Novel biomarkers for prediction of atonic postpartum hemorrhage among ‘low-risk’ women in labor

Background Postpartum hemorrhage (PPH) is the primary cause of maternal mortality globally, with uterine atony being the predominant contributing factor. However, accurate prediction of PPH in the general population remains challenging due to a lack of reliable biomarkers. Methods Using retrospective cohort data, we quantified 48 cytokines in plasma samples from 40 women diagnosed with PPH caused by uterine atony. We also analyzed previously reported hemogram and coagulation parameters related to inflammatory response. The least absolute shrinkage and selection operator (LASSO) and logistic regression were applied to develop predictive models. Established models were further evaluated and temporally validated in a prospective cohort. Results Fourteen factors showed significant differences between the two groups, among which IL2Rα, IL9, MIP1β, TNFβ, CTACK, prenatal Hb, Lymph%, PLR, and LnSII were selected by LASSO to construct predictive model A. Further, by logistic regression, model B was constructed using prenatal Hb, PLR, IL2Rα, and IL9. The area under the curve (AUC) values of model A in the training set, internal validation set, and temporal validation set were 0.846 (0.757–0.934), 0.846 (0.749–0.930), and 0.875 (0.789–0.961), respectively. And the corresponding AUC values for model B were 0.805 (0.709–0.901), 0.805 (0.701–0.894), and 0.901 (0.824–0.979). Decision curve analysis results showed that both nomograms had a high net benefit for predicting atonic PPH. Conclusion We identified novel biomarkers and developed predictive models for atonic PPH in women undergoing “low-risk” vaginal delivery, providing immunological insights for further exploration of the mechanism underlying atonic PPH.


Introduction
Postpartum hemorrhage (PPH), the leading cause of maternal mortality, affects 3%-10% of deliveries and accounts for nearly 27% of maternal deaths worldwide (1,2).Researchers at numerous academic institutions have investigated related risk factors and developed tools to assess the risk of PPH, which merely stratified the risk of PPH based on clinical characteristics.However, emerging evidence suggests that these tools have limited predictive performance.For example, approximately 40% of women classified as high-risk individuals did not experience postpartum bleeding (3,4).Although common risk factors increase the risk of PPH, our clinical data demonstrate that even low-risk women are susceptible to this condition.Additionally, the amount of bleeding is often underestimated and the importance of this symptom is overlooked, leading to delayed diagnosis and treatment and causing severe adverse outcomes among these women.The insufficiency in our current understanding of PPH highlights the need to explore novel indicators and establish a comprehensive predictive model.
Uterine atony continues to be the primary cause of PPH, accounting for approximately 70% of cases (5).Effective uterine contraction is crucial in reducing PPH, while factors that impede proper contraction can lead to atonic PPH and even death in severe cases.Almost all women giving birth are at risk of bleeding due to uterine atony.Compared to cesarean section, where timely measures can be taken, vaginal deliveries pose a greater risk to maternal and child safety due to the more sudden onset of PPH and limited access to emergency resources, especially in lowresource countries.
Recent studies have shown the crucial role of immune cells and cytokines in maintaining pregnancy and initiating labor (6,7).Previous research found a correlation between certain cytokines and preterm birth, with some markers being able to predict delivery timing and adverse pregnancy outcomes (8,9).The onset of labor is accompanied by substantial changes in fetomaternal physiology that facilitate successful delivery.These changes include the disruption of immune tolerance within the fetal membrane and placenta due to immune infiltration, as well as changes in the maternal internal immune environment (10,11).Notably, evident chemotaxis and infiltration of neutrophils and macrophages into the myometrium occur after labor begins, leading to increased expression of inflammatory cytokines (12,13).The secretion of cytokines plays a crucial role in the initiation and progression of uterine contractions and labor during pregnancy (14, 15).However, the association between maternal inflammatory cytokines in labor and uterine atony has not been investigated.Besides, recent studies have suggested that hemogram and coagulation parameters may also be related to this inflammatory response.
The predictive value of clinical features for uterine atony among low-risk women undergoing vaginal delivery is extremely limited.Currently, no clinical characteristics can accurately predict postpartum atony in the general population.Therefore, this study was initiated to analyze plasma cytokines in women undergoing vaginal delivery and establish a predictive model by integrating hemogram and coagulation parameters, providing valuable insights for identifying atonic PPH in the general population.

Patients
Utilizing our extensive biobank, we conducted a nested casecontrol study to identify plasma biomarkers associated with atonic PPH in 'low-risk' primiparous women from April 2022 to November 2023.The inclusion criteria were as follows: primipara with full-term singleton infant, fetal head presentation, and age range of 18 to 34 years.The following exclusion criteria were applied: (1) women with autoimmune diseases; (2) women with pregnancy comorbidities and complications potentially associated with PPH (including preeclampsia, poorly controlled or medically controlled gestational diabetes mellitus, placental abruption, scarred uterus, coagulopathy, and induced labor); (3) women who delivered low-birth-weight infants or macrosomia clinically diagnosed after birth; (4) PPH cases caused by placental factors (identified through ultrasound and placental examination); (5) traumatic PPH cases with severe lacerations of the birth canal; (6) women with intrauterine infection; (7) cases positive for group B streptococcus.Given the high prevalence of obstetric lacerations, it is worth noting that most minor ones do not result in significant PPH.Therefore, our study included individuals with mild obstetric lacerations, defined as mild cervical lacerations (<1cm) or first-degree lacerations of perineum.Furthermore, these lacerations would be confirmed by a senior doctor and not considered as the primary cause of bleeding.
The blood samples were collected during the latent phase of the first stage of labor, which was defined as a period characterized by regular uterine contractions and progressive changes of the cervix, including some degree of effacement and slower progression of dilatation up to 5 cm (16).PPH was diagnosed based on excessive bleeding (≥ 500 ml) in the first 24 h after birth.This study employed volumetric methods instead of traditional approaches to accurately quantify blood loss (17).In brief, intrapartum blood loss was measured using a V-shaped blood collection bag with scale lines and warning lines, which allows the blood loss to be read at any time (Supplementary Figure S1).Atonic PPH was diagnosed based on specific criteria: vaginal bleeding (≥ 500 ml) following spontaneous placental expulsion and identification of a flaccid uterus during physical examination.
Eligible women who experienced atonic PPH after vaginal delivery were assigned to the atonic PPH group, while the remaining pregnant  women without PPH were matched to the atonic PPH group in a 1:1 ratio using propensity scores.Gestational age, BMI (both before pregnancy and at delivery), and newborn birth weight were considered as predictors for PPH outcome during sample collection, and logistic regression was used to calculate the propensity score.Propensity score matching (PSM) was performed using maximized execution performance without replacement, with a caliper width set at 0.02 standard deviations of the logit of propensity score.Finally, a training cohort comprising 40 women diagnosed with atonic PPH and their matched controls was established.Similarly, in the prospective validation cohort, samples were collected from 41 women who experienced atonic PPH and matched with 41 normal controls.The baseline characteristics were extracted from the electronic medical record, while data on obstetric and perinatal complications were collected through timely follow-up by trained nurses.Written informed consent was obtained from each subject, and this study was approved by the Ethical Review Committee of the local hospital (2022KY 063).The study flow is shown in Figure 1.

Sample collection
Specimens were collected during the latent first stage after spontaneous labor.Samples of 6 ml of peripheral blood were taken from each participant via venipuncture using EDTA anticoagulant tubes for detection.After centrifugation at 3000 rpm for 10 min at 4°C, the supernatant plasma was separated and stored at −80°C.

Cytokine profiling
The frozen-thawed plasma was centrifuged and diluted 2-fold with a diluent.Subsequently, a 50 mL sample was taken for testing.The Bio-Plex Pro Human Cytokine Screening 48-Plex Panel (Bio-Rad, #12007283) was used to detect the levels of 48 cytokines (Table 1).The detailed procedures for biomarker measurement methods can be found in Additional File 1: Detailed Methods.

FIGURE 1
Flow diagram illustrating the process of participant screening and enrollment.FGR, fetal growth restriction; HDP, hypertensive disorders of pregnancy; GDM, gestational diabetes mellitus; IUFD, intrauterine fetal death; PPH, postpartum hemorrhage; PSM, propensity score matching.Representative assay working ranges, assay sensitivity, and precision are presented in Additional Files (Supplementary Table S1).

Routine laboratory tests
Results of the last laboratory test before delivery, including whole blood cell analysis and coagulation tests, were obtained upon patients' admission.The complete blood count parameters were analyzed using Cell-Dyn 3700 (Abbott Diagnostics, Santa Clara, CA, USA).Coagulation tests were conducted using the Sysmex CS 2500 System coagulation analyzer (Siemens Healthcare Diagnostics, Erlangen, Germany).All measurements were performed within a 2-h time frame after blood sampling.Neutrophil-to-lymphocyte ratio (NLR), monocyte-to-lymphocyte ratio (MLR), platelet-tolymphocyte ratio (PLR), systemic immune-inflammation index (SII), and LnSII were calculated based on these laboratory test parameters, which were previously reported to be related to systemic immune response.The SII was obtained by multiplying the neutrophil count by the PLR and LnSII was the natural logarithm of SII.
concentration of each sample was determined by constructing a linear standard curve using the provided software, in accordance with the manufacturer's instructions.

Statistical analysis
Statistical analysis was performed using SPSS 26.0 (IBM, Armonk, NY, USA), R 4.3.1 (R Foundation for Statistical Computing, Vienna, Austria), and GraphPad Prism 9.0 (GraphPad Software, San Diego, CA, USA).For continuous variables, normally distributed data are presented as mean ± standard deviation, and non-normally distributed data are presented as median and interquartile range.Independent samples t-test was employed for normally distributed continuous variables, whereas Mann-Whitney U test was used for nonnormally distributed ones.Paired comparisons were conducted using the Wilcoxon paired test.Categorical variables are presented as counts (percentages) and were compared using chisquare analysis or Fisher's exact test.Statistical significance in cytokines was determined at P< 0.05.Multiple comparisons in hemogram and coagulation parameters were adjusted for using a false discovery rate (FDR)<0.1.Differentially expressed cytokines, hemogram, and coagulation parameters were incorporated into subsequent variable screening and the construction of predictive models.
Considering the limited number of observed events, we used least absolute shrinkage and selection operator (LASSO) regression with tenfold cross-validation to mitigate potential collinearity and overfitting issues.A higher l value would shrink smaller coefficients towards zero, retaining only the most influential predictors.To ensure model conciseness, we selected the l min resulting in nine predictors to construct predictive model A. Then, we further analyzed significant variables selected by LASSO regression using logistic regression to identify predictors to construct model B. Risk assessment nomograms for atonic PPH were generated.The models established using the training set were validated.Internal validation was performed using the bootstrap method (1000 times), while temporal validation used prospective cohort data.Discrimination of the model was evaluated by drawing a receiver operating characteristic curve (ROC) and calculating the area under the curve (AUC).Calibration curves were plotted to visualize the calibration of the nomograms and assessed using Hosmer-Lemeshow (H-L) chi-square statistics.Additionally, the clinical utility of the predictive models was assessed by conducting decision curve analyses (DCA) to demonstrate the potential net benefit of the model at various threshold probabilities.

Baseline characteristics and pregnancy outcomes
In the retrospective cohort, a total of 5285 women underwent vaginal deliveries.After applying the predetermined inclusion and exclusion criteria and considering the availability of blood samples, we included 40 'low-risk' women with atonic PPH and matched them with 40 normal controls.For temporal validation, we prospective recruited 41 women who developed atonic PPH and matched them with 41 normal controls (Figure 1).The baseline characteristics, including maternal age, BMI, blood pressure, gestational age, newborn birth weight, and other characteristics were comparable in both two pairs of compared groups (Table 2).

Analysis of maternal circulating cytokines revealed dysregulation in five cytokines associated with atonic PPH
Table 1 provides a summary of the cytokine expression profile in both the atonic PPH and control groups.And to compare differences in cytokine levels across different groups, overall cytokine data was visualized in a heat map representation (Supplementary Figure S2).Besides, pairwise Pearson's correlation analysis between cytokines was performed in the training cohort (Supplementary Figure S3).Among these cytokines, the expression levels of IL2Ra, IL9, TNFb, and MIP1b in the PPH group were significantly lower than those in the control group.Conversely, CTACK exhibited significantly higher levels in atonic PPH than in the control group.Violin plots were generated for these five cytokines (Figure 2).

Construction of the predictive models
The nine factors selected by LASSO were all used to construct predictive model A (Supplementary Table S3).Subsequently, to further control the influence of confounding factors, the above nine factors were analyzed using logistic regression.Finally, only prenatal Hb, PLR, IL2Ra, and IL9 were determined to be independent predictors (P< 0.05), as shown in Supplementary Table S4.These four factors were used to develop model B  GDM cases refer to women who can manage their blood glucose levels through diet and exercise, without medication.This involves achieving fasting blood glucose level < 5.3 mmol/L (95 ml/dl), postprandial 2-hour blood glucose level < 6.7 mmol/L, and maintaining HbA1c levels within a controlled range of 6% (42 mmol/mol).
b Mild obstetric lacerations are defined as first-degree perineal lacerations or cervical lacerations (<1cm).The included.c The intrapartum blood loss was collected in a graduated collection bag, which was subsequently removed when the patient left the delivery room.
d The total blood loss within 24 hours was calculated by summing the intrapartum blood loss (collected in a bag) and the postpartum blood loss in the ward (measured by weight).

Predictive nomograms for atonic PPH
The "rms" package was used to draw nomograms for atonic PPH in women with vaginal delivery (Figure 6), which could predict the risk of atonic PPH in individual patients.In clinical application, the individual score for each predictor could be acquired and then added together to obtain the total score.Finally, on the number axis of the total score, the corresponding predicted probability projected downward would be the risk of atonic PPH for individuals.

Temporal validation
ELISA was employed to quantify IL2Ra, MIP1b, TNFb, CTACK, and IL9 levels in maternal plasma obtained from a prospective cohort.Relevant predictive laboratory parameters (Prenatal Hb, Lymph%) were extracted from electronic medical records and hemogram-derived indices (PLR, LnSII) were calculated.The results revealed significant lower levels of IL2Ra (P<0.001),MIP1b (P<0.001),TNFb (P<0.001), and IL9 (P<0.001) in the atonic PPH group compared to the control group.Conversely, CTACK levels were found to be significantly elevated in women with atonic PPH compared to the matched control group (P<0.001).Furthermore, lower levels of Prenatal Hb (P = 0.004), PLR (P = 0.007), and LnSII (P = 0.005) were observed in women diagnosed with atonic PPH, while Lymph% (P = 0.026) showed a significant increase in this group (Table 4, Figure 7).

Evaluation of models Discrimination
The AUC values for model A in the training set, internal validation set, and temporal validation set were 0.846 (0.757-0.934), 0.846 (0.749-0.930), and 0.875 (0.789-0.961) respectively.The corresponding AUC values for model B were 0.805 (0.709-0.901), 0.805 (0.701-0.894), and 0.901 (0.824-0.979).ROC curves was plotted for both model A and model B (Figure 8).The accuracy, sensitivity, and specificity of model A in the training cohort were 0.813, 0.850, and 0.775 respectively, while in the validation cohort they were 0.756, 0.902 and 0.625 respectively.The corresponding values for model B in the training cohort were 0.750, 0.725 and 0.775, whereas in the validation cohort they were measured as 0.841, 0.854 and 0.829.

Calibration
For model A, the P values of the H-L test were 0.088 in the training set and< 0.001 in the prospective validation cohort.For model B, the corresponding values were 0.827 and 0.158.Calibration curves for both models are presented in Figure 9.The results indicate that model A exhibited poor fitting performance on the prospective validation set.Furthermore, the calibration curve revealed that the predicted values of model A demonstrated greater deviation from the ideal calibration curve.In contrast, model B exhibits superior calibration on the prospective validation set.

Clinical utility
DCA was used to evaluate model A and model B in both the training set and the temporal validation set (Figure 10).Both models showed better net benefit for predicting atonic PPH across all potential thresholds (0% to 80%).This suggested that both models achieved a more optimal balance of clinical intervention for atonic PPH, regardless of the chosen risk threshold.

Discussion
Using LASSO and multivariate logistic regression, we developed predictive models A and B for atonic PPH in "low-risk" women based on samples from a retrospective cohort and matched controls.The performance of the models was assessed through both internal and temporal validation.The findings of this study suggested that model B was promising for clinical implementation.
Early identification of risk factors allows obstetric healthcare professionals to adequately prepare for potentially life-threatening hemorrhages (1).Numerous relevant risk factors are currently used in predictive models, with almost all mentioned in the literature being clinical indicators.The variables frequently included in predictive models include parity, antenatal hemoglobin level, antepartum bleeding, maternal age ≥35 years old, gestational age, newborn weight, multiple pregnancies, obesity, a history of previous cesarean section, and placenta previa (18).For instance, the California Maternal Quality Care Collaborative (CMQCC) developed risk stratification tools for PPH based on major clinical and laboratory risk factors (19).A retrospective study involving 10,134 patients demonstrated that this tool correctly identified 80% of severe PPH cases, but its specificity was only 60% (3).Besides, limited research has been performed specifically focusing on PPH in low-risk populations without apparent risk factors related to PPH.The PPH in these women, however, does not meet the expected low levels.There is thus still an urgent need to identify highly effective novel biomarkers for predicting PPH.
Prior to delivery, immune cells accumulate in the maternal reproductive system and at maternal-fetal interface through cytokine secretion, establishing an inflammatory environment (20)(21)(22).These cytokines play a pivotal role in the initiation of labor and their dysregulation may contribute to PPH.Jiang et al. reported significantly elevated levels of basic FGF, IL1a, IL1b, IL1ra, IL2Ra, IL16, IL18, M-CSF, MIP1a, b-NGF, TRAIL and SCF were associated with an increased risk of atonic PPH (23).Gallo et al. found significantly higher concentrations of IL16, IL6, IL12/IL23p40, MCP1 and IL1b in the PPH group compared to the control group (24).Both studies identified significant biomarkers including IL16 and IL1b associated with PPH.In this study, we observed notable levels of IL2Ra, TNFb, MIP1b, IL9 and higher level of CTACK in atonic PPH group.The discrepancy between our study results and previous studies (23,24) may be attributed to differences in the populations included and the timing of specimen collection: specimens collected by Jiang et al. were obtained in late third trimester (37-41 weeks), while those collected by Gallo et al. were within three days prior delivery.However, the samples in this study were collected during the latent phase of the first stage of labor, characterized by the presence of regular contractions.Cytokine concentrations undergo continuous changes throughout pregnancy and labor progression.Further research is needed to investigate the correlation between cytokine levels at different gestational time points and their association with PPH.
Hemogram and coagulation parameters have significant predictive value for PPH.A previous case-control study showed that individuals with a fibrinogen level below 2 g/L are significantly predisposed to severe bleeding (25).Another retrospective study identified PLT< 150×10 9 /L and an increased APTT ratio as independent risk factors for PPH (26).SII has been proven to have prognostic and diagnostic value in various diseases including infection, tumor development, and cardiovascular disease, indicating its potential significance for assessing immune status and the inflammatory response (27-29).However, the relationship between SII and pregnancy-related outcomes remains unclear.Our findings suggest that prenatal Hb levels, Lymph% PLR, and LnSII showed certain predictive value for atonic PPH.

A B
Predictor selection using LASSO regression analysis with tenfold cross-validation.In contrast to the case for other pregnancy-related complications, the prediction of PPH is theoretically more accurate as parturition approaches.The present study explored the incidence of atonic PPH from a novel perspective, specifically investigating the correlation between maternal biomarkers and the incidence of PPH at the presence of regular uterine contraction.By using peripheral blood samples collected during the latent first stage after spontaneous labor, this study identified novel biomarkers for atonic PPH.Secondly, to address potential issues of multicollinearity compromising model robustness when solely using a logistic regression model, this study initially screened for factors showing intergroup differences and subsequently adding LASSO to logistic regression analysis.This study revealed low prenatal Hb, PLR, IL2Ra, and IL9 as factors independently predictive of atonic PPH.The predictive model based on these four factors exhibited favorable validation outcomes in a temporal validation cohort, emphasizing its clinical utility and suggesting its potential as a viable model for predicting atonic PPH.
The limitations of this study include a relatively small sample size and data derived from a single center, which may introduce selection bias.Future research should focus on larger sample sizes, multi-center data, and prospective trials to further identify and confirm risk factors for atonic PPH in the general population.This study did not include women who underwent cesarean section, since the onset of delivery is closely related to inflammatory responses and selective cesarean section is unrelated to this process.Additionally, owing to different immune characteristics between women undergoing spontaneous labor and induced labor (30), this study does not involve the population undergoing induced labor.The cohort data used in this study only include primiparous women because the labor characteristics of primiparous and multiparous women differ.Further studies should expand the demographic category.For both model A and model B, the AUC values obtained from prospective validation were higher than those from the training set, which may be attributable to changes in detection methods.In the training set, Bio-Plex Pro Human Cytokine 48-Plex Screening Panel was used to simultaneously determine the   Frontiers in Immunology frontiersin.orgexpression levels of 48 cytokines.However, targeted validation involving ELISA was performed on targeted cytokines in the prospective cohort.This study found that intergroup differences in cytokine levels detected by ELISA were more pronounced than those observed in the training set.In future research, our established predictive models should be validated using consistent testing methods.

Conclusion
In conclusion, our research findings suggested that several plasma cytokines are dysregulated during the latent phase of labor in "low-risk" women who later develop atonic PPH.By using LASSO and logistic regression models, we were able to accurately assess the risk of atonic PPH in prospective cohorts.This approach allows early identification of high-risk individuals among pregnant women diagnosed with a "low risk" of bleeding, thereby significantly reducing the incidence and mortality rates associated with PPH in vaginal deliveries.
(A) LASSO coefficient profiles of the 14 risk factors were created against the log (l) sequence.(B) Tuning parameter (lambda, l) selection of deviance in the LASSO regression based on the minimum criterion (left dotted line) and the 1-SE criterion (right dotted line).In the present study, predictor selection was performed according to the minimum criterion (including Prenatal Hb, Lymph%, PLR, LnSII, IL2Ra, MIP1b, TNFb, CTACK, and IL9).LASSO least absolute shrinkage and selection operator, SE standard error.

FIGURE 9
FIGURE 9 Calibration curves of model A and model B for predicting atonic PPH.(A) Calibration curves for the training cohort of predictive models A and B The red line (Bias corrected line for model A) and the green line (Bias corrected line for model B) represent the performance during internal validation by bootstrapping (B = 1000 repetitions).(B) Calibration curves for the prospective validation of predictive models A and B The red line (Apparent line for model A) and the green line (Apparent line for model B) represent the original performance in a prospective cohort.

TABLE 1
Comparisons of concentrations of cytokines between the atonic PPH and matched control group in training cohort.

TABLE 2
Baseline characteristics of participants.

TABLE 2 Continued
Continuous variables are reported as mean ± standard deviation or median (interquartile range), while categorical variables are presented as number (percentage).BMI, Body mass index; GDM, gestational diabetes mellitus; PPH, postpartum hemorrhage; ART, assisted reproductive technology.

TABLE 3
Hemogram and coagulation indices in atonic PPH and matched controls in training cohort.

TABLE 3 Continued
Data are presented as mean ± standard deviation or median (interquartile range).The P value was adjusted using the Benjamini-Hochberg procedure to control the false discovery rate (FDR qvalue) for multiple testing.

TABLE 4
Validation of differentially expressed markers in a prospective cohort.