Discussion: “Maternal death in the 21st century: causes, prevention, and relationship to cesarean delivery” by Clark et al

      In the roundtable that follows, clinicians discuss a study that is published in this issue of the Journal in light of its methods, relevance to practice, and implications for future research. Article discussed:
      Clark SL, Belfort MA, Dildy GA, Herbst MA, Meyers JA, Hankins GDV. Maternal death in the 21st century: causes, prevention, and relationship to cesarean delivery. Am J Obstet Gynecol 2008;199:36.e1-36.e5.

      Discussion Questions

      • What are the research objectives?
      • What is the study design?
      • How would you describe the study population?
      • What are the limitations of retrospective studies?
      • What is confounding by indication?
      • What are the criteria for the assessment of causality?


      Patients might be shocked to learn that women in the United States still die in childbirth. Fortunately, this is a rare event. Yet, the subject is timely; this month, the Journal Club discussed a study that focused on the maternal mortality rate and how it might be further reduced. An important aspect was the potential contribution of cesarean delivery to the maternal mortality rate, especially when we consider the trend toward elective cesarean delivery. Clark et al tapped a large database from the Hospital Corporation of America to examine maternal deaths that occurred among close to 1.5 million deliveries. Their findings sparked a lively conversation.


      Gross: In the United States, the cesarean delivery rate continues to rise, with most recent rates eclipsing 30%. As a result, any complication that is related to the surgery is an important area of study. Not only is the delivery rate increasing to its highest levels, but also the rate of increase is at an all-time high, with rates increasing for patients of all ages, races, and ethnic groups. Cesarean delivery remains the most commonly performed major surgery in the United States today.
      When we counsel patients about cesarean delivery, we should always describe risks and benefits. Risks typically should include surgical complications such as bleeding, infection, and damage to internal structures. It has been my experience that the catch phrase “increased risk of death” is included uniformly, but I personally find it somewhat difficult to pin down this risk in exact terms and numbers. It can also be somewhat difficult to describe to patients exactly why they might be at risk for dying. For that reason, the new study by Clark et al was a very interesting read.
      Gross: What are the authors' research objectives? Why is this area of study important?
      Lewkowski: The authors used a large database from a hospital network to examine the causes of maternal death in the 21st century. In addition, they were hoping to determine how many maternal deaths were preventable and to evaluate a possible causal relationship between cesarean delivery and maternal death.
      I do believe that this is an important area of study. Maternal death, although a rare event, still occurs in pregnancy. However, the causes change as medical practice evolves and new technologies and treatments emerge. It is important to identify these changes to develop suitable interventions.

      Study Design

      Gross: What was the authors' study design? Was it appropriate?
      Scifres: This study design was effectively a case series of all maternal deaths that were recorded in a large hospital-based registry between January 1, 2000, and December 31, 2006. The authors set out to answer questions regarding the causes of maternal death since 2000, the preventability of these deaths, the number of deaths that were attributable to cesarean delivery, and whether methodic changes in the healthcare system could reduce the maternal death rate in the United States. This study design can answer the basic question about causes of maternal death and, to the extent that preventability can be determined based on a review of maternal medical records, whether these deaths could have been avoided.
      The issue of whether a maternal death was attributable to cesarean delivery brings up an important limitation of a case-series design. One thing the authors could not do is thoroughly explore risk factors for maternal death. This would have to be accomplished with either a retrospective cohort or case-control design. In a retrospective cohort study, the records of women who underwent cesarean delivery, which was the exposure of interest in this example, would be compared with those of women who delivered vaginally to determine whether maternal death, which was the outcome of interest, was more prevalent in 1 group. In a case-control study, researchers would begin with the outcome of interest by combing through the records of women who died at delivery and women who survived to determine which exposures, including cesarean delivery, were significantly associated with death. With either design, multivariable logistic regression would be used to assess each variable's potential for confounding. This would determine whether cesarean delivery or other components were associated independently with maternal death.
      Gross: What types of problems might be associated with retrospective data extraction? What actions were taken to reduce the possible shortcomings, and were these successful?
      Chrusciel: Studies that use retrospective data extraction are limited by the content and accuracy of the medical record. Accuracy of recorded information can be checked only by interviewing all persons who were involved in that particular patient's care. The authors validated the diagnosis-related group (DRG) for each case by reviewing the medical record of each woman who died. In 16 cases in which the medical record left a question regarding preventability or cause of death, 2 authors reviewed the documents; in all cases, they were in agreement as to what had occurred. The authors were also able to assess accuracy of the medical record by interviewing providers to clarify the circumstances surrounding each of 3 maternal deaths. However, this does not address the issue of maternal deaths that were coded incorrectly and, as a result, were assigned to DRGs other than those examined in the study (codes 370-375, 378, 380, 381, 383, 384). The researchers' assumption that the DRG codes were accurate would have been strengthened by providing information on the DRG accuracy of records that they reviewed.
      Gross: How might the authors have assessed the accuracy of the DRG codes?
      Chrusciel: The investigators could have determined what proportion, if any, of the records that they had reviewed with the relevant codes (370-375, 378, 380, 381, 383, 384) appeared to be coded inaccurately. Did they see any records that should have carried a different code? The investigators could have taken a random sample of the records that they used and assessed whether they were assigned accurately.
      Gross: Describe the population that were studied in this article. Was this an appropriate population to examine?
      Tse: Study subjects came from 1 of the largest healthcare delivery systems, the Hospital Corporation of America. It is in 20 states across the country and includes primary, secondary, and tertiary care facilities. However, only 2 academic centers were involved. The authors claim that previous analyses of the Hospital Corporation of America's cesarean and operative vaginal delivery rates suggested the patient population is roughly representative of the United States as a whole. Nonetheless, academic centers are vastly underrepresented. This presents a problem, because a large proportion of operative vaginal deliveries are taking place in our academic centers, which is more so than in private centers, where cesarean delivery is more predominant. Although the choice of this population was convenient and appropriate for this study, the data might not translate to academic centers. Thus, the generalizability of the authors' conclusions could be weaker for some regions of the country.
      Gross: What are the 3 classifications of maternal death that were studied in this article? Can you provide examples from each category?
      Lewkowski: Maternal death was classified as either a direct obstetric death, an indirect obstetric death, or a nonobstetric death. Direct obstetric deaths are caused by a complication that develops directly as a result of pregnancy, such as pulmonary embolism, eclampsia, or peripartum hemorrhage. Included in this category are complications that are due to interventions that are aimed at treating these medical conditions. An indirect obstetric death is due to a medical condition that is present before pregnancy but is exacerbated by the pregnancy. An example is a preexisting heart condition that is worsened by pregnancy-related fluid shifts and increased circulating volume. A nonobstetric death is a death during pregnancy or the peripartum period that is caused by an accidental or incidental cause that is not related to pregnancy or its management (fatal trauma, for example).
      Gross: What is confounding by indication? What steps would reduce this phenomenon?
      Chrusciel: Confounding by indication occurs when patients with the same indication receive different treatment. Most commonly, this happens when management of severe cases of a disease differs from treatment of less severe cases. It is expected that cases that receive a specific treatment will be systematically distinct from cases that did not receive that precise treatment. For example, patients who receive the most medical care may be the mostly likely to have adverse outcomes, because the patients who receive the most care are typically the sickest.
      With respect to cesarean delivery, the authors addressed confounding by indication by determining whether maternal death was related to cesarean delivery or to the indication for the procedure. To further reduce this phenomenon, the authors could have used a case-control design to compare cases that resulted in maternal death with similar cases in which the woman survived.


      Gross: What were the authors' findings with regards to maternal mortality rates and causes? Did they represent the population of the entire United States?
      Tse: The authors' observed maternal mortality rate was 6.3/100,000 over a 6-year time period, which is approximately 50% lower than the Centers for Disease Control and Prevention's reported national rate of 13/100,000 for 2006. Their lower rates may have reflected the disproportionate number of private hospitals that were included in the study sample. Cesarean delivery rates in the study were comparable with national rates, nearing 31%. The frequencies of the major causes of death among study subjects (preeclampsia, embolic events, and postpartum hemorrhage) were comparable with national statistics. However, the ethnic composition of the study population was different from that of the entire United States; the ethnic distribution of maternal deaths recently reported by the Morbidity and Mortality Weekly Report has been more skewed consistently towards African American women. This was not true among study subjects.
      Gross: An important fact is that the maternal mortality rate in this country is actually climbing, which is a more recent trend. Somewhat surprisingly, this study found a mortality rate that was roughly 50% of the reported national rate.
      Gross: How confident are you, as a reader, that cases of maternal death could be judged “preventable” or “nonpreventable” from the data extraction process?
      Scifres: The issue of a death being “preventable” or “nonpreventable” is certainly a complicated one. Maternal deaths might be preventable on several different levels. As the authors of this study pointed out, some cases of maternal death judged to be “potentially preventable” resulted from the actions—or inaction—of nonmedical personnel. If you examine deaths such as motor vehicle accidents, alcohol abuse, or suicide, these types of maternal deaths potentially could be prevented by public health measures (for example, efforts that improved compliance with appropriate safety restraint devices or improved access to substance abuse or mental health services).
      Specifically, the authors were interested in identifying deaths that resulted from inappropriate or inadequate actions by healthcare workers or nonmedical personnel around the time of the death. This was performed by retrospective review of the patient's chart and interviews with healthcare staff members when chart review did not provide the information that was needed to determine preventability. It is always very difficult to ascertain retrospectively whether an event was truly preventable, because we cannot know whether a different course of events would have occurred had a different set of choices been made by the involved providers. The authors acknowledged the subjective nature of this judgment, and it is reasonable to assume that an experienced physician could identify cases in which the standard of care was not followed.
      Interviewing healthcare providers who were involved in a maternal death brings up the important topic of recall bias. Maternal death is an uncommon and stressful event. Possibly, a provider's ability to recall details of a case that ended with a death might be different from that of a provider engaged in a similar case with a good outcome.
      Gross: What are the most common causes of maternal death? What assumptions are made regarding the ability or inability to reduce maternal mortality rate based on these causes?
      Tse: As noted, in this study, the 3 most common causes of maternal death were preeclampsia, amniotic fluid embolism, and obstetric hemorrhage. According to the investigators, the most common preventable errors that contributed to the preeclampsia mortality rate were inattention to blood pressure control and to signs and symptoms of pulmonary edema. Belated attention to clinical signs of hemorrhage and subsequent hypovolemia were said to be the most common errors in the care of women who died from postpartum hemorrhage.
      Cardiac disease was the fourth most common cause of maternal death; pulmonary thromboembolism is the fifth. The authors suggested that pulmonary embolic events are an optimal target for large-scale intervention. They refer to population studies of adult surgical patients, in which venous thromboembolism (VTE) was reduced by 70% with routine mechanical or medical thromboprophylaxis. However, these patients are not necessarily analogous to obstetric patients, who have multiple other factors that contribute to their risk for VTE. Still, the authors assumed that the risk reduction would be equal if thromboprophylaxis were used universally in women who undergo cesarean delivery.
      The investigators also assumed that the cesarean delivery mode was related causally to the development of VTE, which may or may not be the case, given the multiple factors that could have led to the decision to proceed with the cesarean delivery in the first place. These other elements might also have contributed to the development of VTE in these patients. Without taking these aspects into account, it is difficult to state that the maternal mortality rate can be reduced by changing only 1 intervention.
      Gross: I think the authors made the assumption that certain rates of complications, such as preeclampsia and amniotic fluid embolism, have a baseline occurrence rate that is likely unchangeable. The risk and prevalence of VTE can both be reduced, and I think that is the reason that they are targeting that complication as an area of potential improvement.
      Gross: One of the authors' major goals was to determine the causal relationship of cesarean delivery and maternal death. What is the difference between proving causation vs association, and does this study definitively demonstrate a causal relationship between cesarean delivery and maternal death?
      Rampersad: Causation is very complex and difficult to prove vs association, which is easier to prove. Association states that there is a statistical relationship between 2 variables but does not specify the nature of the relationship. To prove a causal relationship, 3 things need to be proved: an association exists, variable x precedes y (a situation known as temporality), and correlation between the 2 variables is not a result of outside influence or confounding variable. Other methods or criteria have been used to demonstrate a casual relationship. The most often used technique applies criteria set forth by Sir Austin Brandon Hill in a 1965 article called The Environment and Disease: Association or Causation.
      • Hill A.B.
      The environment and disease: association or causation?.
      It describes 9 factors:
      • Strength: Is the risk large?
      • Consistency: Have the results been replicated by different researchers?
      • Specificity: Is the exposure associated with a very specific disease as opposed to a wide range of diseases?
      • Temporality: Did the exposure precede the disease?
      • Biological gradient: Are increasing exposures associated with increasing risks of disease?
      • Plausibility: Can a credible scientific mechanism explain the association?
      • Coherence: Is the association consistent with the natural history of the disease?
      • Experimental evidence: Does a physical intervention show results that are consistent with the association?
      • Analogy: Can we draw a relationship to a similar result?
      One aim of this study was to determine the extent to which cesarean delivery is causally, as opposed to associatively, related to maternal death. The investigators identified 4 deaths that were attributed to cesarean delivery; 3 deaths were from injured blood vessels and subsequent hemorrhage, and 1 resulted from sepsis that was due to bowel injury. Although they did not provide much information regarding the surgery, we can say from previous studies and our own experience as surgeons, that this seems plausible. It is associated temporally in that the exposure preceded the outcome, and this type of injury previously has been demonstrated to have similar complications.
      Let us look at the 9 factors that were identified by Sir Brandon Hill with respect to cesarean delivery and death in this study:
      • I think it is fair to say that there is a strong association between cesarean delivery and injury to vessels and to the bowel and to subsequent hemorrhage or infection, respectively.
      • The results are consistent across a wide range of surgeons' experiences and studies.
      • Cesarean delivery is not associated specifically with maternal death.
      • Cesarean delivery precedes the diagnosis of vessel and bowel injury, and both are seen rarely without surgical intervention.
      • A dose-response relationship exists.
      • It is plausible that injury to vessels during surgery will lead to hemorrhage and death, if the hemorrhage is severe.
      • There is coherence. The association is consistent with the natural history of the disease.
      • Interventions do work. If a vaginal delivery is performed, then intraabdominal blood vessels and the bowel will not be injured. Or, if injuries are repaired in a timely fashion, if blood products are given when needed, and if appropriate antibiotics are administered in the case of sepsis, death might be avoided.
      • An analogy: Other surgical procedures can lead to injury of vessel or the bowel leading to death.
      As to whether deaths resulted from cesarean-induced vessel and/or bowel injury, there is congruence for 8 of 9 factors (specificity is not present), which makes a convincing case for causation.
      In addition, 7 deaths from thromboembolism were stated to be related causally to cesarean delivery. I think their evidence proves an association with cesarean delivery and VTE, but not causation. VTE can occur at any time during pregnancy, and the diagnosis after cesarean delivery does not say definitively when the clot occurred. Many other confounders could also be associated with the cause of VTE after cesarean delivery; besides, 2 cases occurred during vaginal deliveries. Therefore, cesarean delivery might increase the probability of a VTE, but that does not prove causation.


      Gross: What did the authors conclude regarding the practice of thromboprophylaxis? Will these conclusions change or affirm your current practice in any manner?
      Scifres: Several conclusions regarding thromboprophylaxis are made in this study. First, the authors concluded that, if either mechanical or surgical prophylaxis were assumed to reduce the incidence of venous thrombosis by 70%, as suggested by other studies, then 5 of the 7 deaths that were due to pulmonary embolism in women who underwent cesarean delivery would have been prevented by a policy of universal thromboprophylaxis. They also suggested that, with appropriate universal thromboembolism prophylaxis, the excess risks of death that were due to cesarean delivery might be eliminated virtually. The authors reported that they are implementing a policy of universal thromboprophylaxis with pneumatic compression devices for women who undergo cesarean delivery.
      This study affirms my current practice, which is very similar to that outlined in the article. At our institution, we use pneumatic compression devices as prophylaxis during and after cesarean delivery for most patients. We reserve the use of anticoagulant prophylaxis for patients with multiple risk factors because of a concern for bleeding, as discussed in the article. Although this study does affirm our current practice, I think it is important to discuss the limitations of the data on which we base the decision about whether to offer prophylaxis. If you examine this data set, fatal pulmonary embolism was more common among patients who underwent cesarean delivery than among those patients who delivered vaginally. However, this study is unable to confirm that venous thrombosis is associated independently with cesarean delivery. Multiple confounding factors may be present and include prolonged bed rest and maternal age. In addition, with a retrospective review like this one, we are uncertain about which patients may or may not have received thromboprophylaxis before pulmonary embolism. I agree with the authors' conclusion that some deaths from venous thrombosis are likely preventable. However, important questions about which patients require prophylaxis and the optimal prophylaxis regimen for different groups of patients still remain.
      Gross: Given the suggested reduction in maternal death that is achieved by thromboprophylaxis, should we be taking this information into account when counseling patients about the risk/benefit ratio of repeat cesarean delivery vs vaginal birth after cesarean delivery?
      Rampersad: The benefit of thromboprophylaxis in patients who undergo cesarean delivery has not been proved, and the most recent Cochrane Review on this issue concluded there was insufficient evidence to base a recommendation.
      • Gates S.
      • Brocklehurst P.
      • Davis L.J.
      Prophylaxis for venous thromboembolic disease in pregnancy and the early postnatal period.
      So, I do not think we should use this information currently in counseling patients with a previous cesarean delivery. Patients should be told that the mortality rate from cesarean delivery might be increased when compared with a vaginal delivery.
      Gross: One reason we encounter more cesarean deliveries is this phenomenon of “cesarean on demand” or the primary elective cesarean delivery. These are performed on the basis of maternal choice rather than on any specific obstetric or medical indications.
      Gross: Would the information presented in this study alter your patient counseling with regard to the decision to forgo labor and have a primary elective cesarean delivery?
      Scifres: I think this is important information for patients who request elective primary cesarean delivery. Although the increased rate of maternal death is low, its existence has been found in several studies. So, I think it is worth discussing.
      Gross: Is this study saying that initiating thromboprophylaxis in every patient who has a cesarean delivery can effectively even the playing field with regard to maternal mortality?
      Rampersad: No, I think that the investigators believe this is 1 area in which an improvement can be made. I do not think that it will equal the playing field of cesarean delivery vs vaginal delivery. We are not sure whether we will make an impact by doing thromboprophylaxis, because there are only 2 small randomized control trials.
      Gross: So in essence, I think this study helps to introduce a very important issue that must be studied: How the universal use of thromboprophylaxis in patients who undergo cesarean delivery influences the maternal mortality rate, when compared with vaginal delivery. Like any large case series, this model hopefully will pave the way towards trials that can answer the question at hand more definitively, namely prospectively performed randomized trials. Whether we agree or disagree with the rising rates of cesarean delivery and the concept of cesarean delivery on demand, we must be well-versed in the potential complications of the procedure, particularly because many of our patients are seeking it.


        • Hill A.B.
        The environment and disease: association or causation?.
        Proc R Soc Med. 1965; 58: 295-300
        • Gates S.
        • Brocklehurst P.
        • Davis L.J.
        Prophylaxis for venous thromboembolic disease in pregnancy and the early postnatal period.
        Cochrane Database Syst Rev. 2002; 2 (CD001689)

      Linked Article

      • Maternal death in the 21st century: Clark et al
        American Journal of Obstetrics & GynecologyVol. 199Issue 1
        • Preview
          The article below summarizes a roundtable discussion of a study that was published in this issue of the Journal in light of its methods, relevance to practice, and implications for future research. Article discussed: Clark SL, Belfort MA, Dildy GA, Herbst MA, Meyers JA, Hankins GDV. Maternal death in the 21st century: causes, prevention, and relationship to cesarean delivery. Am J Obstet Gynecol 2008;199:36.e1-36.e5. The full discussion appears at , pages e7-e11.
        • Full-Text
        • PDF