MCHB/EPI Miami Conference — December 7 - 9, 2005

Methods Can Help — Transcript

DAVID HOWARD: Morning, everyone. My name is David Howard, and I just recently finished my Ph.D. at Johns Hopkins in Epidemiology and currently a post-op fellow at the Carolina Population Center at UNC Chapel Hill. And this--the work I'm presenting was sort of the leadoff paper in my thesis. And so the title is: "Bias May Help to Explain the Paradoxical Positive Dose-Response Relationship between Incarceration during Pregnancy and Infant Birth Weight."

So a few studies have reported a positive dose-response relationship between the quantity of exposure to prison during pregnancy and birth outcomes. Two prior studies using the same data from North Carolina focused on birth weight. One prior study by Corder Adel looked at preterm birth. And of course, you know, in light of stereotypes about the prison environment, you know, these findings were somewhat paradoxical. And the studies, however, by Martin et al in particular, which were the ones that--those two studies reported and presented a graph and showed what, you know, they referred to as, you know, a linear dose-response relationship between, you know, incarceration during pregnancy and birth outcomes. However, those two studies by Martin et al in particular used the methodology for quantifying exposure to prison during pregnancy that my committee felt was susceptible to bias. Next slide.

So the method used by Martin et al was to use the number of weeks of pregnancy spent incarcerated. And so basically you'd take the--one way is to take the gestational age at delivery and subtract from that the gestational age at admission, or you can take the date of delivery and then subtract from that, you know, the date of admission, and you get the number of weeks of pregnancy spent incarcerated. And so that was their exposure variable, and their outcome variable was infant birth weight in graphs.

The problem in epidemiological language was a lack of independence between their exposure measurement and their outcome measurement. And in more straightforward language, here's the problem. So if you look at it this way, if two women enter prison at the same point in pregnancy and one delivers preterm and one delivers at term, the one who delivers preterm is necessarily going to have a lower number of weeks of pregnancy spent incarcerated because of the way you calculated. And so you're going to have a spurious association between the number of weeks of pregnancy spent incarcerated and preterm delivery. And then preterm delivery in turn is correlated with infant birth weight. And so the dose-response relationship reported by Martin et al may have been spurious. Next slide.

So now you might be asking, you know, why were we so focused on the dose-response relationship. And it's because of the potentially significant policy implications because--no, sorry, not yet--because if the real truth is that there really is a linear positive dose-response relationship between the quantity of exposure to prison during pregnancy and infant birth weight, then what that means is that no matter at what stage in pregnancy a woman enters the criminal justice system, it would always be better to rush her into prison as quickly as possible, if that were the truth. That would be the implication. And that's why we were really concerned about the, you know, about looking--about re-evaluating this dose-response relationship. Next slide, next one.

So the study officially of this paper, which as I said is the leadoff paper in my dissertation, was to examine whether there's a linear dose-response association between incarceration during pregnancy and infant birth weight when a less biased methodology for quantifying exposure to prison during pregnancy is employed. Next slide. So the study population consisted of pregnant women initiating incarceration in Texas state prisons. The study period was from January 1, 2002 , to December 31, 2004 . Conclusion criteria are that the woman had to have delivered within the study period. Exclusion criteria, I mean, the basic point of that is that we only looked at women who entered state prisons. We didn't look at jails or federal prisons or, you know, they're state jails, county jails, federal prisons. We focused exclusively on state prisons. Next slide.

Data sources. We utilized an electronic medical record system maintained by the Correctional Managed Care Division of the University of Texas Medical Branch , Galveston . We used paper-based delivery records from University of Texas Medical Branch , which is where all the inmates deliver. And we used paper-based records from the Department of Child Protective Services in Texas . Next slide.

Now, how did we choose to quantify exposure to prison during pregnancy? And what we chose to use instead was the gestational age at admission to prison. And the reason being is that we wanted to use a measure to quantify exposure to prison during pregnancy that didn't intrinsically rely on delivery. You see what I'm saying. And so you can estimate gestational age at admission to prison, you know, at the point of admission to prison. You don't have to know anything about when they deliver. In other words, delivery doesn't have to have occurred for you to be able to estimate the gestational age at admission to prison. Whereas with the number of weeks of pregnancy spent incarcerated, you can't determine that measure until delivery has occurred.

Next slide. So in terms of modeling, we used weighted linear regression. And we used weighted just because we took a census of all the deliveries to pregnant women in state prisons it 2003, but we only took a sample of the deliveries in 2004 and in 2002. So we took one-sixths of the births in 2002 and 2004 and a census in 2003. So we had 16 months worth of deliveries that we looked at. So we had to take into consideration the fact that in 2004 and in 2002, you know, not all the births were sampled. So that's why we used weighted linear regression.

We adjusted for maternal race, age, gravidity, years of school completed, history of tobacco use, history of substance abuse, history of alcohol use, and the presence of chronic disease at admission to prison. Gestational age at admission, which was our methodology--we actually modeled it two different ways. One is a categorical variable with cut points at gestation at week 13, 20, 27, and 34, and as a continuous variable subject to linear supply and transformation, again with cut points at 13, 20, 27, and 34. We mainly utilized the second methodology when we were constructing our curves. And the first methodology is--it's much easier to interpret. So that's why we used both methodology. I mean, categorical variables are much more easier to interpret when you present them, but when you're drawing curves, it's sometimes better to represent your exposure variable as a continuous variable, especially when you're looking for dose-response relationships.

Conceptual framework, very quickly, so you can see that, you know, we were looking at gestational age and admission to prison in relation to birth weights as one of our many analyses. And so we conceptualized the number of prison prenatal care visits and gestational age at delivery as potential mediating variables that we would want to adjust for if we saw a dose-response relationship using our measure. Next slide.

So going on to the results, these are some descriptive characteristics. So I have set up a table by the gestational age at admission to prison in categorical form. So we had 52 people--52 women who entered prison in the first trimester, 78 during the weeks 14 through 20, 74 weeks 21 through 27, 85 who entered weeks 28 to 34, and 78 who entered prison past week 34. The racial distribution within those categories--so for all the categories of gestational age and admission to prison, you know, whites were the majority except for women who entered prison during weeks 14 through 20. You can see that in that subgroup the majority--blacks represented the majority. But in every other subgroup, you know, whites represented the greatest portion of that subgroup.

In terms of some risk factors, chronic disease present at admission, the percentages are in parentheses, and you can see that, you know, this is a population obviously at very high risk. You can see that these factors, that chronic disease present at admission to prison, pre-incarceration, reported substance abuse, pre-incarceration STD history, I mean, you can see the rates are high all across the board regardless of when in gestation the women came in. Next slide.

Now, so these results are from our regression model. So infant birth rate births is the timing of incarceration during pregnancy or gestational age at admission to prison among female inmates in Texas state prisons who delivered between January 1, 2002 , and December 31, 2004 . In this model birth weight was modeled as a continuous variable in grams. And we represented gestational age at admission as four dummy variables, with weeks one through 13 being--so those women who came in first trimester would referred category.

And we constructed five models. And in each of the five models, this represents a different extent to which we had adjusted for covariant. In model one, that's the base model adjusted for nothing. In model five we're adjusted for all those factors which I said in a previous slide. And so focusing on model five, which is the most adjusted model we looked at, so you can see that none of the coefficients--so actually in parentheses at the top--it's a stagnant error; it's not the P-values. I didn't catch that. So you can see first of all that, I mean, for none of the coefficients were the--was there statistical significance. You know, if you look at the--if you look at--I mean what's supposed to be happening under the hypothesis of a positive dose-response relationship is that as you enter prison later and later, which means you're spending less and less time during your pregnancy in prison, then you are--then your birth weight should be less and less under that hypothesis. And so what should be happening is that the coefficients as you get from weeks 14 through 20 to 21 through 27, 28 through 34 and past 34, the coefficients--which again are all relative to the first trimester, should be getting more and more negative. And when you look at the coefficients, you don't see that. And when we conducted a linear trend test to see if these coefficients were consistent with a linear relationship, you can see the P-values were insignificant. Next slide.

Now graphically, so what we wanted to do was compare our methodology with the prior methodology graphically. So we utilized the methodology used by Martin et al on our data, and that showed in the dashed curve--so that's birth weight versus number of weeks of pregnancy spent incarcerated. And you can see we got a result extraordinarily similar to hers, you know. When you look at that curve, you say, wow, seems to be a linear positive dose-response relationship. But when you look at our curve using our methodology, gestational age at admission to prison, which is a solid curve, you can see that--and the solid curve is in reference to the top axis--you can see that really especially for women who came in after the first trimester, essentially flat. There's really no statistical evidence--and you know the previous slide is consistent with this--of a linear positive dose-response relationship between the quantity of exposure to prison during pregnancy and birth outcomes.

And then as an additional analysis we thought, well, maybe if we look at our--if we do our analyses separately by race, you know, who knows. Maybe within one particular race there might actually be a linear positive dose-response relationship and not in the others and then that race might be the one that's driving, you know, the overall trends. Next slide.

So we did that. So we stratified our analyses by race. Next slide. And so we developed these three curves. And you can see that, I mean, the main point of this curve is not the difference between the curve, but the fact that none of the curves are consistent with the idea of a linear positive dose-response relationship. Next slide.

So the conclusion was that among the current sample of incarcerated pregnant women in Texas , there was no linear positive dose-response relationship between exposure to prison during pregnancy as quantified by the gestational age of admission to prison and infant birth weight. And so we feel that there's a bias in using the number of weeks of pregnancy spent incarcerated to quantify exposure to prison during pregnancy. Next slide.

Some overall strengths of the study: It was the largest cohort of incarcerated pregnant women in any single study conducted so far in the United States . I mean we had a sample of 367 deliveries to 362 women, whereas in the Martin study, their group of incarcerated pregnant women, which was actually collected over four years, was 168. Use of medical record data as opposed to birth certificate data was a small strength. It wasn't as big as we expected just because medical record data is very dependent on sort of--I mean you find a lot more information, but it's not always consistently recorded by physician, and I mean, you know, these women were treated by different physicians depending on the time of year and, you know, when they came in, and so although we found a lot more information than we have gotten from birth certificate data, there were some--a lot of those extra variables that we found wasn't consistently recorded through the birth--through the medical records. And then another strength is what we believe is a less biased methodology of quantifying exposure to prison during pregnancy.

Overall limitations are equally important. From this study we can't draw conclusions regarding the comparison between incarcerated and nonincarcerated pregnant women. Most adjustment variables were self-reported. And we didn't have access to pre-incarceration medical records. So we weren't able to look at other factors like maternal weight gain because we weren't able to reliably determine what their pre-pregnancy weight was because all these women came in already pregnant. And things like for those women who came in after the first trimester, we weren't able to determine whether they got first trimester prenatal care because we didn't have access to their pre-incarceration medical records. So certain factors we really wanted to look at, we couldn't look at. Next slide.

And this just acknowledges all the people who helped me. And yep, that's about it. Thanks.