AMCHP 2005 ANNUAL CONFERENCE
DELIVERING RESULTS, IMPROVING PREGNANCY & BIRTH
February 19-23, 2005
DAN THOMPSON: I’m going to talk about pregnancy-related mortality associated with obesity in Florida for 1999 through 2002. Okay, in 1996 Florida initiated the pregnancy associated review process, we call it PAMR, to improve surveillance and analysis of pregnancy-related mortality. And the reason we did that was to develop a systematic approach to examining maternal deaths and the reasons we did that were because we saw some CDC publications confirming underreporting of maternal deaths and we had questions about the relationship of maternal death to changes that were occurring in the healthcare delivery systems. And probably the biggest thing that moved us ahead to do this was a cluster of maternal deaths were observed in a Florida county and that--there was a lot of public--there was some media attention on that and so we decided to get more formal about our pregnancy-associated mortality review.
Okay, and then we used the CDC and ACOG expanded definition of maternal mortality, and that definition is, “death of a woman from any cause while she is pregnant or within one year of termination of pregnancy regardless of the duration and site of the pregnancy.” So that’s the definition we were using and then in PAMR, pregnancy associated deaths are reviewed by the PAMR review team. And the review team is a multidisciplinary committee of 31 professionals and it includes, physicians, nurses, nurse midwives, social workers and researchers. The review team meets quarterly to discuss the abstracted cases and they examine trends, common elements and they formulate potential strategies to address these factors. And the team members are volunteers and they volunteer their time and they also provide their own travel costs. So they get together once a quarter and review the cases. Each case is reviewed for these issues.
And these issues are medical problems, medical problems during pregnancy--history of medical problems, medical problems during pregnancy, medical problems associated with labor and delivery, medical problems during the post-partum period, nutritional issues, access to prenatal care, substance abuse, absence of prenatal risk assessment--I’ll explain a little bit more about that later--lack of social support, problems with housing, mental health problems, family violence or neglect, social issues, transportation problems, problems with provision or design of services, environmental or occupational hazards and concerns about family planning access. These are all things that they look at in the review.
Okay, after the review, the team gives a final classification to each case as pregnancy-related, possibly pregnancy-related or not pregnancy-related, so they review all these cases and then they put them in these three categories. Okay, but what I haven’t told you is what are the cases they review when they do all this, so this is how we select the cases for this review. So the division of Family Health Services, working with the office of Planning Evaluation and Data Analysis, we implemented a selection criteria to maximize the identification of pregnancy-associated deaths. And women are included on this surveillance list of pregnancy-associated deaths by any of the following four ways. So basically we’re going to look at several data files we have available to us and based on certain criteria, select records from those files to give to the PAMR team so they can make those determinations. And one group is, we select records where the response on the death certificate is, “Yes,” to the question on the death certificate, “If female, was there a pregnancy in the past three months.”
I think this question has changed on the--on the new birth certificate that--that we began using Florida in March of 2004, but on the old birth certificate prior to March of 2004, this is the question we were looking at. Then we also choose death records if the ICD-10 code indicates a death classified as due to pregnancy, childbirth and the properium. That’s as close as I can get on that one. So we look at the ICD-10 codes on the death record and pick them out that way too. Then another criteria we use is there’s a matching birth or fetal death record within 365 days prior to the woman’s death, and we have computer programs that search the records and match them this way and we‑-we choose those records too. And then the last thing we do is we have a university prenatal risk screening instrument in Florida and it’s universal, but in reality, only 50 percent of women who give birth have this screening done, but we have a file of those records too, and if one of those records matches the death records, then we choose those records too, so that’s another case-finding tool. So any records identified by those--by those four methods are abstracted and forward to the PAMR Committee.
Okay, then the case abstraction process is modeled after the fetal and infant mortality review process, and the abstraction forms capture information from a medical and social history, prenatal labor and delivery, post-partum, social service, care coordination and terminal events records. And we hire FEMR abstractors to do this, so they’re skilled abstractors, they abstract the infant mortality review records also and we hire them to do the--the PAMR records for us. So they collect a lot of information for the--for the review process. Okay, so now we’re going to--that’s--that’s just background, so now we’re going--then I’m going to actually talk about the data. In the four year period of ’99 through 2002 we had 143 deaths classified by the PAMR review as pregnancy-related in Florida . Now, in the review process, anecdotal evidence suggested that obesity might be associated with increased risk of pregnancy-related mortality, so we--the purpose of this analysis is to quantify the relationship between obesity and risk of pregnancy-related mortality in Florida based on this, based on the 143 deaths.
Okay, this--you can’t see this, but this shows that the distribution of pregnancy-related cause of deaths, the biggest category is other at 24 percent, the next biggest is embolism at 20 percent, and after that we have cardiomyopathy at 15 percent. Here’s a table, you might be able to read that better. And these are sorted by the--by the largest at the top, so that’s just some background information about the cause of death. Okay, so what we did was a matched case-control analysis and generally, in a case-control analysis, cases are persons with the attribute of interest, in this case death, and controls are persons that do not have the attribute of interest, so these are--controls would be persons who didn’t die. And then inferences are made by comparing the cases to the controls, that’s just generally what--what you do with a case-control study. And with a matched case-control study you match the cases and the controls and in this analysis cases are pregnancy-related deaths and controls are women screened with a health start prenatal risk screen, so both the cases and the controls were pregnant, so they’re similar in that way.
Since they were both pregnant, they were both at risk for pregnancy-related death. For each case we pick four match controls and they were randomly selected from the controls that matched and they were matched on education, age, marital status and race. So for each case we have four persons who have the same education, age, marital status and race and that’s the date set we use to do the analysis. So matching is a method used to control for the confounding influences of the variables. In this instance we’re controlling for education, age, marital status and race by matching. So there were 143 PAMR deaths in Florida in the four years and for this analysis 28 per--28, or 20 percent, of these deaths were excluded due to missing data for height, weight, age, race, marital status and education, so we had 115 cases left for the analysis. Pre-pregnancy height and weight data are on the PAMR record for cases and it’s collected on the prenatal screening record for the control, so we had the height and weight data for both the cases and the controls.
And this was used to calculate the body mass index for cases and controls, and the formula for body mass index, of course, is weight divided by the square of height, where weight is in kilograms and height is expressed in meters. So here’s a table of our cases and controls and you can tell by reading the normal weight category there that cases are 25 percent--25.2 percent of the cases are normal weight and a much higher percentage, 47.2 percent of the--of the controls are normal weight. So the cases are less likely to be normal weight; the controls are more likely to be a normal weight. And if you turn that around you could say the cases are more likely to not be a normal weight and the controls are more likely to be the normal weight. So here’s a graph of--of that and you can see that in the normal weight category, the bar for controls is at 47.2, much higher than the bar for--for the cases. So the cases tend to have higher percentages and the higher BMI category as compared to the controls and as I said, the 25.2 percent of the cases are in the normal BMI category and 47.2 percent of the controls are in the normal BMI category.
Okay, so when you have matched case-control data you need to use conditional logistic regression to calculate the adjusted odds ratios, confidence intervals and P values and this is a method, it’s--it’s a--it’s a method--it’s a variation of logistic regression which is used for--for the situation when you have matched case-controls. It’s--it’s very similar to logistic regression. So here’s the--here’s the end result. In the adjusted odds ratio column, you can see that we have sort of a dose response there where the referent category is normal weight and overweight has an odds ratio of almost two and it’s significant, and then the odds ratios go up from there and they’re all significant until you get up to obese three, which has an odd ratio of 5.12 and it’s highly significant. And the confidence intervals there also show that--that they’re significant, because they--they all exclude one, except for the underweight category. The underweight category is almost significant, but not quite.
So that’s basically our results. And so we have a dose response and we have an association with obesity and overweight and almost an association with underweight. So this is odds ratio of death associated with obesity. And here’s a graph of the same thing and so you can see the odds ratios go up with each increasing category of obesity and the odds ratio are slightly--are elevated with underweight, but as I said, the--the 2.47 for underweight is not significant, not quite significant. So I’ve already said all this. Okay, the limitations here are--we had missing values which resulted in exclusion of 20 percent of the cases, so if the excluded cases were substantially different from the cases in the analysis, the results could be biased. Also the BMI is based on self-reported pre-pregnancy height and weight and there could be reporting bias, since it’s likely that people tend to underestimate their weight. But if the important bias affects both cases and controls equally, the odds ratios will tend not to be affected since the biases cancel in the calculations, but of course, they may not be biased to the same extent in the cases and controls, so that could be a problem too. So the conclusions are, the adjusted odds ratios indicate that high BMI is associated with greatly increased risk of pregnancy-related death. Since the cases and the controls were matched, these odds ratios are not influenced by differences in race, age, education or marital status between the cases and the controls. So we control them for those things, so it’s not differences in those things, between the cases and the controls because they were matched. Okay, that’s it.