你在这里

利用两个经修改的预测模型个体化评估早产风险

European Journal of Obstetrics & Gynecology and Reproductive Biology, pages 42 - 48

Abstract

Objectives

To construct two prediction models for individualized assessment of preterm delivery risk within 48 h and before completed 32 weeks of gestation and to test the validity of modified and previously published models.

Study design

Data on 617 consecutive women with preterm labor transferred to a tertiary care center for threatened preterm delivery between 22 and 32 weeks of gestation were analysed. Variables predicting the risk of delivery within 48 h and before completed 32 weeks of gestation were assessed and applied to previously published prediction models. Multivariate analyses identified variables that were incorporated into two modified models that were subsequently validated.

Results

Two modified prediction models were developed and internally validated, incorporating four and six of the following variables to predict the risk of delivery within 48 h and before completed 32 weeks of gestation, respectively: presence of preterm premature rupture of membranes and/or vaginal bleeding, sonographic cervical length, week of gestation, fetal fibronectin, and serum C-reactive protein. The correspondence between the actual and the predicted preterm birth rates suggests excellent calibration of the models. Internal validation analyses for the modified 48 h and 32 week prediction models revealed considerably high concordance-indices of 0.8 (95%CI: [0.70–0.81]) and 0.85 (95%CI: [0.82–0.90]), respectively.

Conclusions

Two modified prediction models to assess the risk of preterm birth were constructed and validated. The models can be used for individualized prediction of preterm birth and allow more accurate risk assessment than based upon a single risk factor. An online-based risk-calculator was constructed and can be assessed through: http://cemsiis.meduniwien.ac.at/en/kb/science-research/software/clinical-software/prematurebirth/ .

Keywords: Nomogram, Preterm birth, Premature labor, Prediction tool.

Introduction

Preterm births account for 75% of perinatal mortality and more than half the long-term morbidity[1] and [2]. Preterm labor is now thought to be a syndrome initiated by multiple mechanisms, including infection or inflammation, utero-placental ischemia or hemorrhage, stress, and other immunologically-mediated processes [3] . The observation that several pathways are involved in its pathogenesis may explain why premature delivery has proved so difficult to predict and prevent[4] and [5]. A short cervical length and a raised cervicovaginal fetal fibronectin (fFN) concentration are the strongest predictors of preterm birth[6] and [7]. In addition several other accepted risk factors for preterm delivery include previous history of preterm birth, decidual hemorrhage manifested as vaginal bleeding, and infection[8], [9], and [10]. Although preterm labor is one of the most common reasons for hospitalization of pregnant women, identifying women with preterm contractions who will actually deliver preterm is challenging and often imprecise [5] . Correspondingly, investigators have attempted to elucidate the factors that are associated with preterm delivery in order to prompt consideration of interventions when the risk of preterm delivery is high.

Nomograms depict predictive models which estimate the probability of a specific outcome [11] . They are widely applied tools in clinical practice and can be used to improve patient counseling and treatment planning [12] . However, only a few nomograms have been published in obstetrics[13], [14], and [15]. Recently two models to predict preterm delivery within 48 h and before completed 32 weeks of gestation in women with preterm labor were constructed [13] . The aim of the present study was to construct two modified prediction models for assessment of preterm birth risk by incorporating relevant predictive parameters in order to allow individualized and accurate prediction of preterm birth.

Materials and methods

Patients

In this cohort study data were abstracted from a large single institution's database that was prospectively maintained and electronic medical records were reviewed. In this study all consecutive women between 22 and 32 weeks of gestation who were admitted for threatened labor to our tertiary perinatal center between 2007 and 2012 and delivered at the institution were eligible for inclusion. The diagnosis of threatened preterm delivery was based on clinical evidence of painful uterine contractions and/or cervical dilatation. Women were included, regardless of whether amniotic membranes were intact or ruptured. Gestational age was assigned on the basis of the last menstrual period and confirmed by first or early second-trimester sonography. Women whose pregnancies were complicated by preeclampsia, fetal growth restriction, in utero fetal death, or major fetal anomaly, and multiple gestation pregnancies were not included. The diagnosis of preterm labor was generally based upon clinical criteria of regular painful uterine contractions accompanied by cervical change shortening or contractions visible during contraction monitoring. Initial evaluation included cardiotocography, assessment of patient's past and present obstetrical and medical history, and assessment of gestational age. A speculum examination was performed and swabs for bacteriological cultures and for fetal fibronectin (fFN) were obtained. The presence of vaginal bleeding was documented. The diagnosis of preterm premature rupture of membranes (PPROM) was assessed clinically based on visualization of amniotic fluid in the vagina and confirmed by detection of placental alpha microglobulin-1. Trans-vaginal and trans-abdominal ultrasound examination to assess cervical length, maternal and fetal anatomic abnormalities, confirm the fetal presentation, assess amniotic fluid volume, and estimate fetal weight was performed. Serum blood samples were taken including blood count and assessment of serum C-reactive protein (CRP) levels (normal value <0.5 mg/dl). According to international guidelines two courses of betamethasone were administered to women with preterm uterine contractions who were considered high-risk (based on cervix length and fFN) for preterm birth before completed 32 weeks of gestation accompanied with administration of a tocolytic agent i.e. atosiban or hexoprenalin-sulfate. Appropriate antibiotics were administered to women with positive bacteriological culture results or clinical signs of infection and to women with PPROM. Time to delivery was calculated by the date and time of admission to the institution and the date and time of birth. Outcome variables were defined as delivery within 48 h and delivery before completion of 32 weeks of gestation.

Statistical analysis

Statistical methods are provided as annexed files.

Results

Patients

In total, 682 patients with premature labor who were admitted to the tertiary perinatal center at the Department of Obstetrics and Gynecology of the Medical University of Vienna between January 2007 and October 2012 and delivered at the institution were identified. Of these patients, 617 met the inclusion criteria, had all necessary variables documented, and were selected for analysis. Patient characteristics of the previously published Toulouse cohort and the Austrian cohort are provided in Table 1 . Both cohorts were mainly composed of Caucasian women. One of the main differences between the two cohorts was a significantly higher rate of women who were admitted before 24 weeks of gestation in the Austrian cohort. Delivery rate within 48 h after transfer was slightly higher in the Austrian cohort, whereas delivery rates before completed 32 weeks of gestation were comparable between both cohorts. Mean (SD) CRP serum levels in the Austrian cohort was 1.3 (2.1) mg/dl, a positive fFN test was obtained in 128 (24.0%) of patients and the mean (SD) observed time between admission and delivery was 5.6 (5.4) weeks. Cumulative rates of women still pregnant after admission in relation to gestational week of delivery and time between hospital admission and delivery of the Austrian cohort are shown in Fig. 1 .

Table 1 Patient characteristics of the Austrian (N = 617) and the published Toulouse (N = 737) cohorts.

Parameter Austrian cohort Toulouse cohort
Maternal age at diagnosis (years)    
 Mean 30 29
 
Gestational age at time of admission (weeks)    
 Mean 27 29
N (%) < 24 weeks 145 (24) 17 (2)
N (%) 24–28 weeks 278 (45) 265 (36)
N (%) 29–32 weeks 194 (31) 455 (62)
 
History of preterm delivery or late miscarriage    
N (%) 95 (15.4) 103 (14.0)
 
Cerclage    
N (%) 37 (6.0) 21 (2.8)
 
Clinical characteristics at admission    
 
 PPROM
  N (%) 159 (25.8) 172 (23.3)
 
 Vaginal bleeding    
  N (%) 37 (6.0%) 61 (8.3)
 
 Functional cervical length (mm)    
  Mean 18 17
  N (%) < 15 mm 292 (48.0) 284 (38.5)
  N (%) 15–25 mm 160 (26.3) 194 (26.3)
  N (%) > 25 mm 156 (25.7) 259 (35.1)
 
 Contractions requiring tocolysis    
  N (%) 562 (91.1) 577 (78.3)
 
Outcome    
 Delivery within 48 h 104 (16.9) 157 (21.3)
 Delivery before 32 weeks of gestation 300 (48.6) 317 (43.0)

PPROM: preterm premature rupture of the membranes.

gr1

Fig. 1 Time to delivery and rates of pregnant women during the study period (N = 617).

External validation of the Toulouse prediction models

First we intended to test the validity of the two prediction models published previously by Allouche and colleagues [13] . Thec-indices (95%CI) for the 48-hour model and the 32-weeks model were 0.77 [0.71, 0.82] and 0.81 [0.78, 0.84] in the Austrian validation cohort compared to 0.73 [0.66–0.80] and 0.72 [0.67–0.78] in the Toulouse validation cohort, respectively. The miscalibration values for both models were −0.28 (p < 0.001) and 0.16 (p = 0.11), respectively. Results suggest a significant miscalibration of the 48-hour model and a good calibration of the 32-weeks model, when tested in the Austrian validation cohort. For the 48-hour model the large scale calibration index was −0.29 (p = 0.03), suggesting a significant overestimation of the average event rate (number of preterm deliveries). For the 32-weeks model the large scale calibration index was −0.08 (p = 0.39), suggesting good estimation of the average event rate. These results can be seen graphically in the calibration plot ( Fig. 2 ).

gr2

Fig. 2 Calibration plots of the external validation of the Toulouse prediction models on the Austrian cohort. The dashed line indicates perfect calibration of predictions, while the observed calibration is shown with a smoothed curve (dotted line) and with subgroup estimates of deciles (triangles) (A) 48 h model (B) 32 weeks model.

Construction of two modified prediction models

In a subsequent step we intended to improve the performance of the two published prediction models. Two novel prediction models were created and additional predictors for preterm delivery were incorporated stepwise. The following variables were investigated for both models: PPROM, presence of vaginal bleeding, contractions requiring tocolysis, sonographic cervical length, CRP, fFN test results, and history of preterm delivery or late miscarriage. After testing for nonlinearity and after variable selection based on the results of multivariable analysis ( Table 2 ), the model for predicting preterm birth within 48 h was reduced to the three variables PPROM, CRP and a nonlinear function for sonographic cervical length. Nonlinearity was not observed in any of the variables of the 32-weeks model and variable selection yielded a model incorporating the following six variables: PPROM, presence of vaginal bleeding, sonographic cervical length, week of gestation, serum CRP, and fFN. No clinically meaningful and statistically significant interactions were found in none of the two modified models. The results of the multivariate regression analyses are shown in Table 2 . Based on the results of the multivariate regression analyses two novel nomograms for delivery within 48 h after admission and for delivery before completed 32 weeks of gestation were created and are provided in Figs. 3 A and 4 A, respectively. Fig. 5 A and B provide examples of predictions of two women presenting with preterm labor. We developed an internet-based interface and provide an electronic user-friendly prediction tool that can be assessed online by copying the following link: http://cemsiis.meduniwien.ac.at/en/kb/science-research/software/clinical-software/prematurebirth/ .

Table 2 Multivariate logistic regression analyses of variables associated with preterm delivery of the Austrian cohort.

Parameter Delivery within 48 h (N = 592) Delivery before 32 weeks (N = 476)
OR 95%CI p-Value OR 95%CI p-Value
PPROM (yes vs. no) 3.30 1.88–5.79 <0.0001 34.36 14.67–80.48 <0.0001
Presence of vaginal bleeding 11.94 2.61–54.59 0.0008
Contractions requiring tocolysis
Sonographic cervical length 0.95 0.92–0.97 <0.0001
 9 vs. 16 mm 1.99 1.60–2.47 <0.0001
 26 vs. 16 mm 0.63 0.48–0.82
Weeks of gestation at admission 0.84 0.77–0.92 <0.0001
CRP a 1.47 1.26–1.71 <0.0001 1.38 1.18–1.62 <0.0001
Fetal Fibronectin (positive vs. negative) 2.09 1.16–3.76 0.0096
History of preterm delivery/late miscarriage

a Per value doubling.

OR: odds ratio, 95%CI: 95% Confidence interval, PPROM: premature, preterm rupture of the membranes, CRP: C-reactive protein.

gr3

Fig. 3 Modified prediction model of preterm birth within 48 h after admission. (A) Nomogram. Instructions: Predictive power of maternal serum and amniotic fluid CRP and PAPP-A concentrations at the time of genetic amniocentesis for the preterm delivery Locate patient's variable on the corresponding axis. Draw a line to the points axis. Sum the points. Draw a line from the total points axis to the delivery before 48 h after admission probability axis (B) Calibration plot. The dashed line indicates perfect calibration of predictions, while the observed calibration is shown with a smoothed curve (dotted line) and with subgroup estimates of deciles (triangles).

gr4

Fig. 4 Modified prediction model of preterm birth before completed 32 weeks of gestation. (A) Nomogram. Instructions: Locate patient's variable on the corresponding axis. Draw a line to the Points axis. Sum the points. Draw a line from the total points axis to the delivery before 48 h after admission probability axis (B) calibration plot. The dashed line indicates perfect calibration of predictions, while the observed calibration is shown with a smoothed curve (dotted line) and with subgroup estimates of deciles (triangles).

gr5

Fig. 5 Assessment of preterm birth risk using an online calculator in two patients with 15 mm cervical length. (A) Sample Patient 1 (B) Sample Patient 2.

Internal validation of the modified prediction models

The c-index (95%CI) for the 48-hour model and the 32-weeks model was 0.8 [0.70–0.81] and 0.85 [0.82–0.90], respectively. The miscalibration values for both models were −0.06 (p = 0.6) and −0.06 (p = 0.48), respectively and suggest a good calibration when tested in the internal validation cohort. For the 48-hour model and the 32-weeks model large scale calibration indices of 0.0002 (p = 0.99) and −0.003 (p = 0.98) were assessed, respectively, suggesting accurate estimation of the average event rate (number of preterm deliveries). The calibration plots illustrate how the prediction of the models compared with the actual outcomes observed in the validation set (Figs. 3 B and 4 B).

Comments

In this study two modified prediction models for individualized assessment of preterm birth risk were constructed and validated. The models allow accurate risk assessment and can be used for individualized patient counseling. An online calculator was provided to assess the risk of preterm birth within 48 h and before completed 32 weeks of gestation [13] .

Traditional methods for predicting women destined to deliver preterm rely upon single factors such as obstetrical history, demographic and clinical characteristics, or premonitory symptoms and tend to be inaccurate[5] and [25]. The Toulouse prediction models for women with preterm labor were recently published and provide information that can be used to help assess the individual and patient-specific likelihood of preterm birth within 48 h and before completed 32 weeks of gestation. The prediction models provide both the patient and the counseling obstetrician with accurate information to discuss expected outcome and management individually. Nevertheless, established clinical predictors of preterm birth were not incorporated into these models. Therefore we intended to modify the two previously published models by incorporating additional clinically relevant variables in order to improve the models’ performances and accuracy.

When comparing the baseline clinical characteristics most parameters were similar in both cohorts. The most significant difference was that women of the Austrian cohort were admitted at earlier gestational age and that the rate of women admitted at less than 24 weeks of gestation was higher. Nevertheless, when the data of the Austrian cohort was applied to the Toulouse models high concordance indices were observed. While the 32-week model was well calibrated, significant miscalibration was observed in the 48-h prediction, overestimating the average number of preterm deliveries. This might be explained by the lower preterm birth rate within 48 h in the Austrian cohort when compared with the cohort that was originally used for model construction.

A number of biologic markers in serum, amniotic fluid, and cervical secretions have been shown to be valuable to predict preterm delivery. Variables such as cervicovaginal fFN and markers of inflammation are widely used in clinical practice to detect presence of utero-placental infection and the likelihood of delivery in women with threatening preterm birth[7], [8], and [26]. The main modification between the previously published and the revised models are that CRP and fFN were incorporated into the novel prediction models. fFN was found to be associated with an increased likelihood of preterm delivery before 32 weeks (OR: 2.09). This is not surprising as fFN is a marker of integrity of the amniotic membranes and is accepted as one of the most powerful biochemical preterm birth predictors identified to date [27] . It is well established that fFN adds prognostic information in patients with preterm uterine contractions to that provided by sonographic cervical length, with a notably high predictive value and is therefore widely used in clinical routine[7], [28], [29], and [30]. Serum CRP is an acute phase protein and widely used to monitor inflammatory status. Signs of systemic infection as reflected by elevated CRP serum levels are commonly associated with presence of intrauterine infection, typically causing cytokine secretion, rising concentrations of prostaglandins, increased myometrial contractility, and preterm birth [27] . Previous studies showed that intrauterine infection might account for 25–40% of preterm births and that CRP is useful to predict the risk of preterm birth[5], [6], [26], and [27]. In our study cohort CRP was the strongest predictive parameter for delivery and was therefore incorporated into both modified models.

Maternal history of previous preterm birth is commonly reported to confer to increased risk in subsequent pregnancies[31] and [32]. Interestingly, previous preterm birth was not independently associated with increased risk of preterm birth and was therefore not included into our modified prediction models. This might be related to the stronger effect of other investigated variables.

Internal validation of the two modified models revealed considerably high concordance probabilities and good calibration suggesting accurate prediction of preterm birth. As shown inFig 3 and Fig 4, the probability of preterm birth within 48 h and before completed 32 weeks of gestation predicted by the model seems similar to the actual probability of preterm birth. We think that the parameters that were incorporated into the modified prediction models allow precise assessment of preterm birth risk. The models may be especially useful for institutions that routinely obtain fFN and serum CRP in women with threatening preterm birth. We think that the strong predictive effect of these parameters and their incorporation into the modified risk models seems biologically plausible and clinically meaningful. Nevertheless, future studies should test if the modified models are equally valid among different populations.

The usefulness of a nomogram is that it incorporates several risk factors and maps the predicted probabilities into points on a scale in a graphical interface. Moreover, because the probability outcomes in the nomogram are continuous, women can be given specific estimates of preterm birth risk that reflect the many different combinations of characteristics that different individuals may have. We provide the probability of preterm delivery in two sample patients with preterm labor based on the modified prediction tools that elucidate the heterogeneity that may exist within patients assigned to a specific risk group based on a single variable (e.g. cervical length). Fig. 5 shows individual preterm birth risk when nomogram variables are entered into the online risk calculator. Two patients with the same sonographic cervical length show distinctive differences in preterm birth risk based on multiple variables allowing more accurate prediction.

One of the strength of this study is that it was performed on a large cohort of women. As illustrated by the calibration curves, the predicted probabilities were similar to the observed probabilities both for external validation of previously published models but even more for the modified nomograms. The modified models may be especially useful for institutions that routinely obtain fFN and serum CRP in women with threatening preterm birth. Another strength is that the nomogram can be easily used when all parameters are available and we provide an online risk calculator. Although our results are confirmatory to previous reports, our study has potential limitations. Typically for data accumulated over several years, heterogeneity of clinical management is likely (2007–2012).

In conclusion, we present two modified prediction models to assess the risk of preterm birth within 48 h and before completed 32 weeks of gestations. The models were internally validated and we provide an online calculator. The clinical models allow accurate assessment of the actual risk of preterm delivery for women with threatened preterm delivery. Results can be used for patient counseling and might help to individualize patient care. Future studies should investigate the external validity of the presented models.

Condensation

Two prediction models to accurately assess the risk of preterm birth were constructed and validated. An online calculator can be used for individual risk assessment.

References

  • [1] M.C. McCormick. The contribution of low birth weight to infant mortality and childhood morbidity. N Engl J Med. 1985;312:82-90 Crossref
  • [2] H. Blencowe, S. Cousens, M.Z. Oestergaard, et al. National, regional, and worldwide estimates of preterm birth rates in the year 2010 with time trends since 1990 for selected countries: a systematic analysis and implications. Lancet. 2012;379:2162 Crossref
  • [3] R. Romero, J. Espinoza, J.P. Kusanovic, et al. The preterm parturition syndrome. BJOG. 2006;113(Suppl 3):17-42 Crossref
  • [4] J.D. Iams, R.B. Newman, E.A. Thom, et al. Frequency of uterine contractions and the risk of spontaneous preterm delivery. N Engl J Med. 2002;346:250-255 Crossref
  • [5] Lockwood. Predicting premature delivery—no easy task. N Engl J Med. 2002;346:282
  • [6] R.L. Goldenberg, J.F. Culhane, J.D. Iams, R. Romero. Epidemiology and causes of preterm birth. Lancet. 2008;371:75-84 Crossref
  • [7] R. Gomez, R. Romero, L. Medina, et al. Cervicovaginal fibronectin improves the prediction of preterm delivery based on sonographic cervical length in patients with preterm uterine contractions and intact membranes. Am J Obstet Gynecol. 2005;192:350-359 Crossref
  • [8] L.L. Klein, R.S. Gibbs. Use of microbial cultures and antibiotics in the prevention of infection-associated preterm birth. Am J Obstet Gynecol. 2004;190:1493 Crossref
  • [9] J.H. Harger, A.W. Hsing, R.E. Tuomala, et al. Risk factors for preterm premature rupture of fetal membranes: a multicenter case-control study. Am J Obstet Gynecol. 1990;163:130 Crossref
  • [10] M.S. Esplin, E. O’Brien, A. Fraser, et al. Estimating recurrence of spontaneous preterm delivery. Obstet Gynecol. 2008;112:516 Crossref
  • [11] A. Iasonos, D. Schrag, G.V. Raj, K.S. Panageas. How to build and interpret a nomogram for cancer prognosis. J Clin Oncol. 2008;26:1364-1370 Crossref
  • [12] S. Polterauer, Q. Zhou, C. Grimm, et al. External validation of a nomogram predicting overall survival of patients diagnosed with endometrial cancer. Gynecol Oncol. 2012;125:526-530 Crossref
  • [13] M. Allouche, C. Huissoud, B. Guyard-Boileau, R. Rouzier, O. Parant. Development and validation of nomograms for predicting preterm delivery. Am J Obstet Gynecol. 2011;204(242):1-8
  • [14] M.M. Costantine, K. Fox, B.D. Byers, et al. Validation of the prediction model for success of vaginal birth after cesarean delivery. Obstet Gynecol. 2009;114:1029-1033 Crossref
  • [15] W.A. Grobman, Y. Lai, M.B. Landon, et al. Development of a nomogram for prediction of vaginal birth after cesarean delivery. Obstet Gynecol. 2007;109:806-812 Crossref
  • [25] A. Sotiriadis, S. Papatheodorou, A. Kavvadias, G. Makrydimas. Transvaginal cervical length measurement for prediction of preterm birth in women with threatened preterm labor: a meta-analysis. Ultrasound Obstet Gynecol. 2010;35:54-64 Crossref
  • [26] K.T. Ozer, Z.N. Kavak, H. Gökaslan, K. Elter, T. Pekin. Predictive power of maternal serum and amniotic fluid CRP and PAPP-A concentrations at the time of genetic amniocentesis for the preterm delivery. Eur J Obstet Gynecol Reprod Biol. 2005;122:187-190 Crossref
  • [27] R.L. Goldenberg, J.C. Hauth, W.W. Andrews. Intrauterine infection and preterm delivery. N Engl J Med. 2000;342:1500-1507
  • [28] H. Honest, L.M. Bachmann, J.K. Gupta, J. Kleijnen, K.S. Khan. Accuracy of cervicovaginal fetal fibronectin test in predicting risk of spontaneous preterm birth: systematic review. BMJ. 2002;325:301 Crossref
  • [29] A.B. Boots, L. Sanchez-Ramos, D.M. Bowers, A.M. Kaunitz, J. Zamora, P. Schlattmann. The short-term prediction of preterm birth: a systematic review and diagnostic metaanalysis. Am J Obstet Gynecol. 2014;210:54
  • [30] G.J. van Baaren, J.Y. Vis, F.F. Wilms, et al. Predictive value of cervical length measurement and fibronectin testing in threatened preterm labor. Obstet Gynecol. 2014;123:1185-1192 Crossref
  • [31] C.Y. Spong. Prediction and prevention of recurrent spontaneous preterm birth. Obstet Gynecol. 2007;110:405-415 Crossref
  • [32] A.G. Care, A.N. Sharp, S. Lane, D. Roberts, L. Watkins, Z. Alfirevic. Predicting preterm birth in women with previous preterm birth and cervical length ≥25 mm. Ultrasound Obstet Gynecol. 2014;43:681-686

Footnotes

a Department of Obstetrics and Gynecology, Medical University Vienna, Vienna, Austria

b Karl Landsteiner Institute for General Gynecology and Experimental Gynecologic Oncology, Vienna, Austria

c Center for Medical Statistics, Informatics and Intelligent Systems, Medical University of Vienna, Vienna, Austria

lowast Corresponding author at: Department of Obstetrics and Gynecology, Medical University, Waehringer Guertel 18 20, A 1090 Vienna, Austria. Tel.: +43 1 40400 2962; fax: +43 1 40400 2911.