Secondary analysis of the WOMAN trial to explore the risk of sepsis after invasive treatments for postpartum hemorrhage

Abstract Objective To examine the association between the use of invasive treatments for postpartum hemorrhage and the risk of sepsis and severe sepsis. Methods Secondary data analysis of the WOMAN randomized controlled trial, including 20 060 women with postpartum hemorrhage in 21 countries. Logistic regression with random effects was used. Results The cumulative incidence was 1.8% for sepsis and 0.5% for severe sepsis. All‐cause mortality was 40.4% in women with severe sepsis versus 2.2% for women without. After adjusting for bleeding severity and other confounders, intrauterine tamponade, hysterectomy, and laparotomy increased the risk of sepsis (aOR 1.77 [95% CI 1.21–2.59], P=0.004; aOR 1.97 [95% CI 1.49–2.65], P<0.001; and aOR 6.63 [95% CI 4.29–10.24], P<0.001, respectively) and severe sepsis (aOR 2.60 [95% CI 1.47–4.59], P=0.002; aOR 1.97 [95% CI 0.83–2.46], P=0.033; and aOR 5.35 [95% CI 2.61–10.98], P<0.001, respectively). Conclusion In this secondary data analysis, certain invasive treatments for postpartum hemorrhage appear to increase the risk of sepsis. Further research is needed to confirm this finding and investigate the role of prophylactic antibiotics during these procedures. The harms and benefits of such interventions must be carefully weighed, both in treatment guidelines and during individual patient management. Trial Registration: ISRCTN76912190


| INTRODUCTION
Every day, approximately 830 women die worldwide as a result of pregnancy and childbirth. 1 Most maternal deaths (99%) occur in lowand middle-income countries (LMICs), 1  Risk factors for peripartum infections include pre-existing maternal conditions (obesity, diabetes, malnutrition, severe anemia) and factors related to childbirth (cesarean delivery, prolonged rupture of membranes, multiple vaginal examinations, placental retention). [5][6][7] Interest in postpartum hemorrhage (PPH) as a risk factor for sepsis has recently been sparked by an association found in population-based studies in high-income countries. [8][9][10][11] It is possible that invasive treatments for PPH, such as intrauterine tamponade and hysterectomy, The World Maternal Antifibrinolytic (WOMAN) trial recruited women with PPH in 21 countries. 15 Tranexamic acid was shown to reduce mortality from hemorrhage by 19%. Sepsis was noted to be an important complication. The aim of the present study was to examine this large dataset to determine the association between invasive treatment for PPH and sepsis.

| MATERIALS AND METHODS
We conducted a secondary analysis of all women recruited to the WOMAN trial to assess the association between invasive treatment for PPH and sepsis. Our primary outcome was sepsis, defined by the study authors at the onset of the trial. Our secondary outcome was sepsis with organ dysfunction. Main exposures were manual removal of placenta, hysterectomy, brace sutures, artery ligation (individually or a combination of the uterine artery, ovarian artery, internal iliac artery), intrauterine tamponade, and "laparotomy for other reasons." The WOMAN trial is a randomized, double-blind, placebocontrolled trial to evaluate the effect of tranexamic acid on mortality or hysterectomy in women with PPH. Details have been described elsewhere. 15 In brief, 20 060 women were recruited in 193 facilities in 21 countries. Women over 16 years were eligible if they had a clinical diagnosis of PPH (estimated blood loss >500 mL after vaginal delivery or >1000 mL after cesarean, or any blood loss sufficient to compromise hemodynamic stability). Baseline characteristics included maternal age, place and type of delivery, complete expulsion of the placenta, primary cause of bleeding, and use of uterotonics. Hemodynamic instability at entry was based on clinical signs (e.g. low blood pressure, tachycardia, falling urine output). The primary outcome was a composite of death from all causes or hysterectomy. Secondary outcomes included complications (renal failure, cardiac failure, respiratory failure, hepatic failure, sepsis, and seizures) and surgical interventions to treat hemorrhage (intrauterine tamponade, embolization, brace sutures, arterial ligation, hysterectomy, and laparotomies done for other reasons). Outcomes were recorded from medical records at time of death, at discharge, or 42 days postpartum, whichever occurred first.
Sepsis in the WOMAN trial was defined as infection plus systemic inflammatory response, in line with the previous adult definition of sepsis that was still in use when inclusion started in 2010. 16 In the present analysis, a new variable "severe sepsis" was created as sepsis plus organ dysfunction. Note that this "severe sepsis" variable equals the term sepsis as defined by the 2016 Sepsis-3 consort. 4 Organ dysfunction in the WOMAN trial was diagnosed as follows: renal failure required either a rise in serum creatinine of greater than or equal to 26 μmol/L (0.29 mg/dL) within 48 hours, rise in serum creatinine of 50% or greater known or presumed to have occurred within the past 7 days, urine output less than 0.5 mL/kg/h for more than six consecutive hours, or (in those with pre-existing renal disease) a serum creatinine rise of 200% or more from index serum creatinine or serum creatinine increased to 350 μmol/L (4 mg/dL). Cardiac failure required the presence of typical signs or symptoms (e.g. orthopnea, hepatojugular reflux) or a reduced left ventricular ejection fraction, relevant structural heart disease, or diastolic dysfunction. Respiratory failure required a partial pressure of oxygen less than 60 mm Hg (8.0 kPa) on room air, sea level. Hepatic failure required deterioration in liver function with changes in mental status and coagulopathy. No other types of organ dysfunction were recorded (Table S1 holds "guidance on diagnosing complications").
Percentages and medians were used to describe the data. Risk factors for sepsis were examined using logistic regression with random effects to account for clustering by facility. Main exposures were manual removal of placenta, hysterectomy, brace sutures, artery ligation (individually or a combination of the uterine artery, ovarian artery, internal iliac artery), intrauterine tamponade, and "laparotomy for other reasons." We built a comprehensive model including all possible confounders (age, type of delivery, hospital delivery, primary cause of hemorrhage, and markers of bleeding severity: estimated blood loss, systolic blood pressure, and hemodynamic instability). New organ dysfunctions are thought to result from sepsis rather than preceding it and were therefore excluded as risk factors. The same model was rerun with severe sepsis as the outcome.
Comprehensive logistic regression models with random effects were also used to assess the effect of (severe) sepsis on all-cause

| 233
Cornelissen eT Al. mortality. A priori confounders were age, type of delivery, signs of hemodynamic instability, place of delivery, prophylactic use of uterotonics, and primary cause of hemorrhage. Organ dysfunction lies on the causal pathway and was excluded.
All analyses were carried out using Stata version 14 (StataCorp, College Station, TX, USA). Odds ratios, 95% confidence intervals, and P values from likelihood ratio tests are presented. We considered a P value of <0.05 to be significant.

| RESULTS
A total of 20 060 women with PPH were included in the WOMAN trial, with a mean age of 28 years. Data on sepsis were available for 20 018 (42 missing, 0.2%). During the period of observation, 483 women died (all-cause mortality 2.4%). There were 365 cases of sepsis reported (1.8%), of which 104 met our criteria for severe sepsis (0.5% of the total population). Most women delivered in hospital (87.9%) and almost all received prophylactic uterotonics (96.2%). Almost one-third (29.1%) of women were delivered by cesarean. Data collection was nearly complete for most variables. Table 1 presents an overview of the population characteristics and missing values.
In univariate analysis, all surgical interventions to treat PPH showed evidence of an association with sepsis (

| DISCUSSION
In this analysis of 20 060 women with PPH, invasive treatments to manage PPH such as intrauterine tamponade, hysterectomy, and laparotomy appear to increase the risk of sepsis.
Comparison with published data on sepsis is difficult because of differences in methods and definitions. However, the incidence of 1.8% in our study seems high compared with previous figures. 6,17,18 This supports the hypothesis of an increased sepsis risk in women with PPH and is in line with previous studies. 8,11,19 Explanations for an increased risk of sepsis include an observed cesarean delivery rate of 29.1%, which is higher than one would expect based on published national rates for cesarean delivery. 20 Cesarean delivery is a known risk factor for both PPH and sepsis. 7,12 Moreover, hemodynamic instability is a result of PPH and causes hypoperfusion, which impairs the natural defense mechanisms of the body against infection. Nevertheless, after controlling for these and other possible confounders, the invasive management of PPH in itself appears to carry an infectious risk.
As consequences of sepsis are grave (15% mortality for any type of sepsis, 40% in those with severe disease) and its global impact big (11% of all maternal deaths 2 possibly contributing to more 18 ), prevention is of the utmost importance. We are unable to evaluate the effect of prophylactic antibiotics as their use was not recorded. Guidelines on PPH are heterogeneous in their recommendations on antibiotics. [12][13][14] However, drawing on evidence for similar interventions such as surgery for spontaneous abortion, cesarean delivery, and hysterectomy, 7,21,22 antibiotic prophylaxis seems likely to be beneficial.
Hysterectomy is known to increase the risk of sepsis, possibly due to opening the contaminated vaginal vault. 22 The high risk associated with "laparotomy for other reasons" might be, at least partly, explained by reversed causation where second-look laparotomies and washouts are performed for women with an intra-abdominal infection. However, the protocol did aim to specifically record surgical interventions to treat hemorrhage. Women with PPH who undergo surgery might be more vulnerable to infection than women who have elective surgery, since they present as an emergency with possible additional risk factors (e.g. hemodynamic instability, nonhospital delivery).
Concerns about infections linked to intrauterine tamponade have been raised since the 1950s, but the currently used balloon tamponades are deemed safer in general than gauze packing. All major 234 | Cornelissen eT Al.
T A B L E 1 Characteristics of the study population and univariate analysis of risk factors for sepsis and severe sepsis (n=20 018). While there were few reported adverse events, the review notes that harms were not well characterized. However, our results raise the possibility that tamponade could increase the risk of life-threatening sepsis. This calls for good-quality primary research into the benefits and risks of tamponade and investigation into antibiotic prophylaxis, which is not currently part of the guidelines.

Type of delivery
Of course, a lifesaving intervention should never be withheld to avoid the possible complication of sepsis, but clinicians should be aware of the risk and the importance of prevention, early recognition, and correct treatment.
To our knowledge, the present study is the first to look at possible explanations for an increased infectious risk in women with PPH.
All recent publications that showed an association between sepsis and PPH were from studies in high-income countries with low levels of maternal mortality [8][9][10][11] or only included women who had undergone cesarean delivery. 19 The WOMAN trial is a large multicountry study predominantly in LMICs. Recorded variables were well-chosen for low-resource settings and data collection was nearly complete for most of them. Severity of bleeding is likely to be an important confounder and was corrected for in our model using three different variables: estimated blood loss, systolic blood pressure, and hemodynamic instability. Plausible physiologic mechanisms, large numbers of events, and small P values make it highly unlikely that results are purely due to random error. Random effects were included in the model to account for clustering by facility.
There are also important limitations to consider. This was a secondary data analysis and the original study was not specifically designed to investigate the incidence or risk factors for sepsis. Clear diagnostic criteria were provided but sepsis was a secondary outcome collected from patient records. It was only measured during hospital stay. Longer admissions following surgical interventions for PPH provide more opportunity for sepsis to be diagnosed, and the effect is thus potentially
than causing it. Indeed, for 11% of hysterectomies the stated reason was "to remove a severely damaged, ruptured, or infected uterus." Unfortunately, we were unable to separate out the hysterectomies performed as treatment for infection. Furthermore, "laparotomy for other reasons" might be second-look laparotomies for abdominal infections.
The original study did not collect information on comorbidities such as (gestational) diabetes or antibiotic usage, which are likely to be effect modifiers, although unable to explain the associations we saw.
In conclusion, PPH and postpartum sepsis remain important causes of maternal mortality and morbidity worldwide. In this large multicountry study, women who received certain invasive treatments to manage PPH appeared to face an increased risk of sepsis, which carried a high case-fatality rate. Primary research is urgently required to investigate this finding further and examine ways to reduce the risk, including clearer guidelines on the use of prophylactic antibiotics.
In the interim, potential harms and benefits of these interventions, particularly intrauterine tamponade, should be carefully weighed when developing policy and in clinical management of women.

AUTHOR CONTRIBUTIONS
IR proposed the study subject. IR, HS, BF, SN, and SA were key contributors to data acquisition. IR, LC, and SW developed the study design and LC and SW analyzed the data. LC and SW drafted a first version of the manuscript, which was then critically revised by IR, HS, BF, SN, and SA. All authors provided final approval for the submitted manuscript.

ACKNOWLEDGMENTS
The initial WOMAN trial was funded by the London School of Hygiene