Ninth Annual Maternal and Child Health Epidemiology Conference / December 10-12, 2003
DR. WILLIAM JOHNSON: There are a number of slides in this presentation that I’m not going to discuss specifically. They were purposely included to provide you with some backgrounds primarily in the form of descriptive statistics. I’m going to focus on some characteristics of the Yuma County health data system and also on some results from all variant models, which I think are fundamentally more interesting. I’d like to recognize my colleague, whom you met this morning during the plenary session, Dr. Mary Rimsa, who’s sitting over here, who’s the pediatrician who works with me on this project and for those of you who are discussing fetal death presentation this morning if you didn’t hear it, had to do with pediatric death more generally in Arizona, and it’s a topic that we do plan to pursue as one of the elements in the research agenda for the community health data system that I’m going to discuss now. Concept of a community health system is a voluntary community effort that collects information from all willing providers in a community, including insurers, state agencies and employers, and then the data from each data partner is merged creating a data set with an individual record for each child and tracking that child over time, including all the information that’s available.
Obviously, there are HIPPA requirements to maintain confidentiality and encrypt the data and quite a bit of work on our part has gone into that, including crafting legal contracts with each data partner, which are HIPPA compliant. As of the moment, which is more or less the fourth year of the project, we have approximately 63,000 children in the human data set with data for three plus years and it’s continuing. The financial sponsors of the project are the Flynn foundation, which is a local Arizona foundation, and believe it or not, Arizona State University. We think the advantages of this approach to data collection or in part that it provides information when a question is asked rather than waiting until the question is asked and then going out and designing the data collection protocol to acquire the information that you need, depending on how often you want to update the system, the time lags are relatively short and from an economic standpoint, this form of data collection is very cost effective because in fact the cost of acquisition have already been paid. A zero cost of acquisition is about as low as the price gets and the only real investment you have to make is the cost in initially developing the protocols to merge the data and the actual process of maintaining the data set.
One of the unique advantages of the data set, which we’ve just begun to realize is the fact that unlike most of the available sets it actually measures changes in *cohort over time. The original rational for the creation of this data set came out of an evaluation that we did several years ago of prenatal care in Arizona, using administrative data the Flynn Foundation saw what we had done and they were very concerned initially with insurance coverage for children and asked me if a I could do something similar to what we had done on health start, but on a community-wide basis and track insurance coverage. As you can see, we do track insurance coverage, accessed by the way, is the nonomic for the manage care medicate system in Arizona, and in these particular data it includes the S-chip kits. This shows you the distribution by ethnic group and it points out something that’s rather unique about this data set. Almost 70 percent of the children in Yuma County, which is in the southwest corner of the state and borders Mexico and California, are Latino. We also have a significant, but small population of Native-American children from a tribe that lives in the Yuma County area.
It’s given us the opportunity to address questions about children in these ethnic groups that have not been answerable from other data sets because as the recent task force report in JAMA pointed out on Latino children specifically, in most cases the numbers of children in these ethnic groups are so small that the researchers are frequently forced to combine Latino, Native-American and other smaller ethnic groups in the United States into one set. In the year 2001, taking just one year of the data, you can see the kind of classifications we can make in terms of utilization of health care. Both separation into traumatic and non-traumatic, insured and not insured, E.D. and non E.D. and so forth, and we have so many children in the data set, and remember this is only one year out of multiple years that we have the opportunity to do that. Just expressing it in it a different way a vin diagram with the over lapse and you can look at that at your leisure. One of the conventional stories and it really relates very clearly to the preceding talk, which I thought was very interesting, is that there is a lot of literature that tends to attribute differences among ethnic groups to cultural differences, attitudes and so forth and so on, and comes to some conclusions which actually may or may not be true, but which are sometimes difficult to test because of the lack other information.
The conventional wisdom on Yuma County before we began this study was approximately 25 percent of the children in Yuma county were uninsured. That was stated in a number places and very publicly discussed in Arizona as a major problem and the reason for that was people looking at national statistics and saying what are the characteristics of uninsured children, they tend to be Latino, they tend to be poor, et cetera, et cetera. Most of the Latino children in Yuma County are relatively poor. This is an area whose primary industry is agriculture. There are a lot of migrant workers. The town of San Luis, which is just outside of the town of Yuma, but within the county, for example, is one of those border towns where you walk down the street and if you don’t stop walking you’re in Mexico, and therefore, there is a great deal of cross border migration that goes on and there are many people who are citizens of Mexico who work in Yuma County and go back and forth across the border every day to work in the agricultural organizations. These are some of the descriptive things you can look at. I want to talk about the multivariate models. I’m sure many of you are familiar with this model. The variables in that top row are the ones that we’ve included in the model. We don’t have socio economic status for example because it’s not available for the most part from the information that’s contained in the records of the health care providers although we do have information that would provide some general kind of socio economic characteristics.
We are doing geocoding and looking at different neighborhoods and census dated to try to get some proxies for that. We haven’t done any surveys in Yuma and therefore, we don’t know a lot about health care beliefs, knowledge, education and so forth. We do have quite a bit of information on basis demographics, health insurance, residence, medical conditions and although this has chronic medical conditions, we actually have quite a adequate description of people’s health profiles and in many cases we now have data on cohorts of children where we data for each of three years. The first study that we did after the descriptive reports on Yuma was done by Mary and I and we wanted to look at the use of the emergency department for non-urgent care, not exclusively for Latino children but including those factors in the multivariate models. These are standard multivariate logistic regression models. It turned out that it was quite true as the convention of wisdom suggests that the proportion of children who used the emergency department is much higher if their uninsured. It was not true, however, that most of the children who used the emergency department are uninsured. That has two very different implications in terms of the burden of care on emergency departments and it seems very simple when you think about, but in going back and looking at the literature on emergency departments, it’s not a point that’s often made.
One thing that was more interesting to us was the fact that since we have information on the use of pediatric care, we get the data from the largest pediatric group practices in town in Yuma that children who had any access to pediatric care, all else equal, were 73 percent less likely to use the E.D. if they insured and 93 percent less likely if they were uninsured and that was sort of a very dramatic result for in some cases relatively few visits to a pediatrician. The fact that adolescents are the most likely age group, all else equal, to use the emergency department, I doubt is surprising to anybody, are age groups truncate at age 19. If you control for insurance coverage and need for care and so forth, it turned out that again, contradicting some of the conventional wisdom, the Latino children were no more likely to use the emergency department then the children in the other ethnic groups.
The next stage we went to, was to look at children who relied almost entirely and I guess in this paper it’s literally children whose only source of care for non-urgent care was the emergency department because we have such large data sets we were able to separate out children who receive care in the emergency department for emergent or traumatic conditions and look solely at the children who use the E.D. and in this case who only use the E.D. as their only source of primary care. The ethnic that were most likely in terms to the odd ratios to use the E.D. as their only source of primary care were the Native-American children. Relatively small number of children, so this is one of these rate questions, rather than a burden question. Latino children, again, were 70 percent in this case less likely to rely on the emergency department if they were uninsured, so they didn’t use the emergency department as much as the other ethnic groups even if they were uninsured as their source of primary care. Among the insured children there were no significant differences between the Latino children and the reference group, which were the white non-latino children. Those are two things from the cross sectional data.
Now, we’ve begun looking at the longitudinal data and we are continuing data. We’ll soon have 2002 and most of 2003, but as you can see here we have a cohort of almost 14,000 children, we have data on each of three years for those children and you can see how this distributes out among the others. This provides some very interesting opportunities, the first of which we’re trying to take advantage of is to look at patterns of insurance coverage over time. Those of you who study health insurance, I’m sure are aware of the fairly well known differences with different questions on the CPS about how you ask people about health insurance, whether it’s an immediate question or for the past year and so forth and people recognize that health insurance coverage is not necessarily a continuous state in time, but for many people involves a lot of variation over time. This is for the period 1999 to 2001 and this is just for the Latino children and again we see that 89 percent of the children are insured throughout this time period. In other words, there is no recorded spell. No time at which they received any form of care, in which they were not insured at the time that they received that care.
There is a tiny fraction of the children who were uninsured throughout this period of time and here these are very conservative measures, because what we call uninsured is any spell of uninsurance in general, and so that continuously uninsured actually means any child who had any spell of uninsurance in any of time periods is counted as if they were never insured. The importance of that is really to begin to look at the impact on health care. What does it mean when they have interrupted spells of health care coverage and also from a policy standpoint, Arizona has a policy of requiring reestablishment of eligibility for medic-aid at various time periods, presumably the legislature has required that in order to try to minimize cost, but actually if you look at some of the periods of interruption and what we hear from people in Yuma and other places there is the potential for huge administrative cost that serve very little purpose by requiring this continuing reestablishment of eligibility. These are simply some more descriptive data, as I said I won’t focus on that, but you’re welcome to look at that. In terms of where we’re going from here.
We have a current project in cooperation with the fluid dynamic section of the engineering school at Arizona State University where we’re looking at agricultural burning on the one hand for which they have data, which is very interesting. They actually have real-time maps showing the plumes of particulates going over various geographic locations in Yuma mostly in San Luis (inaudible) and of course, we the incidents of respiratory conditions among children in those areas and we’re mapping and comparing those two sets of events. More generally, the prevalence and treatment of asthma that will be focused on the access data because the access data is the only data set that has complete information on the use of pharmaceuticals. As I mentioned the pattern of insurance, and the effects of changing insurance coverage on axis over care.
The new project known as MHIP is financed by St. Lew’s health initiatives in Phoenix and also partially by Arizona University, so much more ambitious community health data system. This everybody not just children and this Maricopa County, which as you’ve been told has about four million people and the fifth largest city in the United States. At the moment, we have begun to receive data from Medic-aid and from all but one of the major hospital systems in Maricopa County. We’re not quite sure how big this data set is going to get. Although, we just ordered some new servers. So we think we know how big this data sets going to get. One of the challenges in Maricopa County is to get the outpatient data, but this is multi-year project that’s going to take a while, but it very quickly is developing a database that’s available to study a number of the questions including the one’s we’ve studied in Yuma, Maricopa integrated health care system is one the largest county operated hospital systems in the United States, serves primarily a low-income population and we already know that we have for example several thousand Native-American patients on record in Maricopa and the Maricopa system.
If people would like copies of these papers or additional information that’s included, you can either write to me or to Mary or send us Emails at these addresses, and I’d like to thank the previous speakers. I’ve given a lot of talks in a lot places and I always have the horror of being the last speaker because I’ve heard too many times about how people would really appreciate the fact if I would cut my talk in half because there isn’t any time left. So I would like to publicly acknowledge their courtesy and their efficiency and I’ll also quit on time. Thank you.