Predictors of maternal mortality and near-miss maternal morbidity

Abstract

Objective:

To identify risk factors for life-threatening maternal outcomes.

Study design:

Hospital charts were reviewed for cases of maternal mortality or near-miss and for controls overmatched 1:3. Significant risk factors were identified through simple and best subsets multiple logistic regression.

Result:

Eight cases of mortality and 69 near-miss cases were found. Significant risk factors with their odds ratios and 95% confidence intervals are: age 35 to 39 years (2.3, 1.2 to 4.4) and >39 years (5.1, 1.8 to 14.4); African-American race (7.4, 2.5 to 22.0) and Hispanic ethnicity (4.2, 1.3 to 13.2); chronic medical condition (2.7, 1.5 to 4.8); obesity (3.0, 1.7 to 5.3); prior cesarean (5.2, 2.8 to 9.8) and gravidity (1.2, 1.1 to 1.5 per pregnancy). In multivariable logistic regression, race remained significant while controlling for other significant factors and markers of socioeconomic status.

Conclusion:

Some risk factors can be modified through medical care, education or social support systems. Racial disparity in outcome is confirmed and is unexplained by traditional risk factors.

Introduction

Although the maternal mortality ratio has significantly decreased from 582 maternal deaths per 100 000 live births in 1935 to a current level of approximately 8.9,^{1, 2} it is far above the United States Healthy People 2010 goal of 3.3.³ To achieve this goal it is important to understand the factors that contribute to the risk of maternal death. The low rate of maternal mortality leads to a lack of statistical power and, in turn, difficulties in identifying and assessing potential risk factors. Studies have investigated severe acute maternal morbidity or ‘near-miss’ as a surrogate for evaluating obstetric care.^{4, 5, 6, 7} There is great variation in published studies in defining near-miss.^{8, 9} A 1-year prospective multi-center study defined ‘near-miss’ as an acute organ system dysfunction, which if not treated appropriately, could result in death.^{10, 11} Several retrospective studies have defined ‘near-miss’ as obstetric intensive care admissions¹² or prolonged hospital stays and re-admissions.¹³ Another large retrospective study estimated the incidence and predictors of severe obstetric morbidity, which was defined as severe pre-eclampsia, eclampsia, hemolysis, elevated liver enzymes, low platelets (HELLP) syndrome, severe hemorrhage, severe sepsis and uterine rupture.⁷ Geller et al.^{14, 15} have proposed a scoring system to classify women with severe morbidity and near-miss morbidity. These studies have suggested that the identification of risk factors for severe morbidity may improve maternal care.

Traditional reports of risk factors for maternal mortality have relied on vital statistics data, which are known to have limitations with respect to misclassification bias, under-reporting bias and a lack of information on important confounding variables. Our objective was to identify risk factors for maternal mortality or ‘near-miss’ events that may be unavailable from vital statistics data.

Methods

Cases of maternal mortality and near-miss morbidity from the Weiler Hospital of the Albert Einstein College of Medicine/Montefiore Medical Center, an inner city Regional Perinatal Center, from January 1995 through June 2001 were identified through quality improvement records, intensive care unit (ICU) admission records and a computerized search of the medical record database using ICD-9 discharge diagnoses. We defined near-miss morbidity as ICU admission, emergency unplanned return to the operating room or delivery room for hemorrhage, eclampsia, emergent hysterectomy, cardiac arrest, cerebral anoxia, shock and embolism. Maternal deaths were defined as those occurring in pregnancy, or within 1 year of pregnancy due to a pregnancy-related cause. We selected three unmatched controls from the same day that each case delivered.

For each case and control, we collected demographic, medical and obstetric data. Demographic data included age, race, level of education, marital status, Medicaid insurance and employment status. Medical data comprised of height and weight used to calculate body mass index, hematocrit at intake and a history of a chronic medical condition such as hypertension, diabetes, asthma, cardiac disease, sickle cell disease and cerebrovascular disease. Finally, obstetric data consisted of gravidity, parity, estimated gestational age, number of prior cesarean deliveries and level of prenatal care. Prenatal care was determined to be adequate, intermediate or inadequate using the Kessner index,¹⁶ calculated from the number of prenatal visits and the gestational age at the first and last visit. The study was approved by the Institutional Review Board of the Montefiore Medical Center.

A series of univariate logistic regressions were performed with Systat v10 to identify significant (P<0.05) single risk factors for maternal mortality or near-miss. Best subsets regression was used to identify several unbiased multiple logistic regression models using Mallow's C_p criterion. For this, we used the REG procedure of SAS v8, adapted for logistic regression as discussed in Hosemer and Lemeshow.¹⁷ Satisfactory models identified by this procedure were examined with a view to understand race as a risk factor when adjusted for other factors. All models passed the goodness-of-fit criteria suggested by Hosemer and Lemeshow,¹⁷ including Pearson's χ², deviance and the Hosemer–Lemeshow statistic (10 categories).

Results

Within the study period, there were 25 837 deliveries at Weiler hospital. Ninety-two cases were identified, eight maternal mortalities and 84 near-misses. The records for 77 cases, 8 deaths and 69 near-misses, were available for review. The majority (52%) of maternal deaths and near-misses resulted from hemorrhage. Pre-eclampsia and eclampsia were the second leading cause at 14.3%. Other causes of maternal mortality and near-miss included anesthesia accidents and infection. Surgical complications resulted in three near-misses but no maternal deaths (Table 1).

Table 1 Causes of maternal mortality and near-miss mortality

We first examined the effects of possible risk factors individually. Among maternal demographic factors, race/ethnicity was the strongest predictor for maternal death or near-miss (Black race odds ratio (OR) 7.4, 95% confidence interval (CI) 2.5 to 22.0 and Hispanic ethnicity OR 4.2, CI 1.3 to 13.2 as compared to non-Hispanic White women). Women who died or had near-misses were also significantly more likely to be older than control women (OR 2.3, CI 1.2 to 4.4 for 35 to 39 years and OR 5.0, CI 1.8 to 14.4 for ⩾40 years, compared to <35 years). There were no statistically significant differences associated with level of education, marital status or public medical insurance status between cases and controls (Table 2).

Table 2 Maternal demographic factors

As detailed on Table 3, medical factors associated significantly with maternal death or near-miss included obesity (body mass index>29) (OR 3.0, CI 1.7 to 5.3), a past history of a significant medical condition (OR 2.7, CI 1.5 to 4.8) and a prior cesarean (OR 5.3, CI 2.8 to 9.8). An increasing number of prior pregnancies was also associated with an increased risk. For each additional pregnancy, the OR for adverse outcome was 1.3 (CI 1.1 to 1.5). The level of prenatal care was not found to be associated with risk of maternal death or near-miss.

Table 3 Medical and obstetrical factors

The smallest multiple regression model identified by best subsets regression included race, maternal age, past medical history and previous cesarean as explanatory variables. All variables achieve statistical significance in this model. In this model, the OR for Black race is 7.1 (CI 2.2 to 22.5), and the OR for Hispanic ethnicity is 4.2 (CI 1.2 to 14.5) with White race as the reference group in both cases (Table 4).

Table 4 Multivariable models

To investigate the relationship between race and other traditional risk factors, we tried adding obesity and gravidity to the minimal model. In this model, the OR for Black race is 5.2 (CI 1.6 to 17.3), while that for Hispanic ethnicity is 3.4 (CI 1.0 to 11.9).

Finally, we added the following variables: Medicaid status, education level and marital status to the minimal model. We were interested in these variables as possible markers of socioeconomic status. None of them reached statistical significance. Adding these variables widens the confidence limits for race, but leaves the lower limits almost unchanged.

Discussion

During the study period at our institution, we found a relatively large number of maternal deaths. The high incidence we report may be explained by several factors. The intensive case finding used in this study, using multiple local sources, that is ICD-9 codes from computerized databases, quality improvement records and ICU admissions is likely an important contributor. Several reports have shown that vital statistics data alone under-report maternal deaths as much as 30 to 60% compared to enhanced surveillance.^{18, 19, 20, 21} Furthermore, our institution is a regional perinatal center, which accepts referral of women at high risk of morbidity and death.

We found that maternal mortality resulted from hemorrhage, pre-eclampsia/eclampsia, medical conditions, thromboembolic complications, accidents related to anesthesia and infection. As our near-miss cases came from these categories as well, it might be that these women may serve as a useful proxy in the study of risk factors for prevention of maternal mortality.

Obese women were three times more likely to die or suffer severe morbidity in our study. Although in our multivariable regression models, the effect of obesity was accounted for by other factors, obesity is a known risk factor for adverse health outcomes and could be causally related. In the obstetrical population, obesity is associated with factors important in morbidity and mortality. These include thromboembolic disease, hypertension, cesarean delivery, obstetrical hemorrhage and anesthesia complications. With the rising rates of obesity in the US population, especially in children, this risk factor and its relation to maternal outcome deserves more intensive study. The relationship of maternal undernutrition to maternal mortality, especially in developing countries, has also been highlighted as an area in need of further investigation.²² Women with a prior cesarean delivery are also at increased risk of death or near-miss morbidity. This finding coupled with the rising cesarean delivery rate in the United States raises concerns about the potential for a rise in the MMR in the future rather than a decline.

The disturbing finding of marked racial disparity in maternal morbidity and mortality is consistent with that reported previously by others.^{23, 24, 25, 26} In multiple regression analysis, this difference could not be explained by other risk factors that were found to be significantly associated with adverse outcome in univariable analysis. These included age, obesity, history of a chronic medical condition, prior cesarean delivery and gravidity. Education level, marital status and public medical insurance status, factors traditionally associated with socioeconomic status, could not explain the disparity. Others have also not found an explanation for racial disparity in outcome among established risk factors.²⁵

Racial disparity has been reported in several other health indices. The rates of preterm delivery, low birth weight and infant mortality are higher in black women. It is clear that we are far from understanding the causes of racial disparities in maternal outcome. Such understanding will likely require study beyond traditional medical and socioeconomic factors. Considerable controversy exists about the biological reality of race.²⁷ Nevertheless, in our study, as in others, race or ethnicity, as defined in ordinary social terms, is identified as a substantial risk factor for adverse maternal outcome. Since race and ethnicity rather consistently emerge as important factors in both obstetric and other medical situations, investigation of the causation is strongly indicated. Future research strategies should foster a multidisciplinary approach.²⁸ The reasons for racial disparities in medical outcomes need to be understood so that patient care can be optimized and modifiable causes addressed.

Other risk factors are potentially modifiable through appropriate medical care, public health interventions and education of both women and their health-care providers. Obesity remains a difficult, but in principle, a modifiable factor. Other chronic medical conditions can be improved through research and better access to medical care. Efforts can be made to tighten criteria for cesarean delivery. Pregnancy at a more advanced age than desired by a woman may stem from pressure to postpone a family for career development and unplanned pregnancies. The former could be addressed through more enlightened social policies and the latter through better education and access to contraception. Gravidity beyond the family size desired by a woman can also be addressed through education and better access to contraception.

There are several limitations to our study. Our sample size is small, reflecting the rarity of mortality and near-miss. There were 15 cases of near-miss where charts were not available for review. Without the medical record, we did not have information about this group. We recognize that this represents almost 20% of the near-miss group and is a potential source of bias. We suspect that missing medical records may be a random phenomenon in our institution as we also were unable to retrieve records on several initially selected controls. It is possible that cases with missing records may have had medical legal action making them harder to retrieve and therefore potentially sicker than the studied group. We were able, however, to obtain records on all cases of maternal death, of which many had medical legal action. We speculate that the missing records did not change results, although this cannot be completely discounted.

There have been multiple published definitions of near-miss utilized nationally and internationally⁸ This limits the generalizability of our data as well as others. In our study, we chose to use a modification of Mantel¹⁰ and Waterhouse¹² based on systems used in our hospital as criteria for near-miss. Geller et al.¹⁵ published two validated scoring systems after our data had been collected. Those systems assign a point value to five different events; organ failure, ICU admission, transfusion of more than 3 units of blood, extended intubation and surgical interventions. In reviewing our near-miss cases, all 69 cases had one of these events. However, 11 of 69 did not have sufficient ‘points’ by Geller's criteria to be considered near-miss. All of these cases were surgical interventions that did not receive greater than 3 units of blood. Most were prompt hysterectomies for hemorrhage. In addition, in our institution, most critically ill obstetrical patients are managed in the labor and delivery suite with a team of maternal fetal medicine specialists and anesthesiologists. This makes the criteria of ICU admission very institution dependent. Therefore, using ICU admission as a criteria may not accurately reflect the severity of the patient's condition. By Geller's criteria, in her four-factor model, an ICU admission de facto meets criteria for near-miss. Clearly, there is need for standardized criteria in this area.

In summary, we found that older age, obesity, a woman's race or ethnicity, the number of prior pregnancies, presence of a medical condition and a prior cesarean delivery were all predictors of near-miss morbidity when considered one at a time. As in many medical conditions, racial disparity in outcome was unexplained by traditional risk factors. The problem of maternal morbidity and mortality will not be easily solved. Potentially modifiable risk factors should be addressed through education and public health interventions.