A high HEMSTOP bleeding score is a major independent risk factor for postpartum hemorrhage: a prospective cohort study

Abstract

Background

Postpartum hemorrhage (PPH) is a common yet potentially serious complication of delivery. Recently, the HEMSTOP bleeding score has emerged as a tool for detecting preoperative inherited bleeding disorders. Our objective is to investigate its relevance within the obstetrical context for predicting PPH occurrences among unselected pregnant women.

Methods

Prospective cohort study conducted in a tertiary maternity hospital between 2014 and 2016 including women without any known bleeding disorder nor antithrombotic therapy who completed the HEMSTOP questionnaire before delivery. Primary outcome was primary PPH ≥ 500 mL following delivery. We studied the bleeding score ranging from zero to seven both as a continuous and an ordinal variable. We used multivariable analysis with adjustment on PPH risk factors. We also estimated the measures of the bleeding score prognostic accuracy.

Results

PPH occurred in 116 of 2536 women (4.6%). Elevated bleeding scores were associated with increased PPH risk (adjusted RR = 1.58; 95% CI, 1.23 to 2.02; P < 0.001 for a continuous score) demonstrating a progressive rise in risk alongside score escalation (adjusted RR = 1.58; 95% CI, 1.01 to 2.46 for a score of one, adjusted RR = 2.11, 95% CI 0.86 to 5.20 for a score of two and adjusted RR = 7.20, 95% CI 2.54 to 20.41 for a score of three compared with a score of zero; P < 0.001). The area under the curve for the bleeding score accuracy in predicting PPH was 0.56 [95% CI 0.52 to 0.61].

Conclusions

The HEMSTOP bleeding score is weakly predictive of PPH. Women scoring ≥ three are at high risk of PPH.

Background

Postpartum hemorrhage (PPH) concerns 5% of deliveries in high-income countries and remains the leading cause of maternal morbidity and mortality worldwide [1, 2]. Consequently, it is crucial to identify women at higher risk to initiate active management of delivery and early recognition and treatment of PPH. Among the potentially detectable and treatable maternal risk factors for PPH are the inherited mild-to-moderate bleeding disorders (MBDs). These non-severe bleeding phenotypes, such as hemophilia carriership, von Willebrand disease and platelet disorders, can remain occult until delivery and be involved in the occurrence of PPH and progression to severe PPH [3]. They are probably under-diagnosed in PPH, masked by an additional obstetrical cause or responsible for part of the 5 to 40% of PPH with unidentified cause [4, 5]. Two recent studies exploring hemostasis in women after severe PPH found high proportions of MBDs of 23% and 56%, but women included in these studies were referred to hemostasis specialist because of an atypical course of PPH [6, 7]. Thereby, little is known about the role of MBDs in PPH, especially in unselected parturients.

Three quarters of women with a bleeding diathesis have unclassified bleeding disorders without any biological abnormalities highlighted [8]. Moreover, some hemostasis parameters usually not considered as pathological have been shown to be risk factors for severe PPH [9]. Finally, prophylactic administration of hemostatic therapy following abnormal biological tests in the pre-interventional period is rarely undertaken [10]. This has shifted the focus from routine biological screening to clinical evaluations of bleeding risk, prompting the use of standardized questionnaires to receive the history of bleeding diathesis and calculate a bleeding score, as recommended by both national and international guidelines [10, 11]. Only two studies evaluated the prognostic performance of a clinical questionnaire on bleeding diathesis for the occurrence of PPH in the general population. Among 1147 unselected women, Gillissen et al. did not find an association between a high bleeding score and a PPH ≥ 1000 ml [12]. However, they used a Vicenza-based questionnaire which is not intended for screening MBDs in the general population. By contrast, in a recent retrospective cohort study including 3588 unselected pregnant women, Zec et al. tested the HEMSTOP questionnaire previously developed for identifying MBDs before invasive procedures in the general population [13]. Brief and effective in the surgical context, this questionnaire is proposed in the French guidelines for pre-interventional hemostatic assessment [10, 14,15,16]. The authors showed a significant crude relationship between the score and the presence of a PPH ≥ 1000 ml, in a secondary analysis and with only one score cut-off tested [13]. To assist clinicians in identifying parturients with a propensity for bleeding and to facilitate appropriate hemostatic management during delivery, further research into the HEMSTOP score in this context is essential.

The purpose of this study was to use prospective data and multiple adjusted analyses to determine the predictive value of the HEMSTOP bleeding score for the occurrence of PPH in unselected parturients.

Methods

The source population consisted of all pregnant women in a tertiary-level maternity teaching hospital in France (Louis-Mourier Hospital, Colombes). We prospectively and consecutively recruited all the women who underwent an elective obstetric pre-anesthesia clinic from October 2014 to November 2015, which is a standard of prenatal care within the third trimester of pregnancy in France. Approval for the study was obtained by the ethics committee HUPNVS, Paris, France (n°IRB00006477). All the women included were informed about the study and did not indicate their opposition to participate. Women were routinely asked to complete a paper-based HEMSTOP bleeding questionnaire which was integrated into the anesthesia record. We included all women who thoroughly and accurately completed the questionnaire, allowing for the calculation of a bleeding score. The follow-up was terminated on the end of April 2016 to include all deliveries. We excluded women with a known congenital or acquired bleeding disorder since a bleeding tendency and a higher risk of PPH compared to the general population are already demonstrated among those women [3, 17, 18]. We also excluded women with abnormal placentation because of a specific peripartum management. Detailed information on sociodemographic characteristics, medical and obstetrical history, prenatal care, delivery, and obstetric hospitalization was collected from review of the women’s medical charts.

The main exposure variable was the HEMSTOP bleeding score. The creation of the HEMSTOP questionnaire from a consensus reached by a group of anesthesiologists and hemostasis specialists has been previously described [14]. HEMSTOP is an acronym standing for Hematoma, hEmorrhage, Menorrhagia, Surgery, Tooth extraction, Obstetrics and Parents. It is composed of seven binary questions regarding tendency to bruises or hematomas, tendency to prolonged or unusual bleeding requiring medical care, bleeding after dental extraction requiring medical care, abnormal bleeding after surgery, family members with congenital hemostatic disorder, menorrhagia requiring medical care and prior PPH (Table 1). Three answers to each item were available: “yes”, “no” or “situation never encountered”. Each positive answer scores one. The HEMSTOP bleeding score thus ranges from zero to seven. We studied the score as a continuous variable and as an ordinal variable, with three different cut-offs to be considered as abnormal (≥ one, ≥ two, ≥ three). We used the two first cut-offs because they have shown good prognostic performance in the surgical context and the third because it is suggested in the French guidelines for pre-interventional hemostatic assessment [14,15,16]. We also explored the bleeding items separately as secondary exposures variables.

Table 1 HEMSTOP Questionnaire

The primary outcome was primary PPH, defined as a blood loss ≥ 500 mL following delivery. Blood loss was measured on a standardized basis by the use of collector bags in all women and additional suction canisters for cesarean delivery.

Characteristics of anesthesiologist-directed biological hemostatic assessment according to the bleeding score were described: type and timing of hemostasis tests ordered and referral to a hemostasis specialist. A recent platelet count performed at the sixth month of pregnancy was available for the pre-anesthesia clinic, as part of the mandatory prenatal screening in France [16]. We compared observed practices to national and international guidelines on pre-interventional hemostatic assessment, which advise against routine biological testing for MBDs screening and recommend it solely for women with a clinical bleeding diathesis, in conjunction with a specialized consultation [10, 11]. For this analysis, we excluded women with prior biological assessment of secondary hemostasis and also those for whom a hemostasis biological assessment was ordered due to a language barrier or medical conditions such as gestational thrombocytopenia or pre-eclampsia. Characteristics of specific hemostatic management planned for delivery according to the bleeding score were also recorded for all women. Finally, administration of neuraxial anesthesia or analgesia according to the bleeding score was evaluated.

Statistical analysis

The association between the bleeding score and the occurrence of PPH was explored using uni and multivariable logistic regressions. The confounding factors included in the multivariate analysis were PPH risk factors selected on the basis of the literature and the results of the univariate analysis. Overall, there was missing data, exclusively on covariates, for 62 women (2.4%) included in the multivariable analysis. We used multiple imputation by chained equations to address missing data according to Rubin’s rules and randomly created 20 different datasets.

To measure the predictive value of the bleeding score for the occurrence of PPH, we also estimated the area under the receiver operator characteristic (ROC) curve (AUC). We additionally calculated the sensitivity, the specificity, the Youden index (sensitivity + specificity—1) and the positive and negative predictive values for each cut-off of the bleeding score tested to identify the optimal decision threshold [19].

Finally, we performed a sensitivity analysis of the association between the bleeding score and the occurrence of PPH excluding the item ‘prior PPH’ in uni and multivariable regressions and measures of accuracy. This analysis was intended to avoid confusion bias since a history of PPH, beyond being potentially secondary to a hemostasis disorder, is a risk factor for uterine atony [20].

The study sample had a power > 80% to show a relative risk of 2 of PPH in women with a bleeding score ≥ one, with a bilateral alpha risk of 0.05. The associations quantified as relative risks and the test performance parameters of the bleeding score were expressed with their 95% confidence intervals. All tests were two-tailed with P values < 0.05 defined as statistically significant.

Data are reported in accordance with Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD) guidelines (www.tripod-statement.org) (see Additional file 1, reporting checklist).

All analyses and graphs were performed using the STATA software v13.1 (StataCorp, College Station, TX).

Results

Over the one-year and seven months study period, our analysis included overall a cohort of 2536 women with no known bleeding disorder, with an available bleeding score calculated prior to the childbirth and who delivered in the study center (Fig. 1).

Fig. 1

Study population

In the cohort study, 116 women (4.6%) experienced PPH. Among these women, 12 (10.3%) had blood losses ≥ 1000 mL, 12 (10.3%) were transfused with red blood cells and 14 (12.1%) had intrauterine tamponade. No women had arterial ligation or embolization or hysterectomy. Identified PPH risk factors are shown in the Table 2. Uterine atony was the leading cause of PPH for 76 women (65.5%), followed by perineal wound or surgical cause for 23 women (19.8%), placental retention for 16 women (13.8%) and missing cause for one woman. None of the women had a hemostasis disorder reported as the cause of PPH.

Table 2 Characteristics of women, pregnancy and delivery according to the occurrence of PPH (N = 2536 women)

The distribution of the bleeding score is detailed in the Table 3. Biological hemostasis assessment was more frequently performed among women with a high bleeding score, from 6.7% of the women with a score of zero to 33.3% of those with a score of four (Table 3). Only standard screening tests (platelet count, activated partial thromboplastin time and prothrombin ratio) were prescribed and all requested to be performed on admission to the labor ward or on the day of cesarean delivery. No woman was referred to a hemostasis specialist and no specific hemostatic management was planned for delivery. Finally, neuraxial anesthesia or analgesia was not administered differently according to the score (see Additional file 2, Table A1).

Table 3 Association between the bleeding score and items and PPH; hemostasis assessment according to the score

In the multivariable as in the univariate analysis, the incidence of PPH was significantly associated with the bleeding score (adjusted RR = 1.58; 95% CI, 1.23 to 2.02; P < 0.001) (Table 3). The risk of PPH increased exponentially as the bleeding score raised (aRR = 1.58, 95% CI 1.01 to 2.46 for a score of one, aRR = 2.11, 95% CI 0.86 to 5.20 for a score of two and aRR = 7.20, 95% CI 2.54 to 20.41 for a score of three compared with a score of zero; P < 0.001). Two bleeding items were significantly associated with the occurrence of PPH: bruises or hematomas and prior PPH.

The AUC for the accuracy of the bleeding score to predict PPH was very low: 0.56 [95% CI 0.52 to 0.61] (Fig. 2). The best cut-off of the bleeding score to predict PPH according to a maximized Youden index was ≥ one (Table 4).

Fig. 2

ROC curve for the association between the bleeding score and PPH (N = 2536 women). AUC = Area Under the receiver operator characteristic Curve; PPH = PostPartum Hemorrhage; ROC = Receiver Operator Characteristic. The low AUC observed indicates a poor performance of the bleeding score in predicting PPH

Table 4 Sensitivity, specificity, Youden index, PPV and NPV of abnormal bleeding scores for PPH

The sensitivity analysis excluding the item ‘prior PPH’ showed similar results to those of the main analysis (see Additional file 3,Tables A2 and A3). In this analysis, the AUC was 0.56 [95% CI 0.51 to 0.60] (see Additional file 4, Figure A1).

Discussion

The HEMSTOP bleeding score is significantly but weakly predictive of primary PPH in unselected parturients. Since the risk of PPH increases exponentially with the score, a high score ≥ 3 is a major independent risk factor for PPH, with a sevenfold increase in risk compared with a score of zero. Two bleeding items were weakly associated with PPH: tendency to bruises or hematomas and prior PPH.

Implications

Previous studies exploring the associations between various bleeding symptoms and the occurrence of PPH among unselected pregnant women showed that a history of epistaxis, bleeding after surgery and prior PPH were associated with PPH [12, 21]. Cutaneous bleeding, although being significantly associated with PPH in our study, was not highlighted in those previous studies.

Our findings are consistent with those of a recent retrospective cohort study conducted by Zec et al. and including 3588 unselected parturients [13]. The authors showed a significant crude association between a hetero-assessed HEMSTOP score ≥ two and the occurrence of a PPH ≥ 1000 ml and found similar performance parameters to ours for the score to predict PPH. In our study, through the relative risks and positive predictive values we calculated, we observed a significant association and a gradual relationship between a high bleeding score and the occurrence of PPH. This result shows that high scoring women have a great risk of PPH. On the contrary, negative predictive values of the score were close to (1-prevalence) and sensitivities low. This result indicates that a negative score does not allow to exclude PPH.

In another retrospective cohort study, the hetero-assessed HEMSTOP score effectively differentiated 38 patients with hemostatic disorders needing perioperative precautions according to a hemostasis clinic from 70 healthy volunteers [14]. In this study, the bleeding score cut-off of ≥ two showed the best prediction performance for screening MBDs with a sensitivity of 89.5% and a specificity of 98.6%. Also, in a recent prospective multicenter study including 931 women scheduled for surgery, the sensitivity and specificity of a hetero-assessed HEMSTOP score ≥ two were 82% and 80% respectively for the detection of a biological hemostasis abnormality consistent with an increased bleeding risk [15]. However bleeding disorders of unknown cause were not considered in this analysis.

By contrast, we had no data on characterized MBDs among our study population and we thus do not know their relation with the bleeding score and PPH. We yet found a significant association between the severity of the bleeding phenotype, calculated with the score, and the occurrence of PPH. This association was weak, with a very low AUC, which could be explained by two factors. First, most PPH occur from obstetrical causes in healthy women and are thus not related to a bleeding tendency [5]. Also, some women with MBDs do not suffer from PPH, probably due in part to the hypercoagulability of pregnancy [3]. Regarding the best cut-off to predict PPH, a score ≥ one maximized the Youden index in our study. But this score was reached by 25.8% of our study population and was associated with only a 1.5-fold increase in the risk of PPH compared with a score of zero. In comparison, the cut-off of ≥ two was reached by 4.7% of the women included, was associated with a two-fold increase of PPH and presented a greater specificity. A score cut-off ≥ two therefore seems more relevant in terms of clinical importance for a reasonable rate of patients in our study population. However, given the low prediction performance of the HEMSTOP score for the occurrence of PPH, the score should probably not be used as a standalone tool but rather integrated into a risk‐stratified approach. Quantifying the risk of PPH with a pre-delivery clinical prediction model integrating other risk factors could help clinicians to identify women at greatest risk [22]. In the 1.3% women of our study population scoring three or more, we found a sevenfold increased risk of PPH compared with a score of zero. This finding strongly suggests that these women should be classified with high risk of PPH, irrespective of other risk factors.

The binary categorization of women with or without established diagnosis of MBD among those with a bleeding diathesis is biologically questionable and of little clinical relevance. First, three quarters of women with MBDs have normal results in hemostasis tests and are therefore diagnosed with bleeding disorder of unknown cause [8]. Second, in common MBDs such as type I von Willebrand disease and carriers of hemophilia, there is a mismatch between coagulation factor levels and bleeding phenotype [23, 24]. Finally, low yet non-deficient hemostatic capabilities and blood group O are independent risk factors for severe PPH [9]. In our study, we found a quantitative association between the HEMSTOP score, initially designed as a screening tool for MBDs, and the risk of PPH. According to our results, the HEMSTOP bleeding score could be considered as a quantitative risk factor for obstetric hemorrhage in unselected parturients rather than a binary screening tool.

In accordance with national and international guidelines for pre-interventional hemostatic assessment, hemostasis tests for MBDs screening were not routinely ordered by anesthesiologists in women without bleeding symptoms in our study, with only 6.7% of them explored [10, 11, 16]. On the contrary, we observed that only a quarter of the women with a significant bleeding diathesis, i.e. a HEMSTOP score ≥ three according to national guidelines, had biological hemostasis testing prescribed [16]. These results contrast with guidelines which recommend ante-partum hemostasis assays and specialist advice in these women [10, 11, 16]. Several reasons can explain the non-adherence with guidelines we observed. The diagnostic process of MBDs is complex and challenging in terms of resources and techniques, particularly in a limited time frame and during the third trimester of pregnancy [25]. Also, anesthesiologists may question the ability of the bleeding score to detect clinically-relevant MBDs. Finally, the safety and medicolegal considerations for neuraxial anesthesia, a major concern for anesthesiologists, are, apart from platelets, limited by the absence of safe, validated hemostatic thresholds [26]. Thus, in our study, neuraxial anesthesia was commonly administered, even in women with a high bleeding score. Among the few women who were explored for significant bleeding diathesis, only standard tests were performed. None of these women received specific peri-interventional hemostatic management. This result is in accordance with a meta-analysis on perioperative management after standard hemostasis screening in 27 606 patients which showed that less than 0.6% had a corrective treatment given [10].

Considering our findings and existing literature, we suggest a customized and pragmatic application of the HEMSTOP questionnaire and ante and peripartum management based on the bleeding score. Firstly, parturients with a HEMSTOP score of two could undergo standard hemostasis tests, including von Willebrand factor activity, to exclude the most serious or frequent bleeding disorders [25]. If these tests are considered as normal, these women could simply benefit from prophylactic administration of tranexamic acid during delivery, as recently suggested by international guidelines [11, 27]. For those with abnormal results or a bleeding score of three or higher, a consultation with a hemostasis specialist should be considered to further evaluate their condition and to establish an individual care protocol for delivery, in collaboration with anesthesiologists and obstetricians [10, 11]. For women scoring three or more and therefore at high-risk of PPH, the timing and mode of delivery in a high-level of maternal care unit should probably be agreed by the multidisciplinary team [3]. Coagulation monitoring, early recognition of PPH, immediate availability of specific hemostatic therapies and transfusion products may be discussed as standards of care. Secondly, it would be advantageous to conduct the HEMSTOP scoring and thus subsequent evaluations early in pregnancy within the follow-up by obstetricians or midwives. Results should be made available to anesthesiologists. Lastly, it would be prudent to incorporate the HEMSTOP score into a broader assessment of PPH risk that includes other risk factors [22].

Strengths

Our study has several strengths. All women included completed the HEMSTOP questionnaire prior to childbirth, a brief and straightforward questionnaire that is easily administered across the general population. Also, self-assessment of bleeding symptoms reduces the workload for healthcare professionals. Incidence of primary PPH in our study (4.6%) was similar to the national profile for level III maternity units (i.e. with a neonatal intensive care unit within the hospital) [28]. We excluded women with a previous diagnostic of bleeding disorder because of a potential confounding factor being a specific delivery management. To avoid another confusion bias, we performed a sensitivity analysis excluding the item on prior PPH.

Limitations

Our study also has some limitations. The specific reporting of bleeding symptoms varies depending on the population studied and on the mode of administration of the questionnaire [13, 15, 29, 30]. Self-assessment of the bleeding diathesis, as in our study, may lead to a misclassification of women due to misunderstanding, but interviews are also subject to response biases. Likewise, the bleeding history may vary across individuals and populations due to age, the number of hemostatic challenges previously encountered, ethnicity, education and cultural background [29]. Therefore, the monocentric nature of our study may question the generalizability of our results to other populations and settings. Also, we cannot exclude a potential selection bias in our study. Indeed, 22% of the women from our source population had a missing, unanswered or inconsistently answered questionnaire. These women not included may have more social markers such as foreign nationality or low level of education than our study population. Since socioeconomic status is both a PPH risk factor and slightly correlated with the presence of a MBD, the association we investigated may have been biased through our recruitment [31]. In addition, women from our source population who had their pre-anesthesia consultation performed during an antepartum hospitalization because of a pregnancy complication were not recruited. We may therefore have missed women with antepartum hemorrhage secondary to a bleeding tendency [3]. Such selection biases may also limit the external validity of our findings. However, the direction of those biases would be towards a decrease of the association between the bleeding score and the occurrence of PPH, which does not challenge our results. Finally, despite being frequently observed in women with MBDs, we could not study secondary PPH because we had no data on its occurrence [3]. Thus, a composite outcome of antepartum, primary or secondary PPH could have added value in a comprehensive assessment of bleeding tendency.

In conclusion, the calculation of the HEMSTOP bleeding score in unselected pregnant women is of added value in the assessment of primary PPH risk. Additional research is necessary to refine the mode of administration of the questionnaire and to optimize the management strategies for parturients who register a high bleeding score.