1

2

3

4

5 Q1

6

7

8

9

10

11

121314151617181920212223242526

European Journal of Obstetrics & Gynecology and Reproductive Biology xxx (2015) xxx–xxx

Q2

G Model

EURO 8985 1–7

Risk factors for preterm delivery: do they add to fetal fibronectintesting and cervical length measurement in the predictionof preterm delivery in symptomatic women?

Gert-Jan van Baaren a,*, Merel M.C. Bruijn a, Jolande Y. Vis b, Femke F. Wilms c,Martijn A. Oudijk d, Anneke Kwee d, Martina M. Porath e, Guid Oei f,Hubertina C.J. Scheepers f, Marc E.A. Spaanderman f, Kitty W.M. Bloemenkamp g,Monique C. Haak g, Antoinette C. Bolte h, Caroline J. Bax h, Jerome M.J. Cornette i,Johannes J. Duvekot i, Bas W.A. Nij Bijvanck j, Jim van Eijck j, Maureen T.M. Franssen k,Krystyna M. Sollie k, Frank P.H.A. Vandenbussche l, Mallory Woiski l,Patrick M.M. Bossuyt m, Brent C. Opmeer n, Ben W.J. Mol o

a Department of Obstetrics and Gynaecology, Academic Medical Centre, Amsterdam, Netherlandsb Department of Clinical Chemistry and Haematology, University Medical Centre Utrecht, Utrecht, Netherlandsc Department of Obstetrics and Gynaecology, Catharina Hospital, Eindhoven, Netherlandsd Department of Obstetrics and Gynaecology, University Medical Centre Utrecht, Utrecht, Netherlandse Department of Obstetrics and Gynaecology, Maxima Medical Centre, Veldhoven, Netherlandsf Department of Obstetrics and Gynaecology, University Hospital Maastricht, Maastricht, Netherlandsg Department of Obstetrics, Leiden University Medical Centre, Leiden, Netherlandsh Department of Obstetrics and Gynaecology, VU University Medical Centre, Amsterdam, Netherlandsi Department of Obstetrics and Gynaecology, Erasmus Medical Centre, Rotterdam, Netherlandsj Department of Obstetrics and Gynaecology, Isala Clinics, Zwolle, Netherlandsk Department of Obstetrics and Gynaecology, University Medical Centre Groningen, Groningen, Netherlandsl Department of Obstetrics and Gynaecology, Radboud University Nijmegen Medical Centre, Nijmegen, Netherlandsm Department of Clinical Epidemiology, Biostatistics and Bioinformatics, Academic Medical Centre, Amsterdam, Netherlandsn Clinical Research Unit, Academic Medical Centre, Amsterdam, Netherlandso Department of Obstetrics and Gynaecology, The Robinson Institute, School of Paediatrics and Reproductive Health, University of Adelaide, Australia

A R T I C L E I N F O

Article history:

Received 31 August 2014

Received in revised form 22 April 2015

Accepted 19 May 2015

Keywords:

Preterm labour

Prediction

Fetal fibronectin

Cervical length

Pregnancy

A B S T R A C T

Objective: To assess whether patient characteristics add to the fetal fibronectin test and cervical length

measurement in the prediction of preterm delivery in symptomatic women.

Study design: A nationwide prospective cohort study was conducted in all ten perinatal centres in the

Netherlands. Women with symptoms of preterm labour between 24 and 34 weeks gestation with intact

membranes were invited. In all women qualitative fibronectin testing (0.050 mg/mL cut-off) and cervical

length measurement were performed. Only singleton pregnancies were included in this analysis. Logistic

regression was used to construct two multivariable models to predict spontaneously delivery within

7 days: a model including cervical length and fetal fibronectin as predictors, and an extended model

including all potential predictors. The models were internally validated using bootstrapping techniques.

Predictive performances were assessed as the area under the receiver operator characteristic curve

(AUC) and calibration plots. We compared the models’ capability to identify women with a low risk to

deliver within 7 days. A risk less than 5%, corresponding to the risk for women with a cervical length of at

least 25 mm, was considered as low risk.

Results: Seventy-three of 600 included women (12%) had delivered spontaneously within 7 days. The

extended model included maternal age, parity, previous preterm delivery, vaginal bleeding, C-reactive

protein, cervical length, dilatation and fibronectin status. Both models had high discriminative

performances (AUC of 0.92 (95% CI 0.88–0.95) and 0.95 (95% CI 0.92–0.97) respectively). Compared to the

Contents lists available at ScienceDirect

European Journal of Obstetrics & Gynecology andReproductive Biology

jou r nal h o mep ag e: w ww .e lsev ier . co m / loc ate /e jo g rb

* Corresponding author at: Department of Obstetrics and Gynaecology, Academic Medical Centre, 1105 DE, Amsterdam, The Netherlands. Tel.: +31 205664256.

E-mail address: g.j.vanbaaren@amc.uva.nl (G.-J. van Baaren).

Please cite this article in press as: van Baaren G-J, et al. Risk factors for preterm delivery: do they add to fetal fibronectin testing andcervical length measurement in the prediction of preterm delivery in symptomatic women? Eur J Obstet Gynecol (2015), http://dx.doi.org/10.1016/j.ejogrb.2015.05.004

http://dx.doi.org/10.1016/j.ejogrb.2015.05.004

0301-2115/� 2015 Elsevier Ireland Ltd. All rights reserved.

2728 In

29 Q3

30 an31 m32 sy33 ar34 as35 8036 of37 ar38 ef39 [640

41 tr42 de43 im44 ris45 la46 ce47 oc48 pr49 as50

51 re52 an53 w54 pr55 fe56 w57 ce58 de59 34

60 M

61 Se

62

63 co64 Ne65 ob66 fe67 w68 m69

70 la71 m72 th73 7

74 as75 co76 w77 pr

7879808182838485868788899091

92

93949596979899100101102103104105106107108109110111112113114115116117118119120121122123

G.-J. van Baaren et al. / European Journal of Obstetrics & Gynecology and Reproductive Biology xxx (2015) xxx–xxx2

G Model

EURO 8985 1–7

troduction

Spontaneous preterm delivery occurs in 5–13% of all deliveriesd remains the main cause of perinatal mortality and neonatalorbidity in the developed world [1]. Most women withmptoms of preterm labour before 32 weeks gestational agee transferred to a perinatal centre and receive treatment, such

tocolysis, magnesiumsulphate and corticosteroids. Yet, about–95% will not deliver within 1 week after presentation, and half

these women deliver at term [2–5]. Consequently, many womene unnecessarily transferred and exposed to the potential sidefects of tocolysis and corticosteroids in both mother and child–8].To reduce overtreatment and costs of unnecessary referrals and

eatment, accurate identification of those women who will notliver on the short term – i.e. within 7 days – is of the utmostportance. On the other hand, knowledge about the short termk to deliver helps to identify women with symptoms of preterm

bour who need immediate treatment and referral to a perinatalntre in order to optimize circumstances if preterm deliverycurs. Despite many different predictors and risk factors ofeterm delivery known in the literature, it remains difficult tosess this risk in individual women [1,9,10].Several studies, including our nationwide cohort study, have

ported promising results of the combination of cervical lengthd fetal fibronectin measurements to predict preterm delivery

ithin 7 days in symptomatic women [2,11,12]. It is possible thatedictions will improve even further when other prognosticatures are added to these two measures. The aim of this studyas to evaluate whether patient characteristics, when added torvical length and fetal fibronectin, improve the prediction oflivery within 7 days in symptomatic women between 24 and

weeks of gestational age.

ethods

tting

We used data collected in the APOSTEL-I study, a nationwidehort study conducted in all ten perinatal centres in thetherlands between December 2009 and August 2012. Thejective of this study was to evaluate the predictive value of the

tal fibronectin test in addition to cervical length measurement inomen with symptoms of preterm labour. The study design andain results are presented in detail elsewhere [13,12].

In short, we included women with symptoms of pretermbour between 24 and 34 weeks of gestational age with intactembranes. Exclusion criteria were cervical dilatation of more

We obtained fetal fibronectin and cervical length measurementin all participating women. Cervical length measurement insymptomatic women until 34 weeks is standard care in theNetherlands and is performed by clinically active physicians, whoare trained in transvaginal sonography and perform cervical lengthmeasurements themselves. In order to standardize these measure-ments, all participating physicians were given a set of instructionsand a pocket card with illustrated examples of accurate cervicallength measurements. We used the qualitative Rapid fetalfibronectin TLIIQ analyser (Hologic) with a 0.050 mg/mL cut-offfor the fetal fibronectin test. In addition, data collection includedmaternal characteristics, results from digital examination andlaboratory tests, treatment related to preterm labour, and details ofdelivery and neonatal outcomes.

Data analysis

For current analyses we only included singleton pregnancies.The outcome of interest was spontaneous preterm delivery within7 days. We developed two prediction models using logisticregression: the first model with cervical length and fibronectinas predictors and a second model with candidate predictors addedto cervical length and fetal fibronectin. These candidate predictorswere selected based on medical literature, and can be found inTable 1 [1,9,10]. The variables parity and previous preterm deliverywere combined in a new categorical variable because nulliparouswomen cannot have had a previous preterm delivery; we usednulliparous women as the reference and multiparous with andwithout a history of previous preterm delivery as the othercategories.

Approximately 1600 data points (15%) for the predictors weremissing, ranging from 0% missing values for maternal age to 41%for C-reactive protein. Missing data were interpreted as missing atrandom (MAR). Since it is well documented that exclusion ofincomplete cases can yield biased results, we corrected the missingvalues using multiple imputation (10 times) [14–16]. The variablesmaternal age, BMI, number of contractions, C-reactive protein,cervical length and dilatation were analysed as continuousvariables, whereas all other variables were dichotomous. Beforedeveloping the logistic models, linearity between these continuousvariables and the outcome were checked using cubic splineanalyses [17]. Variables were transformed in case of significantnon-linearity.

For the model including all candidate predictors univariablelogistic regression analysis was used to evaluate the individualassociation between the candidate predictors and delivery within7 days. Because of the low incidence of spontaneous pretermdelivery within 7 days, we did a pre-selection of predictors for

model with fibronectin and cervical length, our extended model reclassified 38 women (6%) from low

risk to high risk and 21 women (4%) from high risk to low risk. Preterm delivery within 7 days occurred

once in both the reclassification groups.

Conclusion: In women with symptoms of preterm labour before 34 weeks gestation, a model that

integrates maternal characteristics, clinical signs and laboratory tests, did not predict delivery within

7 days better than a model with only fibronectin and cervical length.

� 2015 Elsevier Ireland Ltd. All rights reserved.

124125126127128129

an three centimetres, previous treatment with tocolysis withindays before inclusion, and contra-indications for tocolysis, such

suspected intra-uterine infections, fetal distress or lethalngenital abnormalities. Women who had iatrogenic deliveryithin 7 days after inclusion were also excluded. All womenovided written informed consent before entering the study.

Please cite this article in press as: van Baaren G-J, et al. Risk factorscervical length measurement in the prediction of preterm deliverdx.doi.org/10.1016/j.ejogrb.2015.05.004

inclusion in the multivariable logistic regression analysis. We usedthe Akaike Information Criterion, meaning that predictors with alevel of significance smaller than 0.157 were selected [18]. Weused backwards elimination to arrive at a parsimonious modelusing the same Akaike Information Criterion [18]. We performedthis selection in each imputation set separately to account for

for preterm delivery: do they add to fetal fibronectin testing andy in symptomatic women? Eur J Obstet Gynecol (2015), http://

130

131

132

133

134

135

136

137

138

139

140

141

142

143

144

145

146

147

148

149

150

151

152

153

154

G.-J. van Baaren et al. / European Journal of Obstetrics & Gynecology and Reproductive Biology xxx (2015) xxx–xxx 3

G Model

EURO 8985 1–7

Table 1Baseline demographic and clinical characteristics of the total study population.

Missings (before

imputation)

n (%)

Delivery within 7 days

N = 73 (12%)a

Delivery beyond 7 days

N = 527 (88%)a

Total

N = 600 (100%)a

Age, year (mean � SD) 0 30.9 � 5.4 29.2 � 5.3 29.4 � 5.4

Caucasian race (n (%)) 0 40 (55) 303 (57) 443 (74)

Body-Mass Index, kg/m2 (median (IQR)) 65 (11) 22.4 (20.7–26.3) 22.3 (20.1–25.3) 22.3 (20.2–25.5)

Nulliparous (n (%)) 0 52 (71) 250 (47) 302 (50)

Previous preterm birth <37 weeks 0 13 (18) 130 (25) 144 (24)

Smoking (n (%)) 48 (8) 9 (12) 79 (15) 88 (15)

Vaginal bleeding (n (%)) 13 (2)

None 35 (48) 478 (91) 513 (86)

Minimal 24 (33) 42 (8) 66 (11)

Active 14 (19) 6 (1) 20 (3)

Contractions (nr per 30 min, median (IQR)) 123 (21) 6.0 (3.0–9.0) 6.0 (4.0–9.0) 6.0 (4.0–9.0)

Laboratory examination at study entry

C-reactive protein (mg/L) (median (IQR)) 253 (42) 19.0 (6.0–35.0) 7.2 (3.0–17.6) 8.0 (3.1–20.9)

Urinary tract infection (n (%)) 117 (20) 14 (19) 89 (17) 103 (17)

Fibronectin status positive (n (%)) 25 (4) 67 (92) 204 (39) 271 (45)

Vaginal examination at study entry

Cervical length, mm (mean � SD) 13 (2) 9.9 � 8.4 28.4 � 11.2 26.1 � 12.5

Dilatation, cm (median (IQR)) 140 (23) 1.0 (1.0–2.0) 0.0 (0.0–1.0) 0.0 (0.0–1.0)

Effacement (n (%)) 169 (28)

0% 2 (3) 184 (35) 186 (31)

25% 2 (3) 106 (20) 108 (18)

50% 18 (25) 171 (32) 188 (31)

75% 11 (15) 44 (8) 55 (9)

100% 40 (55) 23 (4) 63 (11)

Position cervix (n (%)) 202 (34)

Posterior 21 (29) 294 (56) 315 (53)

Median 42 (58) 161 (31) 203 (34)

Anterior 11 (15) 73 (14) 84 (14)

Consistency (n (%)) 193 (32)

Firm 2 (3) 128 (24) 130 (22)

Moderately soft 35 (48) 285 (54) 320 (53)

Very soft 35 (48) 115 (22) 250 (42)

Engagement (fetal station) (n (%)) 231 (39)

Hodge 1 51 (70) 382 (72) 433 (72)

Hodge 2 9 (12) 51 (10) 60 (10)

Hodge 3 13 (18) 95 (18) 108 (18)

Data are presented as number of patients (%) of the total study population for categorical and dichotomous variables and mean � standard deviation (SD) or median (IQR) for

continuous variables.a After multiple imputation.

155156157158159160161162163164165166

167

168169170171172173174175176

differences between imputation sets. Predictors for the finalmultivariable model were selected using the majority method,which means that predictors were selected if in at least 5 out of10 imputed data sets [19].

The estimates for the regression coefficients of both multivari-able models were based on a combination of the regressioncoefficients in each imputation set separately, using Rubin’s rules.To avoid a too optimistic performance of the model withoverestimated regression coefficients, we used bootstrappingtechniques in which the predictor selection was repeated as well.One hundred bootstrap samples of equal size as the original data(n = 600) were drawn from the original dataset with replacement,allowing for multiple sampling of the same individual. Thisresulted in a shrinkage factor that we applied to the regressioncoefficients [17,20].

The two models were compared in terms of overall fit,discrimination and calibration. Model fit of the models wasexpressed with Nagelkerke R2. Improvement in model fit by addingpatient characteristics to the fibronectin test and cervical lengthwas tested with the X2 test. The area under the receiver operatingcharacteristics curve (AUC) was used to express the models’ abilityto discriminate between women who delivered within 7 days andthose who delivered after 7 days. We studied the calibration of themodels with a calibration plot, showing the agreement betweenthe observed and the predicted risks of delivery within 7 days.

Please cite this article in press as: van Baaren G-J, et al. Risk factors fcervical length measurement in the prediction of preterm deliverydx.doi.org/10.1016/j.ejogrb.2015.05.004

To evaluate whether adding patient characteristics to cervicallength and fibronectin improves the capability of the model toidentify low-risk patients we compared the two models in terms ofreclassification [21]. A low risk was defined as a risk of pretermdelivery within 7 days below 5%, which corresponds with a cervicallength of at least 25 mm. The reclassification table summarizes thenumber of women who were correctly reclassified from high riskto low risk and from low risk to high risk in the prediction model,when compared to the model using only cervical length and fetalfibronectin [21]. Data analyses were performed in R version 2.10.0(The R foundation for Statistical Computing, 2009) and SPSSversion 20.0 (Chicago, IL, USA).

Results

We included 714 women in our nationwide cohort. Weexcluded 108 women who had a multifetal gestation, and another6 women who underwent labour induction or had an electivecaesarean section within 7 days after study entrance, leaving600 women in the study group. Seventy-three women (12%)delivered within 7 days after presentation. Baseline characteristicsof the study population are presented in Table 1, which showssummary statistics after imputation. BMI and C-reactive proteinshowed a non-linear relationship with the risk of preterm delivery,

or preterm delivery: do they add to fetal fibronectin testing and in symptomatic women? Eur J Obstet Gynecol (2015), http://

177 an178 lin179

180 fib181 as182 (9183 fib184 de185 Co186 sh187 0.188 Fi189 te190 an191

192 an193 pr194 el195 be196 pr197 ab198 ag199 va

200201202203204205206207208209210211212213214215216217218219220

Figan

inc

th

ob

G.-J. van Baaren et al. / European Journal of Obstetrics & Gynecology and Reproductive Biology xxx (2015) xxx–xxx4

G Model

EURO 8985 1–7

d we applied a logarithmic transformation to better approachearity.In the prognostic model with cervical length and fetalronectin, each as sole predictors, longer cervical length wassociated with a lower risk of preterm delivery (OR 0.85 per mm5% CI 0.81–0.89); regression coefficient �0.16). A positiveronectin test was associated with a higher risk of pretermlivery (OR 5.45 (95% CI 2.05–14.5); regression coefficient 1.70).efficients are presented after shrinkage with an averagerinkage factor of 0.98. The model’s AUC was 0.92 (95% CI89–0.95). The calibration plot of the model is shown ing. 1a. Considering a risk equal or greater than 5% as a positivest result, the model showed a positive predictive value of 33%,d a negative predictive value of 99%.Table 2 shows the results of the univariable and multivariable

alyses performed to create a model based on all potentialedictors. According to the predefined rules for backwardsimination parity and previous preterm delivery would haveen eliminated from the model, but since these are well knownedictors we included them into the model. The final multivari-le model is shown in Table 3. It includes the variables maternale, parity and previous preterm delivery, multifetal gestation,ginal bleeding, cervical length, dilatation, log C-reactive protein

221

222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262

. 1. Calibration plots of the multivariable model, including only cervical length

d fetal fibronectin as predictors (above) and the final multivariable model,

luding all significant predictors (below), demonstrating the agreement between

e risks of delivery within 7 days as predicted by multivariable analysis and the

servation of delivery within 7 days.

Please cite this article in press as: van Baaren G-J, et al. Risk factorscervical length measurement in the prediction of preterm deliverdx.doi.org/10.1016/j.ejogrb.2015.05.004

and positive fetal fibronectin test. The model is presented aftershrinkage with an average shrinkage factor of 0.83.

Adding patient characteristics to a model with cervical lengthand fibronectin significantly increased model fit (chi-square 45,7 degrees of freedom; p < 0.001). The model’s AUC was 0.95 (95%CI 0.92–0.97). The calculated probabilities derived from ourmultivariable model ranged from 0% to 99%. The calibration plotof the model is shown in Fig. 1b. Considering a risk equal or greaterthan 5% as a positive test result, the multivariable model showed apositive predictive value of 30%, and a negative predictive value of99%.

A total of 59 women (10%) were reclassified by using patientcharacteristics in the prediction of preterm delivery in addition tofibronectin and cervical length; 38 women (6%) were reclassifiedfrom low risk (risk < 5% of delivery within 7 days) to high risk and21 women (4%) were reclassified from high risk to low risk.Preterm delivery within 7 days occurred once in both thereclassification groups.

Thus, in total, 17 women (3%) were unnecessarily classified ashigh risk using patient characteristics in the prediction of pretermdelivery (Table 4).

Comment

In this study we assessed whether a prognostic model thatintegrates patient characteristics, cervical length and fetalfibronectin predicts preterm delivery more accurately than thecombination cervical length and fetal fibronectin. The modelconsisted of the variables maternal age, parity and previouspreterm delivery, vaginal bleeding, C-reactive protein, a positivefetal fibronectin result, cervical length and dilatation. Although themodel showed excellent discriminative ability, with an AUC of0.95, the model was unable to identify additional low risk womenas compared to the combination of cervical length measurementand fetal fibronectin testing.

We used a well-described, large, nationwide cohort fordeveloping this prediction model. Relatively few data weremissing, and especially the information of cervical length andfibronectin status, which was known in almost all cases, adds valueto this study. Unfortunately some information was lacking. Firstinformation about previous cervical operations and polyhydram-niosis were not available in the dataset. Second, information aboutthe women who did not give informed consent. However, as thestudy is observational and the additional fibronectin test is not thatinvasive, we believe that just a minority of women did not giveinformed consent. Moreover, in some centres not all women wereregistered. This will however be a random sample. We thereforethink that this will not interfere with our study results.

Another limitation of this study is the relatively small numberof events, which limited our ability for building a multivariablemodel. Even though it has been suggested that for each candidatepredictor studied at least ten events are required, Vittinghoff andMcCulloch described that this number could be lower in certaincircumstances [22]. The small number of events reduces power,and known predictors might not act as a predictor in our model.The low number of events also precluded tests for interaction inthe developed prediction models. We also regret that we were notyet able to perform an external validation of the developedprediction models.

A point of discussion is the potential influence of tocolysis onour study results, considering that we wanted to predict the risksof preterm delivery in untreated women. In general, tocolysis isonly given for the first 48 h after admission and these drugs have ashort half-life [23–25]. We therefore consider it unlikely thattocolysis has influenced the results of this study. Also, it has never

for preterm delivery: do they add to fetal fibronectin testing andy in symptomatic women? Eur J Obstet Gynecol (2015), http://

263

264

265

266

267268269270

Table 2Results of the univariable and multivariable analyses of predictors of spontaneous preterm delivery within 7 days in symptomatic women. Pooled estimates based on

imputed data.

Univariable Multivariable*

OR 95% CI OR 95% CI

Age (year) 1.06 1.01–1.11 1.09 1.01–1.17

Caucasian race 0.90 0.55–1.47 N/A N/A

Log Body-Mass Index (kg/m2) 2.04 0.47–8.99 N/A N/A

Parity and previous PTD

Nulliparous Reference Reference

Multiparous noPTD 0.29 0.14–0.61 0.48 0.17–1.37

Multiparous PTD 0.45 0.23–0.88 0.23 0.17–1.72

Smoking 0.83 0.36–1.89 N/A N/A

Vaginal bleeding

None Reference Reference

Active 31.46 11.39–86.95 19.62 4.61–83.51

Minimal 7.91 4.27–14.66 4.27 1.75–10.42

Contractions (nr per 30 min) 0.97 0.90–1.05 N/A N/A

Laboratory examination at study entry

Log C-reactive protein (mg/L) 1.98 1.49–2.62 1.69 1.16–2.45

Urinary tract infection 1.12 0.56–2.24 N/A N/A

Fibronectin status positive 17.16 6.99–42.15 3.42 1.07–10.94

Vaginal examination at study entry

Cervical length (mm) 0.83 0.79–0.86 0.88 0.84–0.92

Dilatation (cm) 3.25 2.43–4.37 1.51 0.99–2.32

Effacement

0% Reference

25% 1.04 0.09–12.55 N/A N/A

50% 8.63 1.73–43.17 N/A N/A

75% 20.78 3.70–116.65 N/A N/A

100% 139.99 29.82–657.21 N/A N/A

Position cervix

Posterior Reference

Median 3.73 1.86–7.50 N/A N/A

Anterior 2.34 0.52–10.39 N/A N/A

Consistency

Firm Reference

Moderately soft 7.89 0.97–63.98 N/A N/A

Very soft 19.67 2.77–139.59 N/A N/A

Engagement (fetal station)

Hodge 1 Reference

Hodge 2 1.77 0.21–15.28 N/A N/A

Hodge 3 1.07 0.42–2.71 N/A N/A

PTD = preterm delivery.* If the variable had p-value < 0.157 in the univariable analysis, it was considered in the final (multivariable) model.

G.-J. van Baaren et al. / European Journal of Obstetrics & Gynecology and Reproductive Biology xxx (2015) xxx–xxx 5

G Model

EURO 8985 1–7

been proven that tocolysis given during the first 48 h of admissionfor threatened preterm labour delays delivery after 7 days [26].

In our model, both multiparity and a history of preterm deliveryappeared to have a protective effect on delivering preterm

271272273274275276277278279

Table 3Final multivariable model for predicting spontaneous preterm delivery within

7 days in symptomatic women.

Betaa OR (95% CI)

Intercept �4.08

Age (year) 0.068 1.07 (0.99–1.15)

Parity and previous PTD

Nulliparous Reference

Multiparous noPTD �0.61 0.54 (0.19–1.54)

Multiparous PTD �0.38 0.68 (0.25–1.85)

Vaginal bleeding

None Reference

Active 2.47 11.88 (2.83–49.80)

Minimal 1.21 3.34 (1.38–8.06)

Log C-reactive protein (mg/L) 0.43 1.54 (1.09–2.19)

Fibronectin status positive 1.02 2.78 (0.91–8.54)

Cervical length (mm) �0.11 0.90 (0.85–0.94)

Dilatation (cm) 0.34 1.41 (0.93–2.14)

PTD = preterm delivery. AUC = 0.95 (95% CI 0.92–0.97).a Coefficients were shrunken with an average shrinkage factor 0.83.

Please cite this article in press as: van Baaren G-J, et al. Risk factors fcervical length measurement in the prediction of preterm deliverydx.doi.org/10.1016/j.ejogrb.2015.05.004

compared to nulliparous women as the reference group. Therelation is inconsistent with what is known from the literature, asseveral studies have showed that women with a previous pretermdelivery have a greater change to deliver preterm [10]. Yet, womenwho have experienced a previous preterm delivery may be moresensitive to any complaints indicating another preterm delivery, asare their clinicians. These women may visit the hospital with lesssevere complaints than women without a history of pretermdelivery, because they are more worried about their pregnancy oras they have been advised by their physician.

Several studies have reported promising results of combiningcervical length and fetal fibronectin measurements to predictspontaneous preterm delivery in symptomatic women, including

Table 4Reclassification table, showing the number of women reclassified from high risk to

low risk and vice versa using the final multivariable model compared to the model

with only cervical length and fetal fibronectin as predictors.

Model CL

and fFN

Final model

Low risk High risk Total

Low risk 352 (3 PTD – 1%) 38 (1 PTD – 3%) 390 (4 PTD – 1%)

High risk 21 (1 PTD – 5%) 189 (68 PTD – 36%) 210 (69 PTD – 33%)

Total 372 (4 PTD – 1%) 227 (69 PTD – 30%) 600 (73 PTD – 12%)

CL = cervical length; fFN = fetal fibronectin; PTD = preterm delivery <7 days.

or preterm delivery: do they add to fetal fibronectin testing and in symptomatic women? Eur J Obstet Gynecol (2015), http://

280 ou281 fe282 un283 ou284 w285 ca286 ha287 sc288 w289 la290

291 re292 (<293 ce294 pr295 de296 ad297 ris298 tr299 al300 in301 re302 pr303 pa304

305 ch306 va307 fe308 [3309 va310 pr311 ne312 re313

314 be315 ch316 de317 an

318 Et

319

320 Am321 an

322 Fu

323

324 Ne325 (Q4 Z326 ce327 m

328 Co

329

330 An331 m332 Gv333 BN334 m335 M336 Jv337 M338 Jv

339

340

341

342343344345346347348349350

351

352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413

G.-J. van Baaren et al. / European Journal of Obstetrics & Gynecology and Reproductive Biology xxx (2015) xxx–xxx6

G Model

EURO 8985 1–7

r own APOSTEL-1 study [2,11,12]. Measuring cervical length andtal fibronectin has been shown cost-effective by reducingnecessary healthcare costs without compromising neonataltcome [27,28]. The study presented here addressed the question

hether prediction of preterm delivery in symptomatic womenn be further improved by adding predictors. Previous studiesve mainly reported risk assessment in early pregnancy as areening method for the general or high risk pregnant population,hereas we concentrated on women with symptoms of pretermbour [1,9,10].

Additional patient characteristics in our final model did notsult in a more precise prediction of women who are at low risk5%) of spontaneous preterm delivery. The combination ofrvical length measurement and fetal fibronectin testing is asecise, easy to interpret and clinically useful to facilitate clinicalcision-making regarding admission, in utero transfer andministration of antenatal corticosteroids and/or tocolysis. Highk patients may benefit from longer hospitalization, and other

eatment possibilities for preterm labour. On the other hand, itlows clinicians to avoid unnecessary transfers and interventions

low-risk groups, and thereby reduce medical costs. It alsoduces the dilemma of repeated steroids administration later inegnancy [29]. Moreover, the results could be used in counsellingrents for which preterm labour is anticipated.Individual risk assessment of preterm delivery remains

allenging and we should focus on expanding the development,lidation and implementation of prediction models. Quantitativetal fibronectin might add to the prediction of preterm delivery0,31]. We need to standardize the collection of well-definedriables, since they proved to have a good predictive value in deediction of preterm delivery. Cost-effectiveness analyses areeded to evaluate whether the use of prediction models leads to aduction of unnecessary healthcare costs.

In conclusion, in women with symptoms of preterm labourfore 34 weeks gestation, a model that integrates maternalaracteristics, clinical signs and laboratory tests, did not predictlivery within 7 days better than a model with only fibronectind cervical length.

hics approval

The study protocol was approved by the ethics committee of thesterdam Medical Centre on the 21st April 2009 (MEC 08/363),

d was published thereafter.

nding

This investigator-initiated study was funded by Thetherlands Organisation for Health Research and Development

onMw, grant number 80-82310-98-09056). All decisions con-rning the study design, execution, analyses and reports wereade solely by the investigators.

ntribution

Conceived and designed the experiments: BM JV MO BO, JvdP.alysed the data: GvB BO JV. Wrote the first draft of the

anuscript: GvB. Contributed to the writing of the manuscript:B, JV, FW, MO, AK, MP, GO, MSc, MSp, KB, MH, AB, CB, JC, HD,B, JvE, MF, KS, FV, MW, WG, JvdP, PB, BO, BM. Agree with

anuscript results and conclusions: GvB, JV, FW, MO, AK, MP, GO,Sc, MSp, KB, MH, AB, CB, JC, HD, BNB, JvE, MF, KS, FV, MW, WG,dP, PB, BO, BM. Enrolled patients: GvB, JV, FW, MO, AK, MP, GO,Sc, MSp, KB, MH, AB, CB, JC, HD, BNB, JvE, MF, KS, FV, MW, WG,dP, PB, BO, BM. Obtained funding for the trial: JV, BO, JvdP, BM.

Please cite this article in press as: van Baaren G-J, et al. Risk factorscervical length measurement in the prediction of preterm deliverdx.doi.org/10.1016/j.ejogrb.2015.05.004

Conflicts of interest

None to declare.

Acknowledgements

First we would like to thank all women that participated in thestudy. We also greatly acknowledge the efforts of all Dutchobstetrical residents, gynaecologists and midwifes in the perinatalcentres that helped us to include women during their shifts. We aregrateful for the help of all the laboratory specialists and analystswho performed the fibronectin tests. Furthermore, this studywould not have been possible without the great effort of theresearch staff and research nurses and – midwifes of the DutchObstetrical Consortium.

References Q5

[1] Slattery MM, Morrison JJ. Preterm delivery. Lancet 2002;360(9344):1489–97.

[2] Gomez R, Romero R, Medina L, et al. Cervicovaginal fibronectin improves theprediction of preterm delivery based on sonographic cervical length in patientswith preterm uterine contractions and intact membranes. Am J Obstet Gynecol2005;192(2):350–9.

[3] McPheeters ML, Miller WC, Hartmann KE, et al. The epidemiology of threat-ened preterm labor: a prospective cohort study. Am J Obstet Gynecol2005;192(4):1325–9.

[4] Sanchez-Ramos L, Delke I, Zamora J, Kaunitz AM. Fetal fibronectin as a short-term predictor of preterm birth in symptomatic patients: a meta-analysis.Obstet Gynecol 2009;114(3):631–40.

[5] Sotiriadis A, Papatheodorou S, Kavvadias A, Makrydimas G. Transvaginalcervical length measurement for prediction of preterm birth in women withthreatened preterm labor: a meta-analysis. Ultrasound Obstet Gynecol2010;35(1):54–64.

[6] French NP, Hagan R, Evans SF, Mullan A, Newnham JP. Repeated antenatalcorticosteroids: effects on cerebral palsy and childhood behavior. Am J ObstetGynecol 2004;190(3):588–95.

[7] Oei SG, Oei SK, Brolmann HA. Myocardial infarction during nifedipine therapyfor preterm labor. N Engl J Med 1999;340(2):154.

[8] van Veen AJ, Pelinck MJ, van Pampus MG, Erwich JJ. Severe hypotension andfetal death due to tocolysis with nifedipine. BJOG 2005;112(4):509–10.

[9] Wen SW, Smith G, Yang Q, Walker M. Epidemiology of preterm birth andneonatal outcome. Semin Fetal Neonatal Med 2004;9(6):429–35.

[10] Goldenberg RL, Culhane JF, Iams JD, Romero R. Epidemiology and causes ofpreterm birth. Lancet 2008;371(9606):75–84.

[11] Defranco EA, Lewis DF, Odibo AO. Improving the screening accuracy forpreterm labor: is the combination of fetal fibronectin and cervical length insymptomatic patients a useful predictor of preterm birth? A systematicreview. Am J Obstet Gynecol 2012.

[12] van Baaren GJ, Vis JY, Wilms FF, et al. Predictive value of cervical lengthmeasurement and fibronectin testing in threatened preterm labor. ObstetGynecol 2014;123(6):1185–92.

[13] Vis JY, Wilms FF, Oudijk MA, et al. Cost-effectiveness of fibronectin testing in atriage in women with threatened preterm labor: alleviation of pregnancyoutcome by suspending tocolysis in early labor (APOSTEL-I trial). BMC Preg-nancy Childbirth 2009;9:38.

[14] Donders AR, van der Heijden GJ, Stijnen T, Moons KG. Review: a gentleintroduction to imputation of missing values. J Clin Epidemiol 2006;59(10):1087–91.

[15] Greenland S, Finkle WD. A critical look at methods for handling missingcovariates in epidemiologic regression analyses. Am J Epidemiol 1995;142(12):1255–64.

[16] van der Heijden GJ, Donders AR, Stijnen T, Moons KG. Imputation of missingvalues is superior to complete case analysis and the missing-indicator methodin multivariable diagnostic research: a clinical example. J Clin Epidemiol2006;59(10):1102–9.

[17] Harrell Jr FE, Lee KL, Mark DB. Multivariable prognostic models: issues indeveloping models, evaluating assumptions and adequacy, and measuring andreducing errors. Stat Med 1996;15(4):361–87.

[18] Royston P, Moons KG, Altman DG, Vergouwe Y. Prognosis and prognosticresearch: developing a prognostic model. BMJ 2009;338:b604.

[19] Vergouwe Y, Royston P, Moons KG, Altman DG. Development and validation ofa prediction model with missing predictor data: a practical approach. J ClinEpidemiol 2010;63(2):205–14.

[20] Steyerberg EW, Bleeker SE, Moll HA, Grobbee DE, Moons KG. Internal andexternal validation of predictive models: a simulation study of bias andprecision in small samples. J Clin Epidemiol 2003;56(5):441–7.

[21] Steyerberg EW, Vickers AJ, Cook NR, et al. Assessing the performance ofprediction models: a framework for traditional and novel measures. Epide-miology 2010;21(1):128–38.

for preterm delivery: do they add to fetal fibronectin testing andy in symptomatic women? Eur J Obstet Gynecol (2015), http://

414

415

416

417

418

419

420

421

422

423

424

425

426

427

428429430431432433434435436437438439440441

G.-J. van Baaren et al. / European Journal of Obstetrics & Gynecology and Reproductive Biology xxx (2015) xxx–xxx 7

G Model

EURO 8985 1–7

[22] Vittinghoff E, McCulloch CE. Relaxing the rule of ten events per variable inlogistic and Cox regression. Am J Epidemiol 2007;165(6):710–8.

[23] Goodwin TM, Millar L, North L, Abrams LS, Weglein RC, Holland ML. Thepharmacokinetics of the oxytocin antagonist atosiban in pregnant womenwith preterm uterine contractions. Am J Obstet Gynecol 1995;173(September(3 Pt 1)):913–7.

[24] Ferguson JE, Schutz T, Pershe R, Stevenson DK, Blaschke T. Nifedipine phar-macokinetics during preterm labor tocolysis. Am J Obstet Gynecol1989;161(December (6 Pt 1)):1485–90.

[25] Tsatsaris V, Cabrol D, Carbonne B. Pharmacokinetics of tocolytic agents. ClinPharmacokinet 2004;43(January (13)):833–44.

[26] Haas DM, Caldwell DM, Kirkpatrick P, McIntosh JJ, Welton NJ. Tocolytictherapy for preterm delivery: systematic review and network meta-analysis.BMJ 2012;345(January):e6226.

Please cite this article in press as: van Baaren G-J, et al. Risk factors fcervical length measurement in the prediction of preterm deliverydx.doi.org/10.1016/j.ejogrb.2015.05.004

[27] Rose CH, McWeeney DT, Brost BC, Davies NP, Watson WJ. Cost-effectivestandardization of preterm labor evaluation. Am J Obstet Gynecol2010;203(3):250–5.

[28] van Baaren GJ, Vis JY, Grobman WA, Bossuyt PM, Opmeer BC, Mol BW. Cost-effectiveness analysis of cervical length measurement and fibronectin testing inwomenwiththreatenedpretermlabor.AmJ ObstetGynecol2013;209(5):436–8.

[29] Vis JY, Wilms FF, Kuin RA, et al. Time to delivery after the first course ofantenatal corticosteroids: a cohort study. Am J Perinatol 2011;28(9):683–8.

[30] Kuhrt K, Unwin C, Hezelgrave N, Seed P, Shennan A. Endocervical and highvaginal quantitative fetal fibronectin in predicting preterm birth. J MaternFetal Neonatal Med 2014.

[31] Abbott DS, Radford SK, Seed PT, Tribe RM, Shennan AH. Evaluation of aquantitative fetal fibronectin test for spontaneous preterm birth in symptom-atic women. Am J Obstet Gynecol 2013;208(2):122–6.

or preterm delivery: do they add to fetal fibronectin testing and in symptomatic women? Eur J Obstet Gynecol (2015), http://