We sought to evaluate a recently proposed protocol whereby transabdominal ultrasound of the cervix might be used as a prescreen to select women to undergo or to forgo measurement of cervical length via transvaginal ultrasound (CL vag ).
This was a prospective cohort study. Measurements of cervical length via transabdominal ultrasound (CL abd ) and CL vag were made in women with singleton pregnancy during routine obstetrical ultrasound examination at 18 0/7 to 23 6/7 weeks of gestation. The transabdominal screen was considered positive if CL abd was ≤36 mm with the maternal bladder full or ≤35 mm with the bladder empty, or adequate imaging of the cervix could not be obtained. Sensitivity, specificity, predictive values, and likelihood ratios of a positive screen to detect a short cervix (CL vag ≤25 mm) were calculated.
An interim analysis identified several technical problems with CL abd measurements, so the protocol was extensively revised. Under the revised protocol, 1580 women were included. Adequate views of the cervix were obtained via transabdominal imaging in 46% of subjects with the bladder empty and 56% with the bladder full. The correlation between CL abd and CL vag was poor (r = 0.38). Of the 17 patients with a short cervix, 15 had suboptimal transabdominal exams (screen positive) and 2 had CL abd ≤35 mm with bladder empty (screen positive). Sensitivity of the screen was 100% (95% confidence interval, 80.5–100%) but specificity was only 32.2% (95% confidence interval, 29.9–34.6%) and screen positive rate was 66.3%. Several technical problems and limitations of transabdominal imaging of the cervix are shown.
Using modern, high-resolution ultrasound equipment, we were unable to adequately image the cervix via transabdominal ultrasound in half the cases. Although we confirmed that a CL abd cutoff value of 35-36 mm is appropriate for detection of short cervix, the technique for measuring CL abd is fraught with technical problems. Practitioners must validate the technique in their own practice before adopting this or similar prescreening protocols. We decided not to adopt this protocol.
A short uterine cervix in the second trimester is associated with increased risk of early preterm birth (PTB). Because there are effective treatments such as cerclage or vaginal progesterone to reduce the risk of PTB in women with a short cervix, universal ultrasound screening of cervical length (CL) in asymptomatic women with singleton pregnancies is reasonable and appears to be both cost-effective and cost-saving.
Transvaginal ultrasound is required for accurate measurement of CL. With transabdominal imaging, adequate views of the cervix are frequently unobtainable, especially when the bladder is empty or the cervix is short. Filling the bladder to improve the imaging artifactually lengthens the cervix, which might reduce the detection of a short cervix. Furthermore, even when adequate images are obtained, the CL via transabdominal ultrasound (CL abd ) often correlates poorly with CL via transvaginal ultrasound (CL vag ).
Transvaginal examination imposes burdens on the patient (need to undress, discomfort, embarrassment, additional time, additional cost) and on the practice (additional time for sonographer and chaperone, equipment expense, resources to sterilize transducers between exams). Despite the limitations of transabdominal ultrasound, it would be highly desirable if CL abd could be used as a prescreen to determine which women should undergo transvaginal examination.
Friedman et al recently reported findings that would support such a prescreening. They proposed that transvaginal screening could be reserved for those with either CL abd ≤36 mm (before voiding the bladder) or CL abd ≤35 (postvoiding) or with inadequate CL abd imaging. Their results suggested that such a protocol would have >96% sensitivity for detection of a short cervix (CL vag ≤25 mm) and would allow 40% of women to avoid transvaginal screening. However, their 35- to 36-mm cutoff was selected retrospectively and the test performance was evaluated in the same subject group from which it was derived. Before such a prescreening protocol can be recommended for widespread use, the cutoff values and test performance need validation in an independent cohort.
The purpose of the present investigation was to prospectively evaluate the test performance characteristics of the CL cutoff values proposed by Friedman et al.
Materials and Methods
Our ultrasound practice has been accredited by the American Institute of Ultrasound in Medicine (AIUM) since 2002. Following guidelines of AIUM and the American Congress of Obstetricians and Gynecologists (ACOG), we routinely image the cervix transabdominally on all second- and third-trimester ultrasounds. For women with history of a spontaneous PTB, we perform serial CL vag screening from 16 0/7 to 23 6/7 weeks, as recommended by ACOG and the Society of Maternal-Fetal Medicine (SMFM). In June 2012 we began offering universal CL vag screening to women without prior PTB, consisting of a single exam at 18 0/7 to 23 6/7 weeks at the time of a routinely scheduled ultrasound exam (usually for fetal anatomy survey), consistent with SMFM guidelines. Women were allowed to opt out of CL vag screening.
Prior to initiating the study, we were granted a waiver of the requirement for informed consent by the Western Institutional Review Board (IRB) because this was viewed as a quality improvement study and the protocol involved no deviations from routine care other than placing calipers on static images to measure CL abd and collection of data onto forms without patient identifiers.
We recorded CL abd and CL vag measurements for all women who underwent CL vag screening in our office starting July 1, 2013. The IRB-approved data collection form included gestational age, history of spontaneous PTB, singleton vs multiple pregnancy, presence or absence of cerclage, whether the patient had a prior CL vag screening in our office, and status of the maternal bladder (full or empty) during the CL abd exam.
Inclusion criteria for this study were: singleton pregnancy, gestational age 18 0/7 to 23 6/7 weeks, no cerclage in place, and no prior CL vag exam (to exclude multiple observations in the same woman).
All exams were performed using GE Voluson 730 or E8 machines (General Electric Healthcare, Wauwatosa, WI). We used variable-frequency 1- to 5-MHz or 6-MHz transducers for transabdominal imaging and 6- to 12-MHz or 5- to 9-MHz transducers for transvaginal imaging.
The technique for CL abd imaging followed that of Friedman et al. Before study initiation, each sonographer had a 1-on-1 inservice to review the publication and the technique, with instructions to obtain sagittal images with attention to these landmarks: the cervical/vaginal interface, the internal cervical os, the external cervical os, the outline of the cervical corpus, and the full length of the cervical canal. To give sonographers and physicians familiarity with the process, we had a 2-week run-in practice period before initiation of the study. Sonographers were instructed to record the CL abd or to enter a code for “suboptimal” if the landmarks could not be obtained. We did not specify whether the bladder should be full or empty for the CL abd exam, but we recorded the bladder status.
Vaginal ultrasound was performed after completion of the abdominal exam, usually by the same sonographer who had performed the abdominal exam. The technique for CL vag imaging followed that of Berghella et al. Briefly, with the bladder empty and the patient in stirrups, the sterile-sheathed transducer was inserted to touch the cervix, withdrawn until the image blurred, and gently advanced just enough to clear the image. Three sagittal images were obtained at rest and 3 during and after fundal pressure; the shortest value was recorded. Each of our sonographers and physicians had obtained a Certificate of Competence in Cervical Assessment from the Fetal Medicine Foundation (London, United Kingdom) before initiation of the study.
Sample size was based on an expected 10% prevalence of short cervix (CL vag ≤25 mm). We calculated that 1500 subjects would yield 95% confidence intervals (CIs) of ±4% around a sensitivity of 96%, ±3% around a specificity of 40%, and 1.5% around a negative predictive value of 98% as reported by Friedman et al. Interim analyses were planned upon completion of 800 and 1200 subjects.
At the first interim analysis, several problems with the abdominal ultrasound technique were noted and we revised the protocol to address these ( Results section). Exams prior to the protocol revision were designated phase 1 (training) and are not included in the main analysis for this report. In essence, we rebooted the study after the revision, with a new phase 2, which forms the bulk of this report.
Statistical analyses were performed with software (Stata, version 13.1; StataCorp, College Station, TX). Continuous variables were tested with t test or paired t test and categorical variables with χ 2 . Two-tailed P values < .05 were considered significant.
To test the protocol of Friedman et al, each abdominal exam was categorized as either screen negative (low risk) or positive (suspicious). An exam was considered positive if: (1) CL abd was ≤36 mm with bladder full; (2) CL abd was ≤35 mm with bladder empty; or (3) transabdominal imaging was recorded as suboptimal regardless of bladder status. The test characteristics (sensitivity, specificity, predictive values, likelihood ratios) of a positive result in the prediction of short cervix (CL vag ≤25 mm) were calculated. Receiver operating characteristic curves were plotted, summarizing the association of CL abd with short cervix.
From July 2013 through May 2014, we performed 3113 paired CL abd and CL vag exams. Figure 1 summarizes the exclusions and the final sample size in each study phase. During the study period, 13.4% of patients opted out of CL vag screening and are not included in this report.
At the first planned interim analysis (after 800 eligible cases), there were 8 cases with short cervix (CL vag ≤25 mm), 3 of which would not have been predicted by the protocol of Friedman et al (CL abd 37, 39, and 49 mm). Upon reviewing these cases and others, we noted several problems with our technique for obtaining CL abd .
First, in 2 of the cases with discordant results, both the transabdominal and initial transvaginal images appeared to show a normal CL, as illustrated in Figure 2 . However, with the application of fundal pressure during the transvaginal exam, a funnel opened at the internal os and the residual CL was then short (24 mm in case 1, 20 mm in case 2). In other words, transvaginal imaging at rest appeared falsely reassuring until fundal pressure was applied. We were concerned that transabdominal ultrasound might similarly yield falsely reassuring findings unless fundal pressure was also applied as part of the exam.
Second, we noted that CL abd images were usually obtained at the beginning of the exam, often ≥30 minutes prior to the CL vag images. We were concerned that the time delay might have contributed to the poor overall correlation observed between CL abd and CL vag (r = 0.33).
Third, we found significant differences between sonographers in the percentage of exams that they recorded as suboptimal.
Fourth, we found that the rate of suboptimal exams rose dramatically over the early course of the study, from <1% in the first month to 4%, 24%, and 19% in subsequent months, indicating the presence of a learning curve during the first 2 months.
Fifth, we found that many exams recorded a CL abd value based on the length of the cervical corpus, even though the cervical canal was not clearly demonstrated. Case 2 in Figure 2 is an example of this.
We revised the protocol based on these findings. First, we specified that CL abd should be measured before, during, and after fundal pressure, the same as for CL vag measurement. Second, we specified that CL abd should be the last imaging obtained during the transabdominal exam, immediately before the transvaginal exam. Third, we specified that a physician must accept or reject the adequacy of the CL abd images before the transvaginal exam was performed. Finally, we allowed the inclusion of exams in which the external cervical os was not demonstrated, providing that at least 40 mm of the cervical canal contiguous with the internal os were visible.
With adoption of the new protocol, we terminated phase 1 and began anew with phase 2. The following analysis is based on the 1580 subjects from phase 2 who met all the inclusion criteria.
Mean gestational age at the time of the exam was 19.6 ± 1.1 weeks. History of spontaneous PTB was reported by 73 women (4.6%).
Imaging of CL abd was judged suboptimal in 50.8% of all exams. Suboptimal CL abd imaging occurred significantly more often with the bladder empty (54.0%, 1080 exams) than with the bladder full (44.0%, 500 exams, P < .001). Suboptimal CL abd imaging also occurred significantly more often with CL vag ≤25 mm (88.2%, 15 of 17 cases) than with CL vag >25 mm (50.4%, 788 of 1563 cases, P < .005).
Examples of technical problems, pitfalls, and artifacts leading to suboptimal imaging of CL abd are shown in Figures 3 to 6 . These 30 cases were selected by a random sampling algorithm from the entire phase 2 dataset. Several problems with transabdominal imaging are evident, including shadowing or other obscuration by the fetus (cases E, M, N, O, V), bladder edge-artifact obscuring parts of the cervix (cases F, G, J, L, X, and Dd), unexplained shadowing obscuring the cervical canal (case T), or inability to see all or part of the cervical canal for other reasons (cases A, B, C, P). Also evident in these figures is that bladder filling often enhances clear imaging of the cervix (cases L, R, S, U, Y, Z, Aa) but not always (cases B, J, Dd). CL abd was recorded as suboptimal in 13 (43%) of these cases (cases C, E, F, G, H, J, M, N, O, P, V, X, and Aa), including the 2 in which the cervix was short (CL vag ≤25 mm) by transvaginal examination (16 mm in case E and 24 mm in case G).
In exams with a full bladder and adequate CL abd , mean CL abd was 39.0 ± 6.7 mm and CL vag was 41.3 ± 6.0 mm ( P < .001 for the difference, paired t test). With an empty bladder and adequate CL abd , mean CL abd was 39.8 ± 6.9 mm and CL vag was 41.8 ± 6.0 mm ( P < .001, paired t test). There was no significant difference in the mean CL abd comparing exams with full vs empty bladders ( P = .16, 2-sample t test).
There was a moderately poor correlation between CL abd and CL vag ( Figure 7 ) (r = 0.38), similar to phase 1 (r = 0.33, P = .14, Fisher r-to-z transformation test). The regression line relating CL abd to CL vag was not significantly different for exams with full bladders vs empty bladders (interaction P = .53 in multiple regression). Bland-Altman plots ( Figure 7 , B and C) show a small systematic bias (CL vag slightly longer than CL abd ) but wide confidence limits (±15-16 mm difference between CL vag and CL abd ) regardless of bladder status, indicating poor agreement between the 2 measures.
Short cervix (CL vag ≤25 mm) was found in 17 cases (1.1%). Of these, 15 (88%) had suboptimal CL abd imaging and were therefore considered screen positive. The CL abd values in the other 2 cases were 27 mm and 34 mm, also screen positive. Thus, the sensitivity of the protocol was 100% (95% CI, 80.5–100%), mostly because of suboptimal transabdominal imaging of the cervix. The other diagnostic performance parameters of this protocol for the prediction of CL vag ≤25 mm are summarized in Table 1 . The receiver operating characteristic curves relating CL abd to short cervix are shown in Figure 8 . The 35- to 36-mm cut point proposed by Friedman et al had high sensitivity but poor specificity in our data. Among the entire cohort, 66.3% were screen positive.
|Variable||Transvaginal cervical length a|
|Transabdominal cervical length||Normal (>25 mm)||Short (≤25 mm)|
|>36 mm (screen negative)||533||0|
|≤36 mm, bladder full (screen positive)||107||0|
|≤35 mm, bladder empty (screen positive)||135||2|
|Suboptimal (screen positive)||788||15|
|Test characteristic||Point estimate||(95% CI)|
|Positive predictive value||1.58%||(0.9–2.5%)|
|Negative predictive value||100%||(99.3–100%)|
|Positive likelihood ratio||1.47||(1.43–1.53)|
|Negative likelihood ratio||0||ND|
|Odds ratio||ND||(2.1 to ND)|