A risk model to predict severe postpartum hemorrhage in patients with placenta previa: a single-center retrospective study

Cheng Chen, Xiaoyan Liu, Dan Chen, Song Huang, Xiaoli Yan, Heying Liu, Qing Chang, Zhiqing Liang


Background: The study aimed to establish a predictive risk model for severe postpartum hemorrhage in placenta previa using clinical and placental ultrasound imaging performed prior to delivery.
Methods: Postpartum hemorrhage patients were retrospectively enrolled. Severe postpartum hemorrhage was defined as exceeding 1,500 mL. Data collected included clinical and placental ultrasound images.
Results: Age of pregnancy, time of delivery, time of miscarriage, history of vaginal delivery, gestational weeks at pregnancy termination, depth of placenta invading the uterine muscle wall were independent risk factors for severe postpartum hemorrhage in placenta previa. A model to predict severe postpartum hemorrhage in placenta previa was established: P=Log(Y/1−Y), where Y =−6.942 + 0.075 X1 (age) +1.531 X2 (times of delivery) + 0.223 X3 (time of miscarriage) − 3.557X4 (vaginal delivery: 1 for yes, 0 for no) + 1.753 X5 (0 for <37 weeks, 1 for ≥37 weeks) + 1.574 X6 (Depth of placenta invading uterine muscle wall: 0 for normal, 1 for placenta adhesion, 2 for placenta implantation, 3 for placenta penetration); discriminant boundary value of the prediction model (probability: P) was 0.268. Predicting sensitivity (Se) =0.765 (negative predicting accuracy rate), specificity (Sp)=0.900 (positive predicting accuracy rate), total accuracy rate =0.795, and AUC of ROC curve =0.8419.
Conclusions: The risk prediction model which had clinical and ultrasound imaging information prior to delivery had a high decision accuracy. However, before it can be used in the clinic, multicenter large-sample clinical studies should be performed to verify its accuracy and reliability.