John Owen, Nicole Yost, Vincenzo Berghella, Elizabeth Thom, Melissa Swain, Gary A. Dildy III, Menachem Miodovnik, Oded Langer, Baha Sibai, Donald McNellis, for the National Institute of Child Health and Human Development, Maternal-Fetal
Medicine Units Network. Mid-Trimester Endovaginal Sonography in Women at High Risk for Spontaneous Preterm Birth. JAMA. 2001;286(11):1340–1348. doi:10.1001/jama.286.11.1340
Author Affiliations: Department of Obstetrics and Gynecology, University of Alabama at Birmingham (Dr Owen); Department of Obstetrics and Gynecology, University of Texas Southwestern Medical Center, Dallas (Dr Yost); Department of Obstetrics and Gynecology, Thomas Jefferson University, Philadelphia, Pa (Dr Berghella); George Washington University Biostatistics Center, Bethesda, Md (Dr Thom); Department of Obstetrics and Gynecology, Wake Forest University, Winston-Salem, NC (Ms Swain); Department of Obstetrics and Gynecology, University of Utah, Salt Lake City (Dr Dildy); Department of Obstetrics and Gynecology, University of Cincinnati, Ohio (Dr Miodovnik); Department of Obstetrics and Gynecology, University of Texas, San Antonio (Dr Langer); Department of Obstetrics and Gynecology, University of Tennessee, Memphis (Dr Sibai); and the National Institute of Child Health and Human Development, Bethesda, Md (Dr McNellis). Dr Dildy is now with the Louisiana State University, Baton Rouge; Drs Miodovnik and Langer are now with Columbia University, New York City, NY; Dr Sibai is now with the University of Cincinnati, Ohio; and Dr McNellis is retired.
Context Although shortened cervical length has been consistently associated
with spontaneous preterm birth, it is not known when in gestation this risk
factor becomes apparent.
Objective To determine whether sonographic cervical findings between 16 weeks'
and 18 weeks 6 days' gestation predict spontaneous preterm birth and whether
serial evaluations up to 23 weeks 6 days' gestation improve prediction in
Design, Setting, and Participants Blinded observational study performed between March 1997 and November
1999 at 9 university-affiliated medical centers in the United States in 183
women with singleton gestations who previously had experienced a spontaneous
birth before 32 weeks' gestation.
Observation Certified sonologists performed 590 endovaginal sonographic examinations
at 2-week intervals. Cervical length was measured from the external os to
the functional internal os along a closed endocervical canal. Funneling and
dynamic cervical shortening were also recorded.
Main Outcome Measure Spontaneous preterm birth before 35 weeks' gestation, analyzed by selected
cutoff values of cervical length.
Results Forty-eight women (26%) experienced spontaneous preterm birth before
35 weeks' gestation. A cervical length of less than 25 mm at the initial sonographic
examination was associated with a relative risk (RR) for spontaneous preterm
birth of 3.3 (95% confidence interval [CI], 2.1-5.0; sensitivity = 19%; specificity
= 98%; positive predictive value = 75%). After controlling for cervical length,
neither funneling (P = .24) nor dynamic shortening
(P = .054) were significant independent predictors
of spontaneous preterm birth. However, using the shortest ever observed cervical
length on serial evaluations, after any dynamic shortening, the RR of a cervical
length of less than 25 mm for spontaneous preterm birth increased to 4.5 (95%
CI, 2.7-7.6; sensitivity = 69%; specificity = 80%; positive predictive value
= 55%). Compared with a single cervical measurement at 16 weeks' to 18 weeks
6 days' gestation, serial measurements at up to 23 weeks 6 days significantly
improved the prediction of spontaneous preterm birth in a receiver operating
characteristic curve analysis (P = .03).
Conclusions Cervical length assessed by endovaginal sonography between 16 weeks'
and 18 weeks 6 days' gestation, augmented by serial evaluations, predicts
spontaneous preterm birth before 35 weeks' gestation in high-risk women.
Preterm birth is the most important cause of infant morbidity and mortality
and complicates 11% of all pregnancies in the United States.1
Most (80%) of these births result from either spontaneous labor or membrane
rupture.2 Since the development of neonatal
intensive care units, most neonatal deaths associated with prematurity occur
in infants born at less than 32 weeks' gestation, but significant morbidities
including sepsis, respiratory distress, and necrotizing enterocolitis do not
abate until 35 weeks' gestation, after which neonatal outcomes are generally
good.3,4 To date, a prior preterm
birth is one of the strongest and most consistent predictors of prematurity,
and the risk of recurrence is inversely proportional to the gestational age
of the prior delivery.5,6
Endovaginal ultrasound is a reliable technology for imaging the cervix
and lower uterine segment during pregnancy.7,8
While there is ample evidence that a shortened cervical length is associated
with preterm birth,9- 18
it is not known when this risk factor becomes apparent in pregnancy or whether
the adverse cervical ultrasound findings develop over time. Moreover, most
of the current data linking cervical length to subsequent preterm birth have
been collected beyond 20 weeks' gestation.9- 11,14,15
Importantly, most of the available data have been collected either in unselected,
or without physician masking, which means interventions were applied on the
basis of the sonographic findings without a control group for comparison,14,15,17,18 thus
rendering the predictive value of the cervical sonographic findings uncertain.
The importance of longitudinal observations and the natural history of cervical
characteristics in the mid-trimester have also not been well defined.12,17- 19
Our objective was to determine whether cervical characteristics visualized
with endovaginal sonography as early as 16 weeks' through 18 weeks 6 days'
gestation or longitudinally up to 23 weeks 6 days' gestation would predict
spontaneous preterm birth in women with a previous spontaneous preterm birth
before 32 weeks' gestation. From the standpoint of efficacy and other biological
considerations, certain interventions (cerclage) might be more effective if
applied early in gestation (ie, before 24 weeks). We hypothesized that endovaginal
sonography could identify women whose cervical anatomy would make them candidates
for future mid-trimester clinical intervention trials of preterm birth prevention.
This study was performed at 9 university-affiliated centers, all members
of the National Institute of Child Health and Development, Maternal-Fetal
Medicine Units Network, between March 1997 and November 1999. Women with singleton
pregnancies who had experienced at least 1 prior spontaneous preterm birth
before 32 weeks' gestation were eligible; funding was not available to study
a concurrent, low-risk control population. If obstetric records were not available,
a history consistent with spontaneous preterm birth (preterm labor or membrane
rupture) and a birth weight of less than 1500 g were
deemed satisfactory criteria. Women with chronic medical or obstetric problems
that might result in an indicated preterm birth (eg, hypertension, red blood
cell isoimmunization), a history of substance abuse, or uterine anomalies
were ineligible. Women who received a cerclage because of a clinical history
of cervical incompetence were also excluded. The institutional review board
at each center approved the study and potential participants who gave written,
informed consent could be enrolled as long as their first endovaginal sonogram
would be performed between 16 weeks' and 18 weeks 6 days' gestation.
Gestational age was determined by comparing a certain last menstrual
period (if available) with a sonographic evaluation at or before 18 weeks'
gestation. Concordance between the biometric parameters and the menstrual
date of 7 days or less confirmed the last menstrual period; otherwise, the
biometric data were used. After the initial endovaginal sonographic evaluation,
biweekly visits were scheduled to end no later than 23 weeks 6 days' gestation
with a maximum of 4 sonograms per patient.
All sonograms were performed by physicians, ultrasound technologists,
or research nurses who received uniform training and certification before
patient enrollment. Each sonologist reviewed a training videotape of 8 complete
mid-trimester endovaginal sonograms that demonstrated all the required measurements
and subjective assessments. The videotape was accompanied with a detailed
written description of each examination. Each sonologist independently performed
10 endovaginal examinations on unselected patients in the mid-trimester. The
primary investigator critiqued the videotapes and accompanying data sheets
to identify deficiencies. When necessary, the sonologist was asked to submit
additional taped examinations and data sheets demonstrating correction of
any previously identified deficiencies. From the videotaped examinations,
the primary investigator also approved the ultrasound unit(s) at each center.
Each sonographic examination was performed according to a defined protocol:
patients were asked to empty their bladder and then placed in a dorsal lithotomy
position. The endovaginal probe covered by a sterile, lubricated condom was
inserted and advanced along the vaginal canal until an adequate sagittal image
of the cervix could be visualized. The probe was withdrawn slowly until the
image blurred and then the insertion pressure was increased only enough to
restore an adequate image.11,20
An adequate image for the measurement of cervical length was defined as the
visualization of the internal os, external os, and endocervical canal.20
Cervical length was measured with electronic calipers as the linear
distance between the external os and the functional internal os along a closed
endocervical canal (Figure 1). However,
if the endocervical canal appeared to be curved, cervical length was also
assessed as the sum of the lengths of 2 contiguous linear segments, placed
along the endocervical canal, connecting the external os and functional internal
os. If the maximum deflection of canal curvature (defined as the distance
between a line connecting the internal os and external os and the maximum
excursion of the 2 linear components) was at least 5 mm, the recorded cervical
length measurement was the sum of the 2 linear segments (Figure 2A); otherwise, the single linear distance measurement was
recorded (Figure 2B).
Cervical length measurements were performed 3 times. The sonologist
assessed the overall quality of the 3 images and recorded the cervical length
associated with the image that in his/her opinion was associated with the
subjectively best image. However, if the cervical length differed on images
of similar overall quality, the shortest observed cervical length was recorded.11 If a normal-appearing internal os could not be recognized,
the image was further assessed for either funneling or a poorly developed
lower uterine segment. Funneling required prolapse of the membranes through
a dilated endocervical canal to the level of the functional internal os. Funnel
depth was measured from the functional internal os to its "shoulder," visible
more cephalad toward the lower uterine segment (Figure 1 and Figure 2B).
To be characterized as a funnel, the measured depth had to be at least 5 mm.
A poorly developed lower uterine segment precluded a cervical length
measurement because the internal os could not be visualized as a discrete
structure. This subjective diagnosis was characterized by the presence of
an unusually long cervix (generally >50 mm), an s-shaped
endocervical canal, an increased distance between the bladder reflection and
the amniotic cavity, 2 different echogenic areas in the cervix, and an apparent
internal os located appreciably cephalad to the inferior edge of the bladder
reflection. For analyses of cervical length as a continuous variable, cases
of poorly developed lower segments were arbitrarily assigned a cervical length
of 62 mm, which was 1 mm greater than the longest measured cervical length
After baseline assessments were performed, fundal pressure was applied
for 15 seconds along the axis of the canal by the sonologist21
who maintained the standard sagittal view of the cervix to detect any fundal
pressure–induced dynamic changes in the cervix. If the cervix appeared
to shorten, a funnel developed (or increased in size), or if a poorly developed
lower uterine segment resolved as a result of the fundal pressure, repeat
measurements were obtained. Sonograms lasted a minimum of 5 minutes to detect
spontaneously occurring dynamic changes, which also prompted repeat measurements.
Examinations were videotaped for quality assurance.
According to the study protocol, the results of each scan were not made
available to the patient's managing physicians, except in cases of complete
placenta previa (placental tissue visualized extending >1 cm on both sides
of the internal os) or fetal death. The reason for any notification was recorded.
As part of continuing quality assurance, a sample of the videotaped examinations
was selected from each participating center proportional to its enrollment.
The videotapes and data sheets were reviewed by the primary investigator in
conjunction with another subcommittee member blinded to the pregnancy outcome.
If any measurements or subjective assessments were deemed incorrect, the responsible
sonologist and study coordinator were notified and asked to reexamine the
videotape and make appropriate corrections. Initially, examinations were chosen
at random. However, with increasing experience we also developed criteria
for selected reviews that included all cases of funneling, cervical lengths
less than 20 mm or greater than 50 mm, poorly developed lower uterine segments,
spontaneous or fundal pressure–induced dynamic changes, and cases in
which the physician was notified.
The primary outcome criterion for this study was a spontaneous preterm
birth before 35 weeks' gestation, defined as a birth that resulted directly
from either preterm labor or spontaneous membrane rupture before the onset
of labor. Deliveries effected for maternal or fetal reasons were coded as
indicated. As part of the study design, we performed a sample size calculation
based on the following assumptions. Since appropriate mid-trimester sonographic
pilot data were unavailable, sample size was based on data from the Preterm
Prediction Study,11 which collected endovaginal
sonographic data at both 24 weeks' and 28 weeks' gestation. We assumed the
following: spontaneous preterm birth rate before 35 weeks for high-risk women
with a cervical length of 25 mm or greater would be 10%; a cervical length
less than 25 mm would occur in 20% of women; and the incidence of spontaneous
preterm birth before 35 weeks would be 30%. Considering also a desired effect
size of a relative risk (RR) of 3.0 for spontaneous preterm birth before 35
weeks (based on the presence or absence of a cervical length of <25 mm,
2-tailed α = .05, β = .20), 170 patients would have to be studied.
Data were analyzed using SAS version 7.0 (SAS Institute Inc, Cary, NC).
Categorical variables were compared using χ2 or the Fisher
exact test, and continuous data were compared with the Wilcoxon rank sum test.
Logistic regression was used to model the relationship between cervical length
and spontaneous preterm birth controlling for funneling, recognition of dynamic
shortening, and the slope of cervical length over time on serial evaluations
(as derived from linear regression models). Receiver operating characteristic
curves were used to compare the performance of varying cervical length cutoffs
for the prediction of spontaneous preterm birth before 35 weeks. Statistical
significance was represented at P<.05.
Since serial examinations were performed, we also compared spontaneous
preterm birth either with cervical length at the initial examination or the
shortest cervical length observed at any examination. Similarly, we analyzed
cervical length before and after any dynamic shortening occurred. Thus, for
any given patient, up to 4 different cervical lengths could be analyzed: (1)
the length at the initial evaluation before dynamic shortening; (2) the initial
length after dynamic shortening and considering the serial evaluations; (3)
the shortest observed length before dynamic changes; and (4) the shortest
observed length after dynamic shortening.
From all the participating centers, 236 women were initially thought
to be eligible for this study based on their stated obstetric history. A total
of 24 patients were found to be ineligible on review of their records. An
additional 19 patients declined to participate, plus 6 more patients verbally
agreed to participate but did not keep their first sonogram appointment and
therefore were not enrolled. We did not collect outcome data on these 25 women
who were eligible but not enrolled. From the original enrollment of 187, 4
patients were excluded because they were lost to follow-up.
A total of 590 endovaginal sonographic examinations were performed on
our study population of 183 women between March 1997 and July 1999. The median
duration of the sonographic examinations was 5.3 minutes (range, 4-18 minutes)
and the median number of scans per patient was 3 (range, 1-4). Nine of the
183 women underwent a single sonographic evaluation. Of these 9, 3 delivered
within 2 weeks of their first scan, before their next scheduled study visit.
Of the 590 sonographic evaluations, 576 (98%) were videotaped according to
protocol, and 466 (79%) of the taped examinations were later reviewed. After
study inception, 4 women received a cerclage by their managing physicians.
The 183 women in the study had a mean maternal age of 26 years (SD, 5 years);
119 (65%) were African American, 26 (14%) were white, and 38 (21%) were Hispanic.
The earliest prior delivery occurred at a mean of 24 weeks' (SD, 4.8 weeks')
gestation; 135 had a single prior preterm birth, 37 had 2 prior preterm births,
and 10 had more than 2 prior preterm births. On review, we determined that
1 patient had not experienced a prior spontaneous preterm birth before 32
weeks but, rather, had experienced an indicated preterm birth.
The mean gestational age at delivery was 35.2 weeks (SD, 6.3 weeks).
A total of 48 (26%) women experienced a spontaneous preterm birth before 35
weeks; 35 (19%) before 32 weeks; 29 (16%) before 28 weeks; and 20 (11%) before
24 weeks. An additional 5 women underwent an indicated preterm delivery at
31 weeks' to 34 weeks' gestation for obstetric complications. Of the 48 spontaneous
births before 35 weeks, 34 (71%) were associated with preterm labor and 14
(29%) were associated with preterm membrane rupture.
A total of 29 women (16%) had a poorly developed lower uterine segment
throughout their entire initial evaluation. Since these women had been arbitrarily
assigned a cervical length of 62 mm, the median baseline cervical length at
the first scan was 37 mm (range, 0-62 mm); the 10th percentile was 26 mm and
the 5th percentile was 23 mm. The relationship between cervical length at
the initial evaluation and spontaneous preterm birth before 35 weeks was modeled
with logistic regression. Women with shorter cervical lengths had correspondingly
higher rates of spontaneous preterm birth before 35 weeks (P<.001). From the regression model, we determined that the odds
of spontaneous preterm birth before 35 weeks decreased by 24% for each 5-mm
increase in baseline cervical length. We then examined various cervical length
cutoffs for their predictive accuracy (Table 1).
Since we had arbitrarily assigned a numeric cervical length value to
women with a poorly developed lower uterine segment, we evaluated separately
the predictive value of this finding. Of the 29 women with a poorly developed
lower uterine segment throughout their initial evaluation, only 3 (10%) experienced
a spontaneous preterm birth before 35 weeks compared with a 29% rate if the
lower uterine segment was not poorly developed (P
In 9 cases, the sonologist notified the managing physicians after the
sonogram had been performed. A total of 5 of these 9 cases were suspected
placenta previa and were reported according to study protocol. However, in
the other 4 cases, the protocol was not followed. Three were due to specific
cervical findings (cervical bending, funneling, and internal os dilation)
and in 1 case, the physician requested that the cervical length measurement
be unmasked. Considering the potential for bias associated with physician
notification and the 1 patient who had not previously experienced a prior
spontaneous preterm delivery, we determined the effect of omitting these 10
women from the analysis of the initial sonographic evaluation. The RR for
cervical length less than 25 mm and spontaneous preterm birth before 35 weeks
increased slightly from 3.3 to 3.6.
Funneling was noted in 16 patients (9%) at their initial evaluation.
These women were significantly more likely to have a spontaneous preterm birth
before 35 weeks (56% vs 23%; P = .004). However,
women with an observed funnel also had a significantly shorter cervical length
(median, 26 mm vs 38 mm if no funnel was observed; P<.001).
Because cervical length was such a strong predictor of spontaneous preterm
birth before 35 weeks, we also evaluated the finding of a funnel as a potential
independent predictor. The presence of a funnel was not a significant independent
predictor, controlling for cervical length in a logistic regression model
(P = .24). We also included either the presence of
funneling or a cervical length cutoff of less than 25 mm in a contingency
table with spontaneous preterm birth before 35 weeks and observed a lower
RR of 2.7 (95% confidence interval [CI], 1.7-4.3) and a lower positive predictive
value of 59% than when we used an isolated cervical length cutoff of less
than 25 mm (Table 1).
During their first sonographic evaluation, 16 (9%) of 183 women had
observed dynamic changes. A total of 9 followed fundal pressure and 7 were
spontaneous. The cervical lengths of these 16 women shortened from a mean
of 49 mm (median, 62 mm) to a mean of 30 mm (median, 25 mm); 2 of these women
also developed a funnel. In 6 cases, the initially observed, poorly developed
lower uterine segment resolved. Similar to our analysis of funneling, we included
dynamic changes in a logistic regression model with cervical length and observed
a trend toward dynamic changes as a significant independent predictor of spontaneous
preterm birth before 35 weeks (P = .054). We also
considered dynamic changes in a contingency table with a cervical length cutoff
of less than 25 mm and spontaneous preterm birth before 35 weeks. As with
funneling, the inclusion of dynamic changes at the initial evaluation did
not improve the predictive accuracy of a cervical length cutoff of less than
25 mm (RR, 2.4; 95% CI, 1.5-3.8; positive predictive value, 52%).
Serial evaluations demonstrated that cervical length shortened from
a median of 37 mm at the first scan to a median of 32 mm at the fourth scan.
For each of the 174 women with at least 2 sonographic evaluations, we computed
the rate of change of cervical length by fitting a linear regression line
to their observed cervical length measurements. The median rate of shortening
in this group was 1.1 mm per week. Removing the 41 women who had a poorly
developed lower uterine segment and therefore an assigned cervical length
of 62 mm at any time during their initial and serial evaluations, we observed
a median cervical length shortening of 0.9 mm per week. The 44 women who experienced
a spontaneous preterm birth before 35 weeks shortened their cervixes at a
median rate of 2.5 mm per week compared with a rate of 1.0 mm per week in
the 130 women who did not (P = .03).
To determine the effect of serial observations on the predictive accuracy
of endovaginal sonography, we first included the shortest observed cervical
length for each patient in a logistic regression model with spontaneous preterm
birth before 35 weeks as the dependent variable. In this analysis, the shortest
ever observed cervical length before dynamic changes was a significantly better
predictor than the baseline cervical length at the first scan. We further
analyzed the information from serial evaluations by including the slope of
the derived regression line of cervical length over time before dynamic changes
in a logistic regression model, alone, and also with the cervical length at
the first evaluation. The slope of length over time was not a significant
predictor of spontaneous preterm birth before 35 weeks (P = .07). However, after controlling for initial baseline length, the
slope became a statistically significant predictor in the regression model
(P = .002).
Since previous reports examined the relationship between static cervical
length measured beyond 20 weeks' gestation and spontaneous preterm birth,
we performed a secondary analysis of all sonographic evaluations performed
at or beyond 21 weeks' gestation prior to any dynamic changes. If a patient
had undergone 2 studies during this gestational period, the former was preferentially
selected. In this subgroup of 142 women, the RR of a cervical length less
than 25 mm and spontaneous preterm birth before 35 weeks was 3.5 (95% CI,
1.9-6.5). The associated sensitivity and specificity were 46% and 87%, respectively.
Finally, we examined the clinical utility of the shortest observed cervical
length not considering dynamic changes on serial scans, using a cutoff of
less than 25 mm for the prediction of spontaneous preterm birth before 35
weeks (Table 2). Since cervical
lengths tended to shorten over time, more than 4 times as many women (n =
53 vs n = 12) were found to have a cervical length of less than 25 mm during
We considered the additional effect of spontaneous or fundal pressure–induced
dynamic changes that were observed on serial examinations. We included the
shortest observed cervical length after any dynamic changes occurred in a
logistic regression model with spontaneous preterm birth before 35 weeks as
the dependent variable and found that it was a significantly better predictor
than the shortest observed cervical length at any scan prior to dynamic changes.
From the regression model, we determined that the odds of spontaneous preterm
birth before 35 weeks decreased by 43% for each 5-mm increase in the shortest
observed cervical length after dynamic changes.
Considering these dynamic changes, the median rate of shortening remained
1.1 mm per week and 0.9 mm per week after removing the 41 women who had a
poorly developed lower uterine segment. We also analyzed cervical length over
time after dynamic changes using logistic regression. In this case, the slope
was a significant predictor of spontaneous preterm birth before 35 weeks by
itself (P<.001) and also after controlling for
initial baseline cervical length (P<.001). These
analyses confirmed that the inclusion of dynamic changes (ie, shortening)
observed on serial evaluations significantly improved the predictive accuracy
of endovaginal sonography for a spontaneous preterm birth.
We then examined the summary predictive values of postdynamic change–cervical
length measurements at a cutoff of less than 25 mm (n = 60) for the prediction
of spontaneous preterm birth before 35 weeks. Table 2 contains the summary predictive values for a cervical length
cutoff of less than 25 mm at the baseline scan, the shortest observed cervical
length on serial scans before any dynamic changes, and the shortest observed
cervical length considering dynamic changes from the serial endovaginal sonographic
Figure 3 depicts the receiver
operating characteristic curves of the baseline cervical length at 16 to 18
weeks' gestation prior to dynamic changes and the shortest observed cervical
length observed on serial evaluations after dynamic changes. The latter measurement
represented a statistically significant improvement over the former with regard
to the use of cervical length as a screening test for the prediction of spontaneous
preterm birth before 35 weeks (P = .03).
We performed a prospective, blinded observational study to determine
if endovaginal sonography of the cervix at 16 weeks' to 23 weeks 6 days' gestation
would predict spontaneous preterm birth with sufficient accuracy to justify
mid-trimester intervention trials in high-risk women. As a single measurement,
cervical length of less than 25 mm at 16 to 18 weeks' gestation was a significant
predictor of spontaneous preterm birth before 35 weeks, and the inclusion
of dynamic shortening and serial observations of cervical length improved
the predictive values. We conclude that the natural history of cervical anatomy
during midpregnancy can contribute significant information as to the risk
of subsequent spontaneous preterm birth.
We had previously recognized that in approximately 10% of mid-trimester
endovaginal sonograms, the cervical anatomy appeared atypical and, in particular,
a normal-appearing internal os could not be readily identified. This led to
our characterization of a poorly developed lower uterine segment, which in
some cases resolved into a measurable cervix. If resolution occurred during
the sonogram, this represented a dynamic change. We recognized that, in a
few of these transient cases, the cervical length measurement after dynamic
change was actually shortened (<25 mm). In other cases, the poorly developed
lower uterine segment persisted throughout the entire examination, but resolved
before the patient's next visit. In no cases did this finding persist during
all scheduled evaluations. We observed that the incidence of poorly developed
lower uterine segments decreased from 16% at the first scan to less than 2%
by the fourth evaluation. The finding of a poorly developed lower uterine
segment throughout the entire scan appeared to be protective and justified
our decision to consider it as a "long" cervix in the analyses. Although primarily
a subjective diagnosis, we have summarized diagnostic criteria and believe
that it represents a reproducible observation with biological significance.
Our findings challenge previous reports that funneling at the internal
cervical os is a useful predictor of preterm birth.11,22,23
We were impressed by the wide range of biological variability associated with
funneling, which might limit the reproducibility of this finding. For example,
some women did not have a distinctly recognizable shoulder above the functional
internal os, depicted in schematic diagrams of funneling, and thus caliper
placement was operator dependent. Measurement of funnel width as the distance
between shoulders would also have been problematic since some women had only
1 recognizable shoulder. In other cases, asymmetric shoulders occurred, so
it was the sonologist's choice as to which one was used for funnel depth measurement.
Based on these observations, we included funneling as a categorical variable
in the analyses.
Although women with a funnel had significantly shorter cervical lengths
than women with no observed funnel, our analyses confirmed that most, if not
all, of the preterm birth risk was related to cervical length. We postulate
that some cervixes shorten through the process of funneling, but that the
remaining functional length is more important than the precise method of shortening.
However, because our sonographic examination windows were necessarily limited
(nominally, 20 minutes of real-time observations over 6 weeks), it is plausible
that some women with a shortened cervical length had previously experienced
funneling that was never observed.
Dynamic change, after controlling for cervical length, was only a marginally
significant predictor of preterm birth; however, dynamic cervical length shortening
during serial evaluations significantly improved the prediction of preterm
birth. Fundal pressure as a provocative maneuver has been evaluated in women
at risk for cervical incompetence.21 We purposely
excluded women from our study who had undergone cerclage for a clinical history
of cervical incompetence and also recorded unprovoked, spontaneous dynamic
shortening.23 Thus, patient selection likely
explains why fundal pressure–induced dynamic changes were not commonly
observed. Spontaneous dynamic changes were also uncommon and, as independent
findings, did not further improve the predictive value of shortened cervical
length for spontaneous preterm birth. We conclude that cervical length is
the single most important sonographic finding for preterm birth prediction
in high-risk women. Although the precise mechanisms by which the cervix shortens
and contributes to spontaneous preterm birth may ultimately be shown to have
a differential impact on specific interventions, our results support the concept
that for the prediction of spontaneous preterm birth, the means by which the
cervix shortens may not be as important as the fact that it does shorten.
Gestational age at examination, obstetric history, concurrent risk factors
for preterm birth (eg, multiple gestation), subsequent uncontrolled interventions,
the gestational age used to define the preterm outcome, and other aspects
of study design likely explain the observed variance among published reports
on the significance of cervical length measurement for preterm birth prediction.9- 14,16- 18,23
For these reasons, we defined prematurity as delivery before 35 weeks, which
is more clinically relevant than the traditional 37 week end point; included
only women with a prior early spontaneous preterm birth; and masked the sonographic
results. We chose this study population primarily because it is one of the
largest and most readily identified groups at risk. Although women with multiple
gestations also comprise a sizable and homogeneous risk group, the mechanisms
by which spontaneous preterm birth occurs in multiple gestations may be different
than the pathways that lead to recurrent spontaneous preterm birth in singleton
To be clinically useful, the measurement of cervical length should be
reproducible and associated with reasonable thresholds for intervention. From
our quality assurance reviews, we were satisfied that our training and certification
resulted in standardized measurements of cervical length among participating
sonologists. Since reports of cervical length assessment with endovaginal
sonography have become increasingly common in recent years, it is likely that
many centers have developed their own training and certification protocols.
Nevertheless, from our sonologist certification process, we recognized a learning
curve associated with this technique and caution against the use of cervical
length assessment by sonologists who have not had appropriate supervised experience.
With regard to intervention thresholds, we recognize that the relationship
between cervical length and spontaneous preterm birth functions along a continuum
as depicted in a receiver operating characteristic curve (Figure 3). Therefore, no single cervical length cutoff can completely
discriminate between eventual term and preterm births. Depending on the risks,
effectiveness, and costs of a particular intervention, different thresholds
may be appropriate. We believe that a cervical length cutoff of less than
25 mm represents an optimum threshold for inclusion in future mid-trimester
intervention trials of cerclage. However, our findings do not support the
concept of a "normal" vs "abnormal" cervical length, which is oversimplified.
Our findings support the concept that cervical "competence" likely represents
a continuum,11,24- 27
and that the mechanisms that underlie the syndrome of spontaneous preterm
birth are multifactorial and incompletely understood.28
Further investigations combining endovaginal sonography and other markers
of spontaneous preterm birth may increase our understanding of these mechanisms
and permit a more individualized and biologically focused approach to preterm
birth prevention. Until properly designed trials of cerclage or other interventions
prove a benefit from the finding of a "short" cervix in the mid-trimester,29- 32 we
recommend that cervical length measurement in women with a prior spontaneous
preterm birth remain investigational.