Externally validated nomogram for predicting short-term pregnancy outcome of singleton pregnancies with fetal growth restriction (FGR)

Introduction

Fetal growth restriction (FGR) will be used to describe fetuses with an estimated fetal weight (EFW) that is less than the 10th percentile for gestational age, whereas the term small for gestational age (SGA) will be used exclusively to describe newborns whose birth weight is less than the 10th percentile for gestational age.1 2 FGR included early FGR (<32 weeks) and late FGR (≥32 weeks) with different parameters.2 FGR is characterised by the failure of the fetus to achieve its normal growth potential and is associated with perinatal morbidity and mortality.3–5 FGR infants have been reported to be associated with an increased risk of adverse perinatal outcomes (APOs) and half of stillbirths are due to FGR in utero.6 7 Gestational age (GA) is considered as the strongest predictor of postnatal development.8 9 However, in a clinical context, attempts to prolong early-onset FGR pregnancies have to be balanced against the risk of intrauterine demise. One of the main challenges of antenatal care is to identify at-risk fetuses to enable optimum surveillance, timely delivery and even timely termination of pregnancy. For those with adverse pregnancy outcomes, such as stillbirth or iatrogenic labour induction, termination of pregnancy (TOP) before the third trimester would reduce physical and psychological damage to the pregnant woman. Predictive algorithms for the selection of FGR pregnancies have preliminarily been well developed.10 11 Nevertheless, the prediction models can only identify FGR patients, without predicting the pregnancy outcomes. So, to develop an available predictive model to predict the prognosis of singleton FGR pregnancies for accurate and individualised prognosis assessment was very important.

The nomogram has been widely used as a predictive method in disease in recent years.12 13 It meets the requirements for an integrated model, plays a part in the drive towards personalised medicine13 and is convenient for clinicians to use in prognosis prediction.14 15

In the current study, the primary outcome of FGR was TOP, which included intrauterine fetal death and therapeutic lethal induction. We selected the FGR singleton pregnancies as the research objects and developed a nomogram to predict TOP, in singleton pregnancies of FGR in China. An available nomogram for predicting the prognosis of singleton FGR pregnancies in Chinese women was preliminarily developed and externally validated.

Materials and methods

Patients and study design

A retrospective study was conducted on 301 singleton FGR pregnancies at the Peking University People’s Hospital (Beijing, China) from January 2010 to September 2021 as the training set. Inclusion criteria included the following: singleton FGR pregnancies; newborns whose EFW and birth weight were both less than the 10th percentile for gestational age; with definite pregnancy outcome. Exclusion criteria were as follows: twin pregnancy, no pregnancy outcome, pregnancies with chromosome abnormalities, intrauterine infection and the missing data of the clinical factors such as age, history of induced abortion, history of caesarean section, history of abnormal pregnancy, amniotic fluid, umbilical artery flow, fetal anomaly, labour presentation, maternal complication, history of allergy, in vitro fertilisation and embryo transfer (IVF-ET), pre-pregnancy body mass index (BMI), umbilical cord abnormal, placenta abnormality, anaemia and albumin level. From January 2010 to September 2021, an external cohort of 321 singleton FGR pregnancies at the Affiliated Hospital of Qingdao University (validation dataset) were collected, using the same inclusion and exclusion criteria. The excluded patients in each group and the data flow chart were detailed in figure 1. The study was censored on 10 September 2021. The primary outcome in this paper was defined as TOP, which included intrauterine fetal death and therapeutic lethal induction with indications.

Figure 1

Request permissions

Flow chart of study participants in training and external validation groups. APOs, adverse perinatal outcomes; AUC, area under the receiver operating characteristic curve; DCA, decision-making curve; FGR, fetal growth restriction.

Results

Nomogram of prediction model

The prognostic nomogram integrated all significant factors including history of abnormal pregnancy, umbilical artery flow, fetal anomaly, labour presentation and history of caesarean section for pregnancy outcome of singleton FGR pregnancies in the training cohort as shown in figure 2. For a given patient, points were assigned to each of the predictor variables in the nomogram and a total score was derived from the sum of present variables. The total score corresponds to a predicted probability of APOs of singleton FGR pregnancies.

Figure 2

Request permissions

Nomograms for predicting 28th week and 34th TOP-free gestation week (GW) in singleton pregnancies with FGR. In order to evaluate the TOP-free gestation rate of each singleton FGR patient, the score of each variable was calculated by the value of the ‘points’ axis, and the sum of the values of all variables was corresponding to the number of the ‘total points’ axis. The vertical line of the total score was corresponding to 28th TOP-free GW and 34th TOP-free GW. AEDF, absent end-diastolic flow of umbilical artery flow; FGR, fetal growth restriction; REDF, reverse end-diastolic flow of umbilical artery flow; TOP, termination of pregnancy.

Internal validation of the prediction models

The performance of the final model was assessed through discrimination and calibration. In the internal validation of training cohort, there was a C-index of 0.859 (95% CI: 0.81 to 0.90) for predicting TOP, which included intrauterine fetal death and therapeutic lethal induction, a C-index of 0.92 (95% CI: 0.86 to 0.98) for predicting stillbirth and a C-index of 0.87 (95% CI: 0.83 to 0.92) for predicting therapeutic lethal induction with indications. The sensitivity was 71.43%, specificity was 87.35%, positive likelihood ratio was 1. 29 and negative likelihood ratio was 0.07. P value of likelihood ratio test was <0.05. The AUCs (area under the receiver operating characteristic curves) for the 28th and 34th TOP-free gestation week (GW) were 0.90 and 0.89 (figure 3A,B), respectively. The calibrations of the nomogram predicting the 28th and 34th TOP-free GW showed an optimal agreement between the prediction by nomogram and actual observation (figure 4A,B).

Figure 3

Request permissions

Area under the receiver operating characteristic curve (AUC) values of the nomogram. (A) AUC of nomogram in the training group for predicting 28th TOP-free gestation week (GW) in singleton FGR pregnancies. (B) AUC of nomogram in the training group for predicting 34th TOP-free GW in singleton FGR pregnancies. (C) AUC of nomogram in the validation group for predicting 28th TOP-free GW in singleton FGR pregnancies. (D) AUC of nomogram in the validation group for predicting 34th TOP-free GW in singleton FGR pregnancies. FGR, fetal growth restriction; GA, gestational age; TOP, termination of pregnancy.

Figure 4

Request permissions

The calibration curve for predicting pregnancy outcome of singleton FGR pregnancies. (A) 28th TOP-free gestation week (GW) in the training cohort. (B) 34th TOP-free GW in the training cohort. (C) 28th TOP-free GW in the validation cohort. (D) 34th TOP-free GW in the validation cohort. Nomogram-predicted probability of overall survival is plotted on the x-axis; actual overall survival is plotted on the y-axis. FGR, fetal growth restriction; GA, gestational age; TOP, termination of pregnancy.

Risk score model indicated strong association with clinical characteristics in singleton FGR pregnancies

We further analysed the distribution of patients in the low-risk and high-risk groups estimated by the nomogram scores. A median cut-off value (88 points) was applied to stratify singleton FGR pregnancies into a high-risk group (n=43, score ≤87 points) and a low-risk group (n=258, score ≥89 points). The clinical factors of training and validation cohorts between high-risk and low-risk groups were presented in the heatmap (figure 5A and D). The results showed that there were significant differences between the high-risk and low-risk groups in terms of history of abnormal pregnancy, umbilical artery flow and fetal anomaly (p<0.05). The prognostic status of training and validation cohorts was shown in figure 5B and E. Obviously, it was observed that most APOs of singleton FGR pregnancies were distributed in the high-risk part. Both in training and validation cohorts, prognostic analysis in the form of Kaplan-Meier curve showed that the high-risk group had a significant shorter GA than the low-risk group (figure 5C and F, p<0.05).

Figure 5

Request permissions

Differential clinical factors and predictive accuracy of singleton FGR pregnancies in high-risk and low-risk groups. (A) Heatmap and clinical features of high-risk and low-risk groups in the training group. The samples are ordered by risk score, and the score decreases from left to right. *P<0.05, **p<0.01 and ***p<0.001. (B) Risk score distribution in low-risk and high-risk groups in the training group. (C) Kaplan-Meier survival analysis of the low-risk and high-risk group in the training group. (D) Heatmap and clinical features of high-risk and low-risk groups in the validation group. The samples are ordered by risk score, and the score decreases from left to right. *P<0.05, **p<0.01 and ***p<0.001. (E) Risk score distribution in low-risk and high-risk groups in the validation group. (F) Kaplan-Meier survival analysis of the low-risk and high-risk group in the validation group. FGR, fetal growth restriction.

The clinical decision-making curve (figure 6) shows that within a threshold probability from 3% to 49%, patients could benefit from the application of this predictive model.

Figure 6

Request permissions

The clinical decision-making curve of the predictive model for singleton fetal growth restriction pregnancy outcome.

Discussion

FGR infants, those with an EFW that is less than the 10th percentile for GA,18 have been reported to be associated with increased risk of short-term and long-term APOs.19 Predictive algorithms for the selection of FGR pregnancies have preliminarily been well developed.10 11 Nevertheless, these prediction models can only identify FGR pregnancies, without predicting the pregnancy outcomes of these FGR patients. Prediction of FGR patients’ outcome is an important step of a multidimensional approach, which includes adequate management, termination in time or long-term follow-up of these newborns. Apart from only monitoring FGR patients regularly during pregnancy, early prediction of FGR prognosis was a key to clinic decisions. In the present study, we developed a nomogram to predict APOs in singleton FGR pregnancies.

The independent risk factors of singleton FGR pregnancies’ APOs selected by univariate analysis in the present study were somewhat different from those in the previous studies. Umbilical arterial flow, fetal anomaly, history of abnormal pregnancy, labour presentation and history of caesarean section were significantly related to the short-term pregnancy outcome of singleton FGR pregnancies in the training group in this study. One or more of these indicators have been included in previous studies20; however, the combination of these five indicators included in a predictive model was the first time in this study. Absent end-diastolic flow of umbilical artery flow or reverse end-diastolic flow of umbilical artery flow (REDF) was recognised as a sign of severely impaired placental perfusion and was an indicator of adverse outcome.21 22 Nevertheless, a previous study showed that in early-onset FGR, up to 30–32 weeks’ gestation, umbilical artery Doppler was usually not part of management protocols.23 In this paper, the results showed that REDF was an important independent factor associated with the prognosis of singleton FGR pregnancies (table 2). The clinical decision-making curve (figure 6) shows that within a threshold probability from 3% to 49%, patients could benefit from the application of this predictive model.

The Growth Restriction Intervention Study showed better neurological outcome when timely decisions are made in early FGR in a randomised trial based on a combination of computerised cardiotocography and ductus venosus (DV) Doppler assessment.24 DV was an important factor in predicting the outcome of fetuses. We could not include DV in establishing the model owing to the missing data. In future studies, adding DV in the model may further improve the prediction accuracy of the model.

Doppler abnormalities can predict the occurrence of complications in the short term, but normal fetal Doppler values at the time of diagnosis do not exclude their occurrence in the long term.25 26 Especially in the case of late-onset SGA (>32 weeks’ gestation), umbilical artery Doppler is commonly normal.27 For these reasons, counselling of parents with an affected fetus at that time might not be very accurate and this uncertainty may arouse anxiety and distress in parents.28 Other parameters that aid the detection of cases at higher risk of APOs were of great importance.

FGR fetuses have been reported to be complicated with structural abnormalities or in the middle trimester, abnormal soft indicators of ultrasound, such as intestinal echo enhancement, may occur at a high rate of 37%, but the study did not rule out genetic abnormalities.29 In the absence of chromosomal karyotype abnormalities, the incidence of ultrasound abnormalities in FGR was about 25%; femur shortening, omphalocele and abdominal wall fissure were the most common.30 31 Fetal chromosomal abnormalities account for 15~20% of the causes of FGR, and triploid and aneuploid are the most common.31 Therefore, when FGR fetuses are associated with structural abnormalities or abnormal ultrasonic genetic markers, interventional prenatal diagnosis, chromosomal microarray and karyotype analysis are recommended. In this paper, we included only singleton FGR pregnancies with non-chromosomal abnormalities and found fetal structural abnormalities were related to the APOs of singleton FGR patients, most of those did not have an indication of induced labour in terms of the structural abnormalities themselves. Therefore, for those non-chromosomal abnormal singleton FGR pregnancies that had structural abnormalities without indications for induced labour, other indicators need to be considered to determine the final indication of induced labour.

History of abnormal pregnancy like stillbirth increased the risk of other abnormal pregnancy outcomes in the subsequent pregnancy such as FGR placental abruption, caesarean delivery and preterm delivery.32 In the present study, we found that history of abnormal pregnancy was related to APOs of singleton FGR pregnancies. Abnormal labour presentation was related to the causes of stillbirth during labour.33 In this paper, we first found that breech/transverse position was an independent pregnancy prognostic factor of singleton FGR pregnancies, revealing that abnormal labour presentation may be a sign of APOs of singleton FGR pregnancies, though it may not be a cause of the APOs. The result of our paper showed that the history of caesarean section may be related to APOs of FGR patients. So, for those singleton FGR pregnancies with a history of abnormal pregnancy, abnormal labour presentation or history of caesarean section, pregnancy monitoring and strict management should be further strengthened.

A median cut-off value (88 points) was applied to stratify single pregnant women into a high-risk group and a low-risk group. Though the results showed that there were significant differences between the high-risk and low-risk groups in terms of clinical factors such as umbilical artery flow, fetal anomaly and history of abnormal pregnancy, individual differences in singleton FGR pregnancies after redistribution were found in the high-risk group and low-risk group. So, the results further demonstrated that individual clinical factor was difficult to accurately determine FGR patient outcome; an algorithm based on several related clinical factors may be more useful to predict the prognosis of singleton FGR pregnancies. It was found in this study that within a threshold probability from 3% to 49%, singleton FGR patients could benefit from the application of this predictive model.

For the clinical implications of this work, first, as for the fetus at high risk of TOP predicted by the prediction model, pregnancy should be closely monitored and treated more aggressively. Second, the prediction model for death in this paper mainly could predict the short outcome of fetuses and clinical trials should be further taken to demonstrate whether this predictive model could improve the outcome of fetuses with FGR. We hypothesised that after verification of the present findings in prospective studies, proactive perinatal clinical protocols, taking into account this predictive model when deciding on the time of termination of FGR patients, might reduce physical and psychological harm to FGR pregnant women.

There were some limitations in the current research. First, FGR patients were not divided into early-onset FGR and late-onset FGR in the cohorts due to the limited number of cases. Second, twin pregnant women were not included in this study. Third, there were short-term and long-term APOs of FGR patients, but we only predicted the short-term pregnancy outcome of the FGR patients, so long-term APOs could be further predicted in future research and clinical trials should be taken to demonstrate whether this predictive model could improve the outcome of fetuses with FGR. Fourth, owing to the limited sample size, APO in this paper was defined as TOP, which included intrauterine fetal death and therapeutic lethal induction with definite indications given by prenatal diagnostician. It may lead to bias and it makes more sense to predict intrauterine fetal death in the future study. Fifth, samples of the training and validation sets came from completely two different hospitals, which may lead to some bias. In the future study, further enlarging the sample size may help reduce the bias. In addition, the research was a retrospective study, and prospective validation was needed to verify the promotion and application of the model. Finally, there is a risk of heterogeneity of the study variables and population; further optimisation of this model in a national multicentre study is needed.