Cervical length measurement for the prediction of preterm birth in multiple pregnancies: a systematic review and bivariate meta-analysis

Authors


Abstract

Objectives

To review the literature on cervical length as a predictor of preterm birth in asymptomatic women with a multiple pregnancy.

Methods

We searched MEDLINE, Embase and reference lists of included articles to identify all studies that reported on the accuracy of cervical length for predicting preterm birth in asymptomatic women with a multiple pregnancy. We scored study characteristics and study quality, and extracted data in order to construct two-by-two tables cross-classifying cervical length and preterm delivery. Meta-analysis using a bivariate model was performed. Summary receiver–operating characteristics (ROC) curves were generated for various test characteristics and outcome definitions.

Results

We found 21 studies reporting on 2757 women. There was a large variation in gestational age at measurement, cut-off point for cervical length and definition of preterm birth. The summary ROC curve indicated a good predictive capacity of short cervical length for preterm birth. Summary estimates of sensitivity and specificity for preterm birth before 34 weeks’ gestation were 78% and 66%, respectively, for 35 mm, 41% and 87% for 30 mm, 36% and 94% for 25 mm and 30% and 94% for 20 mm.

Conclusions

In women with a multiple pregnancy, second-trimester cervical length is a strong predictor of preterm birth. In the absence of effective preventive strategies, there is currently no place in clinical practice for cervical length measurement in this population. However, future studies should evaluate preventive interventions in women with multiple pregnancies and a short cervix, and cervical length should be measured in any trial studying preventive strategies in multiple pregnancies. Copyright © 2011 ISUOG. Published by John Wiley & Sons, Ltd.

Introduction

Despite the enormous advances in neonatal care during recent years, preterm birth remains the major cause of handicap in children without congenital anomalies or genetic disorders. Many studies have been conducted into the clinical relevance of cervical length as a predictor of preterm birth. In both symptomatic and asymptomatic pregnant women, it has been shown that transvaginal sonography of the cervix is able to identify women who are at increased risk for preterm birth1. This finding has led to randomized controlled trials in which women with asymptomatic cervical shortening during the second trimester were treated with progesterone or placebo. In the first trial that was reported, daily vaginal progesterone turned out to reduce the risk of preterm birth by 40%2. Similar results with progesterone had previously been obtained in women with a singleton pregnancy and a history of preterm birth3, 4. In unselected women with multiple pregnancies, progesterone has appeared to be ineffective in reducing preterm birth rates so far5, 6.

The mechanism of preterm birth in women with a multiple pregnancy is likely to differ from that in women with singleton pregnancies. The human body often seems to lack the natural capacity to carry multiple fetuses to term, whereas the cause of preterm birth in singleton pregnancies is more likely to be found in individual maternal or fetal factors. However, it might be possible that these individual factors also play a role in a minority of women with a multiple pregnancy. Consequently, tests used to predict preterm birth in women with a singleton pregnancy may be useful in women with a multiple pregnancy, although the relative over-distension of the uterus in multiple pregnancies might affect the cervix and therefore the predictive value of cervical length measurement.

We conducted a systematic review to assess whether sonographically measured cervical length can be used to predict preterm birth in multiple pregnancies. We limited our search to asymptomatic women in the second trimester. We planned a meta-analysis using bivariate regression analysis, accounting for correlation between sensitivity and specificity.

Methods

Electronic searches

With the help of a librarian, we carried out electronic searches of MEDLINE and Embase from inception (1966 and 1947, respectively) to January 2009. The search strategy consisted of MeSH or key terms related to multiple pregnancy, cervix and preterm birth. We checked reference lists of relevant studies to identify cited articles not captured by electronic searches and contacted authors of primary studies where contact addresses were available. We aimed to identify studies that reported on cervical length as a predictor of preterm birth in asymptomatic women with a multiple pregnancy. We did not use any restricting criteria for the study design. Reference Manager 11.0 (Thomson ISI ResearchSoft, Carlsbad, CA, USA) databases were established incorporating results of all searches.

Selection of studies

Articles identified through the initial search were first screened on title and, where available, on abstract by two independent reviewers (A.L. and M.H.V.). Studies were included if they reported on women with a multiple pregnancy in whom cervical length was sonographically measured during pregnancy, and for whom gestational age at birth was known. When studies could not be excluded based on their title or abstract, a full manuscript was obtained. These were then independently scored by two reviewers (A.L. and M.H.). When a manuscript was written in a language other than English or Dutch, the article was translated by a colleague with both fluency in the language and expertise in the subject area. At this stage a fourth inclusion criterion was added: the reviewer had to be able to subtract a two-by-two table cross-classifying cervical length and gestational age from the article. In cases where this was not possible, but where the results that were presented indicated that the original data would allow for the generation of a two-by-two table, the authors were contacted by e-mail and/or post. We included studies examining pregnant women at any level of risk and in any healthcare setting. Any disagreements were resolved by consensus or a third reviewer (B.M.). No language restrictions were applied.

Data extraction

For all articles that were included, two reviewers (A.L. and M.H.) independently extracted descriptive data (first author and year of publication), study design characteristics, test characteristics (transvaginal, transperineal or abdominal ultrasound and cut-off values for cervical length and gestational age at measurement), definition of outcome (cut-off values for preterm birth) and data on study quality and test accuracy, using a pre-designed and piloted data extraction form.

The two reviewers independently assessed all included manuscripts for study quality according to an adapted version of QUADAS7. Studies were classified as having an adequate description of the testing procedure when at least the following two aspects were mentioned: (1) the bladder was emptied before testing; (2) whether or not funneling was included in the cervical length.

Data synthesis

We calculated sensitivity and specificity with 95% confidence intervals (CIs) for each study individually, created forest plots to explore heterogeneity for sensitivity and specificity and plotted their combined results in receiver–operating characteristics (ROC) space (sensitivity vs. 1 − specificity). Bivariate regression analysis was used to obtain summary estimates of sensitivity and specificity and their 95% CIs, and to construct summary ROC (sROC) curves8. With a bivariate regression model summary estimates can be calculated simultaneously for sensitivity and specificity within a single model. Sensitivity and specificity within a study are often negatively correlated, owing to implicit variation of threshold values. The bivariate regression model statistically incorporates the negative correlation that might exist between sensitivity and specificity. Variation or heterogeneity between the results of the studies included in the meta-analysis can be the result of differences in thresholds, but also of chance, bias due to flawed design, different clinical subgroups and unexplained variation. When necessary, the bivariate model uses a random effects approach, due to which clinical heterogeneity beyond chance is accounted for.

Different studies often reported a different threshold value to define a positive test result for cervical length, and several studies also reported accuracy for multiple threshold values. In addition, studies used different threshold values for gestational age at delivery to define preterm birth. If we estimated accuracy for a single combination of threshold values for cervical length and preterm birth, this estimate would be based on only a limited number of studies. Furthermore, it is not clear which is the appropriate threshold for either definition. In order to evaluate accuracy measures over the whole range of possible thresholds, however, we did not limit our analysis to a single threshold value, but estimated accuracy measures for all reported threshold values by assuming that the shift in accuracy (higher sensitivity and lower specificity) due to different thresholds is accounted for by the correlation term, as specified in the bivariate model.

As the recommended bivariate modeling approach cannot appropriately account for covariates (such as threshold) with multiple observations from the same study, we used the following strategy to statistically incorporate differences in cervical length thresholds and definitions of preterm birth in the model. We performed three types of analysis, and integrated their results into summary estimates for different cervical length and preterm birth threshold combinations. The first type includes all reported accuracy estimates, irrespective of threshold values for cervical length and preterm birth or gestational age at which cervical length was measured; the second type analyzes the bivariate model for four different cervical length thresholds (20, 25, 30 and 35 mm); and the third type fitted the bivariate model for three preterm birth thresholds (29, 34 and 37 weeks).

In order to avoid results being biased towards studies reporting on many different thresholds, we estimated each model in 100 stratified bootstrap samples, in which only one accuracy estimate from each study was randomly selected. For each parameter the average overall estimates from 100 bootstrap samples is reported. The results of the model were used to estimate summary ROC curves, where the increase in sensitivity and decrease in specificity reflect the shift in threshold value of cervical length in the model, resulting in separate ROC curves for different criteria to define preterm birth (< 34 weeks, > 34 weeks). In a subgroup analysis, we also performed separate analyses for twins and triplets.

Results

Figure 1 summarizes the flow of studies through the review. Out of 382 potentially relevant abstracts, 21 studies screening 2757 women were included. Authors of a further 13 studies were contacted about unclear information in their articles. None of the contacted authors provided information that could clarify our uncertainties, the main reason given being that ‘data were no longer available’. These studies were therefore not included in our analysis.

Figure 1.

Flow chart to illustrate study method.

Twelve studies were prospective and nine studies were retrospective cohort studies. Sample sizes ranged from 14 to 383 women. Table 1 shows the study characteristics of all included studies. Most studies reported on several measurement times, cut-offs for cervical length and cut-offs for preterm birth, allowing for multiple two-by-two tables to be extracted from one study. The number of two-by-two tables per study ranged from 1 to 20.

Table 1. Study characteristics
StudyInclusion criteriaExclusion criterianGA at testing (weeks)Cut-off points for cervical length (mm)Cut-off points for GA at birth (weeks)
  1. ART, assisted reproductive technology; DES, diethylstilbestrol; GA, gestational age; IUGR, intrauterine growth restriction; PTB, preterm birth; TTTS, twin-to-twin transfusion syndrome.

Asymptomatic twin pregnancies
 Arabin et al.20Twin pregnancyIatrogenic PTB15315 to 19 + 6,15, 20, 25, 3036
     20 to 24 + 6,  
     25 to 29 + 6  
 Fait et al.21Twin pregnancyFetal anomalies,2015–173533
  (selective reduction placental abruption,    
  from triplet) IUGR,    
   pre-eclampsia    
 Gibson et al.22Twin pregnancyFetal anomalies,8218, 24, 28, 3220, 22, 2535
   TTTS    
 Goldenberg et al.23Twin pregnancyCerclage,14722–24,2532, 35, 37
   fetal anomalies,  27–28  
   placenta previa    
 Guzman et al.24Twin pregnancyCerclage,11715–20,2028, 30, 32, 34
   iatrogenic PTB  21–24,  
     25–28  
 Imseis et al.25Twin pregnancyCerclage,8524–263534
   iatrogenic PTB    
 Klein et al.26Twin pregnancyFetal anomalies,22320–2525, 30, 3534
   ruptured membranes,    
   vaginal blood loss,    
   maternal pathology    
 McMahon et al.27Twin/triplet pregnancyCerclage,10920, 243032
   ruptured membranes,    
   iatrogenic PTB    
 Naba28Twin pregnancy1424–3425, 3037
 Robyr et al.29Twin pregnancy treatedIatrogenic PTB13716–2620, 25, 30, 3528, 32, 34
  for TTTS     
 Sayin et al.30Twin pregnancyCerclage,43422–2415, 20, 2533
   iatrogenic PTB    
 Soriano et al.31Twin pregnancy,Uterine anomalies,4418–243534
  nulliparous, DES exposure,    
  ART selective reduction,    
   iatrogenic PTB    
 Souka et al.32Twin pregnancyCerclage, TTTS21222–2415, 25, 35, 4528, 30, 32, 34
 Sperling et al.33Twin pregnancyIatrogenic PTB,3832321, 26, 31, 3628, 32, 33, 34, 35
   cerclage,    
   prior conization    
 Vayssiere et al.34Twin pregnancyCerclage,34521–23,25, 3032, 35
   fetal anomalies,  26–28  
   TTTS,    
   placenta previa    
 Yang et al.9Twin pregnancyCerclage,6518, 22, 2625, 30, 3532, 35, 37
   placenta previa,    
   preterm labor,    
   vaginal blood loss,    
   iatrogenic PTB    
Asymptomatic triplet pregnancies
 Guzman et al.35Triplet pregnancyCerclage,5015–20,20, 2528, 30, 32
   iatrogenic PTB  21–24,  
     25–28  
 Maslovitz et al.36Triplet pregnancy,Selective reduction,3614–202532
  trichorionic iatrogenic PTB    
 Maymon et al.37Triplet pregnancyCerclage,3423, 26, 292533
   selective reduction,    
   uterine contractions    
 Missfelder-Lobos et al.38Triplet pregnancy2920–26NoneNone
 To et al.39Triplet pregnancy3822–2415, 25, 3033

Quality assessment

Table 2 summarizes the results of the quality assessment. Of the 16 studies with an adequate description of the test, the study by Yang et al.9 was the only one that included funneling in cervical length. Several authors referred to previous publications by others for a description of the testing technique. Most often cited in this context were studies by Andersen et al.10 and Iams et al11. Practitioners were blinded to the results of cervical length measurement and no interventions based on cervical length were performed in only five studies. In 15 studies patients who had either a cerclage or an indicated preterm birth were excluded from the analysis.

Table 2. Study quality
StudyDesignAdequate description of test procedureCut-off points determined with ROC curvePractitioner blinded to measurement resultsIntervention based on cervical lengthExclusion of patients with intervention or indicated PTB
  • *

    Cut-off point determined with receiver–operating characteristics (ROC) curve at 27 mm, but adjusted to 30 mm for optimal clinical usefulness.

  • No cut-off points used, original data available.

  • Measurements ≥ 5 mm were blinded to the caregiver, patients with measurements < 15 mm received treatment. PTB, preterm birth.

Arabin et al.20Prospective cohortYesUnclearNoYesYes
Fait et al.21Retrospective cohortYesNoUnclearNoYes
Gibson et al.22Prospective cohortYesYesYesNoYes
Goldenberg et al.23Prospective cohortYesNoYesNoNo
Guzman et al.24Prospective cohortNoYesNoYesYes
Guzman et al.35Prospective cohortNoYesNoYesYes
Imseis et al.25Retrospective cohortYesYesNoYesYes
Klein et al.26Retrospective cohortYesUnclearUnclearNoUnclear
Maslovitz et al.36Retrospective cohortYesNoNoUnclearYes
Maymon et al.37Prospective cohortYesYesUnclearUnclearYes
McMahon et al.27Prospective cohortNoYes*YesNoYes
Missfelder-Lobos et al.38Retrospective cohortYesNoNoYesUnclear
Naba28Prospective cohortNoUnclearUnclearUnclearNo
Robyr et al.29Retrospective cohortNoNoNoYesYes
Sayin et al.30Retrospective cohortYesNoNoYesUnclear
Soriano et al.31Prospective cohortYesUnclearYesNoYes
Souka et al.32Retrospective cohortYesNoNoYesYes
Sperling et al.33Prospective cohortYesNoNoYesYes
To et al.39Retrospective cohortYesNoNoYesYes
Vayssiere et al.34Prospective cohortYesYesYesNoNo
Yang et al.9Prospective cohortYesNoNoYesYes

Data analysis

Figure 2 shows an ROC space of the individual studies, as well as the bootstrapped sROC curve plot for all studies in asymptomatic women with a twin or triplet pregnancy. Studies in which cervical length was measured before 20 weeks show an overall high specificity, but sensitivity in these studies does not exceed 70%. The overall sROC curve shows a moderate accuracy, with sensitivity being slightly better than specificity.

Figure 2.

Summary receiver–operating characteristics curve for all studies (twin and triplet). Reported accuracy: preterm birth at: equation image, 34–37 weeks; equation image, 30–34 weeks; equation image, < 30 weeks; gestational age at cervical length measurement: equation image, < 20 weeks; equation image, 20–24 weeks; equation image, > 24 weeks; cervical length cut-off: equation image, 20 mm; equation image, 25 mm; equation image, 30 mm; equation image, 35 mm. _____, Estimated accuracy, summary receiver–operating characteristics curve.

Figure 3 shows summary point estimates for sensitivity and specificity with 95% CIs for four different cut-offs of cervical length. The sensitivity and specificity for birth before 34 weeks are 78% and 66% for 35 mm, 41% and 87% for 30 mm, 36% and 94% for 25 mm and 30% and 94% for 20 mm. As is to be expected, cervical length cut-off is positively correlated with sensitivity, but negatively correlated with specificity.

Figure 3.

Summary receiver–operating characteristics space for different cut-off points of cervical length. Reported accuracy for cervical length in individual studies: equation image, 35 mm; equation image, 30 mm; equation image, 25 mm; equation image, 20 mm. Summary receiver–operating characteristics per cut-off: equation image, 35 mm; equation image, 30 mm; equation image, 25 mm; equation image, 20 mm. _____, 95% CIs.

Figure 4 shows the accuracy of the test per cut-off point for preterm birth. Accuracy was slightly better for preterm birth before 30 weeks than after 30 weeks and overall accuracy of cervical length measured after 24 weeks was slightly better than measurements before 24 weeks (results not shown).

Figure 4.

Summary receiver–operating characteristics curves for different cut-off points of preterm birth. Reported accuracy: preterm birth at: equation image, > 34 weeks; equation image, 30–34 weeks; equation image, < 30 weeks; gestational age: equation image, < 20 weeks; equation image, 20–24 weeks; equation image, > 24 weeks; cervical length cut-off: equation image, 20 mm; equation image, 25 mm; equation image, 30 mm; equation image, 35 mm. _____, Estimated accuracy for preterm birth (PTB) at > 34 weeks; equation image, estimated accuracy for PTB at 30–34 weeks; equation image, estimated accuracy for PTB at < 30 weeks.

Subgroup analysis showed that estimates of sensitivity and specificity were similar for twins (sensitivity 55%, specificity 90%) and triplets (sensitivity 51%, specificity 89%). When analysis was restricted to blinded studies only, cervical length measurement was less sensitive (35%), but this was not statistically significant. Exclusion from the analysis of the studies in which treatment for twin-to-twin transfusion syndrome (Robyr et al.29) or selective reduction (Fait et al.21) was performed, did not change estimates of sensitivity or specificity.

Discussion

We reviewed the literature on cervical length measurement for the prediction of preterm birth in asymptomatic women with a multiple pregnancy. Bivariate meta-analysis of the available data showed a strong association of short cervical length with preterm birth in multiple pregnancies. However, owing to a large variation in gestational age at measurement, cut-off points for cervical length and definitions of preterm birth, and due to the lack of effective treatments, no specific recommendations can be made for the application of this test in clinical practice. Sensitivity of the test is generally low and increases with higher cut-off points for cervical length and preterm birth, at the expense of specificity. Specificity on the other hand is relatively high and increases with an earlier gestational age at testing.

This review has some strengths and limitations. We carried out extensive literature searches without language restrictions and systematically assessed the quality of studies. We have included all available studies reporting on the accuracy of cervical length, for a variety of definitions of preterm birth and thresholds for a positive test result and for different gestational ages at which cervical length was measured, as at present it is unclear which is the most appropriate value to use. This heterogeneity in study characteristics could not, however, be incorporated into the bivariate model commonly used for meta-analyses of diagnostic studies of test accuracy. An adjusted approach based on stratified bootstrap analysis was used to avoid the results being biased towards studies reporting accuracy for multiple thresholds for cervical length, preterm birth and/or gestational age at measurement. Until more sophisticated methods become available to deal with this within-study heterogeneity, this approach may be a viable alternative for evaluating the impact of this heterogeneity on diagnostic accuracy.

Many of these methodological problems may be overcome by using individual patient data meta-analysis. From such data sets, population and patient characteristics, testing conditions and outcome variables can be made as uniform as possible before applying meta-analysis. Currently, initiatives in this field are being employed12. Individual patient data meta-analysis may also allow mono- and multichorionic twins to be analyzed separately, as the risk for preterm birth is strongly influenced by chorionicity13.

In most of the studies included in this review the results of cervical length measurement were not blinded to the caregiver, and interventions were performed based on these findings. Irrespective of whether or not patients with an intervention are included in the analysis, not blinding the caregiver causes information bias. However, we did not observe clinically relevant differences between the potentially biased studies and the studies in which information on cervical length measurement was unknown to the clinician managing the patient.

In spite of the large amount of research that has already been conducted in this subject area, effective strategies for the prevention of preterm birth in multiple pregnancies have yet to be established. Although in one trial progesterone appeared to be effective in both singleton and twin pregnancies with second-trimester cervical shortening, the proportion of twin pregnancies in that trial was only small (24 twin vs. 226 singleton pregnancies) and the effect of progesterone in women with a twin pregnancy was not statistically significant2. Three large trials in multiple pregnancies did not show a reduction in preterm birth after the use of progesterone5, 6, 14. In fact, there was a statistically non-significant trend towards more preterm births in the progesterone group in these trials. Cervical length measurements were not reported in any of the studies.

Two treatment strategies that have been explored in the past even seem to have a detrimental effect on unselected multiple pregnancies. A Cochrane review showed that women with an uncomplicated twin pregnancy who were hospitalized for bed rest had a significantly higher risk of delivering before 34 weeks (odds ratio (OR), 1.8; 95% CI, 1.01–3.3)15. In an individual patient data meta-analysis, cerclage in multiple gestations was also found to increase pregnancy loss or death before discharge from the hospital (OR, 5.9; 95% CI, 1.1–30)16. Another meta-analysis showed a significant increase in preterm birth at less than 35 weeks in twin gestations after cerclage (relative risk, 2.2; 95% CI, 1.2–4.0)17.

An intervention that deserves further research is the vaginal pessary. Retrospective studies on this treatment have shown promising results, and several randomized controlled trials are currently being conducted18. One of these trials will include only women with a multiple pregnancy and will incorporate a cervical length measurement before the start of treatment in the second trimester19.

Despite the lack of an effective intervention seen in unselected populations of women with twin pregnancy, the strong predictive value of cervical length measurement for preterm birth possibly allows the identification of a subgroup of women with a multiple pregnancy that can benefit from preventive treatments such as progesterone. Past efforts to find a preventive treatment for preterm birth have mostly focused on women with a multiple pregnancy in general. Although it is true that this entire group is at increased risk for preterm birth as compared to singleton pregnancies, it is very likely that the individual risk varies and that a subgroup of women with a multiple pregnancy is at an even higher risk for delivering preterm. Just as progesterone and cerclage only seem to have an effect in selected subgroups of women with a singleton pregnancy, these and other interventions may only be effective in a selected proportion of women with a multiple pregnancy.

In this meta-analysis we did not address the risk of delivery within 14 days based on cervical length. This could be an interesting subject for future research, as cervical length may help to select patients who are likely to benefit from treatment with a course of corticosteroids.

In summary, the results of this meta-analysis show that in asymptomatic women with a multiple pregnancy, measurement of second-trimester cervical length can be used to identify a group of women who are at increased risk for preterm birth. Sensitivity, however, is low, indicating that a large percentage of women with a multiple pregnancy will deliver prematurely in spite of a long cervix in the second trimester. In view of the fact that cerclage is known to increase complications, a high specificity is also important, as a limited specificity would increase the exposure of women with a low risk of preterm delivery to unnecessary and potentially harmful interventions. For example, a cut-off of 35 mm with a sensitivity of 78% and specificity of 66%, and a prevalence of preterm delivery of 20% would indicate that 60% of women with a cervix shorter than 35 mm would not deliver before 34 weeks. We think that in future trials on preventive strategies for preterm birth in multiple pregnancies, a blinded second-trimester measurement of cervical length should be part of the study protocol. In this way, planned subgroup analyses can ascertain whether women with a short cervix will benefit from treatment more than others. If preventive treatments are then found to be effective in these women, cervical length measurement could establish a role in the clinical management of women with a multiple pregnancy.

Ancillary