American Journal of Epidemiology Advance Access originally published online on January 7, 2008
American Journal of Epidemiology 2008 167(6):646-652; doi:10.1093/aje/kwm345
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
PRACTICE OF EPIDEMIOLOGY |
Variation between Last-Menstrual-Period and Clinical Estimates of Gestational Age in Vital Records
From the Division of Reproductive Health, National Center for Chronic Disease Prevention and Health Promotion, Centers for Disease Control and Prevention, Atlanta, GA
Correspondence to Dr. Cheng Qin, National Center for Chronic Disease Prevention and Health Promotion, Centers for Disease Control and Prevention, 4770 Buford Highway NE, Mailstop K-23, Atlanta, GA 30341 (e-mail: caq9{at}cdc.gov).
Received for publication April 6, 2007. Accepted for publication November 1, 2007.
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONReferences |
|---|
An accurate assessment of gestational age is vital to population-based research and surveillance in maternal and infant health. However, the quality of gestational age measurements derived from birth certificates has been in question. Using the 2002 US public-use natality file, the authors examined the agreement between estimates of gestational age based on the last menstrual period (LMP) and clinical estimates in vital records across durations of gestation and US states and explored reasons for disagreement. Agreement between the LMP and the clinical estimate of gestational age varied substantially across gestations and among states. Preterm births were more likely than term births to have disagreement between the two estimates. Maternal age, maternal education, initiation of prenatal care, order of livebirth, and use of ultrasound had significant independent effects on the disagreement between the two measures, regardless of gestational age, but these factors made little difference in the magnitude of gestational age group differences. Information available on birth certificates was not sufficient to understand this disparity. The lowest agreement between the LMP and the clinical estimate was observed among preterm infants born at 28–36 weeks' gestation, who accounted for more than 90% of total preterm births. This finding deserves particular attention and further investigation.
gestational age; pregnancy
Abbreviations: LMP, last menstrual period; NCHS, National Center for Health Statistics
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONReferences |
|---|
An accurate assessment of gestational age is vital to research and surveillance in maternal and infant health. This assessment presents a particular problem when research and surveillance use population-based records. The quality of the gestational age measurement that is derived from the birth certificate, the primary source of population-based information on the status of infants at birth, has been in question (1).
There are two types of gestational age measurement in birth certificate files: gestational age based on maternal last menstrual period (LMP) and gestational age based on clinical estimation. The LMP-based estimate is subject to errors resulting from late ovulation, bleeding or spotting during early pregnancy, and erroneous recall of LMP. The source of a clinical estimate, on the other hand, is largely unknown and may include antenatal or newborn assessment.
Sources of error associated with LMP-based gestational age have been studied extensively. Various methods have been proposed for correction of errors in the measurement of LMP-based gestational age, such as exclusion of records with implausible birth weights for gestational age (2–4), but different methods produce different results (5). A composite measure using both the LMP and the clinical estimate of gestational age has been examined in which the LMP-based estimate is replaced with the clinical estimate when a discrepancy exists between the two (6). However, the composite measure is also in question due to the uncertain derivation of the clinical estimate.
Because there is no gold standard available for validating the different approaches using only birth certificate data, a good understanding of existing problems associated with current gestational age measurements in population-based data becomes important for researchers who are using these data. Our purpose in this analysis was to assess the agreement between LMP and clinical estimates in vital records across durations of gestation and US states according to a variety of characteristics and to explore the reasons for disagreement.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONReferences |
|---|
The 2002 US public-use natality file compiled by the National Center for Health Statistics (NCHS) Vital Statistics Cooperative Program was used for the analysis. Here, the gestational age variable that is coded in completed weeks is primarily based on the LMP. In 1989, the clinical estimate of gestational age was added to the US standard birth certificate to provide an additional measure of the length of pregnancy.
Conservative data edits are conducted by the NCHS to identify those records that have a high probability of containing erroneous data. During these data edits, the LMP-based estimate is replaced with the clinical estimate when the date of the LMP is unknown or when the birth weight is extremely high or low for the gestational age (see Appendix table). These records account for approximately 5 percent of the data. Details on the NCHS edits of gestational age data have been published elsewhere (7).
Our analysis was restricted to the 3,187,904 singleton livebirths occurring in 49 US states and the District of Columbia in 2002, after the following exclusions. California was excluded because information on the clinical estimate is not collected on its birth certificate. Records with missing data for birth weight (0.08 percent) and LMP-based gestational age (0.19 percent) were also excluded. Records with unknown or unstated clinical estimates (0.38 percent) were categorized as having missing values in the analysis. Those records for which the LMP-based estimate had been replaced with the clinical estimate by the NCHS, as described above, were also excluded (5 percent).
We examined the difference between the LMP estimate and the clinical estimate in the following ways. First, we plotted birth weight distributions for births in different gestational age groups (20–27, 28–31, 32–36, and 37–42 weeks). We present the birth weight distribution for 28–31 weeks' gestation as an illustrative example (figure 1), because at this gestational age, the bimodal birth weight distribution by the LMP-based gestational age measurement is the most prominent (1, 8). Using the national natality data, the birth weight distributions were unimodal at 20–27 weeks because of truncation of the data and at >34 weeks because the "second mode" merged with the main distribution (6).
|
Next, we examined the discrepancy between the LMP estimate and the clinical estimate using both gestational age and birth weight information. Given that birth weight is more accurately measured than gestational age and birth weight is not normally distributed at early gestational ages (1, 6, 8), we examined numbers of births and median birth weights in the four traditional gestational age groups (20–27, 28–31, 32–36, and 37–40 weeks) for the following three scenarios: 1) the numbers of weeks of gestation as determined by the LMP estimate and the clinical estimate were not the same; 2) the LMP estimate and the clinical estimate disagreed by >1 week; and 3) the LMP and clinical estimates disagreed by >2 weeks.
We then examined the degree of agreement between the two measures for each gestation week from 20 weeks to 42 weeks when the numbers of weeks of gestation by the LMP and the clinical estimate were the same; when the LMP and the clinical estimate were within 1 week of each other; and when the LMP and the clinical estimate were within 2 weeks of each other. In addition, the agreement between the two measures was examined separately for each US state at gestations of 28, 31, and 40 weeks. The main purpose of the examination by state was to evaluate the overall pattern of variation across the country, not to make a comparison between the states.
To further explore factors associated with the discrepancy between the LMP estimate and the clinical estimate, we examined the number of births and the percentage of disagreement by gestational age group and maternal characteristics, as well as ultrasound use during pregnancy. Maternal age was categorized as <20, 20–34, or 
35 years. Maternal education was categorized as less than or equal to high school versus any college education. Trimester of prenatal care initiation was categorized as first trimester (first–third months), second trimester (fourth–sixth months), or third trimester (seventh–ninth months), or no prenatal care/unknown. Birth order referred to the number of surviving liveborns and was characterized as 1, 2 or 3. Use of ultrasound was categorized as yes or no. Information on the date of the ultrasound examination is not included on the birth certificate.
Finally, to better understand the variation between the LMP-based estimate and the clinical estimate between the gestational age groups, we conducted Cox regression analysis. The purpose of this subanalysis was to examine rate ratios for disagreement between the gestational age groups, and the time was set to be a constant. The event was the disagreement between the LMP estimate and the clinical estimate; gestational age estimates were considered to disagree when they differed from each other by more than 2 weeks. The independent variable in the crude model was LMP-based gestational age group, in four categories (20–27, 28–31, 32–36, and 37–42 weeks (reference group)). Possible confounding factors that we examined in the adjusted model include maternal age, maternal education, initiation of prenatal care, order of livebirth, and use of ultrasound. Results from the models with and without missing values for each of the factors were very similar, so observations with missing values were excluded from the final models.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONReferences |
|---|
Among records for which the clinical estimate-based and LMP-based gestational ages disagreed, there were fewer clinical estimate-based records for births with <37 weeks' gestation and more clinical estimate-based records for births occurring at 37–42 weeks' gestation (table 1). Compared with the clinical estimates, median birth weights by the LMP-based estimates were higher for births occurring at <37 weeks' gestation and lower for births occurring at 37–42 weeks' gestation when the LMP and the clinical estimate disagreed by more than 1 or 2 weeks. When the LMP and the clinical estimate disagreed with each other, the median birth weight for births at 37–42 weeks' gestation by the LMP-based estimate (3,374 g) was almost the same as that by the clinical estimate (3,375 g).
|
Agreement between the two measures varied substantially by gestational age (figure 2). It was higher among the extremely preterm (20–27 weeks) and term (37–41 weeks) infants and was lower among the midgestation (28–36 weeks) infants.
|
Among the 25 states that had 50,000 or more singleton livebirths in 2002, the lowest and highest percentages of records with the same gestational age by the LMP- and clinical estimate-based measures were 26.0 and 45.9 at 28 weeks, 14.8 and 49.2 at 31 weeks, and 52.2 and 71.1 at 40 weeks. The overall agreement between the two measures was lowest at 31 weeks' gestation, higher at 28 weeks' gestation, and highest among the term infants, and this was true for each of the individual states. Similar results were found when the LMP estimate and the clinical estimate were within 1 or 2 weeks of each other and for states with fewer than 50,000 singleton livebirths.
The degree of disagreement between the LMP estimate and the clinical estimate varied substantially by maternal characteristics and ultrasound use during pregnancy and by gestational age group (20–27, 28–31, 32–36, or 37–42 weeks) (table 2). Higher disagreement was found at 28–31 weeks' and 32–36 weeks' gestation for each of the factors. Teenage pregnancy, low maternal educational attainment, late prenatal care initiation, high birth order, and no ultrasound use during pregnancy were associated with disagreement between the two measures. Among preterm infants (gestational age <37 weeks), the disagreement between the LMP estimate and the clinical estimate in women with no or unknown prenatal care was lower than that in women who began prenatal care during the second or third trimester.
|
Rate ratios from the Cox regression showed that, compared with 37–42 weeks' gestation, the disagreement between the LMP and clinical estimates was 8.39 times higher at 28–31 weeks, 5.64 times higher at 32–36 weeks, and 3.97 times higher at 20–27 weeks (table 3). In the model adjusting for maternal age, maternal education, prenatal care initiation, livebirth order, and ultrasound use during pregnancy, the rate ratios were 7.79, 5.40, and 3.84, respectively. The fact that there was only a slight difference between the rate ratios in the adjusted model and those in the unadjusted model suggests that the disagreement between the LMP and clinical estimates among the gestational age groups was not confounded by maternal risk factors or ultrasound use during pregnancy. However, when examined individually, each of these factors showed a significant independent effect, regardless of gestational age group (p < 0.001; results not shown).
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONReferences |
|---|
The difficulties associated with gestational age estimates have been addressed in numerous publications (8–16). However, our study is unique in two ways. First, we investigated and found a systematic disparity between the LMP estimate and the clinical estimate of gestational age for all of the states. Second, we examined the difference between the two estimates across gestational age groups and the effect of various potentially confounding factors. We found that agreement between the LMP estimate and the clinical estimate of gestational age varied substantially across durations of gestation. More importantly, preterm births were more likely than term births to have disagreement between the two estimates. We found that timing of prenatal care initiation, maternal educational level, ultrasound use, livebirth order, and maternal age were independently associated with disagreement between the LMP and clinical estimates, regardless of gestational age. However, controlling for these factors did not result in meaningful differences in the relative risk of disagreement between the two estimates by gestational age group.
In this study, we found that when the LMP- and clinical estimate-based gestational ages disagreed, the clinical estimate-based numbers of births and median birth weights were lower for preterm infants and higher for term infants in comparison with the LMP-based estimates. This suggests that the discrepancy between the LMP and clinical estimates was not random. The clinical estimate "corrected" the LMP in both directions. That is, when the LMP- and clinical estimate-based gestational ages disagreed, fewer or lighter infants were categorized as preterm and more or heavier infants were categorized as term by the clinical estimate than by the LMP. It suggests a closer association between the clinical estimate and the birth weight.
We also found that agreement between the LMP estimate and the clinical estimate was lowest among very preterm and moderately preterm infants (28–36 weeks), higher among extremely preterm infants (20–27 weeks), and highest among term infants (37–42 weeks). This pattern was observed for all of the states. Similar findings have been reported by Alexander et al. (15) and Mustafa and David (16). Using the 1989–1991 South Carolina public-use livebirth files, Alexander et al. found that the overall agreement between the LMP estimate and the clinical estimate was 47 percent, but it varied considerably by gestational age (15). Mustafa and David, using the 1989–1991 Illinois computerized birth record data, found that, for more than 40 percent of the cases in the range of 27–34 weeks, the clinical estimate exceeded LMP-based values by more than 2 weeks, and the highest agreement between the LMP estimate and the clinical estimate was 72 percent at 40 weeks (16).
Special attention should be given to the low agreement between the two measures at 28–36 weeks' gestation. In our data, based on the LMP-based estimate of gestational age, 94 percent of all preterm infants (<37 completed weeks) born in the United States in 2002 were born at 28–36 weeks' gestation. Of these infants, 11 percent were born at 28–31 weeks and 89 percent were born at 32–36 weeks. In studies using vital records, the estimated percentage of preterm births was lower when gestational age was based on the clinical estimate or the LMP/clinical estimate composite measurement than when it was based on the LMP (6, 17). An accurate assessment of the gestational age of these infants is important in epidemiologic and etiologic studies of preterm birth, which are crucial for developing effective prevention measures and interventions.
It is unclear why the discrepancy between the two measures was most prominent for the mid-gestational-age groups. It is possible that misrecall of the LMP and bleeding or spotting during early pregnancy occur more frequently in women of low socioeconomic status, and preterm birth is associated with low socioeconomic status (17). Our analysis did show that women with teenage pregnancy, low educational attainment, late initiation of prenatal care, high birth order, and no ultrasound use during pregnancy were more likely to have disagreement between the two measures among preterm births. The reasons for the better agreement at 20–27 weeks' gestation, however, remain unclear. Early ultrasound examination could be an explanation. Unfortunately, the birth certificate does not specify when during pregnancy the ultrasound examination was performed; thus, the effect of ultrasound use on the discrepancy between the two measures cannot be accurately assessed. Initial data editing at the NCHS, such as the exclusion of some records containing extremely implausible birth weights for gestational age, may also have some effect, but the similar findings from the two studies using state vital-record data (15, 16) suggest a limited effect of the NCHS initial data editing. The lower disagreement between the two gestational age measures for no and unknown prenatal care among preterm births may be due to the absence of any prior information other than the LMP. Hence, there was probably little clinical information upon which to base the clinical estimate.
The clinical estimate measure itself, if used appropriately, is the best obstetrical estimate of gestational age. It is based on the integration of various sources of information collected by a clinician. The accuracy of a gestational age derived from the clinical estimate on the birth certificate is not easy to evaluate, however, mainly because its influencing factors involve the skills and performance of the clinician conducting the estimate, the level of health care, and uncertainty regarding its source (prenatal examination or neonatal assessment). It is possible that errors associated with the LMP measure cause underestimation of gestational age, while problems associated with the clinical estimate cause overestimation, resulting in an increased discrepancy between the two measures at 28–36 weeks' gestation. However, the exact causes of the discrepancy remain unclear. A recent study (18) on preterm and postterm births using Canadian and US livebirth files found that the preterm birth rate in the United States derived using the clinical estimate was lower than that derived using the LMP and was closer to that in Canada. The rate ratios for perinatal mortality comparing preterm infants with term infants were comparable between the United States and Canada only when the clinical estimate was used (18). The fixed second birth weight distribution found in another recent study (6) with a peak incidence at approximately 3,000 g at early gestational weeks suggests that these births are probably term births. Although our study was not able to determine which of these two gestational age measurements gives more accurate estimates in vital records, the findings of low agreement between the LMP and clinical estimates of gestational age among infants at 28–36 weeks' gestation and relatively better agreement at 20–27 weeks' gestation deserve careful attention and further investigation.
There is no clear explanation for the high level of agreement between the two measures among term infants. Although misrecording of the LMP was a possible reason for the disagreement, errors in recording are likely to have been random given the large size of the data set and therefore uniform across gestational ages. As such, there is no reason to expect a particularly lower frequency of recording errors occurring among term deliveries. The little difference found in this study between the rate ratios in the adjusted and unadjusted models suggests possible other explanatory factors than what we can identify on the birth certificate.
Besides the disagreement between the two estimates across gestational age groups, we found substantial variation by state. It is interesting that our findings of low agreement around midgestation and higher agreement at lower gestation and among term babies were remarkably similar for all of the states. Clearly, understanding the variation across states requires more information than the birth certificate can provide. The composition of the population may have an impact on the differences in gestational age measures. For example, misclassification of gestational age has been shown to be greater among African Americans than among Whites (8, 9). Other systematic or organizational factors may also be operating, and the source of the clinical estimate may vary between hospitals and states. For example, some hospitals may rely more on early ultrasound examination for the clinical estimate than others. In their study using the South Carolina public-use livebirth files, Alexander et al. (15) found that hospital size and trimester of prenatal care were associated with differences between the LMP and clinical estimates.
In summary, we found substantial disagreement between LMP-based gestational age estimates and clinical estimates across gestational ages and among the states. Information available on the US birth certificate is not sufficient for us to understand this disparity. Comparison of data entered both on the birth certificate and in medical records, if possible, might provide information on factors such as irregularity of the menstrual period, use of medication that may interrupt or delay ovulation, the date of the ultrasound examination, and pregnancy complications such as early-pregnancy bleeding. This information could help to clarify the relation between correctly and incorrectly classified gestational ages, which would be particularly helpful in understanding the second mode distribution of births at early gestational ages. Examining a sample of representative hospitals might offer insight into how clinical estimates are performed and recorded and into variations in clinical estimates across hospitals. This could provide useful information for understanding the variations in clinical estimation of gestational age by state and region. The 2003 revised birth certificate has specific instructions for determining the clinical estimate that should help to standardize and ensure more reliable and consistent estimation of gestational age. Although no definite conclusion can be drawn directly from this study as to which gestational age estimate is less biased, we believe that further discussion will lead to in-depth investigations that may provide direct evidence.
APPENDIX TABLE. Plausible birth weight/gestation combinations used to edit records for the National Center for Health Statistics public-use data files
| |||||||||||||||||||||
|
ACKNOWLEDGMENTS |
|---|
The authors thank Dr. Lucinda J. England, Dr. William M. Callaghan, and Joyce A. Martin for reading the manuscript and making helpful suggestions.
Conflict of interest: none declared.
|
References |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONReferences |
|---|
- David RJ. The quality and completeness of birthweight and gestational age data in computerized birth files. Am J Public Health (1980) 70:964–73.
[Abstract/Free Full Text] - Kramer MS, Platt RW, Wen SW, et al. A new and improved population-based Canadian reference for birth weight for gestational age. Pediatrics (2001) 108:E35. (Electronic article).[CrossRef][Medline]
- Alexander GR, Himes JH, Kaufman RB, et al. A United States national reference for fetal growth. Obstet Gynecol (1996) 87:163–8.[CrossRef][Web of Science][Medline]
- Zhang J, Bowes WA Jr. Birth-weight-for-gestational-age patterns by race, sex, and parity in the United States population. Obstet Gynecol (1995) 86:200–8.[CrossRef][Web of Science][Medline]
- Parker JD, Schoendorf KC. Implications of cleaning gestational age data. Paediatr Perinat Epidemiol (2002) 16:181–7.[CrossRef][Web of Science][Medline]
- Qin C, Dietz PM, England LJ, et al. Effects of different data editing methods on trends in race-specific preterm delivery rates, United States, 1990 –2002. Paediatr Perinat Epidemiol (2007) 21:41–9.[CrossRef][Web of Science][Medline]
- National Center for Health Statistics. Computer edits for natality data, effective 1993. Instruction manual, part 12. (1995) Hyattsville, MD: National Center for Health Statistics.
- Vahratian A, Buekens P, Bennett TA, et al. Preterm delivery rates in North Carolina: are they really declining among non-Hispanic African Americans? Am J Epidemiol (2004) 159:59–63.
[Abstract/Free Full Text] - Suthutvoravut S, Hogue CJ, Guyer B, et al. Are preterm black infants larger than preterm white infants, or are they more misclassified? J Biosoc Sci (1989) 21:443–51.[Web of Science][Medline]
- Alexander GR, Tompkins ME, Altekruse JM, et al. Racial differences in the relation of birth weight and gestational age to neonatal mortality. Public Health Rep (1985) 100:539–47.
- Tentoni S, Astolfi P, De Pasquale AD, et al. Birthweight by gestational age in preterm babies according to a Gaussian mixture model. BJOG (2004) 111:31–7.[Web of Science][Medline]
- Joseph KS, Kramer MS, Allen AC, et al. Implausible birth weight for gestational age. Am J Epidemiol (2001) 153:110–3.
[Abstract/Free Full Text] - David RJ. Population-based intrauterine growth curves from computerized birth certificates. South Med J (1983) 76:1401–6.[Web of Science][Medline]
- Platt RW, Abrahamowicz M, Kramer MS, et al. Detecting and eliminating erroneous gestational ages: a normal mixture model. Stat Med (2001) 20:3491–503.[CrossRef][Web of Science][Medline]
- Alexander GR, Tompkins ME, Petersen DJ, et al. Discordance between LMP-based and clinically estimated gestational age: implications for research, programs, and policy. Public Health Rep (1995) 110:395–402.
- Mustafa G, David RJ. Comparative accuracy of clinical estimate versus menstrual gestational age in computerized birth certificates. Public Health Rep (2001) 116:15–21.
- Kramer MS, Seguin L, Lydon J, et al. Socioeconomic disparities in pregnancy outcome: why do the poor fare so poorly? Paediatr Perinat Epidemiol (2000) 14:194–210.[CrossRef][Web of Science][Medline]
- Joseph KS, Huang L, Liu S, et al. Reconciling the high rates of preterm and postterm birth in the United States. Obstet Gynecol (2007) 109:813–22.[CrossRef][Web of Science][Medline]
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  C. K. Shapiro-Mendoza Infants Born Late Preterm: Epidemiology, Trends, and Morbidity Risk NeoReviews, June 1, 2009; 10(6): e287 - e294. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. B. Morse, H. Zheng, Y. Tang, and J. Roth Early School-Age Outcomes of Late Preterm Infants Pediatrics, April 1, 2009; 123(4): e622 - e629. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||