This retrospective observational study included women who had a singleton pregnancy, had received routine prenatal care during their first, second and third trimesters and had delivered a baby after 32 weeks at Guangzhou Women and Children’s Medical Center between January 2013 and December 2016. Another retrospective cohort including 565 pregnant women who delivery a baby between 2017 and 2018 was used for external validation. The gestational age (GA) was calculated by a combination of crown-rump length (CRL) and last menstrual period (LMP). Study participants were excluded due to the following patient variables: the presence of structural defects that were suspected at the time of routine scans and/or confirmed postnatally; termination of pregnancy, intrauterine death and stillbirth before 32 weeks; no knowledge of the first day of the LMP or the lack of a regular menstrual cycle duration of 28 days plus or minus 4 days. During the initial database search, 2478 women met the inclusion criteria. Of these women, 220 met either the exclusion criteria or had incomplete data for recorded factors, leaving 2258 women for the analysis. The Medical Ethics Committee of Guangzhou Women and Children’s Medical Center approved this study.
Maternal characteristics and clinical variables
Data on maternal characteristics were collected retrospectively from the medical records and included maternal age, pregestational maternal height and weight, nulliparity (no previous deliveries after 24 weeks gestation), conception method (spontaneous or via an assisted reproductive technique), maternal medical history (chronic hypertension, diabetes mellitus, renal disease, autoimmune disease or coagulation disorders), and obstetrical history (e.g., previous stillbirth, miscarriage or fetal anomaly). Within the study, the main clinical variable was maternal blood pressure (BP), which was measured at the time of the first trimester ultrasound (11+0 to 13+6 weeks of gestation) with an automatic blood pressure monitor (OMRON HBP-9020, Kyoto, Japan). BP was measured with the woman comfortably seated after a 5-minute rest. The mean arterial pressure (MAP) was calculated as diastolic BP + (systolic BP-diastolic BP)/3.
Maternal blood biomarkers
Maternal serum free-β-human chorionic gonadotropin (HCG) and pregnancy-associated plasma protein A (PAPP-A) levels were measured at the time of the nuchal translucency scan (11+0 to 13+6 weeks). The measured concentrations of the two hormones were converted to the multiple of the maternal-weight-adjusted gestation-specific median for the local Chinese population, followed by log10 transformation (log10 PAPP-A MoM and log10 fβ-hCGMoM, respectively(15). These concentrations were measured using a time-resolved 1234 Delfia® (Wallac, Turku, Finland).
Transabdominal ultrasound with Doppler evaluation was performed during pregnancy using a Voluson Expert E8 (GE Healthcare), using curvilinear 2.0 to 5.0 MHz transducers. CRL was measured in a true mid-sagittal plane with the genital tubercle and the fetal spine longitudinally in view, in the GA range of 11+0 weeks to 13+6 weeks. In the second trimester (24-28 weeks), the biparietal diameter, head circumference, abdominal circumference and femur length were measured. The second-trimester estimated fetal weight (EFW) was calculated based on the Hadlock formula(16). The umbilical artery-pulsatility index (UA-PI) was calculated from a free-floating portion of the umbilical cord. The middle cerebral artery (MCA) was measured at the axial view of the fetal head, in the inner one-third of its course to the circle of Willis. If some patients had 2 or more measurements of UA-PI or MCA-PI, the mean values would be used. Usually, if the screening for fetal anomalies at 22–26 weeks’ gestation is normal, and fetal growth and fundal height increase in line with gestational age. The next ultrasound examination will be scheduled at 30-32 weeks’ gestation, and then the next four weeks (34-36 weeks), then 38-40 weeks.
Fetal growth restriction
FGR was defined as a birth weight <the 10th percentile for GA with abnormal Doppler indices (either UA-PI >the 95th percentile or MCA-PI <the 5th percentile for gestational age)(17), and/or as a birth weight of less than the 3rd percentile(18) according to local standards(19), regardless of the Doppler status before delivery. Late-onset FGR was defined as FGR that was newly diagnosed at greater than 32 weeks gestation(20).
Continuous variables were analyzed using the unpaired Student’s t-test, while categorical variables were analyzed using the Pearson χ2 test. The Z-score was calculated by dividing the difference between the observed value and the gestational age-specific mean with the standard deviation (SD). The observed measurements of CRL, MAP, UA-PI, HC/AC and EFW were expressed as the respective Z-scores corrected for gestational age. If one woman had repeated test results in one trimester, then the mean Z-score was calculated. Multivariate logistic regression analysis with backward stepwise elimination was used to determine which maternal factors and aspects of the obstetrical history significantly contributed to predicting late-onset FGR and accuracy was assessed by receiver-operating characteristic (ROC) curve. The predicted probabilities from the regression model were documented, and the detection rates (DRs) for a 10% false positive rate (FPR) were calculated. The probabilities in the validation dataset were computed using the formula: P= 1/(1+e (-(predictor))), in which the predictor is the sum of the regression coefficients multiplied by their predictor variable values from the training dataset. The discriminative ability of the model is expressed as AUC. The performance of classification model is evaluated by confusion matrix that was composed by: True Positive (TP), False Positive (FP), True Negative (TN) and False Negative (FN). The categories are obtained by specifying the threshold on the probabilities which is obtained from the training dataset. From the confusion matrix, accuracy, sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) can be calculated.
Accuracy= (TP+TN) / (TP+FP+TN+FN)
Sensitivity= TP / (TP+FN)
Specificity =TN / (TN+FP)
PPV= TP / (TP+FP)
NPV = TN / (TN+FN)
The statistical software package R version 3.4.1 was used for all data analyses.