Background
The maternal mortality ratio (MMR) is a robust indicator of health care quality, inequality and inequity in and among countries [
1]. Most maternal deaths are preventable in low, middle and high resource settings, as was the case for 47% of maternal deaths in Suriname between 2010 and 2014 [
2,
3]. To develop prevention strategies, accurate data on the number of maternal deaths and insight into underlying causes are essential [
2,
4,
5]. However, the assignment of a reliable underlying cause of death and the subsequent classification can be a challenge [
6].
The World Health Organization (WHO) aimed to create uniform maternal death classification guidelines to enhance usability, improve comparability and decrease coding errors [
7‐
9]. Therefore, the WHO launched the International Classification of Diseases-Maternal Mortality (ICD-MM) in 2012, an application of International Classification of Diseases-10
th edition (ICD-10) to classify deaths during pregnancy, childbirth and the puerperium [
7].
Difficulties in attributing the underlying causes can result in inconsistencies in classification in and among countries, despite using the ICD-MM guidelines [
8,
10]. When a European expert panel reviewed pregnancy-related deaths across 13 European countries, they identified 14% more maternal deaths than what the national registries of the individual countries included [
11]. Classification is especially complicated when comorbidities occur, and the start of the chain of events resulting in maternal death has to be determined [
10]. Consequently, underlying cause attribution may vary, or the causes are unknown or unclear, resulting in underreporting. This is not only an issue in low- and middle-income countries but also in high-income countries and was reported by various Maternal Death Review (MDR) committees, including those from Suriname, Jamaica and the Netherlands [
3,
12,
13].
Therefore, this study aimed to assess the applicability of the ICD-MM guidelines by investigating the classification of maternal deaths in one country and across three countries. First, the cause of death as determined by the attending physician was compared to the assessment of the Surinamese MDR committee. Second, cases were shared with the national MDR committees from Jamaica and the Netherlands, and their assessments were compared to the findings of the Surinamese MDR committee. Following these findings, the classification difficulties are discussed, and recommendations for improving the ICD-MM guidelines’ applicability and international comparability of maternal mortality are provided.
Discussion
This study explored consistency in classifying pregnancy-related deaths in Suriname at two levels. First, underlying cause attribution by the attending physicians, and the Surinamese MDR committee was compared; conclusions differed in 47% of cases. Second, the classification of three national MDR committees of Suriname, Jamaica and the Netherlands were compared applying the WHO ICD-MM guidelines to the same cases. There was 15% disagreement among these committees on whether selected pregnancy-related deaths met the criteria to be defined as maternal deaths. They achieved moderate agreement (k = 0.53) on classifying cases as direct, indirect or unspecified, with greater consensus between the Surinamese and Jamaican MDR committees (k = 0.69) than the Surinamese and Netherlands MDR committees (k = 0.48). The MDR committee of the Netherlands, a high-income country, classified more deaths as unspecified than those from the middle-income countries of Suriname and Jamaica. There was higher concurrence among the three MDR committees in underlying cause attribution to abortive outcomes, obstetric hemorrhage and indirect maternal deaths, but only fair agreement on a mix of cases (other direct obstetric causes and unspecified).
The large difference (47%) in underlying cause attribution for maternal death between the attending physicians and the Surinamese MDR committee is not unusual. Similar differences were also seen in Malawi, where poor agreement between healthcare providers and the research team on maternal death classification was reported [
20]. Another study found a 40% difference in underlying cause attribution in a multi-country survey that compared health provider findings with external reviewers among Low- and Middle-Income Countries (LMIC) [
6]. The abovementioned examples illustrate the importance of multidisciplinary case discussion and consensus-based underlying cause attribution.
Besides inconsistent underlying cause attribution, poor coding of pregnancy-related deaths, misidentification, or misclassification can result in inadequate certification and is associated with underreporting [
2,
12,
21]. Due to underreporting, vital statistics could miss at least 50% of the maternal deaths [
22]. Hence, since maternal death certificates are also completed by non-obstetricians (e.g. in the rural interior or when indirect maternal deaths occurred), all clinicians would benefit from training to correctly complete death certificates.
The MDR committees in our study encountered specific challenges for which no clear guidance was available from the ICD-MM guidelines. These included (1) determining the fact of pregnancy with limited evidence; (2) inclusion of deaths from suicide, especially in early pregnancy and (3) whether and how to count maternal deaths outside the country of residence. It is unclear what the minimally acceptable evidence of pregnancy should be without medical confirmation and under which circumstances information from verbal autopsy alone could be used to confirm pregnancy. While the ICD-MM classifies suicide during pregnancy and puerperium as a direct maternal death, this is clearer for puerperal psychosis and postpartum depression than for events early in pregnancy [
7]. The trigger for suicide may be social/circumstantial (partner rejection, domestic violence, unintended pregnancy), rather than clinical (pre-existing mental disorder or hormonal changes impacting mental health) [
23,
24]. In addition, the ICD-MM guidelines do not elaborate on how to classify maternal deaths from suicide (direct vs indirect) when underlying mental disorders existed [
23]. Finally, opinions differed in this study on the inclusion of a resident who had been under local care but died in another country. As no global guidance exists on whether to count such events in the country where the women dies or the country of residence, there is a chance that these cases are not reported at all (excluded in the country where she died and not reported in the country where she lived). Since all births are included in the national birth registry (denominator), we suggest including the mother also in the country where she died (numerator). Importantly, in these situations information is ideally exchanged between countries to facilitate local reporting and sharing of “lessons to be learned”.
Consensus between the MDR committees of Suriname and Jamaica was higher than between those of Suriname and the Netherlands. The cases the Dutch committee considered unspecified but were assigned other diagnoses by the other committees had limited information on the disease course, and lacked confirmatory diagnostic tests such as laboratory results, ultrasounds, Computed Tomography (CT) or Magnetic Resonance Imaging (MRI) scans compared to the cases with more agreement. Advanced diagnostics were often unavailable due to financial or logistic constraints, such as the minimal laboratory capacity in the rural interior areas. In these cases, the MDR committees in LIMC must often rely on clinical judgement to make a diagnosis. Practicing medicine with greater uncertainty regarding diagnosis and treatment outcomes and fewer possibilities to provide evidence-based care is more commonplace in LMIC and possibly explains the more consistent results between the MDR committees of the two middle-income countries.
Classification into type of maternal death (direct, indirect and unspecified) differed in 42% of cases, only achieving moderate agreement among the three MDR committees. Dividing maternal deaths into direct and indirect conditions is pragmatic as preventive programs to avert direct deaths differ from indirect deaths [
25]. However, this division has been questioned by the MDR committees in the United Kingdom (UK) and the Netherlands, especially for women with concurrent direct and indirect comorbidities [
26]. In both middle and high-income countries, several pre-existing conditions such as obesity, diabetes mellitus, and hypertensive diseases are increasing and the risk of pregnant women to develop direct and indirect complications of pregnancy (e.g. postpartum hemorrhage, eclampsia, cardiovascular diseases) [
26‐
28]. This coexistence of multiple conditions in an individual is known as multimorbidity and is one of the challenges of modern medicine [
29,
30]. These conditions obfuscate the strict demarcation between direct and indirect deaths and reduce their relevance. Instead, adding multimorbidity categories, such as (non)communicable diseases and (pre-existing) mental disorders to the ICD-MM guidelines would be more pertinent.
We conducted a sensitivity analysis to explore whether consensus improved with the exclusion of (1) cases without consensus among the MDR committees in the classification as maternal deaths, and (2) cases with incomplete information. As expected, the exclusion of the cases with uncertainty improved the level of agreement. These exclusions strengthened the consensus that already existed between the Surinamese and Jamaican MDR committees. However, since differences are small, these analyses suggest that, even with limited information, MDR committees can reach reliable conclusions on the probable types and underlying causes of maternal deaths.
Our data showed that when the cycle of events leading to death had fewer incidents (Fig.
3), underlying cause attribution was more straightforward (as with abortion-related and obstetric hemorrhage). Selecting the initiating event from a chain of multiple events is more difficult in complex cases, resulting in a discrepancy in underlying cause classification in our study. Two high-income countries, the United Kingdom (UK) and the Netherlands, also reported such differences in underlying cause attribution [
10]. Their MDR committees discussed selected cases where disagreement was expected during a meeting attended by most members of both committees
. While the Netherlands classified a death by the primary underlying pathology, the UK more pragmatically focused on the acute fatal complication [
10]. They suggested that decision-making may be guided by what best informs local practice in the absence of global guidance. However, this approach could result in heterogeneity and complicates comparison among countries.
Reliable underlying cause attribution may be improved by combining clinical data with autopsy findings [
31,
32]. However, autopsy for maternal death is seldom performed in low resource countries such as Suriname, where only two cases were investigated [
3]. It may be useful to revisit verbal autopsy techniques to improve collection and interpretation of information on signs, symptoms and risk factors [
33]. Another possible option is the minimally invasive autopsy. This includes collection of blood, cerebrospinal fluid and tissue samples for histologic and microbiologic analysis [
34]. This option could be explored to assist in identifying the underlying causes of maternal death.
Strengths and limitations
This study’s strength is its unique comparison of the classification of the same cases by physicians and (inter)national MDR committees from three different settings applying the WHO ICD-MM guidelines. Limitations include difficulties in interpreting cases with limited information and, possibly, by a high-income country being unfamiliar with the different contexts of LMIC. The inter-rater reliability should be carefully interpreted as the overall kappa may not be reliable for rare observations, such as group 1 (abortive outcomes) and group 4 (pregnancy-related infections).
Acknowledgements
We thank the Surinamese MDR committee members who reviewed the 2010-2014 cases (the gynecologists/obstetricians : H. Kanhai, G. Essed, P. Goerdin, R. Charles, H. Karansingh, R. Tjon A Fat, L. Olmtak, M. Sietaram, O. Ramkhelawan, K. Ramkhelawan, S. Mohan, M. Dipoikromo, gynecology residents then: F. Rigters and S. Cornelisse, internal medicine specialists: K. Waldring, S. Vreden, A. Niekoop, J. Adhin, P. Issa and A. Punwasi, anaesthesiologists: M. Tjon Sie Fat, D. Nahar, neurologist: S. de Jong and midwives: M. Fitz Jim, S. Abente, S. Holband, A. Naarden, J. Kloppenburg). We are very grateful that Raez Paidin (then medical student) translated the cases in English. We also thank the Jamaican MDR committee members who volunteered to classify these cases (the obstetrician D. McDonald and L. Campbell). We are thankful to the Dutch MDR committee members who voluntarily reviewed the cases (the gynaecologists/obstetricians: J. Schutte, J. van Roosmalen, S. Kuppens, T. van den Akker, J. Stekelenburg, N. Schuitemaker and J. Zwart). We also thank Dr. Peter Zuithoff for his advice in the statistical analysis.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.