Predictive performance of the competing risk model in screening for preeclampsia

Background The established method of screening for preeclampsia (PE) is to identify risk factors from maternal demographic characteristics and medical history; in the presence of such factors the patient is classified as high-risk and in their absence as low-risk. However, the performance of such approach is poor. We developed a competing risks model which allows combination of maternal factors (age, weight, height, race, parity, personal and family history of PE, chronic hypertension, diabetes mellitus, systemic lupus erythematosus or antiphospholipid syndrome, method of conception and interpregnancy interval), with biomarkers to estimate the individual patient-specific risks of PE requiring delivery before any specified gestation. The performance of this approach is by far superior to that of the risk scoring systems. Objective To examine the predictive performance of the competing risks model in screening for PE by a combination of maternal factors, mean arterial pressure (MAP), uterine artery pulsatility index (PI), and serum placental growth factor (PLGF), referred to as the triple test, in a training dataset for development of the model and two validation studies. Study design The data for this study were derived from three previously reported prospective non-intervention multicenter screening studies for PE in singleton pregnancies at 11+0 – 13+6 weeks’ gestation. In all three studies, there was recording of maternal factors and biomarkers and ascertainment of outcome by appropriately trained personnel. The first study of 35,948 women, which was carried out between February 2010 and July 2014, was used to develop the competing risks model for prediction of PE and is therefore considered to be the training set. The two validation studies comprised of 8,775 and 16,451 women, respectively and they were carried out between February and September 2015 and between April and December 2016, respectively. Patient-specific risks of delivery with PE at 0.95, >0.90 and >0.80, respectively, demonstrating a very high discrimination between affected and unaffected pregnancies. Similarly, the calibration slopes were very close to 1.0 demonstrating a good agreement between the predicted risks and observed incidence of PE. In the prediction of early-PE and preterm-PE the observed incidence in the training set and one of the validation datasets was consistent with the predicted one. In the other validation dataset, which was specifically designed for evaluation of the model, the incidence was higher than predicted presumably because of better ascertainment of outcome. The incidence of all-PE was lower than predicted in all three datasets because at term many pregnancies deliver for reasons other than PE and therefore pregnancies considered to be at high-risk for PE that deliver for other reasons before they develop PE can be wrongly considered to be false positives. Conclusions The competing risks model provides an effective and reproducible method for first-trimester prediction of early-PE and preterm-PE, as long as the various components of screening are carried out by appropriately trained and audited practitioners. Early prediction of preterm-PE is beneficial because treatment of the high-risk group with aspirin is highly effective in the prevention of the disease.