- Top of page
Objective To determine the value of cervicovaginal fetal fibronectin as a marker for preterm delivery, a previously published meta-analysis was updated.
Study design Selection criteria confined the analysis to English-language original reports of prospective studies including women at <37 weeks' gestation with intact amniotic membranes. For the outcomes of delivery <37 or <34 weeks' gestation or delivery within 7, 14, or 21 days after fibronectin sampling, we calculated sensitivity and specificity rates for each study, for subgroups of studies, and for all studies combined.
Results A total of 40 studies were included. Statistical heterogeneity was seen in the majority of calculations of combined results and a random-effects model was used in these cases. For the outcomes of delivery <37 and <34 weeks' gestation, overall sensitivity rates were 52% and 53%, and overall specificity rates were 85% and 89%, respectively. For the outcomes of delivery within 7, 14, and 21 days, we calculated sensitivity rates of 71%, 67%, and 59% and specificity rates of 89%, 89%, and 92%, respectively. For the subgroup of women with symptoms of preterm labour, sensitivity rates for delivery <37 and <34 weeks' gestation or delivery within 7, 14, and 21 days of 54%, 63%, 77%, 74%, and 70% and specificity rates of 85%, 86%, 87%, 87%, and 90% were calculated.
Conclusion Cervicovaginal fetal fibronectin is an effective short-term marker of preterm delivery, especially in women with symptoms of preterm labour. Because results appear to be heterogeneous in different studies, caution should be taken when they are applied to a specific population.
- Top of page
Preterm birth, which is responsible for 70% of perinatal mortality and nearly half of long-term neurological morbidity1, is probably the most challenging problem in modern obstetrics and gynaecology. The identification of women at risk for preterm delivery would allow the initiation of important interventions to delay delivery and to improve neonatal outcome, such as maternal transfer to a tertiary-care centre, tocolysis, and corticosteroid therapy.
Although a number of risk factors for preterm delivery have been identified, attempts to determine the risk of preterm delivery with scoring systems based on historical and epidemiological data have been unable to reliably identify women who will ultimately deliver preterm.
In a recent clinical management guideline published by the American College of Obstetricians and Gynecologists2 in which strategies to identify or prevent preterm birth were reviewed, it was concluded that there are no current data to support other approaches, such as the use of salivary oestriol, home uterine activity monitoring, or bacterial vaginosis screening. Conversely, it was also concluded that current evidence supports screening for risk of preterm delivery by maternal obstetric history, cervical ultrasonography, and fetal fibronectin (FFN) testing.
Fetal fibronectin is a major component of the extracellular matrix of the maternal–fetal interface, also known as the choriodecidual junction. Its frequent presence in cervicovaginal secretions before 20 weeks' gestation is thought to be caused by the absence of a complete fusion between the fetal membranes and the decidua. However, in a recent large study3 of quantitative FFN testing early in pregnancy, it was shown that individual levels of FFN in the first half of pregnancy vary considerably, and that values exceeding the 90% percentile at 13–20 weeks' gestation identify women at risk for preterm delivery even at this early period of pregnancy.
The appearance of FFN in cervicovaginal secretions in the late second and early third trimester is thought to represent disruption of the chorion–decidua interface. Choriodecidual infection appears to be the major cause of this disruption, the subsequent presence of cervicovaginal FFN, and spontaneous preterm birth.
A large body of literature confirms the value of FFN testing to determine the risk of preterm delivery. In our previous meta-analysis4 including 27 studies, cervicovaginal FFN appeared to be an effective short-term marker of preterm delivery, especially in women with symptoms of preterm labour. The literature search for this review was done in November 1997, and several diagnostic studies have been published in the meantime. We therefore planned to carry out this update of our previous review to find out if the value of FFN testing to estimate the risk of preterm delivery has been confirmed in the most recent literature.
- Top of page
This update was done using the same methodological framework as in the original meta-analysis4 and only a summary of methods is presented here. In March 2002, we updated our literature search using MEDLINE from 1966 and EMBASE from 1989 to identify all literature included under ‘fibronectin’ and ‘preterm’ or ‘premature infant’, ‘labour’, ‘delivery’, or ‘birth’. The following criteria were used to select studies for inclusion: article—original published English-language report; study design—prospective study; population—pregnant women <37 weeks' gestation with intact amniotic membranes; intervention—detection of fetal fibronectin from cervical or vaginal specimens; and preterm delivery by any available definition as outcome. We scanned all abstracts from the computer print-outs, the retrieved full-text reports, and the references from each retrieved report to determine whether studies met our inclusion criteria.
To determine the clinical homogeneity of the studies included, we extracted patient inclusion and exclusion criteria, criteria for a positive fibronectin test, mean gestational ages of fibronectin sampling, and information regarding inclusion of multiple gestations. If mean gestational ages of fibronectin sampling were missing in individual studies, we estimated them from patient inclusion criteria.
We then extracted total numbers of women with results, numbers of true positive (TP), true negative (TN), false positive (FP), and false negative (FN) results for the outcomes delivery before 37 or 34 weeks' gestation, and total numbers of fibronectin samples, numbers of TP, TN, FP, and FN results for the outcomes delivery within 7, 14, or 21 days of fibronectin sampling. When numbers of TP, TN, FP, and FN results were not available, we tried to recalculate them from the reported results.
We chose sensitivity and specificity rates to assess the diagnostic accuracy of cervicovaginal FFN because they are not influenced by the different prevalence of preterm delivery that we expected to find in the different studies and because they are commonly used and well known to the general reader.
We calculated sensitivity and specificity rates for each study and sensitivity and specificity rates with 95% confidence intervals5 for all studies combined and for subgroups of studies according to the following characteristics: patient inclusion criteria—women admitted with symptoms of preterm labour, asymptomatic women at low or high risk for preterm delivery and inclusion of women with multiple gestations; sampling period—mean gestational age of fibronectin sampling; and sampling frequency—single or serial sampling.
Because we expected sensitivity and specificity rates to be heterogeneous between studies and between subgroups of studies, a statistical test of homogeneity5 was carried out for each analysis to address the statistical validity of aggregating the trials. This test assumes that differences in the results of the individual trials are due to chance alone, i.e. that all trial results are homogeneous. This method provides a statistic following a χ2 distribution with n−1 degrees of freedom, where n is the number of trials. A P value > 0.05 suggests that the assumption of statistical homogeneity is not violated.
To calculate overall results for each outcome and all subgroups, we generally used a fixed-effects model. In such models, the overall effect is obtained as a weighted average of the individual trial effects. However, if significant heterogeneity was observed (which we defined more restrictively as a P value of the homogeneity test of ≤0.10 because of the low power of the test), we used a random-effects model instead. Trials with zero cells were not included in the calculation of combined results.
- Top of page
A total of 40 studies were included in our meta-analysis. The total numbers of women included were 11,282 and 10,805 for the main analyses of delivery <37 and <34 weeks' gestation and the total numbers of samples were 4633, 4453, and 3568, for the main analyses of delivery within 7, 14, and 21 days of sampling.
Due to space limitations, the lists of studies included and excluded, and detailed results of the qualitative and quantitative analyses of the studies included, are not presented in this paper.
Women at average risk or no specific risk for preterm delivery (low-risk group) were included in 15 studies, women assumed to have an increased risk for preterm delivery (high-risk group) in five, and women with symptoms of preterm labour (preterm labour group) in 21 studies (in one study, both low-risk women and women with preterm labour were included and analyzed separately).
In 32 studies a laboratory ELISA assay was used, in six studies a solid-phase immunogold assay, in one study both a laboratory ELISA and a solid-phase immunogold assay, and in another study a prototype dipstick membrane immunoassay was used. Samples were defined as positive if the fibronectin concentration was ≥50 ng/ml in 27 studies, ≥40 ng/ml in one study, and cut-offs were not specifically stated in two studies. In three studies, results were shown for both a threshold of 50 and 60, 40 and 100, or 25 and 50 ng/ml. To be consistent with the results of all other studies, we extracted only the results obtained with a threshold of 50 (or 40) ng/ml.
In 25 studies, cervical and/or vaginal samples were taken only once (single-sampling subgroup), and in 15 studies samples were repeated after a specified period of time (serial sampling subgroup). In six of the 15 studies with serial sampling, however, each sample was separately analysed so that these studies were included in the single sampling subgroup. In two other studies with serial sampling, results were presented both by subject and by swab. To be consistent with the other studies included, we extracted only results for delivery <37 weeks' gestation analysed by subject, and results for delivery within 7, 14, and 21 days analysed by swab.
Tables 1 and 2 summarise the sensitivity and specificity rates with 95% confidence intervals for the main analyses and for all subanalyses. For one subanalysis, no study was available, and in three subgroups only one study was included.
Table 1. Fetal fibronectin as a predictor for delivery before 37 or 34 gestational weeks, combined results.
|Study groups||Delivery <37 weeks||Delivery <34 weeks|
|Sensitivity (%), 95% CI||Specificity (%), 95% CI||Sensitivity (%), 95% CI||Specificity (%), 95% CI|
|All studies||52 (42–63)||85 (82–87)||53 (37–70)||89 (86–92)|
|Women with preterm labour|
|All women with preterm labour||54 (43–65)||85 (81–89)||63 (37–90)||86 (79–93)|
|Single sampling||26 (5–46)||90 (86–94)||41 (23–58)||94 (91–96)|
|Serial sampling||71 (60–82)||76 (71–84)||78 (57–99)||57 (39–74)|
|Low-risk women, single sampling||26 (5–46)||90 (86–94)||40 (21–59)||94 (91–97)|
|High-risk women, single sampling||–||–||23 (14–32)†||94 (93–96)†|
|Low-risk women, serial sampling||67 (51–82)||77 (70–85)||68 (52–83)†||55 (23–87)|
|High-risk women, serial sampling||78 (63–93)||78 (63–93)||92 (62–100)a||59 (47–71)a|
|Singleton gestations, single sampling||15 (12–17)†||89 (83–95)||20 (17–24)†||96 (92–100)|
|Multiple gestations, single sampling||49 (0–100)||85 (55–100)||46 (0–100)||86 (73–99)|
|Singleton gestations, serial sampling||74 (63–85)†||85 (81–89)†||92 (62–100)a||59 (47–71)a|
|Multiple gestations, serial sampling||72 (36–100)||63 (40–87)||68 (52–83)†||55 (23–87)|
|Sampling at ≤23 weeks, single sampling||15 (12–17)†||84 (66–100)||19 (15–23)†||94 (91–96)|
|Sampling at 24–27 weeks, single sampling||39 (0–98)||86 (70–100)||41 (17–65)||92 (89–96)|
|Sampling at ≥28 weeks, single sampling||12 (5–18)†||99 (97–100)†||32 (19–46)||95 (93–98)|
Table 2. Fetal fibronectin as a predictor for delivery within 7, 14, and 21 days after sampling, combined results.
|Study groups||Delivery <7 days||Delivery <14 days||Delivery <21 days|
|Sensitivity (%), 95% CI||Specificity (%), 95% CI||Sensitivity (%), 95% CI||Specificity (%), 95% CI||Sensitivity (%), 95% CI||Specificity (%), 95% CI|
|All studies||71 (57–84)||89 (84–93)||67 (51–82)||89 (85–94)||59 (39–79)||92 (88–95)|
|Women with preterm labour||77 (67–88)||87 (84–91)||74 (67–82)†||87 (83–92)||70 (55–85)||90 (86–94)|
|Asymptomatic (low-risk or high-risk) women||63 (26–90)a||97 (97–98)a||51 (33–70)†||96 (92–100)||44 (27–61)†||96 (91–100)|
The overall result shows that the sensitivity rate for the outcome delivery <37 weeks' gestation is 52% and only slightly higher for the outcome delivery <34 weeks' gestation (53%). In contrast, the specificity rates for both outcomes appear to be higher (85% and 89%, respectively). For the outcomes of delivery within 7, 14, and 21 days, we calculated sensitivity rates of 71%, 67%, and 59%, and specificity rates of 89%, 89%, and 92%.
In only 12 of a total of 35 analyses, the test of homogeneity for either the sensitivity or specificity rate resulted in a P value > 0.10, and in only three analyses was this true for both rates (see Tables 1 and 2).
- Top of page
The results of this update confirm the conclusion of our original meta-analysis i.e. that FFN appears to be one of the most effective markers for preterm delivery, especially in women with symptoms of preterm labour.
Fetal fibronectin is a moderately sensitive, but sufficiently specific marker for preterm delivery occurring before 37 or 34 weeks' gestation. Sensitivity and specificity rates are higher, if delivery within a specified period of time after FFN sampling is used as outcome. The sensitivity rate of FFN is highest up to seven days in advance of delivery and seems to decline with increasing intervals, whereas specificity rates show a slight increase.
Overall, sensitivity rates appeared to be higher in women with symptoms of preterm labour than in asymptomatic women, and specificity rates slightly lower.
Not unexpectedly, the subgroup analyses of serial as compared to single sampling, which was done for the outcomes of delivery before 37 and 34 weeks' gestation, resulted in higher sensitivity but lower specificity rates. Using the outcome of delivery <34 weeks' gestation, the sensitivity of FFN testing seemed to be lower in high-risk women than in low-risk women if FFN is sampled once, but appeared to be higher with serial sampling. Results for women with multiple gestations did not allow a conclusive interpretation, because confidence intervals—as a result of both heterogeneity and small study sizes—tended to be very wide. Also, results within groups of studies ordered by mean gestational ages of FFN sampling did not follow an obvious trend.
We observed that sensitivity and specificity rates tended to vary considerably between different studies. To measure the degree of heterogeneity between the studies included, homogeneity tests were done for all analyses and, if necessary, a random-effects model was used. We also tried to reduce heterogeneity by summarising groups of studies sharing similar study characteristics. However, in only 12 of a total of 35 analyses, the test of homogeneity for either the sensitivity or specificity rate resulted in a P value >0.10, and in only three analyses was this true for both rates. Thus, the goal to reduce heterogeneity, by performing subgroup analyses and to draw more reliable conclusions from the results of these groups, was only partially met.
One of the most striking differences between this update and the original meta-analysis is that sensitivity rates in the subgroup of women with preterm labour tended to be markedly lower in this update and results tended to be much more heterogeneous. Both trends may be explained by the inclusion of recent studies using FFN testing in women with a more heterogeneous definition of preterm labour. Also, in both this subgroup and the meta-analysis in general, decreasing performance rates may also be a sign that the body of literature about FFN testing has reached a more mature stage.
The absence of heterogeneity in some subgroups allowed a more reliable interpretation of the performance of FFN testing in specific circumstances. Single FFN testing in asymptomatic women with singleton gestations, for example, had a very low sensitivity of 15% or 20% if preterm delivery <37 or <34 weeks' gestation were used as outcome. Sensitivity rates with serial sampling in the same group of women, however, were 74% or 92%, respectively, albeit with specificity rates of 85% or 59%. Fetal fibronectin testing in asymptomatic women earlier in the second trimester also consistently showed low sensitivity rates of 15% or 19%, respectively. Although statistical heterogeneity was absent in some subgroups of studies using the outcome of delivery <34 weeks' gestation, interpretation was complicated by the fact that the studies included in the low-risk, serial sampling and the multiple gestations, serial sampling subgroups were identical and there was only one (identical) study included in the high-risk, serial sampling and the singleton gestations, serial sampling subgroups. Other subgroups with absent statistical heterogeneity can be seen in Table 2 when delivery <7, 14, or 21 days was used as outcomes. Sensitivity rates of FFN testing to predict delivery within 14 or 21 days were 51% or 44% in asymptomatic and 74% or 70% in symptomatic women, with specificity rates following an inverse trend.
This confirms our initial conclusion that FFN is an effective short-term marker to predict preterm delivery, especially in women with preterm labour.
Fetal fibronectin is, of course, only a single marker, and combinations of FFN with other potent risk factors, such as a history of a previous preterm delivery or cervical ultrasonography, may further increase the predictive capacity of the model containing FFN alone.
Although identifying women who will deliver prematurely could well lead to meaningful therapeutic interventions, mere identification does not necessarily result in improved outcomes. If the hypothesis that the presence of cervical FFN is an early marker for membrane bacterial colonisation proves correct, potential treatment might include antibiotics, which have failed to show any effects in heterogeneous collectives of women with preterm labour or bacterial vaginosis. To test this hypothesis, the promising results of a subgroup analysis6 of the recent study of the National Institute of Child Health and Human Development, showing a non-significant reduction in preterm birth in women with both bacterial vaginosis and/or Trichomonas vaginalis and a positive FFN test, need to be confirmed.