Severe maternal morbidity surveillance: Monitoring pregnant women at high risk for prolonged hospitalisation and death

Abstract Background There is no international consensus on the definition and components of severe maternal morbidity (SMM). Objectives To propose a comprehensive definition of SMM, to create an empirically justified list of SMM types and subtypes, and to use this to examine SMM in Canada. Methods Severe maternal morbidity was defined as a set of heterogeneous maternal conditions known to be associated with severe illness and with prolonged hospitalisation or high case fatality. Candidate SMM types/subtypes were evaluated using information on all hospital deliveries in Canada (excluding Quebec), 2006‐2015. SMM rates for 2012‐2016 were quantified as a composite and as SMM types/subtypes. Rate ratios and population attributable fractions (PAF) associated with overall and specific SMM types/subtypes were estimated in relation to length of hospital stay (LOS > 7 days) and case fatality. Results There were 22 799 cases of SMM subtypes (among 1 418 545 deliveries) that were associated with a prolonged LOS or high case fatality. Between 2012 and 2016, the composite SMM rate was 16.1 (95% confidence interval [CI] 15.9, 16.3) per 1000 deliveries. Severe pre‐eclampsia and HELLP syndrome (514.6 per 100 000 deliveries), and severe postpartum haemorrhage (433.2 per 100 000 deliveries) were the most common SMM types, while case fatality rates among SMM subtypes were highest among women who had cardiac arrest and resuscitation (241.1 per 1000), hepatic failure (147.1 per 1000), dialysis (67.6 per 1000), and cerebrovascular accident/stroke (51.0 per 1000). The PAF for prolonged hospital stay related to SMM was 17.8% (95% CI 17.3, 18.3), while the PAF for maternal death associated with SMM was 88.0% (95% CI 74.6, 94.4). Conclusions The proposed definition of SMM and associated list of SMM subtypes could be used for standardised SMM surveillance, with rate ratios and PAFs associated with specific SMM types/subtypes serving to inform clinical practice and public health policy.


| INTRODUC TI ON
Substantial changes in maternity care were introduced in industrialised countries in the middle decades of the 20th century, including improvements in the organisation of obstetric services and widespread use of antibacterial agents, ergometrine, and blood transfusion. [1][2][3] The substantial reduction in maternal mortality that followed led to a perception that pregnancy and childbirth had become safe and mostly risk-free. However, maternal mortality remains a concern among vulnerable subpopulations even in high-income countries, and severe maternal morbidity (SMM) is recognised to affect an important fraction of women. [3][4][5][6][7] Although maternal death represents a more extreme and serious outcome than severe maternal illness, its rarity resulted in SMM becoming an important focus for public health surveillance and epidemiologic investigation in high-income countries. 3,[6][7][8] Recent changes in maternal characteristics in such countries, including increases in age and pre-pregnancy weight, have raised new concerns about temporal trends in SMM. 9,10 Reviews of SMM cases show that, as with maternal death, the most common preventable factor is suboptimal care, including failures in diagnosis and delays in treatment. [11][12][13][14] The World Health Organization has recommended that maternal health surveillance focus not only on maternal mortality but also on severe acute maternal morbidity, in order to identify priorities for intervention. 15 The World Health Organization defines severe maternal complications as 'potentially life-threatening conditions', maternal near-miss as 'a woman who nearly died but survived a complication that occurred during pregnancy, childbirth or within 42 days of termination of pregnancy', and severe maternal outcomes as maternal near-miss cases and maternal deaths (per WHO terminology, severe acute maternal morbidity is synonymous with maternal near-miss). 15,16 The WHO definition of near-miss (or severe acute maternal morbidity) notwithstanding, there is little international consensus on the components of SMM, and studies on SMM typically include variable lists of maternal diseases, interventions, and organ failure types (without clear specification of how these meet a prespecified definition). [4][5][6][7] Large differences in SMM conditions included in different studies lead to incomparable SMM frequencies and an inability to benchmark population rates of SMM. The objectives of this study were to propose a comprehensive definition of SMM, to use the definition to create an empirically justified list of component SMM types and subtypes, and to use this list to examine SMM rates in Canada. The proposed definition of SMM, and its components, could serve to spur efforts towards creating an international consensus for SMM surveillance and for benchmarking maternal health outcomes in populations.

| Historical note
The Canadian Perinatal Surveillance System (CPSS) began monitoring SMM approximately 15 years ago, 17 and the Canadian Classification of Health Interventions (CCI). 19 However, even that list was flawed in some respects; one shortcoming was the exclusion of severe pre-eclampsia and HELLP syndrome cases due to coding limitations in the early version of ICD-10 CA. This problem was identified by the CPSS in 2010 19 and subsequently rectified by the Canadian Institute for Health Information; cases of severe pre-eclampsia and HELLP syndrome can now be identified in Canadian hospitalisation data from 2012 onwards. Inclusion of conditions that did not necessarily represent

| Definition of SMM
Severe maternal morbidity was defined as a set of heterogeneous maternal conditions known to be associated with severe illness and with prolonged hospitalisation or high case fatality. The definition was operationalised using an eclectic approach based on diagnostic, interventions and organ failure codes (see below).

| List of SMM types and subtypes
The components of SMM were chosen through consensus by a multi-disciplinary group of experts, who evaluated each candidate component in terms of feasibility of surveillance and validity (per the above-mentioned definition of SMM). All SMM types and subtypes included in the 2010 list and others proposed for inclusion by the multi-disciplinary group were evaluated in terms of frequencies, temporal trends, case fatality rates (ie death during the delivery admission), and (prolonged) length of hospital stay using hospitalisation data from Canada for the years 2006-2015. The consensus assessments involved several meetings of the multidisciplinary group during which empirical data on each potential SMM type and subtype were reviewed, including the relevant code(s) in ICD-10CA and CCI.

| Data source
Information on these hospital deliveries (2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015) was obtained from the Discharge Abstract Database of the Canadian Institute for Health Information, which contained records for approximately 98% of all deliveries in Canada (excluding Quebec). The database included information routinely abstracted from medical charts by trained personnel using standardised definitions and processes. 20 Details regarding maternal and infant characteristics, labour and delivery events, and diagnoses and procedures were documented, with diagnoses coded using ICD-10CA, and procedures coded using CCI. The validity of the information in the Discharge Abstract Database maternal and newborn records has been routinely assessed and shown to accurately reflect information contained in medical charts. 21,22

| Components of SMM: SMM types and SMM subtypes
With SMM in any population defined as the frequency (incidence for new conditions, prevalence for pre-existing ones) of heterogeneous maternal conditions known to be associated with severe illness, prolonged length of hospital stay, or high case fatality, we used diagnostic, intervention, and organ failure codes to identify eligible maternal diseases (eg eclampsia), interventions (eg hysterectomy), and conditions that signified organ failure (eg acute renal failure).
Length of hospital stay was assessed using mean and median duration of hospital stay and the proportion of women with a prolonged length of hospital stay (≥7 days). Each candidate condition considered as signifying a potential SMM was evaluated by examining rates, temporal trends, length of stay, and case fatality rates for the

| Statistical analysis
The frequencies of composite SMM, SMM types, and SMM subtypes were expressed using rates and 95% confidence intervals (CI). Case fatality rates and proportions of women with a prolonged hospital stay (≥7 days) were calculated similarly. Rates of composite SMM were estimated within categories of maternal age, parity, plurality, mode of delivery, and other factors, and contrasts between categories of a determinant were quantified using rate ratios and 95% confidence intervals. The population attributable fractions (PAF 23 ) for prolonged hospital stay and maternal death associated with composite and specific SMM types/subtypes (or, in other words, the fraction of prolonged hospital stays and deaths that could be prevented by eliminating composite SMM or a specific SMM type/subtype) were estimated using the formula.
where PAF denotes the population attributable fraction, p denotes the proportion of cases with prolonged hospitalisation/death due to composite SMM or a specific SMM type/subtype, and RR denotes the rate ratio contrasting the rate of prolonged hospitalisation or death among women with composite SMM or with a specific SMM type/subtype vs women without SMM or without that specific SMM type/subtype. Analysis was carried out using SAS version 9.1 (SAS Institute).

| Ethics considerations
Privacy considerations required the suppression of cells with small values (1)(2)(3)(4); in such cases, rates were provided as ranges calculated using 1 and 4 as the numerator. Since the study was based on   Table S1. Table 3 shows the frequency, case fatality rate, and length of  Table 4. SMM rates were also significantly higher among nulliparous women,women with increasing parity, multi-fetal pregnancy, or previous caesarean delivery; and among women with labour induction or caesarean delivery. Women who received epidural anaesthesia had lower rates of composite SMM than those who did not (12.5 vs 19.0 per 1000 deliveries).

| RE SULTS
The frequency, case fatality rates, and length of stay for broad types of SMM are shown in Table 3 (SMM types are not mutually exclusive). The most common types of SMM included severe preeclampsia, eclampsia, and HELLP syndrome,severe haemorrhage; surgical complications; maternal intensive care unit admission; and hysterectomy. Table 5  Case fatality rates were highest among women with cardiac arrest and resuscitation (241.1 per 1000), hepatic failure (147.1 per 1000), and dialysis (67.6 per 1000), and among those with cerebrovascular accidents (51.0 per 1000). Women with several different SMM subtypes had an extended hospital stay, with ≥40% having a hospital stay ≥7 days among those with placenta praevia requiring blood transfusion, pulmonary oedema and heart failure, disseminated intravascular coagulation, acute renal failure/dialysis, evacuation of incisional haematoma requiring transfusion, acute fatty liver requiring transfusion, assisted ventilation, or sickle cell anaemia with crisis Table 5.
Rate ratios and PAFs for each SMM type are presented in Table 6.
The rate ratio for a prolonged hospital stay among women with any SMM (vs those without) was 14.5, while the rate ratio for prolonged hospital stay among women admitted to an ICU (vs those not admitted to an ICU) was 24.4. The rate ratio for maternal death among women with any severe maternal morbidity (vs those without) was 459.1, and among those admitted to ICU (vs those not admitted to ICU), the rate ratio for death was 461.7. The PAF for maternal death associated with any SMM was 88.0% (95% CI 74. 6

| Principal findings
We used a priori knowledge and empirical support from frequencies, temporal trends, case fatality rates, and length of hospital stay to de- procedures. Case fatality rates were highest among women with cardiac arrest and resuscitation, hepatic failure, those receiving dialysis, those with cerebrovascular accidents, and those with cardiac conditions. SMM was associated with a PAF of 47% to 18% for prolonged hospitalisation and a PAF of 88% for maternal death.

| Strengths of the study
The strengths of our study and the proposed SMM surveillance framework include reliance on multi-disciplinary input and evidentiary support from contemporary data on deliveries. The hospitalisation data source (viz., the Discharge Abstract Database of the Canadian Institute for Health Information) has made changes in ICD-10CA coding (eg introduction of a code for severe pre-eclampsia in 2012) and other important aspects of data collection (linked mother and infant records, extraction of gestational age at delivery, etc), which have facilitated comprehensive monitoring of SMM.

| Limitations of the data
Limitations of our study include an inability to identify some clinically relevant cases based on ICD-10CA codes (eg extreme obesity) and to distinguish between some pre-existing and acute complications arising in pregnancy. We were unable to identify women who received more than one transfusion (a measure of more severe morbidity than TA B L E 1 Frequency, length of hospitalisation, and case fatality rates associated with overall severe maternal morbidity (SMM), and for selected SMM types and subtypes evaluated for the new 2018 list of SMM, Canada (

| Interpretation
Conditions included as components of severe maternal morbidity vary considerably in the literature. 3 Notes on selected diagnostic and procedure codes:  (Table 3).
Published studies of population rates of SMM fall into three primary types, which differ based on the conceptual framework used Abbreviations: CI, confidence interval; ICU, intensive care unit; DIC, disseminated intravascular coagulation; ARF, acute renal failure. *If the numerator of the rate was >0 and <5, a range was provided (assuming a numerator of 1 and 4) instead of actual value (95% confidence interval not provided).
disease-based, intervention-based, and organ system failure-based criteria) to identify cases of SMM in routine hospitalisation data that include diagnosis, intervention codes, and other information.
The list of SMM types and subtypes used in our study was similar, although perhaps more comprehensive than the maternal mor-  TA B L E 6 Frequencies of specific severe maternal morbidity (SMM) types, and associated rate ratios and population attributable fractions (PAF) for case fatality and prolonged length of hospital stay (LOS), Canada (excluding Quebec), 2012-2016 but this was not a repudiation of the definition. Also, women with such severe illnesses and a LOS < 7 days were included as cases of SMM as it is possible to be severely ill, receive life-saving intervention, and recover fairly quickly.
We faced some challenges, however, in translating our SMM definition into an operational list of severely morbid conditions. One limitation arose from our reliance on ICD-10CA codes, which do not capture all conditions of interest with sufficient accuracy. Extreme obesity, a maternal condition which could potentially satisfy our definition of SMM, is an example: ICD-10CA includes a code for obesity but not extreme obesity. Additionally, obesity, which did not satisfy the SMM definition in terms of case fatality rates and prolonged length of stay, was captured in only a small fraction of women: less than 2% vs an expected frequency of over 10%. 10,33 We omitted conditions such as malignancy in pregnancy, which were associated with prolonged length of stay and high case fatality, from the SMM list because the associated burden of illness appeared to be mostly unrelated to pregnancy, and we were unable to identify cases in which the course of the malignancy was aggravated by pregnancy. On the other hand, we included conditions such as maternal ICU admission, which did not identify any additional deaths but did carry a high risk of prolonged length of stay. 34 Finally, we encountered a few conditions (eg diabetes mellitus, asymptomatic HIV infection) where the prolonged length of hospital stay was likely associated with management or socio-economic issues rather than severe morbidity per se.
The rate ratios and PAFs associated with specific types of SMM in our study can be used to inform clinical practice and public health policy. From a public health standpoint, PAFs for maternal death show that cardiac conditions, haemorrhage, obstetric embolism, obstetric shock, DIC, and surgical complications are the priorities to be addressed in terms of SMM prevention. A substantial reduction in maternal mortality would likely result from a reduction in these SMM. Adverse temporal trends or geographic differences identified in our study also provide an impetus for action, whether nationally or at the provincial level. Audit of maternal deaths and SMM cases is a worthwhile undertaking that is being increasingly discussed in clinical circles, and such activities could help focus attention on prevention of SMM and maternal death through improved care.

| Conclusions
We combined a priori clinical knowledge, prolonged length of stay, high case fatality, and expert consensus to identify a set of severely