will include the shortcut links to the major systems. It will also include links to our data policy Web sites, including those of the committee here, as well as our Data Council, and some other statistical policy documents and activities in the federal government.
And then for those interested in the research side, there are links to the refereed literature that NLM maintains. There is also a link to our research in progress which NIH maintains, virtually all of the grants that are awarded.
So at any rate, there is an attempt to kind of pull all of this together. Again, the data sources and the publications and the Web sites can only be as useful as the original data collection. And again, in the race/ethnicity area, I think obviously we have had for three decades now, the standard race and ethnicity categories. I think our challenge has been not so much in using the categories, as in having sufficient information on some of the categories, and some of the subcategories as well.
So we are all very pleased to have you here this morning to give us some ideas about kind of where we stand now with the OMB categories, and what you might see as ways to proceed. Even though the new OMB revision was issued I guess in late 1997, it did give the agencies until 2003 to actually implement the new standards.
And obviously, implementing a change in a standard like that, that has been in use for such a long time is not an easy thing either. And there are a number of questions relating to comparability with previous data, with time series, and with how you bridge previous information with the new information.
So again, we welcome you, and we look forward to all of your thoughts and recommendations.
DR. MAYS: Great, thanks, Jim.
Olivia has joined us. Will you introduce yourself, just so that are part of our subcommittee.
DR. CARTER-POKRAS: I'm Olivia Carter-Pokras. I'm the director of the Division of Policy and Data at the Office of Minority Health, and co-chair as Jim has very kindly mentioned, of the HHS Data Council's working group on race and ethnic data.
DR. MAYS: Great, thank you.
In terms of the goals today in our hearing, the interest we have is in terms of understanding, particularly through the illustration of data, the contributions and limitations of the various federal data sets in providing data on the health disparities of racial and ethnic groups.
What we have done is we have come up with a set of questions that we have asked those who are kind of the designers of the data sets, as well as some of the users of the data sets to share their insights with us, particularly through the use of data or commentary.
I want to share with you what those questions are, just so that you have some background as to why they are presenting what they are presenting. One of the first questions that we asked is what contributions can be made to advance our knowledge of disparities in health and health care using your survey data?
Particularly, in terms of differences in disease rates among the different racial/ethnic groups, whether or not identification of populations that might receive unequal prevention screening or treatment services is of interest to us.
In addition, we want to know do we need additional variables beyond race and ethnicity to begin to document health and health care disparities among racial and ethnic groups? For example, how important is confounding and interactions among the variables as we try and assess disparities? Have such analyses been conducted in your survey data?
One of the questions that we also introduced was the notion of ethnic identity. How can we best measure ethnic identity? Which is a little different than just talking about race and ethnicity.
Also, we wanted to know whether or not it was feasible to link the various data sets to other contextual data. For example, can these data sets be linked to tell us things such as physician supply, neighborhood characteristics? Things that would help us to really further understand some of the causes and consequences of disparities. And if it has been done in their data set, we have asked them to share that or provide examples if possible.
Also, there is an interest in identifying the cost of health disparities. Can some or all of these costs in some way be documented? For example, if health disparities continue, what is the cost to us, both as a society, as well as in terms of a cost in terms of the budget?
Now the committee may be asking some additional questions, and those might range from for example, asking about whether the survey is translated and conducted in various languages, and if so, what are those languages? If not, do they plan to do this? If the survey, for example, asks about language proficiency? Those are also questions that may come up.
Also, does the survey deal with the issue of under count? Some of the samples are affected by the under count of particular racial and ethnic groups. And how does the particular survey handle that?
Also, whether or not the survey does an oversampling of ethnic and racial groups in its current sampling frame. If so, is it area-based through race questions of density? And also whether or not they can comment on the process that is used, if indeed this occurs, for getting a picture of that emergence as a function of the weighting that would happen if there is some oversampling.
So those are the questions that we have shared beforehand, but as you realize as you look at the agenda, they have a limited amount of time. So each of those questions, and in and of itself -- I want to be fair -- they can each take a day on. So they will be somewhere in the range of those questions, and try and give us some highlights.
The place where we're going to start is actually by talking about the Medical Expenditure Panel Survey. The two individuals who are going to start us off today will Steve Machlin, who is a statistician at the Center for Cost and Financing Studies in the Agency for Healthcare Research and Quality. And also, our survey user this morning, as she identified herself, is Marsha Lillie-Blanton, who is a vice president at Kaiser Family Foundation.
Welcome, and we thank both of you.
[Administrative remarks.]
MR. MACHLIN: I'm Steve Machlin. I'm a senior statistician at the Agency for Healthcare Research and Quality. I'm going to do a brief presentation here on the Medical Expenditure Panel Survey, which is what I work primarily on, and what that particular survey has to offer in terms of race/ethnicity breakdown.
Greg Telesaro(?) here on my left, helped me put together this presentation. And he has actually been immersed in these issues a little longer than I have, so he is here to help out if need be.
So let me just dive right now. I'm going to show you now sort of the outline of my presentation. It's not in direct response to the questions Vickie just posed, except throughout the presentation they should be addressed in some form, all of them are in some level. So I took it on myself to organize the presentation in a way that made sense for me.
I'm just going to start with a brief background to MEPS, followed by a quick overview, which most of you are probably familiar with of race and ethnicity revisions. Then I'm going to move into showing the sample sizes that we have to offer in our survey, current and projected in the future. I'll just give a few simple examples of descriptive statistic type things, just to give you a flavor for the kinds of estimates or variables that come from our survey.
But then I'll also give you a more detailed list of the analytic and contextual variables that are of interest, some of them Vickie mentioned, for analysis in conjunction with race/ethnicity. And finally, I'll summarize my talk.
The MEPS is actually a family of surveys with many components. I'm going to be talking exclusively about the household survey component. The survey sampling frame is actually a subsample of the National Health Interview Survey. So NCHS is a co-sponsor of the survey. And being a subsample of that survey, actually if you go to the effort of linking to the health interview survey, it expands the number of variables you can use in an analysis, but you need to be aware that the health interview survey variables are actually for the prior year than for our sample year for MEPS.
MEPS is an in-person household interview. It's a panel survey. Each panel is followed for two years, five interviews gathers data for two years. And each family is interviewed. There is one respondent for the household who answers for all the members.
There are many levels of data that we put out from our survey. The most basic are person level and family level, but we also have event files and condition files, and various other units of analysis that can be linked.
This is just sort of a pictorial display of the MEPS overlapping panel design, as I call it. The first panel was in 1996. Again, the people in that sample were followed from 1996 through 1997. The second panel began in 1997. Those people are followed through 1998, et cetera, on forward.
As far as a calendar year of data goes, starting in 1997, the overlapping panel design kicked in, which means that in any given calendar year is actually information from two different panels combined. In 1997, it would be the second year of the 1996 panel, combined with the first year of the 1997 panel.
In very broad brush strokes, the household component survey purpose, just again, it's a huge survey. There is lots of information, but in very broad brush strokes, it is used to estimate annual health care use and expenditures. The survey is good for looking at distributional estimates across various dimensions, insurance coverage and employment. We get a lot of information on those characteristics and changes over time, as well as selected quality indicators.
The core interview consists of demographics, various health status measures. I mentioned employment and health insurance. We try to enumerate all health care events of the sampled persons, as well as the charges and payments made for those events.
Now I'm going to shift gears here and just give an overview of the race/ethnicity revisions. As most of you here know about the OMB Directive 15, which requires separating Asian-Pacific Islanders into a minimum of two groups, mainly so that Native Hawaiian and other Pacific Islanders could be separately identified.
Also there is a requirement to allow respondents to identify themselves or the people they are responding for as multiple race categories. We are technically in compliance as of 2001. And we followed NCHS' lead in their revisions to the National Health Interview Survey.
A quick overview, the ethnicity questions, the old question had a show card asking whether any of the groups represent the person's origins. The new question asks the person whether they consider themselves or the person they are responding for as Hispanic or Latino, and where their ancestors came from.
The choices for ethnicity, the old choices are on the left, the new choices are on the right. And you can see that the Mexican American category now has been split into more detail in the new choices.
As far as race goes, the old question had a show card saying which race group best identifies the person. Now they ask the person what race or races do you consider yourself to be?
Here are the old choices, six categories. The new choices are many more. Many of these relate to the specific breaking down in detail of the old Asian or Pacific Islander group.
Okay, shifting here now, I'm going to give you some summary tables of the sample sizes MEPS has to offer for these particular race/ethnicity groups. I just want to mention here that MEPS, from a statistical design perspective, is an oversample of blacks and Hispanics. That's just something that dominates or carries through from the fact that we are a subsample of the health interview survey, which is designed to oversample blacks and Hispanics.
What do we mean by oversampling? I just put a couple of pie charts to illustrate the point, just using 1998 MEPS. You can see that on the left, the 1998 unweighted sample, about 24.3 percent of the sample were Hispanic. But when you weight the survey up to represent the US civilian non-institutionalized population, only 11.7 percent of the US civilian non-institutionalized population is Hispanic.
Similarly, blacks are oversampled. You see the yellow chunks of the pie; 14.9 percent of the sample is black, but only 12.6 percent of the country. So this reflects the fact that Hispanics are oversampled at a higher rate than blacks, but they are disproportionately represented in the sample.
In 1997, it wasn't just an oversampling of black and Hispanics. We had some policy-relevant subgroups that we targeted oversamples for as well. These groups included different stratum such as children with activity limitations, and various other things. One of the categories was low income. So to the extent to which minorities may be more likely to be low income, that also has the potential to increase the sample size for a particular minority group.
In 2002, we are going to have a larger overall sample for the survey of 15,000 households from there forward, and that will increase the sample sizes for all groups. And we have built in, designed an oversample of Asians and low income persons as well.
This table just gives you exact sample sizes available from 1996 to 1999 in MEPS for whites, blacks, and Hispanics. You can see there -- in 1997, I should mention, in MEPS was our term a peak year. That's where you will see the largest sample sizes as far as the late 1990s go. And blacks, you can see there are 3,000-5,000 across the period in the sample; Hispanics, 4,600-7,5000. So there is a fairly large sample size to these groups.
Looking at Hispanics broken down by subgroups, you can see the Mexican American sample is by far the largest subgroup. Puerto Ricans, 500-700; Cubans, just a couple of hundred.
We in MEPS have a general rule of thumb. We don't put out estimates for any one group unless there are at least 100 cases, but that's sort of a bare bones minimum, and that's for a gross national estimate. Clearly, when you want to start doing more detailed analyses and breaking groups down by other characteristics, you can run out of samples fairly quickly when you are just starting with a couple hundred.
To complete the 1996 to 1999 sample sizes you can see we had 500-800 Asians and Pacific Islanders, and 100-200 American Indians.
These are estimated sample sizes for the years 2000-2002. As I mentioned earlier, in 2002, we are going to be increasing the overall sample size to 15,000 households, and also having a targeted oversample of Asians and Pacific Islanders. And that, we are estimating, will result in about 9,000 Hispanics, 5,800 black, non-Hispanics, and about nearly 1,000 Asian-Pacific Islanders are expected to have data for.
Okay, I'm going to shift gears and give just a few very simple, descriptive statistics type things, just as some examples of things that are floating around that break down by race/ethnicity. One of the critical variables we collect is information on insurance coverage. This just shows the percent of the population in the first half of the year 2000, of the non-elderly that were uninsured. Hispanic clearly have the highest rate of uninsuredness, about a third, followed by blacks, 23 percent.
This is similar information, but it adds gender to the picture. You can see that both Hispanic males and females have higher rates of uninsuredness than the other race and ethnic groups. And Hispanic males in particular stand out as the highest rate of uninsurance.
This is a simple slide showing the percent of children that were without a usual source of care in 1996. Hispanics had the highest proportion, 17 percent, followed by blacks, 13 percent, and whites, the rate of not having a usual source of care was about half that for blacks.
Okay, this is very simple, but a slide near and dear to my heart, since I work largely on utilization and expenditure data on MEPS. So this is just a very simple average amount spent for medical care by race/ethnicity in 1996. Again, to really understand these numbers, you need to get behind the scenes in that obviously the amount spent for care is going to be a function of whether somebody is insured, their health status, utilization, all kinds of factors.
But just to give you a flavor for the kinds of variables we have, the average per person spent on whites was $2,200 in 1996. That was the highest, and the lowest group were Asian-Pacific Islanders.
This shows the percent of all medical expenditures by race/ethnicity that were paid out-of-pocket. Out-of-pocket payments are often times used as a measure of burden. Whites and Asians, about 1 in 5 of their payments made for health care were made out-of-pocket, versus only about 10 percent for blacks, and 15 percent for Hispanics.
This slide just shows how this race/ethnicity variation in the distribution of medical expenditures by type of service. If you look at the blue parts of the slide, you can see that about half of medical expenditures for blacks were for inpatient services, versus only about one-quarter for Asian-Pacific Islanders, and in between that for the white and Hispanic groups.
Again, those are just very simple descriptive statistics. MEPS can be used to do much more in-depth type of multi-variate analyses. I just put on this slide three examples of research that colleagues of mine have done that are much more in-depth type things.
The first paper by Wigers(?) and Taylor looks at Hispanic and the different subgroups with respect to insurance coverage. The second paper, Wines(?), Ricaz(?), and Cohen do an analysis where they look at the likelihood of having an ambulatory care visit or usual source of care. And they compare whites to Hispanics, and whites to blacks and find that less than half of the difference between race/ethnicity groups is due to differences in insurance and income.
And finally, Monhide(?) and Visnus(?), their paper looks at trends between 1987, which was our prior survey, the National Medical Expenditure Survey in 1996 in insurance coverage for different race/ethnic groups, and found that Hispanics tended to have different patterns.
Okay, I'm going to mention a few other analytic and contextual variables that might be of interest along the lines of things Vickie mentioned. In 1996-2001, there is information on whether the interview was conducted in English or Spanish. Also, if an access problem was reported by a respondent, they were asked was it due to a language barrier with the provider.
For 2002 and beyond we ask for the language spoken at home, whether the sample person was comfortable conversing in English, whether they are foreign-born, and the amount of time they have been in the US. The other variables that could potentially be of interest for analysis of disparities: income, poverty status, various access to care measures, provider characteristics, and we also have a few selected what we call quality indicators such as the receipt of preventive services, and we ask now about care for selected high priority chronic conditions.
Also, I know there is some interest in geographic type analyses, more detailed than just region. That gets into some confidentiality arena, particularly the smaller the geographic unit you want to link. We have a data center at AHRQ which is set up for people to come work on site if they need to link to confidential data.
And for some years we have had the ability for the user to come and link to secondary databases such as the Area Resource File. Anyone interested in using that capability needs to check our Web site for all the information about how you go about it.
In summary, MEPS is in compliance with the OMB directive as of 2001. We have an oversample of blacks and Hispanics each year, and beginning in 2002, an oversample of Asians as well, and a larger overall sample. The sample sizes we have currently are large enough to make estimates for the major race/ethnic groups as a whole, but some of the smaller groups, if you start breaking them down by a lot of detailed variables, you could run into cell size problems.
And MEPS is a very large, comprehensive survey and provides a rich set of analytic variables that can be used in conjunction with race/ethnicity.
DR. MAYS: Great, thank you. Dr. Blanton.
DR. LILLIE-BLANTON: Good morning. I would like to thank the committee for inviting me to speak, to talk about the value of MEPS in measuring racial and ethnic disparities in health and health care.
A few caveats as I begin. First as you will see, I'll talk largely about health care, which is mostly the work that I have been involved in. But also I am a single user. And while I did try to search and get a better sense of how MEPS is getting used by others, basically, I'm giving you my perspective on its value and its use.
And I think one other caveat is in order, and that is that I also serve as a member of the Advisory Council to the agency. So I cannot be viewed as totally an unbiased user of MEPS, but certainly I can't accept either the credit if it does well, or the blame for the things that it does poorly. So with that, let me proceed.
I think there are a couple of question -- Vickie gave us a whole list, but there are a couple that I wanted to focus on. And I think the big issue is one, whether or not MEPS is useful for informing our understanding of differences in health and health care. But also whether or not MEPS is helpful in shaping policy, provider, or patient interventions to eliminate disparities in needed care.
And I want to say that while I certainly think that that is an important, or those are important questions is to ask of any single data set, it is important to us to put in context that MEPS is a single data source of AHRQ, the Agency for Healthcare Research and Quality, and one of many data sources of the Department of Health and Human Services.
And so when you are asking the question of its usefulness, I think you have to look broader in context, and look at its linkages to other data sources that are produced by the department. In particular, I think its relationship or its link to the National Center for Health Statistics data set.
So the issue in my mind when we talk about its utility for shaping policy is also about its linkage to those other data sources that are within the department. And I mentioned the National Center for Health Statistics, but I also think you have got to look at its linkages with data sources of the Center for Medicare and Medicaid Services, as well as data sources of SAMHSA. So I think this is very important to look broadly, and not just focus on a single data set.
There are at least four indicators that I think are important to answer the questions that I have identified. One is the quality of the data source, and I definitely think that is the focus of this committee, and what I will spend most of my time one. But I think there are several other issues that are important, and one is the research agenda for the analysis of that data.
The other is whether or not you've got researchers with skill and interest in using the data, because I think very much that shapes and influences the agenda and what we can learn from it. And lastly, I think it's important to look at issues of translation and dissemination into which the data source is being used to help improve the quality of care, the extent to which that information is being put in the hands of people who can ultimately improve health care use and health outcomes.
Now I'm going to, as a single user, talk about the data set from my own experiences. One, it's because of what I know, but also I think it is helpful to have some examples. I was involved in analysis of MEPS looking at site of medical care, and whether or not racial and ethnic differences persist. It was a study I was involved in with Rose Martinez and Alina Saldinaca(?).
And there were two questions that we were looking at. One is whether or not those differences persist. But secondly, whether the differences that persist were a result largely of differences in insurance coverage that we see among whites, African Americans, and Latinos.
Now we used the MEPS to look at differences between children, and differences among adults age 18-64. And we focused only on those with a regular site of care. So in other words, we eliminated from our analysis those who did not have a usual source of care.
The outcome measure that we looked at was we defined the regular site of care in the ways available to us using MEPS. And using MEPS, we could look at whether or not someone identified an office-based provider, or whether or not they had a hospital-based provider as their regular site of care.
Now unfortunately, I think that there -- the timeliness. While Steve presented data that is being collected, it shows it as a panel survey. In terms of getting that information out to the field so researchers can use it, you are basically looking at people using the 1996 and the 1997 data. And even 1998 data for policymakers is not as current as for those who really want to know what is happening now in the environment.
Now I know this is a tough one to look at, but I want to use this slide to give you examples of a couple of things that I think are very important about MEPS. One is that the sample sizes are more than adequate to do not only bivariate analyses, but also multi-variate analyses. And when you look at a data source, you want to look at both the independent variables, the broader contextual variables that you talked about. And in this case, this study was largely looking at the role of insurance, and whether or not insurance made a difference in the usual source of care.
But we know that there are a number of other variables that affect where people go for care. And socioeconomic position is another one of those, whether it is defined by income, whether it is defined by family income looking at poverty. With this study, we were able to control the whole constant, both insurance, as well as some of those broader socioeconomic factors that we know very much define where people get their care.
And I think that that's one of the important assets of this data set, that it allows you to look broadly contextually at the factors that might influence both health and health care use.
Now in terms of the outcome, I want to mention one other issue. In this case, we defined usual source of care or the site of care to include the emergency room or the emergency department, which I understand now physicians are very adamant about using that term, as well as not to include the emergency department. Because there has been a lot of discussion about the fact that minority populations are more likely to use an emergency department as their usual source of care.
The numbers are in fact minuscule. The proportion is minuscule, but the fact of the matter is there is a wide perception that emergency room use is very high. And so the thinking is that we needed to define an outcome, first to look at what people perceived to be the problem, which is that minority populations, in this case African Americans and Latinos, are more likely to use emergency rooms; and then to not look at that.
For those of you who are interested in the findings of this study, whether you include emergency room, or don't include emergency room, we found that race, independent of insurance coverage, was still related to the usual sources of care used by whites, African Americans, and Latinos. And that finding held true for children and for adults.
This slide looks at the findings just for children. The journal article, which is actually in a packet that I brought for some of the members of the committee, looked at both children and adults. We found that African Americans and Latinos, regardless of insurance, were more likely to have a hospital-based provider as their site of care in 1996.
Now, I think there are a number of issues raised by that finding. And we talk about whether or not the data are important to help us answer questions. First, I think we've got to look at the incentives and the disincentives for where people get care. And why is it that in this current era we still find minority populations more likely using a hospital-based provider as a usual source of care.
But then I think we have to look at the consequences. Are there consequences for having a hospital-based provider as that usual source of care? And those are issues right now I think still we need to do a lot more research on. By consequences I mean are there differences in the patient-provider relationship?
Are there differences in the continuity of care? Are the differences in patient-provider and continuity, does that have some impact on the content of care? A nd if there are some differences in content of care, it may be that we are seeing some differences in outcomes.
So I think that this study, as a single study and an example of use of MEPS raises a whole set of questions that need to be pursued further. But the important thing about this data set is it allowed us to look in-depth, and answer the question of is race still related to where people get their care, understanding the changes that have occurred in the last 20 years in insurance coverage.
Because we do know in the last 20 years that the proportion of the population with coverage, even if it is public coverage, has increased. So I think that this is an important data set that allows us to look both at race, but at some of the contextual variables, as well as outcome.
Now in closing, I want to go back to the first set of questions that I asked and give you my assessment of the contribution of MEPS to understanding health care disparities. First, in terms of data quality, using what I consider a four point scale from excellent, good, fair, poor, something we do a lot in health research, I would rate MEPS as a good data source.
I think we talked about in terms of race/ethnicity, reporting data, self-reported having fairly good estimates for at least four of the five major population groups, but certainly with some limitations for Asian populations, and serious limitations for American Indian and Alaska Natives. So I still would rate it as good.
I think there is room for improvement on some of the contextual variables, issues of fairness, of trust that we know affect health care. So I do think there is room for improvement.
In terms of analysis and the research agenda that is set, I would move it down a notch to what I say is good to fair. I certainly think that using the analysis that I presented and some other research, that it certainly has the data elements that allows you to ask the questions, and to answer some of the more complex questions that we are faced with now.
And I think importantly, what you want to do is look over time. And other than the Health Interview Survey, and even with the Health Interview Survey you don't have the in-depth questions about health care use or expenditures. So I think that in terms of looking over time, we definitely want to know the progress that we are making. And MEPS allows you to do that with comparisons to previous data sources.
Now one of the reasons I say good to fair is I think that in terms of its linkage to other data sets, and collaboration with the other information sources, there is still a lot of growth and room for improvement. Certainly, I think with the Health Interview Survey, because they are using the sampling frame, it allows you to look more broadly and more in-depth at health.
But I think in terms of looking for the CMS, SAMHSA, some of the other agency data sources, there are linkages in information that could be, and should be learned, and work that could be done that I think we need to move forward on.
The other thing I am very concerned about is that the agency, in using MEPS, is still very much presenting data stratified; not stratified by some of the contextual variables that I think we know very much influence our ability to understand are you really looking at racial differences. So when I look at the data sets that are being produced, or the reports that are being produced, I am still seeing black-white-Latino comparisons.
And we know very much, as I said, that both insurance, both socioeconomic status, very much affect the comparisons. We don't present data that is not age-adjusted. And so for us to continue to compare population groups by race without some measure of socioeconomic position doesn't allow you to really understand whether you are looking at a difference by race, or whether you are looking at a difference as a function of some other measure.
And I was very pleased to see that the 2001 Health US takes the Health Interview Survey data and for the health care variables that are in there, there is at least a control for poverty status for each of the measures of both health and health care in a whole set of tables. And I think that MEPS has the capacity to do that, and I think that is the direction we need to move in.
The comparisons by race by themselves are important. We do need to know overall whether or not there are racial differences. But the question we are posed with as a nation is understanding the why. And it's too easy to look at racial comparisons and say, well, this is all a function of race. And if it's a function of other factors, our data has to help point us in that direction. And I think our agencies that are doing the beginning analysis of that data also have to help point us in those directions.
Finally, there are two other issues that I think are related, and that is the research pool, and the translation and dissemination. I would give MEPS a fair to poor in terms of its cultivating researchers who can analyze the data and answer questions about the extent to which racial-ethnic groups vary in health care use.
I certainly think that we are making progress, but when I looked just at a select group of publications on the data, on the Web site of AHRQ, I found 5 out of 51 publications in a select group that looked at racial and ethnic differences in the data set. And that says to me, and that's a select group, from what little I know, I think that probably represents the universe.
That says that we could do a better job at encouraging researchers, not just minority researchers, but researchers in general to use that data set to answer questions that we need to know more about in terms of health and health care use.
Finally, in terms of translation and dissemination, I think I would have to say that the data itself and the agency is doing a poor job in that arena. I certainly think that the agency understands the importance of disseminating, and is making efforts to disseminate. But the link between what you know and what you disseminate is very real.
It's the information base that we have, understanding how racial and ethnic groups vary is limited in our ability to disseminate, to translate that to improvements in health care will also be limited. So I think there is a lot more work to do.
Now I want to in summary, just say that while I have been to some extent critical, I have to close with saying that I do see 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 are not specific to MEPS, but reflect department wide limitations, and that's important to understand.
It is easy to be critical of a part. But what we have to do is look at the whole, because in many cases that part reflects the whole. And so what I have tried to do is be as specific and as concrete as I can, because I do think that it is the whole that we are talking about today, so we are starting with parts and pieces.
Now one more thing I just want to raise, or at least as I close, there are two documents that I think this committee should have. Probably you already have them. One is a Commonwealth Fund report that looks at HHS data. And that is available through a 1-800 number.
The other is a Medical Care Research and Review Journal supplement that the Kaiser Foundation helped to fund. But within that are two of the article that I mentioned, one by Janic(?), the other by Monhec(?) that uses MEPS, but also uses several other data sources, one the National Survey of American Families. Another is a survey that the Kaiser Foundation actually did.
But as you start to look and answer that second question, are the data sources important to help us inform policy, inform practice, inform provider behavior, I think that both of these data sources are important to help answer those questions. And they are both available through our 1-800 number.
Thank you.
DR. MAYS: Great, thank you.
If you all will join her at the table, what we will do now is take questions. I'll start with the committee, and then open it up to the broader audience. Any questions from the committee? Aaron and then Paul.
MR. HANDLER: I have a question of what I didn't hear, rather than what I did hear. A sore point with me, and it's just my own personal point of view is the multi-race question that was included in the 2000 Census. And the reason why it's a sore point with me is because there were 2.5 million people identified as American Indian only on the Census questionnaire nationally, but 4.1 million people identified one race only, plus one or more other races, with one of the other races being American Indian/Alaska Native.
Now both of the presenters spoke about racial data to varying extents. What do you plan to do with the multi-racial data when more than one race is reported? Again, put those people that reported more than one race in an other category? Are you going to have combinations of categories when you present data for people that reported more than one race? Because you are using the old system, and you are not referring at all to the new system.
DR. LILLIE-BLANTON: Well, I haven't begun to do any detailed analyses using the more than one racial category. But I can tell you we produced a report on urban Indian health where we used the new Census data, and we did present in two separate columns, both the persons who identified themselves by single racial category, and those who identified themselves by a single category or in combination.
I do think it is important to not lose those individuals who identify themselves as of that race/ethnicity in combination. I think that it will complicate our analyses. My thinking when I do proceed to do some analyses is that I would do just what I did in this report.
The interesting thing about this report was that it was not looking at race in relation to another factor. So that's it sort of complicates the analysis, because that means you are actually doing your analysis twice, because you are looking at almost two groups, or you could make a decision.
But for African Americans for example, that group, I think it's very small. It's about 2 million out of some 30 million. I'm not sure I have the numbers right, because I haven't looked at this recently. But when you look at Native Americans for example, you are talking about almost a doubling. So we have got a large group.
So a decision could be made for people doing analysis among African Americans to just do one analysis and include anyone who identified African American as a part of their origin. And I think in this country, given the reality of race, it is likely that most of those people will have experiences of being black or African American.
When you move to other racial/ethnic groups, you might not be able to make the same assumption. Color very much drives people's interactions with systems in our society. And so because of that, people who are multi-racial and black often times have somewhat of a different experience than someone who might be multi-racial, and it's not as easy to tell the color variation.
So I think that these are complicated issues, but I fully support people's right to not be defined by a single racial category if that is their choice. And that is essentially what our current Census allows you to do, to have that choice. It is definitely complicated.
MR. MACHLIN: I only have a little to add to that. I don't really know exactly how like public use files will be structured with respect to the multiple race issue. I mean from an ideal standpoint, any individual researcher would have all the detail, and then they could make their own judgments as to what made the most sense to their analysis, or do it a few different ways, and do sensitivity analysis or whatever.
But we certainly will run up against -- part of the driving decision of how things are put out or published in the agency will be like confidentiality issues. If there is a particular combination of races that is rare, that could potentially lead to the identification of a respondent, that sort of detail would be suppressed in a public use file.
But I don't really know how this will play out in terms of policies for putting out public use files. But again, I guess the more detail you collect, the more options it gives researchers to go this way or that, but there is going to be ultimately some constraints.
MR. HANDLER: One thing that bothers me is before the Census was taken there were activists in different communities, different racial and ethnic communities that told their people even if you can identify with more than one race, don't do it, because the data about you will be lost. It will be in an all other category, and your response won't be used.
Now I don't want those people to be right. People were given the opportunity to report in more than race, and we have to use that data. We can't just put those people that reported more than one race in a separate catch all category and they are lost.
DR. LILLIE-BLANTON: And sometimes you do see that when the Census is reporting their data. They will put the single only, and then at the bottom for the category for multi-racial group, as if that's a group that in itself is meaningful. And in many cases it really a combination of lots of people, and its meaning is uncertain.
DR. MAYS: Paul?
DR. NEWACHECK: Thank you. I want to thank the presenters for terrific presentations first.
When I think about MEPS, I think it has a great deal of value for looking at things like usual source of care, insurance coverage. But the really unique contribution of MEPS is the expenditure data. It's the only place where we can get that in a population-based survey sample.
The problem with the expenditure data is that there is a huge amount of variation from person to person. It's a very skewed distribution. Some people have zero expenditures, some have thousands and thousands of dollars. So one of the problems is when we have an expenditure estimate, it tends to have very high standard errors relative to say a point estimate for usual source of care.
This means when we go down looking at racial and ethnic minority groups, particularly for subpopulations, my area of research is in the children's area, we run into problems very quickly. We get I think reasonable estimates if we look at just a broad racial and ethnic categories of black and white, other, if we are looking at all children.
As soon as we look say for example adolescents, which is an important policy group, it falls apart. We get bizarre estimates that come out of it. And that is even with the oversamples that we have, and that's with two years. Like we are using 1996 and 1997 together.
So my question is, what do we do about that? It seems to me there are a couple of choices, and I was asking if you could comment on this, Steve. One would be to further oversample. We already have large oversamples of Hispanics. You don't have particularly large oversamples of blacks.
But one question would be what would be the cost to the survey -- I don't mean the dollar cost -- but what would you lose in terms of shifting more of the non-Hispanic white population to one of these minority groups? Would you have even larger subpopulations to work with? That would be one option.
The other option would be to perhaps think about different analytic techniques that could be used to deal with the high variance problem. Do we take out outliers from the top for example? Do we use smoothing techniques where we take the logs of all expenditures and not report the actual expenditures?
I would appreciate if you could comment on that, because I think this is a big issue for the user community. I'm sure that my area is children, and I'm seeing three or four analogies of trying to it with MEPS now, and I'm thinking of throwing half of them away, because I can't publish data that show bizarre things. I don't feel comfortable doing that.
And I'm sure it applies to many other fields too. So if you could particularly comment on the analytic strategies, I would appreciate it.
MR. MACHLIN: I could relate to everything you are saying, and I think your points are really very well put. They are reality. It doesn't really answer the question, but I just want to start the response with a slight -- something you said reminded me.
I had meant to mention that now that the years are accumulating on MEPS, that in way to try to improve things for some of the smaller groups is to combine data across panels, and that is certainly something that should be considered. It's not a panacea, but it's something I did want to mention that I had forgotten to.
So you start there, but your point is well taken. Even doing that for some small subgroups, when you do analyses of expenditures, the data are highly skewed. By that we mean that there are a lot of people with no expenditures. There is a certain percentage of the population with no expenditures, and then you have a fair number of people with modest.
And then you have this long tail where you have someone that is very sick and had a lot of hospitalizations, and they had $1 million. And if you do an average with $1 million with a whole bunch of $500, the average can end up being the same thing as the 80th percentile. The average and the median are quite different.
It's the same thing why they show median income, because it looks like people make a lot more than they really do by the really wealthy people pulling the averages up. So you are saying you have these wild fluctuations due to high variance because of that fact.
And so other than yes, trying to up sample size, combining years, using transfer information techniques, logging, all the things you mentioned, I think it is worth more analytic work, and every situation differs. And it's true, in MEPS If you get to a particular group, even a big group, if you have one outlier, it can really throw things around.
So I think the analysts, like you have so eloquently put, need to be aware of that, and then make judgments about what you can and can't say. More oversampling, I'm sure there are costs involved, and trade offs. And boss, Steve Cohen, who a lot of you probably know, would be the better person to sort of field that particular idea.
DR. NEWACHECK: As a suggestion, I think it would be very helpful on this issue. I think if there was an agency publication that dealt with this issue of small subgroup analysis, particularly in the issue of race and ethnicity, and provided a guide to users about different analytic approaches, like transformations, removing outliers and such, and the strengths and weaknesses of each of those techniques. Essentially, some guidance to the user community about how do you deal with these circumstances where you get an unexplainable result?
So there is some consistency across users, and we are not all reporting different things. We are then reporting a median, and somebody else is doing a log transformation. To me, that's not very useful, because then it means our studies are really not very comparable. And it seems like there could be ways in which the agency could help to create a more uniform framework and a way of thinking about how to deal with this problem.
MR. MACHLIN: I'll certainly raise this with Steve, we will talk about it. But you would be interested in something in particular relating to expenditures.
DR. NEWACHECK: Well, I think that's where the problem is largely. My use is things like usual source of care, health insurance, it's not a problem.
MR. MACHLIN: Right.
DR. MAYS: Okay, I'm going to have to bring this to a close, so that we can have a bit of a break, and you two might ask your questions on the side, so that we can get to the next presentations.
I want to thank both of you for starting us off in what is definitely the right direction this morning. I appreciate the time that you have taken, and that you have really have shed light on your survey in terms of the committee's point of view in terms of what the issues are that we should consider.
I think to thank Greg also for his participation with Steve. Thank you both very much.
We're going to take a short break, and then we will be back. So we will be gone about five minutes, and then we will be back.
[Brief recess.]
DR. MAYS: Okay, we're going to move on to our next survey, which is the Consumer Assessment of Health Plans. Our presenters have assembled themselves at the front. The presenter for the survey is Judy Sangl, who is with the Agency for Health Care Research Quality, who is a health scientist. And also we will have Jim Moser. Jim Moser is manager at the Barens Group. He is with KPMG Consulting, and will be representing the users this morning.
So thank you both.
DR. SANGL: Good morning. I just want to also point out that with me is my colleague Chuck Darby. He is really considered one of the founding fathers of CAHPS. So if you really have any tough questions, you can ask him.
The CAHPS, I don't know if everybody is familiar with the acronym, is the Consumer Assessment of Health Plans Survey. And it relies on the importance of the consumer perspective. One of the goals of CAHPS was to develop a set of surveys to measure consumers' reports and ratings of their care. And this is across all systems of care, fee-for-service, managed care. And it was to develop reports to consumers on the results of those surveys, as well as evaluate the process and outcome of application of the report. And to also make these products available to the purchasers, plans, and providers.
The CAHPS survey is really a consortium. The products are developed through collaboration of AHRQ and sometimes with CMS. Harvard, RAND, and RTI are the three grantees for the first set of CAHPS. And technical assistance is provided to the users through the Survey Users Network, which we also call SUN, to AHRQ and Westat.
The first thing to know about the survey is it is perhaps set up very differently than a lot of other surveys. There is an adult core, which is 46 items, and also a child core. There are supplemental items that have been developed and tested. For example, there are language and interperter access, items that would be most relevant this committee meeting say. And also a chronic conditions supplemental item set that is most commonly used in the Medicare population.
I think in the set there is a matrix of items, and you can go across populations, seeing which ones are the mandatory sets, which ones are recommended for other items, say the communications core for Medicaid managed care.
And of note is that the CAHPS survey has been translated into Spanish, Russian, Vietnamese, Manadrin, Korean, and Cambodian.
In terms of CAHPS implementation, this is not like many of the other surveys you will be hearing about. AHRQ itself does not conduct the survey. Sponsors conduct the CAHPS survey. And some of the large sponsors include the Medicare program, and it conducts three surveys, one for the managed care population, one for fee-for-service, and one for disenrollment.
Medicaid state programs and SCHIP programs are conducting Medicaid surveys. Depending on the year, there are 35 states that survey the Medicaid populations, and I think maybe five to seven states that are doing it for the State Children's Health Insurance Program.
There are also commercial sponsors. The Office of Personnel Management requires all the plans that contract with the Federal Employees Health Insurance Program to do the survey annually. NCQA now requires the CAHPS survey if plans wish to be accredited. DOD is also another major sponsor. And now approximately 90 million Americans are covered under the plans under which we collect the CAHPS data.
I guess another quick point I wanted to add was that the sponsors can independently choose -- they add the core items, but then the sponsors are the ones themselves that can add the supplemental items. And in some cases they add their own items. I'll come back to that point later.
On the CAHPS survey the race/ethnicity is self-reported, and there are two items. There is a race item, and there are five categories. Unfortunately, the Asian category is missing here. And the second question is on Hispanic or Latino origin or descent. And respondents can choose more than one category.
There are what we call composites and ratings. The set of 5 composites has 17 items. The getting needed care has four items. The getting care quickly has four items. Communication has four items. Helpfulness and courtesy of office staff has two items. Customer service has three. Then there are four overall ratings, one for personal doctor, one for the specialist seen most often, health care in general, and health plan.
Now the supplemental items, for example, there are items on the availability of the interpreter, which obviously can affect a person's assessment of care. There are other items about language spoken at home. If you are able to get access to the interpreter, things like that.
In addition, there are other measures one can construct from the other items in the core set such as percent having a personal doctor, percent seeing a specialist who felt they needed one, percent having an office visit in the last year.
And in terms of trying to think about the committee question of social cost, one can look at differences in gaining access to care, and difference in experience. For example, one can look at if racial groups have differences in getting needed care. If they have more difficulty in getting referrals to specialists. Do they have more or less trouble communicating with their provider?
We know for example, that a lot of times Asians rate their provider communication lower. But if you correct for language spoken at home, most of the time that difference is removed, so that Asian English speakers rate equivalent to whites, but the Asian non-English speakers do not.
And so some of the things to consider when you are looking at the CAHPS data, and some of the data limitations of CAHPS, there are several screens. For example, most of the time the respondent must have had at least one visit to provide some of the ratings. So you don't rate a provider if you have not had an annual office visit. But that independently can be the other measure, that you haven't had an office visit.
Because a lot of times it's the health plan as a unit, there may be a small number of minorities in the plan. And it might be possible to oversample, but very often you do not know the racial/ethnic composition of the plan, so it would be difficult to do.
We know that education and health status tend to be related to assessment of care. Those with lower education tend to rate the plan higher. Those in better health status tend to rate the plans higher. Those are often used as case mix adjustors in CAHPS data.
We know there may be potential response tendencies related to racial/ethnic group membership. For example, there seems to be some maybe anecdotal evidence that Asians don't tend to give the highest ratings on the 0-10 scale when you do the overall ratings, and there has been some exploration of that.
We know that income also affects assessments. New Jersey has done a CAHPS survey of their commercial population and included an income item. Now income is not a core item of CAHPS, but they added it as a supplemental item to their own survey, and found that those with lower income tended to rate higher than those with higher incomes. So trying to disentangle that as another would be useful. And in the next phase of CAHPS, AHRQ will be studying cultural comparability of items.
The one nice thing about CAHPS data is that it is possible to link plan characteristics with the CAHPS data, so that you can perhaps try to get a feel for what is going on now. Now, it may not yet be at the level of detail that one would like to understand what may be best practices where there is less disparities in racial groups.
I and another researcher, Tren Luzan(?) and Nicole Laurie(?) have done some research on both the commercial data and Medicare data, and found huge plan-to-plan variations say in whites versus blacks and white versus hispanics, sometime a 20 point difference in those people having an annual office visit, or those having a personal doctor. And there have not been disparities.
So it would be interesting to go back and find out what are those plans doing well in terms of minimizing disparities, and also find out what is going on in the plans that have huge disparities. I haven't done that yet.
Also, on plans that we have put to NCQA, which I think are about 350 plans, one could get aggregate clinical quality data, the HEDIS data, and compare that with the consumer assessments of plans available. So there are some interesting contextual data that can be added. A lot of times the plan market area may be too aggregate to add other contextual variables like the areas resource files, but then as I said, you have the plus side of the HEDIS or plan characteristics.
Now as I said, the CAHPS is sort of like a network of sponsors. And we have created what is called the National CAHPS Benchmarking Database, or NCBD. And this is a database that is available to researchers. Right now I think in the current database there are about 792 plans that have adult survey data. There are 148 plans that have child survey data. That's a total of 57 sponsors, so you have roughly 350,000 adult responses; 44,000 children responses. And in this case, let me just clarify, you have an adult respondent for the child.
We didn't put on the Web site, but if people are interested in finding out about this database, it would be www.cahp-sun.org. And SUN is the Survey Users Network.
Since you just heard about the MEPS data, it will be interesting to know that one could do other types of analyses, because selected CAHPS items will now be included in the MEPS survey. So one can contrast the usual use expenditure measures with the consumer assessment measures, and that would be linked with the national representative sample.
And just a further point, in the next phase of CAHPS we are going to be going to below the plan level. So for example, there was work on group practice CAHPS, and also individual provider level CAHPS, and also different populations or special populations. There will be the children with special health care needs. The ECHO is the Experience of Care and Health Outcomes, which is really targeted towards behavioral health populations. And we are also in a very early design phase of a CAHPS for nursing home residents.
So Chuck, I don't know if you want to add anything?
DR. MAYS: Great, thank you.
DR. MOSER: Good morning, I'm Jim Moser from the Barens Group of KPMG Consulting, representing a user of CAHPS data. I'm pleased to be here to share some of the results of the studies that I have done.
Judy has already mentioned some of the background of CAHPS. Where I come into this is the Medicare managed care version of CAHPS or MMC for short. And the particular items that I have been looking at most closely are not really the health and doctor ratings, but rather self-reported health status, health conditions, and use of the health care system of these enrollees.
These surveys have been conducted annually since 1997. And I think one of the unique opportunities afforded by these data is that they allow one to identify certain racial/ethnic groups that generally aren't picked up in some of the previous studies such as American Indians, Alaska Natives, Native Hawaiians, and other Pacific Islanders. And so some of the results I'm going to be presenting today look at these six different racial/ethnic groups.
As Judy mentioned, there are a couple of questions on the surveys that ask first about race. And there are five racial categories. Respondents are allowed to check one or more race. And we have found that about 97 or 98 percent of individuals choose just one of those.
There is also a question about Hispanic or Latino origin or descent. We found about 92 percent of respondents give us a usable response to that question. And it is the responses to those two questions that allow us to classify six different racial/ethnic groups. And here you see how we define them in this study.
So I will remind you that we are looking at people who indicate that they are of just one race, so we are not looking here at these multiracial individuals. And the first five lines there are all defined in our study as non-Hispanic Latinos. So it's white, non-Hispanic Latino, black or African American, non-Hispanic Latino and so forth.
And then you get down to the last line, that is where we put all Hispanic Latinos, who technically could be of any race, but we are just restricting those in our study to those who have indicated they are just of one race. And I think more than 90 percent of Hispanic Latinos do choose just one race, so we are not losing very many by doing that. So that's the background of how we are defining these different groups.
If you just look at any one year, and for example, I have some results here for the year 1999, how many individuals responded in each of these racial/ethnic groups, you only have 200 or 300 in some of the smaller racial/ethnic groups such as the Native Hawaiians and the American Indians.
But what we did was put three survey years together, 1997, 1998, and 1999. And this chart just simply shows there are anywhere from 75,000 to up close to 150,000 total respondents. So by lumping all of these surveys together, it allows us to build up the sample size, particularly for some of the smaller groups, so that we can do some good cross-group comparisons.
Here we show the number of individuals who are actually in our analysis sample. So we have 865 American Indians/Alaska Natives by putting three survey years together. And that is a lot better than just looking at one year alone.
I should say that there were some minor differences, if any differences at all, in the questions on the three surveys. And we made a judgment as to whether they were or were not identical in terms of whether we included those questions in our analysis data set.
Just a little bit about how individuals where chosen for CAHPS. These enrollees had to be in a plan for at least 12 months. Generally, a flat number of 600 enrollees in each plan were selected for being surveyed, except if there weren't 600 in the plan, in which case all individuals were chosen.
And as I said, we put three years of data together. In a few instances it just so happened that the same person got interviewed in more than one year. And what we did in that case was just keep their responses for the latest year, so we weren't doubling up any people.
The remainder of my presentation here is going to focus on four different ways that we looked at the data, some demographic characteristics, self-reported health status and health conditions, as well as health care utilization. And then there are some questions on there about smoking, past, current, former smoking. Whether individuals have been advised by doctors to quit smoking. And then I'll show you some statistics on how successful they were in quitting in some cross-group comparisons.
Well, with respect to demographic characteristics, you can see the vast majority are white, and it roughly mirrors what the overall population is. All of these non-white groups account for about 14 percent of Medicaid managed care enrollees, so that's kind of what we are dealing with here.
This shows the gender composition by racial/ethnic group. Just a few comments. Generally speaking, female enrollees outnumber males for every group except American Indians and Alaska Natives. There, male enrollees outnumber females. The greatest disparity is for Native Hawaiians and Pacific Islanders, where almost two-thirds are female.
This shows the age distribution for each group, and it is kind of a busy chart, but I'll just say for any group, if you just focus on the first and the last bar, it shows the percentage that are in the youngest age group, and the oldest age group. And looking at it that way, the so-called youngest age group are American Indians/Alaska Natives, because 15 percent of enrollees are in the youngest group, and only 12 percent in the oldest. On the other hand, the Native Hawaiians and Pacific Islanders are the oldest group.
A little bit about educational attainment. Whites have the highest high school graduation rates, but Asians have the highest college graduation rates. For African Americans, American Indians, and Hispanic Latinos, roughly about half complete high school.
Here are a few tables about -- and these are self-reported health status, as well as health conditions. Again, a kind of a busy chart, but a couple of overall conclusions here. African Americans, Hispanic Latinos, and American Indians/Alaska Natives generally report worse health currently compared with other groups.
Native Hawiians and Pacific Islanders report the greatest improvement in health compared with the previous 12 months at the time the survey was conducted. And American Indians/Alaska Natives report the worst change in health.
There are questions on the survey about five particular serious conditions phrased in the form of has your doctor ever told you that you had. And those are heart disease, cancer, stroke, COPD, and diabetes. Generally speaking, heart disease is a fairly serious condition for all of these groups.
Diabetes is a particular concern to African Americans, American Indians and Alaska Natives, and Hispanic Latinos. And it is also the most prevalent condition for Asians. COPD and stroke are generally less serious compared to some of these other serious conditions. But there are clearly some racial/ethnic differences here.
Now I'm going to talk a little bit about the utilitization of the health care system. The first four lines of this table talk about the distribution of visits to a doctor. And I think a way to interpret it as those who make the fewest visits to doctors' offices tend to be Asians and Native Hawiians/Pacific Islanders, whereas those who tend to make the most frequent doctor visits are African Americans and American Indians/Alaska Natives.
White enrollees are the most likely to visit a doctor or a specialist, as well as to use prescription medicine.
American Indians/Alaska Natives have high rates of use of hospitals and emergency rooms. At the other end of the spectrum there, Asians tended to be low hospital and ER users.
Overall, Asians' use of medical services is generally lowest among all of these groups. American Indians/Alaska Natives have the highest rates of use of such medical facilities and services as in the last three lines there, special medical equipment, special therapy use, and home health care use.
I'm kind of running through these quickly in the interest of time, but clearly, one could peruse these tables at greater length.
This last section, as I said before, is about smoking behavior and advice to quit smoking. This table shows the distribution across three different smoking behaviors as we define it. Those are an individual is either a current smoker, or a former smoker, or has never smoked. And so for any given racial/ethnic/gender group here the percentages should sum up to 100 percent, or at least within rounding.
Just some overall observations, current smoking is most prevalent among American Indians and Alaska Natives, and least prevalent among Asians. Asians and Hispanic Latinos are most likely never to have smoked. And just looking at some gender comparisons, females are less likely than males to be current or former smokers. Again, this is the Medicare managed care population. And most of the differences I mentioned here are statistically significant.
This table looks a lot like the previous one, but the interpretation is a little bit different. The figures here represent percentages of individuals who have heart disease. Again, just a couple of overall observations. Former smokers are more likely to have heart disease than nonsmokers and current smokers.
I thought current smokers would probably be most likely to have heart disease, but if there is any interpretation there, maybe it's some who get heart disease and smoke are sort of frightened into quitting and becoming former smokers; at least, that's one conclusion.
Whites of all smoking behaviors tend to be more likely to have heard disease than compared with other non-white and Hispanic Latino groups. And I should mention that we have constructed tables like this for not just heart disease, but some of the other serious conditions, cancer and diabetes and what not. And again, just in the interest of time, I didn't put them in here today.
This deals with advice by a doctor to quit smoking, smoking cessation counseling. A couple of observations. The differences here between racial/ethnic groups, and even between males and females generally aren't very large, and to the extent that they are measured statistically, they tend not to be very statistically significant.
I think what is probably the most remarkable thing is just the overall level of some of these rates in that anywhere from one-third to one-half of smokers were not advised by a doctor to quit in the last six months. And I think from a public policy perspective, that is kind of a distressingly large figure.
And furthermore, only about half of them received cessation advice, or I should say they only received cessation advice on about half of their doctor visits. But again, not terribly large differences across groups.
This chart shows -- and I'm almost ready to wrap up here -- success in quitting by racial ethnic group, as well as gender. Among those who have ever smoked, Asians have the highest quit rates. The taller the bar there, the greater success in quitting. So Asians, Native Hawaiians, Pacific Islanders tend to have the most success in quitting. On the other hand, African Americans and American Indians/Alaska Natives tend to be least successful in quitting.
Males are generally more successful than females, but those differences tend to be small or not very significant. Overall, it looks like around 75 percent of smokers are successful in quitting.
Just to sum a little bit here, and give you some overall conclusions, Asians tend to have the best reported overall health, and the lowest health care utilization. Whites generally report good health, but nevertheless, are still above average utilizers of health care.
The group with the poorest health tends to be American Indians/Alaska Natives, followed by blacks or African Americans. Hispanic Latinos tend to be average in terms of self-reported health status, as well as utilization of health care.
Females generally report worse health than males of the same group, but females also report lower levels of the five serious health conditions that I identified earlier, compared with males.
Smokers are generally in worse health than nonsmokers. Up to one-half of smokers do not receive any advice to quit smoking from a doctor, and the differences across groups and genders are small. And finally, about three-fourths of persons who have ever smoked eventually do quit, again, with some differences across groups.
That concludes my remarks, and I appreciate the opportunity to share my research with you.
DR. MAYS: Thank you.
Okay, let's open it up to questions. Eugene?
DR. LENGERRICH: Thank you for the presentation. I guess I had a question about how the survey is set up and drawn, probably Dr. Sangl. From looking at these maps, it looks like the states are sometimes doing different things. Is that correct? That they are implementing different portions of the survey, or to different populations?
DR. SANGL: Well, I'm not sure. It may be slightly confusing, but go ahead.
DR. MOSER: I think you can go back to something that Judy pointed out that is fundamentally different about the CAHPS survey than the other surveys that you heard about. That is, CAHPS was put in the public domain and made available to any sponsor who wanted to use it. So CMS is using it. They administer the managed care version, the fee-for-service version. Some state Medicaid programs will so, and so forth.
So that in each state, this simply represents users of some strip in each of those states. There are two states that have not used it in any form.
DR. LENGERRICH: So then when we put them together to do analyses like this, we may not be getting a nationally presentative sample?
DR. MOSER: Exactly, right.
PARTICIPANT: Now the MEPS data collection on selected CAHPS items will be a nationally representative sample, but others may not be.
DR. SANGL: Just one quick point. I think probably the way to think about CAHPS is the populations. There is the Medicare population. There is the Medicaid population, which the states can choose to administer, and then there is the commercial population.
And the one thing I didn't point out, Medicare has a similar benefit level. So you could combine across Medicare without a problem. You might not want to do that for Medicaid, because some of the states add optional benefits. And then the commerical could be all over the map. So you have to think about that if you want to pool.
DR. MAYS: Dr. Coleman-Miller?
DR. COLEMAN-MILLER: I have a few questions for Dr. Moser, and it's only because I have heard so many of the kinds of surveys that you just did, and I guess I just need to try to understand, as you did your surveillance, what cultural dimensions you put into it.
For instance, how did you word the question that only put one race? I didn't hear that. You showed the form, but it was impossible for me to read it from there. The first question.
DR. MOSER: Right. Well, let me see if I can pull it back up here. I think what I showed was pretty much verbatim as how it appears on the survey.
DR. COLEMAN-MILLER: I just couldn't read it.
DR. MOSER: Here is a verbatim reproduction of the question. What is your race? Please mark one or more.
DR. COLEMAN-MILLER: Is everyone clear that everyone who did this could read, and didn't have any language issues? I'm only asking that, because I only see it in one language, and I noticed that you have Hispanic and Asian populations. I was wondering whether there was a language --
DR. MOSER: I believe, Judy, you said that the survey was presented in several different languages.
MR. DARBY: The Medicare managed care, and other CAHPS surveys are in English and Spanish. So both of those versions are offered.
DR. SANGL: This is translated into Spanish, Russian, Vietnamese, Mandarin, Korean, and Cambodian.
DR. COLEMAN-MILLER: That was the CAHPS study, I think.
MR. DARBY: The California Health Care Foundation sponsored the translation of the CAHPS core into those other languages beyond Spanish. English and Spanish has been used with the Medicare population that Jim was reporting on.
DR. MAYS: Just for clarity, isn't this an administered interview, and not --
MR. DARBY: It is a mail survey, but also with telephone follow-up.
DR. MOSER: And proxy respondents are allowed if in fact there is a language issue, just simply to be able to translate the items.
DR. COLEMAN-MILLER: I guess when I hear about whether they smoked or not, we have so many patients who say, no, I don't smoke, and yet they chew so much and have the same exact issues. And I was wondering if that was included?
And just a number of things. They said that their doctor had not advised them not to smoke, and the question is have they been to the doctor? And have they told the doctor that they did smoke? The cultural dimension here is always to try to answer exactly what you want to hear. And so I'm always listening for the part that says I know you are going to say what I want to hear, but I have to ask you questions that cover all of it, so I know I'm getting an accurate 54 percent don't.
DR. MOSER: Well, first of all to your first question, I think we only ask about cigarette smoking, and not other forms of tobacco use on here. But also, I think we are just taking the responses as they have been given to us. So I think you are right, these are responses by individual enrollees. How accurate the information is, what the context is, whether their doctor would have responded the same way about if they had been advised or not is anybody's guess. So I don't know how to really say much more than that.
DR. COLEMAN-MILLER: When I see information that goes out and is published, people accept that without that cavaet that you are talking about. And I suppose that's why all of us are sitting here, because so much goes out without those caveats that we need.
I also noticed that earlier today when they came in, you were talking about Hispanics, whites, non-Hispanic whites, and Hispanic blacks, non-Hispanic black. And that the population base you used. And I'm kind of curious about the different wordings here. Non-Hispanic white, all those are new words for me. I'm new to this.
So I'm wondering what drove that, and how you did not use that, and how others do you use that. I wonder what role the Census had in that, and what role OMB had in that? I'm just curious about all the different terms I hear. It's like a spaghetti group of terms that some use and some don't, and what we do about that.
DR. MOSER: I would comment that one could come up with the categories that you mentioned by looking at the two questions on the questionnaire. We simply adopted the OMB regulations for this item, since it was going to be used and sponsored by a government agency. We used the wording as OMB has prescribed.
DR. MAYS: Kathy Coltin.
MS. COLTIN: Yes, I was curious as the potential, particularly in the Medicare managed care CAHPS of linking the survey data with other CMS data to look at things like for instance whether there are differences in response rates by different racial and ethnic groups. We know that the overall response rate is quite high for the Medicare population, but I don't know how it varies, and how if it does vary, that might actually affect the results, particularly in certain health plans, where were there to be a more representative response from their membership, would their results be different?
DR. MOSER: I don't know what the response rates are. I wish I did. I haven't looked at that. We can link these data to other data sets, and I think that was what you started to say. In fact, in other work that is not reported here, since we know the health plan that these individuals belonged to, we are able to link it up with data on health plans. For example, what their benefit coverages are, what their co-payments, co-insurance, other kind of health plan features are. But that's in research that I am not reporting here today.
MS. COLTIN: Are you able to link to any data sets that have race/ethnicity data in them?
DR. MOSER: Potentially. Again, we haven't tried to do that. This is relatively early in our investigation of these data. There are other data that I would like to see, for example, on socioeconomic variables that aren't in here, that we would like to try to link to, and there is a potential for doing that in the future.
MS. COLTIN: As Dr. Sangl pointed out, if haven't had a visit, you don't respond to most of the items. So it would be interesting to look at that impact too if you could link to the Medicare use file.
DR. MOSER: Actually, some of the data that I talked about with respect to smoking cessation advice, I didn't show all the data that we had in the table that I presented to you, but we did make some adjustments. For example, if an individual didn't see the doctor, or let's say if they did see a doctor, and we knew that, what did the statistics look like? And the percentages were a little different across all the groups, but the across-group comparisons, the conclusions were quite similar to what I presented here.
MS. COLTIN: And, Dr. Sangl, in the commercial population, it seems like there may be an opportunity actually to think about geocoding the data, and seeing whether there are differences in response rates. In the commercial population, the response rates are a lot lower, and that might be a way to look at the impact of race/ethnicity on the representativeness of the samples.
DR. SANGL: Right. I know in the draft Medicaid there was a proposal to actually the states provide the race/ethnicity of the enrollees, so one could work backwards and actually know the composition of the plans. But I'm not sure what the status of that is.
Just a quick question, and we can follow-up later. I think Allen Saslawski(?) at Harvard had done some preliminary look at differential response rates by comparing the enrollment profile that had at least administrative data on race. I'm not sure on the ethnicity. And he did find a differential response.
MS. HELLER: I was just going to comment, and help answer some of the questions that were just posed.
DR. MAYS: Identify yourself.
MS. HELLER: My name is Amy Heller, and I'm the project officer at CMS from the Managed Care Task Survey. One of the things that we are very proud of is our response rate. For the past three years, we have had better than an 80 percent response rate in the managed care suvery. And that does vary slightly by race and ethnicity. Some plans, like with the Chinese Health Plan, CMS doesn't offer the plan in Chinese, the survey. We have a no response.
But also from the other surveys, the fee-for-service survey for example, has a 70 percent rate, but varies slightly by service delivery system. But we do get a very good response rate.
And some of the other questions that were posed were do we look at non-response analysis to see if it varies by race and ethnicity? Yes, that's something we are currently looking at. And also, do we look at things like regional variation? Yes, Harvard is doing a study looking at regional differences, because it is a case mix adjustor for us. And we are also moving towards small area estimation versus regional estimation.
There is a lot of work that is currently ongoing that is supplementing all the work that Jim and some others are doing. And I would be happy to share that with you later.
DR. COLEMAN-MILLER: I have a question as to whether there are any items or you have any data on whether the respondents are US or foreign born? And if they are foreign born, how long they have been in the US? The reason that I'm asking is the categories that were used up here are the not the same categories that are used in many other countries, especially in Latin America.
-- a mail survey, and not a lot of guidance given. I am concerned that when somebody responds to these categories, there is not a lot of guidance given as to how they should identify themselves, or what these categories mean. Because the meanings for a foreign born person would mean something very different than what would be meant in the context here.
DR. MOSER: Unfortunately, there is not a question about whether they are foreign born.
DR. NEWACHECK: I have a question for Mr. Darby and Dr. Sangl. There has been a lot of discussion around the table about response rates and potential for bias. There is also the issue of participation, that is, which plans choose to participate in the national data pool, at least in the commercial population, at least as I understand it.
Could you give us your expert opinion about given the three CAHPS survey samples, the Medicare sample, the Medicaid sample, and the commercial sample, which of those would be amendable or useful to do racial and ethnical analyses, and which of those you probably would not want to do, because of the potential biases, or the likelihood of maybe developing erroneous conclusions because of those biases?
MR. DARBY: I may let the Medicare folks address that. I was thinking that sample could be. I think it really depends state-by-state with Medicaid. Response rates are probably the lowest in Medicaid, but there are some notable exceptions to that.
And again, I think you would have to look at each commercial sponsor to determine. I think we would find there would probably be a number of samples that you could not do the analyses if you would want to, on racial and ethnical subgroups.
And the MEPS survey is including about 15 CAHPS items, and that again, I think would be safe to do the analyses that you would want.
DR. SANGL: Just a quick comment, because we did two parallel analyses, one for the commercial and one for Medicare, and Medicare had the 85 plus percent response rate, and then NVCD had maybe a 45-50 percent response rate on average. But the results were very similar. So I would feel more confident in looking at both, but if you just looked at the commercial alone, and in particular, Medicaid, you may need to be more cautious.
MS. ARIAS: Hi, my name is Elizabeth Arias. I'm at the Mortality Statistics Branch at the NCHS. I have a dual question related to your question on Hispanic origin. Are individuals asked to state which country of origin for instance, either Mexican American, Puerto Rican, or Cuban American? Or are they all grouped together as Hispanic or Latino?
And then if they are, a related question, when they receive this questionnaire, do they get instructions as to what it means to be Hispanic or Latino, such as what you got with the Census form, where Hispanic means that you were born in Cuba, Mexico, Puerto Rico, or South America, or the Spanish speaking Carribean.
MR. DARBY: They are not asked about their country of origin. And there are no other special instructions that are given with the mail survey. Now MEPS, it was reported on earlier, the categories that are used in MEPS. And that of course would apply to the CAHPS items that are included in MEPS. I don't know if Steve or Greg, if they are still here, if they want to add to that, but for MEPS, there is information.
MS. ARIAS: I would just like to make a small point related to my questions. One is that for some of the conditions that you noted, such as diabetes, there is a lot of difference between the Hispanic subgroups. So when you say diabetes is very high among Hispanics and Latinos. It's not the case for all Hispanic subgroups. Similarly, with smoking prevalence, it is very different between Mexican Americans and say Cubans. So that was basically why I asked those questions.
I guess another question would be do you think you might consider changing this question format for future surveys?
MR. DARBY: I would say that we would consider that. One of the key things that we have done again, going back to the fact that this is given to sponsors to use, we have tried to keep the core as short as possible. It is 46 items currently, and everyone wants to add items, rather than take them out as you know, if you have done surveys.
So I would say that it's unlikely that we add that to the core, but it is something that is being considered. We would probably develop it as a supplemental item, and recommend that people include it.
MR. HANDLER: I have a question. Has there ever been a question on the survey asking people who have quite smoking, why they quit smoking? Was it to feel better? Was it because they had a relative that became ill or died because of smoking? Did it become too expensive? You mentioned that health professionals are not encouraging people not to smoke. But the people who did quit, could there be a question why did they quit?
DR. MOSER: All I can say is there are not those why questions on the survey currently. There are really only about four with respect to cigarette use, and they are more or less as have you ever smoked, things of that nature, but not getting into reasons why they may smoke or reasons why they quit.
MR. HANDLER: If we became aware of why people do quit, then we could use those same techniques to get other people to quit.
MR. DARBY: If I could make a point of clarification also. The smoking items are not part of the CAHPS core. They are added to the Medicare managed care version. They are also added to the version that NCQA uses with the HEDIS supplement. I would also like to point out that the data that Jim provided does not cover some of the other items that are in CAHPS.
And again, the core idea is 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 and so forth, those are items that are in the core CAHPS, but we haven't really talked about those.
DR. MAYS: I want to ask two questions. The first is what would it take to actually get the expanded race and ethnicity questions on the core? Because I think part of the issue is one of who has the knowledge base and the expertise to do this. And I would think that the designers of the survey would, that in terms of having that as a part of their core, as opposed to some groups then would have the information and others not would make your survey more usable.
So what would it take to be convincing enough that these should be a part of the core? And then I'll ask my second question, because it's to Dr. Moser.
MR. DARBY: I would think more evidence, like the woman pointed out, that there are differences when you look at country of origin, for example. If there are really differences in the way people report their care, then I think it would be strong evidence that maybe we should add that particular item.
DR. MAYS: Yes, I think that there probably is evidence from different places for those. So that's good to hear, that it's just somebody needs to write a letter and send it to you all with published evidence.
My second question really has to do with as you went through your various tables, there are some instances in which we know, and this is almost getting to Dr. Coleman-Miller's earlier comment that we know for example, there are racial/ethnic differences in the question itself. So for example, just the one on rating one's health will vary by ethnic group.
And the question becomes whether or not there is any kind of data analytic strategy that says in addition to looking at things by race and ethnicity, we are now going to put into the mix, the contextual variables, or what we know about cultural differences on perceptions on the questions. Is there any methodological work going on to help highlight what those differences might be?
DR. MOSER: Not that I'm aware of; certainly not by me. Possibly others are getting into that, but not to my knowledge.
DR. MAYS: Dr. Coleman-Miller?
DR. COLEMAN-MILLER: To follow through on that thought, my question was going to be almost tangential to that. And that is to say that before this survey goes out, does anybody look at it who would culturally say you can't ask someone if they stopped smoking or if they quit smoking, because the answer is going to be yes.
And some people, if you say well, how long has it been since you smoked, it could be five minutes. I just decided to stop smoking a couple of minutes ago. And I have had patients tell me that, no, I don't smoke. I put down no, and then my next question has to be, how long have you not smoked? And when it is minutes, it may seem like an outrageous answer for you, but for them it's I'm now in front of a doctor, and I think I'm going to stop smoking.
So if is some filter that this goes through so that people can say, this isn't the way to ask that question, because you are not going to get the right answer. It's sort of when AIDS was an issue, and you asked young black men, are you gay? It didn't get the right answer, because it wasn't asked correctly.
So I guess this is the same question that Vickie is asking, but do you have a filter that this goes through? Or do you ask these questions based on your cultural background?
DR. MOSER: Well, let me just briefly respond. Probably, I should say that the question about do you currently smoke, or have you quit smoking, they are not asked in exactly that way. I phrased them that way just to make it easier to present the results.
But the questions themselves are more specific about have you smoked in the last 100 days? Or have you ever smoked 100 cigarettes in your life? So there is still some room for whether the answers are truthful or how they are interpreted, but at least they are worded a little more specifically than how I presented them here.
MS. COLTIN: I can also say that as Mr. Darby pointed out, these items are part of the NCQA HEDIS supplement, not part of the core text. And they were actually adopted from the Behaviorial Risk Factor Surveillance Survey, so they are actually worded exactly as they are in that survey. And there has been considerable research done on the performance of those items in that survey.
MR. DARBY: I would also just amplify on that. I think it's an excellent point about designing items for different cultures. I find that survey researchers are probably the worst people to guess what respondents might be thinking when the questionnaire is administered to them.
CAHPS conducted some 450 cognitive interviews on this questionnaire, most of them in English, some in Spanish. We will continue to do more of that. Also, we are about to fund for five more years, a set of grantees to continue the CAHPS development work. As part of that, we are going to be doing a lot of testing of cultural comparability of the items.
You can imagine when you start translating into these other languages, one, the equivalence of the language is one issue, and we do have a standard protocol for translation, by the way. But in addition, the cultural differences in how people view this, Judy pointed out about the response tendencies. The literature is actually all over the place about people's tendencies in responding.
It does appear that possible Asians may be harsher raters, but everything else is pretty mixed about that. But is is an area that we are going to study pretty carefully, because it is very important to know what the respondent is thinking when you ask them the question, or try to know that.
DR. COLEMAN-MILLER: Let me just ask one more question. In terms of the Asian subpopulation, when they receive it, it's either in English or Spanish, right? Is there any plans for translation for any of the Asian subpopulations?
MR. DARBY: Well, there are other translations, Mandarin, Vietnamese, Russian, Cambodian. We also in the next round of CAHPS will be translating it into other languages also, as the demand requires it. We also are looking at translating into more than one version of Spanish. We have been told ultimately that the version we have is too formal, and then by other people that it's too informal. So it's obviously that there is one than one Spanish, and that we need to address that.
DR. MAYS: We are going to break for lunch and return back at 12:45 p.m.
(Whereupon, the meeting was adjourned for lunch at 11:50 a.m.)
DR. MAYS: Welcome to the afternoon session. What I thought we would do is take a couple of minutes and I wanted to check in with the committee on any questions or anything that they wanted to raise before we actually begin the next presentation.
Olivia, there were a couple of things we should really share those because I think those are things for our presenters as they come up because it is coming up as an issue.
DR. POKRAS: Based on the presentations to date and what we expect to hear later this afternoon and tomorrow, one of the questions we had is: What are your particularly survey agencies, what are you doing to provide guidance to those who use your data analysts to use the multiracial data, what kind of guidance you're providing to them? Second, what kind of training are you providing for minority researchers so you can access and use your data?
DR. MAYS: Any other questions or comments? Any from anyone else? As we begin to move into the afternoon, we're actually going to do a variety of things. In addition to hearing from individuals about the surveys, the other thing we're going to do is have a presentation on social economic status.
As you know, one of our questions really is what should we be looking at beyond race/ethnicity. SES has emerged and there is a considerable body of research that suggests that SES is an important variable in terms of looking at health disparities in race/ethnic groups.
We will also have Carolyn Clancy who is going to give us an overview in the police perspective. When I talk about policy, I'm not actually talking about what the rules and regulations are, but it's probably better to think about direction, guidance, and perspectives. She will be with us also this afternoon.
We might as well start with the Medicare Current Beneficiary Survey. Dan Waldo is joining us as the survey person. He is with the Center for Medicare and Medicaid Services. Our user will join us at the table when she arrives.
MR. WALDO: Madame Chairman, members of the committee. Thank you very much. I'm pleased to be able to be here today and tell you a little bit about the Medicare Current Beneficiary Survey, the ways we capture race/ethnicity from respondents to the survey and just a brief overview of the things that you can do with the MCBS involving race/ethnicity and highlight some of the things you can do, but you probably shouldn't.
Let's look at the survey. I'm sure that you're familiar with the MCBS so I'm going to go through the overview fairly quickly to set the stage so that you can understand how we do what it is that we do. MCBS is a survey of nationally-representative sample of people who are enrolled in Medicare.
The universe is the population over the age of 65, and the population for the disabled by social security rules. The survey is conducted by Westat, Inc. for the Centers for Medicare and Medicaid Services, the agency formerly known as HCFA and still frequently known as HCFA in Baltimore.
The principle goal of this survey is to help policy analysts understand how Medicare is serving its target population. Field work for the survey began in the Fall 1991. I'm astonished that we've been in the field as long as we have, but we're still going strong.
The MCBS is a list sample. The sample persons are drawn from a master enrollment list for Medicare that's current as of January of the year. By April, we've selected the panel that's going to be interviewed based on January eligibility. We go into the field to interview those people beginning September of the year and move from there.
This means that we never really capture people who age into Medicare during the year. The youngest of the elderly population that we will be interviewing will be age 66. The sample is drawn to overrepresent enrollees under age 65. The population who are eligible on the basis of disability and to oversample the population aged 80 and older whom we believe to be frailest and most at-risk for high healthcare expenses. In order to minimize ill case, we use a cluster example design.
The respondents are clustered in roughly 100 geographical sampling units or PSUs to minimize filed costs. The PSUs are drawn to produce nationally representative results. We're not in all states. We're not in urban and rural settings in each of the states where we are, but our thought is that we do a pretty decent job of nationally representing the population, something that is matched up by the fact that you weigh up respondent's Medicare costs, we come really, really close to what the actual numbers logged in are.
We think that the difference is because the actual numbers are wrong. They don't agree with us of course. The survey process is fairly routine for this kind of survey. Sample persons remain in the survey for three years worth of work plus an introductory interview and a close-out interview. So we're actually in their house for four calendar years, but we get data from them for three calendar years.
They are interviewed three times a year using computer assister personal interview and we collect information on a wide variety of things. We collect information on demographics, social/economic characteristics, health status and functioning, use of health services, financing of health services which is the riginal genesis of the survey, and interactions with the Medicare program, things like how well the Medicare and You program is delivering information and what kind of information beneficiaries are interested in receiving.
In terms of capturing race/ethnicity in the survey, this is done basically on self-identity. We interview the respondents in their first interview, the intake interview. We ask them to self-identify race/ethnicity.
We do have administrative records that are provided to us by the Social Security Administration on race/ethnicity. We do not rely on that information. Until very recently, there were large gaps in that information. It's based on race/ethnicity identified at the time that the individual applied for social security which means that some of the categories themselves are internally inconsistent.
There is no particular correlation between someone who identified themselves as white in 1998 and someone who identified themselves as white in 1938. There is no attempt to over-represent racial and ethnic groups of enrollees in the MCBS, although we do have one of our primary sampling units in Puerto Rico.
The decision was made early on that we would over-represent on the basis of age rather than race/ethnicity in order to capture populations that were in need of particular types of services. Prior to 1998, we followed the stand OMB protocols for racial identification. As a way of preparation, we had two questions in the MCBS, one of this identifies race and the second identifies Hispanic origin.
In terms of race prior to 1998, in the interview the beneficiary or the respondent was shown a card and asked to identify him/herself with one of the categories: American Indian, Asian or Pacific Islander, Black or African-American, White, and another race. They were asked to specify what that race was.
Including people for whom we simply don't have data or who refused or who didn't know, we come up with about 13,000 enrollees in 1991. By 1997, we were up to about 18,000 sample respondents principally because we kept adding supplements to investigate particular areas such as HMO enrollment and so on.
Subsequent to 1997 or beginning with the 1998 survey in the Fall, we used a different tack. The question was worded the same as before, but we offered new choices to the sample respondents in the show card that allowed us to comply with OMB Directive 15. The interviewers were directed to code all that apply and to prompt respondents for all that apply.
New choices were offered to all sample persons in 1998, not just those in their first interview. In that sense, it was a break from previous practice where we only asked people once. Because we had a new show card, we re-asked everybody. The new show card, again, the question said: Looking at this card, what is your race?
In the case of a proxy respondent, it asked what is the sample person's race, code all that apply. We have the categories that see: American Indian or Alaskan Native, Asian, Black or African-American, Native Hawaiian or other Pacific Islander, White, or another race, and they were asked to specify.
About 241 respondents in 1998 reported identification with more than one race, 866 were missing, refused or didn't know. Almost all of those were missing, and it was basically a quirk of the survey in 1998. As you know, the MCBS contains in its sample both people who are community-based and those who are in institutions.
We do not interview institutionalized sample persons directly, but rather go to a nursing home administrator or someone familiar with the care who is associated with the institution. In 1998, a field decision was made that we would not ask the new questions of people who had already been in the survey and who were institutionalized.
There were 850 of those folks for whom we did not ask the new question in 1998. By 1999, we had asked the new question of everybody. I'm showing 1997 and 1998 data simply because there is a crosswalk that we can look at and we can look at that next.
There is going to be a brief pause here because about half of you have a package that has some tables in it that are labeled Table 1, Table 2, and Table 3. The other half of you do not have those tables. They're tacked in out the back and in an effort to make sure that everybody is engaged and not falling asleep after lunch, I'm going to ask you to identify whether you have a package that starts with Table 1, Table 2, and Table 3 or whether you have to tear out the tables that are there and look at the back.
Raise your hand when you're ready to go. What we have done in these tables is an attempt to crosswalk responses of those people who were in the survey in both 1997 and 1998. Because approximately a third of the panel gets rolled out every year, there are a substantial number of people who stay in.
In fact, there were 10,600 and some respondents who were both in 1997 and 1998 surveys and who received a community interview in both of those years. We have created a table that shows how those people sort themselves across the years. What we found was there was a substantial amount of stability in racial identification between 1997 and 1998 except for those who had identified themselves as American Indians.
There were a total of 113 people who identified themselves as being of more than one race in 1998 who identified themselves with a single race category in 1997. The American Indican respondents tended to sort themselves out much differently and were much more likely to identify with being of more than one race than were those in the other racial categories.
The tables that you have show both the frequencies and row percentages and column percentages that you can look at the data if you want to think about where people with particular racial identifications in 1997 sorted them out in 1998 or where a particular race identification in 1998 came from in 1997.
On the screen here, I'm showing the crosswalk from 1997-1998. Otherwise, the numbers would be way too little. The second table that you have is an effort to show how people who had replied in both 1997 and 1998 associated with the individual race categories that they were offered on the show card.
You can see that there were a substantial number of people who identified themselves on 1997 as American Indian who changed their designations or associated themselves with more than simply the American Indican race in 1998.
Unfortunately, we dot not have a control so that we can tell you what proportion of these people changed as a result of the new race show card and how many of them changed just because respondents forget from year-to-year. I don't think that we will ever have such a control simply because we will not be asking this question every year, but only once on each intake.
The third table that is inserted in here shows those people in 1998, all of 1998, who are community-based who identified with more than one racial category. The attempt here to show the patterns of people who identified themselves with American Indian in that they were much more likely than any other identification to pick another racial identity as well.
The preponderance of them identified themselves as both American Indian and White or American Indian and Black or African-American. Regarding ethnicity in the MCBS from 1991-1997, a single question was asked: Are you or is the sample person of Hispanic origin? Yes or no.
In 1998, we changed the question to read: Are you or is the sample person of Hispanic or Latino origin? Yes or no. The results of this change were negligible. Nobody in 1997 who identified themselves of Hispanic origin changed their mind in 1998. Nobody who identified themselves as not of Hispanic origin changed their mind in 1998 as a result of the words "or Latino."
If we look at the combination of race/ethnicity in the MCBS, you can see that the population shuts themselves out into the bins that you would expect to see. When using the MCBS to study race/ethnicity disparities, you can do a lot of things. Some things, you can do with some confidence. Other things, you should be are u. The things you do with confidence looking at the MCBS is run analyses that involve a single dimension.
Or you can look at common events among the target population. There are other things that you can do, but probably should not do and that is look at rare events or try to break the population into too many cells. Here are some examples of things that you can do with a fair amount of confidence.
Looking at the number of male beneficiaries who had blood test or digital exam for prostate cancer during the previous year. From 1999, we can identify the proportions
of the population, White Non-Hispanic, Black Non-Hispanic and Hispanic population and is not sufficient in size to look at the other categories.
Similarly, you can look for women at the proportion of beneficiaries with a Pap smear or mammogram during the previous year. Again, White Non-Hispanic, Black Non-Hispanic, Hispanic populations provide a robust enough sample size to be able to look at differences.
For the other racial identifications, there is insufficient sample size. Looking at common conditions, hypertension, diabetes, and osteoporosis, it's possible to look at racial and ethnic differences in the prevalence of those conditions.
We can also look at things that are more of interest to our program administrators, position user rates, for example or per capita Medicare expenditures by residents and by race, community-based expenditures and facility care for three of the largest racial groups.
We can look at things like the characteristics of the population who are duly eligible for Medicare and Medicaid and observe as we might have expected that racial minorities and ethnic minorities make up a disproportionate share of people who are eligible for both Medicare and Medicaid.
Now, the sad part is when we start to get into some of the more interesting things looking at race/ethnic different with other categories attached. If we tried to look, for example, at the number of sample persons who had ever been told by a doctor that they had hypertension and we're drawing this now from the 2000 Access to Care file, if we tried to look at age and racial identify of some of the cells and I grayed out in your handout and on the screen, those cells where you have fewer than 50 respondents which is a rule of thumb for squirelly results.
If you look at the mutually exclusive groupings, White Non-Hispanic and Black Non-Hispanic, for most age groups, we have a robust enough sample size that you can look at differences. For the Hispanic population, it is just about as robust with the exception for the population under the age of 44.
When we look at the individual races, individual race identities, only the Black African American and White populations are sufficient in sample size to allow an age break in a discussion of hypertension. When we get to rare events, things get way sparse in a hurry.
If we look at the number of beneficiaries who had ever been told by their physician that they had lung cancer, there are no race/ethnic categories with enough sample size to be able to do a break by age. If you wanted to do this kind of analysis, you would have to collapse the age categories or collapse the race category in order to be able to get up to a decent sample size.
If we look at functional limitations, again, trying to look at that by age and race, you can see that with the exception of the 65-74 year old population, there aren't very many cells that are heavily populated enough to allow you to do robust kinds of calculation and that for categories other than White, Black, or Hispanic, the other racial identifications, there is almost no cell where you can get enough population to do this kind of analysis.
That, basically, is in a nutshell, an overview of the MCBS and the race/ethnicity data that we maintain with it. I'd be happy to answer any questions.
DR. MAYS: Actually, we take questions after both of the presentations. We will invite Joan to join us. Dr. Davanzo is a vice president with Lewin and Associates, the Lewin Group.
DR. DAVANZO: Coming in the middle of Dan's talk, if that was all I heard about the MCBS, I might have been very depressed as a researcher. I want to show you a study that we did where we were able to see a couple of different dimensions, we were able to do multivaried analysis, and we had some pretty interesting results.
The purpose is to provide an example of a study in which the MCBS is used to identify differences. There was a supplement in 1997 on information needs and we looked at the combination of information needs and preventive health behaviors on beneficiaries.
Our research questions wanted to know do beneficiaries have an understanding of the Medicare program, what do that know? This was in 1997 before the National Medicare Education Campaign, and it was right as HCFA was just going into the educate beneficiaries initiative.
This study provided baseline for HCFA in terms of what people knew. Do different subgroups vary in their understanding? We wanted to look at the understanding in relation to the subgroups in relation to preventive health behaviors. The one we chose was flu shot because that was the most prevalent in the data.
We looked at a variety of subgroups. HCFA was very interested in the rural and in low-literacy folks because a lot of the HCFA materials were written at reading levels that were significantly higher than beneficiaries could read. We wanted to look at a variety of different groups, including vision-impaired, hearing-impaired.
It was 18. That sample size is the group that answered all the questions. For the multivariant, we needed to have a full end, and when we went up on a population basis, we were looking at 29 million of the older. We didn't use people in institutions.
What we found out, no surprise, many beneficiaries lack an understanding of Medicare and the biggest knowledge gap was related to managed care. People really didn't understand that. Knowledge levels varied by topic. There were certain topics that people did feel that they had a good understanding of and certain topics they said, no, they had very little understanding.
The receive of the flu shot varied by subgroup. There is a parallel track here. How did we measure knowledge? There were six situations we had self-reported which the question said how much do you know about such and such a thing.
The answers were I know a lot about it, I know a little about it, I know almost nothing about it. That was the person's assessment of what they knew. And then we had measured knowledge which we had some quiz questions and they were true or false. Does Medicare pay for a flu shot?
That was one of the main questions we were interested in. We called that focused knowledge. Can you appeal decisions? Here is the self-reported. The Medicare is general feature. Paying for service. How much do you think you know about services covered by Medicare, Supplemental Insurance?
How much do you know about finding a doctor and how much do you feel you know about staying healthy? There were really different kinds of topics, different levels of complexity, different levels of interest by people. As you can see, people that felt they knew most of what they needed to know, they knew about staying healthy. That's a fairly generic thing.
They didn't know about Medicare to most. The folks that knew just a little of what they wanted to know, we collapsed the categories and HMOs were the ones that knew just a little. What we decided to do was to look at the folks that HCFA was going to have to educate and this was the group that knew little or almost nothing about the topic. That was the group we collapsed.
We broke it down by subgroup. You can see that males and females, there's the dimension and then also within races, there's the dimension of knowledge. But as Dan said, we can only look at those three groups. We can only look at White non-Hispanic, Black non-Hispanic, and Hispanic because everything else is just to small.
But for looking at something as broad as that and with only two dimensions, the dataset is extremely useful. Now, we're back to the flu shot story and as lots of folks have found, African-Americans in 1997 were not receiving flu shots for a variety of reasons.
This is odds ratios. There was a logistic progression of what were the determinants of receiving a flu shot. One equals yes and zero equals no. Controlling for all that stuff. If you look towards the bottom, African-American beneficiaries were 21 percent less likely to receive a flu shot than White non-Hispanic which is the control.
DR. POKRAS: These don't look like odds ratios. They're percentages.
DR. DAVANZO: The way they were presented was along a little grid and the grid didn't copy out. So it was a more likely, less likely. If you had the focused knowledge being the flu shot, you knew that Medicare covered the flu shot, you were 32 percent more likely to receive the flu shot.
What we're saying is knowledge does matter, but it's not the only thing that matters. There's the ethnicity issue here and that was the part that was interesting to HCFA. The application of the knowledge was once we were able to tease out all the different factors, and HCFA actually ended up doing this, but some materials were created specifically for African-Americans.
The survey went on to say there were different questions in later versions that said if you didn't get a flu shot, why not? The various choices were things like my doctor didn't tell me I should, I might get sick, I might get side effects. I didn't think I needed it.
There are a variety of reasons that people didn't receive the shot and they weren't all tied to knowledge. Going into this whole thing, there was the expectation that if you really educate beneficiaries, the behavior would follow and there wouldn't be these little intervening things like race/ethnicity.
The result, then, was to take the created specific materials for African-American beneficiaries that explained the flu shot and explained specifically that it didn't make you sick. That was the thrust of the materials. They wouldn't have known that had people not identified the combined action of the ethnicity and the behavior and the knowledge.
We found out knowledge matters, but it's not everything. If you're going to educate beneficiaries, you have to find out more about them in order to be successful. In terms of using the MCBS measure disparities on the broad level, there is detailed information on service utilization and pharmaceutical which is extremely useful, but if you're interested in the three big groups, it's interesting because it's only the three groups.
The depth of the information is tremendous. There are various supplements about HMO membership and insurance status and functionality. So, all of those kinds of issue, if you're really interested in the three broad groups, you can do a lot of analysis.
It's not appropriate for studying specific subgroups of making precise population estimates. I want to show you something that was actually done that was not just the prevalent.
DR. MAYS: Great. Thank you. We'll ask our two presenters to stay up at the front of the table and what we'll do is start with questions from the committee and then we're going to open it up to the audience.
DR. BREEN: Thank you. I'm a big supporter of Medicare, of course, and the MCBS. My questions are related to, why don't you make it even bigger and better? Since this is a sample for about 30-35 million Americans and it's one of the biggest expenditures that the federal government makes.
In that context, it strikes me that maybe the sample size should be a lisle bigger than 18.000. I'm sure you have financial constraints, but I'm pointing out the obvious that when we're spending so much money and we want to go a good job in delivering the care, it's really a shame not to have more information on population that less likely to be using this set of services maximally.
With that approach, I wondered, is the easiest way to do it to pool several years of data to get more information on the smaller populations? I'm wondering if that's possible. Secondly, if there are any plans to expand the sample or modify the sample to oversample some of the groups?
People who have Medicare are also people who have paid into social security, people who have been working or have been dependents of people who have been employed. They're not a complete cross-section of the US. They're actually doing a little better. The worst-off population is not covered by Medicare. I was just concerned about those things and wondered if you would comment on them.
MR. WALDO: There are a couple of things going on. First, as you pointed out, since the Medicare population is essentially those who are eligible for social security, over time, we're going to expect the proportion of the sample that is race/ethnic minorities will expand, those that occur naturally in our population.
We don't have plans to oversample for race/ethnic groups. That is purely a budget issue for us. The survey has been run on a pretty large shoestring, but a shoestring nonetheless. It's very difficult to get more money. The agency has a total research budget in FY02 of $55 million of which $12 million is used to run the MCBS.
In FY03, the proposal is for a research budget of $24 million of which $12 million is used to run the MCBS. It's hard to squeeze more water out of that stone. In terms of pooling across years, that's quite possible to do and it's facilitated by the structure of our files.
You have to be a little wary of doing so because since beneficiaries stay in the survey for three years, if you were to combine all the observations for 1997 and for 1998, you wouldn't come up with 40,000 observations, you'd come up with something a lot smaller because a number of the respondents would be the same in 1997 and 1998. If you're willing to take a longer view and assume that there hasn't been a whole lot of temporal shift, you could go back to 1991 and pick up there.
DR. NEWACHECK: I was surprised too, to see that you didn't oversample for race/ethnic minority groups in the survey. Most of our large health surveys do do that, the National Health Interview Survey, the Medical Expenditure Panels. It doesn't necessarily have to be particularly more expensive.
You don't have to necessarily just add additional minority populations. You can take your Whites and substitute more minority groups. That's what MEPS does, for example. You have the data because you mention you have the administrative files where people indicate their race and that. That's a good starting point for oversample.
To me, it seems like it would be relatively modest incremental cost to do the kind of oversample we're talking about at least by race. It would be a little more tricky to do Hispanics because you probably don't have that in your master files.
This is a big public expenditure. Race/ethnic minority issues are high on our agenda. Disparities are high on our agenda. As you pointed out when you showed your slides, for some very important although not common health indicators, we just don't have the data to look at it. It seems to me that it would be worthwhile to invest in an oversampling strategy that I don't think would be particularly expensive for the agency.
MR. WALDO: We have the capability to do that. It's not an absolutely fool-proof capability to do that. For example, in the calendar year 2000 access file, just during lunch I ran it. We have in that file 1,100 respondents who identified themselves as being of Hispanic origin. They self-identified as being of Hispanic origin.
The administrative files showed of that same group, 400 people were identified as Hispanic. We can oversample. In this case, in the Hispanic population you can oversample the people who appear on the administrative records as Hispanic, but it's not clear to me whether you would get an oversampel that represented the population you identified as Hispanic or Latino, and how you would make a correction for that later on.
The other caveat in terms of oversampling is that each time you create an oversample population, you diminish the power of the survey. One of the goals of the MCBS was to be able to look at rare events in the population or medical events in the population. We're already limited to that extent.
Oversampling additional populations without increasing the overall sample size is going to diminish our ability to detect those rare events even more. I'm not offering that an argument why we shouldn't do it. I'm just offering it as a caveat to say that it's not necessarily a cost-free.
DR. MAYS: Let me go to Dr. Hertin-Roberts, Ms. Coltin, Olivia, and then Dr. Coleman-Miller.
DR. HERTIN-ROBERTS: I have a specific question of Dr. Davanzo. From the last line you showed us, you showed us percentage differences of knowledge as it differed by race/ethnicity. Did you have a large enough sample, did you do any test of statistical significance on these to see whether these differences were?
DR. DAVANZO: They were.
DR. HERTIN-ROBERTS: You didn't put that up there. So I didn't know.
DR. DAVANZO: There were almost 800 Hispanics in the sample. We weeded down to 10,000. We stayed high on Hispanic and African-American relative to the weed out. They were at .05.
MS. COLTIN: Have you compared the self-reported race/ethnicity data from the MCBS sample with the data on your master enrollment list and if so, what kinds of patterns of agreement or disagreement have you observed? It seems that would be useful for those who might use those data to analyze MedPar data.
MR. WALDO: I mentioned just a moment ago that there seems to be a wide difference between the administrative reported Hispanic population and the self-reported simply because the way the data are reported out by the Social Security Administration is being Hispanic is mutually exclusive of being White or African-American or Black or other races.
There are a substantial number of people in the Medicare population who apparently sorted themselves first as White and then second as Hispanic. That didn't get capture in the administrative files. Other than that, I have the tabulation right here. Unfortunately, I was scribbling it off the screen at lunchtime. The populations sort themselves into broad groups. The biggest difference is in the Hispanic or Latino origin.
DR. POKRAS: Several thoughts came to mind during your presentation. One is I understand about the wish that this was larger, but this is when you compare with NCHS surveys, this is a rather expensive survey. There may be some things that we can learn from the presentations tomorrow that would be helpful in talking about how we can oversample with the existing budget.
I certainly agree with your point on that, $12 million is a substantial amount. This predecessor for this subcommittee was very involved in trying to improve race/ethnic data at Medicare, including bringing social security in, trying to talk about what could be done.
I know HCFA at one point in time actually paid for a questionnaire to send out to folks so that they could re-self-identify and improve the data on there. I don't know how much your survey folks were talking to the folks at SSA to continue to improve the data.
That seems to be a real gap between 1,100 respondents that you say are of Hispanic origin and the administrative file only showing 400. The third thing that came to mind is wondering if you have succeeded in translating your questionnaires. I didn't hear any mention of that.
MR. WALDO: We have the MCBS questionnaire and it's conducted in English and Spanish. The interviewers are both English and Spanish speaking. We identify at the first interview language preference of the respondent and work with that.
Unfortunately, those are the only two languages in which the instrument is written. We have to rely on intermediation and the services of the interviewer for other racial groups. That can be a big problem. We have a huge problem this Fall trying to administer and oversample for the population that were duly eligible in Massachusetts because we couldn't find enough interviewers who spoke Russian and Portuguese. We have not tried to translate the instrument into Chinese. Partly, it's the expense of doing so compared with the number of naturally occurring respondents in the survey.
DR. COLEMAN-MILLER: I actually have four points and I will go through them quickly and would love for you to address each one of them individually. The first one is that you talked about ethnicity, behavior, and knowledge and that you were expanding that and that you went into why they didn't do the flu shots.
But you gave them choices. If you give them four choices and their choice isn't on this, then you allow to expand it further so that someone who had the cultural competency to be able to ask the question for the real reason the person didn't get the flu shot, was that somehow expanded or did you just accept your second set choices as the reasons they didn't get the flu shot?
Another one is that you mentioned here about staying healthy, they knew about staying healthy. When I look at these numbers, I see that most did know about staying healthy and that when we look at the next slide which is Black, I don't have my glasses on real well, but I think that looks like 14 percent.
I'm wondering whether even in all of these we can define what staying healthy looks like. Sometimes it is not the behaviors that we would choose to say staying healthy. When you use a term like this, what staying health really means. The third is that I see this oversampling used all morning and all afternoon.
When I see it being used with the minority community, the oversampling seems to create special problems. The disparity is so significant, I would like those special problems to be disregarded to give us something in the ballpark.
Oversampling seems to be a way to go look at this. Fourth, I'm wondering about the Lewin Group. I don't know anything about them, but I do want to just ask you very specifically, how many minorities are hired full-time there to help you with your decisions?
How many of the mixed population that you're dealing with with this disparity and does the government require that you hire them or do you hire them by choice? I'm really interested. You can make that question number one. That's just a curiosity question.
DR. DAVANZO: Lewin Group has been around for about 30 years. We're a health policy/healthcare management/consulting firm in Falls Church. There are probably 100-120 professional staff. I would say 10 percent are African-American. Maybe 10 percent are Latino. We have a lady from India, Pakistan, Philippines in a practice of 15.
DR. COLEMAN-MILLER: I want to stop you for just one second and say you're 10 percent that you mention, I need to know whether they are the people who go out and do the actual? Are they in an administrative position that says we're going to help you?
DR. COLEMAN-MILLER: Some of them are PhDs, some of master's level. They conduct focus groups and collect primary data for us. They do everything. When we're trying to develop a protocol to go out into field for interviews or something, we always try to have a mixture of folks on the team so that we're asking things in ways that people understand and we're not asking about things that don't matter to folks in the qualitative data collection. That's the example I can tell you.
DR. COLEMAN-MILLER: I want to add one more reason why you might not get the flu shot. Minorities are so able to answer that and I didn't hear it as one of your choices. That's just probably an example of many. I'm just trying to figure this whole data collection thing out and one of the things I'm trying to figure out is how you come up with questions to give you data. That's why I'm pressing this. There is no other reason.
DR. DAVANZO: I think there is a difference between qualitative and quantitative data. We didn't formulate the MCBS. It was formulated a long time ago. Among the choices, there are some open ends that ultimately became choices. My doctor didn't speak my language and there are a lot of choices that evolved in the later iterations of the survey that came out of open ends in the interviews.
It's hard because it's a trade-off between the detail and the realness of qualitative, but within quantitative, it's closed yes, no. It's harder to delineate those fine shades. When you balance the two, you can begin to understand the stuff underneath the question.
MR. WALDO: May I add also that very often the categories of responses that the beneficiary or the respondent is able to choose from is greater than what actually gets tabbed up finally.
Very often analysts will roll responses to categories together. In many of the MCBS questions, there is an other category where the interview will transcribe the response and put it in. Our questions that exist in the MCBS come from two different genre.
One is stuff that we got off the shelf from somebody else. We did this in an effort to make the MCBS responses as comparable as possible with other surveys that had been done. Those questions that are related to the Medicare program specifically are formulated by staff in the agency in consultation.
They use focus groups, we do cognitive testing of the instrument so that by the time the questions get out, they've actually gone through a set of beneficiaries, typical respondents.
DR. COLEMAN-MILLER: Thank you. I just encourage you not to use the old questions. This disparity is so huge now that the old questions might not be the right questions. Somewhere we're not getting the answers that we need. When we get statistics that are up above 200 percent difference, I think that's when we need to not use the old questions, but go to ones that are much more relevant to our communities today. I encourage the Lewin Group to consider that.
DR. WALL: I'm Frank Wall from George Washington University School of Public Health. I want to first express some of my own personal observations and then I will pose my questions. Since the beginning of this morning, I have been listening to the presentations and it's very interesting.
I have to say that I'm a little frustrated. I've been in the field for a long time. Asian-American/Pacific Islander and Native American and the two groups I work with including immigrant refugees. When I hear statistics like this, we are not anywhere to be found.
Two groups are being wiped out. I am a researcher by training. There have to be ways to answer those questions. Early on, one of the speakers said something about Massachusetts. I live in Boston for five years. I work in API clinic. Even from 1980-1990, the census stats tells us that Boston have 5.2 percent API.
You can imagine the rate from 1990 to 2000. Where are we? This is a policy question, but it also has implications for those of us that are in research. We need to really critically think how to answer those questions. Again, I suggest that all of us in here whether it's a federally funded agency, this is a very expensive survey. As an advocate and researcher, I cannot use it.
DR. WARREN: Reuben Warren from the Agency for Toxic Substance and Disease Registry. I ask the question about your Medicaid vs Medicare eligibility. Could you separate the two?
MR. WALDO: Can I separate the two?
DR. WARREN: Did you? You have a Medicare population and a certain percentage of those are Medicaid eligible, do you separate the two?
MR. WALDO: We can support the two. It is not nearly as straightforward as one might hope because there are a number of different classes of Medicaid eligibility among Medicare population. The various laws starting with the balanced, the BBA in 1997 and the BBRA in 1998 created groups of beneficiaries.
Some Medicare beneficiaries are eligible to have all of their co-payments and their premiums paid for by the Medicaid program. Others are eligible to have their premiums only paid by the Medicaid program. Others are eligible to have some portion of home health co-payment aid and the states have the option of extending full Medicaid coverage to any of those classes.
The problem is it's not clear from the administrative records until very recently which enrollees were entitled to which set of Medicaid benefits. So we have in the MCBS both administratively identified Medicaid eligibility and self-identified eligibility. We spend a fair amount of our time trying to sort out what that means.
DR. WARREN: Is that a reasonable way to get at questions of economic status using those as separate? It seems to me that if the communities of color are increasing in numbers over time, and we know that they are, if the health status of those populations are increasing over time, and we know that they are, it seems to me that we can do some really creative work looking at Medicaid vs Medicare eligible. That may be a way to really look in the window of race/ethnicity rather than look at it as just a conglomerate mass.
MR. WALDO: That's one way. The problem with doing that with the Medicare population is that so many people become Medicaid eligible because they've been in a nursing home. You have people of all different SES classes spending down to Medicaid eligibility.
I would think that a more efficacious way of doing it would be to use something like companies like Claritas that develop SES scores for zip codes and associating that with the zip code of the respondent and forming an environmental socio-economic score.
We do ask beneficiaries in the MCBS for family income. It's income for themselves and for their spouse if they live with their spouse. Our experience in getting answers to that question has been spotty because people who are currently eligible for Medicare don't have much problem in talking to you about incontinence, but questions about income are considered to be very personal. We get an awful lot of item non-response to questions about income and an awful lot of responses that are artificially vague.
DR. WARREN: My only suggestion if the minority populations are growing in numbers and the whole question of Medicaid eligibility and opportunity is also going to grow.
DR. PEROW: Good afternoon. I'm Ruth Perow, the executive director of the Summit Health Institute for Research and Education. If I may just put in a minor plug there. The report that was referred to earlier today, the Commonwealth report which is I co-authored, is in the back if anyone would like to get a copy.
I have a commentary and a question for Mr. Waldo. The MCBS is certainly a useful window on the experiences of beneficiaries with the system. It provides information about their use, about their attitudes, about their satisfaction, and to some extent from their perspective, about the treatment they receive: flu shots, Pap smears, and blood tests for prostate cancer.
These are always from the perspective of the beneficiary. Clearly, that suggests that that's half of the picture. In order to really elucidate the disparities, and they are reported extensively by CMS's researchers looking at such things as sygmonoscopies, and heart treatments, and amputations, clear disparities in the kinds of treatments and services that these different groups receive.
In order to complete the picture and it probably lies outside the scope of this inquiry, wouldn't it be important to get information from providers about the services and the treatment and the outcomes that the beneficiaries they treat and see are receiving.
I don't know if you're in a position to talk about what CMS is doing about collecting data from providers, but certainly that will round out the picture in terms of what reality is since we're only really through surveys looking at things from the beneficiary perspective. That may or may not be the real world.
MR. WALDO: CMS has a colorful history with providers and I agree it would be really good to figure out some way to get information from providers matched on with what we're getting. We did a small experiment last year with pharmacies, getting pharmacies to send in their summary of patient medications to match up with our beneficiaries.
From a practical perspective, I think it will be quite some time before we're able to do more with other types of providers, but there is at the same time in a related area, a whole new effort to reach providers and to engage them in a non-specific way to educate them about our programs and to find out what things work and don't work besides the fact that we don't pay anybody enough and go from there.
There may be things in the future that will happen. In the next six months, I wouldn't hold my breath at all. I should also add that the MCBS is not the only tool that CMS has available to look at issues relating to beneficiaries and their interactions with our programs.
We have a series of peer review organizations, one in each state, and they have a contract with the agency. Each of them has a special project that they're working on. Many of them are working on issues specifically about race/ethnic differences in the use of services and access to services and outcomes. We don't necessarily need to put all of the baggage on the MCBS, on this one camel, in order to find out what's going on with the populations.
DR. MAYS: Let me make this our last because Dr. Clancy has joined us and she's up next.
DR. GIBBONS: Chris Gibbons again from CMS. I was going to comment on the last issue, but I won't. I have a clarification and a comment, and a question. Going back to the oversampling issue, I understand what you mean about the administrative dataset being different from other sets, but in my mind, it's not so much an issue of is it perfect or can we do it?
I don't know of any dataset that's perfect or any analysis that can be said to be fool-proof, but in my mind, it's would it better than what we have now? You have clearly said in your very good presentation that what we have now is not very good for minorities, particularly Asians and others.
My question would be, wouldn't oversampling given the budget restraints and analysis constraints, wouldn't it be better than what we have now? As a follow-up on that, I'm a little confused because when it was brought up, you said that that diminishes the power of the survey. One of reasons you had the survey was to detect rare events.
Did I misquote you? On your slide, you said things you should not do, detect rare events. I'm wondering, if this does not sample minority populations adequately, the numbers aren't big enough for most of the analyses you want to do and you're looking for rare events, what populations do you think you'll be able to pick up rare events in?
This one about receipt of the flu shots. I'm African-American. I love the fact that CMS made these cards and things for African-Americans. When I look at this data, it says to me more than 50 percent of every population did not receive a flu shot. In fact, we have a much larger problem than just saying let's do it for one category and we've done something.
African-American is worse and we should do that, but I think we're missing the majority of the problem. We have a systemwide problem here is what this data indicates to me. Finally, I'm interested in Dr. O'Campo's talk because there is much in the literature about SES measures and looking at income.
I'm waiting to see Dr. O'Campo's thoughts on this, but my understanding is that a correct understanding of SES cannot be conferred by a single income fact. SES is actually a merging of three concepts: income, education, and occupational prestige.
Just getting a number figure that you come in would not be the right thing. There is a significant amount of data that shows that it's not actual family income or individual income either, but the relative distribution of income. I'm sure you're aware of that data as well. There are several issues here that I'd be interested to hear your comments.
MR. WALDO: You've made a number of excellent points and I'm looking to the Chair.
DR. MAYS: You can answer. They're asking me about questions.
MR. WALDO: Okay. The question of the power of the survey is straightforward. Anytime you oversample any group, basically, the observation can no longer be used one-for-one. It ends up being weighted. The effect is sample size.
If you take 1,000 people drawn at random, "n" is 1,000. If you take 1,000 people stratified in 2-3 different groups, the effect is less than 1,000 and that's what reduces the ability that the power of the survey.
When we try to detect rare events with a large sample size, you can still do that. The problem is trying to then differentiate within those cases, small events. Is that rare event different qualitatively for someone who is African-American or someone who is Hispanic or someone who's White? That's the part that we can't do.
DR. GIBBONS: You won't be able to pick them up reliably in an African-American or Hispanic. You'll only be able to pick them up in a majority population.
MR. WALDO: Maybe not even in the majority population.
DR. GIBBONS: Precisely. I'm saying how can you even have that as a goal of it anyway?
MR. WALDO: One of the things you can do is to assume or hope that, in fact, racial or ethnic origin of the respondent has no effect on the rare event.
DR. GIBBONS: We already know that that's not the case. It might be a personal bias, but I do think the literature that race has a profound and ethnicity and a whole bunch of other things has a profound impact on what happens with regard to health.
MR. WALDO: There is no question on health. I was speaking of the rare health events. That may be the case. I will say now that the MCBS is not going to be the tool to help define whether there are race/ethnic components to the rare event. That's going to have to be some other type of study. That may be a study where the sample frame is rare medical events, where somebody does a more clinical study of the work.
DR. COLEMAN-MILLER: Many of them have been done already and we can give you all the data. I have it here and have it in our book. There are many, many rare events. It's been proven that there's a racial issue. Some of them 200 percent higher in the racial issues. Let's go back and start back where we already know that there are racial events.
DR. MAYS: One of the things that I'm going to do because there are other questions that are margins is that part of these questions are a theme and I don't want it to be also just your survey. I think as you heard, these are some broader issues. I'm going to bring us to a close now because Dr. Clancy is here and then we have about 30 minutes of discussion after her presentation.
If what you can do in terms of these questions is to bring them to this broad level, we can even discuss these after Dr. Clancy presents. I hope that what everyone is hearing today is also that it's not just one survey. People are beginning to identify several surveys.
I'm also hoping that the message is the call for change that I'm hearing and the ability for us to begin to ask some questions about what's possible. It may be that what we have now may not be the best given where we have to head for Health People 2010.
The questions may be before we hit the mid-career of Healthy People 2010 and we say we haven't hit the mark yet, that we've just introduced these issues now. I hope that part of what the survey people realize as what's going on today that people are giving input as to analytic direction as well as presentations because we're all trying to get to the same place in terms of Healthy People 2010 agendas.
I don't know if people are clear about this, but many of the things that were set as the targets are set on using these surveys which is part of the reason that we're here today. What you didn't hear from, sometimes that's a problem in terms of even some of the targets.
Let us thank you and tell you that we appreciate what you shared with us and what has been generated by people in response to it. I think it's useful for all of us. Thank you very much.
The other thing I'm going to ask is that there's a break and I'm hoping that you two can both stay during the break. It's after this discussion so that if people have very specific questions, they can ask that.
Dr. Clancy is the director for the Center for Outcomes and Effectiveness Research at the Agency for Healthcare Research and Quality and as many of you may know, they are currently also working on looking at issues of health disparities. We're quite happy that she was able to be with us today.
DR. CLANCY: Good afternoon. I'd like to thank the committee for this opportunity to share some thoughts about reducing disparities in Healthcare and what data do we need. When I first talked to Vickie, she told me she was interested in a wish list. I hope that I won't disappoint you.
I will tell you that the biggest challenge I had was trying to figure out what I could add to the really incredible, fine report that Ruth Perow just mentioned a few moments ago from the Summit Health Institute on Research and Education. If you haven't gotten a copy funded by Commonwealth, you will have left without a really fabulous report.
I want to just cover, in general, four broad categories. We had a little computer challenge of our own at home this morning. I don't have hard copies. I'd be happy to e-mail this to anyone. My address is cclancy@ahrq.gov. I'll remind you again at the end.
I want to set the context for disparities in Healthcare as opposed to health and talk about some challenges in the federal data collection process and try to get you to thank with me about what we mean by federal data collection. I do think that that was a very fine contribution of this report I just mentioned, and then speak a little bit about what we don't know and need to learn and finish on the cost of not moving forward, what will we lose if we don't do this?
It's no surprise to people who have been doing work in the field of health services research that there are many of what we might call non-clinical determinants of health outcomes. Although we're very focused on disease processes and the relationship of the processes of care to the outcomes of that care, we've known for a long time that patient characteristics, what kind of practitioner you see, where you get care, patients' preferences, and indeed how clinicians are paid for the services they provide in institutions and so forth, all have a big impact on health outcomes.
Trying to tease all these out hasn't been so challenging, but this is really old news. Of course, healthcare is only one input to quality of life. There are also other inputs that include personal behavior, education, genetics that we're all lucky enough to be born with or not, public health inputs, economics and so forth.
I want to be clear that I'm limiting my remarks to heathcare even though a big challenge that we face collectively is trying to figure out how much of the observed disparities in health that we observe are amenable to improvements in healthcare services.
Based on federal data collection that we already have, we know quite a bit and much of it is fairly depressing. I'm not going to bore you with a very long litany of this because knowing the number of people that I do in this room, I know most of this is very, very familiar to you.
This is analysis of data from NCHS which actually just displays visually causes of excess death among African-Americans and on the left side you see women where cardiovascular disease is clearly the leading cause of excess deaths. On the right side, you see men where cardiovascular disease is also the second cause here to trauma, HIV, and drugs and so forth.
We know that we have huge challenges ahead. This analysis comes from our hospital cost and utilization project. This looks at hospital admissions per 10,000 persons for the older population, 65 and older, and stratifies the results by income. What you can see here is that in these data, are age-adjusted.
Those people who live in a zip code where the average income is under $25,000 are over three times as likely to be admitted to the hospital for an episode of pneumonia that was, in theory, preventable by appropriate vaccines and clearly for those people who live in zip codes who are the median income is over $35,000 have much smaller proportion of avoidable hospitalizations.
It's race/ethnicity, it's socio-economic position. It's many different factors all interacting. The question is how do we actually begin to tease those out? Speaking as a clinician, this is no news for any physician in this country. It can't possibly be news. They see this all the time from the beginning of their training on out.
The real question has been for some time to what extent is the healthcare system own some part of this problem or do all of these problems simply present themselves to healthcare and, therefore, clinicians and providers are in the position of trying to deal with what presented to them.
What you see in this slide are odds ratios indicating predictors of referral for cardiac catheterization. This comes from the infamous study published in the NEW ENGLAND JOURNAL in 1999 by Kevin Schuman and colleagues. They had physicians who were attending professional meetings.
Understand, they were trying to get a diverse sample of primary care physicians. One could also make an argument that those primary care physicians, but one can also make an argument that those primary care physicians who have taken the time out of their schedules and invested in going to a professional meeting may be somewhat above the norm.
They were shown videos of patients where all the socio-economic and occupational characteristics were controlled for. These are actors portraying patients and physicians who agreed to participate in this study could see one of eight different actors. They varied whether the individual was 55 or 70, whether that person was male of female, and whether they were African-American or White.
Having had an opportunity to see the videos. They're incredibly life-like. You feel like you're in the clinic. Because they were actors, they could actually manipulate the script so that literally individuals used the same words to describe their symptoms and so forth which not often how it happens in real life.
For the purposes of the experiment, this was very nicely done. They were able to demonstrate was that these physicians were less likely to recommend invasive diagnostic and therapeutic procedures for older African-American woman. There has been a lot of argument about how the statistics were presented, whether the research was manipulated, the media and so forth.
There is no controversy about the fact that for the older African-American female in this set of scenarios, physicians were significantly less likely to recommend invasive procedures for that individual. So I think the reason that this study generated so much controversy was simply because for the first time, this brought it home to healthcare.
This wasn't about what goes on around heathcare. Suddenly, this was the disparities are here and it's our problem too. This was the more popular connotation from the study which was published several days after the study was reported widely in the press depicting an African-American woman saying, give it to me straight; I can take it, what's wrong with me?
The physician is saying, you're not a White male. At AHRQ, we're intensely interested in the impact of research. This has become a new metric for us. The time to cartoon index.
Just moving along, a year later in the NEW ENGLAND JOURNAL, one of our investigators actually did a study looking at re-profusion therapy and Medicare beneficiaries who just had a heart attack. For those not clinically inclined, this is known as clot busting.
What they did with very careful data because they work very closely with the quality improvement organization, was to identify eligible Medicare beneficiaries, that is to say, those that would benefit from this lifesaving, evidence-based procedure.
They found yet again was that White patients were significantly more likely to receive this treatment than Black patients. Notably, Black women were significantly less likely than anyone else to get this. So, the take-home message from this study in my view are two.
One, once again disparities in care associated with race and significant disparities are clearly demonstrated here. The second is that we're not doing so great overall. It's hard to believe that 59 percent of eligible patients receiving an evidence-based lifesaving treatment represents the Everest of our ambitions and I'm going to come back to this point at the end.
The question is what is happening here. That is really the big challenge before us getting to why. We've come up with one framework that was actually borrowed from one that John Eisenberg developed looking at voltage drops to quality of care.
What are all the possible drop offs between a population-at-large, a healthcare system presumably there to serve their needs, and everyone getting quality of care? The first drop off is do you have insurance, do you have the option of buying insurance?
The second is, if you have the option of buying it, can you afford it, and are you enrolled in insurance? A third step is if you have insurance coverage and you can afford it, are the providers and services that you need covered or do you have a policy that's like one of those hospital gowns; from one dimension it looks like you're covered and everything is fine, but where it really counts, you can see that there are important gaps?
The fourth drop off is informed choice of services and providers available. The fifth is are primary care services accessible? We know that this makes a big difference particularly in receipt of recommended preventive services and so forth. The sixth is, do you have access to the specialty services and institutions that you need finally getting us to quality of care.
I don't know that this is the only framework. It helps my thinking to think about all of the possible steps in a cascade between where we'd like to be and where we are right now. From some researchers at AHRQ, we know that even if income and health insurance coverage were equalized, difference in access to and use of health services would not be eliminated.
In other words, 1/2 to 3/4 of the disparities that we see cannot be explained by these factors. Very importantly for the charge before this committee and before all of us is that it's difficult to identify a single factor that would resolve race/ethnic disparities. It would be really lovely if there were one.
We would just push that button and away we'd go or we couldn't push that button and would begin to collect our evidence and make the case to people who did have the power to push that button, that what's we needed to do and it's not that simple.
For us at AHRQ, this is not an academic proposition. We have been charged by the Congress that starting in 2003 to produce an annual report regarding prevailing disparities in healthcare delivery as it relates to racial factors and socio-economic factors in priority populations.
More than that, our legislation goes on to define priority populations as consisting of people living in rural areas, inner-city areas, low-income groups, minority groups, women, children, the elderly, and individuals with special healthcare needs. This feels like a really difficult thing to get your head around in terms of how we're going to display this in a way that people can understand much less figure out how to interpret the data, you'd be correct.
That's the challenge before us. I can discuss in academic terms the challenges of isolating which factors are most important under what circumstances, but we're actually going to have to produce a public report in about a year. So with that as a back drop, I'd like to just move to talking about the federal data collection process since that's why we're here.
Now, the first issue I'd like to raise for your attention, and I think that Commonwealth Fund supported report did a very nice job about this, is what do we mean by federal data collection? On one level, there are many people who have a very clear connotation in their own heads about what the mandate of this committee is and what your jurisdiction is.
It's about surveys and very clearly defined efforts. What I think this report did was to put it all in one big paper or map that it could also include data collection on claims for services and efforts where the federal government has joined forces with the private sector to assess and improve quality of care.
That is one important point. The second is that data needed for surveillance in monitoring may not be the same as data as data needed to improve healthcare. In theory, they're the same and to the extent that they're the same, that suggests enormous efficiencies in synergy of purpose and that's wonderful.
But it's very clear that in measuring quality of care there has been a great deal of thought given to the criteria that one needs to have an ideal quality measure, it needs to measure something important, it needs to be feasible to collet the data, it needs to be under the control of providers or healthcare systems.
I'm not completely sure that we have same set of criteria clearly developed for reducing disparities even in healthcare. We know from far too many experiences that data on race/ethnicity are not consistently available from healthcare providers. This year, the agency has been funding an exploratory analysis looking at some integrated delivery systems and how they collect these data.
It's been fascinating. The first fascinating piece of information was it took this group several months and multiple conference calls to actually make sure that they were defining their terms in equivalent ways. The second, was finding out that some delivery systems actually have a blanket policy of not collecting these data.
Getting to the why behind that policy is a little trickier, but I don't think that we're going to make much headway until we do understand some of those whys. Some of that relates to the fact that some healthcare providers believe that they're providing equivalent care for everybody.
I don't know what the other reasons are, but I think if we don't know them, we're not going to move too far. The data that are available aren't comparable which speaks to moving forward with compliance with OMB standards. Some of the other challenges are even when there are data on race/ethnicity collected, the collection is often incomplete.
Most of the time, the data on race/ethnicity aren't easily linked to data on those other characteristics I was discussing a few minutes ago such as insurance coverage, employment, income, education, and very importantly, the local context.
For example, if we were to look at disparities in control of pain, most people would agree that this would be a really important thing to do, but it would be very important information and in terms of trying to figure out how to fix the problem when you found what I would say are predictable disparities associated with race/ethnicity to know that in some neighborhoods, pharmacies don't stock those pharmaceutical for a variety of reasons.
Clearly, no news to anyone in this room, our health statistics now tell us that there are disparities, but don't actually tell us how to address why they're there or how to address them. Of course, the time lag here when you're talking about healthcare becomes quite important.
We know that improvements in quality aren't possible without relatively short turnaround feedback of that data. To imagine telling people, and I can tell you as a clinician and anyone in the room who has been in the situation of getting feedback on your performance from two years ago in whatever managed care plan your patients are enrolled, you look at this piece of paper and say, oh. And I should do what about this now?
To the extent that we want to use the federal data collection process to look at improvement, we're going to have to experiment with ways to make some of that data more rapidly available. I mentioned that the agency is going to be sponsoring the first annual report on disparities in healthcare.
We're also at the same time going to be producing a report on the quality of healthcare which gets very interesting. When we turn to the IOM for advice, they suggested to us that one of the six important 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 and so forth.
It's very clear that we're going to have big difficulties with an adequate sample size for most federal datasets. What we don't know and need to find out to move forward, we don't know why disparities occur. We know that that doesn't tell us what is race or socio-economic position or other characteristics of proxy for.
Your skin color, per se, or your sociodemographic characteristics don't, of themselves, to lead providers to recommend one treatment for you or another. What is actually going on here? Again, what proportion of the disparities in health that we see are amenable to improvements in healthcare?
This is not an easy conversation and very much what's up against the boundaries of accountability for healthcare enterprise. We know very little about what local circumstances either ameliorate or increase disparities. I would submit to you and I think the Shier report makes this very clear, that we don't know anything about how to collect these important data that we need respectfully and effectively.
We simply do not have the space in this country right now for individuals to ask all the time of another individual how would you describe yourself even though the answer seems self-evident to them. Finally, we don't actually know how to link evidence of a problem to potential solutions.
My wish list is 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.
To really get at the intersection of these different characteristics since it's pretty clear that there is no single magic strategy, one is left either with the possibility of mandating collection of family income and education data or developing a mechanism to link these with other data sources.
This raises some very serious concerns about privacy and confidentiality, but if we're going to get to why and try to figure out how to fix the problem, it's very important. We need to figure out how to develop and implement a strategy for identifying relevant local characteristics.
I do want to say on this wish list it occurred to me on the way down here that we do have a very important opportunity to learn right now from data collection that's been ongoing for the past couple of years. Many of you may know that the Medicare program mandated collection of the Medicare Health Outcomes Survey over the past couple of years for those involved in plans risk plans or Medicare Plus Choice plans.
This is a survey which will report on a two-year change score in the SF36, but also has huge amounts of sociodemographic data and I think there is an awful lot of important lessons that we could learn from the process of how that worked, where there were problems collecting the data.
There have been some very important improvements made along the way by the survey vendors to try to increase response rate and to get the right answers, but that alone, represents data that are already collected from which we could learn a great deal.
It's going to be very important to explore strategies for rapid release of selected data for improvement purposes. We also need to think very carefully about our strategy for reporting these data. To return to the Medicare Health Outcomes Survey, the question now is given this wealth of sociodemographic data, how do you present the results?
Do you adjust for the differences which some people would say may be exacerbating the pathology or do you report in a stratified fashion? To the extent that you believe that many of these characteristics exert their influence outside the control of healthcare systems and providers, you would say we should adjust for those differences so that we're reporting fairly.
TO the extent that some of those differences actually reflect likely differences in the experiences that individuals have with healthcare systems, then the results should be stratified. The right answer might be and rather than or.
It might be that we want to try both strategies. It's going to be very important for this committee to seriously consider how we get there. We may actually need to rethink our current approach to data collection. We know what all improvement is local, but most of our data are nationally representative.
Are there ways that we can explore partnership opportunities with the state and communities to try to begin to get at this issue? If you think about the distribution of different populations across this country, it's clear that nationally representative data simply are not going to address the needs of many of the ethnic populations that are of concern here.
Finally, we need to learn a great deal more from the users of these data and from other sectors. It's been incredibly intriguing to me that if you live in the Washington area, you don't even need to have children to actually know which elementary schools are considered high quality. We're getting regular reports in the newspaper about how children performed on the educational exams.
Even before that, you just turn to the real estate section and you see it's the first line in the real estate ad. We're very used to the notion of tracking education performance and there is a great deal of interest in thinking about how you deal with socio-economic position and other characteristics of communities in that sector.
It feels like there is an enormous opportunity for an important conversation there particularly for children but also in terms of how different communities are using the data and interpreting them and using them to improve.
Finally, I have a few thoughts on the cost of not moving forward and then we'll get to the part I'm really looking forward to which is your thoughts. If you followed me so far, you would be taking seriously the premise that disparities in healthcare represent a critical opportunity for quality improvement.
What do we lose if we don't do this? The first is that I do believe that if we try to improve healthcare without considering race/ethnicity and socio-economic position, that we run some risk of undermining quality measurement efforts more broadly.
We won't be necessarily be targeting the costs and considerable investment of energy and resources required by data collection to those in greatest need. Predictably as healthcare expenditures are increasing, one can expect that there is going to be some push back on the cost of collecting data for improving quality.
The next question is where do we know we have the greatest opportunities to improve. If we're not liking that conversation to collection of data on race/ethnicity and socio-economic position, I think that we may lose both from the disparities perspective and also from the quality perspective.
Ultimately, if we don't collect these data, we're going to deprive ourselves of important scientific knowledge. The example that comes to mind is that of African-American woman and breast cancer. We know that African-American woman tend to present with more advanced disease than White woman do, those woman who have breast cancer.
There are conflicting results. One studies suggests that mammography may not be entirely protective. There have been some syntheses of existing studies which suggests that access to mammography and follow up of abnormal mammogram is part of the problem, but that there are also other factors going here that we don't understand, that there may be biological differences in the disease and so forth.
We can actually learn from measuring what are important scientific areas for us to invest in biomedical research. Finally, if we don't move forward in collecting these data, we're ultimately going to misallocate resources for quality improvement. If you have limited resources and you want to improve quality of care to raise population health, again, you would ideally like to target those resources to those in greatest need.
If you can't identify those in greatest needs, that's not going to happen. With that, I will thank you for your attention and look forward to your questions. If you want the slides, please send me at e-mail, cclancy@ahrq.gov.
I'd be happy to provide them.
DR. HANDLER: One of the ways to measure health disparities is to use racial data obtained from state birth/death records. You said all organizations receiving federal funds should use the OMB standard. Except for the State of California, the states have not decided to use the multi-race identifier on their birth/death records.
That causes multiple problems. There is no valid way to project outward past the year 2000 population data that the Census Bureau collected in 2000. There is no valid way to calculate birth/death rates for the year 2000 because the denominator is based on multiple race identifiers.
I've spoken to Jim Weed about last week and asked why did OMB ever do this? Didn't they see what was coming? It would be politically incorrect to characterize our conversation any further. Where do we go from here? That's a growing problem and it's only one year away that we're going to have to confront that.
DR. CLANCY: I'm not sure that I have an easy answer. I agree that it's an important problem. I heard someone from Census presenting overall data for how many people actually use the multiple race category in the Census which may or may not be proportionally different from those reporting it on birth records and so forth. My recollection was that it was something on the order of 2.3 percent. It's a small proportion.
DR. HANDLER: It varies by different racial groups though.
DR. CLANCY: Exactly. The other question is whether it varies by different communities. Are people in California, for example, or other areas where you've got multiple diverse populations more or less likely to report that? There is not an easy fix here, but it's worth exploring and we could actually begin to understand what the range of impact that's going to have on our confidence intervals about knowing how important the problems are. I don't see that as a reason to stop.
DR. HERTIN-ROBERTS: I had a question for Mr. Waldo and hearing your comment and your comments, makes me think of this in a broader context. I'm not sure if it's a question or comment. My question for Mr. Waldo was about the PSU. There are about 100 in the survey that was used.
The survey was in place in 1991. The PSUs were drawn to produce nationally representative results. I wonder whether those PSUs have been adjusted over time. The demographics of the population are changing and will change more in the future. Perhaps they're not placed in areas where we could get a nationally representative sample.
Hearing your thoughts, I'm wondering is a nationally representative sample is what we really want. It will artificially deflate the impact of some populations that we see in certain areas. I'm thinking of Texas and California.
If we go only with nationally representative samples, we're going to keep running into the same problem that we keep having where the Asian/Pacific Islanders are underrepresented and don't seem to exist, neither do a significant number of Hispanic.
I'm wondering whether we need to do more than just fine tune our national surveys. Maybe we really need to rethink the way we're collecting this information to make is useful for policy.
DR. CLANCY: I know that for the Medical Expenditure Panel Survey at AHRQ, there has been considerable work in developing techniques to identify people in lower socio-economic groups for the purposes of oversampling in that regard.
When we've increased the sample size, we are reasonably confident that we will be enrolling enough individuals from that strata to make nationally representative estimates. I hear you asking a different question.
It is, should nationally represented, whatever, be the holy grail of all of our efforts? For some issues, the answer may be no. Having said that, the question is how do you get to what is the alternative and so forth? I don't know, but it's worth exploring the question about PSUs and how they've changed.
DR. PARK: I'm John Park. I'm from one of the local governments in Montgomery County in Maryland. Thank you very much for bringing our local issues among other things because as you have said, we locals are the ones who really bring about the changes in healthcare, but we are the ones with the least amount of resources to go about doing that.
I was going to bring up this issue about locally specific data that we really need to have in bringing about those things. I really don't know how you think about as far as bringing the locally valuable information for us to use especially in dealing with racial disparities.
Even in Montgomery County where we are a very affluent jurisdiction, we have huge disparities between Whites and African-Americans and between Whites and Latinos. We have formed interest groups to deal with those things, but our resources are limited because we don't have the data to support what we really need to find out.
DR. CLANCY: This was not a planted question. I'm participating with a steering committee for Latino health initiatives in Montgomery County.
DR. PARK: I wanted to thank you for that.
DR. CLANCY: One of the issues that we're struggling with is that in addition to limited data for many groups, the county has had a very rapid increase in the Latino population over the last decode. It's a very diverse population and data are extremely limited.
What county programs collect when they can is not necessarily consistent with what the hospitals collect is not necessarily consistent with what the state is collecting for some of its programs. None of it is consistent with OMB. This is an area that we're trying to push forward on at a nuts and bolts level such as could you add fields to this hospital discharge and so forth. It's very important.
DR. GIBBONS: I really enjoyed your talk and you made some excellent points. You mentioned about collecting a race/ethnicity data mandate from Congress. This comes up all the time at CMS where I am. It's interesting. We had some discussions a little while ago centered around our managed care plans that work with us and can they really get this and did internal sorts of things.
I'm not a lawyer and don't understand all of this, but at the end of all that, the problem was the perception and the problem among minority groups is they use the ERs differently and it's only a smaller problem than we think. When you look at legislation, we have that mandate and we can do it legally and without problem now, but we are now.
There is a very big and strong perception on the part of these plans that they couldn't do it and were afraid to do it because of the fear of legal things. That's the case in a broader sense. I don't really know, but I thought that was very interesting.
You mentioned your thoughts on if we don't collect this data. I agree with those. From my perspective, I think it even goes further than that. We won't be able to perform the science, but given the current demographic trends and projections in the Medicare population and the growth of minority populations and the shrinking of the now current minority population, if we don't begin to get accurate data, it threatens our accurate and the functioning of healthcare system bemuse we'll continue to do things that are not working for relatively small populations which will become relatively big populations.
Finally, you mentioned allocating the resources to those that best need. This illustrates the prevention paradox. When you look on an individual level, and your comment about research data and quality data being different, I 100 percent agree with that.
This illustrates that. If you take the research approach or the individual approach, that would be a good way to go and it does work, but it you look on a population basis, most times, those incurring the greatest expense of highest intensity are a relatively small proportion of the population, but they consume a lot of dollars.
If you focus on those, you've done nothing for the majority of the population who are also at risk, but lower risk. In terms of the policy perspective, which way do you go? Do you have the more population-oriented approach where you're dealing with people at lower risk, but at risk? Or do you focus on the small groups?
I'm part of a minority and I'm not against that, but in terms of moving population parameters forward, I think we need to do significant more thinking. A high-risk strategy approach may not necessarily get us there.
DR. CLANCY: Just in one response. I think your observation about misperceptions about the legal collection of these data is right on target and there has been a fair amount of work done about that, but I think a lot more education and dissemination that needs to go on around that point.
What we heard is a frank conversation with some leaders from managed care plans 2-3 years ago, trying to get at how would you collect these data. What we heard was that nobody had a clear sense of how to do it both respectfully and efficiently.
For example, putting race/ethnicity data aside for a moment, there was a point where a lot of health plans collected baseline health data. After all, if we're responsible for this defined population, we should actually have a program that helps us identify people that we may want to make more proactive efforts with and so forth.
That turns out to be a great, but really costly idea and hard to justify much of a return on investment. The question is: At what point in the process do you do this? Most people don't want that assessment of race/ethnicity linked too closely to enrollment for obvious reasons even if it's legally okay.
You many create a very strong perception that you're redlining in some fashion or other. the question is, at what point in the process do you collect it? Many people are enrolled in plans and depending on the plan, a third of patients may not show up. How to actually do this in a way that's efficient and gets us where we need to be is one of the challenges.
DR. POKRAS: Thank you, Carolyn, for giving that introduction to my remarks. The Office of Minority Health has been working closely with ARHQ, CMS, and other agencies and other organizations outside the federal government, including the Academic Medicine and Managed Care Forum to investigate these issues.
The AMMCF is a group of academic health centers along with AETNA to investigate whether it makes sense to collect race/ethnicity data. In addition to those collaborations and technical assistance, we also have funded a project at the National Health Law Project to review state laws and regulations to see whether indeed it's legal to collect race/ethnic data.
This is one of the rumors we have heard and one of the reasons that folks give us as to why they're reluctant to collect it. We found very few, four states, prohibiting the collection of race/ethnic data by health plans or insurers at the time of enrollment.
It doesn't at any point in time, it's at the time of enrollment. That offers the opportunity to collect at other points in time. As we heard earlier, Ruth Perow's study reviewed the federal laws and regulations and didn't find anything that prohibited health plans and insurers from collecting the data.
We're also working with seeing what we can do with our pending regulations. The most recent ones are the State Children's Health Insurance Program to allow plans to collect this information. That was published last summer as in interim final rule and a notice proposed rulemaking was published this summer for Medicaid managed care and continuing discussions are going along those patterns.
As we are getting the word out that it is not illegal from the federal point of view to collect the information and there are some very good reasons which have been outlined very nicely by Dr. Clancy why we would like to have the data collected, there are other folks fighting to make sure that the data are not collected.
The racial privacy initiative in the State of California, people need to be aware, those in the room and those listening over the Internet, those working in the anti-affirmative action movement are concerned that the collection of race/ethnic data will actually continue the patterns of differential treatment.
They're trying to avoid that and prohibit the State of California from collecting the information. It's gong to be on the ballot in November for California voters. They have an exclusion for medical research subjects and patients, but that's not well defined.
We're not sure how they're going to be carrying that out. There are various advocacy groups, including the ACLU who is now advertising for somebody to work full-time to gather information in regards to this initiative. Giving you a heads up via the Internet or in the room.
DR. MAYS: We are going to take a break. Let's thank Dr. Clancy for her time and her wish list. We were wishing for a lot of the same things too. That's good. We'll take a 10-minute break.
(Short break at 3:02pm.)
DR. MAYS: Michael Katz from CMS is going to do a quick update in terms of some of the work that CMS is doing on race/ethnicity to bring it back full circle to some of the questions that were raised and then we'll turn to our speaker.
MR. KATZ: My name is Michael Katz. I'm with CMS. I wanted to give a quick update on where we are in terms of population race/ethnicity data. Some of you may be aware that in October 1999, there was a conference on race/ethnicity at the CMS and there was a publication that came out in 2000, the Healthcare Financing Review.
Those numbers were pretty bad. At that time, there were 34,000 American Indians/Alaskan Natives enrolled in Medicare, 300,000 Asian-Americans/Pacific Islanders, and 450,000 Hispanic/Latino populations.
This past December through some of the attempts we have made at updating the data, those numbers are doubled from 1997 through December 2001. There are now over 70,000 American Indians/Alaskan Natives enrolled in Medicare, 600,000 Asian-Americans/Pacific Islanders, and 900,000 Hispanic/Latino populations.
We're doing a lot better and we do try and it's a live issue. We're constantly working on trying to improve the data. We're working with SSA to try to get language data into the mix. SSA has language data in two parts of its organization. We only receive one of those parts.
The part we receive has a language preference for 10 languages. The client data file has over 27 languages. We are on the move and do take this very seriously. Thank you.
DR. MAYS: Our next presentation isn't on a particular data set. It's on a topic. It's on socio-economic status. We've talked about whether or not we need things other than just race/ethnicity to look health disparities in race/ethnic groups.
Ssocio-economic status often is talked about and we worked hard to make sure we would have a presentation on it and multiple race, the two things that people keep referring to. Dr. Patricia O'Campo, who is at Johns Hopkins, was kind enough to rearrange her day.
She taught this morning and rearranged some other things to come over and be with us. Many of you probably know her work in this area as she has been given awards by the IOM as well APHA for some of her work in this area. Thank you for being with us.
DR. O'CAMPO: Good afternoon. Thank you for inviting me to talk about this very important topic as it relates to understanding race and health and racial inequalities and health.
I have about 30 minutes to talk about this topic. If I took the whole 30 minutes, I wouldn't be able to summarize the vast sociological literature that deals with theoretical considerations and measurement of socio-economic position.
It's safe to safe that we can boil it down to a central concept that socio-economic status captures stratification of society into classes of individuals or groups of individuals who have differential access to a variety of resources, economic, political, health, housing and other resources.
That differential access to these resources results in different life chances. Those different life chances impact health directly and result in the gradient that we're all familiar with and it's been demonstrated in hundreds of studies to date. There are a variety of ways to measure socio-economic status.
I'd have to say in the area of health, most of them focus on individual measures. I will, during this presentation, take a little bit of time to talk about measures that up and beyond the individual level. It's important that we try and be very complete in our categorization and measurement of socio-economic status especially if we're trying to understand racial inequalities in health.
I would like to do during this 30 minutes is to briefly review some of the measures and talk about some measures that haven't been used very often in the US, but that may help us more completely characterize socio-economic status and then talk about their relevance in understanding racial inequalities in health.
The measures that we include in our surveys should help us understand whether and why some people have more material resources, power, and prestige than other, whether and how socio-economic status has an impact on how, and how socio-economic status varies by race/ethnicity and how socio-economic status helps us explain racial inequalities in health.
I'll review some of the measures, many we'll be familiar with. Sociologists consider two groups of measures when we're thinking about socio-economic status at the individual level. The first is social stratification measures.
The idea behind that is that we can take individuals in society and order them hierarchically according to education or income. We're all familiar with those measures. That hierarchical gradation describes the different life chances that groups have.
Those at the top have better life chances and also better health. Social class are a different type of measure. They suggest that we can break society up into classes. They're called relational measures. The classes are related to one another.
They usually have opposing interests. When one class benefits, sometimes, it's at the cost of another class. That's why they're related to one another. That's at the root of explaining the inequalities or the way that inequalities and gradations are created.
I can't go into a lot of the theory. There are several very good reviews of socio-economic status and how it's used in health in particular. For example, there was one by Liberatos in the late '70s. In the mid-90s, there was one by Kreeger. Most recently, there's one by Kaplan and Lynch in the social epidemiology textbook that has come out in the year 2000.
Let's start with some of the hierarchical measures. Those we're most familiar with. We have gradation measures that describe economic gradation, income, wealth more accurately captures the total resources that are available to a family or individual more so than income which can fluctuate from year to year. Wealth is more stable.
Power-based measures include occupational try to hierarchies. Prestige-based measures include education. It put an asterisk by the ones we're most familiar with. Education in epidemiology is the one that is used the most followed by income. Let me show you data that show a gradient that almost all of us are familiar with.
Here we have data with intimate partner violence by annual household income. Here we have data on reported crimes related to that and income along the bottom. There is data separate for men and women. Women are victims more than men.
We can see the gradation for both genders. As income is higher, greater than $75,000, the rates are lower and the rates increase as income decreases. This is a familiar pattern in the literature. The same is true for no physician content in the past year. These data come from the NCHS book on socio-economic status and health.
We have data looking at no physician contact in the past year for those 18-64 and these are proportions of men and women ordered by income: poor, middle, and high. We see the familiar gradient. We're less familiar in the US with relational or social class measures of health. I wanted to talk about those because they can add an important dimension to our understanding socio-economic status and health.
It doesn't mean that we shouldn't look at income and education, but in addition we might look at some social class measures. There haven't been too many measures developed to look at social class in health. The ones that are most well-known and used are ones developed by Eric Wright.
He developed measures which look at organization assets. The idea here is that social class identifies two or more groups of people who have opposing interests in different interests. They are differentiated by their ability to have control over ownership assets.
That refers to whether or not you were employed or self-employed. People who own businesses have control over ownership assets. If you're thinking about organizational assets, people in managerial positions or supervisory positions within large organizations have power and access to political power that benefit their particular class interest sometimes at the expense of others.
We can see in the media the discussion of how some of this plays out. The Enron example we've heard about how top management undertook certain actions which benefitted themselves and their class at the expense of those who were under them. We can see how organizational assets work here.
It's no intuitive. So let me tell you about some of the measures that are used to look at organizational assets first. Some of the questions that are used to determine what a person's status is when it comes to organizational assets has to do with whether or not they are a manager, supervisor, or neither.
You can ask a person if they have involvement in policy making of a particular organization. People who are CEOs or chairs of departments would have input into policy making whether it's setting wages for certain groups or benefit levels. They have input in policymaking for their firm.
They also have the ability hire/fire individuals. Managers tend to have powers in both areas. Supervisors have the ability to hire/fire individuals, but they often do not have policymaking powers within their firms. The rest of the people do not have either supervisory or sanctioning or policymaking power.
That's one way to categorize people by social class. An alternative way of looking at it would be categorizing managers according to the number of levels they have beneath them. If you're a chair of department and then have centers and within the centers, there are research projects going on with at least two hierarchical levels beneath you, you can be considered a manager.
First-line supervisors have authority over at least three or more workers, but only have non-supervisory workers beneath them and then you have the rest of the folks. This is a way of looking at ownership assets.
You can characterize large employers as those who have at least 10 employees and they're self-employed. Small employers have somewhere between 2-9 employees and they're self-employed. Then you have people who just self-employed, but may have one other person in their firm and then you have people who are none of those and they are just actually earning wages.
The data is interesting. We don't look at this often in the US, but in Europe, they've been a lot better at looking at socio-economic status. These are data that are showing standardized mortality rates for European population. We see excess deaths per 100,000 persons compared to the professional class which tends to have the lowest mortality rates here.
We can see as we move down the line for people who are somewhat skilled and then unskilled, there are excess deaths and a gradient there as well. As we move down the line of social class, we have greater numbers of excess deaths.
Another example from the US is looking at the National Longitudinal Mortality Survey for 1979-1991. We've categorized people according to professional/managerial vs the rest of the people. We have mortality per 100,000 population of males. It's also stratified by economic sector.
Different economic sectors have different situations when it comes to professionals and non-professionals. In the government, you have higher job security and more benefits for non-professionals than you do in other places.
You might expect that the gradient between non-professionals and professionals would be lower. That's what we see. The lowest rate of inequality here between professionals and non-professionals within the government. The highest level of inequality in the manufacturing and service area which we might expect as well.
There are low rates of unionization and here is where we have some of the lowest wages and benefits. The fastest growing rates of contingent or part-time jobs are here. We might expect to the see the largest gradient here within this sector.
The overall pattern tends to be that professionals compared to non-professionals tend to have lower mortality compared to non-professionals. I want to share one more example that more closely relates to some of the measures I showed you before about manager/supervisor and non-managers and supervisors.
All the slides are going to be put onto the Web site. You'll be able to look at the data more carefully later. These data are concerned with depressive disorder. It comes from the Baltimore Epidemiologic Survey. It's a community-based survey where the population was asked about various mental disorders and the diagnostic interview survey which more closely resembles diagnoses in the DSM3 were asked of the population.
This is DSM3 major depressive disorder. Here are two sets of data. These are the ones that represent differences in major depressive disorder in this particular community by income. The population was divided into the highest, mid, and the lower tertile.
We might expect that the lowest tertile of folks would have the highest rates of major depressive disorder. We might expect a gradient. We see here is those in the lowest tertile have the highest levels of major depressive disorder. There isn't a gradient.
I don' think there is much of a gradient between these two groups here. We do see a difference between those in the lowest tertile and those in the middle and highest in terms of major depressive disorder. We have the social class measures and see some interesting findings.
In the turquoise, you have people who are considered managers. They are the ones who can make policy as well as they can hire/fire. Supervisors just hire/fire and supervise, but they don't give input in policymaking. Then we have those not managers or not supervisors.
We can see that the managers have the lowest levels of major depressive disorders followed by those who are not managers and supervisors. Those people who are supervisors who have the headache of supervising people and are caught in the middle have the highest rates.
That added something up and over what we typically see when we look at just education or income. Knowing a person's social class position can tell us something about the risk at least for mental health disorders. I'd like to spend a little time talking about area-based measures.
As you can imagine, socio-economic status is a very complex issue. We can't completely measure socio-economic status if we just focus on data at the individual level. We're talking about the kinds of data that should be included in surveys and although you don't ask about information about neighborhoods or get objective measures of area-based measures in surveys, you do collect information that will enable you to access this information such as complete information on address.
I am going to talk very briefly about area-based measures. Including data beyond the individual is important for a complete characterization of socio-economic status. Area-based measures can fall into three categories. You can characterize occupational structure such as the proportion of people in a given area that have professional vs non-professional.
You can characterize a particular area by educational structure. The proportion with college degrees vs not. For economic structure, there are a number of things you can look at such as levels of income, wealth. The two are very different and not so highly correlated that looking at both would be advantageous.
You can also look at unemployment levels. We did some of that in Baltimore when we were looking at intimate partner violence during the childbearing year. We first analyzed data based on just the individual information. At a later date, we got information on the area of residence and added that information to our analyses and found the following results.
There are odds ratios for the risk of intimate partner violence during the childbearing year. That's during pregnancy up to about six months post partum. We have two measures of individual levels, education, and income.
We added three indicators of area level economic position, the ratio of homeowners to renters, the unemployment rate and the level of income. We compared lowest quartile for the city as a whole to the rest of the city. This particular population was a low-income population.
Most of the population was receiving Medicaid as a form of insurance. Despite that, we found that these women were living in all areas of the city. As I mentioned before, we had broken the city up into quartiles in terms of income. We found is that these women lived in all four quartiles of the city.
If you're low income, you may not necessarily live in a low-income community. That's important to keep in mind when you're trying to characterize socio-economic status. We found that when all these variables were in the model, individual level income was no longer important.
It is perhaps because the sample was rather uniformly low income. Education was significant and higher education put you at higher risk which is consistent with the theories related to intimate partner violence. I can explain that in greater detail at another time.
It's important to see that even when education and income are in the model, the area level measures were significant as well. The magnitude was actually quite great in terms of thinking about risk of this particular intimate partner violence.
Including measures of area, socio-economic status can help us more completely characterize our population. There was another important lesson to learn with this particular example. Neighborhood SES measures added to our understanding of the risk of intimate partner violence and the magnitude of the neighborhood measures were larger than the individual level SES.
We discovered that there was effect modification and confounding that was present when we added the area-based measures. When we just looked at the individual level data without any neighborhood data in there, the variable raise was not significant.
Baltimore is very black and white. Our sample was really just black and white. Once the neighborhood measures were added into our sample and adjusted for a lot of the neighborhood environment factors, White woman compared to Black women were nine times more likely to be at risk of intimate partner violence. There was effect modification by race that we were able to see when we added the information on area.
The stratification measures in the US are more common, the income and education, but the relational measures may add important information up and beyond the stratification measures that we typically use. The relational measures may capture important issues related to race and ethnic discrimination.
If you're thinking about occupational variables, there are certainly discrimination when it comes to jobs, promotion, as well occupational segregation and some of the relational measure may help us understand the extent to which persons from different race/ethnic groups have access to certain types of jobs.
Area-based measures can help us capture the complexities of socio-economic status. We shouldn't limit ourselves to just understanding socio-economic status at the individual level. Measuring it more than one way may be the most informative.
Much of my work focuses on women's health and the literature confirms some of the issues that SES and health patterns differ by gender. The best measures for women may be different from 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.
The same issues may apply for men. It may be that household socio-economic status is really the best indicator overall. Here we have data from the NCHS, socio-economic status and health where we're looking at lung cancer mortality for those 25-64 by income.
We can see that the gradient that we usually see for men, this is individual income, personal income, not household. We don't see the same gradient for women. We do see differences between those lower income and those higher income, but we don't see this nice gradient here.
The same thing goes for looking at data from the NLMS. I showed you data before for all cause mortality which showed that the difference between the professional and non-professional class was fairly consistent across all economic sectors.
Here, we can see for women when you characterize it according to her professional or non-professional status. The gradients are not as steep as you saw for men. In the government sector, they're almost identical. Something strange happens when you look at finance. Professional women have a higher risk of mortality than non-professional women within finance.
Again, it's not really clear what's going on for women. It's hard without knowing something about households or husband's socio-economic status, it's really hard to know what's going on here. That's something to keep in mind especially when we're looking at women's health issues.
How is social position related to race and race/ethnic disparities in health. I'll talk a little bit about how these measures help us understand what's going on within the various race/ethnic groups. I have some data on all workers' income as a proportion.
Often we see data comparing within groups, comparing White women to White men, Black women to Black men, Hispanic women to Hispanic men and so on. A more interesting comparison is probably comparing everybody to White men. We should all be earning what White men are earning.
If we look at that, then we can see that, in fact, there is quite a bit of disparity across genders and ethnic groups. Within men alone, we can see that for every dollar that a White man earns, an Asian man will earn 92 cents, a Black man 66 cents and Hispanic man 62 cents.
Women earn far less. White women earn 52 cents for every dollar that a White man earns and Hispanic earn 37 cents. You can see there is a lot of disparity across the races. Here we have data for the proportion of workers earning poverty level wages. Poverty is an important consideration.
We can look at woman here. This graph is showing that a very high proportion of women earn a poverty level wage. If they were to go out in 1997 and get a job, somewhere over 50 percent of Hispanic women are earning poverty level wages. Somewhere over 30 percent of White woman are earning poverty level wages. Just over 40 percent of Black women are earning poverty level wages.
You can imagine that for men we have gradients as well. I want to talk about the use of education. That tends to be the socio-economic status factor that we use most in epidemiology. There are limitations to education when we're trying to understand racial disparities in health. Here we have data that comes from Oliver & Shapiro. They have stratified their sample by educational level.
You can see that even if we stratify the sample for Blacks and Whites by educational level, there are disparities in income. If we just adjust for education, we're not really accounting for these disparities here. The differences become more stalk when we consider something like the educational and net worth differentials or wealth differentials.
We can see that even in educational level, those who have an elementary school education, the net worth for the Blacks is $2,500. The net worth for Whites is $25,000. If we adjust our samples for education, we're missing all of these differences here.
Essentially, education alone is not enough when we're thinking about socio-economic status. If we're trying to deal with these issues and trying to account for racial inequalities in health, we do want to do as best a job as we can in completely characterizing socio-economic status. Education alone is not enough.
Income alone may not be enough. Wealth should be considered. You can ask a few questions about wealth. It is on some of the surveys related to retirement, dealing with older populations that do capture information on wealth. Using more than one indicator of socio-economic status at the individual level would be advantageous as well.
Because of the complexities of socio-economic status, we want to account for as much of the difference as we possibly can. It's also important to think about how the measures of socio-economic status are going to differ according to race/ethnic group, the meaning of the variables and expected association to health.
Although, in theory, we're adjusting for something that's independent of race, it's not. Race/ethnicity determine what our status is, in part, in terms of income or what our occupation may be. The patterns that we see of educational attainment or occupation or income distribution are, in part, determined by race.
That's something to consider as well. Ssocio-economic status is not independent of race/ethnicity. Another thing to think about even when we do adjust for the same factor, we may not be doing a proper adjustment. If we think about two people who have the same income, a Hispanic person and White person both earning $40,000 a year.
It may have cost the Hispanic person a lot more to get there than it did the person who is more privileged in this society. It's not clear what we're doing when we adjust for something like income if, in fact, the social and health costs were greater for the person of color.
The choice of measure that we use is going to depend on the health outcome under study. If you're looking at pregnancy outcomes like I tend to do a lot, you might want something that is very proximate to the time period of pregnancy, for example, income. If you're really looking at chronic diseases over lifetime, you might want something more stable like education.
If you're looking at an older population, perhaps wealth measures might be more important there as the cumulative measures of socio-economic status over a lifetime. Whenever we have surveys, we ask about what a person's status is at that moment, what their current income/wealth and occupation may be. That's not the whole story.
Here we have the way that has been adapted from the review that I was mentioning before. We have information over the whole life force from birth to old age. The idea that socio-economic status has an effect over the life is an important idea gaining more support. Some of you may have heard of Barker who says that cardiovascular disease in old is, in part, determined by health status at a younger age.
Perhaps there is a fetal origin to the disease so status at birth is important. When we're thinking about measuring socio-economic status even if we're interested in health issues related to adulthood and old age, we may want to collect information about socio-economic status of parents in childhood or around the time of birth.
This is very important to consider when we're looking racial inequalities of health. The trajectories and the socio-economic status trajectories for different race/ethnic groups are going to differ. Rather than just collecting data at a particular point in time for the health outcome interest and the socio-economic status that's most relevant for that time, we may want to think about collecting socio-economic status for more than one period in the lifetime.
There certainly are numerous individual levels, socio-economic status levels to choose from. Using more than one is highly recommended. Going beyond education and income is important for understanding issues of racial inequality in health. It's important to go beyond just the individual level and begin more routinely to include considerations of area-based measures of socio-economic status.
Although I've shown you data on economic aspects, there are other areas that impact upon health which if surveys allow themselves to capture this data, they would be able to bring into the equation as well to help us understand racial inequalities in health.
We also need more information and basic research on identifying the best measures for socio-economic status for understanding women's health issues as well as understanding various socio-economic status issues related to race/ethnicity.
Finally, if what we want to do is really try and account for socio-economic status across the races to try and understand more completely the issue of racial inequalities in health, we really should consider using multiple measures at the individual and area level. I'll take any questions that you might have.
DR. MAYS: Thank you.
DR. HANDLER: I have a couple of questions and maybe both are related. Fig. 8 that showed lung cancer mortality shows that as income increases, the mortality rate due to lung cancer goes down. I thought that lung cancer is related to smoking.
Maybe the people at the lower income level have a higher proportion of the population smoking and that's why lung cancer rates are higher for lower income people because a higher proportion of them are smoking. Maybe accident deaths would show a similar relationship because the people at the lower socio-economic status engage in riskier behavior, they don't look to the future. You have a couple of things happening simultaneously. I don't know if you totally can explain it by income. There are other things going on at the same time.
DR. O'CAMPO: I can't tell you exactly what proportion of the lung cancer data is attributable to smoking when we try to tease apart the contribution of health behaviors to the gradient in socio-economic status. That's one of the hypotheses. It all has to do with health behaviors, diet, exercise, all of those things.
In fact, there have been studies which have shown that health behaviors don't account fully for the gradient. I can't tell you what proportion of the lung cancer, in particular, gradient is due to health behaviors or smoking. I know that, in general, there have been studies that have looked at the health behavior issue. Even when you account for health behaviors, the gradient remains.
DR. POKRAS: The IOM committee that is currently providing some recommendations and guidance to the AHRQ for Health Disparities in the Healthcare report, one of the issue they wanted to discuss was should they consider an index that combines information from various socio-economic status measures or should they worry about multi-co-linearity when you include multiple measures in a multi-varied analysis? I'd like to see what you think about it.
DR. O'CAMPO: I think it depends on the purpose of the activity. I focused on in this presentation is an adjustment issue. If you want to adjust for socio-economic status, let's put in the kitchen sink. An index might be one issue especially if statistical power is a concern for your particular sample.
The purpose of adjustment is different from trying to identify the reasons for the particular adverse health outcome occurring. In the case of the former where we're just adjusting, I think using an index or multiple indicators even if there is multiple co-linearity shouldn't be a problem.
In theory, you just want to adjust for it. You just want to account for it. You might not be interested in the magnitude of the effect of any one indicator in a health outcome. There is multiple co-linearity, you can trust those estimates, but in some ways it doesn't matter. That's not what you're interested in.
If you're interested in how does social class relate to health outcome and what is the magnitude of effect for education and in common area level and all of that. Including an index will mix all the issues up and you won't be able to tease apart which issue or which indicator or which aspect of socio-economic status is most important.
If there is multiple co-linearity, your estimates will be off. You won't be able to investigate that particular question. If it's really for adjustment, it's less important and using an index is fine. That's my take on it.
DR. HERTIN-ROBERTS: I have a comment and question. Looking at the various dimensions of socio-economic status that you've given us, there have been income, education, and prestige based on occupational status. All of those strike me as very American, European, Western indicators of status and position.
Has there been any exploration of other factors that might be at least as important for some groups? I'm thinking that for some groups, socio-economic status there may be a person who has a very low level occupation that may be a community or religious leader within the community. T hat person would have quite a bit of prestige. Their status would be higher. I don't know whether things like this would make much difference in a huge survey or not, but has that been explored at all?
DR. O'CAMPO: I would ahve to agree with you that there are potentially many ways to capture issues especially related to prestige and power. They may differ across race/ethnic groups. We don't have a lot of information on that. That's an area that needs a lot more exploration qualitatively and quantitatively. I would have to agree with you in terms of sources of power and prestige.
DR. LERG: You mentioned the area-based measure information. I was curious since there is always place on a survey, what area-based measure information should be collected on a survey rather than being linked and brought in through other pools?
DR. O'CAMPO: The community psychologist would say that you should bring in a lot of perceptions data. There have been a number of studies where people are asked about their perceptions of whether their neighborhood is high crime and things like that, social cohesion, and capital items.
Those aren't related to socio-economic status or at least now how I'm characterizing it here. I don't know if any particular socio-economic status area-based measure can really be asked about directly in a survey.
DR. MAYS: When we think about this issue of money and different race/ethnic groups, the variable that talks about household income versus her own income, is there work that looks at how different groups deal with money. It brings into it some of the power relational, gender dynamic issues.
I'm not as clear about is you ask about household income, what does that mean for women in terms of their ability to utilize it? Do they benefit equally? Is marriage better for men than women and income has a lot to do with that?
DR. O'CAMPO: The measures that we're using are not capturing household and power dynamics within households. I don't know of any studies that have really tried to determine control over money and how much that's related to health. I would imagine it would be, but I don't know of any studies that have done that.
DR. SHUE: Ken Shue from the Center to Reduce Cancer Helath Disparities headed by Dr. Harold Freeman. It's important to put the whole issue of SES problem in perspective. We ahve rates on Blacks, Whites, and Asians. We have no national annual rates on the poor by any division of socio-economic status.
In some sense, cancer statistics which are a basis for disparities, don't even really get at these issues. Cancer Institute is now looking towards trying to correct that. They tried to look Census data once every 10 years, look at counties that are historically poor or a number of variables that you mentioned and collect their cancer incidence over time to see what has happened.
One of the studies is the lung cancer study. They've shown that the poor have low lung cancer rates and then they go up. The rich during the '70s had the high lung cancer rates and they go down. What you saw was a snapshot of the later data, but that is just the beginning to the understanding of the socio-economic status problem.
It's as fundamental an issue at the cancer data level. There are no ways of getting right now what the survey rates of the poor are. I just wanted to say we're at a fundamental important point turning the corner, but we have a lot of way to go.
DR. HANDLER: There another problem with using Census data to measure socio-economic status and that's migration. People migrate all time, 1 out of 5 people move from county to county. You can't pigeon hole people because there is so much mobility.
DR. O'CAMPO: There is some data that show that people tend to move to similar kinds of neighborhoods and areas. There shouldn't be too much misclassification if people do move.
DR. COLEMAN-MILLER: I'm interested in whether you have looked at the data that allows for or corrects for socio-economic status and co-morbidity and gender. For instance, the study done by Peter Bach in NY about lung cancer. That study talked about old records and new records in this tertiary care center and was able to correct for all of the socio-economic status factors and found that it still did not matter.
These men were not offered the surgery. It didn't matter whether they were rich, poor, insured. We just simply were not offered the surgery. I think his study validates what you're saying. At the same time, it really does make it so that for all of us, the socio-economic status in correction still doesn't matter and that disparity still sits loud and clear.
We have Marian Gornick with us who also did a study on the issue of amputation of limbs and found that the rate was 240 percent higher whether we were looking at the insured or uninsured patient. There is another element in here that we're missing. I'm wondering if you ahve a way to feather that out above and beyond your information.
DR. O'CAMPO: I do think that the most appropriate measure of socio-economic status is going to differ by health outcomes, but I think the example you gave is a good illustration of why having complete information on socio-economic status even if it doesn't matter is a good thing because then you've accounted for that and you know it's not that.
Now you have to look elsewhere. If you didn't have information or you just had something like education, you wouldn't know if it was other aspects of socio-economic status which might have been contributing to disparities or if you really need to look elsewhere. I still think the complete characterization of socio-economic status can help you in either case whether it matters or not.
DR. GIBBONS: The more I talk to people about this kind of thing, the more I find that we sometimes use the same term but don't mean the same things. We all think we know what we all mean by socio-economic status, we often mean very different things. You gave us plenty of nice ways to measure it, but you didn't define but what you mean by socio-economic status and how you distinguish that from social class if you do?
DR. O'CAMPO: I have actually chosen the term socio-economic position to avoid the status or class thing. In this country, we focus more on status vs a class notion of groups with opposing interests. I'm sure if we were all a bunch of sociologists, we would be arguing about those general terms that I have chosen to use.
It is important to distinguish between class and status measures and if you want to describe them both, you can use like socio-economic position that I've chosen to use.
DR. MAYS: Thank you very much. I'm going to ask our next two presenters to come up. I'll introduce our speakers while they're getting set up. That will get us on our way. Dr. Joyce Abma is a demographer at NCHS. She is in the reproductive statistics branch of NCHS and will be talking to us about the National Survey of Family Growth.
We also have for the behavioral risk factor surveillance system, Dr. Peter Mariolis who is a survey methodologist at CDC. He's with the National Center for Chronic Disease Prevention and Health Promotion the Division of Adult and Community Health. Welcome to both of you.
DR. ABMA: Good afternoon. Hopefully you have a hand out. It doesn't duplicate the slides. It's something to go away with that gives you a few facts of the most recent round of data collection that we've done, and the contact information of the agency, myself, and my staff's e-mail addresses and the Web site for the survey, and some illustrative publications. Feel free to peruse that at your leisure and follow up with any further questions that you have or items of interest later on.
I work on the National Survey of Family Growth. It's a periodic survey of the fertility of US women. I'll give you an idea of the background and the mission of the survey, talk about the measures of race/ethnicity, and a little bit about the sampling designs and hopefully get a chance to show some interesting findings that have come out of the most recent round of data collection.
This slide shows a framework that was developed by demographers in the '80s and it's been used and refined over about the past couple decades to model fertility. I show this because it guides the development of the content of our survey and it has from its inception. What you see there is called the approximate determinants of fertility.
The middle set of squares is the intermediate variables. These most closely affect whether a live birth occurs or not. In the top box, you have variables related to sexual intercourse, timing of first, the frequency, time spent in unions, marriage/cohabitation.
In the second box, you have conception variables that include contraceptive method used, sterilization and infertility. Finally, you have pregnancy outcome variables including miscarriage and stillbirth and adduced abortion. Those variable are all represented on the survey and have been since the beginning and most closely affect whether a live birth occurs or not.
On the left-most box, this is where the race/ethnic measure comes in. It's one of the social factors that are more distal determinants of fertility. The act on live births through the intermediate set of variables. If you have time to read down that list, it also includes things like religion, labor market participation, the standard socio-economic status measures we've been discussing, access to healthcare, more recently, the community environment.
Before I get into the sample size, let me comment that the data is relied upon as a reliable 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 have been five rounds of data collection starting in the early '70s. As you can see on the chart, the survey started out only interviewing ever-married women. In 1982, we began interviewing women of all marital statuses. You can also see that the sample sizes range in the 8,000 area up until Cycle 5 and we made a leap of sorts to almost 11,000.
What you have is a really decent time series for women of all marital statuses dating back to 1982 and for ever-married woman dating all the way back. Let me also mention that the upcoming Cycle 6 is planned for March 2002 and conclude at the end of this year.
The sample design has always been a nationally represented sample of women, the civilian, non-institutionalized population of reproductive age that is approximately 15-44. It's a probability sample, multi-stage, stratified cluster design. Cycles 1-3 were independent.
Cycle 4, the 1988 and Cycle 5, the '95 were linked to the National Health Interview Survey which is another NCHAS survey that you'll hear more about tomorrow. In Cycle 6 this year, we're going back to an independent sample. That linkage has implications you may already be thinking about for analyses.
In order to produce reliable statistics given these sample sizes, we've always included an oversample of Black women. Starting in Cycle 5, we included an oversample of Hispanic women as well. The '95 data is what the findings I show you are based.
It included 198 PSUs. The response rate was 79 percent. The response rate was approximately equal for White, Hispanic, and Black respondents. We ended up with about 6,500 White respondents, about 2,5000 Black women, and about 1,500 Hispanic women.
This is actually the question in the upcoming cycle. It may look familiar to you. It conforms to the OMB Directive 15. The first two items are about Hispanic origin. The third item is asking the respondent to self-identify and choose one or more group that best describes them. The fourth question there is unique to some surveys.
It's asking the respondent if they chose more than one race to choose one that best describes then. Finally, if you still don't have a race identified, the interviewer is asked to record race of respondent by observation. This gets recorded as Black, White, or other.
We end up with very little in-house imputation to do with race/ethnic identification. The other point is that we've always included a choose all that apply format. Dating back to 1973, it's been please select one or more groups. We also asked them to choose the race that best describes them if they're a multiple reporter.
The problem is that even with the larger sample size in Cycle 5, we only had 163 women reporting multiple races. It's not a whole lot to work with to do a lot of time series or sophisticated analysis of race choice. The other thing you may be thinking is we can't produce breakdowns of the smaller race/ethnic subcategory.
For some of the outcomes we look at, for teens, the Hispanic subsample unfortunately gets thin as well, but the strength that we bring to the data users is in terms of at least for the broad race/ethnic categories, we can monitor trends and differentials across a pretty broad range of outcomes in terms of women's reproductive health, pregnancy and childbearing.
In addition, since we do include a fairly rich array of potentially explanatory variables. Modeling of causal hypotheses is possible with the data and it's been used extensively to try to help understand the processes that are going on.
In addition to surveillance, its strength has lied in explanatory analyses. The NSFG is used by Healthy People 2000 and 2010. We supply data for 10 or so objectives mostly in the family planning chapter. The NSFG is funded by and supplies data for the Office of Population Affairs.
We show information about women using Title X family planning clinics. We're able to show characteristics of those women and the women who do not use the clinics who may be in need of such services. We provide information for CDC's HIV prevention program by showing estimates of sexual risk behaviors and some other purposes that will become clear in a minute.
This graph shows one of the classic purposes of the NSFG and that's to complement the vital birth statistics that are collected by NCHS. To arrive at a pregnancy estimate, you have to have a measure of live births, of fetal loss, and of induced abortion.
There are three different sources that go into that chart up there. The NSFG is the one that supplies the fetal loss data. You can see that there are some useful descriptive findings there about disparities between non-Hispanic White, non-Hispanic Black, and Hispanic women in terms of total pregnancy rates.
These are starting points for in-depth analysis because the graphs I've chosen here are two-way descriptive graphs just to give you an idea of overall uses of the data. Here we see the three outcomes I just mentioned, the three possible outcomes of pregnancy by race/ethnicity.
You can see that the shaded part of the pie is fetal loss and if you look at the non-Hispanic White pie, the 67 percent is live births. The middle pie for non-Hispanic Black women, live births is only 50 percent, and a larger proportion of induced abortions.
For Hispanic women, they resemble non-Hispanic White women. There are some pretty stark differences there. Teen pregnancy rates have been a topic of interest for as long as anyone can remember, but very much so currently in terms of social cost of teen pregnancies.
This graph shows that if you look at the left-most set of columns, you see the non-Hispanic White teen pregnancy rate is less than half that of the middle column, the non-Hispanic Black female teen pregnancy rate and less than half that for Hispanic teens.
If you look to the right-most set of bars, the NSFG allows you to further limit the denominator of that pregnancy rate to teens who were sexually active in the past year. It's a more pure population at risk that you're looking at the rate for.
That shows that among this population of teens that are sexually active, there are some contraceptive non-use going on there resulting in the race/ethnic disparities in teen pregnancies. This shows the benefits of having marital history in the data, but cohabitation history.
We asked each woman for the beginning and ending date of each cohabitation she's ever had. We were able to do this study where we looked at the probability that a cohabitation would result in a marriage versus break up. The top line represents the White females. You can see that their transition probabilities are higher at each duration of the cohabitation.
The bottom line is the non-Hispanic Black female line. Their transition to marriage probabilities are lower at each duration. This adds more nuance to that prior slide. It introduces a basic measure of socio-economic status, namely, family income.
You can see that for both Black and White females and particularly in the case of Black females, there is a positive association between family income in the past year and the probability that the cohabitation will end up as a marriage.
This is a basic slide. I just wanted to present it in order to introduce a study that was just done with the data. This shows the percent of singleton babies born between '90 and '93 who were ever breast fed, by race/ethnicity.
You can see there are some disparities in babies ever having been breast fed by the race/ethnicity of the mother with non-Hispanic Black babies, only 1/4 of them having ever been breast fed. There is a study that came out in PEDIATRICS last year and took the data and found out that race had independent effects after controlling for the socio-economic measures on the probability of a baby having been breast fed.
Additionally, they put in a model predicting infant marketability, they included every having been breast fed, low birth weight, and a really decent number of other possible predictors. They found that ever having been breast fed predicted infant mortality as strongly as low birth weight.
Their policy implication there was raising the level of breast feeding among Black women would reduce the gap in infant mortality. NSFG is a source of data on what we call unintended pregnancies. We mean that the mother reported for either came too early at the time of conception or she wanted no further births at all at the time of conception.
Racial and ethnic disparities and unintended births and you can see that non-Hispanic Whites, 27 percent of births that occurred between 1990 and 1994 were unintended compared to 51 percent among non-Hispanic Blacks, and 30 percent among Hispanic women.
We introduce a measure of education. As you can probably predict, the level of education is associated with percentages of births that are unwanted, particularly a pronounced relationship among Black females. You can see that 34 percent of births among Black females with less than a high school degree were unwanted, compared to 11 percent among those with a bachelor's.
This is a file that we have that we can link with the individual level data on characteristics of a community. This shows just one of those measures of the neighborhood, namely, the poverty rate. This shows that this measure of the community socio-economic status is pretty strongly and regularly associated and is by variant table with the percentage of births that were reported as unwanted.
The third set of bars to the right, those who live in low-poverty neighborhoods, only 4 percent of births were unwanted. Those who lived in the high poverty areas, 10 percent of births. I can answer any questions about the contextual file and analysis potential for that.
One thing to note is that contextual data were collected for characteristics of the community at three points in time. You're better able to match the time that your event happened with characteristics of the community at that time.
Cycle 6 is the next survey planned. The main new thing I want to draw your attention to there is that we'll be interviewing males in addition to females. The female survey will stay essentially the same. We'll add the other half of the information that we've been missing for some years. We'll be able to shed light on race/ethnic differentials in such things as father involvement in activities with their children, how well and completely do males report fertility, partnerships and how many partnerships have there been, male contraceptive behavior? What role do they play in preventing unintended pregnancies, male infertility, and STD/HIV risk and transmission. That's about all the time that I should take.
DR. MAYS: We'll have questions afterwards.
DR. MARIOLIS: BRFSS is a shoestring operation, a joint venture of the CDC and health departments in the 50 states, District of Columbia, Guam, Puerto Rico, and the Virgin Islands. This year, for the first time, the Virgin Islands are conducting surveillance so we have 54 surveillance projects.
It's a joint venture which means that the states have a lot of control as well as ownership over the data. Just about any statement you want to make about BRFSS is usually an exception or two. It's 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.
The CDC coordinates the development of an annual set of core questions which are asked by every state. The questions change from year to year, but each year every state is required to ask each of the core questions. Then we have standardized sets of questions on specific topics that we call modules which the states can pick and choose from.
In addition, each state is free to ask any additional questions that it chooses. What we do is try to set standards and monitor the standards for sample design, data collection procedures, and technical support to the projects. Sampling consists of all telephone number of NXS types.
The essential point about the sampling frame is that it does include all telephone numbers which could ring into households. That's not true of any telephone service. The design does have to be RDD in each state. Our data collection guidelines are that there should be 15 callbacks as long as the number remains unresolved distributed over weekday, weeknight, and weekend calling occasions.
One eligible adult who is randomly chosen from each household. We do not allow proxy interviews. States are responsible for data collection. Each state chooses its own data collection or collects it in-house. Once the data are collected and initially edited at the states with a program we provide to them, they're sent to us.
We conduct further editing and at the end of the year, we weight the data and return to the states along with several reports. The states are then able to generate their own reports and studies and we do the same. We make the aggregate data file available to the public.
In 1999, we had almost 160,000 completed interviews, in 2000, 184,450 and in 2001, we expect that we'll have somewhere around 204,000. This is a list to give you an idea of the variety of topics covered in the BRFSS. On the left-hand side are the core for the 2002 questionnaire.
There are a few copies of the questionnaire around. You can get it off our Web site. Even in the core, there are a large variety of topics. That means that no one topic is gone into very deeply, but we do cover a lot of different topics. The option module states may have anywhere from zero. Here is our Web site. The questions and data are available to download. Documentation is also available. My e-mail address is there in case you want to e-mail me about anything.
I am going to look at two questions here. One is alternative ways of measuring race among Hispanics, but how Hispanic respond to the race measure we have. Also, the coding and reviewing and of multi-racial responses. I believe that race is problematic to many Hispanics.
After I wrote that, I spoke to my wife and two children in their 20s about this. I would change this a little bit. I think the race of Hispanics is problematic to non-Hispanics as well. We talk about race. It's a combination of biology and culture.
Hispanic origin is an ethnic category. We make a hard and sharp distinction between race/ethnicity. Many people out there don't and that influences who they answer the question. We allow other race category is an acceptable response. One indicator of the difficulty of the race of Hispanics is that they disproportionately tend to give an answer that is coded as other.
These are the questions that we ask: Are you Hispanic or Latino? Which of the following would you say is your race? Which one or more would you say is your race? We include other on this. 2001 was the first year we allowed people to give more than one race in their answer. For that, we only had one.
We also had if they give more than one race, we say which of these groups best represents your race? Other, is an acceptable category in that question as well. Here are some data from 2002 when people were only allowed to give one answer to the race question.
Hispanic and non-Hispanic. In the bottom cells, you see that 31 percent of the Hispanics were coded as other race. They gave an answer that didn't fit one of the five standard race categories compared to 0.73 percent, less than 1 percent of non-Hispanics.
I've come to think of other as a missing value. You really don't know what it means other than it was not one of the standard race categories. You can see that a lot of the other by the Census Bureau would be coded as White because there are 89 percent Whites. That's not what people say when you ask them.
One of the things I hoped would be positive payoff allowing people to give multiple answers to the race question is that percent other for Hispanics would go down. I had the hope that one possibility was that other meant that people saw themselves as multi-racial, but we had no category for multi-racial.
Our 2001 data show that that's really not the case. In fact, just the opposite happened. In 2000, we had about 30 percent of Hispanics giving other race answer. In 2001, in partial preliminary data, in 50 states plus DC excluding California and Hawaii.
I excluded Hawaii because not only do they have very different racial distribution, but as part of that, they tend to have a fair number of Hispanic Asians which I interpret at Philippines. I thought those Hispanics were pretty different than other Hispanics.
I excluded California because it's one of the exceptions here. They impute race before we get the race data. They really have almost no respondents coded other race or missing. Here we are in 2001, 41 percent of Hispanic gave an other race answer.
Slightly more and the number of non-Hispanics didn't go down. That was almost 1 percent also. The other thing is that this table represents the answers to the question and if people gave more than one race, which one represents your race?
People who originally gave multi answers and then chose a single race coded according to the single race they chose here. These are the same data with multi-racial respondents coded as multi-race. We still have 40 percent of Hispanics coded as other race.
The other thing to note is that really very few people give a multi-racial response. It's almost 3 percent among Hispanics. Among non-Hispanics, it's 1.34 percent. The next question I want to address briefly is among the people who used a multi-racial, for non-Hispanics of those, what distribution in the column are race coded according to people who gave a preference.
Then in the rows are race with multi-racial response coded as multi-racial. The Whites are the people who gave White as their only race or their best race, 99.9 percent gave only White. Only .91 percent were other racial.
If you look at the bottom row, you can see that the multi-racial responses disproportionately tend to come from American Indian/Alaskan Native, Native Hawaiians and a little bit from the other category. For Hispanics, the distribution is similar except that Black Hispanics also tend to be a little high on the percentage giving a multi-racial response.
I have a set of recommendations which most people follow, but not everyone does. Probably everyone here knows enough to do this. Always present results by race/ethnicity with Hispanic as a category never by race alone. Second, never break Hispanics out by race.
You never say never or always. Take it that way. Finally, on the effects of allowing multi-racial choices, there is a minimal direct impact. Only 1.34 percent of non-Hispanics chose multiple races. Only .1 of 1 percent would choose a single race.
It impacts Native Hawaiian/Pacific Islander, American Indian/Alaskan Native, and Hispanic, Blacks the most. There was a significant indirect impact on Hispanics. It increased the percentage choosing other race 10-11 percentage points.
Choosing other race is only to a small extent an alternative to multi-race choices. People who choose other race are predominantly not multi-racial or don't give a multi-racial response to race.
DR. MAYS: Thank you. Let's open it up for questions.
DR. HANDLER: I work for the Indian Health Service and before that, I was a welfare case worker in Spanish Harlem in NYC and spoke to people in their home, interviewed them in Spanish. I could have told you in 1965 that I'm a White person and they're a Hispanic person. It's nothing new. It's been decades that way. We're just finding out, but that's the way they were thinking way back when.
The question I have for Joyce is when you start asking the fathers of the newborns questions, are you going to be asking the race of the father? You didn't list that as one of the items.
DR. ABMA: I went over that too quickly. They're not linked to the females in any way. Some of them may have newborns and others won't. It's just based on how the population is represented that way. They like the females will be asked for their self-identified race with the exact question that you saw there.
DR. COLEMAN-MILLER: I have a question for you about the fertility question. If fertility is your question, then rather than ask about unwanted babies, do you have the opportunity to ask about wanted babies? I know that the socio-economic status for minority women still has a higher infant mortality rate than for the majority female. When you talk about unwanted, it's the opposite of asset mapping. I'm wondering whether you do it also for wanted babies.
DR. ABMA: That's a good point. It's a conglomeration of a bunch of different questions and responses that ends up resulting in the measure of unintendedness that you saw. Based on whether they were using a contraceptive method at the time the pregnancy occured or not, that determines their routing through these questions.
The question ends up being at the time you became question, did you say that you wanted a child at that time, a later time? There is a timing question. That's preceded by the question about when she had the child or at the time of conception, did she want any children in the future at all.
It's a measure that never really uses the term intended or wanted. You can get at the flip side of the coin such as percentages of births that were wanted and timed correctly according to the woman.
DR. COLEMAN-MILLER: Based on the disparity issue looking at the wanted pregnancies that were not live births would be of great concern especially in relation to your unwanted ones. Is there any disparity in the fertility issue? That would be another consideration. There are young people coming to clinics at 16 and 17 year olds saying they can't get pregnant. Is there a fertility disparity?
DR. ABMA: You're referring to reproduction?
DR. COLEMAN-MILLER: From 15-44 is there a fertility?
DR. ABMA: That's not my particular area of expertise, but I don't think there is a significant racial difference in impaired fecundity which is suboptimal fertility or in infertility.
DR. COLEMAN-MILLER: In studying the fertility question, the issue of whether there is a disparity has not been directly addressed?
DR. ABMA: Not by me. If you go to our Web site, you'll find articles by my colleagues and other outside NCHS that have really mined that data in depth and profiled women with impaired fecundity and infertility. There are race/ethnic differences with regard to sterilizing operations and some interesting findings there too.
DR. COLEMAN-MILLER: The other question has to do with your behavior list survey. That's an important survey. It has a lot of potential. I remember that race is problematic for Hispanics. Please know that it's also problematic for the Black population and that I'm not positive about race being biological now with on the new data on gene mapping.
The question of whether race is biologic rather than just cultural which someone just chose to do is really an argumentive one at this point. Saying it's biological is a difficult statement for us to deal with at this point.
DR. KREMGOLD: Barbara Kremgold, Center for the Advancement of Health. In looking at the impact of socio-economic status and early childhood development on health across the life span, it's long concerned me that we have over 40 percent poverty rates for Black and Hispanic children.
When I see your data on the probability of marriage, it's strongly related to income. My question is whether the data that you're presenting has been shown jointly to the part of HHS that deals with welfare reform and an interagency committee with the Department of Labor. It would suggest that the jobs for the father are important in whether marriage in the rates in the probability of marriage.
DR. ABMA: We have a paper that was presented at the AEI seminar on the effects of welfare reform. We have a publications list at the Web site that's in your handout and you can go to the paper there. It has been used for that purpose. It's relevant in the information about cohabitation about marriage and child well-being.
On the one hand, children born out of wedlock, a large proportion of them could be in the union that just isn't a legal marriage, but then if you look at the instability of cohabiting unions, that could have a negative impact on any children within those unions.
DR. KREMGOLD: I was also suggesting as the welfare reform legislation that the jobs programs for minority males would be something that could be discussed jointly in this context.
DR. HERTIN-ROBERTS: I'm puzzled by Dr. Mariolis' statement never break Hispanics out by race. The implication is that because they will not put themselves in the category that we think they should go into. It seems to me that the fact that Hispanics keep identifying themselves as other is data in itself and should be reincorporated into the thinking of how the survey is directed. Rather than saying we're not going to present this information, maybe we need to rethink how we're questioning and considering Hispanics.
DR. MARIOLIS: I wouldn't disagree with that to this extent that you could come up with studies where you would want to look at this issue or make an argument about the race of Hispanics that would make the kind of responses we're seeing applicable to that.
The context that I was coming from is a state health department report that is looking at prevalence of smoking by race or race/ethnicity. There, you do expect the categories to match what's in the state if someone has census data on those. That's a constraint because to the extent that they don't match, then people come back and say what's wrong with your data?
The racial distribution in the BRFSS of Hispanics is not going to come near the racial distribution of Hispanics in the state based on Census data. In addition, if you mix the Hispanics with non-Hispanics and just present straight race, that will also have an effect. That's the context. There are situations where you would want to look at this.
DR. HERTIN-ROBERTS: What concerns me is that what we have is the reality of the person self-identifying in conflict with our surveys. If our surveys are not measuring the reality of self-identification, maybe they need to be adjusted.
DR. MARIOLIS: I fully agree with that. I'm dealing with the data we have now.
DR. CARTER-FOSTER: I just wanted to clarify a few of the issues that came up. There are Healthy People 2010 objectives that relate to infertility and there are differences by race/ethnicity. In regards to the discussion about what do these racial differences mean, there is going to be a session of the secretary's advisory committee on genetic testing Thursday morning at 10am where they're going to have a panel of everyone from geneticists to anthropologists to biologists to talk about whether race/ethnicity data should be collected in the arena of genetic testing and screening. This may be something that you want to pay attention to.
In regards to your hypothesis about what the impact will be of the use of multi-racial categories on the selection of Hispanics using other race. This had been well researched several years ago by the Commerce Department as well as Labor.
When they took the current pilot study they did for the Census, all that research is well documented and summarized as well as references to many other articles on these issues. That's available on the Census Web site. If you look at the alphabet and you go to R, you hit race, it will hot links to these reports as well as similar reports.
The 40 percent is actually something that's been found for decades and seems to be coming up again with the Census data right now. We were hoping that having Hispanic origin go first, would have made an impact on the percentage of Hispanics who select other race. It didn't have as great an impact as we were hoping from these previous test results.
DR. PARK: I have a single track mind about the local data. I called the State of Maryland looking for local level data on behavior risk factor survey and they refused to give it to me. You have states doing the surveys, but isn't it possible that someone could give us the local level data on behavior risk factor survey? The youth BRFSS is done separately and it's done metrowide setting. Can we do the same type of things with the BRFSS?
DR. MARIOLIS: Metrowide?
DR. PARK: Yes, like Baltimore region, Washington region.
DR. MARIOLIS: If you have a high speed Internet connection, you can download the BRFSS data from our Web site. E-mail me. The dataset contains county information with more than 50 respondents. The confidentiality agreement allows us to send you data that includes county for all respondents.
This is a joint venture. There is a surveillance MMWR article coming out very shortly that looks at BRFSS by metro area. If you e-mail me, I'll put you in touch with the person who wrote that.
DR. HITCHCOCK: Joyce, you were talking about race/ethnic data and we were happy to see that went through for a pilot test. Is this the first time? That's really neat. It will be able to address that question.
DR. ABMA: I don't think the wording uses sexual identity. We use that term as a construct. It's within the self-administered portion of the questionnaire along with more of the potentially sensitive questions on the survey. Hopefully, that will promote more full disclosure.
DR. HITCHCOCK: It worked well in the pilot?
DR. ABMA: It did. We didn't have any problems with it and we're optimistic about that.
DR. MAYS: Some of these are reports that forthcoming or out, when you do the presentation on cohabitation and race/ethnicity. Is there any data to talk about what disrupts that? We know that cohabitation differs by gender. Women often believe that living together is going to lead to marriage more than men.
The other thing we know is that for some ethnic groups depending upon the age, the disruptions are not that there won't be a marriage, but the disruptions have to do with things like prison, financial problems and things that help to explain this.
If what we're looking at are health disparities here, can you talk a little bit about in this data, what's the context that will help us to intervene? Do you have acculturation or any of the other variables?
There are very sensitive issues around pregnancy. It's like the difference between the issue of wanted pregnancies vs unwanted pregnancies and what that means in terms of where we intervene. Is it the bad or good people?
DR. ABMA: The whole concept of unintended pregnancy was a way to measure unmet need for family planning services. That originated with the first cycles. What is the difference between the total fertility that women intended and thought was ideal vs the total fertility they received and what accounted for the difference between those two and can we get at percentages of population that are experiencing barriers to family planning services?
That's where the concept of how do we best measure the planning status or attention status of pregnancies evolved. The Cycle 5 is really rich. We have measures that start to get at what you're mentioning. It may not be a good source of analysis for what causes union break ups and the nuances around what leads to an unintended birth and what does that mean and does that differ for people from different backgrounds>
In Cycle 5 with cohabitation, we have pairs of separation within cohabitation. It would allow you to look at separations that were followed by reunions. The same is true of marriages. We have separations with marriage that did not result in divorce. It's not going to get at the in-depth answers that you applied, but it's a springboard for ethnographic or further studies to investigate the patterns we do see.
DR. MAYS: If my focus is on health disparities, and I take the probability, the question would be what do I want to do and what do I want to fix? We have a factor about racial differences and cohabitation, but I don't know what to do next unless what we're saying in the survey is I put that out that some groups don't make it to marriage equal to others.
It's up to somebody else to figure out why. I'm trying to get a sense of some surveys are able to do things really well and others not. There is a message that's here and I'm trying to figure out what's the best source. Do I link your dataset with another? Do I require some cultural competence training of the people doing the data analysis so they can push it as far as possible? Can you link with any other datasets that can give me more contextual information?
DR. ABMA: We do link with the National Health Interview Survey with the Cycle 5 data. You're going to get access to a vast variety of health measures, but perhaps I'm not sure that will get into the areas that you're talking about, but it would still add to the things that you could add to your model to help explain the outcome you're interested in.
The contextual measures refer to the file that we link to the respondent information that contains a wide variety of measures of the characteristics of the neighborhood. Just so you're aware and I'll leave it to you to evaluate whether this helps or not. There are hundreds of different characteristics of the neighborhood 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 unemployment.
There is some potential there, but the survey is valuable for causal modeling of the type that I've described in the breast feeding study. I'm not sure the it applies to the issue that you're getting at.
DR. HERTIN-ROBERTS: If all we're left with is this chart, the question is where are the partners, where are the men? What we know about mortality rates among African-American men, they're dead between disease and violence, they're dying at a much earlier age. It's very difficult to have cohabitation lead to marriage.
I don't know whether the dataset can be linked to other datasets, but I think that this is a particular example of how the data that we have is limited in its interpretation without some other contextual information to be given.
DR. ABMA: You were talking about that. We do have information that will shed light on the reasons for separations within cohabitation and marriages. For the first time in the upcoming cycle, we are asking if the person has self-identified themselves as married or cohabiting and the partner is absent from the household, we have a pretty detailed question asking them to say where they are. This is asked of both men and women. That gets at it a little bit. There may be other reasons that may prohibit union formation or continuation of union.
DR. MAYS: The message is the putting that out in the reports, it's like the people on this side of the table do that, but it's like making sure in terms of the reports being put out having that view of what would be helpful outside in terms of pushing.
To some extent, the numbers and the pathology are there, but it's also having that perspective on what the resiliency is and what to build on in terms of the translational research Dr. Clancy was talking about. Then, how do we move it without having people come in and actually do the analysis themselves.
DR. HANDLER: This was true in the mid-1960s and I guess it's still true today. In Puerto Rico, common law marriage was a legitimate form of a marriage. In NYC, the position was taken that if a marriage of common law variety was begun in an area or jurisdiction where it's a legitimate form or marriage, then NYC recognized it as a form of marriage as long as the people stayed together for seven years. I don't know if the same type of mindset is true today. Is it part of the structure of your survey to consider common law marriage started elsewhere as a legitimate form of marriage?
DR. ABMA: In the items that classify people into marital status, we specify not to include common law marriages.
DR. HANDLER: It does not compute, right>?
DR. ABMA: At least you know what you're looking at. you also have a cohabitation history. You could look at the duration of the union and see if it met that criteria.
DR. HANDLER: The people themselves consider themselves married. There's a break there. They consider themselves married, but you don't consider them married.
DR. ABMA: No.
DR. GIBBONS: I have one comment to the Council as they try to provide leadership and direction to other HHS agencies. This illustrates a point that is going to get more problematic in the future. In this question of race versus ethnicity, they're not the same.
One resource that may be helpful in grappling with this is a paper written by Thomas Levese at Johns Hopkins "Why Should We Continue Measuring Race, But Do a Better Job." The idea is that if we did our health services research along other variables the way we commonly do it among race, it would never be published. It would be shoddy work.
We do some research and then we say we'll control for race. It's not something fundamentally that answers anything. It's telling us about something or poverty is the same thing. Once you control for poverty, what have you done, learned, and know?
You have to go much deeper than that. A lot of these surveys that have Black and White, if you look at the connotation of race in this country, it has largely been political in the broadest sense. As we've moved to a more diverse society, that wasn't good back then, but it broke down when we had Hispanics and others coming in.
When you give a question like that to an African-American, they understand it. My argument would be rather than saying it's unimportant in other populations, a Black person understands that's a race question. I will not say I have European descent or consider myself to do that.
A Hispanic will understand that as an ethnic question. My ethnicity can come from several places. I would think that if you did the same thing Afro-Caribbean or African-American or African, you would get the same kind of disparity. It would behoove the Council to spend time to figure out what is it we're really trying to get by this race/ethnicity data. They're not really giving us the same thing and it's going to be more problematic in the future.
DR. MAYS: Because the hour is late, I'm going to keep my remarks short. It's clear that the surveys that we've heard from are critical to the determination of things such as our target for Health People 2010, resource allocation, and what it is that the public often hears first about a slice of life.
To that extent, it behooves us to really pay attention to the power of these surveys and whether they are addressing what it is that Health People 2010 is designed to do or if they're addressing the actual mission of the some of the offices that are the ones that receive guidance from these surveys.
It seemed like at some points, which was in the lead. We've just gone through a round of guidance around the asking of the question on race/ethnicity. Indeed, what we know is that we've made a lot of progress and we know that our colleagues who are working on these surveys are pushing ahead.
If I were to take our last question, it's really what do we need to know, who do we need to know it for, and how soon do we need to do it? 2010 is going to be upon us. There's a mid-point review that's around the corner. It takes so long for us to get data and to be able to analyze it.
We were getting presentations on data from 1996-1998. We got a little bit in 2000. We're going to be at 2004-2205 before we've actually been able to mine the most recent data that we have in which we see some changes.
You've left this committee with lots of things to think about and lots of challenges, but in representing the committee, what I want to do is share those challenges with you. We have our piece to do, but many of you belong to organizations, write articles, and stand in positions where your voice is stronger than some of ours.
This is a dialogue that we will continue, but hope to continue it as we get input from organizations and individuals and the literature. I hope today has been about the sharing from the individual users and the things that you think are the priorities for this data.
This is federal data. As I like to say, this is your tax dollars at work. You have some ownership of this data. I really appreciate all who came out today to share your voices in helping us to really look at these issues. Thank you very much. We reconvene tomorrow at 8:30 and we have another set of surveys and we also will have Dr. Kington who will join us and a presentation on multiple race. Please join us tomorrow. Thank you.
(The meeting was adjourned at 5:30 pm.)