Apply perinatal and maternal mortality audit and classification in the South African setting

In one line

Maternal and perinatal mortality audit is the disciplined loop — count every death, classify it consistently, find the avoidable factor, recommend a change, implement it, then re-count — and in South Africa that loop runs on two named national engines: the NCCEMD Saving Mothers confidential enquiry and the Saving Babies programme built on the Perinatal Problem Identification Programme (PPIP). The consultant's job is not to memorise rates but to turn a death into a system change.

The arithmetic of the rates themselves and how to read a confidential-enquiry table critically build on the maternity-statistics and critical-appraisal groundwork in maternity-statistics-critical-appraisal; this objective works on the audit machinery — the classifications, the avoidable-factor analysis, and the SA reports that drive policy.

Mechanism & pathophysiology

Mortality audit has no cellular pathophysiology, but it has a mechanism, and getting the mechanism wrong is how units waste effort. The "pathology" being investigated is the pathway from a healthy pregnancy to a preventable death, and the audit cycle is the instrument that dissects it.

The cycle is iterative and never closes: identify every death → collect the data on each → analyse cause and contributory factors → recommend specific, actionable change → implement it → re-audit to confirm the change worked, then start again. A unit that counts deaths but never reaches the implement-and-re-audit arc is doing surveillance, not audit, and its numbers will not move. The WHO frames this as Maternal Death Surveillance and Response (MDSR), launched in 2012 — a continuous action cycle of identify/notify → review → analyse → respond → monitor — and the deliberate emphasis is on the response limb, because counting without acting is the commonest failure mode in resource-limited systems.

The engine that makes the analysis honest is confidential enquiry. An appointed committee identifies the deaths; an independent expert team — assessors who did not provide the care — reviews each anonymised case to determine the cause and, critically, the avoidable factors. Confidentiality is structural, not cosmetic: names of patients and staff are stripped so that clinicians report deaths fully and without fear of medicolegal or disciplinary exposure. The governing ethic is "no name, no blame" — the enquiry exists to learn, not to punish — and the moment a system is perceived as punitive, under-reporting follows and the data corrupt. This is why South Africa's NCCEMD operates under a legal framework that makes maternal death a notifiable condition while protecting the enquiry findings from use in litigation.

The conceptual lens applied to every death is the three delays model (Thaddeus and Maine, 1994): the delay in the decision to seek care (the woman or family does not recognise danger or act), the delay in reaching care (transport, distance, infrastructure), and the delay in receiving adequate care once at a facility. The model's power is that it forces attention onto the system, not just the disease — a woman who dies of eclampsia after a four-hour ambulance wait did not die of "eclampsia" in any actionable sense; she died of a second-delay failure that an audit must name as such. Most South African maternal deaths involve more than one delay, and the third delay (substandard facility care) is the one the health system can most directly fix. The model maps cleanly onto the SA referral architecture: the first delay is a community and antenatal-education problem; the second is an emergency-medical-services and inter-facility-transfer problem (the district-to-regional-to-tertiary chain, where a CHC without a doctor must move a haemorrhaging woman by an ambulance that may be hours away); and the third is a facility-resource-and-skills problem (no blood bank on site, no anaesthetist for an emergency caesarean, a registrar who does not recognise an evolving HELLP). Naming the delay is what makes the recommendation land in the right place — a community-mobilisation intervention will not fix a death caused by an empty blood fridge.

The avoidable-factor analysis is the part that converts a death into a recommendation. Three terms are kept distinct. An avoidable (or modifiable) factor is something which could have caused the death and yet was potentially avoidable. A missed opportunity is a potentially avoidable death where an opportunity to prevent it was present but was missed. Substandard care is poor care which may have resulted in the woman's death. Avoidable factors are sorted into three categories: patient-oriented (late booking, non-attendance, declined care), administrative / health-system (no transport, no blood, no theatre, staff shortages, no ICU bed), and healthcare-provider (failure to recognise, failure to act, wrong treatment). In the South African enquiry, avoidable factors are commonest in the patient-related category — but two cautions matter. First, "patient-oriented" must not become a way to write off a death as the woman's fault: a woman who books late because the clinic is twelve kilometres away and charges her a day's wage in transport has a system problem dressed as a patient choice. Second, the administrative and provider categories together account for the larger share of correctable system failure — and they are where guidelines and training bite. The discipline of the enquiry is to record every avoidable factor for a death, not just the dominant one, because a single death typically has a chain of them (a missed antenatal proteinuria, then a transport delay, then an under-resourced district response) and each link is a separate place to intervene.

A repeated avoidable factor across many deaths becomes a national recommendation: the recurrent finding of "no blood available" drove the requirement for emergency O-negative blood at delivery sites; the recurrence of "failure to recognise and manage obstetric emergencies" drove the ESMOE (Essential Steps in the Management of Obstetric Emergencies) skills-and-drills programme; the recurrence of unsafe spinal anaesthesia in district caesareans drove anaesthetic-safety guidance. This is the audit cycle operating at national scale — the aggregate of thousands of individual avoidable-factor classifications becomes the evidence for a policy that is then re-audited in the next triennium.

Assessment

Auditing well depends on counting the right thing with the right denominator. The definitions are not interchangeable, and an examiner will probe whether the candidate can state them precisely.

Maternal death (ICD-MM / ICD-10) is the death of a woman while pregnant or within 42 days of termination of pregnancy, irrespective of the duration and site of the pregnancy, from any cause related to or aggravated by the pregnancy or its management, but not from accidental or incidental causes. The 42-day window and the exclusion of accidental/incidental causes are the load-bearing clauses.

• Direct obstetric death — results from obstetric complications of the pregnant state, from interventions, omissions or incorrect treatment, or from a chain of events arising from any of these (e.g. PPH, eclampsia, ruptured ectopic, amniotic-fluid embolism).
• Indirect obstetric death — results from previous existing disease, or disease that developed during pregnancy and was not due to direct obstetric causes, but was aggravated by the physiological effects of pregnancy (e.g. rheumatic heart disease, HIV-related infection, pre-existing renal disease). In South Africa the indirect category dominates — driven by HIV-associated non-pregnancy-related infections — which is why a system designed around direct causes alone misses most of the SA problem.
• Incidental (coincidental) death — a death from a cause entirely unrelated to and unaffected by the pregnancy (e.g. a motor-vehicle accident, a homicide). These are excluded from the maternal-death numerator; conflating them inflates the ratio.
• Late maternal death — death from direct or indirect causes more than 42 days but less than one year after termination of pregnancy. Captured separately; relevant as HIV and cardiac disease kill beyond 42 days.

The two indices candidates confuse:

• Maternal mortality ratio (MMR) = maternal deaths per 100,000 live births. It is the risk of death per pregnancy — a measure of obstetric safety. South Africa, like most enquiry systems, reports an institutional MMR (iMMR), restricted to deaths in (and deliveries at) health facilities, because community deaths are incompletely captured.
• Maternal mortality rate = maternal deaths per number of women of reproductive age (typically per 100,000). It folds fertility into the denominator, so it answers a population question, not a per-pregnancy-safety question. Using the rate to compare obstetric care between units is a category error.

The same numerator-versus-denominator discipline governs perinatal indices, where the threshold for counting is birthweight ≥500 g in the South African system (some low-income settings still use ≥1000 g, which is not comparable):

The perinatal period spans from fetal viability (the SA reporting threshold of ≥500 g, ≈22 weeks) through the first seven completed days of life; the neonatal period runs to 28 completed days, split into early (days 1–7) and late (days 8–28). A stillbirth (fetal death) is a baby born with no signs of life at or above the reporting threshold; an early neonatal death is a liveborn baby who dies in the first seven days. The distinction is consequential because it relocates the death in time: an antepartum stillbirth points upstream to antenatal surveillance, an intrapartum (fresh) stillbirth points at labour-ward care, and an early neonatal death points at resuscitation and newborn care. A useful derived statistic is the Perinatal Care Index = PNMR ÷ low-birthweight rate: it normalises mortality for the case-mix of small babies, so a high index signals that babies of a given size are dying who should have survived — that is, poor care rather than merely a sick population. Two units with identical PNMRs but different low-birthweight rates are not delivering equivalent care, and the index is the number that exposes it.

The South African perinatal cause profile, read off the PPIP data, is its own teaching point. The commonest primary obstetric causes of perinatal death are unexplained intrauterine death (most are macerated — they died before labour, pointing at antenatal surveillance and the detection of growth restriction), spontaneous preterm labour (with chorioamnionitis, preterm prelabour rupture of membranes and cervical insufficiency behind it), and intrapartum hypoxia / asphyxia (the most avoidable category — a labour-ward failure of monitoring and timely delivery). Hypertension, antepartum haemorrhage, infection (congenital syphilis remains a named SA cause), fetal abnormality and fetal growth restriction follow. The commonest final neonatal causes are immaturity (prematurity), perinatal hypoxia, infection and congenital abnormality. The audit value is in pairing them: a high "intrapartum hypoxia" count with a low "growth restriction detection" count tells a unit that its labour ward, not its antenatal clinic, is where babies are being lost.

One further definitional pair earns attention. A pregnancy-related death is the death of a woman while pregnant or within 42 days of termination irrespective of cause — it does not require the death to be related to or aggravated by the pregnancy, so it includes the incidental deaths that a maternal death excludes. It is the easier number to capture (no causal judgement needed) and is sometimes the only one available in vital-registration systems; the gap between pregnancy-related and maternal deaths is precisely the coincidental category. The distinction matters when comparing SA enquiry data (true maternal deaths, causally adjudicated) against routine population statistics (often pregnancy-related deaths), because the latter is the larger number.

Reading a confidential-enquiry table is its own skill. A higher iMMR in a tertiary unit than in the districts that refer to it is the expected consequence of receiving the sickest women, not proof of worse care — the unit of audit is the avoidable-factor classification, not the raw ratio. The honest signal in a Saving Mothers table is not the ranking of provinces but the proportion of deaths judged avoidable and which delay dominated, because that is what a recommendation can target. The institutional restriction is itself a known blind spot: the iMMR misses women who die at home or in transit (the first and second delays), so a falling iMMR in a setting where home deaths are rising could conceal a worsening community problem — which is part of why the NCCEMD data, capturing out-of-facility and private deaths, are read alongside, not instead of, the DHIS figure.

Management

The "management" of mortality is the audit response: converting a classified death into a delivered intervention. Structure it immediate → ongoing → long-term, mirroring the cycle.

Immediate — classify the death correctly. Two complementary systems run in parallel.

• ICD-MM classifies maternal deaths into nine groups: six direct — (1) pregnancy with abortive outcome, (2) hypertensive disorders, (3) obstetric haemorrhage, (4) pregnancy-related infection, (5) other obstetric complications, (6) unanticipated complications of management — plus (7) non-obstetric complications (indirect), (8) unknown/undetermined, and (9) coincidental. The classification proceeds through three levels: the type (direct/indirect/unspecified), the group, and the underlying cause. Assigning the underlying rather than the final cause matters: a woman who dies of multi-organ failure after a postpartum haemorrhage is an obstetric haemorrhage death (group 3), not a "multi-organ failure" death — the final mechanism is not the auditable cause.
• ICD-PM classifies perinatal deaths first by timing — antepartum (six groups, A1–A6), intrapartum (seven groups, I1–I7), and early neonatal (eleven groups, N1–N11) — and then mandates linking a maternal condition to every death using the M1–M5 axis (M1 placenta/cord/membranes; M2 maternal complications of pregnancy; M3 other complications of labour and delivery; M4 maternal medical and surgical conditions; M5 no maternal condition). Forcing an M-code onto every perinatal death — even "M5, no maternal condition" — is ICD-PM's central innovation: it couples the dead baby to the mother's care and makes intrapartum, third-delay failures visible as a class. The pilot that tested ICD-PM in South African and UK databases showed that in a large share of perinatal deaths the mother was healthy (M5), which is precisely the finding that redirects attention to intrapartum care quality rather than maternal disease.

Ongoing — analyse contributory factors and write the recommendation. Apply the three-delays lens and the avoidable-factor categories to each death, then write a recommendation that is specific, assigned and measurable — "all women with a booking BP ≥140/90 get same-visit urine protein and a defined escalation" beats "improve hypertension care". The recommendation is the product of the audit; a death that yields no actionable change has not been audited, only recorded.

Long-term — implement and re-audit, and feed the national system. Each facility's PPIP data and maternal-death notifications flow upward: maternal deaths into the NCCEMD / MaMMAS database and Saving Mothers reports; perinatal deaths into PPIP / Saving Babies; and routine denominators (births, deliveries) into the District Health Information System (DHIS). The NCCEMD reports systematically capture more deaths than DHIS because they include deaths outside facilities and in private hospitals, whereas DHIS counts only public-facility deaths — so the two never quite agree, and the corrected figure reconciles them (the national correction adjusts MaMMAS counts against DHIS for provinces where DHIS recorded more, which is why a published iMMR is a "corrected" figure rather than a raw count). The national recommendations that emerge (the maternity care guidelines, ESMOE skills-and-drills training, the requirement for on-site blood and a functioning referral chain) are the re-audit's interventions writ large.

A worked classification

A 27-year-old, HIV-positive, on antiretroviral therapy with a suppressed viral load, books at 30 weeks, develops Pneumocystis pneumonia at 33 weeks, and dies of respiratory failure undelivered. Worked through the machinery: this is a maternal death (she was pregnant at death). It is indirect — the pneumonia is a non-pregnancy-related infection aggravated by, but not caused by, the pregnant state — so it falls in ICD-MM group 7 (non-obstetric complications), not group 4 (which is pregnancy-related, i.e. genital-tract, infection — a common trap). The underlying cause is the HIV-associated NPRI, not the final mechanism (respiratory failure). The avoidable-factor analysis then asks: was there a first-delay factor (late booking at 30 weeks)? A third-delay factor (delayed recognition or treatment of the pneumonia)? The recommendation falls out of which delay dominated. Contrast a 24-year-old who dies of a road-traffic collision at 28 weeks: she is coincidental (group 9) and excluded from the maternal-death numerator entirely. The two cases share the label "death in pregnancy" and diverge at the first classification step — which is exactly the discrimination the audit must get right before any rate is computed.

Guidelines compared

The international and South African machinery overlap but are built for different purposes; a candidate should know which tool answers which question.

The relationship to hold is layered, not competing: ICD-MM and ICD-PM supply the vocabulary, MDSR supplies the cycle, and South Africa's Saving Mothers and Saving Babies are the national implementations — confidential enquiries that classify with (or map to) the WHO systems and add the avoidable-factor analysis and the local recommendations the international codes deliberately omit. A recent direction is the move toward an integrated Maternal and Perinatal Death Surveillance and Response (MPDSR) that audits mother and baby together, reflecting that a dead baby and a sick mother are frequently the same third-delay failure.

The evidence & the controversy

The South African maternal-death picture is dominated by a single epidemiological fact that distinguishes it from high-income enquiries: for most of the past two decades the leading underlying cause was non-pregnancy-related infection (NPRI) — overwhelmingly HIV and its opportunistic infections — an indirect cause that a direct-cause-centric audit would systematically under-prioritise. That is the strongest argument for why South Africa needs its own enquiry rather than importing a high-income template wholesale.

The most recent national data show this picture shifting and a hard-won improvement. In 2022 the corrected number of maternal deaths was 1,062 and the corrected iMMR was 109.6 per 100,000 live births — a marked fall from 148.1 in 2021 and 126.1 in 2020, returning toward the pre-pandemic level of 98.8 in 2019. That spike-and-recovery is almost entirely the COVID-19 signal: COVID-19 pneumonia caused 369 maternal deaths in 2021 versus only 12 in 2022, a vivid demonstration that an enquiry captures pandemic shocks in real time. Stripped of COVID-19, the 2022 cause ranking was NPRI 180 (18.6%), hypertensive disorders 166 (17.1%), obstetric haemorrhage 162 (16.7%), medical and surgical disorders 140 (14.4%, most commonly cardiac), and early pregnancy complications 97 (10%). The decline of NPRI from 37% of deaths in 2021 to under a fifth reflects two decades of antiretroviral scale-up — the clearest illustration that the indirect burden is the one most amenable to a public-health intervention, and a reason the SA enquiry's focus on it was correct.

The sharper question is the avoidability finding and what to do with it. In 2022, 59% of maternal deaths were assessed as possibly or probably preventable by the health system, with the most avoidable being obstetric haemorrhage, anaesthetic and hypertensive deaths — the direct, "old-fashioned" causes that should be the most fixable. The doubling of anaesthetic deaths to 39 (from 17 in 2021), most judged clearly preventable, is the live concern: it points at a skills-and-safety gap in district-level caesarean anaesthesia, not at an intractable disease, and sits against a rising caesarean rate (31.1% in 2022) whose case-fatality, though falling, still runs well above that of vaginal birth. The defensible position is that the direct, avoidable causes are now the binding constraint on the iMMR — South Africa's residual maternal mortality is increasingly a story of system failures (no blood, no theatre, unsafe anaesthesia, missed hypertension) rather than untreatable pathology, which is exactly what makes the SDG target of 70 per 100,000 by 2030 plausible if the audit's recommendations are implemented.

A complementary thread now reshaping audit practice is the shift from counting deaths to counting severe acute maternal morbidity (SAMM) — the "near-miss": a woman who nearly died but survived a life-threatening complication. Near-misses outnumber deaths by an order of magnitude, occur often enough for a single facility to learn from, and carry less medicolegal charge because the woman lived — so they are a richer, faster audit signal than the rare death, and the same avoidable-factor analysis applies. Pairing near-miss review with death enquiry, and integrating maternal and perinatal audit into a single MPDSR loop, is the current direction of travel: a stillbirth and a maternal near-miss in the same labour are usually the same third-delay event seen from two sides, and auditing them together finds the system fault faster than either alone.

On the perinatal side the controversy is the opposite — the rates have been stubborn, with a PNMR of roughly 36/1000 (≥500 g) that has not fallen as fast as the maternal figures, and the three dominant primary causes — unexplained intrauterine death, spontaneous preterm labour and intrapartum hypoxia — include one (intrapartum hypoxia) that is a pure third-delay, care-quality failure and substantially avoidable, and two that are harder. The argument to make is that the most avoidable perinatal death is the intrapartum stillbirth or hypoxic-ischaemic encephalopathy in a normally-grown term baby — a death that ICD-PM's M5 ("no maternal condition") flags precisely because it indicts the labour-ward care, not the mother. This is where audit translates most directly into practice: fetal-heart-rate monitoring discipline, partogram use, and timely caesarean.

Landmark trials & key evidence

The Saving Mothers and Saving Babies reports are confidential-enquiry programmes rather than randomised trials, and that is the point: the evidence for what kills South African mothers and babies, and what is avoidable, comes from systematic audit, not from RCTs — which is why the audit is the evidence base for national obstetric policy.

Exam traps & red flags

• Confusing the ratio and the rate. MMR (per 100,000 live births, per-pregnancy risk) is not the maternal mortality rate (per women of reproductive age, fertility-laden). Comparing units on the rate is a category error.
• Counting the final cause, not the underlying cause. A PPH death that ends in multi-organ failure is an obstetric haemorrhage (ICD-MM group 3) death, not a "multi-organ failure" death — classify the underlying cause.
• Including incidental/coincidental deaths in the numerator. A homicide or a road-traffic death in a pregnant woman is coincidental and excluded from the maternal-death count; including it inflates the iMMR.
• Missing the indirect burden. An audit framework built only around direct causes under-counts the SA reality, where HIV-driven non-pregnancy-related infection has been the leading underlying cause — most SA maternal deaths are indirect.
• Reading a high tertiary iMMR as bad care. A referral unit that receives the sickest women will post a higher iMMR; the auditable unit is the avoidable-factor classification, not the raw ratio.
• Treating audit as counting. Surveillance that never reaches implement and re-audit will not move the numbers — the response limb is the audit.
• Forgetting the maternal condition on a perinatal death. ICD-PM requires an M-code (including M5 "no maternal condition") for every perinatal death; omitting it discards the mother–baby link that exposes intrapartum-care failures.
• Using the wrong birthweight threshold. SA perinatal data count from ≥500 g; a unit using ≥1000 g will report an artificially lower, non-comparable PNMR.
• Mistaking "no blame" for "no accountability". Confidential enquiry protects individuals so that reporting is honest; it then drives system accountability through recommendations — the protection is the mechanism that makes accountability possible, not a substitute for it.

Evidence anchors

• Thaddeus S, Maine D — Too far to walk: maternal mortality in context, Soc Sci Med 1994;38(8):1091–1110
• Allanson ER et al. — WHO ICD-PM pilot in South Africa and the UK, BJOG 2016;123(12):2019–2028
• Moran NF, Naidoo M, Moodley J — Reducing maternal mortality on a countrywide scale: the role of emergency obstetric training (ESMOE), Best Pract Res Clin Obstet Gynaecol 2015;29(8):1102–1118
• WHO — The application of ICD-10 to deaths during pregnancy, childbirth and the puerperium (ICD-MM)
• WHO — The application of ICD-10 to deaths during the perinatal period (ICD-PM)
• WHO Global Health Observatory — maternal mortality ratio indicator definition
• NCCEMD — Saving Mothers Fact Sheet 2022 (South Africa, NDoH)
• Saving Mothers and Babies — maternal mortality, avoidable factors and confidential enquiry (Bettercare)
• Saving Mothers and Babies — perinatal mortality definitions, rates and causes (Bettercare)
• South Africa NCCEMD Saving Mothers 2020–2022 Eighth Comprehensive Triennial Report — the full triennial maternal-death enquiry (named source; figures above are from the 2022 fact sheet).
• Perinatal Problem Identification Programme (PPIP) / Saving Babies — the national facility-based perinatal-death audit database; District Health Information System (DHIS) for routine denominators.