WEDNESDAY, Feb. 12, 2025 (HealthDay News) — For the Preeclampsia Integrated Estimate of RiSk (PIERS) Machine Learning (PIERS-ML) model and the logistic regression-based fullPIERS model, consecutive performance deteriorates over time, according to a study published online Feb. 4 in PLOS Medicine.

Guiyou Yang, from the University of Groningen in the Netherlands, and colleagues examined the performance of two models for predicting the risk for adverse maternal outcomes beyond the first 48 hours following admission for women whose pregnancies are complicated by preeclampsia. Data were included for 8,843 women. The risk differentiation performance of the PIERS-ML model and the fullPIERS model was assessed within a two-week postadmission window. The main outcome measures included trajectories of mean risk of each of the uncomplicated course and adverse outcome groups and daily area under the precision-recall curve (AUC-PRC).

The researchers found that the number of daily outcome events decreased from more than 200 to around 10 in the two-week window. Mean risk was consistently higher in the adverse outcome group versus the uncomplicated course group for both PIERS-ML and fullPIERS models. The AUC-PRC values of the fullPIERS model remained low (0.2 to 0.4), indicating low discriminative capacity. The AUC-PRC for the PIERS-ML model peaked at 0.65 on day 0 and decreased thereafter. The PIERS-ML model generally showed good rule-in capacity for the very high-risk group, with positive likelihood ratios ranging from 70.99 to infinity; rule-out capacity for the very low-risk group was also good, where most negative likelihood ratio values were 0. For other risk groups, performance declined beyond 48 hours.

“Clinicians should interpret consecutive predictions with increasing uncertainty,” the authors write.

Several authors have an intellectual property claim regarding the PIERS-AI machine learning-based suite of models.

Abstract/Full Text

Copyright © 2025 HealthDay. All rights reserved.