Inability to predict postpartum hemorrhage: insights from Egyptian intervention data

Background Knowledge on how well we can predict primary postpartum hemorrhage (PPH) can help policy makers and health providers design current delivery protocols and PPH case management. The purpose of this paper is to identify risk factors and determine predictive probabilities of those risk factors for primary PPH among women expecting singleton vaginal deliveries in Egypt. Methods From a prospective cohort study, 2510 pregnant women were recruited over a six-month period in Egypt in 2004. PPH was defined as blood loss ≥ 500 ml. Measures of blood loss were made every 20 minutes for the first 4 hours after delivery using a calibrated under the buttocks drape. Using all variables available in the patients' charts, we divided them in ante-partum and intra-partum factors. We employed logistic regression to analyze socio-demographic, medical and past obstetric history, and labor and delivery outcomes as potential PPH risk factors. Post-model predicted probabilities were estimated using the identified risk factors. Results We found a total of 93 cases of primary PPH. In multivariate models, ante-partum hemoglobin, history of previous PPH, labor augmentation and prolonged labor were significantly associated with PPH. Post model probability estimates showed that even among women with three or more risk factors, PPH could only be predicted in 10% of the cases. Conclusions The predictive probability of ante-partum and intra-partum risk factors for PPH is very low. Prevention of PPH to all women is highly recommended.


Introduction
Primary postpartum hemorrhage (PPH) is the single largest contributor to maternal mortality worldwide. Throughout Africa and Asia, hemorrhage accounts for 30% or more of all maternal deaths, most of which is PPH [1]. Even if effective antenatal screening existed, hemorrhage often occurs in women with no identifiable antepartum risk factors. Risk factors in the intra-partum period may provide opportunity for intervention, or for active management in the early stages of PPH.
Despite the severe burden of PPH, few studies have examined risk factors predicting PPH in developing countries. The primary objective of this study is to investigate the ability to predict PPH by early screening of women with selected risk factors. We hypothesize that a combination of ante-partum and intra-partum risk factors will be associated with incidence of PPH, and that these factors will have utility for screening and clinical management of labor. The secondary objective of this study is to identify the components of active management of the third stage of labor (AMTSL) that have the greatest impact on PPH incidence. AMSTL is a component of internationally accepted clinical protocol for obstetric management consisting of four components: use of uterotonics, uterine massage, early cord clamping, and cord traction. The importance of AMSTL in PPH is well established, yet the relative importance of each component of the process has not been examined.

Previous Research
Previous research has identified demographic, ante-partum, and intra-partum risk factors that are associated with risk of hemorrhage, but these associations are not predictive. In addition, the majority of these studies focus on women in the United States or Europe who may face different delivery circumstances from women in developing countries. An extensive literature review on PubMed that investigated ante-partum and intra-partum risk factors for PPH was carried out. All of the risk factors found are described below; citations are only for the most recent article published on the topic.
Despite all that is known about PPH risk factors, risk factors associated with primary PPH are not useful in predicting patients who will continue to bleed after firstline therapy [27]. And while previous studies have identified significant associations among ante-partum and intra-partum risk factors, Tsu's paper (1993) is the only published work examining predictive probability of screening for PPH. While predictive values were less than 7%, she found that screening tests for maternal height, parity, and obstetric history could identify one third of PPH cases [28].

Methods
We conducted a secondary data analysis using data from prospective, hospital based operations research in Egypt in 2006. The objective of the original operations research was to compare the current protocol for the AMTSL with the use of 600 μg of misoprostol. The conclusion was that misoprostol should be considered for inclusion in the AMTSL protocol where oxytocin and or ergometrine are not consistently and appropriately used during the 3 rd stage of labor. The original study was analyzed as a pre-/post-intervention study. In the current study, we analyzed all women who experienced PPH as 'cases' and all women who did not experience PPH as 'controls.' Details on design data, collection, and results of the original study are reported elsewhere [29]. Over a 6-month period, 2532 pregnant women were recruited into the study from three university hospitals. Inclusion criteria were: anticipated singleton vaginal delivery, gestational age greater than 36 weeks, and ability to give informed consent. Women who delivered by cesarean section or women with missing information on delivery type were excluded from this analysis, leaving a total of 2510 women.
The primary outcome of interests was incidence of PPH. In accordance with international clinical protocol, PPH was defined as blood loss greater than 500 ml during the first 4 hours after delivery. Following delivery and the clamping of the umbilical cord, a calibrated drape was placed under the women's buttocks for blood loss measurement. This drape remained under the women's buttocks for 4 hours, during which time blood loss was read cumulatively every 20 minutes. Total blood loss was established after the bleeding had stopped. In the original study, women in the misoprostol group were 70% less likely to bleed 500 ml or more (OR = 0.30, 95% CI 0.16-0.56) compared to those in the current practices group. In addition, the women in the misoprostol group needed additional interventions because of postpartum hemorrhage only 11% of the time, compared to 73% among the current practices group.
To evaluate the value of risk factors for screening and labor management, attending physicians collected both ante-partum and intra-partum information using questionnaires that were completed at the time of admission, during labor, after delivery for up to 4 hours, and before discharge. Variables collected at the time of admission for delivery include maternal age (30 years or older), education (literacy), nulliparity, previous antenatal care (no care or high number of visits compared), PPH in a previous pregnancy, history of obstetric complications, intact membranes, anemia, and cervical dilation. Anemia was measured by blood collected after enrollment, and defined as ante-partum hemoglobin at or below 11 mg/dl. Intra-partum variables include vaginal delivery with instruments, episiotomy, labor augmentation, retained placenta, vaginal tears, fetal macrosomia (> 3500 g), length of 1 st and 2 nd stage, and absence of AMTSL. Additionally, the absence of each component of AMTSL was measured separately, including use of uterotonics in the 3 rd stage, early cord clamping, cord traction, and uterine massage.
Ethical review for the study was provided by the Committee for the Protection of Human Subjects at the University of California, Berkeley and the Egyptian ethics committees. The Healthy Mother Healthy Child Project, which was implemented by John Snow Inc. under United States Agency for International Development (USAID) Contract No. 263-C-00-98-0041-00, was responsible for study implementation. Venture Strategies for Health and Development (VSHD) donated the misoprostol used in the operations research (Misotac; Sigma Co., Moubarak industrial City, First Quarter Quesna, Egypt).
Data was entered using Epi-Info version 2002 (Centers for Disease Control and Prevention, Atlanta, Georgia, USA), and statistical analysis was performed with the STATA ® version 8.0 (StataCorp, LP, College Station, Texas, USA). We constructed separate, multiple logistic regression models for ante-partum and intra-partum variables associated with PPH, but not as a stepwise regression. Model 1 included demographic and antepartum variables; Model 2 included intra-partum variables and one variable to represent AMTSL; Model 3 is a variation of Model 2 with AMTSL separated out by each component; and the final model, Model 4, was the combination of demographic/antepartum, intra-partum, and separated AMTSL factors. Separating each of the four components of AMTSL allowed us to assess the relative importance of each component. Odds ratios were provided as the measure of effect. Results from the final model, Model 4, were used in post-model estimation of predictive probability of PPH. In addition, we created cumulative scores based on the number of risk factors (1, 2, 3, or 4+). Sensitivity and specificity were calculated for the cumulative risk factor probabilities and for the combined measure of AMTSL. Receiver operating curves (ROC) were derived from these sensitivity and specificity estimates. Significant associations were established at p-values < 0.05.

Results
The incidence of PPH among the 2510 singleton vaginal deliveries was 3.71% (n = 93). The distribution of demographics, ante-partum, and intra-partum risk factors are listed in Table 1. Odds ratios from bivariable analyses of risk factors and incidence of PPH are reported along with associated p-values and 95% confidence intervals (CIs). Among unadjusted ante-partum variables, history of PPH in a previous pregnancy increased the risk of PPH by almost 69 times. Women who had one antenatal care (ANC) visit, those with no care, and those with higher levels of care also had an increased risk of PPH. No other ante-partum risk factors were significant in bivariable logistic regressions.
Among intra-partum risk factors, episiotomy, labor augmentation, vaginal tears, macrosomia, and length of the first and second stage of labor were each associated with an increased odds of PPH. Women who did not receive AMTSL were 8 times more likely to have PPH. When assessed separately, omission of each component of AMTSL was also significantly associated with PPH risk. Table 2 shows separate adjusted multivariable models for ante-partum risk factors (Model 1), intra-partum risk factors, (Models 2 and 3) and a full model (Model 4) containing both sets of variables. Components of AMTSL were included separately (Model 3) and together (Model 2) in intra-partum models. We used post model probability estimates to predict the probability of developing PPH with each one of the statistically significant risk factors in Model 4 ( Figure 1).    History of PPH, non-use of cord traction, and retained placenta each had a predictive probability of 20% or higher.
We examined whether the percentage of women with 0, 1, 2, 3, or 4+ significant risk factors differed between those who had PPH and those who did not (Figure 2). Although percentages appear different for those with 0, 1, 3, and 4+ factors, the difference was only significant among women with 3 risk factors. Post model probability estimates showed that by screening and diagnosing women with three risk factors, we can identify 10% of women who will develop PPH ( Figure 3). Similarly, screening for 4 risk factors may identify 31% of women who will develop PPH. Table 3 presents the relative sensitivity, specificity, likelihood ratio, and percent correctly classified for women with 1, 2, 3, or 4+ risk factors for PPH. As the number of identified risk factors increases, so does percentage correctly classified. Perhaps because of the low incidence of PPH in the study population, the likelihood ratios are low for all risk factors. Women with 4+ identifiable risk factors had an LR of above 12. Figure 4 displays the relative utility of screening women with 1, 2, 3, or 4+ significant risk factors for PPH. Sensitivity, or the proportion of true positives, is high for using a screening cut point of 1 or 2 risk factors. However, the specificity, or proportion of true negatives, is low at 9% and 47% respectively. This means that if 1 risk factor was used as a cut point, 100% of those who would go on to have PPH would be captured. However, only 9% of those identified as positive would   actually be positive. Thus you would have many people who would be screened as at risk who in reality may not be. Increasing the cut point to 2 risk factors would decrease sensitivity only marginally, but would increase specificity to 47%. Similarly, using 3 risk factors as a cut point would result in few false positives, but would also decrease sensitivity, meaning that some cases would be missed.

Discussion
Results of this study show that most demographic and ante-partum risk factors have little association with development of PPH, with the exception of history of PPH, low hemoglobin, and high usage of ANC. In addition, the predictive probability of ante-partum and intra-partum risk factors for PPH is low, although women with four or more identified risk factors had a predictive probability of over 30%. This probability is higher than those observed by Tsu in Zimbabwe, and indicates that perhaps prediction of PPH is more significantly influenced by ante-partum variables [28].
The results of this analysis confirm our hypothesis that in vaginal, singleton deliveries, both ante-partum and intra-partum factors are associated with the incidence of PPH, but these associations are not predictive. Predictive probability of PPH based on any risk factor remains low. The exception is history of PPH, which had a predictive probability of 70% for incidence of PPH. A high number of antenatal care visits was also associated with increased risk, likely because this serves as a marker for high-risk pregnancies. The utility of screening based on number of risk factors should be considered in conjunction with benefits of positively identifying all possible cases, and costs of over treating.
The findings of this study indicate that AMTSL significantly reduces the risk of PPH. AMTSL has been recommended as a routine management for vaginal deliveries in hospital settings [30,31], but it is rarely used in Egyptian hospitals. An analysis of the management of the third stage of labor in an Egyptian teaching hospital showed that only 15% of the women during the study period received correct AMTSL [32].
To our knowledge, this is the first analysis to examine the utility of each component of AMSTL separately. Our analysis of the components of AMTSL indicates that nonuse of uterotonics in AMTSL had the strongest association with increased odds of PPH after adjustment. Cord traction was also significantly associated with PPH after controlling for other components. Uterine massage and early cord clamping, however, appear to have no independent effect on risk of PPH, which could be due to a lack of variability between each component. We investigated the variability of early cord clamping and cord traction and found that there were some women who received one and not the other. As a result, we decided to include both components in the model. This touches on an overall limitation of this analysis. As many of the identified risk factors are correlated with one another, collinearity of the variables may have biased our inferences. This limitation would likely lead to conservative effect estimates, as we may be over controlling for related factors. In order to account for covariance, we used robust estimator of standard errors.
Low predictive probability values may have been the result of very low incidence of PPH in the study; the incidence of PPH in the study hospitals was 3.7%. Administration of uterotonics and AMTSL in the study hospitals likely contributed to the low rate of PPH.
Despite a 52% reduction in the Egyptian maternal mortality ratio (MMR) during the 1990s, PPH remained the main cause of death. According to the National Maternal Mortality Study [33], primary PPH alone was responsible for 27% of all maternal deaths, making it the single most important cause of maternal death in Egypt. High parity,  low socio-economic status of women, lack of skilled birth attendants, and mismanagement of PPH contribute to high maternal deaths in Egypt [34]. This study advances knowledge on the utility of screening for PPH in a clinical setting. This research has primary implications for screening and subsequent management of those at highest risk. Many of the identified risk factors cannot be prevented or altered, but women who present for delivery with these factors can be closely monitored, and when possible, intra-partum interventions that are significant risk factors should be avoided.

Conclusion
While no single risk factor or combination thereof can reliably predict risk of hemorrhage after delivery, combinations of risk factors can be used for screening and management. Given the limited utility of screening for PPH, program planners and policy makers need to focus on the development and enforcement of protocols for prevention and treatment of PPH by considering the capacity and limitations of each context. The need for further research and policy work surrounding prevention efforts is especially important due to the low predictability of PPH.