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

Measurement of Gestational Age: Challenges to Research and Surveillance

CHENG QIN: Good morning. I'm going to present findings for my analysis on (inaudible) in preterm delivery by (inaudible) the effect of data editing. First, would like to acknowledge my co-workers: Patty Dietz, Lucinda England, Braden Callahan from the Division of Reproductive Health, CDC and Joyce Martin from National Center for Health Statistics.

Our analysis were based on U.S. Natality files 1990 through 2002. The data studied population are non-Hispanic white and the non-Hispanic African/American singleton live birth. In our analysis four data editing methods were applied. There are NCHS, Alexandra, John (inaudible) and LMP clinical estimate. Initial data edits were conducted at NCHS, the other three methods were subjected to these ideas. At the Alexandra method a statistical approach and clinical consultation were combined. First, gestational age distributions were examined for birth groups into 125 gram birth weight intervals. Gestational age values of plus or minus 2.5 standard deviations from the mean were considered initially as cut points for implausible data. Statistically find cut points were then modified by (inaudible) in conjunction with clinical consultation. Birth records (inaudible) were deleted.

The John and the (inaudible) method is based on the assumption that birth weight at each gestational week are normally distributed. The data added in (inaudible) procedures. First, normal probability of birth weight for each gestational week were plotted. Cut of points of birth weight was then selected one (inaudible) from the (inaudible) data began to deviate from the straight lines systematically. Finally, LMP gestational age was supplemented by clinical estimate for infants whose birth weight was greater than the cut off point at a given gestational age. Because of the criteria I'll only define for gestational age between 25 to 35 weeks. The cut point for 25th weeks were applied to gestational weeks of 20 to 24 in our analysis and the records with gestational age greater than 35 weeks were not modified. Records with implausible birth weight gestational age combinations and with missing clinical estimates were excluded.

The LMP clinical estimate method used the two pieces of information on gestational age estimate. If LMP and the clinical estimate are within two weeks together LMP was used otherwise clinical estimate was used. It is based on the assumption that while LMP and the clinical estimate agree with each other the gestational age is correct. If LMP and the clinical estimate do not agree clinical estimate is more reliable. Records with missing clinical estimate were deleted.

California was excluded from our 30 year analysis because of no clinical estimates. New Hampshire and Oklahoma were excluded from the 1990 data because these two states did not report Hispanic in 1990. Missing birth weight and the missing gestational age records were excluded. Records with gestational age less than 20 weeks were also excluded. Percent of records is excluded and replaced by different methods were calculated and the affect of data editing (inaudible) distribution was presented in graphs (inaudible) and methods 1990 through 2002 were reported and percent of change over time by methods and by gestational age stratum were evaluated.

This is a background slide which highlights several key findings from our preliminary analysis. First, it shows that mean distributions by gestational weight shifted toward right as expected. Here's the mean distributions that is birth weight increases as gestation weight increased. Second, unlike the mean distributions the second humps were fixed regardless of gestation. Suggesting they came from a similar population. Third, the second humps around 3,000 grams suggesting they were likely term births or a special group of term infants. Fourth, there were no clear second hump for gestation 25, 26 and 27 weeks which you can see is here, the blue one, the purple and the yellow. They all stopped around here. But we see the incline. We see this incline here for these three gestation weeks. It is because there was no births with birth weight greater than 3,000 grams before 27 weeks that NCHS (inaudible). We would see a second hump if the data was not truncated. Finally, this slide also shows clear second hump for gestation 32 and 33 weeks, which is this one.

This slide shows distribution of birth by gestational age group, 90 percent of births are term infants. It demonstrates that even the small proportion of term births given their absolute large number were moved to a lower gestational age group which is the minority in the whole population could be big enough to make their own peaks.

This slide shows percent of records excluded and replaced by different data editing methods for all gestation. NCHS data was used as the baseline data. Now, African/Americans were affected than white as you can see here, this is African/Americas, this is white, white/African Americans you can't--African/Americans are affected larger. And the last data were modified in 2002 and in 1990. This is 1990 and this 2002. The (inaudible) records with clinical estimate were excluded from the LMP clinical estimate method. The decrease in percent of records excluded by LMP clinical estimate methods indicate increase in the number records having clinical estimate.

This slide shows percent of records excluded and replaced by methods by gestational age stratum for African/Americans. The yellow bars represented records that were excluded and the red bars represent records that were replaces 28 and the 31 weeks were affected most regardless of methods. Alexandra's method had the highest percentage of excludence. As you can see here this is Alexandra's method.

LMP clinical estimate method had the highest percentage of replacement compared to the other two methods. The LMP clinical estimate method had proportionally largest effect on near term birth. Results from whites showed a similar pattern that's now shown here.

By using NCHS data birth weight was truncated at 3,000 grams. There were no records with birth weight greater than 2,200 grams in this group by Alexandra's method and no records-Alexandra's stops here-and no records with birth weight greater than 3,200 grams (inaudible) method which stops here. The long right tail by the LMP clinical estimate method was due to discrepancies between LMP and the clinical estimate. See here. The birth weight distribution was by NCHS data; this is 28 to 31 weeks. The second hump peaked around the 3,000 grams here. It is less pronounced by Alexandra's method and considerably less pronounced but still visible by (inaudible) method, the yellow one. When the LMP clinical estimate was used the second hump is no longer present, which is this line.

This is for 32 to 35 weeks birth. The birth weight distribution was nearly identical for NCHS Alexandra's method. These two methods are almost the same. It became a little tighter by (inaudible) method, the yellow one. When LMP clinical estimate was applied the distribution shifted to the left and become more tighter.

This slide shows preterm delivery trends from 1990 to 2002 by different data editing methods. The dotted lines represent African/Americans and the solid lines represent non-Hispanic whites. The black color represents NCHS method. Pink represent Alexandra's method. Blue represent John (inaudible) method and orange represent the LMP clinical estimate methods. Different data editing method had little effect on trend except one; the LMP clinical estimate method was applied for African/Americans. You'll see here this is for African/Americans otherwise the other three method were pretty much in terms of trend they are similar.

This table shows percent change of preterm delivery between 1990 and 2002 by methods. Percentage change were increased (inaudible) across methods. For African/Americans the change was slightly increased here by the LMP clinical estimate and decreased by other methods. All the changes were statistically significant at five percent reliable.

In summary, we found that by model distribution early gestation weeks disappeared when LMP clinical estimate method was applied. The second hump were fixed and peaked around 3,000 grams. Percent records excluded and replaced decline over time regardless of method suggesting data improvement. Preterm delivery rate non-Hispanic whites regardless of the methods used. Preterm delivery rates slightly increased for non-Hispanic African/Americans by the LMP clinical estimate method, decreased by other methods. Thank you.