Ninth Annual Maternal and Child Health Epidemiology Conference / December 10-12, 2003

What Does the New Data on Children with Special Health Care Needs Tell Us?

GINNY SHARP:  What I was asked to do in this presentation is to do sort of a primmer on how you go about as a State if you want to analyze the data yourself instead of waiting for the data center, or if there are things you want to know about your State's CSHCN population that maybe you won't be able to get out of the data center.  And I will provide a few examples from what we're calling the CSHCN road show that one of our State Department of Health CSHCN program staff and I have been going around the State making presentations and trying to get input from our local CSHCN coordinators, public health nurses and others as to what questions do they have that we might be able to answer from this survey, or can we draw questions from this survey and use them in local areas, as Christi was suggesting and we've already started that in one area.  And also to discuss a few of the analytic issues associated with working with these data at the State level.  Five steps to getting started.  First, go to the website.  Go to the CSHCN survey website, familiarize yourself with the survey instrument and methodology report, download the data files, subset the National files to your own State's data, and then you have to do some merging to make them useful.  Here's what the opening page of the CSHCN survey website looks like.  And if you page down on it you get to a section that says, "View Download" and that's where the documentation is, the full survey instrument, the methodology report and the data sets, and also the variable listings.  Here is the survey bible written by Dr. Blumberg.  It is an incredibly well written document and my biggest fault is I forget to check it.  When I find something that doesn't make sense, then I'll go, "Stephen, what's going on with this variable?" and it'll be page 13, or page 27, and it's all right there.  So first thing, if you have a question, look to this document. 

The whole document is on the website and it's in a PDF so you can just pop it, just download it onto your drive, open it up and search using the find.  You can search for any key word with the find on your Adobe Acrobat.  The files are in a FTP directory and you'll notice there are four separate ones.  And these are all the National data in each of these files, so there's a household file, insurance, interview and screener files.  And some of them are pretty darn big.  So you need to unzip the National files and they're all SAS files, and two of them in particular are rather large.  And you need to translate them into a format for your statistical software.  I am not a Sedan user and after attending last year's training, I am never going to be a Sedan user.  But was lucky enough to discover that Stata can also handle this kind of data and it's a whole heck of a lot easier to use.  But until last February there was not a version that would accommodate all the missing values that are in this data set.  It's not just one missing value, there are like six, seven different letter codes for missing values and you have to be able to import then all accurately to be able to analyze the data correctly.  So then you subset to your State number and there's a codebook for you.  And if you use Stat transfer, it's handy, you can do the translating and the subsetting all in one step.  Adding value labels is rather a pain in the rear if you're not using SAS or Sedan, or if you have SAS on your computer, the nice folks at Stata will send you some code to translate them and put the value labels in, but for those of us that don't have SAS, it's a laborious process and if anybody wants a CD I can send you the National CD with the labels on it in Stata.  So next step is to link the data files.  The household file is where we get the household prevalence, but it also includes some important information.  It includes MSA status, which is the only sub-State geography we have, metropolitan statistical areas.  May or may not be useful to you.  It includes information on family size, on poverty level relative to the Federal Poverty Level and whether or not they get cash assistance. 

And of course a weight for that file.  The screener file is for each child that was screened, these 373,000 kids, it provides a child I.D. number, links them to the household information by the household I.D. number, provides specific demographic information on the child, age, sex, race, ethnicity, and also all of the screener questions, the CSHCN screener questions.  And then it summarizes the result of the screener questions as to whether or not they qualified as a special needs child and whether or not they were interviewed.  And then for those who were identified as special needs kids, the interview file has been created that includes all the answers to the 268 questions, or something, I mean it's a huge number of detailed questions that were asked and it links by the household I.D. number back to the household file where you can link to the MSA status and family size and poverty level, and link back using the child I.D. number to the screener file where you get the child's demographics and the answers to the specific CSHCN and screener questions, although the summary flag is on the interview file.  So you've created your file, now you need to make sense of your data for your State.  I understand that the new version of SAS and SPSS13 both say that they can appropriately replicate the confidence intervals for this type of survey, I haven't tested them, but Christi's group is using SAS and it seems to be working.  So the first step is to try to replicate some of the analyses that are presented on the CSHCN survey website that Stephen has done, including the household prevalence and the screener prevalence.  And if you're setting up your survey data this is the variables that you need to use for the Strata, the primary sampling unit and the weights.  And just make sure that you can replicate those standard areas, as well as the prevalence numbers.  I also recommend that you try to replicate some of the core outcome measures that were published earlier and sent to the States, and Stephen very kindly included his SAS code so you know which variables were used.  Even if you don't program in SAS, it's fairly easy to figure out what variables he used and how he used them.  And then you also need to validate your data.  This is something that sort of got me in a little bit of trouble earlier on, but this is an incredibly complex survey.  And at the time they did it, the 2000 census data weren't out yet and they needed to wait for 9, 10, 11 different things for these tiny little subgroups of the population. 

They still don’t have all the data they need to properly weight for household size.  It hasn't been released for family households with children under 18 in them.  I mean, it's physically impossible for them to have generated totally accurate rates when they don't have the data to do it.  But you do need to check against your own States data sources and make sure your data actually reflects your population.  We validated the household data using primarily census 2000 data and you'll notice that households with children under 18 is not within the 95 percent confidence level of what we would expect to get for our State.  It's just an unlucky guess, I guess.  The screener data, after they readjusted the weights, most of the racial and ethnic groups came in within the 95 percent confidence level of our census estimate.  Remember the census estimate is 2000, the survey data is really 2001, but we're still a little high on the white-only population, and I suspect that's a function of the household size.  So they're all linked.  It's very difficult to validate the prevalent stuff because most of us don't have State data, that's why they did this survey for us.  We have no comparable population based data in Washington for this.  We have some partial validators, we've worked with our Medicaid CAPS survey, we know something about the CSHCN screener who they picked up in our Medicaid population, but that's a very bias sample.  And the screener was partially included in our Burfus 2000, but they left off a question and two parts of other questions, so it's not terribly helpful.  You need to educate yourself about the variables included in this survey and the CSHCN screener itself.  Christi mentioned that these ambulatory pediatrics articles are on their website and one of the variables that threw me for a loop was MSA status.  And every time I present our data around our State they say, "But we don't live in a city."  The people who live in rural Yakima County are not in a city, but it's part of the metropolitan statistical area.  In the West we have humongous counties.  Stratifying by MSA status doesn't tell us much.  In the East you might be better off, but it doesn't work a lot in our huge counties.  So we've done the prevalence’s and that's always interesting to people the difference between household prevalence and population prevalence.  We've looked at sort of the basic demographic characteristics, prevalence rates.  One of the measures that I particularly like is Christi and Deb developed something that I call their complexity measure, instead of just looking at each of the five questions in their screener, they combine them to sort of get an idea of how complex the kid's condition is, so that some kids who qualify only on their question about whether they're managed by prescription medications are really functionally different from those who have functional limitations in their daily activities.  And a lot of the measures that are in this survey fall out very nicely by this complexity measure.  Hopefully they'll put that on their website too.  In our five-year needs assessment from 2000, they identified seven priority areas that we've been trying to figure out whether were making any progress on and for three of these, we have developed some data from the survey that we talk about around the State, the lack of coordination of services, healthcare financing issues and availability of support services.  So which kids need care coordination?

 Well not surprisingly, the kids with functional limitations are much more likely to need care coordination than others.  These lavender colored vertical lines are the 95 percent confidence interval and I implore you please make sure you look at those anytime you do an analysis because there could be what look like huge differences that end up not being statistically significant.  One of the things that surprised us of the 15.7 percent CSHCN parents in Washington who said they needed professional care coordination, 82 percent were actually getting it.  That absolutely floors us.  Insurance coverage.  We tried to compare ourselves to the U.S. as a whole.  This is the lowest level we can go with insurance information because our State does things differently.  Many of our parents don't know whether they have Medicaid or they have private insurance, and so even this we have some questions about, because most of our Medicaid is through a Healthy Options Program that they’re actually insured by a private company.  So, it's very confusing for families.  Ever not insured by household income.  Notice that the only significant difference here is between the 200 to 399 percent of Federal Poverty Level and below 100 percent.  I mean it looks good, but.  Any unmet child health need by family income.  Not surprisingly lower income, as Stephen said, is much more likely to have at least one unmet health need.  This is a series that I wish we had a much larger sample so we could do more with it.  These are some of the services needed by children with special healthcare needs.  Medication most needed, but that's because that was one of the five-screener questions basically, in part, that allowed kids to quality.  One of the big differences, the 27.5 percent is mental health needs and of those 27.5 percent almost 20 percent are not getting the mental health services they need.  This will undoubtedly be a focus in our next five-year needs assessment.  We tried to cut that by age, but the sample just gets too small.  Get too many small cells.  Percent in terms of receiving services, I was asked by one of our regions to look at the percent of ages 3 to 17 receiving special education services.  Not surprisingly there are not a lot of kids who are just on medication who need special ed, but quite a few of the kids who have functional limitations do need special ed services and are getting them, thank goodness.  We have some work around topics.  We need to find a way to get a handle on rural/urban differences, because each time I present in our State, it's "That doesn't apply to us.  We don't have access."  I mean, we've got five counties that don't have any pediatricians or family practice doctors who will take Medicaid.  So we have families driving 200-250 miles to get to a doctor who will take their Medicaid coupons.  We need to find some way to work around that.

I'm hopefully going to try to get somebody to go to the research data center at NCHS and get to the actual zip code data and code it by rural, urban, ruca code I guess it's called, which should be out next year.  Source of health insurance, we're having problems with that still in our State and we're trying to find a way, find some age groups that will work for looking at the utilization and impacts of not getting care.  I would encourage more States to use these data.  As far as I can tell there are just a handful who are currently trying to analyze the data themselves and I think there will be an incredible amount of information and a lot that will come out of the data resource center.  But they're not going to be able to answer all your questions, and they're not going to be able to set it up specific to what each State needs.  We need to find a way to share the work-in-progress.  I didn't know that Rohini had been doing some work on this survey last summer.  I wish I had because I would have been happy to help her.  It would be really helpful if somebody could help us identify appropriate peer States.  I mean I'm sure there are other States out there who organize their Medicaid, who organize their CSHCN programs the way we do.  Are you going to do that?

Christina Bethell:  Well I was just thinking that part of the resource roundtable is to look at what States are doing and States using data, and that's a perfect place to put your materials and the other States can share what they're doing which go beyond the query tool.

Ginny Sharp:  And we will be continuing to generate analyses from our date in response to the questions that are coming up in these regional meetings.  If you want more information on how we're doing the survey analysis, you're welcome to contact me.  The person in our State who is responsible for the needs assessment planning is Stacey DeFries and if you'd like to hear about how we're doing our five-year needs assessment she'd love to talk to you.  Thank you.