Development of an integrated predictive model for the risks of obstetric hemorrhage in multiple pregnancies

Obstetric hemorrhage remains one of the leading causes of maternal mortality worldwide, accounting for up to 28% of maternal deaths [1, 2]. Twin pregnancies are associated with an increased risk of postpartum hemorrhage, with an incidence approximately 7.6% higher than that observed in singleton pregnancies [3].

The principal pathophysiological mechanism underlying the increased risk of obstetric hemorrhage in multiple gestations is excessive uterine overdistension, which impairs myometrial contractility and predisposes the patient to uterine atony, the leading cause of obstetric hemorrhage, accounting for up to 80% of cases [4]. Excessive myometrial stretching results from a greater combined fetal mass, leading to structural and functional alterations of myofibrils and reduced contractile activity. In addition, the larger placental implantation site in twin pregnancies creates a more extensive wound surface following placental separation, potentially increasing blood loss. Gestational anemia further contributes to the risk of obstetric hemorrhage by impairing oxygen delivery to uterine myocytes, thereby compromising effective postpartum myometrial contractions [5].

Marked hemodynamic adaptations characteristic of multiple pregnancies also contribute substantially to the development of postpartum hemorrhage, including a 10–20% increase in circulating blood volume and an approximately 20% increase in cardiac output required to maintain adequate fetoplacental perfusion [6,7]. Additional factors associated with an increased risk of placental abruption include polyhydramnios, preeclampsia, gestational hypertension, and maternal comorbidities [8].

Nevertheless, obstetric hemorrhage develops in only a subset of women with multiple pregnancies, and currently available risk assessment approaches remain insufficiently accurate and reproducible. Therefore, the development of an integrated model for early individualized risk stratification of obstetric hemorrhage in twin pregnancies represents an important clinical objective aimed at reducing maternal morbidity and mortality while optimizing the management of multiple pregnancies and deliveries.

This study aimed to develop an integrated prognostic model for predicting the risk of obstetric hemorrhage in women with twin pregnancies.

Materials and methods

This retrospective cohort observational study was conducted at the at the V.I. Kulakov NMRC for OG&P, Ministry of Health of Russia, between January 2018 and January 2024. The study design and reporting adhered to the core principles of the STROBE statement and TRIPOD+AI guidelines for reporting prognostic models [9, 10]. Pregnancy course and outcomes were analyzed in 590 patients with twin gestations based on available medical records. Clinical, anamnestic, laboratory, and instrumental data were analyzed to identify factors associated with the risk of postpartum hemorrhage.

In the first stage, we evaluated the course and outcomes of twin pregnancies in 590 patients. Patients were stratified according to the occurrence of obstetric hemorrhage during delivery and the postpartum period into two groups: group 1, patients with hemorrhage ≥1000 mL (n=38), and group 2, patients without hemorrhage (n=552). The sample was formed consecutively and included all patients who met the eligibility criteria during the observation period.

Inclusion criteria were twin pregnancy and antenatal care and delivery at the V.I. Kulakov NMRC for OG&P.

Non-inclusion criteria were higher-order multiple pregnancy (triplets, quadruplets), refusal of treatment, a history of severe comorbidities, oncologic or autoimmune disease, and prior organ transplantation. Patients with placenta accreta spectrum disorders, placenta previa, or large uterine leiomyomas were also excluded.

In the second stage, statistical analysis was performed on 88 categorical and continuous variables to identify significant predictors of PPH. The analysis included key indicators of maternal health status, obstetric and medical history, and major gestational complications.

Statistical analysis

Data were analyzed using descriptive and inferential statistical methods. The normality of distribution of continuous variables was assessed with the Shapiro–Wilk test. Non-normally distributed continuous variables are presented as median and interquartile range (Me [Q1; Q3]), and categorical variables as absolute and relative frequencies (n, %).

The choice of statistical tests was guided by the distribution of the data. Between-group comparisons of continuous variables were performed using the non-parametric Mann–Whitney U test. Categorical variables were compared using the Pearson chi-square (χ²) test, or Fisher's exact test when expected cell frequencies were low. Associations between candidate factors and the risk of PPH were assessed by univariable logistic regression, with odds ratios (ORs) and 95% confidence intervals (CIs) calculated for each variable. Variables with extremely low frequency of occurrence (isolated cases) were included in the descriptive statistics but excluded from the regression analysis owing to insufficient statistical power to assess their effect. Factors that reached statistical significance in univariable analysis were entered into a multivariable binary logistic regression model to identify independent predictors of PPH ≥1000 mL, with corresponding ORs and 95% CIs. A nomogram for the personalized risk assessment of obstetric hemorrhage was subsequently developed based on the independent predictors identified in the multivariable model. The discriminative ability of the model was evaluated by receiver operating characteristic (ROC) curve analysis, with calculation of the area under the curve (AUC). A p<0.05 was considered statistically significant. All statistical analyses were performed using IBM SPSS Statistics, versions 26.0–27.0. The nomogram based on the multivariable model was constructed using RMS Package, version 7.1.7.0 (32-bit).

Results

Of 590 patients with multiple pregnancy, postpartum hemorrhage ≥1000 mL occurred in 38 (6.44%), while 552 (93.56%) had no hemorrhage. No statistically significant between-group differences were observed in clinical and demographic characteristics, including age, body mass index, gravidity, parity, or history of miscarriage or induced abortion (p>0.05) (Table 1).

Characteristics related to the course of multiple pregnancy were likewise comparable between groups, including timing of complication detection, degree of fetal weight discordance, and multiple pregnancy–specific complications (p>0.05) (Table 2).

Congenital fetal malformations were significantly more common in the hemorrhage group (18.42% vs. 13.59%; p<0.0001), as was gestational anemia (73.68% vs. 36.23%; p=0.020) (Table 2). Hemoglobin levels before delivery were lower in the hemorrhage group (115.0 g/L vs. 118.0 g/L; p=0.016) (Fig. 1), as was platelet count (179.0×10⁹/L vs. 212.5×10⁹/L; p=0.001) (Fig. 1, Table 3). Thrombocytopenia (platelet count <150×10⁹/L) was identified in 62/590 patients (10.5%). Prophylactic administration of uterotonic agents (oxytocin, carbetocin) – given at the end of the second stage of labor for vaginal delivery or immediately after fetal extraction for cesarean delivery – was associated with a reduced risk of postpartum hemorrhage (p=0.011). Hemorrhage was more frequent among patients who underwent cesarean delivery (81.57% vs. 71.92%; p=0.035) (Table 2). Chorioamnionitis was diagnosed in 2 cases in group 2 and showed no statistically significant correlation with hemorrhage.

Univariate binary logistic regression analysis showed that gestational anemia (OR 2.352; 95% CI 1.12–4.94; p=0.024) and cesarean delivery (OR 2.62; 95% CI 1.13–6.06; p=0.025) were significantly associated with an increased risk of postpartum hemorrhage (Table 4). Conversely, a pre-delivery hemoglobin level above 115 g/L (OR 0.965; 95% CI 0.940–0.991; p=0.008), a platelet count above 179.0×10⁹/L (OR 0.988; 95% CI 0.982–0.994; p<0.0001), and prophylactic uterotonic administration (OR 0.435; 95% CI 0.225–0.844; p=0.014) were associated with a reduced risk of postpartum hemorrhage (Table 4).

Multivariate binary logistic regression was performed to identify independent predictors of postpartum hemorrhage (Table 5). The dependent variable was a binary outcome: presence or absence of obstetric hemorrhage. Predictors demonstrating the strongest statistically significant associations with the outcome in univariate analysis were included in the model. Prophylactic uterotonic therapy showed a trend toward a protective effect that did not reach statistical significance (OR 0.481; 95% CI 0.220–1.052; p=0.067). Cesarean delivery likewise was not an independent predictor (OR 1.752; 95% CI 0.666–4.612; p=0.256).

Multivariate logistic regression identified gestational anemia (OR 3.12; 95% CI 1.39–6.99; p=0.006), pre-delivery hemoglobin below 115 g/L (OR 0.957; 95% CI 0.931–0.985; p=0.002), and platelet count below 179.0×10⁹/L (OR 0.987; 95% CI 0.981–0.993; p<0.0001) as independent, statistically significant predictors of postpartum hemorrhage.

ROC analysis was performed to assess the predictive performance of the multivariate model (Fig. 2). The area under the curve (AUC) was 0.779 (p<0.0001), indicating good discriminative ability. The model's statistical significance was high (p<0.0001), and the 95% CI for the AUC (0.696–0.862) confirmed the robustness of this discriminative estimate.

The optimal cut-off value for clinical application, 0.06587, was selected based on the maximum Youden index, representing the best balance between sensitivity and specificity. At this threshold, sensitivity was 75.7% – correctly identifying 75.7% of patients with the outcome –and specificity was 72.8% – correctly identifying 72.8% of patients without the outcome.

Accordingly, patients with a predicted probability ≥0.06587 can be classified as high risk, while those with a probability <0.06587 can be classified as low risk.

A nomogram (Fig. 3) was constructed from the multivariate logistic regression model to provide individualized prediction of hemorrhage risk. It incorporates the following predictors: anemia (yes/no), hemoglobin level, and platelet count.

The nomogram is a graphical tool for estimating the risk of a binary outcome based on the combination of clinical and laboratory variables included in the logistic regression model. Each predictor value is located on its corresponding scale, and a vertical line is projected from this value onto the "Points" scale to obtain a corresponding point value. The points for all variables in the model are then summed, and the total is used to derive the individual predicted probability of the outcome from the bottom integrated probability scale.

The nomogram thus provides a visual representation of each predictor's contribution to the overall probability of postpartum hemorrhage and can be applied to quantify individual risk in both clinical practice and research settings.

Discussion

Our findings confirm that obstetric hemorrhage in twin pregnancies remains a clinically significant challenge. In the present study, the incidence of hemorrhage ≥1,000 mL was 6.44%, consistent with previous reports on twin pregnancies [3,11], thereby supporting the representativeness of the study cohort. The principal finding of this study was the identification of gestational anemia, hemoglobin concentration, and platelet count as independent predictors of obstetric hemorrhage. These findings are consistent with the recommendations of the International Federation of Gynecology and Obstetrics (FIGO) and the Royal College of Obstetricians and Gynecologists (RCOG) regarding the diagnosis and management of postpartum hemorrhage [12,13], as well as with evidence from large-scale clinical studies.

The WOMAN-2 trial [14], which included 10,620 pregnant women, reported postpartum hemorrhage rates of 6.2% among women with moderate anemia and 11.2% among those with severe anemia. Furthermore, each 10 g/L decrease in pre-delivery hemoglobin concentration was associated with an increased risk of postpartum hemorrhage (adjusted OR 1.29 [95% CI 1.21–1.38]). Similarly, a large population-based study by Al-Zirqi I. et al. [15], which included 307,415 pregnancies, demonstrated that a hemoglobin concentration of <90 g/L was associated with a twofold increase in the risk of severe obstetric hemorrhage.

Several pathophysiological mechanisms may explain this association between obesity and cancer. First, anemia is accompanied by a hyperdynamic circulatory state characterized by an increased heart rate and cardiac output, which may exacerbate blood loss from injured vessels [16]. The activation of hypoxia-sensitive receptors stimulates the sympathetic nervous system to maintain adequate tissue oxygen delivery [17]. Second, reduced hematocrit decreases blood viscosity, thereby increasing the blood flow velocity [18]. Third, anemia alters clot structure and stability, as reduced erythrocyte content is associated with increased susceptibility of fibrin clots to fibrinolysis [19]. Finally, anemia may contribute to uterine atony through impaired myometrial oxygenation [2,8].

In twin pregnancies, these mechanisms are further amplified by the greater circulating blood volume, increased uterine blood flow, more pronounced hemodilution, and hyperdynamic circulatory state. Reduced blood viscosity, altered hemorheological properties, and hemostasis disturbances collectively increase the risk of postpartum hemorrhage [20]. Another important finding of the present study was the identification of reduced platelet count as an independent predictor of postpartum hemorrhage. Even subclinical thrombocytopenia can impair primary hemostasis by reducing the efficiency of platelet plug formation, thereby increasing blood loss.

Notably, the multivariable logistic regression model showed no independent association between cesarean delivery, uterotonic administration, and postpartum hemorrhage. This finding most likely reflects the confounding effect of indications. Carbetocin is routinely administered prophylactically to women considered to be at high baseline risk of postpartum hemorrhage and is therefore used more frequently in clinically complex cases. Consequently, the higher frequency of uterotonic use among women with adverse outcomes should not be interpreted as evidence of a causal relationship but rather as reflecting clinical treatment selection and the inherent limitations of retrospective observational studies.

The integrated prediction model based on the identified independent predictors demonstrated good discriminatory performance (AUC=0.779; p<0.0001; 95% CI 0.696–0.862), indicating its potential clinical utility.

The practical value of this study lies in the development of a nomogram that enables individualized risk assessment in routine clinical practice. Because the model relies on a limited number of routinely measured objective laboratory parameters, its implementation in clinical practice is likely to be feasible.

The limitations of this study include its retrospective design, absence of external validation, limited availability of several potentially important clinical predictors, and possibility of omitted variable bias. Therefore, the proposed model requires further validation in independent patient cohorts.

Future research should focus on prospective multicenter external validation of the model and the expansion of the predictor set to include hemostatic biomarkers and parameters reflecting myometrial functional status. Another promising direction is the integration of the model into clinical information systems with automated real-time risk calculations. Overall, our findings suggest that a personalized approach to risk stratification for obstetric hemorrhage in twin pregnancies, based on objective laboratory parameters, improves predictive accuracy and has the potential to optimize clinical management while reducing the risk of severe maternal complications.

Conclusion

The findings of the present study highlight the multifactorial pathophysiology of obstetric hemorrhage in twin pregnancies. The development of a prognostic nomogram based on the most informative predictors provides a promising tool for individualized risk prediction with important clinical implications for optimizing the management of high-risk pregnancies, improving maternal outcomes, and enhancing cost-effectiveness of healthcare delivery.