Risk factors for postpartum hemorrhage after pregnancy termination for fetal malformations: a retrospective study and internally validated prediction model

Abstract

Purpose

This study aims to elucidate the risk factors contributing to postpartum hemorrhage (PPH) in women undergoing pregnancy termination due to fetal malformations and to construct a predictive model.

Methods

A retrospective study of 268 women after termination for fetal malformations was conducted. Logistic regression identified risk factors for postpartum hemorrhage, and a nomogram was constructed. Model discrimination, calibration, and goodness of fit were evaluated using the AUC, Hosmer–Lemeshow test, and 1,000 bootstrap resampling iterations.

Results

Gravidity, placental adhesion/retention, history of uterine surgery, pregnancy-induced hypertension, and uterine atony were identified as independent risk factors for PPH in women undergoing termination for fetal malformations (P < 0.05). The predictive model demonstrated an overall accuracy of 89.55% (95% CI 85.31–92.67%), with a sensitivity of 37.14% (95% CI 23.17–53.66%), specificity of 97.42% (95% CI 94.50–98.81%), positive predictive value of 68.42% (95% CI 46.01–84.64%), and negative predictive value of 91.16% (95% CI 86.99–94.09%). The discriminative ability of the model was evaluated using the ROC curve, yielding an AUC of 0.8445 (95% CI 0.7731–0.9158, P < 0.0001).

Conclusion

Our proposed model may help identify a subset of women at increased risk of PPH after termination for fetal malformations and support clinical preparedness following internal validation. Further validation in independent cohorts is needed before broader clinical application.

What does this study adds to the clinical work?

This study identifies gravidity, placental adhesion/retention, history of uterine surgery, pregnancy-induced hypertension, and uterine atony as risk factors associated with postpartum hemorrhage following termination of pregnancy for fetal malformations. Although the developed model may help identify women at increased risk and support clinical preparedness during induction and intrapartum management, external validation is still needed before routine clinical use.

Introduction

With the ongoing advancement of obstetric diagnostic technologies, the incidence of pregnancy terminations due to fetal malformations has shown a year-on-year increase [1, 2]. Epidemiological data indicate that the global prevalence of fetal malformations is approximately 3% [3], while in China, the reported prevalence ranges from 4 to 6% [4]. As most congenital anomalies are detected during the prenatal period, a substantial proportion of affected pregnancies are electively terminated after clinical counseling and parental decision-making [5].

Termination of pregnancy due to fetal malformations is an important strategy to reduce adverse perinatal outcomes; however, the incidence of postpartum hemorrhage (PPH) in this population remains high and poses a significant threat to maternal health. In a 10-year cohort of 407 medical terminations for fetal anomaly at ≥ 20 weeks’ gestation, postpartum hemorrhage occurred in 45 of 407 women (11.1%), with manual removal of the placenta and PPH being the most common maternal morbidities [6]. A systematic review and meta-analysis of PPH in women with placenta previa reported an average incidence of 22.3% overall among placenta previa cases, with variation by subgroup (e.g., 27.4% in true placenta previa) [7].

PPH is a severe obstetric complication and one of the leading causes of maternal mortality worldwide [8]. While numerous studies have investigated the risk factors associated with PPH, most existing predictive models have been developed from term deliveries or mixed obstetric populations [9]. However, termination for fetal malformations often involves different clinical circumstances, including a higher proportion of mid- to late-gestation pregnancies, prolonged induction processes, and a higher frequency of placental separation problems. Therefore, the usefulness of term-based models cannot be assumed in this subgroup, supporting the need for a dedicated model. This study aims to address this gap by developing a predictive tool to facilitate individualized clinical preparedness and risk-based management for this high-risk population.

Methods

Study design and participants

This retrospective study included 268 women who underwent vaginal delivery following pregnancy termination for fetal malformations at our institution between June 2021 and December 2024. No identifiable personal information, such as names or addresses, was collected during the study, thereby protecting the patients’ privacy and anonymity. All data were aggregated and reported in summary form only.

Participants were included if they met the following criteria: ① confirmed diagnosis of fetal malformations through prenatal assessment and underwent termination procedures at our hospital; ② age ≥ 20 years with adequate literacy, comprehension, and communication abilities; and ③ voluntary participation with written informed consent. The exclusion criteria were as follows: ① electively terminated after clinical counseling and parental decision-making, and ② incomplete clinical data.

Data collection

Basic clinical information was collected from the electronic medical records of women who underwent vaginal delivery following pregnancy termination for fetal malformations. The collected variables included maternal age, gravidity, parity, history of uterine surgery, history of abortion, placental status, pregnancy-related complications and comorbidities, and uterine atony. According to the timing of clinical ascertainment, gravidity, parity, history of uterine surgery, history of abortion, and pregnancy-induced hypertension were determined from the obstetric history or medical records before or at the start of induction. Placental adhesion/retention and uterine atony were recorded when identified during induction, delivery, or immediate postpartum management. Placental status included placental adhesion, placenta previa, and retained placenta, while pregnancy-related complications and comorbidities included multiple gestation, gestational diabetes mellitus, pregnancy-induced hypertension, and anemia during pregnancy. During the study period, termination management and routine PPH prevention measures were based on our institutional protocol, including routine uterotonic use and close monitoring of bleeding during and after delivery, with escalation according to clinical need, gestational age, cervical status, and clinician judgment.

Among the 268 cases, 35 women who experienced PPH were classified into the hemorrhage group, while the remaining 233 women without PPH comprised the non-hemorrhage group. Individuals in the hemorrhage group met the diagnostic criteria outlined in the 2023 edition of the Guidelines for the Prevention and Management of Postpartum Hemorrhage issued by the Obstetrics and Gynecology Branch of the Chinese Medical Association. As all women included in this study underwent vaginal delivery following pregnancy termination, PPH was defined as blood loss ≥ 500 mL within 24 h after delivery or the presence of signs and symptoms of hypovolemia, regardless of measured blood loss [10]. Intrapartum blood loss was assessed according to the same institutional measurement procedure throughout the study period. Intrapartum blood loss was measured using a volumetric method, including blood collected in calibrated containers and visible blood loss recorded during delivery. Postpartum blood loss within 24 h was calculated using the gravimetric method, with blood-soaked pads or materials weighed and compared with their dry weight according to the formula: blood loss volume = (wet weight − dry weight) × 1.05. Total blood loss was defined as the sum of intrapartum blood loss and postpartum blood loss within 24 h after delivery. The classification of PPH was based on this total measured blood loss and the presence of clinical signs or symptoms of hypovolemia when applicable.

Statistical analysis

Statistical analysis and data visualization were performed using GraphPad Prism version 9.5.0 and R software version 4.3.3. Univariate analysis was conducted for clinically relevant variables, and categorical variables were compared using the chi-square (χ²) test or Fisher’s exact test when cell counts were small. The results are expressed as frequencies and percentages [n (%)]. Multivariable analysis was conducted using binary logistic regression. Variables that remained independently associated with PPH were refitted into a reduced model for nomogram construction. Model calibration was assessed using the Hosmer–Lemeshow goodness-of-fit test, while model discrimination was evaluated by calculating the area under the ROC curve (AUC). Internal validation of the nomogram’s predictive performance was performed using 1000 bootstrap resampling iterations. A p-value < 0.05 was considered statistically significant.

Results

A total of 268 women who underwent pregnancy termination for fetal malformations were included in this study, among whom 35 experienced PPH, yielding an incidence rate of 13.06%. Univariate analysis indicated that maternal age, gravidity, history of abortion, placental adhesion/retention, history of uterine surgery, pregnancy-induced hypertension, and uterine atony were significantly associated with the occurrence of PPH (p < 0.05). In contrast, parity, multiple gestation, pregnancy-related anemia, and gestational diabetes mellitus were not significantly associated with PPH (p > 0.05) (Table 1).

Table 1.

Univariate Analysis of Factors Influencing PPH in Women Undergoing Pregnancy Termination for Fetal Malformations

Variable	Coding	Hemorrhage Group (n = 35)	Non-Hemorrhage Group (n = 233)	χ2	P
Age				4.477	0.049
< 35 years	0	29 (82.9%)	203 (87.1%)
≥ 35 years	1	6 (17.1%)	30 (12.9%)
Parity				0.058	0.811
< 2	0	31 (88.6%)	203 (87.1%)
≥ 2	1	4 (11.4%)	30 (12.9%)
Gravidity				25.85	< 0.001
< 3	0	10 (28.6%)	168 (72.1%)
≥ 3	1	25 (71.4%)	65 (27.9%)
History of abortion				6.645	0.01
Yes	1	25 (71.4%)	112 (48.1%)
No	0	10 (28.6%)	121 (51.9%)
Multiple gestation				0.216	0.642
1 fetus	0	34 (97.1%)	229 (98.3%)
≥ 2 fetuses	1	1 (2.9%)	4 (1.7%)
Placental adhesion/retention				36.6	< 0.001
Yes	1	15 (42.9%)	17 (7.3%)
No	0	20 (57.1%)	216 (92.7%)
History of uterine surgery				22.15	< 0.0001
Yes	1	10 (28.6%)	12 (5.2%)
No	0	25 (71.4%)	221 (94.8%)
Gestational diabetes mellitus				3.183	0.074
Yes	1	11 (31.4%)	43 (18.5%)
No	0	24 (68.6%)	190 (81.5%)
Pregnancy-induced hypertension				13.87	0.0002
Yes	1	9 (25.7%)	15 (6.4%)
No	0	26 (74.3%)	218 (93.6%)
Anemia during pregnancy				0.687	0.407
Yes	1	7 (20.0%)	34 (14.6%)
No	0	28 (80.0%)	199 (85.4%)
Uterine atony				36.21	< 0.0001
Yes	1	11 (31.4%)	8 (3.4%)
No	0	24 (68.6%)	225 (96.6%)

Independent risk factors for pph in women undergoing pregnancy termination for fetal malformations

To identify independent predictors of PPH, variables that were significant in the univariate analysis were entered into a multivariable logistic regression model. PPH occurrence (yes = 1, no = 0) was defined as the dependent variable. In the full multivariable model (Table 2), gravidity (≥ 3), placental adhesion/retention, history of uterine surgery, pregnancy-induced hypertension, and uterine atony remained independently associated with PPH (all p < 0.05). In contrast, maternal age ≥ 35 years and history of abortion were not independently associated with PPH after adjustment for other variables.

Table 2.

Full Multivariable Logistic Regression Model for Identification of Independent Risk Factors for PPH

Factors	β	S.E	Wald χ2	OR(95% CI)	P
Intercept	−3.201	0.4064	62.031	0.041(0.017 − 0.084)	< 0.0001
Age (≥ 35 years)	−0.6980	0.6154	1.287	0.498(0.136 − 1.562)	0.257
History of abortion	−0.2069	0.6736	0.094	0.813(0.196 − 2.882)	0.759
Placental adhesion/retention	1.202	0.5730	4.399	3.326(1.045 − 10.05)	0.036
History of uterine surgery	1.500	0.6173	5.902	4.480(1.319 − 15.190)	0.015
Pregnancy-induced hypertension	1.649	0.5937	7.716	5.203(1.576 − 16.570)	0.006
Gravidity (≥ 3)	1.412	0.6832	4.296	4.103(1.156 − 17.510)	0.039
Uterine atony	1.389	0.652	4.530	4.009(1.104 − 14.580)	0.033

Variables included in this model were selected based on statistical significance in univariate analysis. The model was used to identify independent predictors of PPH after adjustment for potential confounders.

Development of the final prediction model for pph

To construct a clinically applicable prediction model, variables that remained statistically significant in the multivariable analysis (Table 2) were retained and refitted into a reduced logistic regression model. As the final model was refitted using only these predictors, the regression coefficients differ slightly from those reported in the full multivariable model. Compared with the full model, the reduced model included five predictors: placental adhesion/retention, history of uterine surgery, pregnancy-induced hypertension, gravidity ≥ 3, and uterine atony. The complete coefficient table of the reduced model is presented in Table 3. The logistic regression equation was: Logit (P) = –3.294 + 1.171 × (placental adhesion/retention) + 1.410 × (history of uterine surgery) + 1.680 × (pregnancy-induced hypertension) + 1.133 × (gravidity ≥ 3) + 1.392 × (uterine atony), where each predictor was coded as 1 when present and 0 when absent. This reduced model was then used for nomogram construction and performance evaluation.

Table 3.

Final reduced logistic regression model coefficients for the prediction of PPH

Predictor	Β	S.E	Wald χ2	OR	95% CI for β	95% CI for OR	P value
Intercept	−3.294	0.3699	79.317	0.037	−4.093 to −2.630	0.017–0.072	< 0.0001
Placental adhesion/retention	1.171	0.5638	4.318	3.227	0.0307–2.257	1.031–9.555	0.0377
History of uterine surgery	1.410	0.5902	5.703	4.094	0.233–2.567	1.262–13.02	0.0169
Pregnancy-induced hypertension	1.680	0.5921	8.049	5.364	0.489–2.835	1.631–17.04	0.0046
Gravidity ≥ 3	1.133	0.4787	5.607	3.106	0.198–2.096	1.219–8.132	0.0179
Uterine atony	1.392	0.6410	4.713	4.022	0.123–2.659	1.131–14.290	0.0299

Construction of the nomogram

A nomogram prediction model was constructed using the five predictors included in the final reduced logistic regression model: placental adhesion/retention, history of uterine surgery, pregnancy-induced hypertension, gravidity ≥ 3, and uterine atony. In the nomogram, placental adhesion/retention contributed 70 points, history of uterine surgery contributed 84 points, pregnancy-induced hypertension contributed 100 points, gravidity ≥ 3 contributed 67 points, and uterine atony contributed 83 points, as illustrated in Fig. 1.

Fig. 1.

Nomogram for predicting the risk of PPH in women undergoing pregnancy termination for fetal malformations

Discrimination of the predictive model

Based on a cutoff value of 0.5, the model correctly classified 240 of 268 participants, corresponding to an accuracy of 89.55% (95% CI 85.31–92.67%). The sensitivity was 37.14% (13/35; 95% CI 23.17–53.66%), specificity was 97.42% (227/233; 95% CI 94.50–98.81%), positive predictive value was 68.42% (13/19; 95% CI 46.01–84.64%), and negative predictive value was 91.16% (227/249; 95% CI 86.99–94.09%). The discriminative performance of the model was evaluated using the ROC curve, yielding an AUC of 0.8445 (95% CI 0.7731–0.9158, p < 0.0001) (Fig. 2).

Fig. 2.

ROC curve of the predictive model for PPH risk in women undergoing pregnancy termination for fetal malformations

Calibration of the predictive model for pph risk in women undergoing pregnancy termination for fetal malformations

The nomogram model was internally validated using 1000 bootstrap resamples, yielding a mean absolute error (MAE) of 0.047 for the calibration curve, indicating good agreement between the predicted and observed outcomes (Fig. 3). In addition, the Hosmer–Lemeshow goodness-of-fit test produced a p-value of 0.6892 (> 0.05), suggesting that the predictive model demonstrates an adequate fit.

Fig. 3.

Calibration plot of the predictive model for PPH risk in women undergoing pregnancy termination for fetal malformations

Discussion

Pregnancy termination for fetal malformations is an important obstetric procedure; however, PPH remains a serious complication and may threaten maternal safety. In this study, 268 women who underwent pregnancy termination for fetal malformations were included, among whom 35 developed PPH, corresponding to an incidence rate of 13.06%. This rate appears higher than that commonly reported in routine vaginal delivery populations and is comparable to previous evidence from medical termination for fetal anomaly at or beyond 20 weeks’ gestation, in which PPH was reported in 45 of 407 women (11.1%) [6]. Our present study further identified several clinically accessible predictors of PPH and integrated them into a simple prediction model for this specific clinical setting. These findings indicate that commonly available clinical factors may help identify women at increased risk during termination for fetal malformations, where PPH is relatively frequent and clinical management differs from routine term labor. Therefore, the proposed tool may support clinical preparedness in women considered to be at higher risk, while real-time clinical assessment should remain the basis of clinical decision-making.

Previous studies have identified multiple causes and risk factors for PPH, among which uterine atony, retained placenta or abnormal placentation, and pregnancy-related complications are commonly reported as clinically important contributors [11]. Regarding our predictor “history of uterine surgery” in the present model, previous uterine procedures, including abortion-related uterine instrumentation and curettage, may damage the endometrium and myometrium, thereby increasing the risk of abnormal placental adherence or retention in subsequent pregnancies [12]. This mechanism is clinically relevant because placental adhesion/retention was identified as one of the independent predictors retained in the final model. Moreover, placental adhesion/retention may be related to previous curettage or endometrial injury, and incomplete placental separation during the third stage of labor can prolong bleeding, impair effective uterine contraction, prevent timely closure of uterine blood vessels, and consequently increase blood loss [12, 13]. In addition, women with hypertensive disorders of pregnancy are prone to systemic arteriolar spasms, which heighten the risk of placental abruption, and while antispasmodic and antihypertensive treatments are necessary, they may also contribute to uterine atony postpartum, further increasing the risk of PPH [14, 15].

In the present study, the full multivariable logistic regression identified five independent predictors of PPH, which were subsequently retained and refitted into a reduced final model for nomogram construction and performance evaluation. The model’s discriminatory power was assessed using the ROC curve, yielding a promising AUC of 0.8445, and with a cutoff value of 0.5, the model showed high specificity but relatively low sensitivity, indicating that it may be more useful for identifying a subset of women at high risk of PPH than for capturing all women who will develop PPH. In practice, a high predicted risk may support preparedness, including closer monitoring, uterotonic readiness, blood product preparation, early involvement of experienced obstetric and anesthesia teams, and, when abnormal placentation is suspected, referral to centers where multidisciplinary management and prophylactic endovascular procedures are available [16]. However, clinical decisions should remain based on real-time assessment, and external validation is required before broader clinical implementation. Future investigations are also needed to evaluate clinically appropriate cutoffs that may improve sensitivity while preserving acceptable specificity.

The calibration of the nomogram model was evaluated using 1000 bootstrap resamples, yielding an MAE of 0.047, which indicated acceptable agreement between the predicted and observed outcomes. The Hosmer–Lemeshow goodness-of-fit test also showed no significant lack of fit, with a p-value of 0.6892. Taken together, these findings suggest that the model had acceptable internal performance in the present cohort. Compared with existing PPH prediction tools developed mainly from routine delivery populations, this model was derived from women undergoing termination for fetal malformations, a clinical setting in which prolonged induction, placental separation problems, and intrapartum changes may contribute to hemorrhage risk. The final model included both baseline variables and variables identified during clinical management. Gravidity, history of uterine surgery, and pregnancy-induced hypertension were available before or at the start of induction, whereas placental adhesion/retention and uterine atony were identified during induction, delivery, or immediate postpartum care. Therefore, the model should be interpreted as a dynamic clinical risk-assessment model instead of a pre-induction prediction model because placental adhesion/retention and uterine atony may be part of the clinical pathway leading to PPH. Their inclusion may help clinicians recognize increasing hemorrhage risk during management, but these factors should not be interpreted as independent baseline predictors present before induction begins or as indications for aggressive intervention in the absence of clinical bleeding.

Limitations

This study has several limitations. First, this was a single-center retrospective study with a relatively small sample size, as only 35 PPH events occurred. Since the final reduced model included five predictors, the event-per-variable ratio was limited, which might have reduced the stability of the regression estimates and increased the risk of overfitting. In addition, several clinically relevant candidate predictors were evaluated before construction of the final model, and thus, the findings should be interpreted as exploratory. Although bootstrap resampling was used for internal validation, the generalizability, calibration stability, and clinical usefulness of the model require confirmation in larger multicenter cohorts. Second, gestational age was not included in the final prediction model as a complete comparative analysis across gestational-age groups was not available in the current dataset. Given the limited number of PPH events, adding gestational age without stable stratified estimates could have increased model instability. Nevertheless, gestational age remains clinically relevant in termination-related hemorrhage risk and should be further evaluated in larger multicenter studies. Third, fetal malformations were not subcategorized in this dataset; therefore, we were unable to assess whether PPH risk varies by anomaly type. Future prospective studies with larger sample sizes and external validation cohorts are needed to confirm the robustness of this model and to explore whether additional clinical or biological markers, such as the sFlt-1/PlGF ratio, could improve risk assessment in this population.

Conclusion

In this study involving women undergoing termination for fetal malformations, we identified five clinically accessible predictors of PPH and developed an internally validated nomogram with good discrimination and calibration. The model may support risk stratification and preparedness during induction and intrapartum management in this high-risk setting. However, external validation in independent cohorts is required before broader clinical implementation.

Author contribution

All authors contributed to the study conception and design. Material preparation, data collection, and analysis were performed by Cuihua Zhou, Xiuhua Zhang, Qian Rao, and Mei Song. The first draft of the manuscript was written by Cuihua Zhou and Hongbiao Yu, and all authors, including Cuihua Zhou and Shujuan Yang, commented on previous versions of the manuscript. All authors read and approved the final manuscript.

Funding

The authors declare that no funds, grants, or other support was received during the preparation of this manuscript.

Data availability

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Declarations

Comflict of interest

The authors have no relevant financial or non-financial interests to disclose.

Ethical approval

This study was performed in line with the principles of the Declaration of Helsinki. Approval was granted by the Ethics Committee of Nanchong Central Hospital (Approval No. 2024-Review-198).

Consent to participate

Written informed consent was obtained from all individual participants included in the study.

Consent to publish

Not applicable.