AMCHP 2006 ANNUAL CONFERENCE
EARLY CHILDHOOD: BUILDING THE FOUNDATION FOR LIFELONG HEALTH
March 4-8, 2006
PAUL NEWACHECK: I'd like to echo what Michael was saying earlier, this is a very exciting time having data from this survey now available and seeing all the analyses that are being conducted out in the field. And we will have the second iteration of the survey going, as Michael said, next year. And that will provide an opportunity to do actual trend analysis. Because most of the indicators will stay the same between the 2003 and 2007 survey.
So those of you who are survey data users and want to look at trends either in your own state or nationally, this will be a real opportunity for that. And something to look forward to.
So I want to be talking about determinants of oral outcomes of children. I'll start out indicating our research questions. I want to talk about the conceptual model that we developed for this. We spent a fair amount of time on that. And then some preliminary empirical findings, both from multi‑level modeling work that we've done and then also looking at some of the variables that predict oral health outcomes.
Before I get into the questions, I want to tell you about our study team. We have pretty good size group here. Mars Sue Bader, statistician, epidemiologist based in Boston. Susan Fisher‑Owens, pediatrician at UCFS; Jan Weintraub, pediatric dentist; Stewart Ganski, dental epidemiologist; Larry Plata, pediatrician; and Matt Bramlet with the SLATES team at MCHS. He's been working with us on analyzing the confidential data from the survey where we get geographic identifiers not available on the public use data.
This project was supported or has been supported by the National Institute for Dental and Craniofacial Research at NIH. I want to emphasize the results I'll present will be quite preliminary. It's taken a lot longer than we thought to do this analysis partly because we imputed data where there's missing cases, and that took a lot of statistical programming time to create imputation values for those occasions.
And also the multi‑level modeling work ended up being a lot more work than we thought. We're kind of in this very preliminary stage. What I'm going to talk about today are preliminary results.
And I realize this is supposed to be an early childhood but we actually divided our population into two groups the early childhood group and school aged group and we started with the school aged group that's where we are. I'm going to be talking about school aged results even though the session is on early childhood.
So our research questions are, first, what are the child, family and community variables that predict oral health outcomes? And second, what is the role of context or place in explaining oral health outcomes. This is really getting at the multi‑levelness of it. What is the role of factors that operate at the county or state level in determining oral health outcomes for kids, first those that operate at the child or family level.
Those are our two basic questions that we're looking at. And the rationale behind these questions is that that knowledge about what are the determinants and what are the role of contextual factors in determining oral health is important to creating important prevention policy. That's where we're coming from.
This is our conceptual model. I'm going to show it to you in several stages. We start out with this very simple model of ‑‑ it's kind of a little hard to read. But this is the classic epidemiologic triad for oral health where there's three different interactive effects.
There's the host and the teeth. There's the microflora or bacteria in the mouth and the diet. These three factors interact to produce oral health outcomes for kids, name least caries and such. That's the classic model. We know from population health work and epidemiology there are many other factors that play a role beyond what happens within the child itself.
So we expand the model and this first stage we have child level influences that go beyond what's happening in the mouth to include health behaviors and practices. Biologic and genetic endowments, use of dental services physical health, dental insurance, development, other things like that.
And then beyond the child there's the family. And here we have a number of characteristics at the family level that may play a role in influencing oral health outcomes, including socioeconomic status, social support, family functioning, health behaviors, those sorts of things.
And then the next layer out in this ring is the community. And here there's the social environment, the physical environment, healthcare system, culture, social capital, all these things that happen outside of the family. But then could contribute to oral health outcomes. So this is our kind of broad model. But we add on top of that at least conceptually the element of time because all these processes take place over a period of time. They don't happen immediately and they take place within the context of a child's development. So they're happening gradually while the child is developing. So this is our complete model, conceptually. And we actually have paper where we describe this and we build it using the population based health literature. When we actually try to model this, we have to simplify things considerably. In particular, the time element is something we can't really address using cross sectional data. So that's one major, major simplification. But we also don't have measures for all of these different variables that are in the model within the 2003 National Survey of Children's Health. We have a lot, but we don't have all of them. So we're missing some. Of course there's the issue of causality too and the multi‑directional nature of causality and we have to simplify around that. So our empirical model is a gross simplification of the conceptual model. Let me talk about what we do with our empirical analysis, though.
The outcome variable in our case is the parental assessment of the child's oral health status, and that was in five levels, excellent, very good, good, fair, poor. We dichotomize that into excellent, very good versus the other categories for this analysis.
We used a procedure called the GLEMIX for the multi‑level analysis. We used factor analysis to reduce the number of individual variables and down into indexes and condensed variables, because we had so many things to look at. And we used Sudan to do our variance estimation. We're looking at two ages groups, zero to five, six to 17. We had started with six to 17 and that's where we still are. So that's what we're talking about today.
I don't think the results would be dramatically different for the zero to four, at least in the broad kind of way that I'm going to show you.
So let me start with our multi‑level modeling results and then I'll go on to the individual predictive variables that seem to be important. The question here really is what is the effect of place and explaining oral health outcomes, and we can use the survey data to look at the child and the family, for example, and some of the community variables that were collected in the survey. But we know in addition there are factors that operate, say, for example at higher geographic levels like the neighborhood or the county or the state. So there might be, for example, variations in state environmental quality or water quality or fluoridation policies or things like that that could play a role in determining oral health outcomes. So we wanted to try to look at this as well and see how much of the variational oral health outcome for kids is determined at the level of child and family versus at these higher geographic levels.
So we did two models. We did two level multi‑level model where we looked at individuals and states. And then we did a three level model where we looked at individuals counties and states and these are called null models where we basically are looking to see what's the maximum amount of the variation in the oral health outcomes that could be explained by these higher level geographic contributors. So all the factors, whether it has to do with the environment, the social environment, the policy environment, whatever is happening at those levels should be reflected in this null model. Gives us a sense of the maximum contribution of these factors.
You can see from these results that there isn't a lot of contribution of the contextual level variables. In the two level model only one percent of the variation in oral health outcomes can be explained by state level factors. And in our three level model, only 2% of the variability in child health outcomes can be explained by things happening at the county level or the state level.
So what this says is that virtually all of the variational oral health is happening at the individual level that is at the level of the child and family and in the immediately environment and not much is happening at the county or state level. Doesn't mean there's more that couldn't be happening at the neighborhood level if we had a good measure of that but we don't have that in this survey so we're restricted at looking at county and state.
That's our first findings really there isn't a lot happening at the state or county level, at least, in terms of determining oral health outcomes.
So I'll turn now to talk about what we've found at the individual child level and family level and within the immediate community, what variables were important for children. I showed you the conceptual model. We have the three different concentric rings, the child, the family and the community. And we put together a list of variables from the national survey of children's health that reflected each of these different areas, about 25 or so child level variables, about 25 family level, and half a dozen or so community level variables. And then put these into regression equations, series of regression equations to see which ones were statistically significant. And these are the results. These are all of the factors that were statistically significant.
And this is at the child level to begin with. And these are grouped according to roughly the categories that were in that conceptual model. So the important variables include in the biologic area include our proxy measures of birth order, parental health and this is a factor index that includes physical and emotional health of the parents.
Physical attributes, age, race, ethnicity, BMI, presence of special healthcare needs were statistically significant. Use of dental services. Preventive and dental insurance was important. Sleep patterns, exercise patterns, protective head gear use was an important statistical predictor and then child's development as measured by problems in school and index of after school activities was statistically significantly associated with oral health outcomes as well.
So that's the child level. At the family level, we have another set of factors that are significantly related to outcomes, including family structure, family functioning, measured by several indicators. Socioeconomic status. Parents health behaviors and coping skills. We had several indices we created there using factor analysis, exercise, health insurance, coping.
And then family culture as measured by language at home. And so all of these factors were statistically significant in terms of predicting oral health outcomes for kids.
At the community level, we found two major factors to be important. One was an index of social capital. And this is measured by the parental perception of the neighborhood where the child lives, whether the neighbors are helpful, whether they watch out for children in the neighborhood, whether they can be counted on, trusted, and whether the neighborhood is perceived as being safe by families.
There were also two measures of physical safety being safe at home and safe at school that were significantly associated with oral health outcomes.
So those are the sets of factors. So we have quite a few variables that we found to be significant. And these are the broad categories, just summarizing the broad categories that we found to be significant in terms of explaining oral health.
So how well did we do in terms of the explanatory power of these models? Actually not that great at least at the individual level. At the state level we did well. If we look at, for example, state to state variation and oral health outcomes for kids, about 65% of that state to state variation is explained by our model. So that's pretty good, actually. But at the individual level, where there's naturally more variability because of randomness and other factors, we didn't do so well. We only explained about 10 percent of the overall variation with this model even with those 40 or so statistically significant variables. We're not doing that well explaining those outcomes for kids.
So we're a little disappointed there but this is still preliminary, and maybe we'll do a little better later on.
So there are some methodological issues and challenges that we're still working on. One, of course, is to try to explain that other 90% that's left over of the individual variation oral health outcomes. The second is in our multi‑level modeling we looked really just to county and state levels. And we're not really capturing neighborhoods when we do that. Counties are big areas, you know often with hundreds of thousands of individuals. Sometimes millions in them.
And they don't really reflect neighborhoodness. And unfortunately in the national survey of children's health we don't have enough information to really capture the neighborhood or at least in a meaningful way. So that's a limitation of our multi‑level modeling.
I think if we did have that information we'd probably do a better job of explaining some of the variation in outcomes for kids.
And then finally there's this issue of directionality and causality, we're assuming everything goes from the independent variables to the outcome variable as we've defined it but in fact we know that things go back and forth, there are feedback loops, nonlinear relationships and such that we're not really capturing in our analysis. So that's something we'll be working on in the future.
So let me just summarize what our conclusions are. First, our starting point was that it's important to understand variation in children's oral health outcomes as a means of devising effective intervention and policies. And what we found out is there are multiple factors that contribute to oral health outcomes and they do take place at multiple levels. We need to do better at identifying all those factors but we do know they're present. And the analysis really suggests that if we are going to design interventions that are going to be effective for reducing oral health problems for children and especially disparities, we need to take on some kind of multi‑factorial approach. So that's basically where we are at this point. So I'll stop there.