a comparison of lmp-based and ultrasound-based estimates of gestasinal age

A comparison of LMP-based and ultrasound-based estimates ofgestational age using linked California livebirth and prenatalscreening recordsPatricia M. Dietza, Lucinda J. Englanda, William M. Callaghana, Michelle Pearlb, Megan L. Wierb and Martin Kharrazic

aNational Center for Chronic Disease Prevention and Health Promotion, Division of Reproductive Health, Centers for Disease Control and

Prevention, Atlanta, GA, bSequoia Foundation, La Jolla, and cCalifornia Department of Health Services, Genetic Disease Screening Program,

Richmond, CA, USA

Summary

Correspondence:Patricia M. Dietz, 4770 BufordHwy MS K-22, Atlanta, GA30341, USA.E-mail: [email protected]

Dietz PM, England LJ, Callaghan WM, Pearl M, Wier ML, Kharrazi M. A comparisonof LMP-based and ultrasound-based estimates of gestational age using linked Califor-nia livebirth and prenatal screening records. Paediatric and Perinatal Epidemiology 2007;21(Suppl. 2): 62–71.

Although early ultrasound (<20 weeks’ gestation) systematically underestimates thegestational age of smaller fetuses by approximately 1–2 days, this bias is relatively smallcompared with the large error introduced by last menstrual period (LMP) estimates ofgestation, as evidenced by the number of implausible birthweight-for-gestational age.To characterise this misclassification, we compared gestational age estimates based onLMP from California birth certificates with those based on early ultrasound from aCalifornia linked Statewide Expanded Alpha-fetoprotein Screening Program (XAFP).The final sample comprised 165 908 women. Birthweight distributions were plotted bygestational age; sensitivity and positive predictive value for preterm rates according toLMP were calculated using ultrasound as the ‘gold standard’.

For gestational ages 20–27 and 28–31 weeks, the LMP-based birthweight distribu-tions were bimodal, whereas the ultrasound-based distributions were unimodal, buthad long right tails. At 32–36 weeks, the LMP distribution was wider, flatter, andshifted to the right, compared with the ultrasound distribution. LMP vs. ultrasoundestimates were, respectively, 8.7% vs. 7.9% preterm (<37 weeks), 81.2% vs. 91.0% term(37–41 weeks), and 10.1% vs. 1.1% post-term (�42 weeks). The sensitivity of the LMP-based preterm birth estimate was 64.3%, and the positive predictive value was 58.7%.Overall, 17.2% of the records had estimates with an absolute difference of >14 days.The groups most likely to have inconsistent gestational age estimates included AfricanAmerican and Hispanic women, younger and less-educated women, and those whoentered prenatal care after the second month of pregnancy. In conclusion, we foundsubstantial misclassification of LMP-based gestational age.

The 2003 revised US Standard Certificate of Live Birth includes a new gestational ageitem, the obstetric estimate. It will be important to assess whether this estimateaddresses the problems presented by LMP-based gestational age.

Keywords: gestation, ultrasound estimate, LMP estimate, perterm rate, post-term rate.

Conflicts of interest:the authors have declared noconflicts of interest.

Introduction

Problems with the accuracy of gestational age com-puted by last menstrual period (LMP) on birth certifi-cates have been documented.1–5 Evidence of this

inaccuracy is illustrated by birthweight distributionsthat are bimodal at gestational ages <32 weeks, withthe modal birthweight of the second peak consistentwith that of term infants.1,4 Inaccuracy of LMP-based

62

©2007 Blackwell Publishing Ltd. No claim to original US government works. Paediatric and Perinatal Epidemiology, 21 (Suppl. 2), 62–71

mailto:[email protected]

gestational age can be caused by biologically associatederrors in menstrual cycles and by human error in recallor data entry.6,7 Inherent in estimating gestational agewith LMP is the assumption that all women have aregular 28-day menstrual cycle and ovulate 14 daysafter the first day of their LMP. However, becausetiming of ovulation varies, even with accurate recalland data entry of the LMP, estimates of gestational agebased on LMP can be inaccurate. For example, onestudy found that 10% of women had cycles <25 dayslong, 12% were between 31 and 35 days, and 3% were36 days or longer, while 5% were too irregular to say.8

Time from LMP to ovulation is more likely to belonger, as opposed to shorter, than 14 days, resulting inan overestimation of gestational age when usingLMP.9,10 Biologically associated error can also occur ifearly bleeding in pregnancy is thought to be menstrua-tion or if LMP is missing because of amenorrhoea.Clinicians are well aware of the shortcomings of LMP,and standard practice is to base gestational age esti-mates on early ultrasound (<20 weeks) or other factorswhen LMP is uncertain. In addition, clinicians fre-quently substitute ultrasound-based gestational ageestimates for LMP-based estimates when the twodisagree.11

However, while early ultrasound has been estab-lished clinically as the gold standard, questions havebeen raised as to its validity for use in research. Onecommon concern is that ultrasound may introducebiases because it is based on fetal growth, and thuscould systematically result in the assignment ofincorrect lower gestational age estimates for smallerinfants.12,13 Recent studies have found that early(<20 weeks) ultrasound-based gestational age formulasare fairly accurate, with random errors of �10 days[95% confidence interval (CI)].14 In addition, fetuseswith characteristics associated with small fetal size,such as first births and female sex, were found to besystematically dated 1–2 days younger.15,16 Anotherlarge study of singleton pregnancies with ultrasoundexaminations between 12 and 22 weeks found no evi-dence that growth-restricted fetuses were systemati-cally classified incorrectly at lower gestational ages,and that the discrepancy between the LMP-basedgestational age and the ultrasound-based gestationalage was primarily related to ovulation later than theassumed 14 days.17 Thus, while early ultrasound maysystematically underestimate gestational age forsmaller fetuses by 1–2 days on average, this bias isrelatively small compared with the large magnitude of

error indicated by records with implausiblebirthweight-for-gestational age based on LMP.4,9,18 Pre-vious studies comparing LMP-based and ultrasound-based gestational age have found high rates ofgestational age misclassification by LMP. However,these studies have been limited to clinic- or hospital-based samples,9,12,18 to women with reliable LMPdates,12 or to studies outside the US.12,18 Therefore,whether the findings of these studies can be genera-lised to other populations is unknown.

We sought to better understand and characterise themisclassification found with gestational age estimatedby using LMP from birth certificates. Because USbirth certificates do not include information on earlyultrasound, we compared gestational age estimatesbased on LMP from California birth certificates withgestational age estimates based on early ultrasound(�20 weeks’ gestation) from a population-based prena-tal screening programme in which approximately 70%of the State’s pregnant women participate. Unlikeprevious studies, this inquiry benefited from a largesample derived from the cohort of women who deliv-ered in California in 2002.

Methods

The study population was defined as pregnant womenenrolled in the Statewide California ExpandedAlpha-fetoprotein Screening Program (XAFP) whogave birth to a live singleton infant during 2002, andwho had an estimated gestational age based on ultra-sound recorded on their XAFP screening form. TheXAFP is a triple marker screening programme offeredto all women entering prenatal care by 20 weeks’ ges-tation. When maternal blood is drawn for this screen,the medical provider fills out a form dating the preg-nancy based on LMP, physical examination, and/orultrasound, when available. Using SuperMatch 2001software (SuperMATCH Concepts and ReferenceVersion 3.10, Vality Technology Incorporated, March2001), a probabilistic method was employed to linkrecords from the XAFP and Statewide NewbornScreening programmes and birth certificates usingmother’s name, date of birth, social security number,delivery date, XAFP accession date, telephone number,street address, city and zip code. A conservative cer-tainty cut-off was used to minimise false matches.

In 2002, there were 530 926 livebirths in California.Of these, 327 218 livebirth records (62%) linked to anXAFP record from the same pregnancy, with approxi-

LMP and ultrasound estimates of gestational age 63


mately 86% of XAFP records successfully linking to alivebirth record. Failure to link records may haveresulted from data entry errors, pregnancies that didnot end in a livebirth, or women who moved out ofState before delivery. Among the linked records,195 616 (59.8%) women had ultrasound reported onthe XAFP records. After excluding 3238 women withmultiple births, 192 378 women were eligible for thestudy. Of these, we excluded records missing LMP onthe birth certificate (n = 26 249) or with gestational ageat birth of <20 weeks by either LMP (n = 206) or ultra-

sound (n = 30). The final sample comprised 165 908women (50.7% of livebirth records linked to an XAFPrecord, 32.2% of livebirths in California in 2002,Table 1).

LMP-based gestational age at delivery was calcu-lated using LMP and date of birth from the birth cer-tificate. Ultrasound-based gestational age at deliverywas calculated using the ultrasound-based estimate ofgestational age on the date the ultrasound was per-formed, and the date of delivery on the birth certificate.We categorised the two gestational age variables into

Table 1. Maternal demographic andpregnancy characteristics by studyeligibility and inclusion status, Californialivebirths, 2002 Characteristic

Eligible and included(n = 165 908)

%

Ineligible or excluded(n = 349 481)

%

Race/ethnicityWhite 31.8 30.2African American 5.4 6.1Asian 7.8 6.8Hispanic 48.3 51.0Other 6.7 5.8

Age (years)<20 7.4 10.920–24 21.0 24.625–34 58.3 46.9�35 13.2 17.6

Education (years)<12 24.9 31.112 28.1 29.0>12 47.0 39.9

Payment source (delivery)Medi-Cal 36.1 46.7Private 62.2 47.8Uninsured 1.1 2.9Other 0.6 2.5

Month prenatal care began1–2 76.1 65.73–4 22.1 24.35–6 NA 6.4�7 NA 3.6

Parity0 40.6 38.61 32.8 31.3�2 26.5 30.1

Infant birthweight (g)<2500 4.9 5.0�2500 95.1 95.0

LMP-based gestational age (weeks)<37 8.7 9.037–41 81.2 84.242–44 8.0 5.6�45 2.1 1.2

LMP, last menstrual period; NA, not applicable.Due to rounding or missing values totals may not add up to 100%.

64 P. M. Dietz et al.


five groups based on completed weeks: 20–27, 28–31,32–36, 37–41 and �42 weeks.

To explore predictors of inconsistent gestational age,we obtained infant birthweight, race/ethnicity, moth-er’s age, education, source of payment for delivery,and month of entry into prenatal care from the birthcertificate.

We first compared the birthweight distributions foreach gestational age group using LMP-based andultrasound-based gestational age estimates. We alsocalculated the sensitivity and positive predictive valueof the LMP-based gestational age, using ultrasound asthe gold standard. We compared the mean birthweightand whether the infant was placed in a neonatal inten-sive care unit (NICU) for estimates that were concor-dant and discordant for gestational age. For thisanalysis only we divided the group of 20–27 weeks into

two gestational age categories (20–23 and 24–27 weeks)to more closely examine differences. The NICU vari-able was obtained from the Statewide NewbornScreening programme database, and indicates whetherthe infant was in a NICU at the time of specimen col-lection (median time between delivery and specimencollection, 29 h). We compared the demographic char-acteristics of women with inconsistent ultrasound- andLMP-based gestational age estimates. We definedinconsistent as an absolute difference >14 days andused this cut-off to identify gross errors in gestationalage. All demographic characteristics were entered intoa logistic regression model to assess the independenteffects of each risk factor on inconsistent estimates,holding the other characteristics constant. Finally, wecalculated preterm delivery rates for LMP- andultrasound-based estimates overall, and by race/

Figure 1. Birthweight distribution of singletonbirths delivered at 20–27 weeks’ gestationaccording to ultrasound (n = 733) and lastmenstrual period (LMP) (n = 745).

% ofbirths

Birthweight (g)

0

10

20

30

40

400 1200 2200 3000 3800

Ultrasound

LMP

Figure 2. Birthweight distribution of singletonbirths delivered at 28–31 weeks’ gestationaccording to ultrasound (n = 1091) and lastmenstrual period (LMP) (n = 1235).

% ofbirths

Birthweight (g)

05

10152025303540

400 1200 2000 2800 3600 4400

Ultrasound

LMP

Figure 3. Birthweight distribution of singletonbirths delivered at 32–36 weeks’ gestationaccording to ultrasound (n = 11 410) and lastmenstrual period (LMP) (n = 12 499).

% ofbirths

Birthweight (g)

0

10

20

30

40

400

1000

1600

2200

2800

3400

4000

4600

Ultrasound

LMP



ethnicity, age, parity, education, month of entry intoprenatal care, and infant’s sex.

Results

Women included in the sample differed from those notincluded in that they were disproportionately aged25–34 years, more educated and less likely to haveMedi-Cal (California’s Medicaid programme)(Table 1). The women included were also more likely tohave begun prenatal care in the first 2 months of preg-nancy and were more likely to have delivered post-term (�42 weeks’ gestation) based on LMP. The two

groups were similar in racial/ethnic and low birth-weight rates.

Figures 1–3 present the birthweight distributionby LMP- and ultrasound-based gestational age. ForLMP-based gestational ages 20–27 weeks (Fig. 1) and28–31 weeks (Fig. 2), the birthweight distribution isbimodal, whereas the distribution based on ultrasoundgestational age is not, but it has a long right tail. ForLMP-based gestational age 32–36 weeks (Fig. 3), thebirthweight distribution is wider, flatter, and shiftedto the right compared with the ultrasound-baseddistribution. For both LMP- and ultrasound-basedgestational ages 37–41 weeks (figure not shown),

Table 2. Sensitivity and positive predictivevalue of last menstrual period estimate ofgestational age using ultrasound estimatesas the gold standard, total studypopulation and by racial/ethnic groups

Gestational age (weeks)Sensitivity% [95% CI]

Positive predictive value% [95% CI]

All women<37 64.3 [63.5, 65.1] 58.7 [57.9, 59.5]20–27 76.9 [73.9, 80.0] 75.7 [72.6, 78.8]28–31 60.4 [57.4, 63.5] 49.9 [47.1, 52.7]32–36 57.6 [56.7, 58.5] 52.8 [51.9, 53.7]37–41 85.6 [85.5, 85.8] 95.9 [95.8, 96.0]�42 33.6 [31.5, 35.8] 3.6 [3.3, 3.9]

White<37 66.8 [65.3, 68.3] 68.8 [67.3, 70.3]20–27 74.0 [67.1, 80.9] 76.0 [69.2, 82.8]28–31 64.9 [59.2, 70.6] 62.9 [57.2, 68.6]32–36 62.3 [60.7, 63.9] 64.4 [62.8, 66.0]37–41 88.0 [87.7, 88.3] 96.2 [96.0, 96.4]�42 38.8 [35.3, 42.3] 5.8 [5.2, 6.4]

African American<37 71.8 [69.0, 74.6] 63.7 [60.9, 66.5]20–27 76.9 [69.0, 84.8] 83.0 [75.6, 90.4]28–31 58.7 [49.9, 67.5] 52.6 [44.2, 61.0]32–36 61.1 [57.6, 64.6] 52.8 [49.5, 56.1]37–41 83.7 [82.9, 84.5] 95.2 [94.7, 95.7]�42 28.5 [20.7, 36.3] 3.9 [2.7, 5.1]

Hispanic<37 60.9 [59.7, 62.1] 52.2 [51.1, 53.3]20–27 77.2 [73.0, 81.4] 72.5 [68.2, 76.8]28–31 56.6 [52.3, 60.9] 41.7 [38.0, 45.4]32–36 53.0 [51.7, 54.3] 45.8 [44.6, 47.0]37–41 83.4 [83.1, 83.7] 95.6 [95.4, 95.8]�42 29.5 [26.3, 32.7] 2.5 [2.2, 2.8]

Asian<37 68.9 [65.8, 72.0] 62.6 [59.5, 65.7]20–27 86.4 [76.3, 95.5] 80.8 [69.5, 92.1]28–31 62.9 [50.9, 74.9] 68.4 [56.3, 80.5]32–36 64.2 [60.8, 67.6] 57.4 [54.0, 60.8]37–41 89.1 [88.5, 89.7] 97.1 [96.8, 97.4]�42 27.5 [18.8, 36.2] 2.7 [1.7, 3.7]



birthweight distributions overlap and appear normallydistributed. For LMP-based gestational age �42 weeks(figure not shown), the birthweight distribution iswider, flatter, and shifted to the left, compared with theultrasound-based distribution.

According to LMP-based gestational age estimates,8.7% of the infants were preterm (<37 weeks), 81.2%were term (37–41 weeks) and 10.1% were post-term(�42 weeks). In comparison, according to ultrasound-based estimates, 7.9% of the infants were preterm,

Table 3. Mean birthweight and NICU admissions by cross-tabulation of LMP-baseda and ultrasound-basedb gestational age estimates

LMPgestational age (weeks)

Ultrasound

20–23 24–27 28–31 32–36 37–41 42–44 �45 Total

20–23n 134 32 6 9 25 0 0 206Mean birthweight (g) 481 810 d d 3 437 d d 996(SD) (109) (526) (485) (1035)% NICU c 88 83 67 8 0 0 66

24–27n 55 343 59 23 58 1 0 539Mean birthweight (g) 590 857 1291 2 164 3 409 d d 1 213(SD) (255) (281) (430) (531) (516) (900)% NICU 96 98 97 56 2 0 0 84

28–31n 8 107 616 218 286 1 0 1 235Mean birthweight (g) d 915 1395 2 211 3 347 d d 1 946(SD) (257) (364) (524) (435) (942)% NICU 100 98 97 63 3 0 0 69

32–36n 0 17 272 6568 5 560 30 2 12 449Mean birthweight (g) d 953 1595 2584 3 286 3677 d 2 876(SD) (212) (391) (522) (482) (391) (643)% NICU 0 100 93 40 5 0 0 25

37–41n 8 12 40 4245 129 218 1173 12 134 708Mean birthweight (g) d 1482 1760 2 873 3 453 3811 3545 3 437(SD) (1220) (828) (495) (456) (458) (467) (471)% NICU 38 80 85 18 3 7 0 4

42–44n 1 4 15 195 12 539 538 1 13 293Mean birthweight (g) d d 1944 2 865 3 522 3828 d 3 522(SD) (923) (554) (463) (485) (482)% NICU c 75 93 21 3 6 0 4

�45n 5 7 11 152 3 225 76 2 3 478Mean birthweight (g) d d 1309 2 792 3 512 3831 d 3 470(SD) (273) (550) (468) (571) (538)% NICU 100 100 91 24 3 7 0 5

Totaln 211 522 1019 11 410 150 911 1818 17 165 908Mean birthweight (g) 620 887 1463 2 691 3 454 3815 3441(SD) (593) (367) (433) (536) (459) (470) (547)% NICU 89 97 95 32 3 6 0

aLMP from birth certificate.bUltrasound from XAFP screening form.cMissing data.dBirthweight means were not calculated for n < 10.LMP, last menstrual period; NICU, neonatal intensive care unit.



91.0% were term and 1.1% were post-term. Usingultrasound as the gold standard, the overall sensitivity(the percentage of true preterm deliveries correctlyidentified by LMP) was 64.3%, and the positive predic-tive value (the percentage of those found to be pretermby LMP that were ‘true’ preterm) was 58.7% (Table 2).The sensitivity and positive predictive value werehigher for the gestation group of 20–27 weeks than theother preterm groups. They were lowest for the post-term group, with a sensitivity of 33.6% and positivepredictive value of 3.7%. When stratified by race/ethnicity, Hispanics had the lowest sensitivity andpositive predictive value for less than 37 weeks.Whereas whites had similar sensitivity and positivepredictive value, Hispanics and African Americanshad lower positive predictive value than sensitivity,meaning that the number of infants falsely identified aspreterm using LMP estimates exceeded the number oftrue preterm infants missed by these estimates.

In order to evaluate ultrasound as a measure of ges-tational age versus LMP, we compared the mean birth-weights of infants with gestational age estimates thatwere concordant and discordant, using LMP and ultra-sound (Table 3). Among discordant gestational age

groups, mean birthweights categorised by ultrasoundwere closer to the mean birthweights of infants withconcordant estimates than those categorised by LMP.In Table 3, mean birthweights for gestational age cat-egories as determined by ultrasound (columns) weremore similar to one another than were mean birth-weights for gestational age categories determined fromthe LMP (rows). However, some misclassificationamong infants <37 weeks’ gestation based on ultra-sound was apparent (potentially due to clerical error),as mean birthweights increased with increasing LMP-based gestational age among infants with discordantestimates. Examination of percentage of infants in theNICU showed that ultrasound estimates of gestationalage were more consistent with what would beexpected. Preterm gestational age groups determinedby ultrasound had a higher percentage of infants in theNICU than did those determined by LMP.

Overall, 17.2% of gestational age estimates had anabsolute difference of >14 days between the twosources (Table 4); for 4.0% of the records, theultrasound-based estimate was greater than the LMP-based estimate and for 13.2% the LMP-based estimatewas greater than the ultrasound-based estimate.

Table 4. Proportion of women withinconsistenta estimates of gestational ageand adjusted odds ratios for inconsistentestimates by selected maternal andpregnancy characteristics

Inconsistent (%) Adjusted ORb [95% CI]

Race/ethnicityWhite 13.2 ReferenceAfrican American 19.0 1.3 [1.2, 1.4]Asian 13.8 1.1 [1.0, 1.2]Hispanic 20.5 1.3 [1.2, 1.4]Other 14.4 1.1 [1.0, 1.2]

Age (years)<20 22.5 1.7 [1.6, 1.8]20–24 20.9 1.6 [1.5, 1.7]25–34 16.0 1.3 [1.2, 1.3]�35 13.5 Reference

Education (years)<12 22.4 1.4 [1.3, 1.4]12 19.0 1.2 [1.2, 1.3]>12 13.3 Reference

Month prenatal care began1–2 15.3 Reference3–4 22.0 1.5 [1.4, 1.5]

Parity0 15.5 Reference1 17.1 1.2 [1.1, 1.2]�2 19.8 1.3 [1.3, 1.4]

aInconsistent is >14 days absolute difference between LMP estimate and ultrasoundestimate.bAdjusted for all characteristics simultaneously.



African American and Hispanic women comparedwith white women had a greater percentage of recordswith inconsistent LMP and ultrasound gestational age,as did women aged <35 years compared with theirolder counterparts, women with fewer years of educa-tion compared with women with �13 years, andmultiparae compared with primiparae. Women who,according to birth certificate records, entered into pre-natal care in the third or fourth month of pregnancyhad infants with higher rates of inconsistent estimatescompared with women who entered in the first orsecond month. We found the same groups of womenwith higher inconsistent estimates when we stratifiedby LMP estimate > ultrasound estimate and LMPestimate < ultrasound estimate (data not shown).

Preterm delivery rates differed according to mater-nal characteristics when using LMP and ultrasound(Table 5). For example, the odds ratio (OR) for pretermdelivery for African American infants compared withwhite infants was higher for LMP-based at 1.8 [95% CI1.7, 1.9] than for ultrasound-based gestational age esti-

mates at 1.5 [95% CI 1.4, 1.6]. A similar pattern wasfound for education. The OR of 1.2 for preterm deliveryfor male infants compared with female infants was thesame for LMP-based and ultrasound-based gestationalage estimates, and thus there was no evidence thatgestational age based on ultrasound resulted in higherpreterm rates among fetuses known to be smaller, suchas females.

Discussion

Using ultrasound-based gestational age as the goldstandard, this study found evidence of misclassificationof gestational age based on LMP. We found a greaterpercentage of false preterm infants, resulting in inflationof the preterm delivery rate. In addition,AfricanAmeri-cans and Hispanics had a greater percentage of recordswith misclassified gestational age than white women,resulting in inflated racial/ethnic disparities in pretermrates. The same pattern was found for women with less

Table 5. Preterm ratesa and UORs usingLMP- and ultrasound-based gestational ageestimates for selected characteristics

Characteristic

LMP Ultrasound

Preterm rate UOR [95% CI] Preterm rate UOR [95% CI]

Race/ethnicityWhite 7.3 Reference 7.3 ReferenceAfrican American 12.5 1.8 [1.7, 1.9] 10.8 1.5 [1.4, 1.6]Asian 7.3 1.0 [0.9, 1.1] 6.5 0.9 [0.8, 1.0]Hispanic 9.8 1.4 [1.3, 1.4] 8.1 1.1 [1.1, 1.2]Other 9.2 1.3 [1.2, 1.4] 8.4 1.2 [1.1, 1.2]

Age (years)<20 10.8 1.0 [0.9, 1.1] 9.0 1.0 [0.9, 1.0]20–24 8.7 0.8 [0.8, 0.9] 7.6 0.8 [0.8, 0.9]25–34 8.3 0.8 [0.7, 0.8] 7.6 0.8 [0.8, 0.8]�35 10.6 Reference 9.3 Reference

Education (years)<12 10.4 1.4 [1.3, 1.4] 8.3 1.1 [1.1, 1.2]12 9.4 1.2 [1.2, 1.3] 8.3 1.1 [1.1, 1.2]>12 7.8 Reference 7.5 Reference

Month prenatal care began1–2 8.9 Reference 8.1 Reference3–4 8.6 1.0 [0.9, 1.0] 7.3 0.9 [0.9, 0.9]

Parity0 8.8 1.1 [1.0, 1.1] 8.3 1.2 [1.2, 1.3]1 8.0 Reference 7.0 Reference�2 10.0 1.3 [1.2, 1.3] 8.4 1.2 [1.2, 1.3]

Infant genderFemale 8.2 Reference 7.2 ReferenceMale 9.6 1.2 [1.1, 1.2] 8.6 1.2 [1.2, 1.3]

aRates are limited to gestational ages between 20 and 44 weeks.LMP, last menstrual period; UOR, unadjusted odds ratio.



education. The birthweight distributions for gestationalages <32 weeks were bimodal when based on LMP butunimodal when based on ultrasound. While concernshave been raised that ultrasound-based gestational ageresults in misclassification of fetuses smaller thanexpected, we found no evidence of this bias in ourstudy, as the ORs of preterm delivery for male infantscompared with female infants were the same for LMP-and ultrasound-based gestational ages. The majority ofinconsistent estimates for LMP-based post-term infantswere of gestational ages greater than those arrived at byultrasound. This is consistent with our knowledge thatovulation is more likely to occur later than the assumed14 days after the first day of the LMP rather than earlier.

Our finding that the ultrasound-based gestationalage distribution had fewer post-term deliveries is con-sistent with those of other studies.9,18 However, find-ings regarding preterm rates are not consistent: onestudy found higher preterm rates using ultrasoundestimates,18 another found preterm rates to be similarbetween ultrasound and LMP estimates,9 while wefound higher preterm rates with LMP-based estimates.These inconsistent findings suggest that the amountand type of error in LMP-based gestational age canvary depending upon characteristics of the sample anddata collection methods. Some types of error, such asdelayed ovulation, result in overestimation of gesta-tional age, whereas poor recall could cause error ineither direction. The predominant direction of the errorwill determine an overall under- or overestimation ofgestational age compared with the ‘true’ estimate. WithLMP, it is likely that more than one type of error isaffecting the estimate of gestational age and contribut-ing to bidirectional misclassification.

Our study benefited from a large sample size thatincluded a subpopulation of women from the cohortwho gave birth in California in 2002. While this studypopulation may be more representative and have morestatistical power than those based on hospital or clinicsamples,9,12,13,18 characteristics of women included inour sample differed from those not included in severalimportant ways. Our sample included more womenwith post-term gestational age based on LMP, which isa marker for poor dating. Women screened in theXAFP programme who received ultrasound weremore likely to have had post-term LMP dates thanthose who did not receive ultrasound, suggesting thatuncertain dates might have been an indication forthe ultrasound. Therefore, the LMP-based dates forwomen included in our sample may be less reliable

than the LMP-based dates for the general population.If so, the misclassification rate of gestational age fromLMP could be lower in the general population thanfound in this study.

On the other hand, women in our sample were moreeducated, less likely to have Medicaid coverage, older(with the exception of women aged �35 years, who areeligible for amniocentesis without XAFP screening),and entered prenatal care earlier on average thanexcluded women, attributes associated with morereliable LMP dates. It is reassuring that LMP-basedpreterm rates among included and excluded womenwere similar, suggesting that the misclassification ofLMP-based gestational age among preterm infants mayindeed be representative of the general population ofCalifornia. Finally, while we assumed ultrasound to bethe gold standard when estimating sensitivity andpositive predictive value, we found some evidence oferror in ultrasound-based gestational age estimates.Therefore, the sensitivity and positive predictive valueof LMP-based gestational age may be higher for theentire cohort of infants in California than described inour sample.

In conclusion, our study provides evidence that asubstantial amount of misclassification results whenusing LMP-based gestational age estimates, and thismisclassification can lead to inflated preterm deliveryrates. In addition, differences in preterm delivery ratesbetween whites and African Americans, and betweenwhites and Hispanics, can also be inflated. Includingultrasound-based estimates of gestational age on thebirth certificate would help to improve the accuracy ofpreterm delivery rates, yet not all women receive anultrasound before 20 weeks’ gestation. Those whoreceive ultrasound may have uncertain LMP dates (anindication for ultrasound), and are more likely to beprivately insured. The 2003 revised US standard birthcertificate includes a new gestational age item, theobstetric estimate, which is the clinician’s best estimateof gestational age at delivery given available datinginformation, including ultrasound but excluding neo-natal assessments. Validation of this item will beimportant to assess whether it helps address the prob-lems presented with LMP-based gestational age.

Acknowledgements

The California Department of Health Services, GeneticDisease Branch collected the XAFP and NewbornScreening programme records and the California



Center for Health Statistics provided the birth cohortfiles. Allen Hom and Steve Graham of the SequoiaFoundation conducted the record linkage.

References

1 David RJ. The quality and completeness of birthweight andgestational age data in computerized birth files. AmericanJournal of Public Health 1980; 70:964–973.

2 Kramer MS, Platt RW, Wen SW, Joseph KS, Allen A,Abrahamowicz M, et al. A new and improvedpopulation-based Canadian reference for birthweight forgestational age. Pediatrics 2001; 108:E35.

3 Zhang J, Bowes WA Jr. Birth-weight-for-gestational-agepatterns by race, sex, and parity in the United Statespopulation. Obstetrics and Gynecology 1995; 86:200–208.

4 Vahratian A, Buekens P, Bennett TA, Meyer RE, Kogan MD,Yu SM. Preterm delivery rates in North Carolina: are theyreally declining among non-Hispanic African Americans?American Journal of Epidemiology 2004; 159:59–63.

5 Tentoni S, Astolfi P, De Pasquale A, Zonta LA. Birthweightby gestational age in preterm babies according to aGaussian mixture model. BJOG 2004; 111:31–37.

6 Savitz DA, Terry JW Jr, Dole N, Thorp JM Jr, Siega-Riz AM,Herring AH. Comparison of pregnancy dating by lastmenstrual period, ultrasound scanning, and theircombination. American Journal of Obstetrics and Gynecology2002; 187:1660–1666.

7 Waller DK, Spears WD, Gu Y, Cunningham GC. Assessingnumber-specific error in the recall of onset of last menstrualperiod. Paediatric and Perinatal Epidemiology 2000;14:263–267.

8 Rowland AS, Baird DD, Long S, Wegienka G, Harlow SD,Alavanja M, et al. Influence of medical conditions andlifestyle factors on the menstrual cycle. Epidemiology 2002;13:668–674.

9 Wilcox AJ, Dunson D, Baird DD. The timing of the ‘fertilewindow’ in the menstrual cycle: day specific estimates froma prospective study. British Medical Journal 2000;321:1259–1262.

10 Berg AT. Menstrual cycle length and the calculation ofgestational age. American Journal of Epidemiology 1991;133:585–589.

11 ACOG Committee on Practice Bulletins. American Collegeof Obstetrics and Gynecology Practice Bulletin No. 58.Ultrasonography in pregnancy. Obstetrics and Gynecology2004; 104:1449–1458.

12 Henriksen TB, Wilcox AJ, Hedegaard M, Secher NJ. Bias instudies of preterm and postterm delivery due to ultrasoundassessment of gestational age. Epidemiology 1995; 6:533–537.

13 Morin I, Morin L, Zhang X, Platt RW, Blondel B, Bréart G,et al. Determinants and consequences of discrepancies inmenstrual and ultrasonographic gestational age estimates.BJOG 2005; 112:145–152.

14 Chervenak FA, Skupski DW, Romero R, Myers MK,Smith-Levitin M, Rosenwaks Z, et al. How accurate is fetalbiometry in the assessment of fetal age? American Journal ofObstetrics and Gynecology 1998; 178:678–687.

15 Johnsen SL, Rasmussen S, Sollien R, Kiserud T. Fetal ageassessment based on ultrasound head biometry and theeffect of maternal and fetal factors. Acta Obstetricia etGynecologica Scandinavica 2004; 83:716–723.

16 Källén K. Mid-trimester ultrasound prediction of gestationalage: advantages and systematic errors. Ultrasound inObstetrics and Gynecology 2002; 20:558–563.

17 Larsen T, Nguyen TH, Greisen G, Engholm G, Møller H.Does a discrepancy between gestational age determined bybiparietal and last menstrual period sometimes signify earlyintrauterine growth retardation? BJOG 2000; 107:238–244.

18 Yang H, Kramer MS, Platt RW, Blondel B, Bréart G, Morin I,et al. How does early ultrasound scan estimate of gestationalage lead to higher rates of preterm birth? American Journal ofObstetrics and Gynecology 2002; 186:433–437.



a comparison of lmp-based and ultrasound-based estimates of gestasinal age

Documents