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

Measurement of Gestational Age: Challenges to Research and Surveillance

LAURA SCHIEVE: Thank you, so I'm going to talk about a fairly unique group of infants and their gestational age estimates, those born after assisted reproductive technology.

First, ART is defined as treatments or procedures that include the handling of both eggs and sperm or embryos to establish pregnancy and here's a list of treatments considered ART. I'm not going to discuss each of these. I just want to make the point that it is a diverse set of treatments in itself and that it does not include assisted insemination only or use of ovarian stimulation mediations if there is no egg retrieval.

These are the basic steps in an ART treatment cycle. In most cases the patient is given medication to stimulate egg production. After a couple of weeks the eggs are retrieved, combined with sperm and the resulting embryos will be fertilized from one to six days. They'll then be transferred into the uterus and then hopefully implantation, pregnancy and live birth delivery will result.

In the United States ART clinics are mandated to report data for all ART procedures annually to CDC and we've been collecting and reporting these data since 1997 and because it is federally mandated we have a very high compliance rate.

Data reported to the system include patient medical history, some data on the procedure itself, the treatment outcome and if it was a pregnancy we do get data on pregnancy outcome. Now, because the ART providers don't typically care for patients throughout their pregnancy but only through about first trimester they must ascertain the pregnancy outcome through active follow up but still loss to follow up has been less than 1 percent annually and about 10 percent of clinics are selected each year for onsite data validation visits and we have consistently found that the discrepancy rates related to pregnancy outcome and really related to all variables are fairly low. And these are just one year of data validation findings from 2001. In this year 40 clinics were visited, 37 had reported one or more live births and we reviewed nearly all records from those clinics had a live birth and the discrepancy between what the clinic reported to CDC and what we were able to find documented in their medical record was less than 1 percent for number of infants born and there was some errors noted for birth weight and date of birth.

So my objective today is to review the birth weight distribution by gestational age for ART live birth deliveries that were reported to the system. In undertaking the analysis we began with some assumptions. First, in the ART system data on the exact date of fertilization is known and recorded thus we assume that our data errors for gestational age estimate would be substantially less than those based on LMP reporting. Now, we know it's not that the ART system is completely error free but these should be limited to abstracting and recording errors. Date of birth in particular must be obtained through active follow up recorded in the chart and then abstracted and reported to the surveillance system but in vital states as we've been discussing throughout this you also have recall and other errors with LMP. And then our second assumption is that the reporting and recording errors would be predominantly random and thus distributed in keeping with the gestational age distribution. So because you have so many more term births we would expect more term births misclassified as preterm than vice versa.

Between 1999 and 2001 nearly 300,000 ART cycles were initiated. From these we selected those that use the patients own eggs, freshly fertilized embryos and then progressed with successful embryo transfer resulted in a clinical pregnancy and live birth delivery and we excluded a rare complication of heterotypic pregnancies in which there was both an entopic and an intrauterine implantation simultaneously which does happen when you transfer multiple embryos and then we further limited to singleton pregnancies. We also excluded the small percent with missing or invalid gestational age and because we really wanted to get as clean gestational age data--with date of embryo transfer and then we excluded the small number that we're missing birth weight data. Gestational age was calculated as date of birth minus date of fertilization plus 14 such that it would be compatible with what the woman's theoretical LMP date would have been and as I've mentioned date of fertilization is recorded in the medical record and date of birth is obtained via active follow up with mothers or in some cases the obstetric provider and birth weight is obtained at the same time either in grams or pounds and ounces and we converted values to grams. We divided gestational age into four groups, 87 percent were classified as term.

Nearly 11 percent were moderately preterm, 1.3 percent were very preterm and 0.8 percent were very, very preterm with 20 to 27 weeks and here's our birth weight distribution and here we still see for 20 to 27 weeks a slight bimodal distribution and here I've taken some of the data from Patty's analysis and overlaid it and we see this is just from the analysis of LMP, not ultrasound. We see something similar but when we and I think we did this slightly different than you did. I think we chose 2,400 as the (inaudible) here, so the numbers might not match exactly. Right, so our LMP we found 15 percent that fit that second curve whereas ART it was less. It was 9.1 percent and then here's 28 to 31 weeks. Again, we see a slight bimodal distribution however, not as prominent as with the California LMP data and then here's those proportions. Now, the LMP data were twice as likely to be in the second curve than the ART. Okay. So here we go to 32 to 36 weeks. Much more normally distributed as we've been seeing. Again, though, just like she said with ultrasound data when we compare it with the LMP from California we know that the curve is a bit wider and shifted to the left and then here's our curve for 37 to 44 weeks and we didn't break down post term here and this time the curve almost exactly matches what would have been for the California of that same gestational period. And then here we looked at the birth weight distribution for ART infants at every week of gestation between 22 and 40 weeks and in these gestational ages there were at least 15 births in each of these ages.

This chart shows the mean and median birth weights and basically it confirms what we just saw. At the lower gestational ages there's a (inaudible) of the distribution as evidenced by the differential between the mean and median and that tends to then disappear. These are the percentage of births less than 1,500 grams by every week gestation and as you'd expect there's a lot more of them in the younger gestational ages but between 22 and 28 weeks there's really not much trend as the percentage of very low birth weight ranged between 83 percent to 95 percent. And then it began to decrease as expected reaching just one-tenth of a percent by 37 weeks.

Here I've added moderately low birth weight and then finally normal birth weight added on top and that's really the most surprising is that the percentage of normal birth weight births hovers are 10 percent at every gestational age between 22 and 32 weeks and in fact between 22 and 28 weeks you have a greater probability according to (inaudible) that you have a normal birth weight than a moderately low birth weight and in the interest of times I've cut out some signs but just to say if you look at this purple portion, if we looked at just all of the normal birth weight babies and we looked at the distribution of birth weight for normal birth weight, so above 2,500 grams you'd find a similar mean and median. You don't even really see that like the 22 weeks are closer to the 2,500 side whereas the, you know, 38 to 40 are closer to the 4,000 gram side. It's pretty flat lined and mean and median are exactly the same across that and it was even with getting to pretty sparse data.

So in summary the ART gestational age estimates between 20 to 27 weeks and 28 to 31 weeks showed bimodal distribution but it was a lot less than we saw with LMP based gestational age estimates from California vital statistics. Also, among the ART births the high proportion of normal birth weight even at very low gestational ages to us is suggestive of data errors and the pattern of results suggests that term birth were, indeed, more likely to be misclassified and that it appeared somewhat random is that the erroneous gestational ages appeared to have been distributed across the gestational age spectrum.

And just one limitation that I think we need to notice that we also obtained our birth weight from maternal report, so it is subject to some misreporting as well, however, erroneous birth weight I would think would also presumably be random and would not cluster as normal birth weight across gestational ages but would more likely--the birth weight errors is that normal birth weights were erroneously recorded and then evenly distributed and then finally the implication is that the bimodal distribution indicates some degree of data errors and some of that can even be due to reporting and recording differences and finally I'd just like to thank my collaborators.