The research implications of the selection of a gestational age estimation method

Authors

  • Courtney D. Lynch,

    1. Epidemiology Branch, Division of Epidemiology, Statistics and Prevention Research, National Institute of Child Health and Human Development, National Institutes of Health, Department of Health and Human Services, Bethesda, MD, USA
    Search for more papers by this author
  • Jun Zhang

    1. Epidemiology Branch, Division of Epidemiology, Statistics and Prevention Research, National Institute of Child Health and Human Development, National Institutes of Health, Department of Health and Human Services, Bethesda, MD, USA
    Search for more papers by this author

  • Conflicts of interest: the authors have declared no conflicts of interest.

Dr Courtney D. Lynch, Epidemiology Branch, Division of Epidemiology, Statistics and Prevention Research, NICHD/NIH/DHHS, 6100 Executive Blvd, Room 7B03P, Rockville, MD 20852, USA.
E-mail: lynchcd@mail.nih.gov

Summary

There are three primary methods of gestational age estimation: dating based on last menstrual period (LMP), ultrasound-based dating and neonatal estimates. We review the strengths and limitations of each method as well as their implications for research. Dating based on LMP is a simple, low-cost method of estimating gestational age. Limitations associated with the use of menstrual-based dating include reporting problems such as uncertainty regarding the LMP date, possibly due to bleeding not associated with menses, as well as concerns about the incidence of delayed ovulation, which can result in invalid estimates of gestation, even for women with certain LMP dates. Given that most women in the US have at least one ultrasound during pregnancy, it is becoming increasingly common for clinicians to verify menstrual dates using early ultrasound. To calculate gestational age with the use of ultrasound, fetal measurements are compared with a gestational age-specific reference. The primary limitation of this method is the fact that the gestational age estimates of symmetrically large or small fetuses will be biased. Further, given that ultrasound references were developed using pregnancies that were dated according to reliable LMP dates, they are potentially biased in the same direction as dates calculated according to LMP. Neonatal estimates of gestational age have been shown to be the least precise dating method. To highlight the research implications of the choice of a gestational dating method, we used data from the Routine Antenatal Diagnostic Imaging with Ultrasound Study to identify risk factors for post-term delivery. Risk factors for post-term delivery are shown to vary according to the choice of a gestational dating method, suggesting that some findings are an artefact of the choice of a method rather than evidence of causality.

Introduction

In perinatal research, gestational age is included as an endpoint or covariate in many studies. While on the surface, it seems that the calculation of gestational age is very straightforward, all methods of gestational age assessment have strengths and weaknesses that deserve careful consideration. The purpose of this work is to review the various methods of gestational age estimation, including menstrual-based dating, ultrasound-based dating and neonatal estimates, and to highlight the inherent assumptions and research implications associated with each measurement.

Dating based on last menstrual period

Naegele's rule

In the US, the estimated date of delivery for most pregnancies is assigned according to Naegele's rule.1 This is based on the assumption that the average menstrual cycle is 28 days in length with ovulation occurring on day 14. The average pregnancy is assumed to last 280 days from the first day of the last menstrual period (LMP). In practice, the estimated date of delivery is often calculated with a due date wheel that was developed using Naegele's rule, which subtracts 3 months from the first day of the LMP and adds 7 days.

Date of last menstrual period as a proxy for the date of conception

Because fertilisation occurs soon after ovulation, ovulation is often used as a proxy for the time of conception. The problem with this approach is that many women are not aware of when they ovulate. Although fertility awareness methods such as temperature charting and cervical mucus monitoring are available, as well as home-based urine test kits, for detecting the luteinising hormone surge that immediately precedes ovulation, only a small subset of women use them. Even if all women who were trying to conceive took advantage of such methods, an alternative gestational dating strategy would still be needed for the 50% of US pregnancies that are unintended (i.e. mistimed or unwanted).2 Therefore, in the absence of the availability of a first- or early second-trimester obstetric ultrasound, estimating the date of conception based on the date of the first day of the LMP remains the best available option for most women.

The problem of uncertain dates

One problem with the use of menstrual-based dating is that it requires women to be able to report a reasonably accurate LMP date. Not only does the percentage of women who report a specific LMP date vary depending on the population, women with certain dates have been found to be systematically different in many ways from women with uncertain dates. Using 1976–80 data from the Aberdeen Maternity and Neonatal Data Bank, Hall and colleagues found that among the 11 602 women studied, 79% had certain dates (certain within ±1 week), 13% had approximate dates (certain ±2 weeks), and 7% had uncertain dates (certain ±4 weeks).3 Women with approximate or uncertain dates were more likely to be young, primiparous, smokers, of lower educational attainment, and to have long or irregular menstrual cycles compared with women with certain dates.3 Buekens and colleagues found that 16% of the 22 404 singleton deliveries that occurred in 10 Belgium hospitals between 1974 and 1978 had unknown LMP dates.4 Women with unknown dates were more likely to be aged 19 years or younger, aged 35 years or older, of North African nationality, unmarried or illiterate. Wenner and Young found that 34% of the 355 mothers who received care from the Seattle-King County Maternal and Infant Care Program between 1969 and 1979 had non-specific LMP dates.5 Women with a non-specific date were more likely than women with specific dates to experience complications such as maternal proteinuria, caesarean section and a low 1-min Apgar score.5

Digit preference

Even among women who report a specific LMP date, the date may not be accurate. Waller and colleagues examined data from the California alpha-fetoprotein screening programme from 1996 to 1997 and California birth certificate data from 1987.6 They found that at least 7.9% of the screening programme LMP dates and 12.9% of the birth certificate dates appeared to be affected by digit preference. The numbers that were reported more often than expected in both datasets included 1, 5, 10, 15, 20, 25 and 28, with the number 15 being reported 2.5 times more than expected.

The impact of the timing of ovulation and other biological factors

Several biological factors can affect the validity of using LMP dates as a proxy for the time of conception, including the timing of ovulation, the timing of fertilisation relative to ovulation, midcycle bleeding (i.e. bleeding not associated with menses), and bleeding related to the use of oral contraceptives. Ovulation typically occurs 14 days prior to the beginning of the next menses, approximately 17 h after a peak in plasma luteinising hormone.7 The follicular phase, which ends with ovulation, is lengthened among women whose menstrual cycles are longer than 28 days.8 About 30% of women report an average cycle length of at least 30 days.9

Table 1 compares the characteristics of studies that have reported the mean length of the follicular phase (i.e. the timing of ovulation).8,10,11 Even among women with average menstrual cycle lengths, the timing of ovulation can vary. Baird and colleagues reported that only 10% of women with regular 28-day menstrual cycles ovulated on exactly day 14.11 Seventy-five per cent of their participants, however, were found to have ovulated within ±4 days of day 13 (the mode of the recorded follicular phase lengths). One population-based study found that the most frequent length of human gestation is 283 days (rather than 280 days), a finding that suggests that ovulation occurring on days 15 to 17 is possibly typical rather than delayed.12

Table 1.  Comparison of studies reporting the mean length of the follicular phase
Authors and dateDescription of populationSample sizeOvulation detection methodFollicular phase length
Mean ± SD (days)
  1. LMP, last menstrual period.

  2. x = not calculated.

Baird et al.11 (1995)Women trying to conceive during the North Carolina Early Pregnancy Study221Day of luteal transition = day on which the urinary oestrogen to progesterone ratio begins to drop16.3 ± 6.2
Nakling et al.10 (2005)Cohort of singleton pregnancies among women residing in Oppland County, Norway11 238Modelled using LMP and ultrasound-based estimates of gestational agex ± 7
Saito et al.8 (1972)Cohort of patients presenting to Tokyo Medical and Dental University Hospital (approximately a third of whom were infertile)129Basal body temperature shift18.8 ± 8

The timing of fertilisation relative to ovulation as well as atypical bleeding patterns can also influence the accuracy of LMP-based dating. While sperm can survive in the female reproductive tract for up to 6 days, the median survival time for an egg is approximately 12 h.13,14 As such, pregnancy is most likely to take place during a cycle in which intercourse occurs in the 5 days prior to ovulation and/or on the day of ovulation itself.15,16 Midcycle bleeding or bleeding associated with withdrawal from oral contraceptives could also lead a woman to misreport or be uncertain of her LMP date.17,18

Dating based on early ultrasound

Trends in ultrasound use

While gestational dating based on first-trimester ultrasound was once reserved for women with unknown LMP dates, it is becoming increasingly common in the US to use ultrasound to routinely verify women's estimated dates of delivery. Many women in the US have at least one obstetric ultrasound during pregnancy,19 although research has failed to demonstrate the benefit of its routine use in low-risk populations.20–24 US clinicians will often revise a woman's due date if the LMP and ultrasound-based estimates differ by more than: ±7 days up to 20 weeks' gestation, ±14 days from 20 to 30 weeks' gestation, and ±21 days at 30 weeks' gestation and beyond.25

The basis for ultrasound dating

To estimate gestational age with the use of ultrasound, physicians take various measurements of the fetus depending on the woman's reported LMP date. For ultrasounds performed during the first trimester, crown–rump length is used to estimate gestational age, given its rapid growth and linear relation with gestation age during this time period.26 The crown–rump landmarks become visible at approximately 8 weeks' gestation.26 In the second and third trimesters, various combinations of biparietal diameter, head circumference, abdominal circumference and femur (diaphysis) length are used.27 The fetal measurements are compared with age-specific references using standard formulae.

The problems with ultrasound dating

Like menstrual dating, ultrasound-based gestational dating is imperfect. The primary criticism of ultrasound-based dating is that fetal measurements are compared with fetal size references that fail to account for normal variability.28 Using these references, the estimated gestational age of a symmetrically large fetus would be biased upward, although the use of newer dating formulae that rely on multiple fetal parameters is thought to reduce the magnitude of this bias.29 Some researchers have questioned the utility of ultrasound-based dating formulae altogether, as most of them were developed using reliable LMP dates as the gold standard.30 If delayed ovulation leads menstrual-based dating to systematically overestimate gestational length by several days in some women, then ultrasound-based formulae that were developed using women with reliable menstrual dates as the gold standard would likely be biased in the same direction.

Still others have argued that the manner in which the formulae were derived is not a concern, as several recent studies of pregnancies achieved through in vitro fertilisation (in which the dates of conception were known) have shown ultrasound-based dating to be highly accurate, including one study that demonstrated that most of the 38 published formulae for ultrasound-based gestational dating had systematic errors of less than 1 week.29 It should be noted, however, that fetuses conceived through assisted reproductive technologies may not be ideal for assessing the accuracy of ultrasound-based gestational dating formulae, as the growth patterns of fetuses achieved through assisted reproductive technologies may differ from the growth patterns of naturally conceived fetuses.31

Neonatal estimates

In situations where a woman receives little or no prenatal care, gestational age can be estimated neonatally using the Dubowitz or Ballard examinations.32–34 Using standardised scoring systems, clinicians are able to estimate gestational age based on the physical and neuromuscular maturity of the infant. While neonatal (i.e. postnatal) estimates have some clinical utility, they are less precise than obstetric estimates of gestational age; therefore, they are not ideal for research purposes.

The main problem with neonatal estimates of gestational age is with regard to validity. In general, neonatal estimates tend to overestimate the gestation of infants born at <40 weeks, while underestimating the gestation of infants born at ≥40 weeks.35 For example, the Dubowitz examination has been found to overestimate the obstetric estimate of gestational age by as much as 2 weeks in infants <34 weeks of gestation,36 while the overestimation associated with the use of the New Ballard Score is of the order of 2 to 4 days for infants born at 32 to 37 weeks' gestation.34 Similarly, Alexander and colleagues found that 75% of infants who were found to be post-term according to ultrasound were misclassified as term infants by the Ballard examination.35 There is also some suggestion that the extent of the systematic error may differ by race, as the Ballard Score has been shown to systematically overestimate the gestational age of black infants.37

Which method more accurately predicts the date of delivery?

When considering the use of menstrual histories vs. ultrasound for pregnancy dating, it is important to understand that the techniques measure two different entities. One measures the length of pregnancy, and the other measures the size of the fetus. Numerous studies have compared the use of menstrual vs. ultrasound-based dating for gestational age assessment, and most have reported that dating based on early second-trimester ultrasound is superior to LMP-based dating in predicting the actual date of delivery, even among women with certain LMP dates (Table 2).38–45

Table 2.  Studies comparing the accuracy of LMP- and ultrasound-based gestational age estimation in predicting delivery date
Authors and dateResearch questionPopulationDefinition of reliable menstrual historyUltrasound timingFetal biometry used for predictionSummary
  1. LMP, last menstrual period; CRL, crown–rump length; BPD, biparietal diameter; IVF, in vitro fertilisation; EDC, estimated date of confinement.

Campbell et al.41 (1985)Is a single ultrasound-based estimate of gestational age more predictive of estimated delivery date than optimal menstrual history?4246 obstetric patients at King's College Hospital who gave birth to liveborn infants weighing 2500 g or moreKnown date of LMP; regular cycles (±5 days); no oral contraceptive use in the previous 2 months; no unusual bleedingMore than 85% of women had a scan before the 21st week of gestationCRL and/or BPD44.7% with non-specific dates; routine ultrasound dating using BPD at 12–18 weeks of gestation is more predictive of delivery date than optimal menstrual history.
Kramer et al.39 (1988)Does the magnitude and direction of the error between LMP and ultrasound-based estimates vary as a function of the LMP-based estimate?11 045 women who gave birth following spontaneous labour at Royal Victoria HospitalKnown LMP dateEarly second-trimester ultrasound (usually 16–18 weeks’ gestation)BPDUsing ultrasound as the gold standard, positive predictive value of LMP estimate is 0.949 for term infants, 0.775 for preterm infants, and 0.119 for post-term infants.
Rossavik and Fishburne38 (1989)Do various ultrasound-based gestational dating curves give similar results in a population of women with reliable menstrual dates?17 women who had IVF, 6 women with sonographically confirmed ovulation, and 60 women with reliable menstrual datesKnown LMP date; regular cycles; no oral contraceptive use in the previous 2 monthsRandom selection of ultrasound scans throughout pregnancy stratified by timeBPDNo differences by plurality; a LMP-based estimate of gestational age that is supported by findings of a first-trimester pelvic exam may be better at predicting conceptional age than the best ultrasound-based dating curve.
Waldenstrom et al.44 (1990)Among women with optimal menstrual histories, at what point should the ultrasound-based EDC be used in lieu of the LMP-based estimate?4609 women with reliable menstrual histories who went into spontaneous labourKnown LMP date; regular periods (±14 days); no withdrawal bleedingUltrasound at approximately 15 weeks of gestationBPDWhen the EDCs differ by 1 week or less, there is no difference in the ability of LMP and ultrasound to predict delivery date. When the dates are discrepant by more than 1 week, then the ultrasound-based estimate is a better predictor of delivery date.
Kieler et al.45 (1993)Among women with very regular (every 28 days) cycles, is the ultrasound-based EDC more accurate than the LMP-based estimate?1713 women from a multicentre randomised controlled trial who gave birth to liveborn singletonsKnown LMP date; regular cycles (every 28 days); no recent oral contraceptive use; no unusual bleedingWomen from the intervention arm had a scan at 15 weeks’ gestationBPDWhen the difference between the LMP and ultrasound-based EDC is greater than 7 days, ultrasound is more accurate at predicting delivery date. When the difference is 7 days or less, both methods are equally accurate.
Backe and Nakling40 (1994)Is ultrasound-based gestational age estimation more accurate at predicting the date of delivery in a routine clinical setting?1650 Norwegian women who gave birth to singletons via spontaneous deliveriesLMP recalled ±3 days; regular 28 day cycles (±4 days); no recent use of oral contraceptives; no recent pregnanciesPrior to 20 weeks of gestationBPD20% with unreliable dates; ultrasound-based EDC was significantly closer to the delivery date than LMP-based EDC.
Mongelli et al.42 (1996)Which EDC should be used when both a valid LMP and an ultrasound-based estimate are available?34 249 women from the East Midland Obstetric Database who received care prior to 24 weeks of gestationKnown LMP date; regular cycles; no oral contraceptive use in the previous 3 monthsMean age at ultrasound examination was 15.6 weeksBPDAmong women with certain LMP, ultrasound alone is a better predictor of date of delivery than any method that takes the LMP date into consideration.
Tunon et al.43 (1996)Which EDC should be used when both a valid LMP and an ultrasound-based estimate are available? If the estimates are discrepant by less than 7 days, which estimate should be used?14 167 women with reliable menstrual histories who had spontaneous vaginal births of normally formed liveborn infantsKnown LMP date; regular cyclesSecond-trimester ultrasound (15–22 weeks of gestation)BPDEven when the difference between the LMP and ultrasound-based EDC is less than 7 days, the ultrasound-based estimate is better at predicting the date of delivery.

Most of the early work that was conducted comparing LMP with ultrasound dating techniques used fetal head measurements (i.e. biparietal diameter) to estimate gestational age. In these studies, the ultrasounds were performed between 15 and 20 weeks' gestation according to LMP. To be eligible for enrolment, the women in each study were required to have a known LMP date; however, the authors of two of the studies pointed out that some of the women's dates were unreliable.40,41 In every case, the authors reported that ultrasound-based dating techniques are superior to dating based on LMP, particularly with regard to predicting the actual date of delivery.

Mongelli and colleagues42 calculated the estimated dates of delivery for 34 249 singleton pregnancies with reliable menstrual dates according to five methods: LMP only, ultrasound only, and three separate combinations of LMP and ultrasound that replaced the LMP-based date with the ultrasound-based date if the discrepancy was more than 7, 10 or 14 days respectively. They reported that dating based on ultrasound alone was significantly better at predicting the actual date of delivery than LMP-based dating alone or any combination of the two methods. Delivery occurred within ±10 days of the estimated date in 64.1% of the women when menstrual dates alone were used, and in 70.3% of the women when ultrasonography alone was used. However, it should be stressed that delivery occurred on the predicted date in only 3.6% of women when the date was based on LMP and in only 4.3% of women when the date was based on ultrasound.

It is important to note that despite the fact that ultrasound may be slightly better at predicting the actual date of delivery, the real difference in the estimated dates of delivery calculated by the two methods is quite small. Backe and Nakling have suggested that LMP-based dating systematically underestimates the ultrasound-based estimated date of delivery by an average of 2–3 days.40 While the selection of a gestational dating method clearly has implications when studying outcomes such as pre- and post-term birth, the difference is not likely to be clinically relevant in most situations, as physicians often rely on the LMP-based estimated date of delivery unless there is at least a 1-week discrepancy between the LMP-based and ultrasound-based estimates.

Implications for research

An example

A key issue to understand when selecting a gestational age estimation method for research purposes is that the choice may have implications, particularly in studies of pre- or post-term delivery as well as studies of outcomes that are closely related to length of pregnancy. To illustrate this issue, we used data from the Routine Antenatal Diagnostic Imaging with Ultrasound Study (RADIUS) to evaluate the effect of the choice of gestational age estimation method on the identification of risk factors for post-term delivery.20

The RADIUS was a randomised, clinical trial that was conducted from 1987 to 1991 to assess whether the administration of two routine obstetric ultrasound examinations reduces perinatal morbidity and mortality in a low-risk population. Eligibility criteria for participation included: aged 18 years or older, English speaking, date of first day of the LMP known within 1 week, regular menstrual cycles, free of chronic disease, no history of recent oral contraceptive use, and no history of stillbirth or history of delivering a small-for-gestational-age infant. Eligible women who provided informed consent to participate were randomised to receive either standard pregnancy management (routine prenatal care from her selected obstetrician) or two routine obstetric ultrasounds (scheduled for 15–22 and 31–35 weeks according to LMP) in addition to standard pregnancy management.

Among the 53 367 women screened for participation in the RADIUS trial, 21 050 (39%) were found to be eligible. RADIUS investigators randomised 15 530 women, 7812 to the intervention group and 7718 to the control group. We used data from the 7812 women randomised to the intervention group. We excluded 105 women who were lost to follow-up, 60 early pregnancy losses, 12 elective terminations, 68 multifetal pregnancies, 32 women who withdrew informed consent, 574 women who were missing fetal ultrasound measurements from either examination, and 178 women who were of a race/ethnicity other than non-Hispanic white or black. This left 6783 women available for analysis.

Menstrual-based pregnancy length in days was calculated by subtracting the date of the first day of the LMP from the date of delivery. Ultrasound-based gestational length was determined by calculating the estimated gestational age in days at the first routine ultrasound (according to ultrasound dating standards), and then adding the number of days between that date and delivery. Gestational age at the first routine ultrasound was calculated according to the multiple parameter formula of Hadlock et al., which used the head circumference, abdominal circumference, biparietal diameter and femur length measurements recorded during the first routine ultrasound.28 This formula has been shown to be the most accurate ultrasound-based gestational dating formula in studies of women with certain menstrual dates, as well as pregnancies achieved through assisted reproductive technologies. It has been shown to have a systematic error of less than 1 day.29

Post-term delivery was defined as delivery at 42 completed weeks of gestation or later according to three estimation methods: LMP, ultrasound and an obstetric estimate, in which the LMP estimate was replaced by the ultrasound measurement if they differed by more than 14 days.

The incidence of post-term delivery varied according to the gestational age estimation method. It ranged from a high of 5.4% [95% confidence interval (CI) 5.1, 5.7] when calculated according to LMP to a low of 2.5% [95% CI 2.3, 2.7] when calculated with ultrasound. The incidence of post-term delivery was 3.8% [95% CI 3.6, 4.1] using the obstetric estimate algorithm. The mean length of pregnancy calculated according to LMP was 279 days, 1.3 days longer than the ultrasound-based estimate.

These figures are similar in magnitude and direction to those reported by others. Between 1987 and 1989, Reuss and colleagues prospectively followed 1159 pregnant women for a study in which pregnancy length was a primary outcome of interest.46 They found that the incidence of post-term delivery was 9.7% when gestational age was calculated according to reliable LMP dates, compared with 2.8% based on ultrasound evaluations. Using data collected from 3655 women as part of the Pregnancy, Infection, and Nutrition study from October 1995 to May 2001, Savitz and colleagues found that the incidence of post-term delivery was 12.1% according to LMP and only 3.4% according to ultrasound. Further, they found that the LMP-based estimate of gestation was 2.8 days longer on average than the ultrasound-based estimate.47

We used unconditional logistic regression to identify risk factors for post-term delivery (Table 3). The factors that we examined included: maternal race (non-Hispanic white, non-Hispanic black); age (18–20, 21–30, 31 + years); education (less than high school, high school graduate, some college or college graduate, graduate school); smoking status at enrolment (non-smoker, smoker); maternal pre-pregnancy body mass index (BMI; kg/m2 classified as underweight, healthy, overweight or obese); parity (0, 1+); and infant sex (female, male). Variables were retained in the final model if they were found to be significantly associated with post-term delivery, or if they changed the estimate of another factor by more than ±10%.

Table 3.  Predictors of post-term delivery by method of estimation of gestational age, Routine Antenatal Diagnostic Imaging with Ultrasound Study (RADIUS)
 Percentage
(n = 6783)
LMP estimateUltrasound estimateObstetric estimatea
Crude ORbAORcCrude ORbAORcCrude ORbAORc
  • LMP, last menstrual period.

  • a

    LMP-based estimate replaced by ultrasound-based estimate if they differed by more than 14 days.

  • b

    Crude odds ratios [95% confidence intervals (CI)].

  • c

    Adjusted odds ratios [95% CI].

  • Bold = P < 0.05.

Maternal race
 Non-Hispanic white96.21.0 Reference1.0 Reference1.0 Reference1.0 Reference1.0 Reference1.0 Reference
 Non-Hispanic black3.80.9 [0.5, 1.7]0.9 [0.5, 1.5]0.5 [0.1, 1.4]0.3 [0.1, 1.2]0.7 [0.3, 1.5]0.7 [0.3, 1.4]
Maternal age (years)
 18–204.11.6 [1.0, 2.5]1.3 [0.8, 2.1]1.0 [0.5, 2.3]0.8 [0.4, 1.9]1.6 [0.9, 2.6]1.2 [0.7, 2.0]
 21–3067.51.0 Reference1.0 Reference1.0 Reference1.0 Reference1.0 Reference1.0 Reference
 31+28.40.7 [0.5, 0.9]0.8 [0.6, 1.0]1.2 [0.8, 1.6]1.6 [1.1, 2.3]0.8 [0.6, 1.0]1.0 [0.7, 1.3]
Maternal education
 Less than high school3.41.0 Reference1.0 Reference1.0 Reference1.0 Reference1.0 Reference1.0 Reference
 High school graduate25.70.9 [0.5, 1.5]0.9 [0.5, 1.5]1.9 [0.6, 6.2]1.7 [0.5, 5.5]0.8 [0.4, 1.4]0.8 [0.4, 1.5]
 College61.80.8 [0.5, 1.4]0.8 [0.5, 1.4]2.0 [0.6, 6.2]1.6 [0.5, 5.4]0.7 [0.4, 1.3]0.7 [0.4, 1.4]
 Graduate school9.10.7 [0.4, 1.4]0.7 [0.4, 1.4]2.4 [0.7, 8.1]1.7 [0.5, 6.2]0.6 [0.3, 1.2]0.6 [0.3, 1.3]
Maternal smoking status
 Non-smoker87.11.0 Reference1.0 Reference1.0 Reference1.0 Reference1.0 Reference1.0 Reference
 Smoker12.91.3 [1.0, 1.8]1.3 [0.9, 1.7]1.0 [0.6, 1.5]1.0 [0.6, 1.7]1.2 [0.9, 1.7]1.1 [0.8, 1.6]
Maternal body mass index
 Underweight6.01.1 [0.7, 1.8]1.1 [0.7, 1.6]1.2 [0.6, 2.3]1.3 [0.7, 2.4]1.3 [0.8, 2.1]1.2 [0.7, 2.0]
 Healthy71.61.0 Reference1.0 Reference1.0 Reference1.0 Reference1.0 Reference1.0 Reference
 Overweight or obese22.41.4 [1.1, 1.7]1.4 [1.1, 1.7]1.5 [1.1, 2.1]1.6 [1.1, 2.3]1.3 [1.0, 1.8]1.3 [1.0, 1.8]
Parity
 045.61.0 Reference1.0 Reference1.0 Reference1.0 Reference1.0 Reference1.0 Reference
 1+54.40.6 [0.5, 0.8]0.7 [0.5, 0.8]0.3 [0.2, 0.4]0.3 [0.2, 0.4]0.5 [0.4, 0.6]0.5 [0.4, 0.6]
Infant sex
 Female48.91.0 Reference1.0 Reference1.0 Reference1.0 Reference1.0 Reference1.0 Reference
 Male51.10.9 [0.7, 1.1]0.9 [0.7, 1.1]1.4 [1.0, 1.8]1.3 [1.0, 1.8]0.8 [0.6, 1.1]0.8 [0.6, 1.1]

When pregnancy length was estimated according to LMP, maternal age, smoking status, maternal pre-pregnancy BMI and parity were found to be statistically significantly associated with post-term delivery in the univariable analyses. In the adjusted analyses, overweight and obese women had a 40% increased odds of post-term delivery compared with healthy-weight women, while the odds of post-term delivery was 30% lower among multiparae compared with primiparae.

When pregnancy length was based on ultrasound, maternal pre-pregnancy BMI and parity were statistically significantly associated with post-term delivery in the univariable analyses. In the adjusted analyses, overweight or obese women had a 50% increased odds of delivering post-term compared with healthy-weight women. Multiparae, in contrast, had a 70% decreased odds of delivering post-term compared with primiparae.

As expected, the results in which pregnancy length was calculated using the obstetric estimate algorithm were somewhat intermediate to the LMP and ultrasound findings. The two factors that were found to be significantly predictive of post-term delivery were maternal pre-pregnancy BMI and parity. In the adjusted analyses, overweight or obese women had a 30% increased odds of post-term delivery compared with healthy-weight women, while the odds of post-term delivery was decreased by approximately 50% in multiparae compared with primiparae.

Given that the choice of method of gestational age estimation has a clear impact on the identification of risk factors for post-term delivery, it is likely that most of the observed associations are artifacts rather than evidence that the factors are in some way involved in the aetiology of post-term delivery. In the LMP-based analyses, factors associated with longer or more irregular menstrual cycles were associated with post-term delivery; however, in the ultrasound-based analyses, identified factors were related to fetal growth. It has been well documented that there is systematic bias in the assessment of LMP and ultrasound-based gestational age for some subgroups of the population.48–50 These biases often result in somewhat large (>7 days) discrepancies between LMP and ultrasound-based estimates of gestation, which can lead to spurious associations between certain covariates and outcomes related to length of pregnancy, depending on the gestational age estimation method that is chosen.

One important issue to note is that while the impact of a given factor on the estimate of pregnancy length may be small, the joint effect of systematic errors could be quite large for certain subgroups of the population.48 For instance, overweight or obese women have longer and more irregular menstrual cycles compared with normal-weight women, which influences the accuracy of LMP-based gestational age estimates.51,52 In addition, overweight or obese women are more likely to carry larger fetuses, resulting in systematic error in the ultrasound-based estimate of pregnancy length.53

Regardless of the gestational age estimation method chosen, multiparae appear to be less likely than primiparae to deliver a post-term infant. This is consistent with previous work that indicates the mean length of pregnancy is somewhat shorter among multiparae.12 While the odds ratios for post-term delivery were comparable when gestation was calculated using LMP or the clinical-based estimate, the odds ratio for post-term delivery appears to be reduced even further when ultrasound-based estimates are used.

Conclusion

When designing and conducting perinatal research studies, particularly those in which pregnancy length is an outcome or the outcome of interest is related to gestational age, great care should be taken in selecting a gestational age estimation method. Like others, we have shown that study findings can differ according to the method selected. At a very minimum, bias due to the selection of a particular dating method should always be considered as an alternative explanation for associations identified.

Acknowledgements

This work was supported with intramural and extramural funding (HD21017, HD19897, HD21140) from the National Institute of Child Health and Human Development, National Institutes of Health, Department of Health and Human Services.

Ancillary