Usefulness of angiogenic factors in prenatal counseling of late-onset fetal growth-restricted and small-for-gestational-age gestations: a prospective observational study

To create a predictive model including biomarkers and evaluate its ability to predict adverse perinatal outcomes in late-onset small fetuses, ultimately helping to provide individualized counseling at the time of diagnosis. This was a prospective observational study, including singleton pregnancies with an estimated fetal weight (EFW) below the 10th percentile, at a gestational age between 32 + 0 and 36 + 6 weeks of gestation (WG). Variables recorded at diagnosis to predict adverse pregnancy outcomes were: soluble fms-like tyrosine-kinase-1 to placental growth factor ratio (sFlt-1/PlGF), fetal Doppler (umbilical artery and middle cerebral artery), uterine artery pulsatility index (UtAPI), EFW percentile, gestational age, and the presence of maternal risk factors for placental insufficiency. Logistic regression models were developed for the prediction of three co-primary outcomes: composite adverse perinatal outcomes (APO), and the need for elective delivery before 35 or 37 WG. Sixty (52.2%) fetal growth restricted (FGR) and 55 (47.8%) small for gestational age (SGA) were enrolled. Thirteen (11.3%) women needed elective delivery before 35 WG and 27 (23.5%) women before 37 WG. At least one APO occurred in 43 (37.4%) pregnancies. The best marker in univariate analyses was the sFlt-1/PlGF ratio [AUC = 0.932 (95% CI, 0.864–0.999)]. The multivariate model including sFlt-1/PlGF showed a better predictive performance for APO than the multivariate model without sFlt-1/PlGF (P < 0.024). sFlt-1/PlGF is a good predictor of APO at the time of late-onset FGR/SGA diagnosis. Our predictive models may be useful to provide early individualized prenatal counseling in this group of women. Further studies are needed to validate these preliminary findings in a larger cohort.


Introduction
Fetal growth restriction (FGR) refers to a fetus that is not reaching its full growth potential [1]. Although it can be the result of chromosomal abnormalities [2], congenital infections [3] or other major malformations [4], more than two-thirds of FGR cases are caused by a mismatch between placental supply and the nutritional requirements of the fetus. This is the so-called placental-related fetal growth restriction, wherein there is a physiological deficiency in the remodeling of uterine and placental spiral arteries resulting in restricted uteroplacental perfusion [5].
According to the Delphi consensus criteria, late-onset FGR is identified after 32 weeks of gestation (WG) [1,6]. It is the more common presentation of growth restriction (up to 80% of FGR cases), and is generally linked with a milder placental deficit, together with a minor degree of fetal hemodynamic adaptation. Although placental dysfunction is mild, this group has a high risk of deteriorating rapidly, such that they have an elevated risk of stillbirth [7].
An accurate diagnosis of FGR and its discrimination from constitutional small-for-gestational-age (SGA) is key for making an appropriate management of these cases and providing accurate prenatal counseling. To discriminate FGR from SGA, a series of prenatal sonographic markers have been identified, such as an estimated fetal weight (EFW) below the third percentile [8] or abnormal maternal-fetal Doppler assessment [9,10]. These ultrasound findings have been related to a greater risk of stillbirth [11] and a greater post-natal morbidity and mortality [12].
Recent research has been focused mainly on the role of ultrasound in predicting adverse perinatal outcomes in FGR and SGA pregnancies [13][14][15]. Accordingly, consensus definitions for FGR now incorporate Doppler indices of placental dysfunction during pregnancy to provide a more robust assessment of pathological growth restriction [1]. A recent study showed that when late-onset FGR is diagnosed, EFW percentile at diagnosis is the only sonographic parameter associated with adverse pregnancy outcomes and that abnormal uterine artery pulsatility index (UtAPI) is associated with intrapartum fetal distress leading to obstetric intervention [15]. Late-onset SGA fetuses with abnormal UtAPI have a two-to-threefold increased risk of adverse perinatal outcomes (APO). UtAPI and cerebro-placental ratio (CPR) have similar ability to discriminate SGA from FGR [14]. By contrast, umbilical artery (UA) Doppler remains normal in the majority of late-onset FGR and SGA cases, making it less accurate to predict complications [14].
Recently, it has been described that angiogenic factors in maternal serum, such as placental growth factor (PlGF) and soluble fms-like tyrosine-kinase 1 (sFlt-1), are reliable markers for detecting cases at a higher risk of placental insufficiency and poor outcomes [16,17]. The association of these markers with histological signs of placental hypoperfusion in late-onset SGA pregnancies has already been shown [18]. Furthermore, compared to clinical factors and UtAPI and UA Doppler, these markers seem to improve the prediction of adverse outcomes in FGR and SGA pregnancies [19].
Other parameters used in the prediction of adverse outcomes in cases with placental dysfunction are maternal characteristics. The evaluation of maternal age, ethnic origin and history of chronic hypertension or preeclampsia contribute significantly to the assessment of risk for adverse pregnancy outcomes, such as stillbirth or iatrogenic preterm birth [20][21][22].
Doppler studies, maternal characteristics, gestational age, EFW and biomarkers have all shown to be good predictors of adverse outcomes. The combination of all these markers within a predictive model may potentially improve the identification of SGA/FGR cases at a higher risk of adverse outcomes. The aim of this study was to create a model and evaluate its predictive ability for adverse perinatal outcomes in late-onset FGR gestations, ultimately helping to provide early individualized prenatal counseling at the time of diagnosis.

Study design and participants
This prospective observational study was conducted at Hospital Universitari Vall d'Hebron (Barcelona, Spain) between July 2017 and July 2019. The study population consisted of pregnant women referred for fetal growth assessment due to an EFW below the 10th percentile between 32 + 0 and 36 + 6 WG. Exclusion criteria were multiple pregnancy, known fetal chromosomal abnormalities or fetal defects, intrauterine death, or the need for immediate delivery (due to non-reassuring cardiotocography, absent or reversed ductus venosus A-wave, reverse end-diastolic flow in the umbilical artery (UA), absent end-diastolic flow in the UA at the time of diagnosis before 34 WG or placental abruption).
The initial assessment involved the recording of maternal demographic characteristics and medical history, ultrasound assessment of fetal anatomy and EFW (based on biparietal diameter, fetal head circumference, abdominal circumference and femur length) [23], color Doppler for measurement of mean UtAPI, umbilical artery pulsatility index (UAPI) and UA diastolic flow (present, absent or reversed), Doppler for middle cerebral artery pulsatility index (MCAPI) and ductus venosus pulsatility index (DVPI) and ductus venosus diastolic flow (present, absent or reversed A-wave). Maternal serum was extracted by venopuncture at the time of FGR/SGA diagnosis. PlGF and sFlt-1 measurements (pg/mL) were performed by the fully automated Elecsys assay, which is based on an electrochemiluminescence immunoassay platform (cobas e analyzers; Roche Diagnostics, Penzberg, Germany). Gestational age (GA) was determined by fetal crown-rump length measurements at 11 + 0-13 + 6 WG [24]. SGA was defined as an EFW between the third and tenth percentile with no feto-maternal Doppler abnormalities and FGR was defined as an EFW below the third percentile or between the third and tenth percentiles accompanied by feto-maternal Doppler abnormalities [25]. All cases were assessed and followed up at a specific placental insufficiency unit by the same group of experienced consultants, who were not blinded to angiogenic factor levels, since these are routinely measured in our center at the time of FGR/SGA diagnosis to rule out preeclampsia [17]. Nevertheless, the timing and mode of delivery was always based on GA, Doppler findings, conventional visual cardiotocography (CTG) interpretation, and maternal signs and symptoms according to the current hospital protocols, which are based on the latest evidence [25], regardless the results of the sFlt-1/ PlGF ratio. According to that protocol, elective delivery was recommended at > 40 weeks for SGA, > 37 weeks for FGR with antegrade UA flow, > 34 weeks for FGR with AEDF and > 30 weeks for FGR with REDF. CTG indications for elective delivery were: fetal heart rate sinusoidal tracing or absent fetal heart rate variability accompanied by recurrent late decelerations, recurrent variable decelerations or bradycardia [26]. Preeclampsia was defined according to the definition of the American College of Obstetricians and Gynecologists [27]. Delivery at 34 weeks or later will be indicated in cases with PE with severity features, whereas in cases with no severity features, expectant management will be recommended until 37 + 0 weeks. In cases that required elective delivery before 35 weeks, a single course of antenatal corticosteroids (2 × 12 mg betamethasone administered intramuscularly within 24 h) was administered for accelerating fetal lung maturation.

Statistical analysis
The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement was followed for reporting the results [28]. Categorical data are presented as frequency and percentage and continuous data are presented as the median and interquartile range (IQR). First, univariate logistic regression analysis was performed to examine the association between each predictive variable and all examined adverse outcomes.
As outcome measures, we considered the following coprimary outcomes: the need for elective delivery before 35 WG, the need for elective delivery before 37 WG and the presence of adverse perinatal outcomes (APO). APO was defined as any of the following: stillbirth, intrapartum Cesarean section due to non-reassuring CTG, neonatal death, NICU admission for longer than 48 h, respiratory distress syndrome (RDS), bronchopulmonary dysplasia (BPD), neonatal sepsis, retinopathy of prematurity (stages III-IV), periventricular leukomalacia, Apgar score < 7 at 5 min or umbilical artery cord pH < 7. RDS was defined as respiratory rate above 60 or below 30 breaths per minute, grunting on expiration, chest indrawing, central cyanosis, apnea, or the use of surfactant therapy in the neonatal period [34].
Sample size was calculated assuming that the predictive models would yield at least 50% detection of any co-primary outcome. With a 80% power and a significance level of 0.05, the required sample size was 106 individuals.
The odds ratios (OR) with their respective 95% confidence intervals (CI) were calculated. Second, since some predictive variables may be related, further multivariateadjusted analyses were made to identify which predictive variables still had a significant contribution in both predictive models. A forward stepwise multivariate logistic regression analysis was used to identify which variables remained significant predictors of each of the examined pregnancy outcomes. We are aware that not all clinical practices have access to sFlt-1/PlGF measurements. Thus, we decided to create two different multivariate models to make these predictive algorithms applicable in most clinical settings: one model included all variables, and the other model included all variables except sFlt-1/PlGF. Adjusted OR (aOR) and their respective 95% CI were calculated for the variables that remained contributive in both multivariate models. Third, logistic regression analyses were used to derive the formula for calculating the risk for each pregnancy outcome: Y was derived by univariate or multivariate logistic regression analysis, as appropriate. Finally, the risks predicted by logistic regression analysis were used to construct receiver-operating characteristic (ROC) curves for assessing the performance of each individual variable and combinations of these variables for predicting each adverse pregnancy outcome. The resulting areas under the curve (AUC) were compared using the DeLong test [35].
The R statistical software (R Core Team 2018, R Foundation for Statistical Computing, Vienna, Austria) was used for all data analyses. Statistical significance level was set at P < 0.05.

Ethical approval
This study (PR(AMI)349/2016) was approved on June 2, 2017, by the Ethics Committee of Vall d'Hebron Research Institute, Barcelona, Spain. All women provided their written informed consent.

Results
During the study period, 124 pregnant women were consecutively invited to take part in the study, and all agreed to participate. However, four women delivered at other clinical practices and five required immediate elective delivery and were excluded from the analyses (see flowchart in Fig. 1). From the remaining 115 participants, 55 (47.8%) were diagnosed with late-onset SGA and 60 (52.2%) were diagnosed with late-onset FGR. The median GA at diagnosis was 34.4 weeks. Demographic and clinical characteristics of the study cohort are shown in Table 1 and in Tables S1, S2 and S3. Median GA at delivery was 37.2 WG (IQR 37.0-38.6) and 73 (63.5%) women required elective delivery. Further details regarding pregnancy outcomes are shown in Table 2.
Amongst the markers examined in the univariate analyses, UtAPI > 95th percentile, sFlt-1/PlGF, GA at diagnosis and abnormal fetal Doppler were significantly associated with all adverse pregnancy outcomes. In addition, EFW percentile was inversely associated with the need of delivery before 37 WG. The sFlt-1/PlGF ratio was the single marker with the greatest AUC for the prediction of all adverse outcomes. Thus, the predictive ability of sFlt-1/PlGF alone was compared to that of the two multivariate models. All variables were included in the multivariate analysis; however, those that were found to be not significant, were excluded from the model. The results of the logistic regression analyses are shown in Table 3.
Adverse perinatal outcomes occurred in 43 (37.4%) cases, with admission to the NICU and Cesarean section due to non-reassuring CTG being the most frequent APO. For the prediction of APO, statistically significant differences were found when the multivariate model with sFlt-1/PlGF was compared with the multivariate model without sFlt-1/PlGF (P = 0.024). No statistically significant differences were found between the multivariate model with sFlt1/PlGF and the model with sFlt-1/PlGF alone (P = 0.451).
The ROC curves and their AUC values (95% CI) for prediction of all adverse pregnancy outcomes are shown in Fig. 2 and Table 3, respectively. The ability of the multivariate model with sFlt-1/PlGF to predict delivery before 37 WG and APOs was not significantly greater than that of the model with sFlt-1/PlGF alone. Therefore, univariate logistic regression analyses were used to assess the risk for delivery before 37 WG and APOs by sFlt-1/PlGF alone.
The formulas to derive Y for all models can be seen in Table S4.
Detection rates (DR) for fixed 5% and 10% false positive rates (FPR) were calculated to ascertain whether small differences between the AUC could be clinically significant (Table 4). This table shows a very similar DR for the model with sFlt-1/PlGF alone and for the multivariate model including sFlt-1/PlGF; however, the multivariate model without sFlt-1/PlGF consistently showed a poorer performance than the two previous models.
One example of the application of the formulas derived in this study can be seen in Table S5. For easy use of the individual risk assessment, we created an Office Excel sheet (Microsoft, Redmond., Washington, USA) that can be readily used in any clinical setting (Supporting Information, Document S1).

Discussion
This study demonstrates that the use of sFlt-1/PlGF ratio in risk assessment for prenatal counseling in late-onset FGR and SGA pregnancies may help to predict adverse pregnancy outcomes at the time of diagnosis. Two multivariate models were constructed (one including sFlt-1/PlGF and one without sFlt-1/PlGF) and their predictive ability was compared to that of the model with sFlt-1/ PlGF alone. No significant differences were found between the multivariate model with sFlt-1/PlGF and the model with sFlt-1/PlGF alone in terms of predicting elective delivery before 37 WG and APOs. However, when comparing both multivariate models, the one without sFlt-1/PlGF had a significantly poorer performance than the one with sFlt-1/PlGF for predicting elective delivery before 37 WG and APOs. In the univariate analysis, the sFlt-1/PlGF ratio was the best marker for predicting delivery before 35 WG, with a greater AUC than that of UtAPI, although not significantly.
Recently, two algorithms with sFlt-1/PlGF have been published, showing a good performance for predicting adverse outcomes in early onset FGR [36,37]. Nevertheless, A recent publication made among 175 gestations with SGA explored the potential value of a combined predictive model including maternal risk factors, EFW centile, fetal Doppler assessment and the combination of estriol and PlGF. This group showed a detection rate of 62% for the prediction of APO with the combined model [38]. In that study, sFlt-1/PlGF was not measured, and the prediction of elective delivery was not assessed; therefore, the results cannot be compared to our study.
A previous study conducted in 198 SGA pregnancies between 30 and 40 WG concluded that PlGF and sFlt-1 values at diagnosis can predict adverse outcomes with a similar performance to that of Doppler parameters [39]. In addition, the combination of angiogenic factors and Doppler parameters did not improve the predictive ability of sFlt-1/PlGF or Doppler. In that study, for APO prediction, the ROC curves for CPR, PlGF multiples of the median (MoM) and the combination of both were 0.652, 0.656 and 0.684, respectively. However, the definition of APO in that study was different from the definition given in this study, and therefore, comparisons between AUC and DR were not possible.
Another study conducted in 62 SGA pregnancies at more than 34 WG showed that sFlt-1/PlGF > 38 at diagnosis was associated with a lower birth weight (2045 g vs 2405 g, P < 0.001). In that study a positive correlation between sFlt-1/PlGF and UtAPI was also found [40].
These studies provide evidence that the sFlt-1/PlGF ratio is a good marker for identifying SGA fetuses at a greater risk of adverse outcomes, but do not provide a model to determine a patient's specific risk for elective preterm delivery or the occurrence of APOs. Except for these studies, research published to date on predicting pregnancy outcomes is mainly focused on feto-placental Doppler findings and the progression of these findings; however, in late-onset FGR and SGA, the progression of Doppler findings is not as predictable as in early onset FGR and SGA [14]. Therefore, the uncertainty regarding pregnancy prognosis added to the fact that a larger number of ultrasonographic examinations are usually performed in FGR and SGA pregnancies, render both late-onset FGR and SGA deeply stressful pregnancy conditions for parents, as well as a healthcare burden [41].
The multivariate model with sFlt-1/PlGF and the model with sFlt-1/PlGF alone showed a similar ability to predict delivery before 37 WG; therefore, both models may potentially be used to evaluate a patient-specific risk. However, when different approaches with comparable results are available, the simplest approach should be used according to the statistical principle of parsimony [42], and thus, the use of sFlt-1/PlGF alone is preferred for individual risk assessment. The multivariate model without the sFlt-1/PlGF ratio showed a good overall performance for predicting APOs and may; therefore, be useful when the sFlt-1/PlGF ratio is not available. In addition, for predicting delivery before 35 WG, the univariate model with UtAPI may be used when the sFlt-1/PlGF ratio is not available.
The main strengths of this study are its prospective design, and the incorporation of the sFlt-1/PlGF ratio for assessing potential adverse pregnancy outcomes at the time of diagnosis in a cohort of late-onset FGR and SGA pregnancies. In addition, since sFlt-1/PlGF might not be available at diagnosis in all clinical settings, we also provide a model based on maternal characteristics and ultrasonographic markers, which allows a risk calculation without including the sFlt-1/PlGF ratio. We acknowledge some limitations of this study. First, our results could be of great value for individualized prenatal counseling; however, they could be overfitted since they were developed in a single population. In addition, we did not perform any bootstrapping procedure which would have provided stable estimates with low bias [43]. Therefore, these preliminary findings should undergo external validation in a larger cohort before being used in clinical practice. Second, sFlt-1/PlGF results were not blinded to the investigators, which may have influenced the decisionmaking process for planning follow-up or steroid administration for fetal maturation in some cases, even though management was intended to be undertaken without taking into consideration sFlt-1/PlGF results. Nevertheless, the indication for elective delivery was based strictly on EFW, Doppler assessment and CTG findings in all cases. Finally, as this was an observational study, our findings cannot be used for planning delivery or tailoring followup. These decisions should be based exclusively on current evidence-based protocols guided by EFW, CTG and Doppler findings.

Conclusions
sFlt-1/PlGF seems to be a good predictor of adverse perinatal outcomes at the time of late-onset FGR/SGA diagnosis. One formula combining maternal characteristics and ultrasonographic findings, and another including only sFlt-1/PlGF were developed. Our predictive models may be useful to provide early individualized prenatal counseling in this group of women. Further studies are needed to validate these preliminary findings in a larger cohort.