Competing‐risks model for prediction of small‐for‐gestational‐age neonate from maternal characteristics, serum pregnancy‐associated plasma protein‐A and placental growth factor at 11–13 weeks' gestation

Abstract

Objectives To expand a new competing‐risks model for prediction of a small‐for‐gestational‐age (SGA) neonate, by the addition of pregnancy‐associated plasma protein‐A (PAPP‐A) and placental growth factor (PlGF), and to evaluate and compare PAPP‐A and PlGF in predicting SGA. Methods This was a prospective observational study of 60 875 women with singleton pregnancy undergoing routine ultrasound examination at 11 + 0 to 13 + 6 weeks' gestation. We fitted a folded‐plane regression model for the PAPP‐A and PlGF likelihoods. A previously developed maternal history model and the likelihood models were combined, according to Bayes' theorem, to obtain individualized distributions for gestational age (GA) at delivery and birth‐weight Z‐score. We assessed the discrimination and calibration of the model. McNemar's test was used to compare the detection rates for SGA with, without or independently of pre‐eclampsia (PE) occurrence, of different combinations of maternal history, PAPP‐A and PlGF, for a fixed false‐positive rate. Results The distributions of PAPP‐A and PlGF depend on both GA at delivery and birth‐weight Z‐score, in the same continuous likelihood, according to a folded‐plane regression model. The new approach offers the capability for risk computation for any desired birth‐weight Z‐score and GA at delivery cut‐off. PlGF was consistently and significantly better than PAPP‐A in predicting SGA delivered before 37 weeks, especially in cases with co‐existence of PE. PAPP‐A had similar performance to PlGF for the prediction of SGA without PE. At a fixed false‐positive rate of 10%, the combination of maternal history, PlGF and PAPP‐A predicted 33.8%, 43.8% and 48.4% of all cases of a SGA neonate with birth weight < 10th percentile delivered at ≥ 37, < 37 and < 32 weeks' gestation, respectively. The respective values for birth weight < 3rd percentile were 38.6%, 48.7% and 51.0%. The new model performed well in terms of risk calibration. Conclusions The combination of PAPP‐A and PlGF values with maternal characteristics, according to Bayes' theorem, improves prediction of SGA. PlGF is a better predictor of SGA than PAPP‐A, especially when PE is present. The new competing‐risks model for SGA can be tailored to each pregnancy and to the relevant clinical requirements. © 2020 International Society of Ultrasound in Obstetrics and Gynecology

Competing‐risks model for prediction of small‐for‐gestational‐age neonate from estimated fetal weight at 19–24 weeks' gestation

“…The EFW likelihood updates the prior distribution of birth‐weight Z ‐score and GA at delivery. In the high‐risk cases, the joint distribution is shifted towards earlier GAs and lower birth weights, resulting in a higher risk for SGA, as we have demonstrated previously 11–14 .…”

“…In the competing-risks model for prediction of SGA, the performance of screening by maternal characteristics and medical history is improved by the addition of second-trimester EFW. This study provides further evidence that SGA is a spectrum disorder [11][12][13][14] . The Z-score of EFW has a continuous association with Z-score of birth weight and GA at delivery; EFW and birth weight are correlated linearly, and this association becomes steeper for earlier GAs.…”

“…We have proposed recently a new competing-risks model for the prediction of SGA [11][12][13][14] . This new approach is based on the concept that SGA is a two-dimensional spectrum disorder whose severity is reflected continuously in both gestational age (GA) at delivery and Z-score of birth weight for GA.…”

“…The first step was a maternal history model that defined a patient-specific joint distribution of Z-scores of birth weight and GA at delivery 11 . The second step was the addition of the first-trimester biomarkers' multivariate likelihood according to Bayes' theorem [12][13][14] . The model enables us to compute risks for any chosen cut-off.…”

Machine learning improves early prediction of small‐for‐gestational‐age births and reveals nuchal fold thickness as unexpected predictor

“…In terms of predicting SGA, there have been studies that used ML models. Features used for prediction in these past studies include ultrasound biometrics measurements, 15 , 16 umbilical Doppler blood flow, 17 , 18 pregnancy risk factors, sociodemographic, maternal characteristic and medical history, 19 , 20 pregnancy associated plasma protein A and placental growth factor 21 …”