The association between body mass index and postpartum hemorrhage after cesarean delivery

We aimed to evaluate the association between obesity and postpartum hemorrhage (PPH) after cesarean delivery (CD). This was a retrospective cohort study using a multicenter database of 20 hospitals in the United States. We analyzed 27,708 patients undergoing CD from 2015 to 2019. The exposure of interest was BMI, and the primary outcome was PPH (estimated blood loss [EBL] ≥ 1000 mL). Simple logistic regression was used to evaluate the relationship between obesity and intrapartum complications. Multivariable logistic regression was used to adjust for any confounding demographic variables. Hosmer and Lemeshow’s purposeful selection algorithm was adapted to develop a multivariable logistic regression model of PPH. Analyses were conducted using STATA 16.1 (College Station, Texas) with p ≤ 0.05 considered significant. BMI exerted a significant effect on the frequency of PPH (p = 0.004). Compared to patients with BMI 18.5–24.9 kg/m2, patients with BMI between 25 and 59.9 kg/m2 had an increased odds of PPH. The odds of PPH in patients with BMI > 60 kg/m2 was not increased compared to patients with BMI 18.5–24.9 kg/m2. Obesity was associated with a decreased odds of blood transfusion (aOR 0.73, 95% CI 0.55–0.97). In conclusion, higher BMI was associated with PPH yet a lower odds of transfusion after CD.

for missing BMI at delivery, approximately 65% were missing data for ≥ 60% of analysis-relevant variables. An additional 2374 patients in the dataset with vacuum-or forceps-assisted delivery were also excluded to ensure that patients who had a vaginal delivery were not included in the analysis. A total of 27,708 patients were eligible for inclusion in the analysis. CD were reported by 20 hospitals across the Eastern (N = 5; 18.2%), Western (N = 7; 48%), and Central (N = 8; 33.8%) United States. The requirement for informed consent was waived by The George Washington University Committee on Human Research Institutional Review Board due to the retrospective nature of the study, and all data were de-identified prior to analysis. The study was determined to be research that is exempt from IRB review by The George Washington University Committee on Human Research Institutional Review Board under Department of Health and Human Services (DHHS) regulatory category 4 (IRB number NCR203136). All methods were carried out in accordance with relevant guidelines and regulations. No experimental protocols were performed as this was a retrospective study.
The main exposure of interest was BMI at time of delivery. BMI at delivery was sorted into six categories (BMI below 18.5, 18.5 to 24.9, 25.0 to 29.9, 30.0 to 34.9, 35.0 to 39.9, and above 40) according to guidelines from the National Institutes of Health and World Health Organization 1,15 , of which 5 were employed for descriptive and demographic analysis (Table 1). In the primary analysis, maternal BMI was treated as a dichotomous categorical variable to evaluate obesity (BMI ≥ 30 kg/m 2 ). The primary outcome was PPH, defined as EBL ≥ 1000 mL for the purpose of analysis. Additional outcomes such as estimated gestational age (EGA) less than 37 weeks, intrapartum complications including pre-eclampsia and chorioamnionitis, abnormal hematologic parameters including thrombocytopenia, and pharmacologic interventions including oxytocin use were evaluated as secondary outcomes. Descriptive statistics included counts, proportions, means, and standard deviations calculated as necessary using standard definitions. To evaluate the association between BMI classification and other patientlevel variables, we used chi-square or Fischer's exact test (when N ≤ 5) for categorical variables and analysis of variance for continuous variables 16 .
We employed simple logistic regression to evaluate the relationship between obesity and intrapartum complications, including PPH. We used multivariable logistic regression to adjust for any confounding demographic characteristics that were associated with obesity at the 0.1 alpha level. To determine the demographic and intrapartum variables most associated with PPH, we adapted Hosmer and Lemeshow's purposeful selection algorithm to develop a multivariable logistic regression model of PPH 17,18 . Candidate variables were screened with simple logistic regression, using the Wald test and a p-value cutoff of ≤ 0.25 to exclude variables from consideration. During the iterative process of variable selection, covariates were sequentially removed from the model if they were not significant at the 0.1 alpha level and did not exert a confounding influence. We tested for multicollinearity between covariates by determining the variance inflation factor. A literature review was also used to support the choice of covariates as potential confounders influencing the odds of PPH [19][20][21][22][23] .
EBL was not reported for 1918 deliveries (6.92%), maternal age was not reported for 227 deliveries (0.82%), and EGA was not reported for 451 deliveries (1.63%), with approximately 9% of deliveries missing any combination of these variables. Because these data could not be assumed to be missing completely at random, we treated them as missing at random and employed multiple imputation by chained equations with 10 total imputations to account for statistical uncertainty while addressing missing data 24 . Missing data were also observed for race/ ethnicity, marital status, insurance type, and parity. Missing data from these variables were included in the multivariable models as dummy variables. All analyses were conducted using STATA 16.1 (College Station, Texas) with p ≤ 0.05 considered significant.

Results
Our cohort included 27,708 CD from 20 hospitals in the Universal Health Services (UHS) system across the United States. The majority of patients included in our study were White (55.0%) and between ages 20 and 34 years (75.5%); 49.1 percent were married. A large proportion of the patients in our cohort had private insurance (42.4%) and delivered at hospitals in the West Coast (48.0%). The majority of patients (79.9%) had no prior CD. We compared demographic and medical characteristics across 5 BMI categories, with several differences noted between groups. The prevalence of asthma, pre-gestational diabetes, and chronic hypertension increased as BMI increased between groups (p < 0.001 for all) ( Table 1).
Concerning our primary outcome of interest, obesity in its dichotomized form (defined as BMI ≥ 30 kg/m 2 ) did not significantly influence the odds of EBL ≥ 1000 mL. Increasing BMI category was, however, associated with EBL ≥ 1000 mL (p < 0.001) ( Table 2). While categorical obesity did not influence the odds of EBL ≥ 1000 mL, we conducted a sensitivity analysis to gauge the extent to which the extremes of BMI, relative to BMI 18.5-24.9 kg/ m 2 , influence the odds of EBL ≥ 1000 mL in this population. The odds of EBL ≥ 1000 mL increased with BMI between 25 and 59.9 kg/m2 relative to that observed in patients with BMI 18.5-24.9 kg/m 2 in a multivariable model (p = 0.004). The greatest odds of EBL ≥ 1000 mL was observed among patients with BMI 50-59.9 kg/m 2 (aOR 1.69 [1.28-2.21]), relative to patients with BMI 18.5-24.9 kg/m 2 (  (Table 3).
We found significant associations between BMI category and the frequency of secondary maternal outcomes in our cohort including pre-eclampsia (p < 0.001), gestational diabetes (p < 0.001), and any antibiotic use (p = 0.001) ( Table 4). Considering secondary neonatal outcomes, BMI category was significantly associated with preterm delivery (p < 0.001) and NICU admission (p < 0.001) (    (2000;1999). C Excluded during prescreening at the α = .25 level. D Excluded during the iterative process at the α = .10 level.  (Table 6).

Discussion
In our cohort of 27,708 CD across the United States, we observed an association between increasing BMI and EBL ≥ 1000 mL after adjusting for maternal demographic characteristics known to be associated with PPH. The observation that further discretized BMI was associated with increased odds of EBL ≥ 1000 mL while categorical obesity was not suggests that the BMI cutoff of ≥ 30 kg/m 2 BMI to define obese versus non-obese is not itself predictive of PPH. We found an association between obesity and other intrapartum complications, including pre-eclampsia and gestational diabetes. Despite an increased odds of EBL ≥ 1000 mL, we found a decreased odds of blood transfusion in patients with obesity after CD. Our findings contrast with prior work investigating the relationship between obesity and PPH. Butwick et al. conducted a large population-based cohort study that found an increased odds of hemorrhage after vaginal delivery and decreased odds of hemorrhage after CD for women with obesity; however, the effect was small and without strong evidence of a dose-response relationship between BMI and PPH 9 . In addition, the absolute event rate for PPH was 2.8%, which was lower than the frequency observed in our cohort (11.9%) 9 . Paglia et al. found a reduced risk of severe PPH among women with obesity after controlling for mode of delivery; of note, this was a case-control study of women with PPH who received blood components, indicating severe hemorrhage 11 . Our study focused on EBL ≥ 1000 mL, which may or may not have clinical effects.
Interestingly, despite an increased odds of PPH, we observed a decreased odds of blood transfusion among women with obesity. This finding implies that the increased blood loss we observed in patients with obesity may lack clinical relevance, and is consistent with prior work demonstrating greater blood volume and hemoglobin mass in individuals with obesity 25 . This finding also lends support to the notion that estimated blood loss is an unreliable indicator of clinically significant hemorrhage, and variation exists among hospitals in the use of quantitative over estimated blood loss. One formula devised for estimating allowable blood loss prior to transfusion includes allowable blood loss as directly proportional to the patient's estimated blood volume, implying www.nature.com/scientificreports/ that patients with obesity with a higher estimated blood volume are able to tolerate a higher estimated blood loss prior to requiring blood transfusion 26 . However, there is little consensus in the literature regarding the impact of obesity on maternal blood volume, with one study reporting similar circulating blood volume between gravidas with and without obesity 27 . We found that women with obesity in our cohort more frequently used oxytocin, dinoprostone, misoprostol, and magnesium sulfate, lending support to the theory that the mechanism by which obesity could increase PPH risk is related to uterine atony. Future work may examine variation in the etiology of PPH among women with and without obesity, as well as variations in PPH management based on BMI. The impact of obesity on maternal blood volume and the threshold for estimated blood loss at which clinically significant changes in hematologic parameters or transfusion risk occurs in women with obesity should also be investigated further due to our finding that despite a higher estimated blood loss, transfusion was not more frequent among women with obesity.
Limitations of our study include limitations common to studies using large databases, including potential misclassification errors and missing data. To address this limitation, we treated missing data as missing at random and employed multiple imputation by chained equations to account for statistical uncertainty while addressing missing data. Our dataset was unable to distinguish quantitative blood loss from estimated blood loss, which limits the accuracy of the blood loss reported and may contribute to variation in reporting blood loss among the hospitals in our dataset. Pre-pregnancy BMI and gestational weight gain were not reliably captured in our dataset, yet may affect pregnancy outcome more significantly than BMI at the time of delivery. Our study is also limited by the lack of ability to capture postpartum complications that occur after hospital discharge, such as delayed PPH, which is a potential explanation for the lack of association we observed between obesity and venous thromboembolism in our study despite the known increased risk of venous thromboembolism among gravidas with obesity 6,28 .
Despite these limitations, our study has a number of strengths. We included a large number of patients across multiple sites in different geographic regions across the United States, increasing the generalizability of our results. We excluded patients with missing BMI data since this variable was the focus of our study. We focused on including only CD to eliminate the potential confounding effect of mode of delivery in the relationship between BMI and blood loss. Our findings have significant clinical implications due to the rising prevalence of obesity in the United States as well as the high number of CD performed annually, currently accounting for more than 30 percent of all deliveries in the United States 29 . Our results demonstrate several potential risk factors for PPH after CD that can be used for population-based risk assessment, which can aid clinicians in identifying groups of patients that are at highest risk for PPH as part of an ongoing effort to lower hemorrhage-related morbidity and mortality.
In conclusion, we found an association between BMI and estimated blood loss after CD, with elevated BMI associated with an increased frequency of PPH. Despite a higher estimated blood loss, we did not find a higher frequency of blood transfusion among women with obesity after CD, implying that there may be no difference in clinically significant blood loss between women with and without obesity (Supplementary Information). Future work may investigate the amount of blood loss at delivery associated with maternal morbidity among patients with obesity, as well as variation in the etiology and management of PPH based on BMI.

Data availability
The dataset analyzed during the current study is not publicly available due to the sensitive nature of de-identified electronic health record data. To request more information about the data analyzed for this study, the principal investigator (Dr. Homa Ahmadzia) or the corresponding author (Dr. Julia Whitley) may be contacted.