To explore and construct a risk prediction model for refractory postpartum hemorrhage in postpartum women and to preliminarily evaluate its predictive efficacy.

A total of 500 postpartum patients who met the inclusion and exclusion criteria and were admitted to Linhai Maternal and Child Health Hospital from January 1, 2024 to January 1, 2025 were selected as the study subjects, aged 20~42 years [(30.71±5.02) years]. They were randomly divided into a training group (n=350) and a validation group (n=150) in a 7∶3 ratio. The incidence of refractory postpartum hemorrhage in the training group was assessed, and patients were divided into a refractory postpartum hemorrhage group (n=32) and a non-refractory postpartum hemorrhage group (n=318). Risk factors for refractory postpartum hemorrhage were identified, and a Nomogram prediction model was constructed. Receiver operating characteristic (ROC) curves were used to measure the predictive accuracy of the model, and clinical decision curve analysis (DCA) was used to evaluate the clinical utility of the model.

There were statistically significant differences between the two groups in history of scarred uterus, placenta previa, placental abruption, placental adhesion/invasion/penetration, coagulation dysfunction, uterine atony, prenatal serum coagulation factor Ⅶ (FⅦ) and fibrinogen (FIB) (P<0.05). Multivariate logistic regression analysis showed that history of scarred uterus, placenta previa, placental abruption, placental adhesion/implantation/penetration, coagulation dysfunction, uterine atony, prenatal FVII≥161 U, and prenatal FIB≥3.0 g/L were independent risk factors for refractory postpartum hemorrhage in women who had given birth (P<0.05). The area under the curve (AUC) of the nomogram model was 0.862 (95%CI: 0.833-0.925) in the training group and 0.834 (95%CI: 0.754-0.887) in the validation group. ROC curve analysis determined an optimal risk probability cutoff of 28%. At this threshold, the sensitivity and specificity were 81.17% and 78.64% in the training group, and 75.33% and 72.54% in the validation group, respectively. DCA analysis showed that when the threshold probability set by clinical decision-makers was 28%, the model had a higher net benefit than the full-intervention or no-intervention strategy, suggesting that it has certain clinical decision-making value.

The occurrence of refractory postpartum hemorrhage in postpartum women is associated with factors such as history of scarred uterus, placenta previa, placental abruption, placental adhesion/implantation/penetration, coagulation dysfunction, uterine atony, FVII, and FIB. Constructing a predictive model can provide a reference for subsequent risk assessment and intervention studies of high-risk groups.