Conceived and designed the experiments: JK GL SG PB MK. Performed the experiments: MK. Analyzed the data: GK JK GL SG PB MK. Wrote the paper: GK JK GL SG PB MK.
The authors have declared that no competing interests exist.
Women continue to die unnecessarily during or after pregnancy in the developed world. The aim of this analysis was to compare women with severe maternal morbidities who survived with those who died, to quantify the risk associated with identified factors to inform policy and practice to improve survival.
We conducted a national cohort analysis using data from two sources obtained between 2003 and 2009: the Centre for Maternal and Child Enquiries maternal deaths database and the United Kingdom Obstetric Surveillance System database. Included women had eclampsia, antenatal pulmonary embolism, amniotic fluid embolism, acute fatty liver of pregnancy or antenatal stroke. These conditions were chosen as major causes of maternal mortality and morbidity about which data were available through both sources, and include 42% of direct maternal deaths over the study period. Rates, risk ratios, crude and adjusted odd ratios were used to investigate risks factors for maternal death. Multiple imputation and sensitivity analysis were used to handle missing data.
We identified 476 women who survived and 100 women who died. Maternal death was associated with older age (35+ years aOR 2.36, 95%CI 1.22–4.56), black ethnicity (aOR 2.38, 95%CI 1.15–4.92), and unemployed, routine or manual occupation (aOR 2.19, 95%CI 1.03–4.68). An association was also observed with obesity (BMI≥30 kg/m2 aOR 2.73, 95%CI 1.15–6.46).
Ongoing high quality national surveillance programmes have an important place in addressing challenges in maternal health and care. There is a place for action to reverse the rising trends in maternal age at childbirth, and to reduce the burden of obesity in pregnancy, as well as ongoing recognition of the impact of older maternal age on the risks of pregnancy. Development and evaluation of services to mitigate the risk of dying associated with black ethnicity and lower socioeconomic status is also essential.
Globally, reducing maternal mortality has been recognised as an important challenge facing all governments and international agencies
Nevertheless, because maternal deaths in the developed world are still uncommon, identifying factors that can be addressed to prevent death may be difficult precisely because cases are rare. Comprehensive and lengthy surveillance is needed to generate sufficient information to guide changes in policy or practice. It is increasingly being recognised that the additional study of severe maternal morbidity can complement enquiries into maternal deaths and is therefore of increasing importance to service providers and policymakers in the area of maternal health
The aim of this analysis was to compare the characteristics of women with a range of specific severe maternal morbidities who survived with those who died to quantify the risks associated with identified factors in order to inform policy and practice to improve survival.
The London Multi-centre Research Ethics Committee approved the UKOSS general methodology (04/MRE02/45) and the studies of individual severe morbidities (04/MRE02/46, 04/MRE02/71, 04/MRE02/72, 04/MRE02/73, 07/H0718/54). Surveillance of maternal death through CMACE is a form of national audit and does not require Research Ethics Committee Approval. Collection of data by CMACE was approved by the National Information Governance Board.
Data concerning women who died and women who survived from five specific maternal conditions were analysed: eclampsia, antenatal pulmonary embolism, amniotic fluid embolism, acute fatty liver of pregnancy, and antenatal cerebral stroke. These conditions were chosen for the pragmatic reason that they represent major causes of maternal mortality and morbidity about which data were available through both UKOSS and the confidential enquiry into maternal deaths carried out by the Centre for Maternal and Child Enquiries (CMACE). Data for the analysis were obtained from two separate sources: information about women who died was obtained from the CMACE confidential enquiries into maternal death database and information about women with severe morbidity who survived was obtained from the UKOSS database.
Cases of severe maternal morbidity for these purposes were defined as women with eclampsia, antenatal pulmonary embolism (PE), amniotic fluid embolism (AFE), acute fatty liver of pregnancy (AFLP) and antenatal stroke (for definitions see
Data source | Start of data collection | End of data collection | Duration of data collection (months) | Number of Cases (%) | Estimated number of maternities |
|
|
||||||
All deaths | CMACE | 01/01/2003 | 31/12/2005 | 36 | 60 (60) | 2,114,004 |
All deaths | CMACE | 01/01/2006 | 31/12/2008 | 36 | 40 (40) | 2,291,493 |
|
|
|||||
|
||||||
Amniotic fluid embolism | UKOSS | 01/03/2005 | 28/02/2009 | 48 | 48 (10) | 3,032,560 |
Acute fatty liver of pregnancy | UKOSS | 01/02/2005 | 31/08/2006 | 19 | 56 (12) | 1,147,848 |
Antenatal pulmonary embolism | UKOSS | 01/02/2005 | 31/08/2006 | 19 | 138 (29) | 1,147,848 |
Eclampsia | UKOSS | 01/02/2005 | 28/02/2006 | 13 | 214 (45) | 779,442 |
Antenatal stroke | UKOSS | 01/07/2007 | 31/12/2008 | 18 | 20 (4) | 1,168,233 |
|
|
*Maternal deaths from eclampsia (n = 14), antenatal pulmonary embolism (n = 28), amniotic fluid embolism (n = 30), acute fatty liver of pregnancy (n = 3) and antenatal stroke (n = 25).
Data from Office for National Statistics, Key Population and Vital Statistics.
**Number of women after exclusion of fatal cases for the five morbidities.
Cases for UKOSS studies were additionally ascertained by contacting clinicians in intensive care units, specialist liver units and radiology departments, as appropriate. Very few additional cases (n = 3) were identified through these additional sources. Maternal deaths were also ascertained through UKOSS, and routine cross-checking with CMACE data showed a high level of case ascertainment. However, for the purposes of this analysis, all maternal death cases were excluded from UKOSS data to ensure these data were limited solely to cases of morbidity where the woman survived.
The methodology of the confidential enquiries into maternal deaths has been described in detail previously
UKOSS data were collected during different durations and time periods from CMACE. However all the UKOSS data were collected during the 2003–2008 CMACE data collection period. As maternal deaths are rare, we took the approach of using the CMACE data available over the full time period in order to maximize the statistical power of these analyses.
The incidence of maternal mortality and severe maternal morbidity and the ratio of survivors to deaths were calculated for each condition with 95% confidence intervals (CI). The denominator used was the number of maternities (women delivering one or more live or stillborn infants) during each study period, estimated from the most appropriate UK birth data available (
In order to investigate trends in continuous variables, unadjusted relative risks with 95% CIs were calculated across groups and Spearman correlation was used to examine the associations between age and BMI and the risk of death. We further investigated the potential factors underlying mortality differences in severe maternal morbidities by using a logistic regression analysis. Factors were included where there was a pre-existing hypothesis or evidence to suggest that they may be associated with maternal mortality. In order to adjust for any effect relating to the individual morbidities, we included an adjustment factor for each condition in the analysis. We developed a full regression model by including both potential explanatory and confounding factors. We tested continuous variables for departure from linearity by the addition of quadratic terms to the model and subsequent likelihood ratio testing. We calculated adjusted odds ratios with 95% CI.
The factors included in the model were maternal age, parity, body mass index (BMI), smoking during pregnancy, ethnicity, and socioeconomic classification based on occupation. Occupation was classified according to the Office for National Statistics socio-economic classification
Data were missing for ethnicity and BMI for between 12% and 23% of cases. We investigated two different methods of analysis to account for this. In a first analysis we included all participants, with creation of a categorical indicator variable for missing responses (missing indicator). The second analysis included all participants with missing responses by using multiple imputation. Missing data for BMI, type of employment, age, parity, smoking and ethnicity were imputed using chained equations
Maternal pre-existing physical and psychiatric conditions were reported only for the 2006–08 CMACE data. There were additionally fewer missing data for these women (7% missing for BMI, 20% missing for occupation). We therefore repeated the multivariate analyses with both missing indicator and multiple imputation models in a subset analysis, using only data on the population of women who died from 2006 to 2008 (n = 40) and the women who had a severe maternal morbidity and survived (n = 476) to investigate the role of pre-existing maternal conditions in death following severe morbidity.
In addition, we assessed the robustness of the analysis, given the limitation of the missing data, by undertaking a series of sensitivity analyses. In each case we assumed specific extreme scenarios and apportioned missing values accordingly.
We assessed the additive effect of the presence of multiple risk factors on the risk of death in a model including all factors found to be significantly associated. The final model included maternal age over 30, BMI equal or over 30 kg/m2, black Caribbean or African ethnicity and unemployed, routine or manual occupation.
We used Stata version 10 software for all analyses (StataCorp, College Station, TX).
One hundred maternal deaths from one of the five specific causes under investigation were identified from the CMACE database to have occurred between 2003 and 2008. A total of 476 women with severe morbidity who survived were identified through UKOSS (
The estimated mortality and morbidity rates are shown in
Incidence of maternal death per 1 000 000 maternities (95%CI) | Incidence of survival from severe morbidity per 1 000 000maternities (95%CI) | Ratio of survivors to deaths (95%CI) | |
Amniotic fluid embolism | 6.8 (4.6–9.7) | 15.8 (11.7–21.0) | 2.3∶1 (1.2∶1 to 4.6∶1) |
Acute fatty liver of pregnancy | 0.7 (0.1–2.0) | 48.8 (36.9–63.4) | 70∶1 (18∶1 to 634∶1) |
Antenatal pulmonary embolism | 6.4 (1.4–9.2) | 120 (101–142) | 19∶1 (11∶1 to 101∶1) |
Eclampsia | 3.2 (1.7–5.3) | 275 (239–314) | 86∶1 (45∶1 to 185∶1) |
Antenatal cerebral stroke | 5.7 (3.7–8.4) | 17.1 (10.5–26.4) | 3.0∶1 (1.3∶1 to 7.1∶1) |
Numerator data obtained from UKOSS and CMACE databases and denominator data from the Office for National Statistics, Key Population and Vital Statistics 2007. Office for National Statistics, Newport.
Increasing maternal age and body mass index were significantly associated with the risk of death (
Maternal deaths n = 100, n(%) | Survivors n = 476, n(%) | Crude OR |
Adjusted OR |
Adjusted OR |
|
Age (years) | |||||
<30 | 32 (32) | 269 (57) | 1 | 1 | 1 |
30–34 | 40 (40) | 120 (25) | 2.80 (1.68–4.68) | 2.89 (1.56–5.36) | 2.58 (1.36–4.90) |
≥35 | 27 (27) | 85 (18) | 2.67 (1.51–4.71) | 2.04 (1.01–4.11) | 2.36 (1.22–4.56) |
Missing | 1 (1) | 1 (0) | |||
Parity | |||||
Nulliparous | 46 (46) | 286 (60) | 1 | 1 | 1 |
Multiparous | 53 (53) | 189 (40) | 1.74 (1.13–2.70) | 0.75 (0.44–1.28) | 0.76 (0.45–.1.29) |
Missing | 1 (1) | 1 (0) | |||
Ethinicity | |||||
White | 65 (64) | 352 (74) | 1 | 1 | 1 |
Black Caribbean and African | 19 (20) | 45 (9) | 2.44 (1.34–4.44) | 2.40 (1.14–5.06) | 2.38 (1.15–4.92) |
Other minority ethnic groups | 15 (15) | 72 (15) | 1.20 (0.64–2.23) | 1.16 (0.54–2.51) | 1.31 (0.65–2.76) |
Missing | 1 (1) | 7 (1) | |||
BMI (kg/m2) | |||||
<30 | 53 (53) | 327 (67) | 1 | 1 | 1 |
≥30 | 24 (24) | 84 (18) | 1.83 (1.06–3.14) | 1.57 (0.83–2.97) | 1.71 (0.91–3.19) |
Missing | 23 (23) | 65 (14) | |||
Occupational classification | |||||
Managerial | 20 (25) | 113 (24) | 1 | 1 | 1 |
Intermediate occupation | 19 (23) | 101 (21) | 1.06 (0.54–2.10) | 1.41 (0.65–3.07) | 1.33 (0.59–2.95) |
Manual or unemployed | 42 (52) | 207 (43) | 1.14 (0.64–2.05) | 2.33 (1.13–4.80) | 2.19 (1.03–4.68) |
Missing | 19 (19) | 55 (12) | |||
Smoking during pregnancy | |||||
Yes | 13 (14) | 89 (19) | 0.67 (0.35–1.26) | 0.62 (0.29–1.32) | 0.58 (0.27–1.26) |
No | 82 (86) | 376 (79) | 1 | 1 | 1 |
Missing | 5 (5) | 9 (2) |
*Odds of death.
In the analysis restricted to the 2006–8 maternal death data, 107 (22%) of the women who survived had a pre-existing medical condition compared with nine (22%) of the women who died (
Women who died | Survivors | |
n = 40 n (%) | n = 476 n (%) | |
Previous history | ||
Cancer | 0 | 3 (1) |
Previous Thrombotic event | 0 | 3 (1) |
Current disease | ||
Asthma | 1 (2) | 39 (8) |
Auto immune disease | 1 (2) | 3 (1) |
Congenital or aquired Cardiac disease | 1 (2) | 7 (1) |
Diabetes, endocrine disorders | 2 (5) | 12 (2) |
Epilepsy | 1 (2) | 3 (1) |
Essential hypertension | 3 (7) | 8 (2) |
Haematological disorders | 0 | 5 (1) |
Renal or urological disease | 0 | 5 (1) |
Others | 0 | 19 (4) |
Overall | 9 (22) | 107 (22) |
Note that some women had more than one condition.
Maternal deaths n = 40, n(%) | Survivors n = 476, n(%) | Crude OR |
Adjusted OR |
|
Age (years) | ||||
<30 | 10 (25) | 269 (57) | 1 | 1 |
30–34 | 11 (27) | 120 (25) | 2.85 (1.17–6.90) | 4.29 (1.63–11.2) |
≥35 | 19 (48) | 85 (18) | 4.96 (2.23–11.0) | 5.26 (1.86–14.5) |
Missing | 0 | 1 (0) | ||
Parity | ||||
Nulliparous | 19 (48) | 286 (60) | 1 | 1 |
Multiparous | 20 (50) | 189 (40) | 1.59 (0.83–3.06) | 0.62 (0.28–1.35) |
Missing | 1 (2) | 1 (0) | ||
Ethinicity | ||||
White | 26 (65) | 352 (74) | 1 | 1 |
Black Caribbean and African | 8 (20) | 45 (9) | 2.41 (1.03–5.64) | 3.01 (1.08–1.42) |
Other minority ethnic groups | 6 (15) | 72 (15) | 1.13 (0.45–2.84) | 1.49 (0.50–4.45) |
Missing | 0 | 7 (1) | ||
BMI (kg/m2) | ||||
<30 | 24 (60) | 327 (67) | 1 | 1 |
≥30 | 13 (33) | 84 (18) | 2.11 (1.03–4.32) | 2.73 (1.15–6.46) |
Missing | 3 (7) | 65 (14) | ||
Occupational classification | ||||
Managerial | 7 (17) | 113 (24) | 1 | 1 |
Intermediate occupation | 8 (20) | 101 (21) | 1.28 (0.45–3.65) | 1.80 (0.59–5.55) |
Manual or unemployed | 17 (43) | 207 (43) | 1.33 (0.54–3.29) | 3.27 (1.03–10.4) |
Missing | 8 (20) | 55 (12) | ||
Smoking during pregnancy | ||||
Yes | 7 (20) | 89 (19) | 1.06 (0.45–2.50) | 1.06 (0.36–3.13) |
No | 28 (70) | 376 (79) | 1 | 1 |
Missing | 5 (20) | 9 (2) | ||
Pre-existing condition | ||||
Yes | 9 (22) | 89 (19) | 0.83 (0.39–1.80) | 0.88 (0.39–1.99) |
No | 31 (78) | 376 (79) | 1 | 1 |
Missing | 0 | 9 (2) |
*Odds of death.
To explore the effects of potential biases related to the missing values in the model including all the women who died and survived, we tested the sensitivity of our multivariable model to a series of assumptions about the missing values (
|
|
|
BMI (kg/m2) | All missing BMI values assumed to be <30 kg/m2 | |
<30 | 1 | |
≥30 | 1.30 (0.70–2.40) | |
BMI (kg/m2) | All missing BMI values assumed to be ≥30 kg/m2 | |
<30 | 1 | |
≥30 | 1.99 (1.17–3.36) | |
BMI (kg/m2) | Restricted to the period 2006–08 for CMACE data. All missing BMI values assumed to be <30 kg/m2 | |
<30 | 1 | |
≥30 | 2.60 (1.11–6.12) | |
BMI (kg/m2) | Restricted to the period 2006–08 for CMACE data. All missing BMI values assumed to be ≥30 kg/m2 | |
<30 | 1 | |
≥30 | 1.65 (0.75–3.63) | |
Ethnicity | All missing ethnicities assumed to be white | |
White | 1 | |
Black African or Caribbean | 2.25 (1.07–4.71) | |
Indian, Bangladesh, Pakistan, Other minorities | 1.08 (0.50–2.30) | |
Ethnicity | All missing ethnicities assumed to be Black Caribbean or African | |
White | 1 | |
Black African or Caribbean | 2.55 (1.27–5.11) | |
Indian, Bangladesh, Pakistan, Other minorities | 1.16 (0.54–2.49) | |
Employment | All missing occupational codes assumed to be managerial | |
Managerial | 1 | |
Non managerial | 1.17 (0.58–2.36) | |
Manual or unemployed | 1.85 (1.01–3.40) | |
Employment | All missing occupational codes assumed to be manual or unemployed | |
Managerial | 1 | |
Non managerial | 1.41 (0.65–3.06) | |
Manual or unemployed | 2.14 (1.07–4.27) |
*adjusted for age, parity, ethnicity, BMI, employment and smoking.
Analysis of the combined effects of the risk factors present showed that the odds of death associated with these severe maternal morbidities increased progressively in the presence of more than one of the risk factors identified (
Number of risk factors present | OR [95%CI] |
0 | 1 |
1 | 1.35 (0.67–2.75) |
2 | 2.77 (1.33–5.76) |
3 | 4.40 (1.76–11.0) |
4 | 8.45 (0.49–149) |
Risk factors included: age ≥30; unemployment, routine or manual occupation; black Caribbean or African ethnicity; and a BMI equal or over 30 kg/m2.
This is the first study to compare on a national basis the characteristics of women who die in pregnancy with those who survive after experiencing one of a series of severe maternal morbidities. This analysis is uniquely possible because of the systems that exist in the UK to collect information about these women
Previous analyses have shown, independently, that women from minority ethnic groups are more likely to suffer from a severe morbidity in pregnancy
There is a growing trend in developed countries for childbearing to occur at a later time in women's lives
Obese women were also more likely to die following severe maternal morbidity in our analysis. The analysis was limited by missing data but the association of obesity with death was significant when the analysis was performed including only the group of women who died between 2006 and 2008, when the data were more complete, even after adjustment for the presence of pre-existing medical conditions. Obesity is an increasingly important public health problem throughout the developed world. Obese pregnant women generally require care from a wide range of health professionals, have more complex pregnancies and require more interventions
Few studies have compared socioeconomic differences in mortality according to the cause of death
This study has highlighted two factors previously unreported as being linked with progression from morbidity to death, depressive illness and learning or intellectual disability. Women with depressive illness were over-represented by five-fold amongst the women who died from severe obstetric morbidity. Recent reports
Our analysis assumes that severe morbidity per se is a better outcome than mortality, whereas the morbidity itself may also represent a failure of management of less severe morbidity. Severe morbidity may be associated with long-term disability as a consequence of, for example, stroke associated with hypertensive disorders of pregnancy, or hypoxic brain injury following resuscitation from amniotic fluid embolism
Collection of the cases was performed nationally across the UK for both the women who died and those who survived. The UK is, to our knowledge, the only country where both the necessary data collection systems exist to allow this type of analysis. In this analysis we took advantage of data which had already been collected to allow us to rapidly investigate the progression from morbidity to death without the need to carry out a new study which would take several years to complete. However, the use of existing data has limitations. Some major causes of direct maternal death such as sepsis were not included because data were not collected by the UK Obstetric Surveillance System during the time period for which maternal death data were available. Similarly, we were unable to include cases of hemorrhage, because there were insufficient data in the CMACE database to identify cases with the same definitions as used in UKOSS. However, the specific range of conditions studied accounted for 42% of direct maternal deaths over the 2003–2008 period in the UK and cover all other major groupings of cause of death. Cases of indirect maternal death were not included; these results are therefore only generalizable to women suffering from direct obstetric morbidity. Ongoing surveillance of a full range of disorders causing both maternal mortality and morbidity would allow analysis of a wider range of conditions in the future; however, the marginal benefits of such an approach need to be weighed against the expense and burden to clinicians of reporting a much larger number of cases, compared with the current targeted approach focusing on specific morbidities in a changing programme.
UKOSS data were collected during different durations and time periods from CMACE. However all the UKOSS data were collected during the 2003–2008 CMACE data collection period and we believe that significant bias associated with these different data collection timings is unlikely. As maternal deaths are rare, this approach maximizes the statistical power of these analyses and therefore improves the validity of the results. However, the two sets of data were collected separately and therefore the number of comparable data items is limited. It is also possible that there was differential case ascertainment between the two systems, although since both systems use several methods to ensure maximal case ascertainment, we believe this is unlikely to have led to significant bias.
Data about BMI and occupational status were missing for a substantial proportion of women, particularly for those who died; we therefore performed multiple imputation analyses assuming that data were missing at random. As more data were missing in the CMACE collection we included the outcome (death or severe maternal morbidity) in the multiple imputation model as this method has been shown to provide the best results when dealing with missing data
The messages from this study can be used to inform actions to reduce maternal mortality throughout the developed world. Ongoing high quality national surveillance programmes still have an important role to play in addressing new challenges in maternal health and care. Women from vulnerable populations in high resource countries remain at increased risk of maternal death in the presence of severe maternal morbidities. This study has identified that women with a history of depressive illness and intellectual disability are over-represented amongst women who die, suggesting a need for rapid referral systems for women with these co-morbidities and pregnancy morbidity. There is a clear place for public health action to reverse the rising trends in maternal age at childbirth and clinical action to mitigate its effects, and to reduce the burden of obesity in pregnancy. Further research is needed to address weight management prior to, during and after pregnancy. In addition, development and evaluation of services to mitigate the risk of dying associated with being of black Caribbean or African ethnicity and being unemployed or from routine or manual socioeconomic groups is essential. It is not clear whether the increased risk of death is related to difficulties in access to maternal care through physical (location) or cultural factors. There is thus a place for more in depth studies to determine exactly why the presence of these factors makes women more likely to die. As the latest figures from the World Health Organisation indicate
(DOCX)
This study would not have been possible without the contribution and enthusiasm of the UKOSS and CMACE reporting clinicians who notified cases and completed the data collection forms. We would also like to thank Matthias Pierce and Maria Quigley for their statistical advice.