Beyond maternal death: Improving the quality of maternal care through national studies of near-miss maternal morbidity

Aims

1. To implement a national programme of study of near-miss maternal morbidity to complement confidential enquiries into maternal deaths.

2. To use mixed methodologies to improve the evidence base for disease prevention and treatment and to inform commissioning of maternity services.

3. To use the data to develop recommendations for best practice to prevent and manage near-miss maternal morbidities.

Objectives

1. To determine the incidence of the specific morbidities most commonly leading to maternal death in the UK.

2. To assess the contribution of existing risk factors to disease incidence and identify steps that may be taken in clinical practice to address these factors to reduce incidence.

3. To describe how the conditions are managed and describe any variations in management, exploring the impact that different management strategies or interventions have on outcomes and costs, in order to make recommendations for best practice to improve outcomes for all women.

4. To describe the outcomes of the conditions for mother and infant and identify any groups in which outcomes differ.

5. To investigate which factors influence the risk of death and how these might be addressed to prevent death.

6. To explore whether an external confidential enquiry or a local review approach can be used to investigate and improve the quality of care for affected women.

7. To assess the longer-term impacts of near-miss maternal morbidity for women, their babies and families.

Research questions

• What are the experiences of women who have near-miss maternal morbidity in the UK?

• How does this experience impact on their social and family relationships and future childbearing plans?

• Are there any aspects of the care experience that were particularly good or bad or could be improved?

The sample

We aimed for a maximum variation sample of women living in the UK42,43 covering a wide range of conditions, based on the principal causes of direct maternal deaths identified in recent maternal death enquiry reports. 2 We sought to include women with a range of ethnic and socioeconomic backgrounds and at varying times after their ‘near-miss’ event. Interviews continued until thematic saturation was reached. Our overall sample included 36 women, 10 male partners and one lesbian partner (Table 1). The majority of partners were interviewed together.

Recruitment packs were distributed through a number of routes to ensure a wide, varied sample. Routes included support groups, the National Childbirth Trust, social network forums [Mumsnet (www.mumsnet.com) and Netmums (www.netmums.com)], newspaper advertisement, intensive care clinicians contacted through the Intensive Care National Audit & Research Centre (ICNARC), an advertisement in the UKOSS newsletter and word of mouth. To try and reach a wider ethnic minority population, we had the recruitment packs translated into Bengali and distributed through a consultant in an east London hospital. Note, however, that the final sample had limited ethnic diversity (see Table 1). We sought to include as broad a range of conditions and time distance from the event as possible as we were keen to understand the longer-term effects of a near-miss maternal morbidity. All women had experienced a severe life-threatening complication in pregnancy [thrombosis or thromboembolism, hypertensive disorders of pregnancy, haemorrhage, amniotic fluid embolism (AFE) or sepsis]. Our sample included broad socioeconomic diversity. Interviews took place between 2010 and 2014.

After obtaining informed consent, one of the authors (LH) interviewed participants in the setting of their choice (usually their home). Participants were asked about their or their partner’s experiences of pregnancy and life-threatening illness. The interview started with an open ended narrative section during which respondents described what had happened, followed by a semistructured section with prompts to explore any relevant issues that had not already emerged, including their recovery and family life since their near-miss. The interviews were all audio- or video-taped and transcribed verbatim.

The analysis reported here focuses on the implications that emerged from the data for quality improvement, commissioning and clinical practice across the care pathway for women and their families. Verbatim quotations are used to illustrate themes that emerged from the data.

Analysis

The transcripts were read and reread, a coding frame was constructed and the data coded. Anticipated and emergent themes were then examined across the whole data set as well as in the context of each person’s interview. A qualitative interpretive approach was taken, combining thematic analysis with constant comparison. 44,45 NVivo 9 (QSR International, Warrington, UK) was used to facilitate the analysis.

Implications for quality improvement, commissioning and clinical practice

This section pulls together all the implications for quality improvement, commissioning and clinical practice that have been identified from analysis of the data. Implications were identified across the care pathway:

• the emergency

• experiences of intensive care

• transfer from intensive care

• contact with the baby

• communication

• the aftermath

• follow-up

• support in the community

• emotional recovery

• counselling

• impact on the family – partners

• impact on the family – family and family life

• information and support

• good practice.

Teaching and learning points

The themes identified in the interviews were summarised as a series of teaching and learning resources. They are intended as a resource for clinicians across the care pathways of these women and their families, including midwives, obstetricians, nurses, anaesthetists, intensive care specialists, health visitors and GPs. Below are the key learning points from each of the summaries. A fuller description and clips are available on www.healthtalk.org. 46 Note that themes relevant to both analyses are repeated.

Amniotic fluid embolism

Amniotic fluid embolism although rare, remains one of the leading causes of direct maternal mortality in high-income countries,2,55,56 characterised by sudden unexplained cardiovascular collapse, respiratory distress and disseminated intravascular coagulation. The rarity of AFE together with the fact that clinical diagnosis of the condition is one of exclusion makes it difficult to obtain reliable information concerning incidence, risk factors, management and outcomes. Previous reviews56,57 suggest that the reported total incidence of AFE varies from 1.9 per 100,000 to 7.7 per 100,000 maternities; the reported incidence of fatal AFE cases varies from 0.4 per 100,000 to 1.7 per 100,000 while the case fatality rate owing to AFE ranges from 11% to 43%. There is little consistency in the factors reported to be associated with the occurrence of AFE and very limited data available regarding factors associated with serious maternal outcomes.

Placenta accreta/increta/percreta

Three variants of abnormally invasive placentation are recognised: placenta accreta, in which placental villi invade the surface of the myometrium; placenta increta, in which placental villi extend into the myometrium; and placenta percreta, in which the villi penetrate through the myometrium to the uterine serosa and may invade adjacent organs, such as the bladder. Placenta accreta, increta or percreta are associated with major pregnancy complications58 and are thought to be becoming more common59 owing to a number of factors including rising maternal age at delivery and an increasing proportion of deliveries by caesarean. 60,61 This finding is of particular concern in the context of increasing rates of caesarean delivery and older maternal age at childbirth. 62,63 However, the risk associated with these factors has not been quantified on a population basis in studies using robust clinical and pathological definitions. There is also limited information to guide the optimal management of this condition.

HELLP/ELLP syndrome

Haemolysis, elevated liver enzymes and low platelets syndrome is a serious complication of pregnancy that usually occurs in women who have signs of pre-eclampsia. 64 In the absence of haemolysis, the condition has been called ELLP syndrome. 65,66 Incidence estimates vary widely and there has been no comprehensive study of the risk factors for this complication. There is also debate over the optimal management of the syndrome, particularly with regard to women who develop the condition remote from term. 66,67

Uterine rupture

Uterine rupture is a complication of pregnancy associated with severe maternal and fetal morbidity and mortality. In high-income countries it most commonly occurs in women who have previously delivered by caesarean section. 68 This observation has led to debate about the optimal management of labour and delivery in women who have delivered by caesarean section in previous pregnancies. Women with a previous caesarean delivery have generally been encouraged to attempt a trial of labour in subsequent pregnancies,69 but recent reports of an increased risk of morbidity, particularly owing to uterine rupture, are thought to have contributed to a marked decrease in some countries in the number of women attempting vaginal birth after caesarean section. 70

Research questions

The following questions were addressed in all studies:

• What is the incidence of the condition?

• Are any factors (e.g. age, parity, obesity, ethnicity, socioeconomic status, multiple pregnancy, late or poor attendance for antenatal care) associated with an increased risk of developing the condition?

• How is the condition managed in the UK?

• What are the outcomes of the condition for women and their babies?

In addition, the following condition-specific questions were addressed.

Case and control definitions

The case and control definitions used for each of the studies were as follows:

Data collection

Cases of the specific near-miss morbidities were notified through the monthly card mailing of UKOSS from all hospitals with obstetrician-led maternity units in the UK. On reporting a case, clinicians were sent a data collection form requesting further details to confirm the case definition and ascertain information regarding potential risk factors, diagnosis, management and outcomes. AFE cases were identified between 1 February 2005 and 31 January 2014; placenta accreta/increta/percreta cases were identified between 1 May 2010 and 30 April 2011; HELLP/ELLP syndrome cases were identified between 1 June 2011 and 31 May 2012; and uterine rupture cases were identified between 1 April 2009 and 30 April 2010.

For the uterine rupture study, controls were obtained from a random sample of obstetrician-led maternity units in the UK in month 4 and month 12 of the study, weighted by the total number of births. The time and day on which reporting clinicians were asked to select controls were randomly identified using data on birth date and time from one county of England (Leicestershire) to try and provide a representative sample of women delivering during each 24-hour period and on different days of the week. Clinicians were then asked to complete a data collection form for these controls. For the placenta accreta/increta/percreta and HELLP/ELLP syndrome studies, clinicians who reported a case were asked to select and complete a data collection form for two controls, identified as the two women meeting the control definition.

All data requested were anonymous and obtained from the woman’s medical records. If complete forms were not returned, up to five reminders were sent. The data were double entered into a customised database. Duplicate reports were identified using information on the woman’s year of birth and expected date of delivery. Cases were checked to ensure that they met the case definition and controls were checked to confirm they had been selected correctly. When data were missing or a data validity check highlighted a problem, the reporting clinicians were contacted to provide or check the information.

Study power

Amniotic fluid embolism: the analysis had 80% power at the 5% level of statistical significance to detect odds ratios (ORs) of ≥ 1.7 and ≥ 2.6, assuming putative risk factors have a prevalence of 40% and 5%, respectively.

Placenta accreta/increta/percreta: the actual number of cases and controls identified during the study gave an estimated power of 80% at the 5% level of significance to detect ORs between 1.9 and 3.3, assuming a prevalence range for potential risk factors of between 5% and 40% in the control women.

HELLP/ELLP syndrome: the actual number of HELLP cases and controls identified gave an estimated power of 80% at the 5% level of significance to detect ORs between 1.8 and 2.9, assuming a prevalence range for potential risk factors of between 5% and 40% in the control women.

Uterine rupture: over the 13-month study period, we anticipated identifying 200 cases (based on an estimated incidence of 1 in 4000 maternities4) and 600 controls. A ratio of three controls per case was planned in the study proposal to maximise the power of the study given that uterine rupture is a rare condition and the number of cases would be limited by disease incidence. Assuming that 10% of women with a previous caesarean section delivering in the UK are induced with prostaglandin and/or receive oxytocin in their labour, and with a 3 : 1 ratio of controls to cases, 106 cases and 316 controls would give an estimated power of 80% at the 5% level of statistical significance to detect a 2.5-fold increase in the odds of uterine rupture in women with a previous caesarean section who have prostaglandin labour induction and/or oxytocin used in labour.

Analysis

The overall incidence with 95% confidence intervals (CIs) of each condition was calculated using the most recently available national birth data that corresponded most closely with the particular study period, as a proxy denominator for the number of maternities during the study period. Denominator data to calculate the incidence and 95% CIs of placenta accreta/increta/percreta in women with and without a previous caesarean delivery and in women with and without placenta praevia diagnosed prior to delivery were estimated using the proportions of women in these various categories observed in the control women together with the most recently available birth data. To calculate the incidence and 95% CIs of placenta accreta/increta/percreta in women with a previous caesarean with and without placenta praevia diagnosed prior to delivery, we used an estimate of the incidence of placenta praevia in women with a previous caesarean delivery (1.2%), derived from a 2010 systematic review. 75 To calculate the incidence with 95% CIs of uterine rupture in women with and without a previous caesarean section, the most recently available birth data were used together with an estimate of the proportion of women in the UK who had previously delivered by caesarean section (15%), derived from the rate in a group of population-based controls comprising women giving birth in the UK in 2005–6. 8 Information on the proportion of women with a previous caesarean delivery planning a vaginal or caesarean section delivery in their current pregnancy, estimated from that observed in the control women, was used to estimate the denominator for calculation of the incidence and 95% CI of uterine rupture according to planned mode of delivery in women with a previous caesarean section. Denominator data to allow calculation of the incidence and 95% CI of uterine rupture in women with a prior caesarean delivery planning a vaginal delivery according to whether labour was induced with or without prostaglandin and/or oxytocin were also estimated using the proportions observed in the control women.

Potential risk factors for each condition were investigated using unconditional logistic regression to estimate ORs and 95% CIs. A full regression model was developed by including both explanatory and potential confounding factors in a core model if there was a pre-existing hypothesis or evidence to suggest they were related to the condition in question; the findings are reported as adjusted odds ratios (aORs) and 95% CIs. Continuous variables were tested for departure from linearity by the addition of first-order fractional polynomials to the model and subsequent likelihood ratio testing. When there was evidence for non-linearity, continuous variables were presented and treated as categorical in the analysis. When there was no evidence of departure from linearity, continuous variables were presented as categorical for ease of interpretation, but have been treated as continuous linear terms when adjusting for them in the analysis. Plausible interactions were tested in the full regression model by the addition of interaction terms and subsequent likelihood ratio testing on removal, with a p-value of 0.01 considered evidence of significant interaction to account for multiple testing. The chi-squared test, Fisher’s exact test or Wilcoxon rank-sum test, as appropriate, were used to compare groups. Logistic regression using robust standard errors (SEs) to allow for the non-independence of infants from multiple births was used when comparing infant outcomes. All analyses were carried out using Stata statistical software version 11 (StataCorp LP, College Station, TX, USA).

Amniotic fluid embolism

HELLP syndrome

Placenta accreta/increta/percreta

Uterine rupture

Risk factors for uterine rupture after prior delivery by caesarean section

A total of 139 (87%) of the uterine ruptures occurred in women who had previously delivered by caesarean section. Table 9 shows the characteristics of these women compared with control women. Women who had two or more previous caesarean deliveries had a higher odds of having a uterine rupture than women with only one previous caesarean delivery (aOR 3.02, 95% CI 1.16 to 7.85), as did women who had an interval of < 12 months compared with ≥ 24 months between their last caesarean section and their last menstrual period in their current pregnancy (aOR 3.12, 95% CI 1.62 to 6.02). There was no evidence to suggest a departure from linearity in the relationship between odds of rupture and number of caesarean deliveries, with the odds of rupture increasing by 3.02 (95% CI 1.62 to 5.63) for every one additional caesarean delivery. However, there was evidence of a non-linear relationship in the association between uterine rupture and caesarean section pregnancy interval, with the odds of rupture appearing to plateau for intervals beyond 12 months (Figure 1).

TABLE 9 - Risk factors for uterine rupture in women with prior delivery by caesarean section

BMI, body mass index.

Percentage of those with complete data.

Adjusted for all factors in the table apart from socioeconomic group, previous uterine incision type(s), previous caesarean uterine closure type(s) and twin pregnancy. When adjusting for age, BMI, parity and number of previous caesarean deliveries, these variables have been treated as a continuous linear term in the analysis.

Risk factor	n (%)a of cases with a previous caesarean (N = 139)	n (%)a of controls (N = 448)	Unadjusted OR (95% CI)	Adjustedb OR (95% CI)
Sociodemographic factors
Age (years)
< 35	94 (68)	313 (70)	1	1
≥ 35	45 (32)	134 (30)	1.12 (0.74 to 1.68)	1.47 (0.89 to 2.45)
Ethnic group
White	94 (69)	325 (75)	1	1
Non-white	42 (31)	111 (25)	1.31 (0.86 to 2.00)	1.12 (0.68 to 1.84)
Socioeconomic group
Managerial and professional occupations	33 (30)	108 (32)	1
Other	77 (70)	226 (68)	1.12 (0.70 to 1.78)
BMI at booking (kg/m2)
< 25	56 (42)	173 (40)	1	1
25–29.9	43 (33)	132 (31)	1.01 (0.64 to 1.59)	1.12 (0.65 to 1.91)
≥ 30	33 (25)	127 (29)	0.8 (0.49 to 1.31)	0.73 (0.41 to 1.30)
Previous obstetric and medical history
Parity
1–2	116 (83)	385 (86)	1	1
≥ 3	23 (17)	62 (14)	1.23 (0.73 to 2.07)	1.1 (0.57 to 2.14)
Number of previous caesarean deliveries
1	121 (87)	368 (82)	1	1
≥ 2	18 (13)	79 (18)	0.69 (0.40 to 1.20)	3.02 (1.16 to 7.85)
Previous caesarean uterine incision type(s)
All low transverse incisions	120 (99)	390 (98)	1
Any non-low transverse incisions	1 (1)	8 (2)	0.41 (0.05 to 3.28)
Previous caesarean uterine closure type(s)
All double	75 (90)	241 (91)	1
All single	5 (6)	17 (6)	0.95 (0.34 to 2.65)
Mixture of double and single or other closure type	3 (4)	7 (3)	1.38 (0.35 to 5.46)
Previous uterine surgery
No	124 (90)	394 (88)	1	1
Yes	14 (10)	52 (12)	0.86 (0.46 to 1.60)	0.92 (0.43 to 1.96)
Previous uterine performation
No	137 (100)	446 (100)
Yes	0 (0)	1 (0)
Current pregnancy
Twin pregnancy
No	139 (100)	444 (99)
Yes	0 (0)	4 (1)
Interval between last caesarean section and last menstrual period (months)
≥ 24	71 (52)	294 (67)	1	1
12–23	35 (26)	99 (22)	1.46 (0.92 to 2.33)	1.38 (0.80 to 2.38)
< 12	31 (23)	48 (11)	2.67 (1.59 to 4.50)	3.12 (1.62 to 6.02)
Placenta praevia
No	136 (98)	445 (99)	1	1
Yes	3 (2)	3 (1)	3.27 (0.65 to 16.40)	28.19 (4.03 to 197.39)
Macrosomia (birthweight ≥ 4000 g)
No	117 (89)	382 (86)	1	1
Yes	14 (11)	62 (14)	0.74 (0.40 to 1.36)	0.85 (0.42 to 1.73)
Planned mode of delivery
Elective caesarean section	20 (15)	250 (56)	1	1
Vaginal	116 (85)	198 (44)	7.32 (4.40 to 12.19)	19.37 (8.53 to 43.98)

FIGURE 1.

Risk of uterine rupture according to the interval between the last caesarean section and start of current pregnancy. a, Adjusted for women’s age as a continuous variable, ethnicity, body mass index as a continuous linear term, parity as a continuous linear term, number of previous caesarean deliveries as a continuous linear term, previous uterine surgery, placenta praevia, macrosomia and planned mode of delivery.

The presence of placenta praevia increased the odds of rupture (aOR 28.19, 95% CI 4.03 to 197.39), although note that this finding is based on a very small number of women and should be interpreted with caution. The odds of rupture was also raised in women who planned to have a vaginal delivery in their current pregnancy compared with women who planned to deliver by elective caesarean section (aOR 19.37, 95% CI 8.53 to 43.98). This finding was irrespective of whether or not the women who planned to have a vaginal delivery had their labour induced and/or received oxytocin in labour (Table 10). However, the women who had prostaglandin labour induction and/or oxytocin used in labour appeared to have raised odds of rupture compared with the women who laboured without prostaglandin induction or oxytocin in labour (see Table 10). No significant interactions were found.

TABLE 10 - Risk factors for uterine rupture in women with prior delivery by caesarean section

Percentage of those with complete data.

Adjusted for woman’s age as a continuous linear term, ethnicity, body mass index as a continuous linear term, parity as a continuous linear term, number of previous caesarean deliveries as a continuous linear term, previous uterine surgery, interval between last caesarean section and last menstrual period as a categorical term, placenta praevia and macrosomia.

Labour either not induced or induced without prostaglandin.

Risk factor	n (%)a of cases with a previous caesarean (N = 139)	n (%)a of controls (N = 448)	Unadjusted OR (95% CI)	Adjustedb OR (95% CI)
Planned elective caesarean section delivery	20 (17)	250 (58)	1	1
Planned vaginal delivery and
laboured without prostaglandin inductionc or oxytocin in labour	52 (45)	132 (31)	4.92 (2.82 to 8.60)	12.74 (5.44 to 29.87)
labour induced with prostaglandin and oxytocin not used in labour	10 (9)	9 (2)	13.89 (5.06 to 38.10)	35.91 (10.38 to 124.28)
laboured without prostaglandin inductionc but oxytocin in labour	28 (24)	33 (8)	10.61 (5.38 to 20.91)	35.36 (13.38 to 93.41)
labour induced with prostaglandin and oxytocin used in labour	6 (5)	5 (1)	15.00 (4.21 to 53.48)	52.05 (11.30 to 239.84)

While dinoprostone (prostin, propess, prostaglandin E) was the agent used for all the control women induced with prostaglandin, dinoprostone was used in 82% and misoprostol in 18% of the women induced with prostaglandin who had a uterine rupture. Intrauterine death was the indication for induction for all of the women who received misoprostol.

A further exploratory analysis found that among women with a previous caesarean delivery, uterine rupture was no more likely to occur in women who had not had a previous vaginal delivery compared with those who had (79%, 110/139 of women with uterine rupture had not had a previous vaginal delivery compared with 74%, 331/446 of control women, aOR 1.43, 95% CI 0.65 to 3.13).

Data sources

Analyses

Stata 11 statistical software was used for all analyses. Frequencies of demographic and clinical variables were tabulated for all studies and compared across groups using chi-squared tests for categorical variables and a Student’s t-test or Wilcoxon rank-sum test as appropriate for continuous variables.

UK Obstetric Surveillance System

A national prospective case–control study of all peripartum women diagnosed with severe sepsis (including septic shock), irrespective of the source of infection, was carried out in all obstetrician-led maternity units in the UK from 1 June 2011 to 31 May 2012.

As there is currently no standardised definition for severe sepsis in pregnant and peripartum women, the study definition for surveillance purposes was developed based on previous literature and by consensus of the UKOSS steering committee. 10 In the non-obstetric population, consensus definitions of sepsis severity (SIRS, sepsis, severe sepsis and septic shock) were developed in 1992. 119 However, these definitions and subsequent improvements are often not applicable to pregnant and peripartum women because clinical signs and symptoms of severe infection differ in this population. Specifically, SIRS can be a sign of ruptured membranes and changing biochemistry associated with labour and delivery, as well as a clinical marker of severe infection. Therefore, the clinical parameters of SIRS in the presence of an infection are often altered in the obstetric population. We adopted the ‘obstetric SIRS’ criteria from a 2001 study of severe obstetric morbidity4 and took into account clinical management (level 2 or level 3 critical care)120 and whether or not the woman died. The full case definition for this study is listed in Box 2. Controls were women who did not have severe sepsis and who delivered immediately before each case in the same hospital. For women transferred to higher-level hospitals, controls were drawn from the delivery hospital. The source population was, thus, all women giving birth in the UK.

The study included a descriptive analysis of the incidence, causative organisms, sources of infection and outcomes of severe sepsis, and a case–control analysis of factors associated with severe sepsis and septic shock. In order to assess risk factors for developing severe sepsis, all cases were compared with non-septic controls. To assess the risk of progression to septic shock, cases with a diagnosis of septic shock were compared with all other cases with severe sepsis that did not develop into septic shock.

The incidences of severe maternal sepsis and septic shock with 95% CIs were calculated using the number of maternities reported in the most recent national birth data78,90,91 as the denominator. Women with signs and symptoms of sepsis prior to delivery were classified as antepartum cases. Sources of infection, causative organisms, and sepsis severity characteristics were tabulated for all cases and stratified according to partum status, as pathogenesis is known to differ between pregnant and postpartum women. 121

For risk factor analyses, sociodemographic, medical history and delivery characteristics with a priori evidence of an association with sepsis were compared between cases and controls, and between cases with and without septic shock. Sources of infection and causative organisms were also assessed as risk factors in the latter comparison. The proportion of missing data in this study was very low; the only variables with substantial missing data (> 1%) were source and organism of infection and socioeconomic group. In order to account for the missing data for sources of infection, causative organisms and socioeconomic group, the subcategories of ‘unknown’ and ‘no laboratory-confirmed infection’ were included for these variables in all analyses.

The odds of severe sepsis and septic shock associated with each risk factor were estimated using univariable unconditional logistic regression and were then adjusted using multivariable unconditional logistic regression. For both the severe sepsis and septic shock outcome groups, factors were adjusted in two stages. First, all a priori sociodemographic and medical history factors, with the exception of previous caesarean delivery and previous pregnancy problem (as these were dependent on parity) and partum status (as the control population was only women who had delivered), were included in a primary model. Second, delivery factors were then adjusted for a priori risk factors using a more parsimonious approach in order to avoid overadjustment or substantial colinearity given the large number of variables. Results were adjusted only for a priori factors from the primary model that were known risk factors, were significant in the primary model at a p-value of < 0.05, or were plausible confounders as identified in previous literature. Delivery characteristics were evaluated for postpartum cases only, as this set of risk factors pertained specifically to delivery.

Results of both stages of adjustment are reported as uORs and aORs and their 95% CIs for severe sepsis. For ease of presentation of risk factors for progression to septic shock, results are reported only for factors included in the final adjusted models. LR-tests with a significance level of p < 0.01 were used to check for interactions between variables; no significant interactions were identified in the final adjusted models.

Within a 1-year study period, we anticipated approximately 316 cases of severe sepsis based on an estimated incidence of 4 per 10,000 maternities. 4 For the severe sepsis risk factor analysis, with two controls per case, and for a risk factor prevalence of at least 5% in control women, the study was estimated to have had 80% power at p < 0.05 (two-sided) to detect a statistically significant OR of ≥ 2.3. The actual number of cases and controls identified during the study period of 12 months generated an estimated power of 80% at the 5% level of significance to detect an OR of ≥ 2.1, for the same risk factor prevalence level. For the septic shock risk factor analysis, for a risk factor prevalence of at least 15% in women without septic shock, the analysis had 80% power at the 5% level of significance to detect an OR of ≥ 2.6.

Aberdeen Maternity and Neonatal Databank

The total study population comprised 89 cases of uncomplicated maternal sepsis, 14 cases of severe maternal sepsis and 412 control women. The majority of all cases occurred postpartum and before hospital discharge [98.9% (n = 88) of women with uncomplicated sepsis and 92.9% (n = 13) of women with severe sepsis]. One woman had uncomplicated sepsis antepartum and one women with severe sepsis was diagnosed with an infection during delivery, which later progressed to septic shock postpartum. Sepsis case rates for the 23-year study period are shown in Figure 2. Although rates of severe sepsis remained relatively constant over the study period, rates of sepsis overall have increased significantly since 2003 (p = 0.002) compared with the previous two decades. Given the increase in awareness of sepsis, we cannot exclude that this represents an increase in diagnosis/reporting rather than an increase in disease.

FIGURE 2.

Incidence of all sepsis and severe sepsis cases per 1000 maternities from 1986 to 2009 in Aberdeen (3-year rolling averages). Reproduced with permission from Acosta et al. , © The Authors. BJOG An International Journal of Obstetrics and Gynaecology © 2012 RCOG.

Over the 23-year study period, the proportion of spontaneous vaginal deliveries decreased while the proportion of operative vaginal deliveries and caesarean sections increased significantly (p < 0.001, p = 0.009 and p < 0.001, respectively). Despite these trends, sepsis case rates remained predominantly the highest among women who had a caesarean section throughout the study period, although rates increased in all modes from 2003 onwards. Antibiotic usage (before discharge from the labour ward) in the overall study population increased steadily from 1986 to 1996 (p < 0.001) and plateaued from 1997 to 2009. Although there was no significant difference between cases and controls in the total proportion of antibiotic usage across all modes of delivery (Table 11) among the women with sepsis who had a caesarean section, 38.5% received antibiotics during pregnancy or delivery, which was significantly lower than the 70.2% of control women who had a caesarean section and received antibiotics (p < 0.001). In addition, antibiotic usage among control women who had a caesarean section increased significantly over time (p = 0.003), whereas antibiotic usage among cases who had a caesarean section did not increase over time. There was no significant difference in antibiotic usage among women in the case and control groups who had a manual placenta removal or an operative vaginal delivery.

After adjusting for changes over time and the other factors in each model, factors significantly associated with both uncomplicated sepsis and severe sepsis were younger age, multiparity, anaemia, operative vaginal delivery and caesarean section (compared with spontaneous vaginal delivery) (see Table 11). Obesity was significantly associated with uncomplicated sepsis; however, this association was not present with severe sepsis. Additionally, operative vaginal delivery and pre-term birth were associated with uncomplicated sepsis, whereas induced labour was associated with severe sepsis.

Intensive Care National Audit & Research Centre

In the CMP database between 2008 and 2010, there were 646 pregnant or recently pregnant women who met the case definition for severe sepsis, which represented 14.4% of maternal critical care unit admissions and 10.6% (n = 474) had septic shock. The absolute risk of maternal critical care admission with severe sepsis was 4.1 per 10,000 maternities (95% CI 2.9 to 5.6).

Characteristics of the cohort are shown in Table 12. Women had a significantly higher risk of admission to critical care with severe sepsis if they were aged < 20 years or ≥ 40 years than if they were aged 25–29 years. Increased risk of severe sepsis was also significantly and progressively associated with increasing deprivation based on area of residence.

The source of infection could be identified from the primary reason for admission to the critical care unit for 598 women (92.6%). Frequencies of the reported source of infection among women admitted with severe sepsis are shown in Figure 3. The most common source of infection was pneumonia/respiratory infection (n = 257; 40%). Of these, only 27 were identified as laboratory confirmed cases of AH1N1 influenza (2009 and 2010 were AH1N1 influenza pandemic years).

FIGURE 3.

Source of infection among pregnant or recently pregnant women admitted to critical care units in England, Wales and Northern Ireland 2008–10.

Of all severe maternal sepsis admissions, 4.6% (n = 29) died. The absolute risk of acute hospital mortality of women admitted was 1.8 per 100,000 maternities (95% CI 1.1 to 2.8). Pneumonia/respiratory infection was the most common source of sepsis among women who died (n = 12; 41.4%). Factors associated with mortality are shown in Table 13. After adjustment for a priori factors and changes over time, deprivation, being overweight or obese, respiratory dysfunction and haematological dysfunction were significant independent risk factors for mortality.

UK Obstetric Surveillance System

During the study period, there were a total of 365 confirmed cases of severe sepsis out of 780,537 maternities in the UK,11–13 representing an incidence of 4.7 per 10,000 maternities (95% CI 4.2 to 5.2). Seventy-one women (20%) developed septic shock, which represents an incidence of 0.91 per 10,000 maternities (95% CI 0.71 to 1.15). Data were collected on 757 controls.

Laboratory-confirmed infection was reported for 233 (63.8%) severe sepsis cases and a source of infection was identified for 270 cases (74.0%); 60 cases (16.4%) had neither a source of infection nor causative organism identified. The distribution of sources of infection, causative organisms and severity characteristics are shown in Table 14 and Figure 4. Overall, the largest proportion of cases was due to genital tract infection (31.0%) and the most common organism causing infection was Escherichia coli (21.1%). However, the distributions of both the infection source and the causative organism differed significantly between women with antepartum and postpartum sepsis (p < 0.0001 for both), as did the risk of septic shock. Readmission (for reasons other than delivery) also differed significantly between the two groups: 108 (48%) women with postpartum sepsis were readmitted, compared with six (5%) women with antepartum sepsis (p < 0.0001). Of all cases, 286 (78%) received level 2 or intensive care and five women died (see Table 14). Of the women who died, two had infection with E. coli and three women had an unknown causative organism. Twenty-nine women (8%) with severe sepsis had either a miscarriage or a termination of pregnancy. For women diagnosed with severe sepsis antenatally, 5 of 137 infants were stillborn (3.6%) and 7 died in the neonatal period (5.1%). Fifty-eight infants (42.3%) were admitted to the NICU.

FIGURE 4.

Distribution of causative organisms according to (a) source of infection; and (b) mode of delivery. Stacked bars represent the number of cases with specific causative organisms according to infection source and mode of delivery categories. Data are mutually exclusive.

There were 296 cases with recorded dates and times for the first sign of SIRS and the severe sepsis diagnosis. For 245 (83%) severe sepsis cases and for 49 (85%) septic shock cases, there were < 24 hours between the first sign of SIRS and the diagnosis of severe sepsis, and for 264 (89%) severe sepsis cases and for 55 (95%) septic shock cases there were < 48 hours between the first sign of SIRS and the diagnosis of severe sepsis. Additionally, for 16 (50%) women with a group A streptococcal infection there were < 2 hours and for 24 (75%) women there were < 9 hours, between the first sign of SIRS and the diagnosis of severe sepsis.

A priori sociodemographic and medical history characteristics of women with severe sepsis compared with control women are listed in Table 15. After adjustment and compared with controls, women who were of black or other minority ethnic origin, were primiparous, had a pre-existing medical problem, or had a febrile illness or were taking antibiotics in the 2 weeks prior to presentation were at significantly increased odds of severe sepsis. There was no statistically significant association between premature rupture of membranes and severe sepsis in either antenatal cases (n = 20, aOR 1.72, 95% CI 0.98 to 3.02) or postnatal cases (Table 16). In addition to significant a priori factors, the following factors significantly increased the odds of severe sepsis in women with postpartum sepsis: having an operative vaginal delivery (aOR 2.49, 95% CI 1.32 to 4.70), having a pre-labour caesarean section (aOR 3.83, 95% CI 2.24 to 6.56), having a caesarean section after the onset of labour (aOR 8.06, 95% CI 4.65 to 13.97), or having a complication of delivery (aOR 1.69, 95% CI 1.09 to 2.63) (see Table 16). Of note, of all women who had a caesarean section, 96.6% of cases and 94.8% of controls received prophylactic antibiotics at delivery.

Multiple pregnancy (aOR 5.75, 95% CI 1.54 to 21.5) and group A Streptococcus as the causative organism (aOR 4.84, 95% CI 2.17 to 10.8) were significantly associated with an increase in the odds of progression from severe sepsis to septic shock. Before adjustment for group A streptococcal infection, spontaneous vaginal delivery (aOR 3.85, 95% CI 1.35 to 10.96) and operative vaginal delivery (aOR 3.12, 95% CI 1.03 to 9.57) were significantly associated with an over threefold increase in the odds of progression to septic shock.

Research questions

• What is the relative cost-effectiveness of treatment for peripartum haemorrhage with uterine compression sutures, factor VIIa and major vessel ligation or embolisation?

• What additional data should be collected in a primary study of therapies for near-miss maternal morbidity to inform an economic evaluation?

Structure of cost-effectiveness model

A decision-tree model of the acute clinical pathway of women (mean age 32 years; interquartile range 29–36 years) with PPH was developed using TreeAge Pro Suite 2015 software (TreeAge Software, Inc., Williamstown, MA, USA). This type of model is recommended for acute conditions that need immediate effective interventions. 133 A time horizon of 12 months was used in the economic model. Treatment success was defined as no requirement for either a further therapy to treat PPH or hysterectomy. The model commenced when first-line treatment for PPH failed and one of the four second-line treatments was used as potential therapies to stop the bleeding. If the second-line therapy was successful, the model stopped and costs and effects were calculated for that therapy. If second-line therapy failed, the woman either underwent hysterectomy or was treated with one of the remaining second-line therapies as a third-line therapy. After receiving third-line therapy, the final possible outcomes were successful treatment or hysterectomy. We assumed that if third-line therapy failed, women underwent hysterectomy and not a fourth-line treatment, as this is most likely in clinical practice. 2,128

Estimation of baseline probabilities

Baseline probabilities were extracted from the UKOSS study. 10,129 Table 17 presents the probabilities used in the model for each intervention that were computed as the number of occurrences divided by the total number of women receiving a particular therapy. Probabilities are presented either as intervention specific or common to all regimes. For the intervention-specific probabilities, the probability of a successful outcome after second-line treatment and the probability of hysterectomy after a failure of second-line treatment were extracted from the epidemiological study. The probability of a failure outcome after second-line therapy was estimated as residuals (i.e. 1 – probability of a successful outcome). The probability of having a particular third-line therapy was also estimated as the residual. The probabilities of having a particular third-line treatment and the outcomes after third-line treatment were common to all regimes. No clear evidence about the probability of use of a particular intervention as third line compared with the remaining ones was observed in the study by Kayem et al. 129 Therefore, we assumed that women were equally likely to receive one of the remaining therapies as third-line treatment. We pooled the number of hysterectomies performed across all interventions used as third line, to estimate the probability of having hysterectomy after third-line treatment.

Incorporating treatment effect in the economic model

The treatment effect in the model was introduced by modifying the baseline probability of successful outcome after uterine compression sutures by a treatment modifier reflecting the treatment effect of uterine compression sutures against one of the remaining second-line therapies. The baseline probability of uterine compression was used because it was the most commonly used intervention among the 272 women identified in the UKOSS study. As treatment modifier, we used the OR of successful outcome after second-line treatment between pelvic vessel ligation, rFVIIa or interventional radiology compared with uterine compression sutures. We used the UKOSS data to obtain ORs from a logistic regression using a binary dependent variable indicating whether or not second-line therapy was successful and treatment indicators as well as a set of women’s characteristics and mode of labour as explanatories. The ORs (95% CIs) estimating treatment success for rFVIIa, pelvic vessel ligation and interventional radiology compared with uterine compression sutures were estimated to be 0.173 (0.068 to 0.469), 0.169 (0.053 to 0.537) and 2.533 (0.531 to 12.087), respectively. Baseline probabilities of successful outcome after second-line rFVIIa, pelvic vessel ligation and interventional radiology were converted first into odds before being multiplied by the OR and were converted back into probabilities when informing the parameters in the model.

Costs

The direct health-care costs associated with the second-line interventions were the only costs included in the model and were calculated using a detailed costing analysis. Any other health-care costs (e.g. length of stay, hospital readmissions, primary care visits) were assumed to be similar across interventions and, therefore, cancelling out in the calculation of the incremental costs and hence were excluded from the model. This is based on the clinical assumption that the outcomes of successful treatment of PPH do not differ in a treatment-dependent manner. The main resource use and cost drivers associated with the interventions were elucidated through interviews with a haematologist, two obstetricians and an anaesthetist all based in UK hospitals. The results from the interviews suggested that staff, equipment, key drugs and a set of general measures common to all regimes were the main cost drivers. General measures included common resources to all therapies such as intra-arterial monitoring, control venous pressure monitoring, active warming and fresh-frozen plasma. Unit costs associated with the resource use were primarily obtained from national sources when available134 and from the literature. If no published cost data were readily available, commercial companies were contacted to obtain such information. All unit costs were expressed in 2012–13 Great British pounds and discounting was not implemented given that only intervention costs were included in the model. A summary of the mean interventions costs used in the model is presented in Table 18.

Base-case utility values

Evidence suggests that women who have a PPH but have a successful outcome after second- or third-line treatment experience a complete recuperation at 1-year follow-up with quality-of-life levels similar to women of the same age from the general population. Utility values associated with a successful outcome health state in the model were therefore extracted from responses to the European Quality of Life-5 Dimensions-3 Level (EQ-5D-3L) instrument135 in the Health Survey for England in 2011 from women aged < 40 years in the general population. The utility value of a successful outcome was based on 359 young women’s responses and we applied the UK value set to estimate utilities obtaining a mean (SE) utility value of 0.923 (0.008). We sought published sources to identify quality-of-life weights associated with women of childbearing age undergoing hysterectomy in an emergency situation by conducting a literature search from inception to May 2014 in MEDLINE, Cumulative Index to Nursing and Allied Health Literature and EMBASE using a combination of the terms quality of life, fertility, infertility and hysterectomy and found no evidence of published utilities. We searched the national health surveys in the UK and USA to explore whether or not such data could potentially be available. We found that in the 2003 US Medical Expenditure Panel Survey (MEPS),136 women who completed the EQ-5D-3L were asked whether or not they had undergone a hysterectomy (which may or may not have been related to PPH) in the previous year. Full details are available from the MEPS survey web site. 137 We identified 148 women under the age of 40 years who underwent a hysterectomy (for any reason) in the 2003 MEPS dataset and estimated a mean (SE) utility weight of 0.744 (0.024). We used the UK value set to calculate utilities from the EQ-5D-3L responses obtained in MEPS. 138 We had originally anticipated using data from the study reported in Chapter 9 to produce utility estimates, but the questionnaire response rate was too low to be considered representative.

Analytical methods

All analyses are carried out from the perspective of the UK NHS and are reported in terms of incremental cost-effectiveness ratios (ICERs) and net-benefit statistics. The primary health outcome measure in the economic analysis was QALYs but the results of the analysis are also presented using hysterectomy avoided. Therefore, ICER results are presented using cost per QALY gained and cost per hysterectomy avoided. QALYs for each treatment strategy were estimated attaching a utility value to the successful outcome or hysterectomy health states and assuming women spent 12 months in that state. Dominated strategies (one strategy is found to be both cheaper and produces more QALYs) were eliminated before conducting any incremental analysis. Uncertainty in the model parameters (probabilities, treatment effect and utility parameters) was evaluated through a probabilistic sensitivity analysis (PSA). 139 In PSA, each input parameter in the model is assigned a specific statistical distribution for which values are randomly drawn generating a large number of pairs of mean costs and effects for each strategy. These estimates are then used to create an empirical distribution of the differences in costs and effects between interventions.

We generated 10,000 random draws and used beta distributions for both probabilities and utilities. 140 As we had access to patient-level data to inform the uncertainty around treatment effects of the interventions, we used the empirical distribution of the ORs reported in Chapter 5, Sensitivity analysis, from 10,000 bootstrap samples for the PSA. We reported the PSA summary results for the total costs and effects per patient associated to each second-line regime to treat PPH for our base case in line with current recommendations. 140 Uncertainty around cost-effectiveness results is presented using cost-effectiveness acceptability curves (CEACs) that indicate the probability that a particular intervention is cost-effective for different thresholds of willingness to pay (WTP) for a QALY or hysterectomy avoided. Current thresholds of WTP for a QALY as suggested by National Institute for Health and Care Excellence (NICE) were used in this study to determine value for money. 141

To estimate the impact of individual parameters on the results of the analysis, we performed an analysis of covariance. Using this approach, the proportion of the sum of squares in the output parameters (i.e. incremental costs and incremental QALYs) explained by uncertainty in each input parameter was identified. 140

Costs were treated deterministically in the economic model, as there were no reliable stochastic estimates to inform SEs. We implemented a one-way sensitivity analysis to determine how the main base cost-effectiveness results varied when the total costs associated with uterine compression sutures, pelvic vessel ligation, rFVIIa and interventional radiology were increased/reduced by 20%, 40% and 60%, as informed by the ranges in Table 18.

Although our literature search found no published evidence on quality-of-life weights associated with women of childbearing age undergoing hysterectomy in an emergency situation, it identified proxy utilities associated with the health state of convalescence after hysterectomy in women with heavy menstrual bleeding. 142 In addition, the 2011 US MEPS143 had also available hysterectomy treatment information and data on the generic quality of life Short Form questionnaire-12 items144 that we translated into EQ-5D-3L utility values using a validated a mapping algorithm. 145 We evaluated the robustness of our original base-case utility estimates running the economic model using these two additional sources of utility information.

Base-case summary results

Table 19 presents a summary of the base-case probabilistic analysis ranking all interventions in ascending order of costs. Both pelvic vessel ligation and rFVIIa were dominated strategies (more expensive and produce fewer QALYs/avoid fewer hysterectomies) by interventional radiology and uterine compression sutures. Therefore, pelvic vessel ligation and rFVIIa were removed and not evaluated further in the base case with the incremental analysis comparing interventional radiology versus uterine compression sutures. The interventional radiology strategy was more expensive but also produced more QALYs than uterine compression sutures, with an estimated incremental cost of £2827 (95% non-parametric CI £1909 to £3948) and incremental QALYs of 0.013 (95% CI –0.006 to 0.032). The ICER representing the value of the additional benefit of interventional radiology compared with uterine compression sutures was estimated to be £210,585 per QALY gained. This indicates that for current NICE thresholds of WTP for QALY gained, interventional radiology does not represent a good use of NHS resources. This was also corroborated by the results of the net monetary benefit analysis that indicated uterine compression sutures had the largest monetary benefit of £12,843, assuming a WTP of £20,000 per QALY gained. Similar results were obtained using hysterectomies avoided as the main outcome measure but in this case, interventional radiology had an estimated ICER of £37,533 per hysterectomy avoided.

Analysis of uncertainty

The CEACs associated with uterine compression sutures and interventional radiology, using QALYs or hysterectomy avoided as measures of outcome in the economic evaluation, are presented in Figure 5. Using QALYs, the use of uterine compression sutures as a second-line treatment strategy always had a higher probability of being cost-effective than interventional radiology for any value of WTP for QALY gained. When using hysterectomy avoided as the outcome measure, interventional radiology had a higher probability of being cost-effective than uterine compression sutures at values of WTP > £30,000 per hysterectomy avoided.

FIGURE 5.

Cost-effectiveness acceptability curves indicating the probability that uterine compression sutures or interventional radiology is cost-effective for different values of WTP for health gained. (a) QALYs as the measure of outcome; and (b) hysterectomy avoided as the measure of outcome.

The results of the analysis of covariance suggested that the probability of success after uterine compression sutures had the largest contribution to the uncertainty around incremental costs explaining 71.7% of the variability of incremental costs and 19.8% of the variability in incremental QALYs. The utility estimate associated with the hysterectomy state contributed to 44.7% of the variability of the incremental QALYs. The remaining parameters did not contribute greatly to the uncertainty around cost-effectiveness results.

Sensitivity analysis

Reducing or increasing uterine compression suture costs by 20%, 40% or 60% did not affect the baseline cost-effectiveness results and uterine compression sutures were still the most cost-effective option for any threshold of WTP. A similar output was obtained when the costs of pelvic vessel ligation and rFVIIa were reduced/increased. Increasing the costs of interventional radiology improved the cost-effectiveness of uterine compression sutures to an almost asymptotic CEAC towards a probability of 1. However, if the costs of interventional radiology were reduced by 40%, it became the most cost-effective intervention for thresholds of WTP > £19,000 per QALY gained but asymptotically to a maximum probability of 70%. Interventional radiology became the most cost-effective intervention for any threshold of WTP per QALY gained if the costs of interventional radiology were reduced by 60%.

The CEACs of uterine compression sutures compared with interventional radiology using alternative utility estimates to inform the quality-of-life weights associated with women undergoing hysterectomy in an emergency situation suggest little impact on cost-effectiveness results compared with the base-case analysis.

Research questions

• What are the characteristics of women with fatal and non-fatal specific severe maternal morbidities?

• What factors are associated with maternal death, including core demographic and pregnancy characteristics (such as maternal age, ethnicity, socioeconomic and smoking status, BMI, parity, multiple pregnancy, inadequate use of antenatal care, pre-existing medical conditions and current and previous pregnancy problems)?

Data sources

Analyses

Maternal characteristics associated with death during pregnancy and childbirth

Increasing maternal age and BMI were significantly associated with the risk of death in the CMACE/UKOSS analysis (Figure 6). Results from the multivariable models are presented in Table 20. The final model using a missing indicator analysis produced very similar effect estimates to those from the multiple imputation analysis. Women who were ≥ 30 years, black Caribbean or African, unemployed or with routine or manual occupation had higher odds of progressing to death. Women who had a BMI ≥ 30 kg/m² were more likely to die from these severe maternal morbidities (aOR 1.71, 95% CI 0.91 to 3.19), although the increased odds were not statistically significant.

FIGURE 6.

Risk of death according to (a) age; and (b) BMI. Bars show risks with 95% CIs.

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 (Table 21), although of note was the high degree of uncertainty around the estimated odds associated with the presence of all four risk factors.

Role of medical comorbidities and other factors associated with direct maternal deaths

Medical comorbidities were found to be significantly associated with direct maternal deaths from eclampsia, pulmonary embolism, severe sepsis, AFE and peripartum haemorrhage, after controlling for a number of known risk factors for maternal mortality (Table 22). The adjusted odds of having pre-existing medical conditions were almost fivefold higher among women who died than women who survived (aOR 4.82, 95% CI 3.14 to 7.40). In addition, we also found substance misuse (aOR 10.16, 95% CI 1.81 to 57.04), inadequate use of antenatal care (aOR 15.87, 95% CI 6.73 to 37.41), previous pregnancy problems (aOR 2.21, 95% CI 1.34 to 3.62), hypertensive disorders of pregnancy during current pregnancy (aOR 2.44, 95% CI 1.31 to 4.52) and Indian ethnic background (aOR 2.70, 95% CI 1.14 to 6.43) to be independently associated with maternal deaths (see Table 22). At a population level, 70% (95% CI 66% to 73%) of the increased risk could be attributed to these six factors, the most important being medical comorbidities (PAF 48.9%, 95% CI 40.5% to 56.2%) followed by previous pregnancy problems, hypertensive disorders of pregnancy, inadequate use of antenatal care services, Indian ethnicity and substance misuse (see Table 24).

Eight out of the 18 medical comorbidities were significantly associated with maternal death (Table 23). The odds of pre-existing musculoskeletal disorders was 12 times higher and the odds of inflammatory/atopic disorders (excluding asthma) was 10-fold higher among women who died than women who survived. Women who died had five times higher odds of suffering from autoimmune diseases and infections, such as sexually transmitted infections, tuberculosis and group B Streptococcus infection, fourfold higher odds of having a pre-existing haematological disorder and more than three times higher odds of essential hypertension than the women who survived. The odds of suffering from mental health problems and asthma were more than two times higher among women who died than women who survived.

The odds associated with maternal death increased by more than three and a half times per unit increase in the ‘risk factor’ score (aOR 3.59, 95% CI 2.83 to 4.56; p < 0.001), after controlling for other variables that are not included in the score. On a population basis, 70% (95% CI 66% to 73%) of the increased risk associated with maternal death could be attributed to the six identified factors, the most important being medical comorbidities followed by previous pregnancy problems, hypertensive disorders of pregnancy, inadequate use of antenatal care services and Indian ethnicity (Table 24).

Secondary data analysis

Two unmatched case–control analyses were conducted using existing UKOSS data collected between February 2005 and January 2013 to examine the associations of ethnic and socioeconomic groups with severe maternal morbidity in the UK. Two sets of cases and controls were used. Cases included in the analysis of socioeconomic inequalities were women who suffered from one of the following six conditions of severe maternal morbidity directly attributable to pregnancy causes: AFE,32 AFLP,30 eclampsia,28 peripartum hysterectomy,31 peripartum haemorrhage129 and uterine rupture. 181 For the ethnic inequality study, in addition to these six conditions, cases were included from a further five conditions: antenatal pulmonary embolism,29 stroke in pregnancy,74 placenta accreta,72 HELLP syndrome54 and severe sepsis. 99 These data sets represented the entirety of data available on direct maternal morbidities and control women at the time the analyses were conducted. The controls were women who delivered immediately before the cases in the same hospital. 158 In total, 1144 cases and 2256 controls were included in the socioeconomic inequality study182 and 1753 cases and 3310 controls were included in the ethnic differences study. 154 For the included sample sizes, assuming a prevalence of exposure of 1.1% (the prevalence of the smallest ethnic minority group, the Bangladeshi group), the analysis had 80% power to detect an OR of ≥ 1.8 associated with severe maternal morbidity at p < 0.05.

Maternal occupation (or husband’s/partner’s occupation if information about women’s occupational status was not available) was used to categorise women into four socioeconomic groups: managerial/professional, intermediate, routine/manual and unemployed (additional category), using the National Statistics Socio-economic Classification. 183 The ethnic groups were classified based on self-reported ethnicity noted in medical records according to the UK national census classification. 149 Potential confounders were those that were adjusted for in previous studies – pregnancy-related factors such as parity, problems during current and previous pregnancies, multiple pregnancy, medical comorbidities, utilisation of antenatal care, maternal age, BMI and smoking status. 152,155,158

Univariable analyses were conducted to assess the association of the exposure variables socioeconomic status and ethnicity, and other independent variables with the outcome (severe maternal morbidity). Multivariable unconditional logistic regression analyses were performed to control for confounders. For the socioeconomic analysis, the potential confounders were added to a model that included the outcome and socioeconomic status variable one at a time with subsequent likelihood ratio testing. All variables found to be associated with the outcome at p < 0.05 (two-tailed) and those identified as important confounders from the literature were included in the final model.

The multivariable models for ethnicity analysis included all variables that were found to be associated with the outcome at p < 0.1 (two-tailed) in the univariable analyses and those identified as important potential confounders in previous literature. Four models were built in a hierarchical fashion. Model 1 adjusted for anaemia in current pregnancy, diabetes in current pregnancy, previous pregnancy problems, pre-existing medical problems and parity. In model 2, smoking status and inadequate utilisation of antenatal care services were added to the other variables. Model 3 included socioeconomic status in addition to those variables included in model 2, and model 4 adjusted for all variables including women’s age and BMI. In addition to controlling for the known risk factors, this approach enabled understanding of their effects on the association between ethnicity and severe maternal morbidity.

Plausible interactions were tested by adding interaction terms and using LR-tests (p < 0.05) and no significant interactions were identified. Information on ethnicity was not available for 1.8% of the sample. Based on the method used in two previous studies, women with unknown ethnicity were included in the ‘white European’ group because the redistributed proportions matched more accurately with the estimated ethnic profiles in the UK population census. 158,184 Although for a majority of the independent variables the number of participants with missing data was < 1% and their distribution did not differ significantly between the cases and controls and among the ethnic groups, for four variables, ‘socioeconomic status’, ‘BMI’, ‘smoking’ and ‘previous pregnancy problems’, the proportion of participants with missing information was > 1%. The data were assumed to be not missing at random and thus coded ‘missing’ as a separate group for these variables. In addition, a series of sensitivity analyses were performed by assuming extreme scenarios and accordingly redistributing the missing observations into the extreme groups; this had no material effect on the findings.

The continuous variables were tested for deviations from linearity,185 which showed non-linear associations between the variables and the outcome. Thus, in the final model the variables were incorporated as categorical variables for the ease of interpretation of the ORs and the advantage of including and testing the missing data as a separate category. All analyses were carried out using Stata 11.

Primary data collection

A national, population-based cohort study was conducted using UKOSS as described in Chapter 3. The cohort included any pregnant woman in the UK of ≥ 20 weeks’ gestation who was of very advanced maternal age. Although very advanced maternal age has generally been used to refer to women aged ≥ 45 years, for pragmatic reasons so as to not overburden reporting clinicians, very advanced maternal age was defined for our purposes as women aged ≥ 48 years at their date of delivery. The cohort was identified through UKOSS between 1 July 2013 and 30 June 2014. The UKOSS reporting clinicians were also asked to identify comparison women, defined as the two pregnant women of ≥ 20 weeks’ gestation who were < 48 years of age at their estimated date of delivery and delivered immediately before the older woman in the same hospital.

The ORs with 95% CIs were estimated throughout using unconditional logistic regression. ORs were adjusted for factors if there was a pre-existing hypothesis or evidence that the factors were potential confounders or mediators of the relationship between advanced maternal age and the outcome in question. To help examine the relative influence of the potential confounders and mediators on the association between maternal age and the outcome in question, models were adjusted in a hierarchical fashion: model 1 adjusted for sociodemographic factors; model 2 additionally adjusted for previous medical history; and model 3 additionally adjusted for relevant pregnancy related factors. ‘Missing’ was included as an extra category for variables that had ≥ 10% of missing data. Continuous variables were tested for evidence of departure from linearity by the addition of first-order fractional polynomials to the model and subsequent likelihood ratio testing. Continuous variables that showed evidence of non-linearity were treated and presented as categorical in the analysis, while those showing evidence of linearity were treated as continuous linear terms when adjusting for them in the analysis but presented as categorical for ease of interpretation. Plausible interactions were tested in the full regression model by the addition of interaction terms and subsequent likelihood ratio testing on removal, with p < 0.01 considered evidence of significant interaction to account for multiple testing.

Women who initially had a multiple pregnancy but then had fetal reduction were classified in the analysis according to the number of fetuses left after the reduction. Spontaneous first-trimester losses in women known initially to have a multiple pregnancy were classified in the analysis according to the post loss number of fetuses. Second-trimester losses in a multiple pregnancy were classified according to the pre-loss number of fetuses in the main analysis, but were not included when examining neonatal outcomes unless they occurred after 24 weeks. Logistic regression using robust SEs to allow for non-independence of neonates from multiple births was used when comparing neonatal outcomes.

Using the most recent national birth data,12,186,187 we anticipated identifying 406 women aged ≥ 48 years at their date of delivery and 812 comparison women. With these numbers of women the study would have had an estimated power of 80% at the 5% level of statistical significance to detect ORs of ≥ 1.5 and ≥ 2.0, assuming outcomes have an incidence of 40% and 5%, respectively. The actual number of older and comparison women identified during the study gave an estimated power of 80% at the 5% level of significance to detect ORs of ≥ 1.6 and ≥ 2.5, assuming the same outcome incidence levels.

Maternal socioeconomic status

A substantial proportion of the sample did not have information on socioeconomic status (17% for cases and 13% for controls); thus, bias owing to missing data could not be ruled out. The results of multivariable regression analysis showed that, compared with controls, the cases were 1.51 (95% CI 1.18 to 1.94) times more likely to have missing socioeconomic information. Although not statistically significant, the odds of belonging to the routine/manual group was 1.17 (95% CI 0.94 to 1.45) times higher and belonging to an unemployed group was 1.22 (95% CI 0.92 to 1.61) times higher among women who suffered from severe maternal morbidity than among the controls (Table 25).

Maternal ethnic group

After accounting for other factors and possible confounders sequentially, the fully adjusted model 4 showed that compared with white European women, the odds of severe maternal morbidity were 83% higher among black African women, 80% higher among black Caribbean women, 74% higher in Bangladeshi women, and 56% and 43% higher in the other non-white (non-Asian) and Pakistani groups, respectively (Table 26). There was remarkably little change in ORs with adjustment for other factors across the models suggesting that other sociodemographic and clinical factors had little or no confounding effects. Women with inadequate utilisation of antenatal care services were twice as likely to be at risk of severe morbidity as women who adequately utilised the services (aOR 1.97, 95% CI 0.96 to 4.04) and the odds of inadequate utilisation of antenatal was higher among black African (OR 4.46, 95% CI 1.47 to 11.48) and Caribbean (OR 4.35, 95% CI 0.49 to 18.19) women than among white European women. In this analysis, which included additional potential confounders, notably inadequate utilisation of antenatal care and had greater statistical power, there was no evidence of any association between maternal socioeconomic status and severe morbidity.

Older maternal age

A total of 233 women of very advanced maternal age were notified to UKOSS and 454 comparison women. The median age of the older women was 49 years (range 48–61 years) while the median age of the comparison women was 31 years (range 16–46 years). Older women were significantly more likely than comparison women to be overweight or obese, to not smoke in pregnancy, have had previous uterine surgery not including previous caesarean section, have previous or pre-existing medical condition(s), be nulliparous, have a multiple pregnancy and have conceived following assisted conception (Table 27).

Table 28 shows the pregnancy complications experienced by the older and comparison women. Unadjusted analysis suggests that older women were more likely than comparison women to have a range of complications including gestational hypertensive disorders, gestational diabetes, PPH, caesarean delivery, iatrogenic and spontaneous pre-term delivery and ITU admission. However, with the exception of gestational diabetes, caesarean delivery and ITU admission, these effects were attenuated and became non-significant after adjustment for confounding or mediating factors. The higher rate of multiple pregnancy and use of assisted conception in the older women explained most of this attenuation (see Table 28). There was evidence of significant interaction between caesarean delivery and parity: the raised odds of having a caesarean delivery were only apparent in nulliparous older women (aOR 9.90, 95% CI 3.64 to 26.92 in nulliparous women; aOR 0.71, 95% CI 0.31 to 1.66 in parous women). No other significant interactions were found. Among the older women who had a caesarean delivery, maternal age was the primary indication for 21% (36/175).

Among the older women, a total of 268 fetuses survived beyond 24 weeks’ gestation (35 sets of twins, three sets of triplets and 189 singletons). Of these 268 fetuses known to be alive in utero beyond 24 weeks, three were stillborn antepartum; two of these were a set of twins, of which both were stillborn, and the other occurred in a women who initially had a triplet pregnancy but had already spontaneously lost one of the triplets in the second trimester before 24 weeks. A further two of the fetuses died shortly after birth following very pre-term delivery (< 28 weeks’ gestation), equating to a perinatal mortality rate of 18.7 per 1000 (95% CI 6.1 to 42.9 per 1000). Although this was more than double the national rate of 7.5 per 1000,95 the difference was not statistically significant (RR 2.5, 95% CI 1.0 to 5.9), noting the limited statistical power of this comparison.

The proportion of fetuses surviving beyond 24 weeks that had a congenital anomaly was similar between the older women and the comparison women (1.9%, 5/263 vs. 1.5%, 7/460; p = 0.702), as was the proportion that had other major complications such as respiratory distress syndrome and severe infection (2%, 4/205 vs. 3.8%, 13/344; p = 0.240). The proportion of fetuses that had a low birthweight (< 2500 g) was higher among those born to older women than comparison women (32%, 85/267 vs. 8%, 38/463; p < 0.001), although this difference disappeared after controlling for gestational age at delivery.

Research questions

• What types of incidents trigger local reviews in the UK?

• What is the quality of local guidance on conducting reviews of severe maternity incidents in the NHS?

• What local approaches are being used to review near-miss maternal morbidity cases and how do these compare with an external case review approach (confidential enquiry)?

Conditions eliciting local reviews

All 211 consultant-led maternity units in the UK were contacted up to three times, via letter, e-mail and telephone, and asked to supply a copy of the list of incidents that triggered a local review as well as a copy of their maternity risk management strategy and/or their incident review procedure during 2012. Definitions for each type of incident were documented a priori. Conditions included sepsis, massive obstetric haemorrhage, abruption, cardiac arrest and severe blood loss. Varying thresholds were accepted for several categories, including ‘severe blood loss’, ‘low Apgar score’, ‘low cord pH’, ‘prolonged inpatient stay’ and ‘prolonged second/third stage’. The conditions included were classified by two researchers individually (AS and Olaa Mohamed-Ahmed) and disagreement resolved by discussion with a third reviewer if required (MK). Incidents that were listed by at least 5% of maternity units were reported under the following categories: maternal, neonatal or organisational incidents. The lists of incidents were compared with incidents recommended for review by the Royal College of Obstetricians and Gynaecologists (RCOG). 191

Protocols for local reviews

The Appraisal of Guidelines for Research and Evaluation (AGREE) II instrument was used to assess the quality of the incident review guidelines obtained from the mailing, using 23 items organised into six domains. 192 The domains were ‘scope and purpose’, ‘stakeholder involvement’, ‘rigour of development’, ‘clarity of presentation’, ‘applicability’ and ‘editorial independence’.

Each item was scored on a 7-point Likert scale ranging from 7 ‘strongly agree’ to 1 ‘strongly disagree’ with a mid-point of 4. An overall assessment of the quality of the guideline was made by calculating the mean average score from the 23 scores for each AGREE criteria. Any score of 0–2.9 was classified as ‘poor quality’, a score of 3–3.9 was ‘average quality’, a score of 4–4.9 was ‘good quality’ and a score of ≥ 5 was considered to be ‘high quality’.

Two researchers (AS and Olaa Mohamed-Ahmed) met to discuss and agree how best to apply each question within the AGREE II criteria to incident reporting within maternity services, prior to evaluating any guidelines. Each guideline was assessed independently by each researcher and if the score for any item differed by > 2 points, they met to discuss and resolve discrepancies in interpretation. Scores were revised accordingly, with a third researcher qualified in midwifery providing clarification on any outstanding issues. Analyses were based on the revised scores. Four additional issues were considered when appraising each guideline: the method(s) used for reviewing an incident, the types of professionals involved in the review, the methods for disseminating the outcomes and whether or not units had a process to audit changes in clinical practice arising from incidents.

Standardised domain quality scores for each guideline were calculated according to the AGREE II instrument standard methods. The possible range for standardised domain scores is 0–100%. Median domain scores with 95% CI and the proportion of guidelines scoring < 30%, 30–60% and > 60% were calculated. Scores for the lowest and highest scoring domains were compared with scores for other domains using the Wilcoxon matched pairs signed rank-sum test for non-parametric data.

External versus local reviews

Six sites were randomly selected following stratification from all NHS consultant-led maternity units in England [two large teaching hospital units (5000+ deliveries per year), two medium units (2–4999 deliveries per year) and two small units (1–1999 deliveries per year)] and on the basis of their Clinical Negligence Scheme for Trusts (CNST) level at the time of selection (one of each size with CNST level 1, the basic risk management level, and one of each size with CNST level 3, the highest risk management level).

Each unit provided a list of all cases of direct (obstetric) near-miss maternal morbidity that had triggered a local review assessment process during the previous 6 months. From these lists, six cases were randomly selected from each unit and full sets of clinical notes, together with their local review report, were requested. In total, 33 sets of anonymised clinical records were obtained and reviewed and the remaining three sets of notes were unavailable. The methodology used for and lessons identified from local reviews were described.

Eleven external reviewers, comprising five obstetricians, four midwives and two anaesthetists, assessed each set of anonymised medical records, using the confidential enquiry approach used by the MBRRACE-UK Confidential Enquiry into Maternal Deaths. 2,147 Two obstetricians, a midwife and two anaesthetists assessed the care of each woman. Five reviewers thus examined each woman’s care. Each primary assessor completed an independent review of the woman’s care, highlighting the lessons to be learned to improve care in the future. This was checked by a second assessor in the relevant specialty. External assessors were located throughout England and, to maintain anonymity, assessors were only asked to review the care of women who had been cared for outside their region. The assessment process and all individual findings were strictly confidential; all assessors were required to sign a confidentiality agreement.

Qualitative and quantitative approaches were used to compare the reviews. The reviews were read and reread, a coding frame was constructed and the data coded. Anticipated and emergent themes were then examined. A qualitative interpretive approach was taken, combining thematic analysis with constant comparison. 44,45 NVivo 9 was used to facilitate the analysis. Once the themes had been identified, the number of reviews identifying each theme was quantified and compared using a chi-squared or Fisher’s exact test as appropriate.

Conditions eliciting local reviews

Of the 211 consultant-led maternity units in the UK, 71% (n = 150) provided a list of incidents that would trigger a local review. The conditions listed were highly variable, although those recommended by the RCOG were most frequently represented. No single incident or condition was recommended for review by every maternity unit.

The majority of units (> 90%) that responded included maternal and neonatal deaths, stillbirths, intensive care admissions, severe blood loss, shoulder dystocia, third- or fourth-degree tears, eclampsia, low Apgar scores and unplanned home birth (Table 29). The most commonly listed maternal conditions that had not been recommended for review by the RCOG included cord accidents (77%), cardiac arrest (69%) and sepsis (64%). Meconium aspiration (14%) and hypothermia on admission (9%) were often listed in addition to the other neonatal conditions recommended for review by the RCOG. A variety of organisational incidents featured on the trigger checklists including inadequate staffing levels (70%) and delay in access to theatre (61%).

Protocols for local reviews

Of all 211 consultant-led maternity units in the UK, 70% (n = 148) provided an incident review protocol or risk management strategy. Trusts or health boards that had more than one unit (n = 22) all indicated that the same guideline was in use across all units. Thus, 120 guidelines applicable to 148 maternity units were included in the appraisal. Many of the guidelines received were maternity specific (n = 82, 68%) and the remainder applied to all incidents, irrespective of speciality, that occurred in a trust or board.

Scores for the ‘scope and purpose’ domain were significantly higher than scores for the other domains (p < 0.001). The median score for this domain was 86% and 112 of the 120 guidelines (93%) scored > 60%. Only 49 guidelines scored > 60% for the ‘stakeholder involvement’ domain, which includes being authored by a range of health professionals or consulting with staff. Scores for the ‘rigour of development’ domain were significantly less than for any other domain (p < 0.001). For the ‘clarity of presentation’ domain, 93% of guidelines scored > 60%. The median score for the ‘applicability’ domain was 56%, but only six guidelines scored < 30%. None of the guidelines referred to the independence of the guidelines from the funding body or any competing interests of the guideline development group members (see Methods, Conditions eliciting local reviews).

Root cause analysis was most commonly suggested as the methodology for reviewing serious maternity incidents (n = 97, 81%), with case review analysis at regular meetings for less-serious incidents. The guidelines for six units (4%) did not state the approach to be used for reviewing incidents. Other approaches to review incidents that were mentioned were significant event analysis (five units), trend analysis (four units) and one unit recommended choosing an approach from systems analysis with a contributory factor framework, brainstorming, the five whys, incident decision tree, fishbone diagram and gap analysis.

Most review guidelines required one or two designated people to review reports of all incidents, such as a risk midwife or manager. Other guidelines designated a team of people to review each incident including senior risk managers, lead clinicians and specialist midwives in risk (see Methods, Conditions eliciting local reviews). Incidents could be escalated to a supervisor of midwives, maternity risk management boards, adverse events committees, patient safety co-ordinators, relevant executive directors or arrangements made to have a committee with external professionals and an independent chairperson, as required.

Lessons learned from incident reviews were most commonly disseminated at meetings, in reports, newsletters and posters (n = 110, 92%), with individual feedback and support given to those involved in the incident. Other means of communication included e-mail, intranet, memos, via line managers, local and multidisciplinary forums, in the minutes of meetings and ‘message of the week’. Only 10 guidelines (8%) did not mention any means of communicating the findings of reviews with staff.

Many guidelines (n = 83, 69%) included details of a process to audit the impact of recommendations arising from incidents. A further 28 (23%) guidelines included mention of having an auditing process or a committee that would consider governance and assurance issues. It was not clear if the audit cycle around incidents would be completed. Only nine units that responded made no mention of audits.

External versus local reviews

The care of 33 women with near-miss maternal morbidities was assessed by the confidential enquiry panel. The majority of women (73%) had severe haemorrhage (> 1500 ml of blood loss).

Local reviews of the care of 28 of the women (85%) had been conducted and for the remainder a local review had not been conducted either because the incident had been assessed locally as not needing a review, or because cases were not selected for review owing to capacity issues. A formal report of the local review process and outcome had been written in only 11 cases (33%) and four of these local reviews used root cause analysis. For 12 cases (36%), a review group had discussed the incident and a summary of the outcome had been noted in a spreadsheet. For an additional five cases the documented local review comprised a timeline of events with brief notes on evidence of good care and if issues had occurred.

The categories of staff involved in conducting reviews varied between units. An individual midwife conducted four reviews, a group of midwives conducted two reviews, and five reviews (15%) involved both obstetricians and midwives. One of these five reviews included an anaesthetist and one included an obstetric trainee. It was not possible to identify the specialty or grade of reviewers in 17 local review reports. In five local reviews (15%) it was documented that the woman concerned had asked the review group to consider specific questions about her care and that she was later notified of the group’s conclusions and action plans.

In the local review reports, action plans had been written for 14 cases (42%). The conclusions from three local review reports (9%) included a recommendation to audit the subsequent change to clinical practice.