Validation of competing‐risks model in screening for pre‐eclampsia in twin pregnancy by maternal factors

Abstract

Objective To examine the predictive performance of the competing‐risks model in screening for pre‐eclampsia (PE) by maternal demographic characteristics and medical history in twin pregnancy, in a training dataset used for development of the model and a validation dataset. Methods The data for this study were derived from two prospective non‐intervention multicenter screening studies for PE in twin pregnancies at 11 + 0 to 13 + 6 weeks' gestation. The first study of 2219 women, which was reported previously, was used to develop the competing‐risks model for prediction of PE and is therefore considered to be the training set. The validation study comprised 2999 women. Patient‐specific risks of delivery with PE at < 34 (early), < 37 (preterm) and < 41 + 3 (all) weeks' gestation were calculated using the competing‐risks model and the performance of screening for PE in the training and validation datasets was assessed. We examined the predictive performance of the model by, first, its ability to discriminate between the PE and no‐PE groups using the area under the receiver–operating characteristics curve (AUC) and, second, calibration, which assesses agreement between the predicted risk and observed incidence of PE. Results The incidence of early PE, preterm PE and all PE in the training and validation datasets was similar (1.8% vs 1.4%, 5.6% vs 5.6% and 7.7% vs 7.2%, respectively) and this was substantially higher than in our previous studies in singleton pregnancies. The training and validation datasets had similar AUCs for early PE (0.670 (95% CI, 0.593–0.747) vs 0.677 (95% CI, 0.594–0.760)), preterm PE (0.666 (95% CI, (0.617–0.715) vs 0.652 (95% CI, 0.609–0.694)) and all PE (0.656 (95% CI, 0.615–0.697) vs 0.644 (95% CI, 0.606–0.682)). Calibration plots of the predictive performance of the competing‐risks model demonstrated that, in both the training and validation datasets, the observed incidence of PE was lower than the predicted one and such overestimation of risk was particularly marked for early PE. Conclusions Discrimination and calibration of the competing‐risks model for PE in a validation dataset are consistent with those in the training dataset. However, the model needs to be adjusted to correct the observed overestimation of risk for early PE. Copyright © 2019 ISUOG. Published by John Wiley & Sons Ltd.

Revised competing‐risks model in screening for pre‐eclampsia in twin pregnancy by maternal characteristics and medical history

“…In a previous study, we evaluated the predictive performance for PE of the competing‐risks model in singleton pregnancies using two validation datasets and demonstrated very good discrimination between affected and unaffected pregnancies and good agreement between predicted risk and observed incidence of PE. In contrast, a validation study of our competing‐risks model for twin pregnancies found that, in both the training and validation datasets, the observed incidence of PE was lower than the predicted one, especially for early PE. In this study, we developed a new model and demonstrated good agreement between predicted risk and observed incidence of PE < 34, < 37 and < 39 weeks' gestation.…”

“…In a study of 2219 twin pregnancies, we proposed that the same competing‐risks model developed in singleton pregnancies can be adapted for use in twins; in dichorionic (DC) and monochorionic (MC) twin pregnancies with the same characteristics as in singleton pregnancies, the distribution of gestational age at delivery with PE was shifted to the left by 8 and 10 weeks, respectively. In a subsequent validation study involving 2999 twin pregnancies, we found that the predictive performance for PE was consistent with that in the training set used for development of the model; however, calibration plots of the predictive performance of the competing‐risks model demonstrated that, in both the training and validation datasets, the observed incidence of PE was lower than the predicted one and such overestimation of risk was particularly marked for early PE. This suggested the need for a model in which the effect of twins relative to singletons in decreasing the gestational age at delivery with PE should increase with gestational age.…”

“…First, 2219 twin pregnancies (training dataset) that were examined at King's College Hospital and Medway Maritime Hospital, UK, between January 2006 and December 2015. Second, 2999 twin pregnancies (validation dataset) that were examined at five hospitals in England (King's College Hospital and Medway Maritime Hospital, between December 2015 and April 2018; Homerton University Hospital, between January 2014 and April 2018; North Middlesex University Hospital, between May 2015 and April 2018; and Southend University Hospital, between June 2015 and April 2018), one hospital in Bulgaria (Dr. Shterev Hospital in Sofia, between January 2013 and April 2018) and one hospital in Spain (Hospital Clínico Universitario Virgen de la Arrixaca in Murcia, between March 2009 and April 2018). Third, the validation dataset of 16 747 singleton pregnancies from the Screening ProgRamme for prE‐Eclampsia (SPREE) study; this was a prospective multicenter study in seven National Health Service (NHS) maternity hospitals in England.…”

“…Three datasets were used for this study. 6 . Third, the validation dataset of 16 747 singleton pregnancies from the Screening ProgRamme for prE-Eclampsia (SPREE) study; this was a prospective multicenter study in seven National Health Service (NHS) maternity hospitals in England 7 .…”

“…The objectives of this study were, first, to develop a new extension of the competing‐risks prior model in screening for PE in twin pregnancies in the original training dataset, second, to examine the predictive performance of this model in screening for PE with delivery < 34 weeks (early PE), < 37 weeks (preterm PE) and at any gestational age (all PE) in twins in a validation dataset, and, third, to demonstrate the application of screening in a mixed population of singleton and twin pregnancies.…”