Risk factors for severe postpartum hemorrhage: a case-control study

Severe postpartum hemorrhage (PPH) is the largest contributor to maternal morbidity worldwide, accounting for 50–75% of all such cases [1‐4]. Consequently, PPH has received increasing attention as a quality indicator for obstetric care. Furthermore, evidence exists that the incidence of PPH is increasing in high-income countries [5‐12]. An increase in the prevalence of known maternal and obstetric risk factors for PPH could play a role, but the supporting evidence from the published studies is limited. For example, in a Canadian study [7], induction of labor, augmentation of labor, and cesarean section partially explained the increasing rate of PPH. These findings may indicate that women undergoing these interventions need closer monitoring for severe PPH in the early postpartum period.

Several risk factors for PPH are known, such as multiple pregnancy, operative delivery and chorionamnionitis, however PPH may occur among patients with no known risk factors [13, 14]. Our ability to reduce the risk of PPH depends on ongoing investigations of previously unaccounted for causes and risk factors.

The primary aim of the study was to evaluate risk factors for severe PPH, taking into consideration pre-pregnancy, antenatal and intrapartum variables.

The source population was defined as pregnant women living in the metropolitan area of the Oslo and Buskerud municipality who were admitted to two university hospitals in Oslo (Ullevaal or Rikshospitalet) or Drammen Hospital for delivery between January 1, 2008 and December 31, 2011. From this source population, we performed a retrospective case-control study. We identified 1064 cases of severe PPH through birth suite records and hospital databases. Severe PPH was defined as blood loss ≥1500 mL or the need for blood transfusion for excessive bleeding at the time of delivery. Blood transfusion for excessive bleeding was defined as a blood transfusion given for a likely PPH ≥1500 mL due to clinical symptoms and signs of anemia or hemodynamic decompensation after delivery. We excluded women who received a blood transfusion because of postpartum anemia, without evidence of excessive hemorrhage. The attending physician or midwife estimated the blood loss visually in all three hospitals. Controls were a random sample of all deliveries without severe PPH from the same source population and period of time as the cases, comprising a total of 2059 deliveries. We selected random controls after removing the cases of severe PPH from the total number of deliveries at the three hospitals. Weighting was done according to the total number of deliveries in each hospital during the study period, resulting in control fractions from Rikshospitalet, Ullevaal, and Drammen of 21%, 62%, and 17%, respectively. Considering two controls per case, we estimated the number of controls needed from each hospital according to the number of delivering women in each hospital compared to the total number. This rendered sampling fractions at Rikshospitalet, Ullevaal, and Drammen of 4.8%, 5.2%, and 4.7%, respectively. The random sample was generated in STATA version 11.0 (Stata Corp LP, College Station, TX, USA).

Registration of patient data was based on information from 1) the hospitals’ medical records; 2) maternity databases (Obstetrix® from Siemens AG, Oslo, Norway and Partus® from Clinsoft, Oslo, Norway), and 3) birth suite records containing labor and delivery outcomes on all deliveries, including the volume of blood loss during delivery. If a woman had more than one delivery, the second and subsequent pregnancies were excluded to limit repeated correlated measurements.

In this study, we distinguished between causes of and risk factors for PPH; no direct causes were included in the risk factor analyses. Causes of severe PPH were classified as Tone (uterine atony, uterine inversion and abruption of the placenta), Tissue (retained placenta and retained parts of placenta, and abnormal placentation), Trauma (uterine rupture, birth canal trauma, and surgical trauma), and Thrombin (coagulation disorders). We registered up to two causes for each case if both were considered to be main causes, except in cases labeled as atony due to retained placenta which were reported as a retained placenta. A retained placenta necessitating a manual or operative delivery of the placenta was classified as a retained placenta. Cases with retained placental tissues diagnosed in the operating theatre or by ultrasound and needing surgical or manual removal, were classified as retained placental tissue. Abnormal placentation was defined as placenta accreta, increta or percreta. We identified cases caused by abnormal placentation post-delivery by reviewing medical records and pathology reports.

Based on literature review, we selected potential risk factors for consideration in our analyses. Pre-pregnancy factors included marital status, ethnicity, uterine anomalies (septated uterus, uni- or bicornuate uterus, uterus didelphys), previous uterine surgery (myomectomy and septal removals), previous cesarean section, previous severe PPH (≥1500 mL), and uterine fibromas. Current pregnancy conditions included maternal age, ethnicity (country of origin), pre-pregnancy body mass index (BMI), anemia in start of pregnancy (hemoglobin ≤9 g/dL), assisted reproductive technology (in vitro fertilization [IVF] or intra-cytoplasmic sperm injection [ICSI]), multiple pregnancy, gestational diabetes (insulin treated or diet regulated), use of anticoagulant medications such as low molecular weight heparin (LMWH) in pregnancy, polyhydramnios, severe preeclampsia or HELLP-syndrome, and premature rupture of membranes (PROM). Intrapartum factors included maternal fever (>38 °C) during delivery, mode of delivery, induction of labor, labor augmentation with oxytocin, and infant birth weight. Maternal age, BMI, and infant birth weight were considered as continuous variables for inclusion in the final model and categorical variables for descriptive purposes. Age was divided into 5-year groups, using 20–24 years as the reference group. BMI was categorized using the World Health Organization (WHO)’s classification, with a BMI of 18.5-24.9 kg/m² as the reference. Infant birth weight was dichotomized into ≥4500 g or <4500 g, according to the definition of fetal macrosomia [15]. Furthermore, we used WHO’s world regions to classify ethnic origin [16].

The data were entered into a database built in EpiData Version 3.1. (EpiData Association, Odense, Denmark). The medical records were re-examined if we observed data outliers or categorization errors for independent and dependent variables. All medical records were reviewed by two investigators (LTN and SP).

For our sample size estimation, we considered maternal age and induction of labor as potential risk factors for severe PPH. For both sample size estimates, we considered a type 1 error of 5%, a power of 80%, and a case:control ratio of 1:2. Regarding maternal age, in Oslo, approximately 18% of parturients are 35 years or older [17]. We hypothesized that women in this age group had a 1.4-fold increased risk of severe PPH compared to younger women [1, 18]. With a minimum detectable odds ratio (OR) = 1.4, we would need at least 656 cases and 1312 controls for a total of 1968 patients. Regarding induction of labor, in Norway, the frequency of labors induced with oxytocin is 3.4% according to figures from The Medical Birth Registry of Norway [17]. In a population-based study from Norway examining risk factors for severe PPH, the total rate of induction of labor was 10.8% [18]. We estimated that the frequency of induction of labor would be around the mean of these rates. Assuming a 1.6-fold increased risk of severe PPH in induced labor [1, 18], we would need at least 698 cases and 1396 controls, totally 2094 patients, when considering a minimum detectable OR = 1.6. Thus, for this case–control study, we decided to include at least 2400 deliveries (800 cases and 1600 controls).

We analyzed the data according to a pragmatic strategy [19], which means priority was not given to a specific etiological hypothesis. Univariable analysis was done to assess candidate variables as risk factors for severe PPH, and the associations between potential risk factors and severe PPH was quantified by the OR and 95% confidence interval (CI). The multivariable analysis was preceded by estimation of collinearity between risk factors. When collinearity existed between two variables, we omitted the one with least clinical relevance. Finally, multivariable logistic regression with manual backward elimination was used to identify independent risk factors for severe PPH [20]. Our criteria for sequential elimination of candidate risk factors were the variables’ strength and significance on the association with severe PPH, and optimal calibration and discrimination of the model. The continuous variables of maternal age, BMI, and birth weight were introduced into the model in its logged form as they were linearly associated with the outcome. However, birth weight is presented dichotomized in Table 3 as this was judged clinically more interesting.

Because we were applying a pragmatic strategy, the predictive accuracy of the model was evaluated by calibration and discrimination [21]. Calibration, which measures the ability of the model to assign the appropriate risk, was evaluated by the Hosmer and Lemeshow (H-L) goodness-of-fit test. A statistically non-significant H-L result (P > 0.05) suggests that the model predicts accurately on average. Discrimination, which measures the model’s ability to differentiate between individuals with and without severe PPH, was evaluated by analysis of the area under the receiver operating characteristic (ROC) curve. If the area under the curve is greater than 0.7, it can be concluded that the model has an acceptable discriminatory capability. All statistical analysis was performed using STATA version 11.0 (Stata Corp LP, College Station, TX, USA). We used Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines in reporting our case–control study [22].

From the source population of 43,105 deliveries, we identified 1064 women with a recorded PPH of ≥1500 mL or blood transfusion, giving a frequency of 2.5% (95% CI: 2.32–2.62). The identified causes of severe PPH are listed in Table 1. The most common cause was uterine atony (60.4%), while we identified retained placenta in 19.8% of the cases. Abnormal placentation was diagnosed post-delivery in 4.4% of the cases.

The study population comprised a total of 1064 cases of severe PPH and 2059 random controls without severe PPH. The distribution of potential risk factors is presented in Table 2. Europe, the United States, and Oceania were countries of origin for the majority of cases and controls (78.8% vs. 81.7%, respectively). The median (interquartile ranges) values for maternal age, BMI, and birthweight were similar among cases and controls; maternal age: 32 (29–36) years vs. 32 (29–35) years respectively; pre-pregnancy BMI: 23.1 (21.0–26.1) kg/m² vs. 22.8 (20.8–25.7) kg/m² respectively; and birthweight: 3546 (3075–3930) g vs. 3465 (3120–3834) g respectively. In the univariable analysis, severe PPH was more likely among women with the following medical and obstetric characteristics: primiparity, women who were married or cohabiting, previous severe PPH, previous uterine surgery, known uterine anomaly, multiple gestation, IVF/ICSI pregnancies, anemia, gestational diabetes mellitus, uterine fibroma, polyhydramnios, receipt of anticoagulant medication, and severe pre-eclampsia or HELLP syndrome. Severe PPH was more likely among women with these intrapartum and delivery characteristics: instrumental vaginal delivery, in-labor cesarean delivery, induced labor, labor augmentation with oxytocin, fever during labor, PROM, and infant birth weight ≥4500 g.

Risk factors independently associated with severe PPH are presented in Table 3. The strongest independent risk factors were a history of severe PPH (adjusted odds ratio (aOR) = 8.97; 95% CI: 5.25–15.33), anticoagulant drugs in pregnancy (aOR = 4.78; 95% CI: 2.72–8.41), anemia diagnosed in the start of pregnancy (aOR = 4.27; 95% CI: 2.79–6.54), severe pre-eclampsia or HELLP syndrome (aOR = 3.03; 95% CI: 1.74–5.27), uterine fibromas (OR = 2.71; 95% CI: 1.69–4.35), and multiple pregnancy (aOR = 2.11; 95% CI: 1.39–3.22). The H-L goodness-of-fit test was non-significant (chi-square = 12.99, P = 0.1122), indicating satisfactory model fit. The area under the ROC curve was 0.7173, indicating an acceptable discriminatory capability.

Finally, we performed an additional sensitivity analysis where the outcome was limited to an estimated postpartum blood loss ≥1500 mL. When considering the same risk factors, the results of the sub-analysis did not change our conclusions (Appendix: Table 4). Goodness-of-fit of the model was indicated by a non-significant H-L test (χ ² = 8.77, p = 0.3618), and a borderline discriminatory capability (area under the ROC curve = 0.6927).

In this case-control study, we evaluated risk factors for severe PPH. A history of severe PPH was the strongest independent risk factor in our study. Our findings suggest that women with increased risk of severe PPH can be identified when antepartum and intrapartum variables are considered. Furthermore, retained placental tissue was a much more frequent cause of severe PPH than previously reported.

The main strength of this case-control study was the quality of the data source. By reviewing medical records, we were able to evaluate potential risk factors for severe PPH using a broad selection of clinical variables which may not be easily retrieved from registries. In addition, reading the medical records enabled us to present accurate information on the causes of severe PPH and especially on the rate of retained placental tissue (including abnormal placentation) in the cases. We lacked data on women with suspected abnormal placentation prior to delivery as there was little awareness around pre-delivery diagnosis of abnormal placentation in Norway during the study period. Abnormal placentation could therefore not be included as a risk factor for severe PPH in our model.

Assessing risk factors in retrospect is a limitation in this study. In order to minimize selection bias, we selected all cases of severe PPH and a random sample of controls from the same source population. There is a possibility that some cases were misclassified. Blood loss was estimated visually in the three hospitals included in our study, and the blood loss may have been underestimated [23, 24]. However, we had further data in medical records to assess the severity of the hemorrhage by need for blood transfusion. There is a possibility of information bias due to misclassification of the exposures. However, neither the resident obstetrician nor midwife were aware of our research questions, therefore the potential for misclassification of risk factors would be non-differential creating a bias toward the null [20]. The sample size provided adequate power to highlight risk factors mildly associated with severe PPH, such as infant birth weight. However, we did not account for multiple testing in the power analysis and the smallest detectable odds ratios may have been underestimated.

We based our results on deliveries in three hospitals in or close to Oslo, Norway, and our results may not necessarily be generalizable to other delivery populations in well-resourced countries. The majority of the women in this study had European, USA, or Oceanian ethnicity, and was non-obese and married. These characteristics reflect a society with a low proportion of women with low socioeconomic status and fewer immigrants compared to other well-resourced countries. Furthermore, the hospitals mainly reflect an urban setting, but urban-rural differences in Norway are small and home births are an exception.

In this study, the strongest risk factors for severe PPH was a history of severe PPH, anticoagulant medication, anemia, severe preeclampsia or HELLP syndrome, uterine fibromas, and multiple pregnancy. Including these risk factors in clinical guidelines could help to identify women with high risk of severe PPH prior to delivery. By identifying these women, adequate resources and staff could be mobilized in preparation for severe bleeding at the time of delivery. Several of the identified risk factors in our study were related to medical and obstetric interventions such as anticoagulant medication, assisted reproductive technologies, labor induction, and labor augmentation with oxytocin. Further studies are needed to better understand the risk-benefit profile of each intervention on maternal outcomes. Risk factors identified in our study could be considered in future studies examining risk prediction models for severe PPH.