Early prediction of preterm birth for singleton, twin,
and triplet pregnancies
Hongzhuan Tan a,b,c , Shi Wu Wen a,b,c,d, * , Xi Kuan Chen b,c , Kitaw Demissie e , Mark Walker b,c a School of Public Health, Central South University, Changsha, Hunan, PR China b OMNI Research Group, Department of Obstetrics & Gynecology, University of Ottawa, Faculty of Medicine, 501 Smyth Rd, Ottawa, Box 241, Canada K1H 8L6 c Clinical Epidemiology Program, Ottawa Health Research Institute, 501 Smyth Rd, Ottawa, Canada K1H 8L6 d Department of Epidemiology and Community Medicine, University of Ottawa, Faculty of Medicine, 501 Smyth Rd, Ottawa, Canada K1H 8L6 e Division of Epidemiology, School of Public Health, University of Medicine and Dentistry of New Jersey, Piscataway, NJ, USA Received 7 April 2005; received in revised form 22 April 2006; accepted 30 April 2006 Abstract Objectives: To create prediction models of early preterm birth for singletons, twin, and triplet pregnancies. Study design: We used a historical cohort study with the 1996 birth registration data for singletons and the 19951997 linked birth/infant death dataset for multiple births of the United States. Preterm birth was dened as gestational age <32 completed weeks. Eligible study subjects were randomly allocated to two groups: one group (80%subjects) for the creation of the prediction models, and the other group (20% subjects) for the validation of the established prediction models. Multivariate logistic regressions were used to establish the prediction models. We further assessed the sensitivity, specicity, positive predictive value (PPV) and negative predictive value (NPV) of the established prediction models with different cut-off values in the validation group. Results: The sensitivity, specicity, PPV, and NPVof the established model were 24.58, 93.54, 5.91, and 98.69%, respectively for singletons, 64.66, 57.04, 16.29, and 92.59%, respectively for twins, and 63.57, 53.58, 42.96, and 72.78%, respectively for triplets. Conclusion: The prediction models of early preterm birth for singleton, twin, and triplet pregnancies created by this study could be useful for obstetricians to identify women being at high risk of preterm birth at early gestation. # 2006 Elsevier Ireland Ltd. All rights reserved. Keywords: Preterm birth; Prediction; Singleton; Twins; Triplets 1. Introduction Preterm birth is the most important cause of perinatal mortality and morbidity in industrialized countries; 6080% of deaths of infants without congenital anomalies are related to preterm birth [1,2]. Preterm birth is also associated with cerebral palsy and other long-term health sequelae [3]. Identifying women at high risk of preterm birth is helpful in obstetric care. It could help in guiding treatment with antenatal steroids on appropriate patients in a timely fashion, and in directing tertiary care referrals. Previous studies have used various strategies to predict preterm birth, including risk scoring systems [4,5], biochemical markers such as fetal bronectin [6], salivary estriol [7], transvaginal sonography [8], vaginal infections [9], clinical characteristics (bleeding, substance abuse, white blood cell count 14,000 cells/ml, and frequency of uterine contractions) [1011], and data mining methods [12] in singletons and/or twins. The sensitivity of these methods was only about 40%, and the positive predictive value (PPV) was usually lower than 30%[4,11]. Although cervicovaginal fetal bronectin test has a better prediction of preterm birth www.elsevier.com/locate/ejogrb European Journal of Obstetrics & Gynecology and Reproductive Biology 131 (2007) 132137 * Corresponding author. Tel.: +1 613 737 8899x73912; fax: +1 613 739 6266. E-mail address: swwen@ohri.ca (S.W. Wen). 0301-2115/$ see front matter # 2006 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.ejogrb.2006.04.038 than other methods, it commonly predicts the preterm birth within 710 days before advanced cervical dilatation [6], leaving little time for decision-making in terms for prenatal treatment. Crane et al. [13] and Goldenberg et al. [14] found that the ability to predict preterm birth will be improved by combining several lab tests, but this strategy will add to the complexity of the diagnosis. Furthermore, preterm predic- tion models with detailed clinical and laboratory data usually conducted in large centers, may not be applicable in the general population. On the other hand, preterm prediction models based on vital statistics and/or hospital discharge data, although they may have a lower PPV, have a higher statistical stability and a better generalizability. These models can serve as a preliminary screening tool, so that the more complicated and expensive laboratory tests can be applied to high-risk patients with improved effectiveness and efciency. The rate of preterm birth is much higher in multiple pregnancies than in singletons [15], and the multiple birth rates have been increasing in the last two decades, resulting in the increasing rate of preterm birth [1,16,17]. The risk factors for preterm birth are also somewhat different in multi-fetal pregnancies from that of in singletons. No previous study on preterm prediction has been conducted for prediction model in singleton, twin and triplet pregnancies simultaneously. Moreover, no previous study has used separate sample to validate the predict model. The objective of this study was to establish effective prediction models of spontaneous preterm birth for singleton, twin, and triplet pregnancies, based on a large population birth registration data from the United States. 2. Materials and methods This study was based on the birth registration data of the United States, compiled by the National Center for Health Statistics, Centers for Disease Control and Prevention. The birth registration data were based on live births and infant deaths up to 1 year, registered in the 50 states and the District of Columbia. The data were coded according to uniform specications, have gone through statistical quality checks, have been carefully edited by the National Center for Health Statistics, and formthe basis for the ofcial birth Statistics of the United States [18]. For singletons, the 1996 linked birth/infant death dataset of the United States, excluding multiple births, were used. For twins and triplets, the 19951997 multiple birth registration of the United States were used. The 1995 1997 multiple births were matched by plurality, state and county of occurrence of delivery, mothers date of birth, date of last menstrual period (LMP), number of prenatal visits, level of education, weight gain during pregnancy, and date of delivery. The matching was successful for 98% of the multiple sets [19]. Available study variables in these databases included socio-demographic information of the parents such as parents age, race, maternal education, and maternal life-style factors such as smoking and alcohol consumption during pregnancy, obstetric history, complica- tions of the pregnancy, labor and delivery complications, and birth weight, gestational age, and other infant outcome variables. The states of California, New York (except New York city), Indiana and South Dakota did not send data on maternal smoking. Early pretermbirth was dened as less than 32 completed weeks of gestation. We elected to use the more restrictive denition because the risk of neonatal mortality and morbidity is much higher in the early preterm birth (20 31 weeks of gestation) than later (3236 weeks of gestation) preterm birth, and the need for prenatal intervention in the early preterm births is also much higher [18]. Gestational age was estimated by the interval between the rst day of LMP and the date of delivery. If the date of LMP was not recorded, or if the calculated gestation weeks fell beyond the duration considered biologically plausible, gestational age estimated by the physician was used. Stillbirths and those births with no information of gestational age were also excluded. The subjects without tobacco use information were excluded because the association of smoking with preterm is denite. Pregnancy was the unit of analysis for this study. Eligible study subjects were randomly allocated to two groups: one group (80% subjects) for the creation of the prediction models, and the other group (20% subjects) for the validation of the established prediction models. We rst described the maternal characteristics and the rate of preterm birth in singletons, twins and triplets. We then created the prediction models by stepwise multivariate logistic regression analysis for group 1 subjects. The dependent variable (Y) was dened as 0 when the infants gestational age 32 weeks or 1 when the infants gestational age less than 32 weeks. The prediction variables included in the initial regression model were maternal age X 1 (X 1-1 : under 20, X 1-2 : 2024, X 1-3 : 2529, 3034 as the referent group, X 1-4 : 35 and over), maternal race X 2 (0: non-black, 1: black), maternal education X 3 (1: under 12 years, 2: 12 years, 3: 1315 years, 4: 16 years and over), marital status X 4 (0: married, 1: unmarried), parity X 5 (0: primiparous, 1: multiparous), prenatal care visit initiation X 6 (13 months as the referent group, X 6-1 : 46 months, X 6-2 : 7 months and over or no visit), maternal cigarette smoking during pregnancy X 7 (0: no, 1: yes), maternal weight gain per week (pounds) X 8 (X 8-1 : under 0.83 (<mean 2 S.D.), X 8- 2 : 0.831.12 (<mean S.D.), 1.131.70 (mean S.D.) as the referent group, X 8-3 : 1.70 and over (>mean + S.D.)), and maternal medical risk factors X 9 (0: no, 1: with a diagnosis of any of the following conditions: anemia, cardiac disease, lung disease, diabetes, genital herpes, hydramnios/oligohy- dramnios, hemoglobinopathy, chronic hypertension, preg- nancy-induced hypertension, eclampsia, incompetent cervix, previous infant more than 4000 g, previous preterm or small-for-gestational-age infant, renal disease, RH H. Tan et al. / European Journal of Obstetrics & Gynecology and Reproductive Biology 131 (2007) 132137 133 sensitization, uterine bleeding, or other medical risk factors). We conducted a preliminary analysis, and found that maternal education level showed a linear relationship with preterm birth, so we entered it as a ranked variable. Other variables, such as maternal age, prenatal care visit initiation, and maternal weight gain, were entered into the models as dummy variables; the category with lowest risk of preterm was dened as reference category. The areas under the receiver operating characteristic curve (ROC curve) generated by logistic regressions model were applied to evaluate the effectiveness of the prediction models. To assess the impact of medical intervention on the observed association, we repeated the analysis after excluding those mothers who had an induction of labor or who had a cesarean delivery with a diagnosis of cardiac disease, lung disease, diabetes, hydramnios, hemoglobino- pathy, chronic hypertension, pregnancy-induced hyperten- sion, eclampsia, renal disease, RH sensitization, placenta previa, or abruption placenta. We further assessed the sensitivity, specicity, PPV and NPVof the established prediction models with different cut- off values generated from ROC curves in the validation group. (The cut-off value is a criterion used to screen the women at high risk of preterm birth or not. If the value calculated from the prediction model with the womens information was equal or higher than the cut-off value, these women were considered as being at high risk of preterm birth; if lower than the cut-off value, these women were considered as being at no risk of preterm birth.) All data analysis was applied in SAS 8.02. 3. Results In the study of singletons, subjects with missing information on gestation age (40,687) or tobacco use (810,849) were excluded, leaving 2,988,844 subjects for analysis. Among them, 80% subjects (2,391,480) were allocated to the prediction group, and 20% of them (597,364) were allocated to the validation group. In the study of twins and triplets, unmatched sets (3850), fetal death (6954), and subjects with no information on tobacco use (657,879) or gestation age (3467) were excluded, leaving 104,536 twin sets and 3868 triplet sets for analysis. Among the twins, 83,673 sets were allocated to the prediction group and 20,863 to the validation group. Among the triplets, 3110 were allocated to the prediction group and 758 to the validation group. Table 1 describes the rate of preterm birth by characteristics of the study subjects in singletons, twins, and triplets. The rate of preterm birth increased substantially with increased plurality. There were also major differences in the rate of preterm birth by maternal characteristics among singletons, twins, and triplets. In particular, the teenagers, black race, lower maternal education, unmarried, prenatal care initiation later, smoking, weight gain too slow or too quick, increased the rate of pretermbirth in singletons, twins and (especially) in triplets (Table 1). Table 2 shows the prediction models for preterm. The areas under ROC curve were 0.73, 0.65 and 0.65, respectively, in singleton, twin, and triplet pregnancies. The predictors of preterm in singletons were young or old maternal age, black race, and unmarried, primiparous, lower education level, later initiation of prenatal care, cigarette smoking, low or high gestational weight gain, and medical complications. The predictors of early preterm for twins and triplets were similar with predictors for singletons (Table 2). Tables 35 describe the validity of the prediction models under different cut-off values. The sensitivity, specicity, PPV, and NPV of the prediction model were 24.58, 93.54, 5.91, and 98.69%, respectively, at the cut-off value of 0.04 for singleton pregnancies. Corresponding gures were 64.66, 57.04, 16.29, and 92.59% at the cut-off value of H. Tan et al. / European Journal of Obstetrics & Gynecology and Reproductive Biology 131 (2007) 132137 134 Table 1 The rate of preterm birth by characteristics in singletons, twins, and triplets Variable Singletons Twins Triplets N * % N % N % Maternal age (X 1 ) Under 20 401067 2.57 7961 18.82 63 49.21 2024 740378 1.71 20624 13.93 266 45.49 2529 823841 1.34 28889 11.37 964 41.49 3034 676407 1.32 29734 9.75 1559 32.01 35 and over 347151 1.73 17328 9.18 1016 29.23 Maternal race (X 2 ) Non-black 2480292 1.21 85126 10.17 3557 33.43 Black 508552 3.73 19410 17.98 311 51.13 Maternal education (X 3 ) Under 12 years 661012 2.38 16521 15.27 204 40.20 12 years 1007321 1.77 33185 12.33 857 38.39 1315 years 657666 1.42 24463 11.13 958 36.64 16 years and over 662845 0.91 30367 9.24 1849 31.69 Marital status (X 4 ) Married 2006217 1.12 75877 9.91 3524 34.14 Unmarried 982627 2.70 28659 16.13 344 42.15 Parity (X 5 ) Primiparous 1254081 1.80 43571 14.13 2355 38.77 Multiparous 1734763 1.52 60965 9.82 1513 28.75 Prenatal care visit initiation (X 6 , month) 13 2391424 1.37 87911 11.05 3507 34.07 46 409662 2.13 11424 12.08 184 42.39 7 and over or no care 187758 3.92 5201 20.27 177 42.37 Maternal smoking (X 7 ) No 2580777 1.54 92067 11.37 3717 34.68 Yes 408067 2.26 12469 13.45 151 39.07 Maternal weight gain per week (pounds X 8 ) Under 0.83 167777 1.53 30469 14.91 675 46.52 0.831.12 749918 1.16 29593 9.58 729 36.90 1.131.70 316960 1.72 30479 8.11 1435 29.34 1.71 and over 254189 3.61 13995 16.64 1029 33.43 Medical complication (X 9 ) No 2226912 1.14 82741 11.24 2845 35.61 Yes 761932 3.07 21795 13.05 1023 32.75 * N = number of subjects. H. Tan et al. / European Journal of Obstetrics & Gynecology and Reproductive Biology 131 (2007) 132137 135 Table 2 The association between preterm birth and predicting variables included in the logistic predicting model Variables Singleton Twin Triplet b a OR b 95% CI c b OR 95% CI b OR 95% CI Maternal age (X 1-1 ) 0.3341 1.40 1.351.45 0.2126 1.23 1.141.33 0.3999 1.49 1.062.08 Maternal age (X 1-2 ) 0.0457 1.05 1.011.08 0.1358 1.15 1.081.22 0.3273 1.39 1.011.92 Maternal age (X 1-3 ) 0.0405 1.04 1.011.07 Maternal age (X 1-4 ) 0.1180 1.13 1.091.16 0.0734 0.93 0.870.99 Maternal race (X 2 ) 0.8802 2.41 2.362.47 0.5058 1.66 1.571.75 0.6358 1.89 1.402.55 Maternal education (X 3 ) 0.1241 0.88 0.870.89 0.0639 0.94 0.920.96 0.1056 0.90 0.820.99 Marital status (X 4 ) 0.3287 1.39 1.351.43 0.1508 1.16 1.101.23 Parity (X 5 ) 0.2420 0.79 0.770.80 0.4969 0.61 0.580.64 0.5509 0.58 0.490.68 Prenatal care visit initiation (X 6-1 ) 0.1335 1.14 1.111.18 0.1576 1.18 1.091.25 Prenatal care visit initiation (X 6-2 ) 0.5152 1.67 1.621.73 0.4526 1.57 1.451.71 Maternal cigarette smoking (X 7 ) 0.2676 1.31 1.271.34 0.1985 1.22 1.141.30 Maternal weight gain per week (X 8-1 ) 0.2221 0.80 0.780.82 0.4605 1.59 1.501.68 0.3471 1.42 1.111.81 Maternal weight gain per week (X 8-2 ) 0.2645 0.77 0.740.80 0.2122 1.23 1.161.32 0.3213 1.37 1.111.72 Maternal weight gain per week (X 8-3 ) 0.8810 2.41 2.332.50 0.4720 1.60 1.501.72 0.1736 1.19 1.061.49 Maternal medical complication (X 9 ) 0.9111 2.49 2.442.54 0.1613 1.18 1.121.24 a b: coefcient of regression indicating the magnitude of change dependent variable (preterm) with corresponding change in independent variable. b OR: odds ratio. c CI: condence interval. Table 3 The validity of predicting model under different cut-off value in different population-in singletons (%) Cut-off value Group 1 Group 2 Total SE a SP b PPV c NPV d SE SP PPV NPV SE SP PPV NPV 0.01 83.17 47.44 2.57 99.41 92.80 47.51 2.54 99.41 83.10 47.46 2.56 99.41 0.02 55.95 76.80 3.86 99.05 55.81 76.87 3.83 99.06 55.92 76.81 3.86 99.05 0.03 36.00 88.95 5.15 98.82 35.96 88.96 5.10 98.83 35.99 88.95 5.14 98.82 0.04 25.19 93.52 6.08 98.68 24.58 93.54 5.91 98.69 25.07 93.52 6.05 98.69 0.05 18.85 95.65 6.73 98.61 18.07 95.62 6.38 98.61 18.69 95.64 6.66 98.61 0.06 13.88 97.12 7.43 98.54 12.99 97.09 6.87 98.54 13.70 97.11 7.32 98.54 0.07 10.04 98.11 8.14 98.50 9.34 98.10 7.50 98.50 9.90 98.11 8.01 98.50 0.08 6.87 98.86 9.12 98.45 6.31 98.86 8.35 98.46 6.76 98.86 8.97 98.46 0.09 5.24 99.20 9.79 98.43 4.67 99.20 8.97 98.44 5.12 99.20 9.59 98.43 0.10 3.83 99.44 10.17 98.41 3.43 99.45 9.30 98.42 3.75 99.44 10.00 98.42 a SE: sensitivity. b SP: specicity. c PPV: positive predictive value. d NPV: negative predictive value. Table 4 The validity of predicting model under different cut-off value in different population in twins (%) Cut-off value Group 1 Group 2 Total SE a SP b PPV c NPV d SE SP PPV NPV SE SP PPV NPV 0.04 99.56 1.19 11.74 95.36 99.58 1.27 11.53 95.90 99.56 1.21 11.70 95.47 0.06 92.70 18.76 13.09 95.12 93.09 18.86 12.91 95.48 92.78 18.78 13.06 95.19 0.08 82.52 35.05 14.36 93.82 83.00 35.43 14.25 94.16 82.61 35.12 14.34 93.89 0.10 65.62 56.42 16.58 92.56 64.66 57.04 16.29 92.59 65.43 56.55 16.52 92.56 0.12 53.72 68.13 18.20 91.77 52.01 68.48 17.58 91.69 53.38 68.20 18.08 91.76 0.14 42.08 77.41 19.74 91.01 39.57 77.86 18.76 90.88 41.59 77.50 19.55 90.99 0.16 30.69 84.82 21.06 90.26 27.97 85.08 19.50 90.14 30.15 84.87 20.76 90.24 0.18 22.38 89.79 22.45 89.76 20.64 90.14 21.30 89.78 22.04 89.86 22.23 89.76 0.20 15.56 93.40 23.73 89.34 14.11 93.73 22.53 89.17 15.27 93.46 23.50 89.35 0.22 11.20 95.57 25.01 89.07 9.88 95.89 23.69 89.17 10.94 95.63 24.76 89.09 a SE: sensitivity. b SP: specicity. c PPV: positive predictive value. d NPV: negative predictive value. 0.10 for twins, and 63.57, 53.58, 42.96, and 72.78% at the cut-off value of 0.34 for triplets, respectively. Supplement analysis excluding subjects with medical intervention yielded similar results (data available upon request). 4. Discussion Based on large population databases, we have created prediction models of early preterm for singleton, twin, and triplet pregnancies, using common demographic variables such as maternal age, race, education, marital status, parity, prenatal care initiations, and cigarette smoking. Excluding those subjects whose pregnancy was probably terminated by medical intervention did not change the results. The risk of preterm birth was different in singletons, twins, and triplets. As a result, we determined the cut-off values from ROC curves from validation groups for singletons, twins, and triplets separately. The results showed that the prediction models were more effective for twins and triplets, because the PPVand NPV were higher in these two models than that in the singleton model. However, probably owing to the small sample, fewer predictors entered into the nal model for triplets. Previous preterm prediction models have used various strategies, focusing on ultrasound, biochemical markers, vaginal infection, and clinical characteristics. Hassan et al. [20] reported that ultrasonographic cervical length can predict preterm birth with the sensitivity of 8.214.7%, specicity of 99.798.8%, and PPV of 47.631.6%. Cervicovaginal fetal bronectin test can predict spontaneous preterm birth with the sensitivity of 43% and specicity of 80% [21]. Iams et al. [11] found that uterine contraction frequency had poor ability to predict the spontaneous preterm birth, with a sensitivity of 9% and PPV of 25%. These prediction methods were complicated and had poor diagnostic performance. Wildschut et al. [22] used socio- demographic factors to predict preterm, and achieved sensitivity of 3.7 and 17.4%, and specicity of 98.6 and 95.0%, respectively, for primiparous and multiparous. All these models have focused on singletons and/or twins and have not been validated by a separate sample. Our large population-based study established statistically stable preterm prediction models for singletons, twins and triplets, using common demographic and life-style variables. We have also examined the impact of different cut-off values on sensitivity, specicity, and PPV. Clinicians and public health workers could conveniently take different cut-off values in their preterm screening program in order to obtain their optional results. Our preterm prediction models were further validated with a separate sample, with the ability to assess their true diagnostic performance when they are being used in other populations. The variables included in our model are common and are often available in obstetric routine practice, so these prediction models could nd wide application for clinicians and public health workers to identify high-risk population for the management of preterm birth with different plurality. For example, if there is a woman with twin pregnancy, who is 31-year-old (X 1 = 0), black (X 2 = 1), has 12 years education (X 3 = 2), is married (X 4 = 0), has no preterm birth history (X 5 = 0), has prenatal care visit initiated in the fth month (X 6-1 = 1), is smoking (X 7 = 1), and gains 1.2 pounds/week (X 8 = 0), we can calculate the probability of preterm birth for this woman as 0.2348 (P = 1/ (1 + exp((1.9154 + 0 + 0.5058 1 0.0639 2 + 0 0 + 0.1576 1 + 0.1985 1 + 0))) = 0.2348), with the pre- term birth prediction model for twins. Although the sensitivity and PPV in our preterm prediction models were lower than those based on detailed clinical and lab testing information [13], it can be used to identify women at increased risk of preterm birth at early gestation, so that the more complicated and costly lab tests or some possible preterm birth preventive measures can be preserved for the most needed patients in a timely fashion. H. Tan et al. / European Journal of Obstetrics & Gynecology and Reproductive Biology 131 (2007) 132137 136 Table 5 The validity of predicting model under different cut-off value in different population-in triplets (%) Cut-off value Group 1 Group 2 Total SE a SP b PPV c NPV d SE SP PPV NPV SE SP PPV NPV 0.18 99.54 2.41 35.14 90.74 99.63 2.04 35.88 90.91 99.50 2.34 35.29 90.77 0.22 92.96 15.26 36.82 80.31 94.05 14.52 37.70 81.61 93.18 15.12 37.00 80.55 0.26 85.91 27.18 38.53 78.41 87.36 25.56 39.23 78.62 86.20 26.87 38.67 78.45 0.30 72.57 45.15 41.28 75.60 76.21 41.51 41.75 76.03 73.29 44.44 41.37 75.68 0.34 60.70 59.28 44.20 73.96 63.57 53.58 42.96 72.78 61.28 58.17 43.94 73.74 0.38 43.65 75.87 49.01 71.71 46.47 73.01 48.64 71.26 44.21 75.32 48.93 71.62 0.42 31.14 84.44 51.53 69.77 36.06 81.80 52.15 69.93 32.12 8393 51.67 69.80 0.46 19.83 90.50 52.58 68.00 22.30 88.55 51.72 67.45 20.33 90.12 52.39 67.89 0.50 14.09 93.65 54.09 67.23 15.61 92.84 54.55 66.67 14.39 93.49 54.19 67.12 0.54 8.25 96.16 53.29 66.36 8.55 94.48 46.00 65.25 8.31 95.83 51.61 66.15 a SE: sensitivity. b SP: specicity. c PPV: positive predictive value. d NPV: negative predictive value. Acknowledgments Dr. Tan and Dr. Chen are International Fellows of the University of Ottawa. Dr. Wen is a CIHR New Investigator. Dr. Walker is a Career Scientist of the Ontario Ministry of Health. We thank Joyce Martin for her support in data acquisition and management. References [1] Goldenberg RL. The management of preterm labor. Obstet Gynecol 2002;100:102037. [2] Mesleh RA, Kurdi AM, Sabagh TO, Algwiser AA. Changing trends in perinatal deaths at the Armed Forces hospital, Riyadh, Saudi Arabia. J Obstet Gynecol 2001;21:4955. [3] Hack M, Fanaroff AA. Outcomes of children of extremely low birthweight and gestational age in the 1990s. Early Hum Dev 1999; 53:193218. [4] Holbrook Jr RH, Laros Jr RK, Creasy RK. Evaluation of a risk-scoring system for prediction of preterm labor. Am J Perinatol 1989;6:628. [5] Owen J, Goldenberg RL, Davis RO, Kirk KA, Copper RL. Evaluation of a risk scoring system as a predictor of preterm birth in an indigent population. Am J Obstet Gynecol 1990;163:8739. [6] Honest H, Bachmann LM, Gupta JK, Kleijnen J, Khan KS. Accuracy of cervicovaginal fetal bronectin test in predicting risk of sponta- neous preterm birth: systematic review. BMJ 2002;325:301. [7] Mauldin JG, Newman RB. Preterm birth risk assessment. Semin Perinatol 2001;25:21522. [8] Honest H, Bachmann LM, Coomarasany A, Gupta JK, Kleijnen J, Khan KS. Accuracy of cervical transvaginal sonography in predicting preterm birth: a systematic review. Ultrasound Obstet Gynecol 2003;22:30522. [9] Meis PJ, Goldenberg RL, Mercer B, et al. The preterm prediction study: signicance of vaginal infections. National Institute of Child Health and Human Development MaternalFetal Medicine Units Network. Am J Obstet Gynecol 1995;173:12315. [10] Macones GA, Segel SY, Stamilio DM, Morgan MA. Prediction of delivery among women with early preterm labor by means of clinical characteristics alone. Am J Obstet Gynecol 1999;181: 14148. [11] Iams JD, Newman RB, Thom EA, et al. Frequency of uterine con- tractions and the risk of spontaneous preterm delivery. N Engl J Med 2002;346:2505. [12] Goodwin LK, Iannacchione MA, Hammond WE, Crockett P, Maher S, Schlitz K. Data mining methods nd demographic predictors of preterm birth. Nurs Res 2001;50:3405. [13] Crane JM, Armson BA, Dodds L, Feinberg RF, Kennedy W, Kirkland SA. Risk scoring, fetal bronection, and bacterial vaginosis to predict preterm delivery. Obstet Gynecol 1999;93:51722. [14] Goldenberg RL, Iams JD, Mercer BM, et al. What we have learned about the predictors of preterm birth. Semin Perinatol 2003;27:185 93. [15] Rydhstroem H, Heraib F. Gestational duration, and fetal and infant mortality for twins vs. singletons. Twin Res 2001;4:22731. [16] MacDorman MF, Minino AM, Strobino DM, Guyer B. Annual summary of vital statistics2001. Pediatrics 2002;110:103752. [17] Alexander GR, Slay M. Prematurity at birth: trends, racial dispa- rities, and epidemiology. Ment Retard Dev Disabil Res Rev 2002;8: 21520. [18] National Center for Health Statistics. 19951997 Matched multiple birth data set. Hyattsville, Maryland: U.S. Department of Health and Human Services, Centers for Disease Control and Prevention; 2000 [NCHS CD-ROM Series 21, No. 12]. [19] Ekwo EE, Moawad A. Maternal age and preterm births in black population. Paediatr Perinat Epidmiol 2000;14(2):14551. [20] Hassan SS, Romero R, Berry SM, et al. Patients with an ultrasono- graphic cervical length < or =15 mm have nearly a 50% risk of early spontaneous preterm delivery. Am J Obstet Gynecol 2000;182: 145867. [21] Revah A, Hannah ME, Sue-A-Quan AK. Fetal bronectin as a predictor of preterm birth: an overview. Am J Perinatol 1998;15: 61321. [22] Wildschut HI, Nas T, Golding J. Are sociodemographic factors predictive of preterm birth? A reappraisal of the 1958 British Perinatal Mortality Survey. Br J Obstet Gynaecol 1997;104:5763. H. Tan et al. / European Journal of Obstetrics & Gynecology and Reproductive Biology 131 (2007) 132137 137