How to identify twins at low risk of spontaneous preterm delivery




The aim of this study was to evaluate transvaginal sonographic assessment of cervical length at 23 weeks as a screening test for spontaneous preterm delivery in order to define a cut-off value that could be used to select twin pregnancies at low risk of spontaneous preterm delivery.


In a prospective multicenter study of 383 twin pregnancies included before 14 + 6 weeks a cervical scan with measurement of the cervical length was performed at 23 weeks' gestation. The results were blinded for the clinicians if the cervical length was ≥ 15 mm. The rates of spontaneous delivery at different cut-off levels of cervical length were determined.


Eighty-nine percent of the twins had dichorionic placentation and 58% were conceived after assisted reproduction. The rate of spontaneous preterm delivery was 2.3% (1.5% for dichorionic (DC) and 9.1% for (MC) monochorionic twins) before 28 weeks and 18.5% (17.1% for DC and 29.5% for MC twins) before 35 weeks. The screen-positive rate was 5% for a cervical length ≤ 20, 7–8% at ≤ 25, 16–17% at ≤ 30 and 34–48% at ≤ 35 mm depending on chorionicity. The false-negative rate (1 − negative predictive value) ranged from 1.2% at 28 weeks to 18.6% at 35 weeks for all twins. Receiver–operating characteristics curves showed that the sensitivity increased with declining gestational age with cut-off levels of highest accuracy at 21 mm for 28 weeks and 29 mm for 33 weeks.


Cervical length measurement at 23 weeks of gestation is a good screening test for predicting twins at low risk of preterm and very preterm delivery, especially in DC twins. The present results suggest that a cut-off of 25 mm should be recommended. Copyright © 2005 ISUOG. Published by John Wiley & Sons, Ltd.


Preterm delivery (PTD) is a significant obstetric problem contributing greatly to perinatal mortality worldwide. One of the main reasons for PTD is the growing rate of multiple gestations. In Denmark the incidence of twin deliveries is now 2.3%. This may be explained by the widespread availability of assisted reproduction (AR) technologies.

Survival of preterm infants is mainly dependent on the gestational age at delivery. Survival thus increases from 33% at 24 weeks' gestation to more than 95% by 32 weeks1, 2.

Prediction of PTD has been tried by various regimens. In a logistic regression model, cervical length has been shown to be the only independent variable among different potential confounders for the prediction of spontaneous preterm delivery (SPTD)3. In twin as well as in singleton pregnancies the risk of SPTD is inversely correlated with cervical length4, 5. The shorter the cervix, the higher is the risk of SPTD. Several studies3, 6–11 have evaluated the ability of different ultrasonographic parameters of the cervix to predict SPTD. Different cut-off values for the cervical length and for the definition of PTD have been used.

The aim of this multicenter study was to evaluate screening for SPTD with transvaginal sonographic assessment of the cervical length in a large cohort of women with twin pregnancies in order to define a cut-off value that could be used to select those twin pregnancies at low risk of SPTD.


This was a prospective multicenter observational study that included five different university centers of fetal medicine (four in Denmark and one in Sweden). From November 1999 to May 2003, all women were included before 14 + 6 weeks of gestation after giving oral and written informed consent. At the time of inclusion a transabdominal or vaginal scan was done to determine chorionicity12.

The cervical scan was performed at 23 weeks' gestation with an empty bladder using a transvaginal probe in a dorsal lithotomy position over a 3-min scan period. The cervix was visualized in the sagittal plane. The canal length, funnel length and width were measured (Figure 1)13, 14. The patients may have had more than one vaginal scan, but only the result of the planned 23-week cervical scan was used in the study.

Figure 1.

Diagrammatic representation of cervical measurements. A, funnel length; B, canal length; C, funnel width.

The results were blinded to the clinicians if the length of the cervical canal was ≥ 15 mm. This cut-off was chosen as women with a cervical length < 15 mm may have an increased risk of SPTD, and the study aim was to detect those women at low risk of SPTD. Otherwise the obstetricians were informed, and the women were treated with bed rest, antibiotics and steroids at the discretion of the obstetricians.

The gestational age of the fetuses was estimated according to the twin with the greatest biparietal diameter at the 18-week scan15. Last menstrual period dates were not used nor were in-vitro fertilization (IVF) dates. Obstetric history, including information about the method of conception, was obtained in connection with the informed content. Data on the obstetric outcome were obtained from each hospital. The study was approved by the Danish ethical committee (KF 11-134/00).

Statistical analysis

The normality of the distribution of cervical length was assessed using the Chi-square test. Chi-square and Fisher's exact tests were used to calculate the differences between groups. Receiver–operating characteristics (ROC) curve analysis was used to discriminate cases with delivery before different weeks from cases delivered after that time at different cut-off levels for the cervical length at 23 weeks of gestation. If the 95% CI included 0.5 there was no significant difference between the two groups. The positive likelihood ratio (i.e. the ratio between the probability of a positive test result given the presence of the disease and the probability of a positive test result given the absence of the disease) was used to compare the true-positive cases with the false-positive cases.

The odds ratio (OR) (i.e. the proportion between the ratio of presence of characteristics and the absence of characteristics in one group compared to another group) was used to determine how many times the ratio of response and non-response in one group was greater or smaller than in another group.


A total of 424 women were enrolled in the study and underwent a cervical scan. Some 228 (54%) were spontaneous pregnancies, while 196 (46%) were conceived after AR. Most pregnancies (85%) were DC; two cases were monoamniotic. Twenty (5%) women had a history of spontaneous miscarriage between 12 and 25 weeks or PTD before 33 weeks. Table 1 summarizes the characteristics of the study population. Women who underwent induction of labor (n = 17) were excluded from the study, including the two women with monoamniotic twin pregnancies, as were women with a prior conization (n = 21) and women who had a cerclage (n = 3).

Table 1. General characteristics of the study population
  1. GA, gestational age.

Pregnancies included424 
Induced labor17 
Cervical cerclage3 
Included in the study383 
Dichorionic placentation 89
Conceived after assisted reproduction 58
Mean maternal age ± SD (years)31.8 ± 7.0 
Previous spontaneous miscarriage 
 Before 12 weeks9725
 Before 25 weeks154
Previous preterm birth 
 At 25–32 weeks51
 At 32–37 weeks62
Mean GA at cervical scan ± SD (weeks)23.2 ± 1.0 
Mean cervical length ± SD (mm)38.3 ± 10.0 
Mean GA at delivery ± SD (weeks)36.5 ± 3.0 

The study group therefore comprised 383 women, 58% of whom were pregnant after AR, with 89% DC placentation and 11% MC diamniotic (MC;DA) placentation. The cervical length was measured at 23 weeks (mean, 23.2 ± 1.0 weeks). The distribution of CL was not normal (P = 0.0001). The median, 5th and 2.5th centiles were 39, 21 and 14 mm, respectively, and the mean was 38.3 ± 10.0 mm (Figure 2).

Figure 2.

Distribution of cervical length in the study group at 23 weeks of gestation. The vertical lines represent normal distribution.

In 12 cases the cervical length was < 15 mm. All were DC pregnancies. Nine (75%) of these cases were conceived by AR, six of them had celestone, two had antibiotics, and three were hospitalized. The remaining nine women were recommended to relax at home. Two women delivered before 24 weeks, five before 32 weeks and four after 36 weeks. None of these women had a cerclage. One of the twin pairs delivered before 24 weeks died in the neonatal period. The rest of the twins were doing well with only minor or no sequelae.

The SPTD rates stratified according to chorionicity and method of conception are shown in Table 2. The incidence of SPTD in the different groups ranged from 2.3% to 18.5% in all the twins and between 9% and 30% among MC twins compared to 1.5% and 17% among DC twins according to the cut-off value of gestational age (Tables 3 and 4). The rate of SPTD before 35 weeks was significantly increased for the MC twins compared to the DC twins, independently of the method of conception (Table 2).

Table 2. Rates of spontaneous preterm delivery at different gestational ages and cervical lengths at the 23-week ultrasound scan in twins according to the method of conception and chorionicity
  1. CL, cervical length; DC, dichorionic; DC-AR, dichorionic twins conceived by assisted reproduction (AR); DC-NAT, dichorionic twins conceived naturally (NAT); MC, monochorionic-diamniotic; MC-AR, monochorionic twins conceived by assisted reproduction; MC-NAT, monochorionic twins conceived naturally; NS, not significant; SPTD, spontaneous preterm delivery.

n8 213  36 126  126/21336/8221/16244/339
SPTD (weeks)
 < 2811342NS3811NS1%/2%NS8%/13%NS2%/2%NS9%/1.5%P < 0.05
 < 32113157NS3843NS3%/7%NS8%/13%NS7%/4%NS9%/5.6%NS
 < 33113199NS41154NS4%/9%NS11%/13%NS9%/6%NS11%/6.8%NS
 < 341133215NS719108NS8%/15%NS19%/13%NS15%/10%NS18%/12.4%NS
 < 352254722NS1131119P < 0.059%/22%P < 0.0531%/25%NS22%/14%NS30%/17.1%P < 0.05
CL (mm)
 < 21113147NS1332NS2%/7%NS3%/13%NS7%/2%NS5%/5.0%NS
 < 26113209NS2676NS6%/9%NS6%/13%NS10%/6%NS7%/8.0%NS
 < 313383918NS4111915NS15%/18%NS11%/38%NS19%/14%NS16%/17.1%NS
 < 365637535NS16444032NS32%/35%NS44%/63%NS36%/35%NS48%/33.9%NS
Table 3. Numbers and rates of screen-negative women with twin pregnancies and rates of spontaneous preterm delivery for different cut-off levels of cervical length at the 23-week ultrasound scan in twins according to chorionicity
CL (mm)Screen-negative rateSensitivity (%)/1 − NPV (%)
n%< 28 weeks< 32 weeks< 33 weeks< 34 weeks< 35 weeks
  1. Values in bold represent rates of spontaneous delivery. 1 − NPV, 1 − the negative predictive value (false-negative); CL, cervical length; DC, dichorionic; MC, monochorionic-diamniotic.

In DC twins  1.5%5.6%6.8%12.4%17.1%
 > 203229540/0.926.3/4.321.7/5.619.0/10.515.5/15.2
 > 253129240/1.026.3/4.526.1/5.423.8/10.220.7/14.7
 > 302818340/1.131.6/4.634.8/5.333.3/9.927.6/14.8
 > 352246660/0.957.9/4.656.5/4.454.8/8.446.6/13.8
In MC twins  9.1%9.1%11.4%18.2%29.5%
 > 20 mm429550.0/7.050.0/4.940.0/7.325.0/14.615.4/26.8
 > 25 mm419350.0/5.050.0/5.040.0/7.525.0/15.015.4/27.5
 > 30 mm378450.0/5.450.0/5.440.0/8.125.0/16.223.1/27.0
 > 35 mm235275.0/4.375.0/4.360.0/8.750.0/17.453.8/26.1
In all twins  2.3%6.0%7.3%13.1%18.5%
 ≤ 203649533.3/1.626.1/4.721.4/5.618.0/11.214.08/16.7
 ≤ 253539244.4/1.430.4/4.528.6/5.624.0/10.719.72/16.1
 ≤ 303188344.4/1.634.8/4.735.7/5.632.0/10.626.76/16.2
 ≤ 352476466.7/1.260.9/3.657.1/4.854.0/9.347.89/14.9
Table 4. Rates of spontaneous preterm deliveries and screen-positive cases at a given cut-off level of the cervical length at the 23-week ultrasound scan with estimation of sensitivity, false-positive rates, false-negative rates and odds ratios for different gestational ages and different cut-off levels of the cervical length in 10 different studies
StudynNulliparous (%)Prior SPTD (%)SPTD (%)Screen-positive rate (%)ORCIPSens (%)FP (%)1 − NPV (%)
< 28 weeks< 32 weeks< 33 weeks< 34 weeks< 35 weeksCL ≤ 20 mmCL ≤ 25 mmCL ≤ 30 mmCL ≤ 35 mm
  1. 1 − NPV, 1 − the negative predictive value (false-negative); CL, cervical length; FP, false-positive (i.e. 1 − specificity); OR, odds ratio; screen-positive, numbers with delivery before the chosen gestational age for a given cervical length; Sens, sensitivity (detection rate); SPTD, spontaneous preterm delivery.

Goldenberg et al.3147 10.2 8.8    17.7  7.062.1–23.30.02053.814.25.0
        32.0 17.7  3.111.3–7.40.01029.812.027.3
Imseis et al.68538.016.0   20.0–33.30.00694.151.52.9
Yang et al.86557.03.0 9.2    9.2  18.662.6–134.80.00850.05.15.1
     9.2     20.0 11.111.8–69.90.01266.73.83.8
        23.1  20.0 10.282.6–40.50.00153.38.011.5
     9.2      64.6  0.082100.061.00.0
        23.1   64.612.91.5–105.90.00593.352.04.0
Souka et al.521529.332.03.8           100.0  
     8.0          47.0  
        17.5       35.0  
Skentou et al.743443.51.9  7.8   10.4  6.062.8–13.3< 0.000135.38.35.7
Vayssiere et al.924239.01.2 5.4     12.8 6.992.2–22.50.00246.210.93.3
 225      14.7  12.9 3.231.3–7.90.02027.310.412.2
Guzman et al.1013156.0  9.2   17.6   4.011.1–14.00.02541.715.16.5
 129     17.2 18.0   3.471.2–9.70.02836.414.213.3
Gibson et al.1680      18.8 7.5  11.401.8–70.00.01028.63.013.3
Soriano et al.1744100.0    20.5    26.762.006.6–634.0< 0.000188.511.13.1
Sperling et al. (present38357.0 2.3     7.6
 study)    6.0   4.7   10.33.4–30.7< 0.000121.43.34.6
       13.0 4.7   7.93.0–21.1< 0.000118.02.711.2
      7.3   7.6  7.82.7–22.80.00128.65.96.0
     6.0    7.6  6.72.5–18.10.00130.46.14.5
        18.5 7.6  4.92.2–10.7< 0.000119.714.116.1
     6.0     16.4 3.01.2–7.30.03634.815.24.7
        18.5  16.4 2.21.2–4.10.01726.814.116.2
      7.3    16.4 3.21.4–7.30.01335.714.95.6

There were no differences in the rate of cases with cervical length below 21, 26, 31 or 36 mm, respectively, between twins conceived by AR compared to twins conceived naturally, or when stratified according to placentation (Table 2).

The screen-negative rate for STPD was estimated for different cut-off levels of cervical length and ranged from 95.3% at a cervical length of 20 mm to 64.8% at a cervical length of 35 mm for all the twins. The figures for MC and DC twins were almost the same (Table 2). The 1 − negative predictive value (NPV) (1 − NPV ≈ false-negative value) ranged from 1.2% to 16.7% among all the twins depending on the gestational age cut-off. The 1 − NPV figures for MC twins were higher than the 1 − NPV figures for DC twins and independent of cut-off values for both SPTD and cervical length at 23 weeks (Table 3).

If we use the cut-off values of 25 mm and 32 weeks there would have been 16 (4.5%) false-negative cases. Five (1.3%) of these delivered before 28 weeks. None of the 16 cases had a history of late miscarriage or PTD. Two of the 16 cases were MC;DA without any ultrasonic sign of twin–twin transfusion syndrome. Both delivered before 28 weeks. There were two cases of fetuses with heart malformations, which delivered at 28 + 4 and 30 + 1 weeks, and one case with serious maternal intrahepatic cholestasis, in which there was spontaneous labor just after 31 weeks.

ROC curve analysis showed that the sensitivity of the cervical length measurement for PTD increases with declining gestational age, with a cut-off level of highest accuracy at 21 mm for 28 weeks (sensitivity 44.4% and false-positive rate at 6.7%) and 29 mm for 33 weeks (sensitivity 35.7% and false-positive rate at 14.9%) (Figure 3).

Figure 3.

Receiver–operating characteristics (ROC) curve analysis showing the sensitivity and false-positive rate (i.e. 1 − specificity) for delivery before different weeks at different cut-off levels of the cervical length at 23 weeks of gestation. For a sensitivity of 44.4% and a false-positive rate of 6.7% at 28 weeks the cut-off level is 21 mm (asterisk). For a sensitivity of 35.7% and a false-positive rate of 14.9% at 33 weeks the cut-off level is 29 mm (cross). ⧫, < 28 weeks; ▪, < 32 weeks; ▴, < 33 weeks; ●, < 34 weeks; ▵, < 35 weeks.

The 1 − NPV was used as an expression of the false-negative screening rate, and it was 1.4% for 21 mm and 28 weeks and 5.6% for 29 mm and 33 weeks.

The positive likelihood ratio for SPTD falls exponentially from 47 to 1 when the cervical length increases from 2 to 60 mm (Figure 4). The LR curves converge around 20 mm.

Figure 4.

The positive likelihood ratio showing the ratio between the probability of delivery before and after a given week in women with a cervical length measured at 23 weeks of gestation. ⧫, < 28 weeks; ▪, < 32 weeks; ▴, < 33 weeks; ●, < 34 weeks; ▵, < 35 weeks.

Thus the LR, the ROC analyses and the 1 − NPV values showed that a 25 mm cut-off value for cervical length at 23 weeks of gestation was best at identifying twin pregnancies at low risk of SPTD.


This study demonstrated that in twin pregnancies the risk of SPTD decreases with increasing cervical length measured at 23 weeks of gestation. These results are comparable to the results of Souka et al.5 and Skentou et al.7 from the Nicolaides group in London. No difference was found in the cervical length between twins conceived by AR compared to twins conceived naturally, nor between MC and DC twins. The rate of SPTD was, however, significantly higher among MC twins compared to DC twins. The higher false-negative screening rate among MC twins compared to DC twins could support the theory that the pathogenesis for SPTD is different in MC and DC twins. To the authors' knowledge, no other studies to date have stratified according to both placentation method of conception.

Table 4 outlines the results of different studies on this issue. The study of Vayssiere and coworkers9 was a prospective multicenter study as was the present study, but it was smaller and had a larger proportion of parous women. They found a sensitivity/false-positive rate of 46%/11% compared to 35%/15% in the present study, and the 1 − NPV was 3.3% compared to 4.7% in the present study with cut-off values of 30 mm and 32 weeks. The study of Gibson et al.16 was a prospective and observational study like the present one, however chorionicity and method of conception are not mentioned. They found a 22 mm cut-off value to be the best predictor for SPTD before 35 weeks with an OR of 11.4 and a sensitivity of 28.6% compared to an OR of 4.9 and a sensitivity of 19.7% in the present study. The study of Soriano et al.17 included only IVF pregnancies, 5% of which were MC. They chose a 35 mm cut-off and found both SPTD and a sensitivity (88%) comparable to the results of Imseis et al.6. The screen-positive rate was lower than in the present study, which may be explained by the different proportions of MC twins. The studies of Souka et al. and Skentou et al. were based on the database from Kings College Hospital, London, UK. The sensitivity/false-positive rate was 35%/8% compared to 29%/6% with a 1 − NPV of 6% compared to 11% with 25 mm and 33 weeks as cut-off values.

The only other study providing information on deliveries before 28 weeks' gestation5 had nearly twice as many very preterm deliveries as in the present study (3.8% vs. 2.3%). This could be due to the different composition of the cohorts, as theirs originated from central London, while the present study was a multicenter study from smaller cities with different rates of MC twins, or it could be due to selection bias.

The smaller studies by Imseis et al.6 and Yang et al.8 differ in having a much higher screen-positive rate when using 35 mm as the cut-off limit, but the NPV was the same.

Screening with transvaginal ultrasonography performed at around 23 weeks' gestation has been debated. Those against point out the lack of effective intervention when a short cervix is detected. Elective cervical cerclage, urgent cervical cerclage (UCC) and emergency cervical cerclage and bed rest with or without tocolysis have been tried. At least five non-randomized studies18–22 and two randomized studies23, 24 have focused on the issue of performing UCC in women with a short cervix. It was only possible to show a decrease in the rate of SPTD in one of these studies24, and the sample size of both studies was small. A meta-analysis therefore concluded that the effectiveness of elective cervical cerclage has not been proven and the role of cerclage when a short cervix was detected at 23 weeks remains uncertain25. For the subgroup of high-risk women with twin pregnancies these results may support a screening at 23 weeks' gestation. It gives the clinician an opportunity to refer appropriate cases to high-risk units.

Generally, in screening we are looking for tests with a high sensitivity and a high NPV. For women with twin pregnancies it is different as far as they have an a priori odds of complication. It is therefore of interest to define the subgroup among twin pregnancies at lower risk of SPTD in order that the obstetrician might reassure these pregnant women and avoid bed rest and other interventions. In this study the false-negative rate (1 − NPV) is acceptably low and the figures are almost the same whether using 25 or 35 mm as cut-off values for the CL. Since the screen-negative rate decreases with higher cut-off values, too many pregnant women are needlessly worried without providing the obstetrician with an increased sensitivity for SPTD. We therefore suggest that a cut-off of 25 mm at 23 weeks should be recommended in screening for SPTD, especially in DC twins. In conclusion, cervical length measurement at 23 weeks of gestation can be used as a good predictor of twin pregnancies at low risk of preterm and very preterm birth.