UBC Faculty Research and Publications

Timing of delivery in a high-risk obstetric population: a clinical prediction model De Silva, Dane A; Lisonkova, Sarka; von Dadelszen, Peter; Synnes, Anne R; Magee, Laura A Jun 29, 2017

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata


52383-12884_2017_Article_1390.pdf [ 528.57kB ]
JSON: 52383-1.0366847.json
JSON-LD: 52383-1.0366847-ld.json
RDF/XML (Pretty): 52383-1.0366847-rdf.xml
RDF/JSON: 52383-1.0366847-rdf.json
Turtle: 52383-1.0366847-turtle.txt
N-Triples: 52383-1.0366847-rdf-ntriples.txt
Original Record: 52383-1.0366847-source.json
Full Text

Full Text

RESEARCH ARTICLE Open AccessTiming of delivery in a high-risk obstetricpopulation: a clinical prediction modelDane A. De Silva1,2 , Sarka Lisonkova1,2, Peter von Dadelszen3,4, Anne R. Synnes5, Canadian Perinatal Network(CPN) Collaborative Group1 and Laura A. Magee4*AbstractBackground: The efficacy of antenatal corticosteroid treatment for women with threatened preterm birth dependson timely administration within 7 days before delivery. We modelled the probability of delivery within 7 days ofadmission to hospital among women presenting with threatened preterm birth, using routinely collected clinicalcharacteristics.Methods: Data from the Canadian Perinatal Network (CPN) were used, 2005–11, including women admitted tohospital with preterm labour, preterm pre-labour rupture of membranes, short cervix without contractions, ordilated cervix or prolapsed membranes without contractions at preterm gestation. Women with fetal anomaly,intrauterine fetal demise, twin-to-twin transfusion syndrome, and quadruplets were excluded. Logistic regressionwas undertaken to create a predictive model that was assessed for its calibration capacity, stratification ability, andclassification accuracy (ROC curve).Results: We included 3012 women admitted at 24–28 weeks gestation, or readmitted at up to 34 weeks gestation,to 16 tertiary-care CPN hospitals. Of these, 1473 (48.9%) delivered within 7 days of admission. Significant predictorsof early delivery included maternal age, parity, gestational age at admission, smoking, preterm labour, prolapsedmembranes, preterm pre-labour rupture of membranes, and antepartum haemorrhage. The area under the ROCcurve was 0.724 (95% CI 0.706–0.742).Conclusion: We propose a useful tool to improve prediction of delivery within 7 days after admission amongwomen with threatened preterm birth. This information is important for optimal corticosteroid treatment.Keywords: Preterm birth, Prediction model, Antenatal corticosteroidsBackgroundPreterm birth is the leading cause of perinatal mortalityand morbidity in Canada and worldwide [1, 2]. One themost effective means to reduce neonatal mortality andmorbidity in preterm infants is antenatal administration ofcorticosteroids [3, 4]. The proven benefits of antenatalcorticosteroids for fetal lung maturation and prevention ofserious neonatal morbidity among women with pretermdelivery have resulted in the inclusion of corticosteroidtreatment as standard obstetric care in industrialisedcountries [3, 5–7]. Treatment guidelines recommend ad-ministering corticosteroids to women who are at risk ofdelivering within the next 7 days when they presentbetween 24+0 and either 33+6 weeks gestation in Canada[5] and the United States [6], or up to 35+6 weeks gesta-tion in the United Kingdom [7].One of the major barriers to appropriate use of ante-natal corticosteroids is that timing of delivery is oftenunknown; approximately half of women who are admittedto hospital with threatened preterm birth remain undeliv-ered after 7 days [8, 9]. This makes it difficult to maximise‘optimal’ use of corticosteroids (i.e., administration towomen who go on to deliver within the next 7 days andnot administering them to women who do not deliverwithin the next 7 days). A recent study from Nova Scotia,Canada, showed that between 1988 and 2012, the propor-tion of women with suboptimal antenatal corticosteroidtreatment (i.e., more than 7 days prior to delivery)* Correspondence: LMagee@sgul.ac.uk4Molecular & Clinical Sciences Research Institute, St. George’s University ofLondon, Rm J0.27, Jenner Wing, Cranmer Terrace, London SW17 0RE, UKFull list of author information is available at the end of the article© The Author(s). 2017 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, andreproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link tothe Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver(http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.De Silva et al. BMC Pregnancy and Childbirth  (2017) 17:202 DOI 10.1186/s12884-017-1390-9increased approximately 5 times (from 7% to 34%) whereasoptimal treatment doubled (from 10% to 23%) [10].Until recently, administration of a single course of ste-roids at <34 weeks was considered both effective andsafe. Increasingly, however, concerns have emergedabout adverse effects following multiple courses and sideeffects following a single course administered in thecommunity in under-resourced settings [11–17]. Multipleantenatal corticosteroid exposure at preterm gestation hasbeen found to affect fetal growth and CNS development[11–14], and to be associated with neurosensory dysfunc-tion in particular among infants born at term [15, 16]. In arecent cluster randomised controlled trial in low- andmiddle-income countries (LMICs), community adminis-tration of antenatal corticosteroids for threatened pretermbirth before 36 weeks was associated with no benefit ofantenatal corticosteroids among babies born preterm, andwith a 12% increase in neonatal mortality and a 45%increase in suspected maternal infection [17].Actual corticosteroid treatment rates vary widely,depending in part on maternal and obstetrical charac-teristics [4] and the difficulty in predicting pretermbirth [18, 19]. Most literature focuses on predictors asindividual tests [20, 21] or clinical biomarkers; [20, 22]however, they have not been found to be as useful inpredicting preterm birth among asymptomatic or nul-liparous women, even when combined [21–24]. Thereis a paucity of literature on prediction models thatcombine multiple determinants of preterm birth, particu-larly within 7 days of admission. Moreover, there is a lackof models that use characteristics that are available uponadmission. Thus, based on clinical characteristics collectedin routine clinical practice, we aimed to identify womenwho were likely to deliver within 7 days after admission tohospital due to a high-risk of preterm delivery.MethodsWe carried out a retrospective cohort study of women ad-mitted to any of the 16 perinatal centers participating inthe Canadian Perinatal Network (CPN) (Additional file 1).The CPN study collected demographic, clinical, andbirth information on women who were both at high-risk of very preterm birth between 22+0 and 28+6 weeksgestation and admitted (or re-admitted) to the partici-pating CPN tertiary hospitals for at least 24 h fromAugust 1, 2005 to March 31, 2011. These women werefollowed-up until delivery.The CPN database collected information on maternaldemographic and behavioural characteristics (e.g., mater-nal age, marital status, smoking history, parity), past med-ical and obstetric history (e.g., number of prior abortionsand preterm births), characteristics of current pregnancy(e.g., fertility treatments, use of drugs or alcohol in preg-nancy, reason for admission to hospital), maternal andfetal surveillance and treatments (e.g., expectant care, useof antibiotics, or partial or completed courses and sub-sequent courses of corticosteroids), and pregnancy out-comes. CPN details including data definitions havebeen published previously [25]. Ethics approval wasobtained centrally as a quality assurance project by theResearch Ethics Boards at the University of BritishColumbia (H05–70359) and at each study site. As such,written consent was not required, and any collected infor-mation was anonymised and de-identified prior to analysis.In the current study, we included women in the CPNdatabase who presented between 24+0 and 28+6 weeks ges-tation with one or more of the following conditions atenrollment: preterm labour, preterm pre-labour rupture ofmembranes (PPROM), short cervix without uterine con-tractions, dilated cervix or prolapsed membranes withoututerine contractions on admission (see Additional file 2for definitions). We also included women who wereadmitted for these conditions occurring de novo at up to34+6 weeks after previous admission to the CPN networkfor other reasons, as these women would also be eligibleto receive corticosteroids. We excluded women whosepregnancies were complicated by intrauterine fetal demiseat the time of admission, twin-to-twin transfusion syn-drome, those with quadruplets, and known fetal anomalydue to differences in their clinical management [26].To create a generalisable model, it is recommendedthat highly subjective factors or those dependent on thehealth care system, local practice, or individual practi-tioners should not be considered as predictors [27, 28].Therefore, we did not include the following variables ascandidate predictors in our study: transfer to anotherfacility, use of cervical cerclage, tocolytic and antibiotictreatment, and the fetal fibronectin test administeredbefore admission.Unadjusted associations between candidate predictorsand delivery within 7 days were tested using chi-squareand Fisher’s exact test (as appropriate for categoricalvariables), or the Student’s t-test or Wilcoxon non-parametric test (as appropriate for continuous variables).For the initial multivariable logistic regression model, weselected candidate predictors that were either associatedwith delivery within 7 days at p < 0.10 [29], or for facevalidity, were known to be clinically relevant to pretermbirth (including alcohol use during pregnancy, and pre-vious preterm birth) (Table 1) [30–32]. Potential interac-tions between candidate predictors were examined. Inthe multivariable model, the association between deliv-ery within 7 days and candidate predictors that werecontinuous (i.e., maternal age and gestational age) wasexamined using clinically meaningful categories, productterms, and splines where necessary. Missing values forall variables in the model were imputed using the MICEmethod of multiple imputation [29, 33]. This methodDe Silva et al. BMC Pregnancy and Childbirth  (2017) 17:202 Page 2 of 10Table 1 Demographic and clinical characteristics of women admitted to a tertiary hospital at 24–34 weeks gestationDelivery within 7 days (N = 1473) Delivery in >7 days (N = 1539) p valueaBaseline Demographic and Medical/Surgical HistoryMaternal age on admission (years)a< 20 74 (5.0%) 45 (2.9%)20–24 219 (14.9%) 213 (13.8%)25–29 403 (27.4%) 404 (26.3%)30–34 466 (31.6%) 509 (33.1%) 0.00235–39 260 (17.7%) 275 (17.9%)40–44 43 (2.9%) 79 (5.1%)≥ 45 8 (0.5%) 14 (0.9%)Nulliparousa 811 (55.1%) 699 (45.4%) <0.001Prior miscarriagea 432 (29.3%) 514 (33.4%) 0.018Prior therapeutic abortion 227 (15.4%) 231 (15.0%) 0.798Prior birth at <37 weeks 213 (14.5%) 321 (20.9%) 0.013Prior birth at 34–36 weeksa 82 (5.6%) 131 (8.5%) 0.090Prior birth at <34 weeksa 161 (10.9%) 190 (12.3%) 0.476Pre-existing medical conditionsPre-existing hypertensiona 26 (1.8%) 48 (3.1%) 0.023Pre-existing diabetes mellitus 20 (1.4%) 24 (1.6%) 0.757Uterine structural abnormalitiesa,c 73 (5.0%) 94 (6.1%) 0.193Renal disease/urologya 7 (0.5%) 20 (1.3%) 0.027Rheumatic disease 14 (1.0%) 8 (0.5%) 0.241Cardiac disease 10 (0.7%) 14 (0.9%) 0.612Cervical procedures 17 (1.2%) 21 (1.4%) 0.723Otherb 202 (13.7%) 261 (17.0%) 0.016Smoking after pregnancy diagnoseda 321 (21.8%) 265 (17.2%) 0.002Missing 18 (1.2%) 5 (0.3%)Alcohol use during pregnancy (socially or at least twice weekly)a 43 (2.9%) 42 (2.7%) 0.838Missing 19 (1.3%) 11 (0.7%)Illicit drug use after pregnancy diagnoseda 69 (4.7%) 52 (3.4%) 0.083Missing 19 (1.3%) 7 (0.5%)Current pregnancyGA on admissiona (weeks)24 + 0–25 + 6 520 (35.3%) 602 (39.1%)26 + 0–27 + 6 593 (40.3%) 616 (40.0%)28 + 0–29 + 6 329 (22.3%) 311 (20.2%) <0.00130 + 0–31 + 6 10 (0.7%) 7 (0.5%)32 + 0–34 + 6 21 (1.4%) 3 (0.2%)Multiple pregnancya 256 (17.4%) 380 (24.7%) <0.001Twins 237 (16.1%) 343 (22.3%)Triplets 19 (1.3%) 37 (2.4%)Reasons for admissionPreterm laboura 776 (52.7%) 489 (31.8%) <0.001Preterm pre-labour rupture of membranesa 597 (40.5%) 483 (31.4%) <0.001Short cervix without contractionsa 125 (8.5%) 548 (35.6%) <0.001De Silva et al. BMC Pregnancy and Childbirth  (2017) 17:202 Page 3 of 10assumes data are missing at random; to test this assump-tion, the cases with complete data were compared tocases with missing variables. Imputation models werebuilt including all possible predictors of the variable tobe imputed and the outcome variable. This was donefive times to generate five completed datasets withplausible values for the missing variable.Backward selection was used to remove candidatepredictors that were not significantly associated withbirth within 7 days (Wald statistic p > 0.05) to obtainthe final model [34].The diagnostic performance of the final model wasassessed in terms of calibration capacity, stratificationcapacity, and classification accuracy [34]. Our goal wasto develop a pragmatic model with high sensitivity, lowfalse negative rate (of <5% based on the precedent set byprenatal diagnostic screening) [35–37], and also highnegative predictive value (NPV), to best identify womenfor whom an immediate administration of steroids maybe suboptimal. Calibration capacity was assessed by acalibration curve that indicated whether the proportionof women who delivered within 7 days in each group ofdeciles of modeled probability corresponded to the pre-dicted probability. Risk stratification capacity wasassessed by a classification table that showed whetherthe model had the capacity to distinguish between high-and low-risk groups, and whether these categories wereclinically meaningful. Finally, classification accuracy wasassessed by the extent to which the women who deliv-ered within 7 days had an increased predicted probabil-ity of doing so, and whether women who deliveredbeyond 7 days had a low predicted probability of deliverywithin 7 days (i.e., sensitivity and specificity, receiver oper-ating characteristic [ROC] curve). Internal validation ofthe model was performed using a bootstrap method onthe model development set with 200 iterations to generatebias-corrected ROC and calibration curves.Sensitivity analyses included assessment of other pre-existing medical and/or surgical conditions that may beassociated with preterm birth, and the impact of missingvalues on the predictive performance of the model byexcluding these cases. To ensure that the modelperformed well for subgroups of women that may beunder-represented in our population, we tested the finalmodel’s accuracy separately for women with singleton vs.multiple gestation, and for women admitted at 24–31 weeksvs. 32–34 weeks gestation. All statistical analyses wereperformed using R 3.1.1 (http://www.r-project.org).ResultsThere were 3012 women with a primary admission at24+0 to 28+6 weeks gestation or with a subsequentadmission up to 34+6 weeks gestation (after a previousadmission for other indications) who were eligible forthe study. Of these, 1473 (48.9%) women deliveredwithin 7 days of admission, while 1539 (51.1%) womendelivered after 7 days. Approximately one-third ofwomen (31.1%) delivered within 48 h (see Additionalfile 3), and 14.2% of women delivered at term gestation.Among women who delivered within 7 days of admis-sion (N = 1473), 846 (57.4%) received steroids onadmission to hospital. Among those who deliveredmore than 7 days after admission (N = 1539), 941(61.1%) received steroids on admission to hospital and598 (38.9%) did not receive steroids on admission tohospital. Thus, 1444 (47.9%) women received optimaltherapy on admission with respect to steroids (by eitherreceiving it and delivering within 7 days [N = 846], ornot receiving it and not delivering within the next7 days [N = 598]) and 1568 (52.1%) women receivedsuboptimal therapy on admission (by not receivingsteroids and delivering within 7 days [N = 627], orreceiving steroids and delivering more than 7 days later[N = 941]). In routine practice, however, it is recognizedthat a waiting period may be used after admissionbecause ongoing assessment allows for re-evaluation ofthe need for antenatal corticosteroids. In our study, 74women did receive antenatal corticosteroids after ad-mission (day 1–7) and delivered within 7 days, while138 women received antenatal corticosteroids afterTable 1 Demographic and clinical characteristics of women admitted to a tertiary hospital at 24–34 weeks gestation (Continued)Dilated cervix or prolapsed membranes without contractionsa 241 (16.4%) 154 (10.0%) <0.001Other associated complications/conditions at admissionAntepartum haemorrhagea 219 (14.9%) 123 (8.0%) <0.001Gestational hypertension (any) 9 (0.6%) 4 (0.3%) 0.170Gestational hypertension with proteinuria 7 (0.5%) 3 (0.2%) 0.217Intrauterine fetal growth restriction 18 (1.2%) 25 (1.6%) 0.604aFactors considered in the full model with a p value <0.10 (as highlighted in bold)bOther conditions include intra-abdominal infection (N = 92), asthma (N = 162), neurologic disease (N = 47), STDs (N = 82), gastrointestinal disease (N = 23),liver disease (N = 36), pre-existing thrombophilia (N = 22), pre-existing thromboembolism (N = 17), psychiatric disorders (N = 132), or other underlying medicalconditions (N = 27)cIncludes leiomyomas (N = 84), bicornuate uterus (N = 54), unicornuate uterus (N = 5), didelphic uterus (N = 13), and other (N = 16)De Silva et al. BMC Pregnancy and Childbirth  (2017) 17:202 Page 4 of 10admission and delivered more than 7 days after thesteroid administration (decreasing the overall optimaluse to 45.6% and increasing overall suboptimal use to54.4% in this cohort).Demographic characteristics and obstetric history,current pregnancy characteristics up to the time of admis-sion, and maternal interventions during hospitalizationamong women who delivered within 7 days from admis-sion vs. those who delivered later are described in Table 1.Women who delivered within 7 days after admission weremore likely to be younger, primiparous, have a history ofspontaneous abortion and preterm birth, have a single-ton pregnancy, smoke during pregnancy, and be admit-ted at higher gestational age. These women were alsomore likely to be admitted for preterm labour, PPROM,and dilated cervix or prolapsed membranes, and lesslikely to be admitted with short cervix without contrac-tions. They were more likely to have associated antepartumhaemorrhage.As expected, adverse perinatal and maternal healthoutcomes occurred more frequently among women whodelivered within 7 days, compared with those who deliv-ered later (see Additional file 4). Only 200 (6.6%) womenhad fibronectin testing data available, and 252 (8.4% over-all, or 37.4% among women with a short cervix) had arescue or elective cerclage for short cervix.There were no significant interactions identifiedbetween candidate predictors. The final model includedeight variables whose associations with delivery within7 days were significant. The crude and adjusted oddsratios and 95% confidence intervals are presented inTable 2. The predictive model equation was as follows:Risk score =12.13 – [0.41 × Maternal age ≥ 40years] –[1.08 × GA] + [0.02 × (GA)2] – [0.54 × Parity] + [0.32 ×Smoking] + [2.00 × Preterm labour] + [1.72 × PPROM] +[1.85 × Prolapsed membranes] + [0.67 × Antepartumhaemorrhage]; the probability of outcome (delivery<7 days) = 1/(1 + e-risk score). Based on this model, forexample, a woman who is 35 years old, nulliparous, non-smoking, and presents at 28 weeks’ gestation with PPROMat admission would have, on average, a 34% probability ofdelivery within 7 days, whereas the same woman present-ing with PPROM and preterm labour would have a prob-ability of 79%. Figure 1a displays a ROC curve for the finalmodel, with AUC = 0.724 (95% CI: 0.706–0.742).The calibration curve (Fig. 1b) showed good calibra-tion capacity of the final model, as the proportion ofwomen who delivered within 7 days in each risk cat-egory corresponded to the predicted risk. Predictedprobabilities that were very low (<20%) and high (>80%)were slightly overestimated.The final model had a relatively good calibration andstratification ability to distinguish between women withhigh vs. low risk of delivering within 7 days (Table 3).Among women with 0–15% predicted probability of de-livery within 7 days, 6% delivered within 7 days, whileamong those with ≥65% predicted probability, 75.4% de-livered within 7 days. Among women who deliveredwithin 7 days, only 1% had a low probability of such deliv-ery, while approximately 25% had a high probability.Table 2 Risk factors included in the final model predicting delivery within 7 days of admissionRisk factor OR [95% CI] Adjusted ORa [95% CI]Maternal age (yr)< 40 Reference Reference≥ 40 0.55 [0.39–0.78] 0.66 [0.45–0.97]ParityNulliparous Reference ReferenceParity ≥1 0.68 [0.59–0.78] 0.58 [0.50–0.68]Smoking during pregnancyb 1.35 [1.13–1.62] 1.37 [1.12–1.67]Gestational age (GA) on admissionc 1.08 [1.03–1.13] cMaternal conditionsPreterm labour 2.38 [2.05–2.76] 7.37 [5.85–9.29]PPROM 1.52 [1.30–1.76] 5.59 [4.42–7.07]Prolapsed membranes 1.75 [1.41–2.18] 6.36 [4.77–8.48]Associated antepartum haemorrhage 2.06 [1.63–2.60] 1.96 [1.53–2.52]OR odds ratio, PPROM preterm pre-labour rupture of membranesaadjusted for all other factors presented in the tablebmissing values were imputedcGestational age was modelled using higher order polynomials (see the equation below)Equation:Risk score =12.13 – [0.41 × Maternal age ≥ 40years] – [1.08 × GA] + [0.02 × (GA)2] – [0.54 × Parity] + [0.32 × Smoking] + [2.00 × Preterm labour] + [1.72 × PPROM] +[1.85 × Prolapsed membranes] + [0.67 × Antepartum haemorrhage]the probability of outcome (delivery <7 days) = 1/(1 + e-risk score)De Silva et al. BMC Pregnancy and Childbirth  (2017) 17:202 Page 5 of 10Different thresholds of probability were considered toidentify women unlikely to deliver within 7 days. Table 4shows a good stratification ability of the model withmeaningful probability categories. We used probabilitycut-offs as follows: ≤15% as our lowest threshold inorder to have a sufficient number of women included inthis category, <27% to minimise the false negativerate to <5% (based on similar prenatal diagnostic test-ing criteria) [35, 38], <50% to be better than the flipof a coin, and ≥65% a threshold with a sufficientnumber of women in this ‘high probability of delivery’category. If the optimal cut-off point were to maxi-mise negative predictive probability (NPV) with achosen acceptable false negative rate of <5%, the opti-mal threshold of predicted probability would be ≤27%(sensitivity 96.2% with the lower 95% CI limit of95.1%; NPV of 89.0%, 95% CI 86.0–91.5). At thisthreshold, corticosteroid treatment would be given to2502 (83.1%) women and withheld from 510 (16.9%), withtreatment that would be optimal in 1871 (62.1%) womenand suboptimal in 1141 (37.9%), a significant differencecompared with values in our cohort of 47.9% and 52.1%,respectively (p-value <0.001). The corresponding negativelikelihood ratio of 0.13 (95% CI 0.10–0.17) indicates ahighly informative test in terms of ruling out deliverywithin 7 days. However, in practice, the predicted prob-ability of delivery within 7 days would best be used as acontinuous value to customise management.Internal validation of the final model using a bootstrapmethod with 200 repetitions yielded an AUC optimismof 0.004, with bias-corrected AUC = 0.720 (95% CI0.702–0.738). The bias-corrected calibration curve didnot exhibit any appreciable changes compared with theinitial curve (see Additional file 5).Fig. 1 Graphical presentation of prognostic performance of the final model to predict delivery within 7 days of admission among a high-riskcohort. a Area under the ROC curve. b Calibration curveDe Silva et al. BMC Pregnancy and Childbirth  (2017) 17:202 Page 6 of 10Sensitivity analyses were performed including pre-existing medical and/or surgical conditions (such asrenal disease, rheumatic disease, cervical procedures,cardiac disease) in the full model but were not foundto be significant predictors. Sensitivity analysis wasalso performed excluding women with missing values.Of the variables in the final model, only smoking dur-ing pregnancy had missing values (N = 23); the resultsremained unchanged after exclusion of those cases(see Additional file 6).We performed additional analyses to examine the ac-curacy of the final model among subgroups of womenthat may be under-represented in our population. Weapplied the final model to women with singleton vs.multiple pregnancy (see Additional file 7); the predictiveaccuracy for each subgroup was similar to the overallmodel performance (AUC 0.71 [0.69, 0.73] and AUC0.77 [0.73, 0.81], respectively). Similarly, we appliedthe model to women admitted at 24–31 weeks vs.32–34 weeks gestation (see Additional file 8). Thepredictive accuracy was similar for women admittedat 24–31 weeks (AUC 0.72 [0.71, 0.74]), however, wecould not test all model parameters for the subgroupof women admitted later due to low sample size.DiscussionMain findingsOur large retrospective cohort study included womenadmitted to a tertiary hospital for conditions that putthem at high risk for preterm birth, primarily at 24+0and 29+6 weeks gestation. Approximately 49% deliveredwithin 7 days after admission, while 51% delivered after7 days (14% delivered at term). We constructed a parsi-monious predictive model to identify women who are atrisk of delivery within the next 7 days in order to opti-mise administration of antenatal corticosteroids. Themodel had a fair predictive accuracy (AUC of 0.73), goodcalibration capacity and stratification ability, and it wasinternally validated using a bootstrap method. Althoughthe predictive accuracy of our model was recognised tobe fair, it is based on information collected in routineclinical care and it would significantly improve optimalTable 3 Stratification ability of the predictive model to identify women with and without the outcome (delivery within 7 days)Delivery within 7 days Delivery after 7 days TotalPredictedprobabilityN Calibration(% in probabilitycategory)Stratification(% of all womenwith outcome)N Calibration(% in probabilitycategory)Stratification(% of all womenwithout outcome)0–14.9% 14 6.0 1.0 219 94.0 14.2 23315.0–19.9% 34 15.1 2.3 191 84.9 12.4 22520.0–49.9% 392 41.2 26.6 560 58.8 36.4 95250.0–64.9% 668 59.7 45.3 450 40.3 29.2 1118≥65.0% 365 75.4 24.8 119 24.6 7.7 484Total 1473 100 1539 100 3012Table 4 Prognostic accuracy of delivery within 7 days at various cut-off points of predicted probabilityPredicted probability (%) <15.0a <27.0 <50.0 <65.0Number of women 233 510 1410 2528N women delivered at ≤7 days (N = 1473 overall) 14 56 440 1108N women delivered at >7 days (N = 1539 overall) 219 454 970 1420Sensitivity 99.1% (98.4–99.4) 96.2% (95.1–97.1) 70.1% (67.7–72.4) 24.8% (22.6–27.1)Specificity 14.2% (12.6–16.1) 29.5% (27.3–31.8) 63.0% (60.0–65.4) 92.3% (90.8–93.5)False positive rate 85.8% (83.9–87.4) 70.5% (68.2–72.7) 37.0% (34.6–40.0) 7.7% (6.4–9.2)False negative rate 0.9% (0.6–1.6) 3.8% (2.9–4.9%) 29.9% (27.6–32.3) 75.2% (72.9–77.4)Positive predictive value 52.5% (50.6–54.4) 56.6% (54.7–58.6) 64.5% (62.1–66.8) 75.4% (71.4–79.0)Negative predictive value 94.0% (90.2–96.4) 89.0% (86.0–91.5) 68.9% (66.3–71.2) 56.2% (54.2–58.1)Positive likelihood ratio 1.16 (1.13–1.18) 1.37 (1.32–1.41) 1.90 (1.76–2.04) 3.93 (3.15–4.91)Negative likelihood ratio 0.07 (0.04–0.11) 0.13 (0.10–0.17) 0.47 (0.43–0.52) 0.82 (0.79–0.84)ae.g., if women with predicted probability <15% were considered at low-risk of delivery within 7 days, the sensitivity of the prognostic tool would be 99.1% whilethe specificity would be 14.2De Silva et al. BMC Pregnancy and Childbirth  (2017) 17:202 Page 7 of 10steroid treatment on admission, (from 47.9% to 62.1% inour cohort) by decreasing the number of women who donot but should have received steroids (false negative rate).The proportion of women with optimal treatment couldbe maximised further with a different probability cut-off,recognising that the ideal trade-off between sensitivity andspecificity will depend on the clinician’s and patient’s viewof potential benefits and harms associated with either pro-viding or postponing treatment.InterpretationOptimising antenatal corticosteroid use requires ac-curate identification of women who will deliver within7 days and benefit from this treatment. However,preterm birth is notoriously difficult to predict [19].Meta-analyses of individual prognostic factors for pre-term delivery showed that fetal fibronectin, absence offetal breathing movements, and cervical length havepotential for diagnostic use [20]. However, recentstudies have shown that fetal fibronectin testing doesnot have an optimal clinical utility, as it does not pre-vent preterm birth nor prevent adverse perinatal out-comes among women with threatened pretem labour,and is further associated with increased cost [21, 23, 24].Only fetal breathing movements as a prognostic testconsistently yielded a likelihood ratio of a highlyinformative test (LR+ >10). However, these meta-analyses included very heterogeneous studies examin-ing spontaneous preterm delivery within 24 h, 7, and10 days among women with preterm labour [18, 20].In contrast to our study, these predictors were exam-ined as individual tests, and not in combination withother determinants of early preterm delivery, and assuch could not be used to estimate the full range ofprobability of delivery. Only a few studies focused dir-ectly on predictive modeling of early delivery amongselected groups of women, including those with mul-tiple pregnancies [39] or those transferred from onehospital to a higher-level hospital [26]. The latterstudy identified predictors of delivery within 48 h oftransfer, of which seven were similar to our result,with the addition of sonographic length of uterinecervix (a shorter length indicated a higher probabilityof delivery). Other studies reporting the associationbetween preterm birth and preterm pre-labour rup-ture of membranes, prolapsed membranes, vaginalbleeding, cervical dilatation, preterm labour, older ma-ternal age (>40 years), higher parity, or smoking areconsistent with our results [20, 32]. However, thesestudies did not focus on predictive probability of pre-term birth for individual women. Inclusion of theseknown key risk factors as predictors in our modelthus increases its face validity.Strengths and limitationsA strength of this study is that it includes a large popu-lation dataset with detailed clinical information collectedat the time of admission to the hospital with threatenedpreterm birth, the point in time at which clinicians mustdecide whether to give antenatal corticosteroids so thatthey will have the maximal effect by 48 h beforedelivery. However, there are limitations to our study.First, the time of admission was not recorded in theCPN database, and thus our outcome potentiallyincluded women delivering within 7 days and 23 hafter admission. Second, the predictive model is basedsolely on patient and pregnancy characteristics at thetime of admission and the reason(s) for admission.The predictive probability of delivery within 7 days isrelevant to the event of admission for hospital care,and needs to be reassessed based on events thatoccur afterwards. For example, a low probability ofearly delivery may lead to a decision not to adminis-ter corticosteroids at admission; however, subsequentruptured membranes during hospitalization wouldprompt reassessment of risk. Unfortunately, we didnot have detailed follow-up information required toconstruct a time-varying model for multiple reassess-ment of the decision about timing of birth. Third, wedid not account for inter-centre variability recognisingdifferences between centres, although we tried tolimit this by excluding any practice-related predictorsto increase generalisability. Fourth, we used internalvalidation to estimate the generalisability of ourmodel. However, external validation (for example, in asimilar cohort of women at risk of preterm delivery,or women admitted at 32–34 weeks for which therewere few cases) is needed to better assess theperformance of the predictive model in other settings.Finally, we did not have reliable information on pre-pregnancy body mass index and assisted reproductivetechnology use, which may constitute important riskfactors (41.9% and 76.6% of values were missing,respectively).ConclusionIn conclusion, we propose a useful tool to improveprediction of delivery within 7 days at the time ofadmission among women with threatened pretermbirth primarily at 24+0 and 29+6 weeks. Until the dis-covery of novel biomarkers can improve prediction,such a tool can significantly improve rates of optimalsteroid use by increasing the proportion of womenwho receive steroids within 7 days prior to delivery,and decrease rates of suboptimal steroid use by notimmediately treating women with a very low probabil-ity of delivery within 7 daysDe Silva et al. BMC Pregnancy and Childbirth  (2017) 17:202 Page 8 of 10Additional filesAdditional file 1: Table S1. A list of all members of the collaborativegroup in CPN. (DOCX 13 kb)Additional file 2: Table S2. Definitions of conditions and variables asused in the Canadian Perinatal Network. (DOCX 13 kb)Additional file 3:Figure S1. Kaplan-Meier curve showing the proportionof women who remained pregnant from the time that they were admitted tohospital and identified as being at risk of delivery within 7 days. (DOCX 26 kb)Additional file 4: Table S3. Pregnancy outcomes among women whopresented at 24–34 weeks. (DOCX 15 kb)Additional file 5: Figure. S2. Corrected calibration curve of the finalmodel after internal validation. (DOCX 27 kb)Additional file 6: Table S4. Sensitivity analysis of the final model afterexcluding women with missing values. (DOCX 13 kb)Additional file 7: Table S5. Sensitivity analysis of the final modelamong singleton and multiple pregnancies. (DOCX 13 kb)Additional file 8: Table S6. Sensitivity analysis of the final modelrestricting to women <32 weeks. (DOCX 14 kb)AbbreviationsAUC: Area under the curve; CNS: Central nervous system; CPN: CanadianPerinatal Network; GA: Gestational age; LMICs: Low- and middle-incomecountries; LR-: Negative likeliehood ratio; LR+: Positive likelihood ratio;NPV: Negative predictive value; OR: Odds ratio; PPROM: Preterm pre-labourrupture of membranes; ROC: Receiver operating characteristic [curve]AcknowledgementsNot applicable.FundingThe Canadian Perinatal Network was funded by the Canadian Institutes ofHealth Research and from the Ontario Ministry of Health and Long-termCare. The funders of the study had no role in study design, data collection,data analysis, data interpretation, or writing of the report.Availability of data and materialsAdditionaly sensitivity analyses and graphs are included as supplementary files.The anonymized dataset analyzed in this current study is available from thecorresponding author on reasonable request.Authors’ contributionsDAD, SL, and LAM conceived the idea for the study. DAD carried out theanalysis and DAD and SL wrote the first draft of the paper. LAM, PvD, andARS provided their content expertise and the writing of the paper, with allauthors approving the final manuscript.Ethics approval and consent to participateThe Canadian Perinatal Network received ethics approval centrally as aquality assurance project by the Research Ethics Boards at the Universityof British Columbia (H05–70359) and at each study site. As such, writtenconsent was not required, and any collected information was anonymisedand de-identified prior to analysis.Consent for publicationNot applicable.Competing interestsThe authors declare that they have no competing interests.Publisher’s NoteSpringer Nature remains neutral with regard to jurisdictional claims inpublished maps and institutional affiliations.Author details1Department of Obstetrics & Gynaecology, University of British Columbia,C420-4500 Oak Street, Vancouver, BC V6H 3N1, Canada. 2School ofPopulation and Public Health, University of British Columbia, 2206 E. Mall,Vancouver, BC V6T 1Z9, Canada. 3St. George’s University Hospitals NHSFoundation Trust, Blackshaw Road, Tooting, London SW17 0QT, UK.4Molecular & Clinical Sciences Research Institute, St. George’s University ofLondon, Rm J0.27, Jenner Wing, Cranmer Terrace, London SW17 0RE, UK.5Division of Neonatology, Department of Paediatrics, University of BritishColumbia, 1R14-4500 Oak Street, Vancouver, BC V6H 3N1, Canada.Received: 3 March 2017 Accepted: 21 June 2017References1. Public Health Agency of Canada. Canadian Perinatal Health Report. Ottawa:Minister of Public Works and Government Services Canada; 2008.2. Magee L, Sawchuck D, Synnes A, von Dadelszen P. Magnesium Sulphate forFetal Neuroprotection Consensus Committee. SOGC Clinical PracticeGuideline. Magnesium sulphate for fetal neuroprotection. J Obstet GynaecolCan. 2011;33(5):516–29.3. Roberts D, Dalziel SR. Antenatal corticosteroids for accelerating fetal lungmaturation for women at risk of preterm birth. Cochrane Database Syst Rev.2006;19(3):CD004454.4. Kazem M, Hutcheon JA, Joseph KS. A population-based study of antenatalcorticosteroid prophylaxis for preterm birth. J Obstet Gynaecol Can. 2012;34(9):842–8. doi:10.1016/S1701-2163(16)35383-X.5. Crane J, Armson A, Brunner M, De La Ronde S, Farine D, Keenan-Lindsay L,et al. Antenatal corticosteroid therapy for fetal maturation. J ObstetGynaecol Can. 2003 Jan;25(1):45–52.6. Gilstrap LC, Christensen R, Clewell WH, D’Alton ME. National Institutes ofHealth Consensus Development Panel. Effect of corticosteroids for fetalmaturation on perinatal outcomes. J Am Med Assoc. 1995;273(5):413–7.7. National Institute for Health and Care Excellence. Preterm labour and birth.2015. Available at: https://www.nice.org.uk/guidance/ng25?unlid=9291036072016213201257.8. Mahony R, McKeating A, Murphy T, McAuliffe F, O’Herlihy C, Foley M.Appropriate antenatal corticosteroid use in women at risk for preterm birthbefore 34 weeks of gestation. BJOG. 2010;117(8):963–7. doi:10.1111/j.1471-0528.2010.02590.x.9. Sanya R, Al Naggar E, Gasim M, Ahmed BI. Use or overuse of antenatalcorticosteroids for suspected preterm birth. J Matern Fetal Neonatal Med.2014;27(14):1454–6. doi:10.3109/14767058.2013.860523.10. Razaz N, Skoll A, Fahey J, Allen V, Joseph K. Trends in Optimal, Suboptimal,and Questionably Appropriate Receipt of Antenatal CorticosteroidProphylaxis. Obstet Gynecol. 2015;125(2):288–96. doi:10.1097/AOG.0000000000000629.11. Crowther CA, McKinlay CJD, Middleton P, Harding JE. Repeat doses ofprenatal corticosteroids for women at risk of preterm birth for improvingneonatal health outcomes. Cochrane Database Syst Rev. 2015;(7): CD003935.12. Murphy KE, Hannah ME, Willan AR, et al. Multiple courses of antenatalcorticosteroids for preterm birth (MACS): a randomised controlled trial.Lancet. 2008;372(9656):2143–51. doi:10.1016/S0140-6736(08)61929-7.13. French NP, Hagan R, Evans SF, Godfrey M, Newnham JP. Repeated antenatalcorticosteroids: Size at birth and subsequent development. Am J ObstetGynecol. 1999;180(1):114–21. doi:10.1016/S0002-9378(99)70160-2.14. Stalnacke J, Diaz Heijtz R, Norberg H, Norman M, Smedler AC, Forssberg H.Cognitive outcome in adolscents and young adults after repeat courses ofantenatal corticosteroids. J Pediatr. 2013;163(2):441–6.15. Asztalos EV, Murphy KE, Willan AR, et al. Multiple Courses of AntenatalCorticosteroids for Preterm Birth Study: Outcomes in Children at 5 Years ofAge (MACS-5). JAMA Pediatr. 2013;167(12):1102–10. doi:10.1001/jamapediatrics.2013.2764.16. Asztalos E, Willan A, Murphy K, et al. Association between gestational age atbirth, antenatal corticosteroids, and outcomes at 5 years: multiple courses ofantenatal corticosteroids for preterm birth study at 5 years of age (MACS-5).BMC Pregnancy Childbirth. 2014;14(1):272. doi:10.1186/1471-2393-14-272.17. Althabe F, Belizán JM, EM MC, et al. A population-based, multifacetedstrategy to implement antenatal corticosteroid treatment versusstandard care for the reduction of neonatal mortality due to pretermbirth in low-income and middle-income countries: the ACT cluster-randomised trial. Lancet. 2015;385(9968):629–39. doi:10.1016/S0140-6736(14)61651-2.De Silva et al. BMC Pregnancy and Childbirth  (2017) 17:202 Page 9 of 1018. Honest H, Forbes CA, Durée KH, et al. Screening to prevent spontaneouspreterm birth: systematic reviews of accuracy and effectiveness literaturewith economic modelling. Health Technol Assess. 2009;13(43):1.19. Honest H, Bachmann LM, Sundaram R, Gupta JK, Kleijnen J, Khan KS. Theaccuracy of risk scores in predicting preterm birth – a systematic review. JObstet Gynaecol. 2004;24(4):343–59. doi:10.1002/uog.202.20. Boots AB, Sanchez-Ramos L, Bowers DM, Kaunitz AM, Zamora J, Schlattmann P.The short-term prediction of preterm birth: a systematic review and diagnosticmetaanalysis. Am J Obstet Gynecol. 2014;210:10. doi:10.1016/j.ajog.2013.09.004.21. Esplin MS, Elovitz MA, Iams JD, et al. Predictive Accuracy of SerialTransvaginal Cervical Lengths and Quantitative Vaginal Fetal FibronectinLevels for Spontaneous Preterm Birth Among Nulliparous Women. JAMA.2017;317(10):1047–56. doi:10.1001/jama.2017.1373.22. Conde-Agudelo A, Papageorghiou A, Kennedy S, Villar J. Novel biomarkersfor the prediction of the spontaneous preterm birth phenotype: asystematic review and meta-analysis. BJOG. 2011;118(9):1042–54. doi:10.1111/j.1471-0528.2011.02923.x.23. Berghella V, Saccone G. Fetal fibronectin testing for prevention of pretermbirth in singleton pregnancies with threatened preterm labor: a systematicreview and metaanalysis of randomized controlled trials. Am J ObstetGynecol. 2016;215(4):431–8. doi:10.1016/j.ajog.2016.04.038.24. Macones GA. Fetal fibronectin testing in threatened preterm labor: time tostop. Am J Obstet Gynecol. 2016;215(4):405. doi:10.1016/j.ajog.2016.07.057.25. Magee LA, von Dadelszen P, Allen VM, et al. The Canadian PerinatalNetwork: A National Network Focused on Threatened Preterm Birth at 22 to28 Weeks’ Gestation. J Obstet Gynaecol Can. 2011;33(2):111–20. doi:10.1016/S1701-2163(16)34795-8.26. Allouche M, Huissoud C, Guyard-Boileau B, Rouzier R, Parant O.Development and validation of nomograms for predicting preterm delivery.Am J Obstet Gynecol. 2011;204(3):242.e8. doi:10.1016/j.ajog.2010.09.030.27. Moons KGM, Altman DG, Vergouwe Y, Royston P. Prognosis and prognosticresearch: application and impact of prognostic models in clinical practice.BMJ. 2009;338(7709):1487–90. doi:10.1136/bmj.b606.28. Moons KGM, Royston P, Vergouwe Y, Grobbee DE, Altman DG. Prognosisand prognostic research: what, why, and how? BMJ. 2009;338(7706):1317–20. doi:10.1136/bmj.b375.29. Steyerberg EW. Clinical Prediction Models: A Practical Approach toDevelopment, Validation, and Updating. New York: Springer; 2009.30. Nykjaer C, Alwan NA, Greenwood DC, Simpson NAB, Hay AWM, White KLM,et al. Maternal alcohol intake prior to and during pregnancy and risk ofadverse birth outcomes: evidence from a British cohort. J EpidemiolCommunity Health. 2014;68(6):542–9. doi:10.1136/jech-2013-202934.31. Bhattacharya S, Lowit A, Bhattacharya S, Raja EA, Lee AJ, Mahmood T, et al.Reproductive outcomes following induced abortion: a national register-based cohort study in Scotland. BMJ Open. 2012;2(4):e000911. doi:10.1136/bmjopen-2012-000911.32. Goldenberg RL, Culhane JF, Iams JD, Romero R. Epidemiology and causes ofpreterm birth. Lancet. 2008;371(9606):75–84. doi:10.1016/S0140-6736(08)60074-4.33. Vergouwe Y, Royston P, Moons KGM, Altman DG. Development and validationof a prediction model with missing predictor data: a practical approach. J ClinEpidemiol. 2010;63(2):205–14. doi:10.1016/j.jclinepi.2009.03.017.34. Royston P, Moons KGM, Altman DG, Vergouwe Y. Prognosis and prognosticresearch: Developing a prognostic model. BMJ. 2009;338(7707):1373–7. doi:10.1136/bmj.b604.35. Malone FD, Canick JA, Ball RH, et al. First-Trimester or Second-TrimesterScreening, or Both, for Down’s Syndrome. N Engl J Med. 2005;353(19):2001–11. doi:10.1056/NEJMoa043693.36. Wapner R, Thom E, Simpson JL, et al. First-Trimester Screening for Trisomies21 and 18. N Engl J Med. 2003;349(15):1405–13. doi:10.1056/NEJMoa025273.37. Wald NJ, Rodeck C, Hackshaw AK, Walters J, Chitty L, Mackinson AM,SURUSS Research Group. First and second trimester antenatal screening forDown’s syndrome: the results of the Serum, Urine and UltrasoundScreening Study (SURUSS). J Med Screen 2003; 10(4): 56–104.38. Spencer K, Spencer CE, Power M, Dawson C, Nicolaides KH. Screening forchromosomal abnormalities in the first trimester using ultrasound and maternalserum biochemistry in a one-stop clinic: a review of three years prospectiveexperience. BJOG. 2003;110(3):281–6. doi:10.1046/j.1471-0528.2003.02246.x.39. van de Mheen L, Schuit E, Lim AC, et al. Prediction of preterm birth inmultiple pregnancies: development of a multivariable model includingcervical length measurement at 16 to 21 weeks’ gestation. J ObstetGynaecol Can. 2014;36(4):309–19. doi:10.1016/S1701-2163(15)30606-X.•  We accept pre-submission inquiries •  Our selector tool helps you to find the most relevant journal•  We provide round the clock customer support •  Convenient online submission•  Thorough peer review•  Inclusion in PubMed and all major indexing services •  Maximum visibility for your researchSubmit your manuscript atwww.biomedcentral.com/submitSubmit your next manuscript to BioMed Central and we will help you at every step:De Silva et al. BMC Pregnancy and Childbirth  (2017) 17:202 Page 10 of 10


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items