Applying the international classification of diseases to perinatal mortality data, South Africa

Abstract Objective To examine the feasibility of applying the International Classification of Diseases-perinatal mortality (ICD-PM) coding to an existing data set in the classification of perinatal deaths. Methods One author, a researcher with a non-clinical public health background, applied the ICD-PM coding system to South Africa’s national perinatal mortality audit system, the Perinatal Problem Identification Program. The database for this study included all perinatal deaths (n = 26 810), defined as either stillbirths (of birth weight > 1000 g and after 28 weeks of gestation) or early neonatal deaths (age 0–7 days), that occurred between 1 October 2013 and 31 December 2016. A clinical obstetrician verified the coding. Findings The South African classification system does not include the timing of death; however, under the ICD-PM system, deaths could be classified as antepartum (n = 15 619; 58.2%), intrapartum (n = 3725; 14.0%) or neonatal (n = 7466; 27.8%). Further, the South African classification system linked a maternal condition to only 40.3% (10 802/26 810) of all perinatal deaths; this proportion increased to 68.9% (18 467/26 810) under the ICD-PM system. Conclusion The main benefit of using the clinically relevant and user-friendly ICD-PM system was an enhanced understanding of the data, in terms of both timing of death and maternal conditions. We have also demonstrated that it is feasible to convert an existing perinatal mortality classification system to one which is globally comparable and can inform policy-makers internationally.


Introduction
High on the global health agenda is the need to accelerate progress towards ending preventable perinatal deaths, defined by the World Health Organization (WHO) as either a stillbirth of weight > 1000 g or after at least 28 weeks gestation, or an early neonatal death in the first 7 days after birth. 1 In developing appropriate intervention strategies to reach this target, the causes of perinatal deaths must be classified in a globally comparable way. 2,3 A recent systematic review identified no less than 81 different systems used to classify perinatal deaths globally, with only 17 systems using the International Statistical Classification of Diseases and Related Health Problems (ICD) codes. 4 Other studies have recognized that multiple, disparate systems impede the ability to understand and achieve accurate estimates of cause of death, hindering effective prevention strategies. 5,6 Of particular importance is the need to focus on the mother-infant dyad, as maternal condition is closely related to perinatal death. 1 The Every Newborn Action Plan recommends that maternal complications be recorded as part of perinatal death registration; however, challenges existed in applying the 10th edition of the ICD (ICD-10) classification system as maternal condition was not linked to perinatal condition. 7 To address these issues, the WHO application of ICD-10 to perinatal deaths (ICD-perinatal mortality or ICD-PM) was published in 2016, 8,9 the first perinatal death classification system developed for application globally. 10 ICD-PM is modelled on the WHO application of the ICD-10 system to deaths during pregnancy, childbirth and the puerperium (ICD-maternal mortality or ICD-MM), 11 and follows all coding rules of ICD- 10. 12 Importantly, the ICD-PM system identifies the timing of perinatal death (i.e. antepartum, intrapartum or neonatal), links causes of death to existing ICD-10 codes and connects maternal condition with perinatal death. 8 One of the aims of ICD-PM is to group ICD-10 codes into clinically relevant and easy-to-use categories. 10 We demonstrate the benefits achieved, in terms of an improved understanding of the data, from the application of ICD-PM codes to perinatal deaths that were previously classified using the South African perinatal mortality audit system, called Perinatal Problem Identification Program.

Data source
South Africa's perinatal mortality audit system 13 records and classifies perinatal deaths at all 588 clinics across the country. Each clinical team performs a mortality review shortly after death and reports the cause of perinatal death (and associated maternal condition) to the classification system. For the purposes of the system, perinatal deaths are defined as either fresh or macerated stillbirth or early neonatal death (age 0-7 days). The primary obstetric cause of death is classified in terms of both lead categories and subcategories according to Box 1. Maternal condition is also recorded, and classified as either healthy (where the examining clinician did not identify any clinical problems) or as one of the medical/obstetric conditions listed in Box 2. Classifications of perinatal death are linked to maternal condition lead categories, but not to subcategories. Data are joined into a national database at the Medical Re-Implementation of ICD-PM codes, South Africa Tina Lavin et al.
search Council Unit for Maternal and Infant Health Care Strategies, Pretoria. Regular auditing of individual clinics is conducted to ensure the completeness and accuracy of the database.
We used all 26 810 perinatal deaths, which occurred during the period between 1 October 2013 and 31 December 2016, recorded in the classification system's database ( Table 1). The start date coincided with the launch of the third version of the system, which had been improved to include gestational age at death.

Conversion to ICD-PM coding
The first author, a non-clinical researcher with a background in public health, studied The WHO application of ICD-10 to deaths during the perinatal period: ICD-PM 9 to learn to apply ICD-PM codes to the classification system's database. The coding conversion took place between November 2017 and January 2018. The second author, a consulting obstetrician, provided guidance and verification on a voluntary basis. The ICD-PM system 2,9,14 classifies mortality according to: (i) time of death, whether antepartum (A1-A6), intrapartum (I1-I7) or neonatal (N1-N11); (ii) the primary cause of perinatal death (e.g. loss of fetal blood: P50); and (iii) the main maternal condition (M1-M4 to describe various complications and conditions, and M5 for healthy mother) at the time of perinatal death.

Ethics
Data were collected with the permission of the South African Department of Health. This analysis was approved by the technical task team who run the database and produce the reports from the South African Medical Research Council/University of Pretoria Maternal and Infant Health Care Strategies unit. This was a secondary analysis and all identifiers of the cases were removed. Ethics approval was given by the University of Western Australia Human Ethics Committee (RA/4/1/7955, 20 November 2015).

Intrapartum deaths
The main causes of the 3725 intrapartum deaths (

Neonatal deaths
The main causes of the 7466 neonatal deaths were complications of intrapartum events (2184 deaths; 29.3%; N4) or low birth weight and prematurity (2128 deaths; 28.5%; N9). All neonatal deaths classified as complications of intrapartum events were due to severe birth asphyxia (N4 P21;  Table 3 provides an example of how ICD-PM codes were applied to neonatal deaths classified by the South African classification system as being due to preterm labour. According to the classification system, most (83.5%, 1772/2121) of the deaths due to preterm labour were associated with a healthy maternal condition. Under the ICD-PM classification, however, 96.5% of these (1710/1772) were associated with a non-healthy maternal condition. For example, cases of perinatal death due to idiopathic preterm labour, premature rupture of membranes, premature rupture of membranes with chorioamnionitis, cervical incompetence and premature rupture of membranes with chorioamnionitis and intact membranes were assigned the codes M3 P03.8, M2 P01.1, M2 P01.1, M2 P01.0 and M1 P02.7, respectively. Only 3.5% (62/1772; iatrogenic preterm delivery for no real reason) of neonatal deaths due to preterm labour associated with a healthy mother, according to the South African classification, are coded as M5 under the ICD-PM system.

Discussion
Here we show that ICD-PM coding improve consideration of maternal complication when classifying perinatal deaths. Previous research in South Africa reported that maternal complications were linked to around one half of all stillbirths and one quarter of early neonatal deaths. 13 According to the South African classification system, 45.7% (8644/18 927) of stillbirths and 27.4% (2158/7883)

Box 2. South African Perinatal Problem Identification Program classification of maternal conditions
No obstetric condition  We managed to classify all neonatal deaths with a primary cause of intrapartum asphyxia with an associated maternal condition using the ICD-PM codes, while the South African classification system only classified 17.4% (512/2942). Several subcategories such as labour-related intrapartum asphyxia, cord around the neck and others as outlined in Box 1 are classified according to the South African classification system as perinatal complications with a healthy mother. Using the ICD-PM system, however, these deaths can be correctly categorized as the result of a maternal condition. Antepartum haemorrhage, because of abruptio placentae or placenta praevia is considered a perinatal condition under the South African classification system, but classified as a maternal condition by the ICD-PM system.
We also show that ICD-PM coding improve consideration of timing of death. A recent systematic review found that 59% of globally reported stillbirths had no information regarding the timing of death, 15 making the appropriate timing of interventions difficult to identify. Further, in some resource-poor settings the timing of a perinatal death may be the only piece of information captured. This information should therefore be a part of any classification system. 16 The application of the ICD-PM coding system to our data revealed a significant burden of deaths occurring during the antepartum period. Further, more than a quarter of early neonatal deaths were due to low birth weight. This highlights the already established importance of investment in antenatal care to reduce perinatal mortality. The  More detailed information for these categories would enhance the alignment of the existing data collection system to ICD-PM Birth trauma (I2) is not captured by the South African system: most deaths due to birth trauma are classified as traumatic assisted delivery or other cause of death not described in classification More detailed information for these categories would enhance the alignment of the existing data collection system to ICD-PM Perinatal Problem Identification Program maternal condition classifications too broad For maternal conditions in the South African system, only lead categories can be linked to perinatal death (i.e. hypertension, obstetric haemorrhage, medical and surgical disorders); no specific details (e.g. proteinuric hypertension, eclampsia, chronic hypertension, etc.) can be linked Improved linkage between perinatal cause of death and certain maternal conditions would allow more specific maternal ICD-PM codes to be applied For deaths related to other complications of labour and delivery (M3, other complications of labour and delivery), a large proportion of cases were classified as unspecified under the code P03.9 fetus and newborn affected by complication of labour and delivery, unspecified. In the South African system, these deaths were classified as labour-related intrapartum asphyxia with no further detail as to the exact labour-related maternal cause of these deaths It may be possible to reduce the number of deaths falling under this unspecified category if South African mortality audits were able to capture more detailed information around maternal causes for complications of labour and delivery, such as those conditions falling under: M3 P03.1 fetus and newborn affected by other malpresentation, malposition, disproportion during labour and delivery; or P03.6 fetus and newborn affected by abnormal uterine contractions and conditions classifiable under O60-O75 High proportion of antepartum deaths classified as unspecified causes with no maternal complication (A6 M5) Initially it appeared that ICD-PM coding was not sufficiently sensitive to identify the causes of these antepartum deaths accurately; however, these deaths were at the highest descriptive level in the South African system. No more information regarding the cause of death was available These deaths were due to unexplained or unknown causes. There could be no improvement in the ICD-PM classification system that would reduce the number of deaths classified as A6 M5 ICD-PM: International Classification of Diseases-perinatal mortality.

Research
Implementation of ICD-PM codes, South Africa Tina Lavin et al.
2016 WHO antenatal care recommendations 17 include an increased number of antenatal care contacts in the third trimester. In response to these recommendations and the increased number of third-trimester stillbirths observed when antenatal care visits had not been made during this period, the number of recommended antenatal care visits was changed in South Africa in April 2017. 18 A commonly cited burden of perinatal mortality is prematurity and prematurity-related causes. 19 However, simply identifying that prematurity is an important contributor to deaths gives no information regarding the optimal timing for interventions. From the ICD-PM classification, we see that 36.7% (1270; coded under A5, disorders related to fetal growth) of deaths due to prematurity (3426; the total of deaths classified as A5, I6 or N9) occurred during the antepartum period, and that 72.7% (923/1270) of these deaths were also related to a maternal complication. This information is invaluable to public health workers and policy-makers in targeting interventions; a heightened awareness of the causes of such deaths allows a focus on preterm-related issues, showing that both obstetric and neonatal interventions are required.
For implementing ICD-PM coding, systematic training of data administrators in the classification of deaths using ICD-PM will be required to ensure familiarity with the new system, as well as consistency across settings. Data administrators will also need to have access to clinicians to discuss cases that do not clearly fit a specific ICD-PM classification. In our experience, however, the ICD-PM system is both clinically relevant and easy to use; for example, the coder for this study does not have a clinical background. There was a high level of agreement between the coder and the verifying obstetrician, with differences encountered in only two cases: (i) premature rupture of membranes with chorioamnionitis (M1 P01.1 according to coder, M1 P02.7 according to obstetrician) and (ii) unexplained uterine death (A3 according to coder, A6 according to obstetrician). This demonstrates the feasibility in implementing the ICD-PM codes to existing data sets by administrators or allied health providers, in consultation with clinicians. Data administrators can be trained in the application of ICD-PM coding under the mentorship of clinicians, an advantage in low-resource settings.
We noted some specific issues with ICD-PM, including mutually exclusive categories, deaths which could be classified under two different ICD-PM codes, multiple contributing factors for cause of death, and causes of death not captured by the South African classification system but considered by ICD-PM codes (or vice versa). Examples of these issues and potential solutions are discussed in Table 4.
As maternal and perinatal outcomes are closely related, both mother and infant benefit from intervention; 2 this is particularly relevant in the management of hypertension and care during the intrapartum period. 3,20,21 However, possible challenges exist with the application of the ICD-PM system to data sets which consider perinatal death and maternal condition separately, introducing issues in the integration of the two systems. The adaption of integrated perinatal and maternal data collection systems may be difficult in poorly resourced settings. For countries that do not have well established death classification systems, future developments could consider autopsy review categories aligned with ICD-PM codes for better consistency between death review and coding stages. For example, the South African classification system could be strengthened to align more closely to ICD-PM as described in Table 4.
In conclusion, by allowing for an increased recognition of the role of maternal condition and the timing of death in perinatal mortality, our conversion of an existing national perinatal mortality data set to ICD-PM codes enhanced our understanding of the data. This work is part of a larger work investigating perinatal deaths in South Africa and the required interventions. 18,22 Our new classification of perinatal deaths could inform the allocation of resources and the timing of interventions. Adopting the ICD-PM coding system internationally would lead to a consistent global perinatal death classification system, which would create comparable data that could inform policy-makers globally. ■