First-trimester prediction of pre-eclampsia: external validity of algorithms in a prospectively enrolled cohort.

OBJECTIVE To evaluate the performance of published first-trimester prediction algorithms for pre-eclampsia (PE) in a prospectively enrolled cohort of women. METHOD A MEDLINE search identified first-trimester screening-prediction algorithms for early-onset (requiring delivery < 34 weeks) and late-onset (requiring delivery ≥ 34 weeks) PE. Maternal variables, ultrasound parameters and biomarkers were determined prospectively in singleton pregnancies enrolled between 9 and 14 weeks. Prediction algorithms were applied to this population to calculate predicted probabilities for PE. The performance of the prediction algorithms was compared with that in the original publication and evaluated for factors explaining differences in prediction. RESULTS Six early and two late PE prediction algorithms were applicable to 871-2962 women, depending on the variables required. The prevalence of early PE was 1.0-1.2% and of late PE was 4.1-5.0% in these patient subsets. One early PE prediction algorithm performed better than in the original publication (80% detection rate (DR) of early PE for 10% false-positive rate (FPR)); the remaining five prediction algorithms underperformed (29-53% DR). Prediction algorithms for late PE also underperformed (18-31% DR, 10% FPR). Applying the screening cut-offs based on the highest Youden index probability scores correctly detected 40-80% of women developing early PE and 71-82% who developed late PE. Exclusion of patients on first-trimester aspirin resulted in DRs of 40-83% and 65-82% for early and late PE, respectively. CONCLUSION First-trimester prediction algorithms for PE share a high negative predictive value if applied to an external population but underperform in their ability to correctly identify women who develop PE. Further research is required to determine the factors responsible for the suboptimal external validity.