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

Measurement of Gestational Age: Challenges to Research and Surveillance

PATRICIA DIETZ: Thank you, Lucy. Bill Sappenfield asked the question earlier what happens if you have LMP and ultrasound for these women. And I'm going to spend the next ten minutes answering his question. I'll be doing a comparison of ultrasound and LMP based gestational age using 2002 California vital records and AFP screening program records that were linked and I'd like to acknowledge my co-authors Lucy England and Bill Callahan who are here today. Michelle Pearl and Megan Weir who are with the Sequoia Foundation and Marty Kharrazi who's with the California Department of Health and Human Services.

I'd like to start by talking about some of the potential errors using LMP and to me they fall into two categories. One is related to person errors such as the recall of dates. I think Joyce earlier today presented some pretty compelling data that this happens, that women can't remember and don't report maybe the day of their LMP or there's digit preference where the first day of the month and the 15 th day of the month are more likely to be reported. There could be misinterpretation of LMP. It's supposed to be the first day of your last menstrual period and women may not think of it in that way and you also might have data entry error. I also think about biological violations of the assumptions that we use when think about using LMP to calculate gestational age in that it's assumed that women will ovulate 14 days after the first day of their LMP. And we know that that isn't true for all women that some women will ovulate before and even more likely that some women will ovulate after 14 days. In addition, there could be breakthrough bleeding where a women thinks that's here menstrual period but it's not or there may be no LMP to report. A women could deliver a baby and become pregnant with the second without ever having a menstrual period. So those are some of the problems inherent when you use LMP to calculate gestational age.

Now, early ultrasound at less than 20 weeks was seen as maybe a better use of measuring gestational age. We know that it's better at predicting estimated date of delivery than LMP but there are some errors with ultra sound as well. Because ultrasound measures fetal size, infants that maybe biologically inherently smaller like females versus males will be dated a little earlier and studies suggest that for early ultrasound the effect is not great. It's about two to three days for those fetuses that would be more likely to be smaller. There's also non-differential error in that there's measurement error which is estimated to be plus or minus one week with early ultrasound and then again we have data entry error.

So the purpose of this study was to compare the gestational age estimates based on LMP with those based on early ultrasound and we assumed in this study that the early ultrasound was the gold standard despite some of the potential errors. For the data sources we took the LMP from the 2002 California birth certificate records and the ultrasound came from the California statewide prenatal care AFP screening records and these two data sources were linked using a probabilistic method and approximately 86 percent of women with and AFP screening between July 2001 and December 2002 linked to a 2002 birth record.

Just to give you some information on the AFP screening program it's statewide but participation is voluntary and it's offered to all women who receive prenatal care before or at 20 weeks of estimated gestational age. It's a triple marker test of screening and in order to interpret these tests you really need to have accurate gestational age and so on the form there's a lot of information that they collect from the physician when the blood is drawn from the woman. They ask for ultrasound. They ask for LMP. They ask for the clinicians' best estimate of age. And not all women participating in this program have had early ultrasound.

So we started out with 530,000 births in California and 327,000 of these linked to a prenatal screen. Of these 195,000 had an ultrasound and then we restricted this analysis to singleton births and then we excluded women or records where the LMP was missing or the gestational age was less than 20 weeks or greater than 44 weeks and also if the birth weight was less than 500 grams. So we ended up with 162,000 women in our sample.

To calculate the gestational age based on LMP we took the date of delivery and subtracted the date of the LMP to come up with days of gestation. For the ultrasound we had to do a few more steps. I mentioned that this form is filled out by a clinician when the blood is drawn and it includes the ultrasound estimate of gestational age at the date the ultrasound was performed. So we took that estimate and, sort of, figured out how old that infant would be and added two weeks so we came up with a false LMP and we took the date of delivery minus that LMP to come up with days of gestation.

So what I'm going to do is I'm going to describe the study population. I'm going to present the birth weight distributions by gestational age. You've seen some of these graphs already. I'm going to present the sensitivity of the LMP based gestational age using ultrasound as the gold standard. We also looked at the characteristics of women who had inconsistent estimates, meaning that the LMP estimate and ultrasound was greater or more than 14 days. And then I'm going to look at the preterm rates by LMP and ultrasound and sort of look at well, how do these errors affect what we see when we look at this data.

This shows you characteristics of the women in our sample compared to all the women who delivered singleton births but were not included in our sample and you can see that the race distribution looks pretty similar although for the Hispanics we have 48 percent and there were 51 percent in those that were excluded.

When we looked at age it's not surprising we find we have a greater percent in the 25 to 34 group and as I said the AFP screening is voluntary and so what you're seeing is women who would be more interested in getting this information. We also had a more educated group of women (inaudible) in the ones that were excluded.

This slide gives you an overall distribution of gestational age. The ultrasound is in yellow and what you can see is for LMP we ended up with a lot more post dates, a lot later gestational age and you can't see it as well in the preterm but we ended up with more preterm with LMP as well. You have a wider and more flatter distribution with LMP compared to the ultrasound.

Next, I'm going to present you all the birth weight by gestational age graphs. Here we have birth weight distribution for 20 to 27 weeks and what you see here is that for LMP you end up with a small second mode of some heavier infants and with ultrasounds you have what is a flat long tail of heavier infants and both of those tails and the second mode suggests some error going on with those measurements of gestational age. Here we have the graph for those 28 to 31 weeks. With LMP again we see that second mode that you've seen quite a lot today and with ultrasound we don't see that second mode suggesting that there's less error but we do see a tail, so there is still some error going on. For 32 to 36 weeks what we see with LMP is a wider more flat distribution and skewed to the right and the ultrasound is more to the left and it's a tighter distribution. For gestational age for 37 to 42 weeks what we see is agreement. Happy to see that. And then when we look at post term 42 to 44 weeks we see the ultrasound is slightly to the right with a little heavier birth weight compared to the LMP.

This slide just shows you the agreement if you're just looking at preterm or term using LMP and ultrasound and what we found is for 5.1 percent of the birth both ultrasound and LMP estimates agreed that they were preterm. We found that for when ultrasound said preterm and LMP said term that happened 2.8 percent and for an ultrasound said term and LMP said preterm it was 3.7 percent which was greater in the preterm area, so when you look at the overall percentages you get a lower percent of preterm if you use ultrasound at 7.9 percent and higher for LMP which was 8.8 percent.

This slide is looking at the sensitivity of LMP in predicting preterm using ultrasound as the gold standard. So if you go down the column for positive preterm delivery that is the column in which the ultrasound said these were preterm and we get 12,880. Of these LMP said 8,360 were preterm. So the sensitivity is 65.1 percent. If you want to know of all the preterms that LMP said these are preterm infants, what percent did ultrasounds say that they were preterm? That was 58.4 percent. So if you're doing a study and you're really trying to look at some risk factors for preterm delivery and using LMP you will have a lot of noise in your data that would lead to some error if we can take these findings and generalize it to other situations.

In this study we looked at the sensitivity and we broke it down looking at the 20 to 27 week, the 28 to 31 and 32 to 36 and what you see is that while the sensitivities a little bit better for the 20 to 27 weeks for all gestational age groups in the preterm area it is still quite low even in the 32 to 36 weeks where we don't have the ability to see a second hump to help us think about what's the error because of the way the distribution works out when you get closer to term but there's still a lot of error going on in the 32 to 36 week group.

This slide we put in here to give you a sense of were we right in saying ultrasound was the gold standard. What this slide shows is the mean birth weight by LMP and ultrasound estimates. So if you go down the first column that says 20 to 27. In that column the ultrasound said the infant was 20 to 27 weeks and in your first row the gestational age based on LMP also said 20 to 27 weeks and we get 810 grams. That's the mean of the group that fell in there. Now, if there was no error with either of them, if you go across that row we should have 810 grams in every cell or if we go down we should have 810 grams in every cell but we don't. And if you go across that row we see LMP says it's 20 to 27 weeks and you see as the ultrasound estimate of gestational age goes up so does the mean birth weight go up in that cell. It's also true that's going on when you look at ultrasound and you go down to the next cell where LMP says it's 28 to 31 weeks. Ultrasound says 20 to 27 but what you find is 940 grams is closer to 810 grams than 1,300 grams is when the LMP said it was 28 to 31. So what we found is, yes, there was noise with ultrasound. There was some error going on but this gave us some confidence that using ultrasound as the gold standard was an okay decision.

Next, we looked at the inconsistent LMP gestational and age and ultrasound estimates and we looked at those greater than 14 days absolute difference. We had 15.4 percent of women with inconsistent estimates, 4.1 percent of those the ultrasound had a higher estimate of gestational age than LMP and an 11 percent the LMP had a higher percentage than the ultrasound and we found not surprising that women who were less than 20 years of age who had less than 12 years of education, African/American women and Hispanic women and women who entered prenatal care not in the first trimester were more likely to fall in those categories.

Here I'm just comparing the ultrasound and the LMP rates and what we find is that when you're looking at 20 to 27 weeks the rate of preterm at that gestational age was similar between the two groups. When you looked at 28 to 31 weeks you start seeing that LMP was a little higher, .8 compared to .6 and when you look at the 32 to 36 weeks you get ultrasound of a 6.9 percent rate and LMP at 7.7 percent. What's also interesting is if you look at the post term dates is that you get a much higher rate when you use LMP.

Here we're looking at the preterm rates by race and ethnicity and what you see is that the rate among whites is actually quite similar and a little bit higher for the ultrasound group than the LMP group, 7.3 compared to 7.2 but when you compare African/Americans you get 10.8 for ultrasound and 12.3 for LMP and for Hispanics you see a similar trend. You get a higher rate when you use LMP compared to ultrasound and I think this is important when you think about when we do our comparisons of ratios of whites to blacks or Hispanics to whites that there is a difference but are we suggesting that the difference might be greater than it truly is.

I'd like to mention some of the limitations thinking about the results of this study. As I showed in the beginning this sample was slightly different from all singleton births in California and that's important because is the LMP era that we're finding in the sample greater than what it would be among all births? And if that is true then we would be saying the LMP era is higher than what it really is compared to ultrasound. And that's really very possible because women tend to get early ultrasound if they're uncertain about their dates. So it is possible that we had a group of women who had more uncertainty about their LMP than of all women who were giving birth in California . On the other hand the error could be less in this group because we had demographic differences that favored women who were more likely to have consistent LMP and ultrasound rates. So it's hard to speculate how these results can be generalized to all women in the vital records system.

So in summary, the ultrasound birth weight distributions differed from the LMP birth weight distributions. The ultrasounds did not have a bimodal distribution for less than 32 weeks and they had a tighter birth weight distribution for 32 to 36 weeks and it had less post term births. The groups with the highest inconsistent estimates included African/Americans and Hispanics, younger women, women with less education and those entering prenatal care later. Our overall sensitivity for preterm delivery using LMP was 65 percent and the predicted value positive was 58 percent and so as I mentioned this has real implications for people looking at risk factors. There's just a lot of noise when you're using LMP to determine preterm delivery. Among preterm groups the sensitivity was lowest in the 32 to 36 week group and as we heard this morning that is the group that's responsible for the majority of preterm deliveries. So we found that there was poor sensitivity in this group and in the mean birth weight supported ultrasound as the gold standard although ultrasound had evidence of some error as well. The ultrasound preterm rates were lower than the LMP base rates and it just raises a question are rates based on LMP inaccurately high? We also a different pattern by race. The white LMP and ultrasound preterm rates were very similar but the African/American and the Hispanic preterm rates were higher when they were based on LMP. So when we look at these ratios between different race and ethnicity groups are we even getting a greater bias? And that's it. Thank you.