The number of publications dealing with ultrasound examination of the cervix during pregnancy has exploded during the last few years. A very quick search of ‘PubMed’ yielded 280 articles on the subject published since 1998. These articles can be divided into five main categories, namely those dealing with: 1) cervical changes observable at ultrasound examination during pregnancies ending at term, and establishment of reference data; 2) prediction of spontaneous preterm delivery in unselected pregnant populations; 3) prediction of spontaneous preterm delivery in pregnancies at high risk of spontaneous preterm delivery; 4) examination technique; 5) treatment of women identified by ultrasound examination of the cervix to be at high risk of spontaneous preterm delivery.
Most published studies show that cervical length decreases1–9, and that cervical width increases1, 4, 8–10, during singleton pregnancies ending at term. However, in two studies (in which it is not clear if both singleton and multiple pregnancies were included) cervical length and width remained unchanged during pregnancy11, 12. The observation period tended to be shorter and to end earlier in the studies reporting no cervical changes. Longitudinal studies of women with singleton pregnancies have shown that there may be different patterns of change in cervical length and width during normal pregnancy, i.e. a continuous shortening and widening of the cervix between mid-gestation and term, or an accelerated shortening and widening starting at around 30 gestational weeks, or a sudden and dramatic shortening and widening at term8, 9. Few studies touch upon changes in the inner cervical os or dynamic cervical changes in pregnancies ending at term1, 3, 13, and few have studied them systematically8, 9. Both dynamic cervical changes and small openings of the inner cervical os do occur during normal singleton pregnancy, but in most cases these changes are not seen until late in the third trimester8, 9. Cervical length and width, and changes in cervical length and width as pregnancy progresses, seem to be similar in nulliparous and multiparous women with normal singleton pregnancy2, 3, 5, 6, 8, 9. In twin pregnancies, too, cervical length decreases with advancing gestation, but after 20 weeks' gestation the cervix is shorter in twin pregnancies than in singleton pregnancies14. This is true even in twin pregnancies going to term (Bergelin and Valentin; data submitted for publication).
Several studies have tried to identify women with singleton pregnancies at high risk of preterm delivery in the general pregnant population by ultrasound examination of the cervix in the second trimester (18–28 weeks)15–19. All show that the shorter the cervix in the second trimester, the greater the risk of spontaneous preterm delivery. Funneling does not seem to contribute further clinical information17. Scrutiny reveals considerable differences in sensitivities, specificities (false-positive rates), and likelihood ratios between the studies cited. For instance, in the study by Heath et al.16 a cervical length of ≤ 20 mm predicted birth ≤ 32 gestational weeks with a likelihood ratio of 8 (sensitivity, 58%; false-positive rate, 7%), whereas the corresponding likelihood ratio in the study by Hassan et al.19 was 18 (sensitivity, 11%; false positive rate, 0.6%). The differences in results can almost certainly be explained by differences in study design (e.g., at what gestational age the measurements were taken, and how preterm delivery was defined), whether or not the results of the ultrasound examinations were acted upon and how they were acted upon, the study populations (e.g., the proportion of women with risk factors for preterm delivery), and possibly in measurement techniques (e.g., whether transfundal pressure was applied). Large differences in the rates of preterm birth (0.8–4.3% ≤ 34 gestational weeks; 2–3.6% ≤ 32 weeks) and parity (58–99.8% of the women were multiparous) clearly demonstrate that there must have been considerable differences between the study populations. In only one study were the results of the cervical examinations not disclosed to clinicians15. Knowledge of the cervical measurements followed by intervention (e.g., cervical cerclage, bed rest, tocolysis) may have biased the results of the remaining studies. Bias cannot be completely excluded even for the two studies in which results were made available to clinicians only if the cervix was 15 mm or shorter, and where results were calculated excluding women who had cervical cerclage because of a short cervix detected at ultrasound examination16, 17. The differences in results with regard to sensitivity, specificity, and likelihood ratio between the studies make it difficult for clinicians to estimate the risk of preterm delivery at a specific cervical length in their own patients. They need to know which sensitivity, false-positive rate and likelihood ratio are applicable to their own pregnant population.
Many studies have examined the possibility to predict preterm delivery in asymptomatic pregnant women at high risk of spontaneous preterm delivery by examining the cervix by ultrasound examination in the second trimester20–22. Women with multiple pregnancy constitute a specific subgroup at high risk23–29. In only one21 of ten studies evaluating cervical ultrasound examination to predict preterm delivery in high-risk pregnancies were clinicians kept unaware of the results of the ultrasound examinations. In the other nine studies20, 22–29 ultrasound results were disclosed and allowed to be acted upon, which means that the sensitivities and specificities reported in these studies are likely to be poor estimates of the truth. In three studies20–22 comprising high-risk singleton pregnancies, in which serial ultrasound examinations of the cervix were performed in the second trimester, cervical length ≤ 24–25 mm at any of the examinations in the second trimester was associated with a small increase in risk of preterm delivery (≤ 34 or ≤ 33 gestational weeks), likelihood ratios being around 3. Cervical length ≤ 24 mm at 16–18 weeks was associated with a likelihood ratio of 9.5 with regard to birth ≤ 34 weeks21. The presence of funneling or the size of a funnel did not contribute any predictive information in addition to cervical length in singleton pregnancies at high risk of preterm delivery20.
Likelihood ratios associated with cervical lengths ≤ 15–20 mm at around 20 gestational weeks in twin pregnancies vary between studies from 8 to 40 with regard to birth ≤ 31–32 weeks23–27. Results for triplet gestations are even more disparate28, 29. Again, the differences in results are likely to be explained by differences in study design, intervention between the test and the outcome, study population, and measurement technique. Results with regard to the predictive value of the presence or size of a funnel in multiple pregnancies are conflicting. Guzman and colleagues23, 28 found cervical length to be at least as good as other ultrasound variables to predict delivery ≤ 34 weeks in twin and triplet pregnancies. Yang et al.24 stated that both cervical length and funneling were independently and strongly associated with birth ≤ 34 weeks in twin gestations.
The technique used when measuring the cervix is important, but it would probably be wrong to say that one measurement technique is more correct than another. However, the sensitivity, specificity and likelihood ratio reported in one particular study are applicable to one's own pregnant population only if one uses an examination technique identical to that used in that particular study. It is important to determine whether cervical measurements obtained using one measurement technique are more predictive of preterm delivery than those obtained using another measurement technique (e.g., if cervical length measurements obtained after transfundal pressure are more predictive than measurements obtained without transfundal pressure; if the use of the shortest cervical length is more predictive than the use of the mean of repeated measurements; if measurements of a curved cervix are more predictive when taken following the curvature of the cervix or when taken as a straight line between the inner and outer cervical os). An article discussing the technique of examining the cervix in the first trimester is published in this issue of the Journal30. Shalev et al. conclude that the gestational sac does not reach the inner cervical os until 12 gestational weeks, and that therefore the inner cervical os cannot be evaluated for the presence or absence of funneling before that time. It remains to be shown how often an opening of the inner cervical os can be detected so early in pregnancy, and whether an opening of the inner cervical os as early as at 12 weeks has any clinical significance.
Also in this issue of the Journal are three articles examining the ability of cervical length measurements taken early in pregnancy (at 11–14, 13–15 and 10–14 gestational weeks) to predict preterm birth31–33. Conoscenti et al.31 and Carvalho et al.32 studied unselected pregnant populations (the former study included only singleton pregnancies; the latter probably included both singleton and multiple pregnancies), whereas Berghella and coworkers33 examined women with singleton pregnancies at high risk of spontaneous preterm delivery. In none of these three studies were the results of the early cervical length measurements available to clinicians. The results showed that cervical length at 11–15 weeks cannot predict preterm delivery in an unselected pregnant population. However, Carvalho et al.32 found that cervical shortening between 11–14 and 22–24 weeks was associated with an increased risk of preterm delivery. Unfortunately, it is unclear how cervical shortening was defined, and if shortening contributed information independent of cervical length. It would be worth exploring in more detail the clinical significance of shortening of the cervix between the first and the second trimesters. According to the study comprising high-risk pregnancies, 5% of the high-risk women had a short cervix (defined as ≤ 24 mm) at 10–13 gestational weeks, a short cervix being associated with a moderately increased risk of birth ≤ 34 weeks (likelihood ratio, 4.7)33. Even though the results of the early examinations were not disclosed to clinicians, 52% of the women at high risk had a cervical cerclage at 12–14 weeks33. This intervention between the test and the outcome is likely to have affected the sensitivity, specificity and likelihood ratio.
When evaluating the properties of a diagnostic test or a screening test, it is extremely important that clinicians are kept unaware of the test results, and that no action is taken on the basis of the results of the test. Action is likely to change the outcome (for the better or for the worse), and so both the sensitivity and the specificity of the test will be affected34. Action taken on the basis of information other than the results of the test to be evaluated may also bias the results34. It is very unfortunate that in most studies evaluating the ability of ultrasound examination of the cervix to predict spontaneous preterm delivery, results of the cervical ultrasound examinations were available to clinicians. In virtually all studies comprising women at high risk, treatment—including cerclage—was instituted to try to prevent preterm delivery. This intervention between the test and the outcome means that the sensitivities and specificities published probably do not reflect the true ability of cervical measurements to predict preterm birth. However, we do have evidence that an ‘abnormally’ short cervix in the second trimester increases the risk of preterm delivery both in unselected singleton pregnancies15, and in singleton pregnancies at increased risk of preterm delivery21. Evidence is weaker for twin and triplet pregnancies, because in all studies on multiple pregnancies, the results were, at least to some extent, available to clinicians, and this may have introduced bias. Very importantly, the magnitude of any risk increase is uncertain, mainly because of inappropriate study designs, where action was taken on ‘abnormal’ test results, but also because the magnitude of a change in risk is likely to depend on the characteristics of the women tested. In a high-risk population, the magnitude of a change in risk might differ between women with different risk factors. For instance, it may differ between those who are at high risk because they have undergone a cone biopsy and those who are at high risk because they have previously given birth preterm. Thus, the mix of women with different risk factors will almost certainly affect the sensitivity, specificity, and likelihood ratio of a test designed to predict spontaneous preterm delivery. The same is true of ‘unselected’ populations, because they will necessarily contain varying proportions of women at high risk of spontaneous preterm delivery. The magnitude of a risk increase (or risk decrease) is extremely important to the practicing clinician, if he/she wants to use ultrasound examination of the cervix to estimate the risk of spontaneous preterm delivery in his/her own patients. A good measure of risk increase/risk decrease is the likelihood ratio [sensitivity/(1 — specificity)]. Post-test odds can be calculated by multiplying the pretest odds with the likelihood ratio. Odds can easily be converted into probabilities and vice versa34, 35. Nomograms simplify calculation of post-test probabilities34, 35: one only needs to know the pretest probability and the likelihood ratio. The odds ratio is another useful measure of risk increase/risk decrease, with nomograms available for easy calculation of post-test probabilities36. However, odds ratios are more difficult than likelihood ratios to intuitively understand. Unfortunately, few studies evaluating diagnostic tests or screening tests (including those evaluating cervical measurements in pregnancy) present results in terms of likelihood ratios or odds ratios. Many do not even provide the reader with information about sensitivity and specificity, both of which are necessary to calculate the likelihood ratio (moreover, because of the trade-off between sensitivity and specificity, information about sensitivity is meaningless if it is not presented together with its specificity, and vice versa). Most studies contain the information necessary to calculate—or at least estimate—sensitivity, specificity and likelihood ratio, but it is often very cumbersome for the reader to find the information, which may be well hidden in the text or in a figure (in preparing this Opinion I had to calculate most likelihood ratios myself). I would like to make a plea for uniform presentation of results of studies evaluating diagnostic tests and screening tests. The following information should always be presented in absolute values as well as in percentages, preferably in the abstract of the article: prevalence of the condition sought, sensitivity, specificity, and likelihood ratio. This is what the reader needs to know to understand the clinical value of a test. Other important information to be given in the abstract is whether or not test results were available to clinicians and acted upon, and the characteristics of the study population. Information about to what extent there was interaction between the test and the outcome—even when action was not taken on the basis of the test results—is important, too, but it might not be possible to give detailed information about this in a short abstract. The same is true of examination technique/test methodology.
What should we do with pregnant women who have been identified to be at high risk of spontaneous preterm delivery because they have a short cervix at ultrasound examination? Should we recommend them sick leave, put them to bed, give them tocolytics and/or antibiotics, or a cervical cerclage? To the best of my knowledge, there is no scientific evidence that sick leave, bed rest, tocolysis or antibiotics prevent preterm delivery in asymptomatic pregnant women with risk factors for spontaneous preterm delivery, be the risk factor a short cervix or something else. In a non-randomized trial, placing a cervical cerclage in asymptomatic women with a cervix ≤ 15 mm detected at routine ultrasound examination at 23 gestational weeks prevented birth ≤ 31 weeks37. Unfortunately, this was not confirmed in a randomized controlled trial conducted by the same research team (Kypros Nicolaides, personal communication, 2002). A randomized controlled trial by Rust and coworkers38, 39 showed no benefit of cervical cerclage in women with an abnormal cervix at ultrasound examination in the second trimester. Their study population probably consisted of a mix of women with and without known clinical risk factors for preterm delivery. According to other randomized trials, cervical cerclage may be of benefit to women who are clinically at high risk of delivering preterm40–42. Measurement of cervical length might lead to avoidance of unnecessary cerclage in women with a ‘long’ cervix42. I have stated above that treatment instituted between test and outcome is likely to alter pregnancy outcome and so affect the sensitivity and specificity of a test to be evaluated for its ability to predict preterm delivery; this may seem contradictory, because at the same time I now state that we have no evidence that some of the treatments we use to try to prevent preterm birth are effective. However, intervention almost certainly has some effect on outcome—positive or negative. For instance, hospitalization with bed rest in twin pregnancy may increase the risk of preterm birth43! Because we often do not know the magnitude or direction of a possible treatment effect, results with regard to sensitivity and specificity become unreliable. Cerclage may change pregnancy outcome for the better or for the worse depending on the patient. In certain high-risk groups it may be beneficial and prolong pregnancy41; in other high-risk groups or in women without known risk factors it might do harm by provoking infection, preterm rupture of the membranes and preterm delivery. Even when results of studies were disclosed to clinicians only if the cervix was ≤ 15 mm16, 17, 26, 27, 29, and even in studies in which women who had a cerclage because of a short cervix were excluded from analysis16, 17, 27, we cannot exclude bias. Even though interaction between test and outcome need not completely negate the value of a study, it does introduce considerable uncertainty about its validity.
Despite the large number of publications dealing with the possibility of using ultrasound examination of the cervix in pregnant women to predict spontaneous preterm delivery, our current knowledge about how (or if) to use cervical ultrasound examination during pregnancy in clinical practice is insufficient. One reason for this is that the design of many studies evaluating the diagnostic properties of cervical ultrasound examination during pregnancy is inappropriate, mainly because actions were taken on the results of the test to be evaluated. Another reason is that we do not fully understand the physiology of term labor or the pathophysiology of preterm labor. The pathophysiology, cervical changes preceding spontaneous preterm delivery detectable at ultrasound examination or by biochemical tests, and effective treatment need to be determined separately for women with different risk factors. For example, they should be determined separately for women with 1) Müllerian abnormality, 2) previous cone biopsy, 3) previous second-trimester loss, 4) previous preterm delivery, 5) multiple pregnancy, 6) a clinical history compatible with cervical incompetence, or 7) no known risk factors.
Unfortunately, current evidence does not support screening of unselected populations for preterm delivery by using ultrasound examination of the cervix at early or mid-gestation. Two studies published in this issue of the Journal suggest that results of ultrasound examination of the cervix in early pregnancy do not predict preterm delivery in unselected pregnant populations31, 32. Moreover, even though there are data showing that a short cervix at mid-gestation is associated with an increased risk of preterm delivery both in unselected populations and in high-risk pregnancies, we lack effective treatment, except perhaps for women with certain risk factors who might benefit from cervical cerclage40, 41. It is important that research in the field of spontaneous preterm delivery continues.