Study population

Using 2 French regional perinatal network databases (MATERMIP in the Midi-Pyrénées district and AURORE in the Rhône-Alpes district), we identified 906 women between 22 and 32 weeks transferred to a tertiary care center for threatened preterm delivery.

The first cohort included 737 women transferred to Paule de Viguier University Hospital of Toulouse (Hospital A) for threatened labor between Jan. 1, 2004, and Dec. 31, 2008. This cohort was used as a training set to develop predictive models. Data for each admission were collected prospectively by the MATERMIP Statistics’ Desk. Paule de Viguier Hospital is the only tertiary care obstetric unit in Midi-Pyrenees, and it is here that the regional perinatal network centralized all the information concerning in utero transfers in that district.

The second cohort included 169 women transferred to Croix-Rousse University Hospital of Lyon (Hospital B) for preterm delivery through the AURORE perinatal network during the same period. Data from these patients were used for validation of the prediction models developed from the first cohort.

In utero transfer was defined as transfer of a pregnant woman from 1 maternity care center to another by ambulance, helicopter, hospital car service, or personal vehicle. Standard care at the study hospital for patients with threatened labor was administered to all patients. This care involved administration of betamethasone as antenatal corticosteroid therapy, antibiotics when preterm premature rupture of membranes (PPROM) was associated with threatened preterm delivery, and tocolytic treatments, depending on uterine activity (calcium channel blockers, β-mimetics or atosiban).

In addition, the level of perinatal care was defined according to the classifications of French regulations on the safety of childbirth (dated Oct. 9, 1998). Level 1 units have no neonatal ward and are not required to have a pediatrician on site. Level 2 units have neonatology facilities to manage infants born at 32 weeks or later (special care nurseries). In these units, a pediatrician must be present during the day and either on call or present on nights and weekends. Level 3 units (tertiary care centers) have an on-site neonatal intensive care unit and at least 1 neonatologist must be present 24 hours a day, 7 days a week.

The diagnosis of threatened preterm delivery was based on clinical evidence of painful uterine contractions and cervical dilatation. Contractions were defined as deflections from a clear baseline with a rounded peak lasting 40 to 120 seconds. 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. If there was a discrepancy of more than 7 days, sonographic gestational age was used.

Women whose pregnancies were complicated by PPROM without uterine activity or cervical changes, preeclampsia, fetal growth restriction, fetofetal transfusion syndrome, in utero fetal death, or major fetal anomaly were excluded. PPROM was defined as leaking of amniotic fluid before the onset of labor. Diagnosis was made by direct visualization of amniotic fluid on speculum examination, nitrazine tests, or rapid identification of insulin-like growth factor-binding protein-1 in cervicovaginal secretions.

Sonographic scans of the uterine cervix were performed with a 4-8 MHz transducer after the patient had emptied her bladder. The internal orifice was identified in the sagittal plane, and the probe was moved until the entire cervical canal was visualized. The cervical length was measured as the distance between the internal and the external orifices identified by sonolucency of the cervical canal. All the sonographic measurements were performed at admission by experienced investigators.

Also, for each woman, the following information was recorded: individual demographic and obstetric data (age, parity, obstetric history, and characteristics of this pregnancy), clinical characteristics at enrollment (vaginal bleeding, uterine contractions requiring tocolysis, and/or PPROM), and sonographic measurement of cervical length. The main obstetrical outcomes were delivery within 48 hours after admission and delivery before 32 weeks. The study was approved on Sept. 4, 2006 (protocol no. 117630) by the institutional review board.

Statistical analysis

To develop well-calibrated and exportable nomograms, we built each model in a training cohort and validated it in an independent validation cohort. Univariate and multivariate logistic regression analyses were used to test the association between routine clinical variables, ultrasonographic length of the uterine cervix, and occurrence of delivery within 48 hours after in utero transfer or before 32 weeks. Clinical variables tested included age, parity, history of previous preterm birth or late miscarriage, multiple pregnancy, gestational age at admission, PPROM, vaginal bleeding, and uterine contractions requiring tocolysis (presence of documented regular uterine contractions of 6 or more per hour).

Backward variable selection was performed to determine independent covariates. Continuous variables were fit using restricted cubic splines to relax the linearity assumptions if necessary. The model performance was quantified with respect to discrimination and calibration. Discrimination (ie, whether the relative ranking of individual predictions is in the correct order) was quantified with the area under the receiver operating characteristic curve or with the concordance index, which is similar to the area under the receiver operating characteristic curve (AUC) but appropriate for censored data.

The concordance index is the probability that, given 2 randomly selected patients, the patient with the worse outcome will in fact have a worse outcome prediction. The concordance index ranges from 0 to 1, with 1 indicating perfect concordance, 0.5 indicating no better concordance than chance, and 0 indicating perfect discordance. We used the bootstrapping method (200 repetitions) to obtain relatively unbiased estimates.

Calibration (ie, agreement between observed outcome frequencies and predicted probabilities) was studied with graphical representations of the relationship between the observed outcome frequencies and the predicted probabilities (calibration curves). These curves represented grouped proportions vs mean predicted probability in groups defined by quantiles. All analyses were performed using the R package with the Survival, Design, Hmisc, Rpart, and Lexis libraries ( http://lib.stat.cmu.edu/R/CRAN/ ).