Obstetric hemorrhage risk assessment tool predicts composite maternal morbidity

Obstetric hemorrhage is one of the leading preventable causes of maternal mortality in the United States. Although hemorrhage risk-prediction models exist, there remains a gap in literature describing if these risk-prediction tools can identify composite maternal morbidity. We investigate how well an established obstetric hemorrhage risk-assessment tool predicts composite hemorrhage-associated morbidity. We conducted a retrospective cohort analysis of a multicenter database including women admitted to Labor and Delivery from 2016 to 2018, at centers implementing the Association of Women’s Health, Obstetric, and Neonatal Nurses risk assessment tool on admission. A composite morbidity score incorporated factors including obstetric hemorrhage (estimated blood loss ≥ 1000 mL), blood transfusion, or ICU admission. Out of 56,903 women, 14,803 (26%) were categorized as low-risk, 26,163 (46%) as medium-risk and 15,937 (28%) as high-risk for obstetric hemorrhage. Composite morbidity occurred at a rate of 2.2%, 8.0% and 11.9% within these groups, respectively. Medium- and high-risk groups had an increased combined risk of composite morbidity (diagnostic OR 4.58; 4.09–5.13) compared to the low-risk group. This established hemorrhage risk-assessment tool predicts clinically-relevant composite morbidity. Future randomized trials in obstetric hemorrhage can incorporate these tools for screening patients at highest risk for composite morbidity.

Statistical analysis. Patient variables between low-, medium-, and high-risk postpartum hemorrhage cohorts defined from the database were compared using Chi-square for categorical variables, independent samples t-test and one-way ANOVA for parametric continuous variables, and Mann-Whitney U/Wilcoxon ranksum test for nonparametric continuous variables.
A composite outcome incorporating obstetric hemorrhage (defined as EBL ≥ 1000 mL by gravimetric method), blood transfusion, ICU admission, and/or additional hemorrhage-related complications (hysterectomy, dilation and curettage) was then compared between the low-, medium-, and high-risk postpartum hemorrhage cohorts. Respective incidences of composite outcome and each individual component by hemorrhage cohort were reported. Statistical diagnostics including sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) and diagnostic odds ratio (dOR = relative risk/RR) were calculated for the composite hemorrhage-related morbidity outcome between medium-and low-risk cohorts, high-and low-risk cohorts, and combined medium-plus high-risk relative to low-risk hemorrhage cohorts.
All statistical analysis was performed using SAS version 9.4 (SAS Institute Inc., Cary, NC) and R version 3.6.2 (R Foundation for Statistical Computing, Vienna, Austria). A two-sided p-value less than 0.05 was considered statistically significant.
Ethics approval. IRB approval through George Washington University, waiver of consent given retrospective nature.

Results
In this cohort of 56,903 women, 14,803 (26%) were categorized as low-risk, 26,163 (46%) as medium-risk and 15,937 (28%) as high-risk of obstetric hemorrhage. Maternal characteristics, by cohort, are summarized in Table 2. Increasing risk cohort was associated with multiple variables not used in stratification. Notably, older age, higher BMI, rate of cesarean delivery, number of prior cesareans, and preeclampsia were all associated with categorization in an increased risk cohort (all respective p < 0.0001; Table 2). Obstetric hemorrhage occurred at a rate of 2.1%, 7.6% and 11.4% within the low-, medium-and high-risk groups, respectively. Composite maternal morbidity was found to occur in 2.2% of the cases characterized as low-risk, 8.0% of medium-risk and 11.9% of high-risk (Table 3). We observed that blood transfusion, ICU admission, and/or additional hemorrhagerelated complications (hysterectomy, dilation and curettage) occurred at a frequency of 0.1% 0.4% and 0.5% over the respective hemorrhage incidences (defined as blood loss of 1000 cc or greater) of 2.1%, 7.6% and 11.4%, respectively. We observed a significantly increased risk of composite morbidity for both the medium-(dOR 3.80; 95% CI 3.37-4.27) and the high-risk group (dOR 5.93; 5.26-6.68) when compared individually to the low-risk group. Importantly, those in the combined medium-and high-risk groups were found to have a 4.58 times higher odds of experiencing a composite morbidity than the low-risk group (p < 0.0001; Table 4). When comparing the combined medium-and high-risk groups to the low-risk group, NPV for composite morbidity was 0.98 and positive predictive value (PPV) was 0.10 ( Table 4).

Discussion
These findings provide insight into the ability of an established hemorrhage risk-assessment tool to predict hemorrhage-related morbidity. This study demonstrates a significantly increased risk of composite hemorrhagerelated morbidity for patients categorized as both medium and high risk for hemorrhage, when compared to the low risk group.
In the United States, the overall rate of severe maternal morbidity (SMM) has increased over two-hundred percent since 1993, driven primarily by an increased utilization of blood transfusion 25 . Obstetric hemorrhage remains one of the leading causes of SMM and accounts for over half of all identified SMM events 1 . One of the most common preventable factors in SMM is failure to identify patients who are at high risk for obstetric complications, which can significantly delay diagnosis, management, and treatment 26 . In response, risk assessment programs and obstetric care bundles have been implemented on a national level to better identify patients who are at the highest risk for obstetric hemorrhage 19,[27][28][29][30][31][32] .
The AWHONN risk assessment tool has emerged as a simple, easy to implement, and low-cost hemorrhagerisk prediction tool that is now widely accepted across the United States. It was designed as a risk-stratification method, identifying patients with low, medium, or high risk for obstetric hemorrhage 29 . Prior studies have suggested that this tool works moderately well to identify patients at highest risk for both obstetric hemorrhage and transfusion 17 , and when implemented, may reduce rates of maternal morbidity 33 . Our study demonstrates that, in addition to predicting obstetric hemorrhage, the AWHONN risk assessment tool identifies women at highest risk for hemorrhage-related morbidity, which may be more clinically relevant. However, it is key to note that within this study, morbidity was primarily driven by PPH. With an excellent diagnostic odds ratio (OR), those categorized as high or medium risk by the AWHONN tool can receive extra attention in preparation for hemorrhage and hemorrhage-associated morbidities. Additionally, with an NPV value of 0.98, this admission questionnaire functions well as a screening test for hemorrhage-related morbidity, reliably separating those who are at low risk, as 98% of those classified as low-risk did not experience a hemorrhage-related morbidity. Though our PPV indicates that a patient classified as medium-or high-risk only has a 9% chance of experiencing a hemorrhage-related morbidity-due to a somewhat rare outcome-this follows with general guidelines for a screening test, with a high NPV of 0.98 and low risk for categorization of patients in a higher risk group. Importantly, this sheds light on the poor diagnostic accuracy of these tools at predicting either hemorrhage or hemorrhage-associated morbidity and sheds light on the need for a more accurate risk assessment tool.
Severe maternal morbidity and mortality remain important public health concerns, affecting 2.9/1000 births, with the majority of morbidity attributed to obstetric hemorrhage 1 . Understanding the ability of existing hemorrhage risk-prediction tools to predict hemorrhage-related morbidity will increase awareness of a patient's highrisk status, encourage early management of hemorrhage, and potentially allow healthcare providers to selectively implement preventative hemorrhage interventions (e.g., active management of the third stage of labor). Further understanding of how this tool predicts those at highest risk for morbidity would also improve our ability to describe and study high-risk patient populations. Importantly, data from this tool should be abundant, as it is already in use on a national level, and is easy to implement and track. Especially in health centers with low levels of maternal care, this tool has the potential to efficiently allocate more resources to monitor those who have an increased risk of hemorrhage. Lastly, clinically relevant composite hemorrhage morbidity and baseline rates are essential to design randomized clinical trials assessing interventions to reduce obstetric hemorrhage.
Strengths of this study include the large number of patients and the use of a multicenter database to capture a diverse maternal population. Our dataset is contemporary, collected from 2016 to 2018, making it generalizable to our modern patient population. The retrospective design of our study poses several limitations, including the loss of complete data, which may have disproportionately affected more complicated, and therefore more hectic, deliveries, where data may not have been entered in entirety. Additionally, hemorrhage risk scores were calculated for each patient at the time of admission to labor and delivery in a prospective manner. An important limitation to our study includes potential error in hemorrhage risk score calculation. Patients who were classified as low risk were identified to have risk factors that would place them in a medium or high risk category. This error is either an error in score calculation or patient demographic data available within our database and www.nature.com/scientificreports/ is a limitation of our retrospective analysis. This might lead to incorrect charting of bedside procedures such as bedside curettage or uterine tamponade device placement. Use of EBL may lead to incorrect categorization of hemorrhage status; however, since this misclassification bias was across all groups this is something that likely did not impact our findings. As this study was retrospective and across multiple clinical sites, technique for measuring EBL was not standardized between hospitals or providers. In addition, it is possible that transfusion may have occurred before delivery as time of transfusion was not documented as a variable thus, potentially changing the risk stratification of participants. Additionally, data were collected regarding ICU admission, however there was no ability to discern the diagnosis leading to admission. It is possible that there was misclassification of hemorrhage risk categories upon admission (for example, a patient with a prior cesarean was mistakenly categorized as low risk) because the hemorrhage risk scores calculated by nurses on labor and delivery and not verified by a physician. Furthermore, there is no universal definition of hemorrhage-related maternal morbidity, which limits the generalizability of our study in comparison to other works, as well as national and global statistics, which may have used other definitions. Lastly, the AWHONN tool recommends that periodic risk assessments should be conducted by facilities that choose to use this tool. Currently, there is no standardized way to evaluate this tool thus its effectiveness may differ by location.
In conclusion, our study demonstrates that the AWOHNN hemorrhage risk-prediction tool not only identifies patients at highest risk for obstetric hemorrhage, but also can be used as a screening tool for those at risk for hemorrhage-related morbidity. Women who were categorized as being at high risk for obstetric hemorrhage using the AWHONN hemorrhage risk assessment tool were six times more likely to experience hemorrhage-related morbidity compared to those who were low-risk. One setback of this tool is the low PPVs indicate a high false positive rate thus, there needs to be additional research done to develop better tools to optimize prediction of obstetric hemorrhage risk. Furthermore, morbidity was primarily driven by PPH, hence the morbidity outcome among the different groups were similar. Thus, further studies looking to assess the accuracy of prediction tools can consider using PPH as the primary way to compare different risk categories.

Data availability
Data will not be available for direct download or analysis given not included in original IRB submission Of note, findings were presented at the 40th Annual SMFM Meeting, February 3-8, 2020; Grapevine, TX.