Ninth Annual Maternal and Child Health Epidemiology Conference / December 10-12, 2003
VIOLANDA GRIGORESCU: Good morning. First of all, I would like to thank the MCH-EPI conference organizing committee for the opportunity to present this analysis, which is kind of first step and it needs to be continued. The analysis was conducted when I was assigned to work with Georgia Division of Public MCH Epidemiology Section, and I would like to acknowledge my colleagues, Mohamed Qayad, Emily Kahn, and Debra Hersh, especially Mohamed Qayad. He’s the guru-- Medicaid guru-- in Georgia. He is responsible for Medicaid data, and Debra Hersh is very interested in pre-term birth. I’d also like to acknowledge my colleagues from CDC Division of Reproductive Health. Dr. *(inaudible), Dr. *(inaudible).
Why we decided to look to maternal morbidity. Some of you have been yesterday in the sessions related to maternal mortality and maternal morbidity, and we discussed that so far we use maternal mortality as the main indicator to measure maternal health. But this represents the definitive consequence of severe maternal morbidity. And we learned in Georgia when I was there and I did the *(inaudible) investigation, that we don’t know enough about maternal morbidity. Is less frequently measured, is more difficult to track at the population level than maternal mortality. Of course, as you know, and I’ve seen faces in the audience, people who are pioneering, looking to maternal morbidity, and publishing different papers. That’s Dr. *(inaudible) and yesterday we thought we’d *(inaudible). Maternal morbidity is very complex, *(inaudible) relates to a preexisting medical condition that affects or is exacerbated by pregnancy or a pregnancy related medical condition. Why we decided to look to maternal morbidity to women *(inaudible) only Medicaid.
We know that poor social economical status may lead to poor health outcomes, and these women who are enrolled in Medicaid, most of them have the-- are a considerably high risk population with poor socio economical status. But also, this was a Georgia priority. Almost 50% of live births in Georgia are Medicaid covered. This is why we decided to look to Medicaid women, and to look to the impact of morbidity on pregnancy outcomes. This *(inaudible) objective is why we decided to do this analysis and I’m not going to read all of this. The main point is we that we didn’t like that point in time to look to maternal morbidity to inform our Medicaid providers and to *(inaudible) collaboration to be able to develop better strategies. We used 2001 Medicaid delivery claims data in patient file linked with birth certificate data. We looked to prevalence of different health conditions, and we decided that we might want to link to pregnancy outcomes given the fact that we talk about MCH Maternal and Child Health.
So we choose to look to pre-term, which as you know, is of national concern and the entire analysis was conducted by using SAS for all *(inaudible) and data analysis. During our data cleaning and exploratory data analysis, we looked to different variables recording *(inaudible) to files, and most of you know that we have *(inaudible) in both files. It was a challenge to understand or to decide which one we should use, and I’m not going over all because the purpose of this presentation is not to talk about the differences between these two files. But we decided after this cleaning and data, and exploratory data to chose maternal age from Medicaid claims data, and ICD-9 codes, of course, because they are not in birth certificates.
And from birth certificates, they *(inaudible) to choose *(inaudible) mother race, and I can give you an example of *(inaudible) race the differences between two files. They don’t match, of course, 1,257 recorded as unknown race in Medicaid we found as being *(inaudible) as white in birth certificates, and 644 recorded *(inaudible) Medicaid *(inaudible) white in birth certificate file. So we felt more comfortable to use race from birth certificates because it is used previously, is national recognized, and we assume that is better coded. And then we did this with other variables, because I said the purpose of this presentation today is not to show you the differences between two files. There were 224, 429 ICD-9 codes in Medicaid claims data, and I’m talking about just inpatient file.
In Medicaid, the way we receive it in Georgia, had two diagnostic fields with one or more ICD-9 codes ranging from one to 14 for one woman. We considered all ICD-9 codes, and we grouped ICD-9 codes into 116 groupings. Those unrelated to maternal health were not considered in the analysis, and when I saw those unrelated, I mean those related to fetal, congenital and *(inaudible) other fetal diagnostics, or let’s say, some women have diagnostics like delivery or c-section only. So, we didn’t look at those, we look only to maternal morbidity. I know this is very debatable and we are open for any criticism, but I was trying to understand what other conditions we may experience, and I’m talking not just about those pregnancy related, I’m talking about chronic conditions, health problems that women have prior to pregnancy. So we kind of use these two groups, big groups, non-pregnancy related, assuming that they are pre-existing, health conditions diagnosed before or during pregnancy but not determined by pregnancy. And I say diagnosed during pregnancy because many women may not know when they are pregnant if they have any problems.
They are diagnosed during pregnancy but doesn’t mean that is pregnancy related, and pregnancy related, those health conditions determined by the pregnancy status. I would like to look to the last one. It was very hard to decide only from ICD-9 codes for some visits or some health conditions if they belong to the first two groups, and some of you who really looked at ICD-9 codes know that it’s hard to look only to a code and to say, “This is previous,” or, “This is during pregnancy.” For some of them it’s not so complicated, but for some of them it is and I can give you an example of anemia. If you look to 280 codes, it is, like, previous you assume because there’s not 600 codes during pregnancy, but if you go back to 600 codes, it is listed as condition classified *(inaudible) 280. Now, the question is, condition classifiable, where is this anemia? Where it occurred? Prior or during pregnancy? Well, it’s very hard to decide without medical history, without knowing a little bit more about this condition, so I took the liberty to keep certain conditions in the third group that I didn’t use in the analysis just because I wasn’t sure if I can have this in either non-pregnancy related or pregnancy related.
And as I said, I know it’s debatable, we are open for any criticism, but I didn’t feel comfortable when I looked to this grouping to pose this condition in either one of the first two. Looking a little bit to demographic characteristics of the population in Medicaid, as we all know and probably everybody I assume before looking at this that we had a little bit more younger women with a mean age of 23 years. I didn’t find *(inaudible) differences between races, and we did the ratio-- black white-- we have more if you remember from the one previous slide. There were more white women than black, and when we did the ratio, the ratio for the entire sample size was .8 and this is the same ratio within each group, like below 20, 20, 30, 40 years of age, and above 35. The differences were not significant even though we can say for those below 20 and above 35, there were more black than white.
And I would like you to keep in mind this ratio because I use this ratio every time when I look to different health conditions just to make sure we understand when *(inaudible) by race, the differences. The prevalence of non-pregnancy related so-called preexisting health conditions. In this *(inaudible) I felt comfortable to look at by no means this is a perfect grouping, by no means this is the only conditions we should look at. This were conditions I found like having *(inaudible) important prevalence and I could look at with a confidence that this are preexisting condition only by using ICD-9 codes, and the *(inaudible). I wasn’t able to understand if there other surgery or just c-section, but I assume most of this are previous c-section. As you can see, *(inaudible) the first one followed by previous hypertension *(inaudible) diabetes and so on. However, when I stratified by race and I used the black/white ratio, the pattern changed and as you can see in the first column, I’m talking about the ratio for the entire sample size.
When we look to different health conditions, the pattern is not the same. Diabetes mellitus is the first one and I’m sure for some of you it’s not a surprise because, you know, this is having the highest racial disparities. Previous diabetes mellitus was followed by previous hypertension, asthma, or *(inaudible) and so on, and *(inaudible) are stratified by race how it looked like. For pregnancy related health conditions, I pulled those you already know and this was a little bit easier because *(inaudible) 600 ICD-9 codes a little bit easier to define if they are pregnancy related. Pregnancy hypertension all stages were the first one with the highest prevalence, followed by premature rupture of membranes, venereal disease, gestational diabetes, infection *(inaudible) cavity and so on. However again, when I stratified by race, the pattern changed. The first one was venereal disease followed by infection of the *(inaudible) cavity, pregnancy hypertension and so one. I looked to hypertension during pregnancy by stage just because I knew we have different stages and we can look at ICD-9 code and actually look at different stages.
It’s not surprising that *(inaudible) hypertension is the first one because we have more women having this *(inaudible) hypertension and less having eclampsia. The same case when I stratified by race the pattern changed radically, it’s reverse. Transient hypertension was *(inaudible) in African/Americans and eclampsia was having the highest disparity. It is true the numbers in eclampsia were a little bit smaller, but still, it’s showing that’s how the differences occurs as we look to stages of hypertension during pregnancy. Well the question was if we look to this maternal health condition, what is the impact on pregnancy outcomes? How does maternal health conditions relate to the health of *(inaudible) baby? So as I said we looked to pre-term birth, we did the *(inaudible) by *(inaudible) analysis. And in our exploratory data analysis *(inaudible) way trying to understand the type of variables we are going to use, and we *(inaudible) developed different models to estimate the worlds ratio for pre-term birth, adjusting for maternal age and race.
And I’ll go back a little bit why we choose to look only to those two variables. After animal *(inaudible), we find out that the health-- looking by groups again, preexisting condition and pregnancy related condition. In preexisting condition on non-pregnancy related condition, diabetes mellitus *(inaudible) up to be the most significant associated with pre-term birth with the world ratio of 2.65, followed by cardio diseases and previous hypertension. In pregnancy related health conditions, severe pre-eclampsia was the most significant one and I don’t think it’s surprising for anybody because we already know and it’s already proved. *(Inaudible) followed by abruptia placentia, premature rupture of membranes, eclampsia, placenta previa, and *(inaudible) cavity. We know that pre-term births is a very complex problem with no simple solution, but the message is that maternal health conditions have an impact on pregnancy outcome *(inaudible) pre-term birth.
We need to consider in further analysis when we look to pregnancy outcomes, maternal health conditions. These findings reflect only Medicaid clients. Non-Medicaid mothers may have different health conditions, and we know still maternal morbidity has many unknowns and is a continuing challenge public health professional. But as I said, if we want to develop better prevention strategies, we should understand it better. The public health implication of this study is to look more, to explore further, to conduct more epidemiological analysis, not only by using Medicaid *(inaudible) using hospital discharge data, to better understand the impact of morbidity on pregnancy outcomes, to identify better prevention strategies. Also this is a way to improve the collaboration among health care providers-- or I should say between health care providers and public health professionals-- to develop better follow-up protocols, and even to be able to develop a *(inaudible) approach to women’s health, and some of you probably already read-- was published in American Journal of Preventative Medicine-- a group of physicians who came up with a very nice approach for a lifespan approach of health care to women.
Of course, the study has a lot of limitations. We looked-- first of all; Medicaid is a billing system and is not for epio research status. It required a long time to do accurate cleaning. We look only to inpatient file. At the beginning, we were very enthusiastic, thinking that we will be able to use out patient file but we find out that there are many missing ICD-9 codes and we were not able to use it. That means we might have underestimated prevalence of different diseases because we didn’t use all files. Also, as I said previously, there are different-- the services may differ by health *(inaudible) status, I’m talking only about Medicaid, but if we explore further and only to hospital discharge data to women covered with other insurance, we might find differences. As I said, this is *(inaudible) it’s just the first step, it’s not the end of the story, and it’s just to give you the challenge to think a little bit more. It doesn’t want to be by any means the best way of looking to maternal morbidity. And I would like to thank you very much for your attention. Thank you.