Related Experiment Video
Updated: Oct 3, 2025
The 4-vessel Sampling Approach to Integrative Studies of Human Placental Physiology In Vivo
Published on: August 2, 2017
Systematic external evaluation of four preoperative risk prediction models for severe postpartum hemorrhage in
Xiaohe Dang1, Guoping Xiong2, Cuifang Fan3
1Department of Obstetrics and Gynecology, Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology, Wuhan 430030, China.
Aim:
To compare and evaluate the validity of the existing risk prediction models for severe postpartum hemorrhage (SPPH) in patients with placenta previa.
Methods:
We conducted a systematic literature review to collect the existing risk prediction models for SPPH in patients with placenta previa, and recruited patients with placenta previa who underwent cesarean section in Tongji Hospital (Wuhan, China) and 4 cooperative hospitals from January 2018 to June 2021. We defined SPPH as total blood loss ≥1500 mL or transfusion packed red blood cell ≥4 U. The risk of SPPH of each patient was predicted by the collected models, respectively. Then we calculated the sensitivity, specificity, coincidence rate (CCR), positive predictive value (PPV), negative predictive value (NPV) and drawn the receiver operating characteristic (ROC) curve and decision curve analysis (DCA) curve of each model.
Results:
This external cohort contained 1172 patients of whom 284 patients (24.23%) experienced SPPH, and 4 risk prediction models were collected in this study. After evaluated by this external cohort, the area under the ROC curve (AUC), sensitivity, specificity, CCR, PPV and NPV of the four models ranged from 0.644 to 0.755, 38.38% to 86.31%, 42.75% to 86.49%, 56.23% to 74.83%, 38.68% to 47.60%, 81.15% to 87.45%, respectively. The model established by Kim JW et al. had the highest sensitivity, NPV, AUC and net benefit, the model established by Lee JY et al. had the highest specificity, CCR and PPV.
Conclusions:
The four prediction models showed moderate predictive performance, the discrimination indicators and benefit indicators of each model were not simultaneously ideal in this population. The prediction models should be further optimized to improve the discrimination ability and benefit, and prospective external validation studies should also be carried out before they are applied to clinical practice.
