Prediction of small-for-gestational-age neonates at 33–39 weeks’ gestation in China: logistic regression modeling of the contributions of second- and third-trimester ultrasound data and maternal factors

This was a retrospective study. A total of 21 092 pregnant women gave birth in Taizhou hospital from January 2017 to March 2021. Figure 1 outlines the data collection process and 5 298 pregnant women were included in this study. The inclusion criteria were single live births at 33 to 41 weeks’ gestation, and a history of at least three ultrasonographic examinations at 21⁺⁰–27⁺⁶, 28⁺⁰–32⁺⁶, and 33⁺⁰–39⁺⁶ weeks at our hospital. Considering the errors caused by different hospitals with different doctors and instruments, the reports form different hospitals were excluded. Maternal characteristics, diseases during the pregnancy, ultrasound data, and delivery information were recorded when the pregnant women came to our hospital for delivery.

The GA of the fetus was determined by the last menstrual period and the crown-to-rump length (CRL) of the fetuses at 11–14 weeks [17]. When the GAs determined by both methods were less than 1 week apart, the last menstrual period was chosen to determine the GA. However, when the GAs determined by both methods were more than 1 week apart, the GA based on ultrasound measurements was included in the analyses. All ultrasound examinations were performed by examiners with a certificate from a medical practitioner. In China, pregnant women younger than 35 weeks were required to participate in the first- and second-trimester screening or NIPT screening, and older and high-risk pregnant women were required to undergo amniocentesis to eliminate major fetal chromosome diseases. When pregnant women from 33 to 40 weeks of gestation came to our hospital for ultrasound appointment, we introduced the intention of this study to them and sign an informed consent form. All pregnant women, admitted to our hospital for delivery, were informed that all information about their pregnancy might be used anonymously in the future and signed informed consent. The study was approved by Taizhou Ethics Committee.

The previously published report showed expected sensitivity towards EFW for detecting neonatal SGA to be 60.6% and specificity 87.6%, whereas for AC this rate was 64.4% and 83.5%, respectively, between 33⁺⁰ ~ 35⁺⁶ gestational weeks [18]. The sample size was calculated by software according to the above rates and a usable sample size of 88 patients was required in the anomalies group, with an allowable error of 0.1 in a two-tailed test with p < 0.05.

All pregnant women were Chinese. Maternal age, height, weight, body mass index (BMI), pregnancy history (the number of pregnancy, number of production, scarred uterus), and disease history during pregnancy were recorded.

Hypertension, diabetes, pre-eclampsia, vaginal inflammation (such as bacterial vaginosis, candida vaginosis, streptococcus lactis vaginitis, trichomonas vaginitis), viral infection (HBV, syphilis, HPV, etc.), intrahepatic cholestasis, thyroid dysfunction, abnormal placental shape, placental hypofunction, anemia or pelvic adhesionism during pregnancy might affect the health and development of the fetuses. As we all knew, pre-eclampsia was a type of hypertensive disorder of pregnancy that could lead to conditions such as fetal growth restriction and placental abruption. Gestational diabetes could cause fetal overgrowth. Vaginal inflammation might lead to infection, causing pelvic inflammatory disease and intrauterine infection. placental hypofunction (placental aging) might lead to hypoxia of the fetuses and even to the arrest of placental development.

The delivery process was also well-documented, and the mode of delivery, fetal distress, vaginal laceration, and the use of low forceps was recorded.

Neonatal sex, weight, GA at birth, and neonatal asphyxia were recorded. According to the latest Chinese standards [19], a fetus weighing less than the 10^th percentile or more than 90^th percentile of the GA (≤ 36 weeks), and weighing less than 2500 g or more than 4000 g after term is considered a SGA neonate or large for GA (LGA) neonate, respectively. Neonates that do not meet the criteria for SGA and LGA statuses are called appropriate for gestational age (AGA) neonates.

All pregnant women had undergone ultrasound examinations at 20⁺⁰–27⁺⁶ weeks, 28⁺⁰–32⁺⁶ weeks, and 33⁺⁰–39⁺⁶ weeks. Head circumference (HC), abdominal circumference (AC), femur length (FL), biparietal diameter (BPD), occipitofrontal diameter (OFD), amniotic fluid index (AFI), placebtak thickness (PT), the ratio of systolic velocity / diastolic velocity of umbilical artery blood flow (S/D), pulsatile index of umbilical artery (PI), resistance index of umbilical artery blood flow (RI) were recorded. The 10^th, 50^th and 90^th percentile values of all ultrasound data were counted according to gestational age.

All ultrasound data in this paper were transformed according to the local median of gestational ages, that was to say MoM values (multiple of the median). The reasons were as follows: At each GA, ultrasound data were not normally distributed. And each region had distinct differences due to climate and diet differences (e.g., northern and southern China). Therefore, MoM values were more appropriate than Z-Score when ultrasound data were transformed according to GA in this region.

The combination of AC, FL, BPD and HC yields a variety of formulas for calculating EFW [6]. According to the ultrasonic data from 33⁺⁰–39⁺⁶ weeks’ gestation, EFW was calculated respectively and finally converted into EFW MoM values (EFW-M) according to GA. ROC curve analysis showed that the formula created by Hadlock (1985) based on AC, FL and HC was the most suitable for this paper (Data not displayed).

Previous article showed that AC growth velocity was better than EFW growth velocity for prediction of SGA neonate between 20 and 36 weeks [11]. The calculation method of AC growth velocity referred to this article.

Multivariate logistic regression was used to establish a SGAp model. Univariate logistic regression analysis was used to analyze the relationship between related variables and SGA neonates. The predicted SGA values were recalculated based on this model for all fetal conditions. The DR of SGA neonates was observed at 5% and 10% FPR. At the same time, The FPR of SGA was observed at the DR of 85%, 90% and 95%.

Data were described in terms of medians/interquartile range (IQR) for continuous variables or numbers (n and %) for categorical variable. Mann–whitney U test, rank sum test or Chi-square test were used to compare differences between groups. Receiver operating characteristic (ROC) curves analysis was performed to evaluate the power of SGAp, AC MoM value at 33⁺⁰–39⁺⁶ weeks (AC-M), EFW-M, EFW-M plus MF, AC value at 33⁺⁰–39⁺⁶ weeks (AC), AC growth velocity, EFW, and EFW plus MF to discriminate SGA neonates. A p < 0.05 was considered to be significant.

A total of 5 298 pregnancies with GAs of 33–41 weeks at birth were included in the study (Fig. 1 shows the inclusion process). The study population included 214 SGA (4.1%), 4828 AGA (91.1%), and 256 LGA (4.8%) neonates. The AGA and LGA neonates, collectively referred to as non-SGA neonates, served as the control group in this study. Basic information on the pregnant women, disease history during pregnancy, delivery, and newborn information are shown in Table 1. Weight, height, and BMI during pregnancy, proportion of boys, gestational week of delivery, and the proportion of streptococcal vaginitis in the women who delivered SGA neonates were significantly lower than those in women who delivered AGA and LGA neonates. On the other hand, the proportions of cesarean deliveries and patients with chronic hypertension and preeclampsia among women who delivered SGA neonates were significantly higher than those among women who delivered AGA and LGA neonates. Correlation analysis showed that SGA was correlated with maternal height, weight, BMI, fetal sex, and gestational disease history (streptococcus lactis vaginitis, gestational hypertension, preeclampsia, intrahepatic cholestasis during pregnancy), while LGA was associated with maternal age, height, weight, BMI, number of pregnancies, number of deliveries, baby sex, and gestational disease history (gestational diabetes, bacterial vaginitis, anemia).

All pregnant women had undergone ultrasound examinations at 20⁺⁰–27⁺⁶ weeks, 28⁺⁰–32⁺⁶ weeks, and 33⁺⁰–39⁺⁶ weeks. In China, the most accurate detection time for abnormal ultrasound findings is 23–25 weeks, so a large number of people choose to undergo ultrasound at 24 weeks’ gestation. All data were grouped according to GA and described as median, 10^th, and 90^th percentiles (Table 2). After all data were MoM value-transformed, rank-sum test analysis showed that the BPD, OFD, HC, FL, AC, and AFI of SGA fetuses were significantly lower than those of AGA and LGA fetuses at 20⁺⁰–27⁺⁶, 28⁺⁰–32⁺⁶, and 33⁺⁰–39⁺⁶ weeks’ gestation (Table 3).

Multivariate logistic regression modeling was conducted for all factors and ultrasound data after transformation. SGA neonates were represented by the dichotomous variable. It was found that the variables significantly correlated with the history of hypertension in pregnant women (X₁, normal blood pressure = 0, gestational hypertension = 1, chronic hypertension = 2), preeclampsia (X₂, no preeclampsia = 0, mild disease = 1, severe disease = 2), OFD MoM value at 21⁺⁰–27⁺⁶ weeks (OFD-M-[21, 27]) (X₃), FL MoM value at 21⁺⁰–27⁺⁶ weeks (FL-M-[21, 27]) (X₄), AC MoM value at 28⁺⁰–32⁺⁶ weeks (AC-M-[28, 32]) (X₅), BPD MoM value at 33⁺⁰–39⁺⁶ weeks (BPD-M) (X₆), FL MoM value at 33⁺⁰–39⁺⁶ weeks (FL-M) (X₇), AC MoM value at 33⁺⁰–39⁺⁶ weeks (AC-M) (X₈), AFI MoM value at 33⁺⁰–39⁺⁶ weeks (AFI-M) (X₉). This model is called the SGA predictor (SGAp) (As shown in the Table 4). SGAp = -59.496–0.784X₁-0.693X₂-9.377X₃ + 7.26X₄ + 14.578X₅ + 14.903X₆ + 8.436X₇ + 26.531X₈ + 2.087X₉.

Univariate logistic regression results also showed that these indexes were closely related to SGA neonates (Table 5). After MoM transformation, the ultrasound data were all between 0 and 2, which were close to the values of hypertension and eclampsia. Univariate logistic regression results illustrated that OFD-M-[21, 27], FL-M-[21, 27], AC-M-[28, 32], BPD-M, FL-M, AC-M had high risk factors for SGA (Table 5).

The SGAp values, AC-M, EFW-M, and EFW-M plus MF of fetuses in the SGA group were significantly lower than those in the non-SGA group (Table 6). The SGAp values were significantly different between the SGA and non-SGA groups (Table 6). The AUCs of SGAp, AC-M, EFW-M, EFW-M plus MF, AC, AC growth velocity, EFW, and EFW plus MF are showed in Fig. 2. The screening efficiency of SGAp, AC-M, EFW-M, and EFW-M plus MF are significantly higher than AC, AC growth velocity, EFW, and EFW plus MF. The DR of SGAp, AC-M, EFW-M, and EFW-M plus MF for SGA neonate screening at 5% and 10% FPR are shown in Table 7. The corresponding FPR of these four indicators at 85%, 90%, and 95% DR are also shown in Table 8.

The present study confirmed that the SGAp, which was constructed using ultrasound data obtained in the second and third trimesters along with data for maternal history of hypertension and preeclampsia, showed better screening ability than EFW. The AUCs of SGAp, AC-M, EFW-M, EFW-M plus MF, AC, AC growth velocity, EFW, and EFW plus MF for SGA neonate screening were 0.933 (95%CI: 0.916–0.950), 0.906 (95%CI: 0.887–0.925), 0.920 (95%CI: 0.903–0.936), 0.925 (95%CI: 0.909–0.941), 0.818 (95%CI: 0.791–0.845), 0.786 (95%CI: 0.752–0.821), 0.810 (95%CI: 0.782–0.838), and 0.834 (95%CI: 0.807–0.860), respectively. However, all eight measures (SGAp: 0.950, 95%CI: 0.932–0.967; AC-M: 0.929, 95%CI: 0.909–0.948; EFW-M: 0.938, 95%CI: 0.921–0.956; EFW-M plus MF: 0.941, 95%CI: 0.924–0.957; AC: 0.874, 95%CI: 0.847–0.900; AC growth velocity: 0.791, 95%CI: 0.746–0.837; EFW: 0.866, 95%CI: 0.839–0.893; EFW plus MF: 0.873, 95%CI: 0.847–0.899) showed more effective screening performance if birth occurred within two weeks of the assessment. The SGA screening efficiency of data transformed by MoM value was significantly higher than that of data without MoM value transformation.

The DR of SGAp, AC-M, EFW-M, and EFW-M plus MF at 10% FPR were 85.8%, 75.8%, 80.0%, and 82.5%, respectively for screening SGA neonates delivered < 2 weeks after the assessment. The FPR of SGA screening by SGAp for 85%, 90%, and 95% DR were 9.4%, 12.4%, and 22.0%, respectively, in deliveries occurring < 2 weeks after the assessment.

The DR of SGAp, AC-M, EFW-M, and EFW-M plus MF for birth at any time were 80.4%, 69.6%, 73.8%, and 74.3%, respectively. The FPR of SGA fetal screening by SGAp for 85%, 90%, and 95% DR were 11.8%, 16.4%, and 27.8%, respectively.

The strengths of this SGA neonatal screening study are as follows: First, based on the local median values for each GA, ultrasound data were MoM value-transformed, increasing their accuracy. Second, the study participants included pregnant women whose babies were born at 33–41 weeks (including preterm delivery). Third, the SGAp was based on prenatal ultrasound and maternal disease data, thereby ensuring better SGA screening than EFW.

The limitations of this study are as follows: First, this was a retrospective study. Of the 21092 pregnant women who gave birth in our hospital, 3/4 had been examined by ultrasound once or twice in our hospital, and most of them were tested in their local women's health care centers. Moreover, the ultrasound data were not incomplete in their local women's health care centers, resulting in a large amount of data loss. Second, the evaluations based on the SGAp model could only be performed after 33 weeks of gestation.

We found that EFW-M and EFW-M plus MF assessments in the third trimester could predict 73.8% and 74.3%, respectively, of SGA neonates delivered at 33–41 weeks’ gestation at 10% FPR. Fadigas et al. used EFW Z-score (EFW-Z) and EFW-Z plus MF data at 35–37 weeks and reported that 63.1% and 66.0% of SGA neonates (< 10th percentile) were screened at 10% FPR [20]. Ciobanu et al. also used EFW-Z obtained at 35⁺⁰ to 36⁺⁶ weeks of gestation for screening SGA neonates (< 10^th percentile) and reported DR of 65.3% and 69.3% at 10% FPR for deliveries at ≥ 35 weeks’ gestation [21]. Bakalis et al. found that EFW-Z plus MF assessments at 30–34 weeks could predict 79.2% and 52.7% of SGA neonates with 10% FPR in deliveries occurring < 5 weeks and > 5 weeks after the assessments, respectively [16]. Overall, the screening effect of EFW-M was similar to that of the EFW-Z. However, it is easier to convert according to the GA.

This is the first study to combine ultrasound and MF data to construct an SGAp model. The SGAp model could screen 80.4% of SGA neonates at 10% FPR in deliveries at 33–41 weeks of gestation. For deliveries that occurred within two weeks of the evaluation, the DR increased up to 85.8%.

In China, it is common for pregnant women to undergo ultrasound examinations five times during pregnancy: at 6–8 weeks, 12–14 weeks, 23–25 weeks, 29–31 weeks, and 34–36 weeks. In addition, some pregnant women in the third trimester may undergo ultrasound examinations every month or even at two-week intervals. Appropriate use of these ultrasound data is extremely important. AC and EFW growth velocities between 20 and 36 weeks of gestation cannot be used for effective screening¹². Therefore, in this study, the most effective ultrasonic data across different stages were superimposed, and a logistic regression model was used to establish the SGAp model. The screening performance of the SGAp model was shown to be better than that of EFW. The three ultrasound data points used in this study were all obtained over a relatively large gestational range, ranging from 20 to 27 weeks, 28 to 32 weeks, and 33 to 39 weeks, improving the convenience of performing SGAp-based assessments in actual clinical practice.