Stillbirths Counted Better Than Mothers
A ten-year audit of DHIS2 returns from five African countries finds facility reporting captures stillbirths close to the modelled national rates — and, in most of the countries, fewer than half of expected maternal deaths.
Every month, tens of thousands of health facilities in Ghana, Kenya, Tanzania, Uganda and Zimbabwe file a return that carries the deaths that happened inside their walls. The returns travel by paper form and clerk and phone line into a national database, and from there into the figures a ministry quotes when it is asked how many mothers are dying. A study published on 20 November in the Journal of Global Health set ten years of those returns against one another and against the global modelled estimates, and found that what the filing system sees depends heavily on which death it is being asked about. Stillbirths arrive close to the modelled national rates, sometimes above them. Maternal deaths, in most of these countries, arrive at a fraction of what the models expect.
The paper is the first multi-country audit of its kind for maternal and newborn indicators, and its headline numbers are blunt. In Kenya, adjusted facility data accounted for 9–26% of the maternal deaths the Maternal Mortality Estimation Inter-Agency Group expects for the country across the study years. Tanzania captured roughly 19–49%, Uganda 29–58%, Zimbabwe 36–54%. Ghana came closest to parity, above 70%, with an adjusted 2017 figure of 322 per 100,000 live births slightly exceeding the modelled 301.
Sixteen indicators, twenty-nine fields
Joseph Akuze and colleagues, most of them based at the London School of Hygiene & Tropical Medicine, Makerere University and the ministries of health of the countries they studied, worked from national DHIS2 repositories rather than from facility records. They selected 16 of the 25 indicators in the Every Newborn Action Plan — those for which a numerator and denominator existed somewhere in the system — and traced the 29 data elements behind them across 2014 to 2023. They then applied three of the four dimensions of WHO’s Data Quality Review framework: availability of the data element, completeness of facility reporting, and internal and external consistency. The fourth dimension, timeliness, could not be assessed at all, because submission dates were not consistently available in the five extracts — a gap the authors call “a substantive limitation rather than a minor one”.
The country files were not alike. Ghana’s extract covers hospitals only, from 2017 to 2022, and its final year is six months long. Kenya’s runs from 2019, Tanzania’s from 2015, Uganda’s from 2014, Zimbabwe’s from 2015. Availability of the 29 fields ranged from 41% in Ghana and Zimbabwe to a peak of 76% in Uganda. Fifteen of the sixteen indicators were computable in Uganda, eleven in Kenya and Tanzania, nine in each of Ghana and Zimbabwe. Four ENAP indicators were absent everywhere: neonatal morbidity, disability after neonatal conditions, small-for-gestational-age births, and any quality-of-care standard. Birth registration data appeared only in Uganda, from 2021, while it was still being piloted.
The ratios that should not be there
The most arresting material sits in the internal-consistency checks, where the study compares ratios that ought to move together. The ratio of the reported stillbirth rate to the reported neonatal mortality rate is expected to sit between 0.7 and 1.0. In Ghana it ran between 5.0 and 12.5; in Uganda between 2.0 and 3.0. The perinatal-to-neonatal ratio, expected at 1.2–2.0, reached 3.0–7.0 in Ghana. Kenya’s ratio of neonates started on kangaroo mother care to low-birthweight infants exceeded 2.0 against an expected 0.4–0.8, a pattern that began in 2020 and suggests either widening eligibility or incomplete detection of low birthweight.
The authors resist the obvious reading. Ghana’s hospital-only frame, they write, means referral facilities carrying complicated cases dominate the returns, and differences in how stillbirth is defined and counted across registers do the rest. Their own conclusion is that these ratios are “best read as hypotheses requiring further investigation rather than established evidence of neonatal death under-reporting”. A related artefact shows up in caesarean section: Ghana’s returns put the rate at 26–32% against a WHO reference of 13–14.5%, which is what a surgical referral hospital looks like when it is asked to speak for a whole population.
Intermittent filing, uneven tiers
Completeness splits sharply by level of the system. Tanzania achieved near-universal reporting across dispensaries, health centres and hospitals, above 95% through 2015–2022. Kenya’s hospitals filed at 95–98% while its lower-level facilities slid from 81% to 54% over five years. Uganda’s hospital reporting climbed from 75% in 2014 to 86% in 2018, then fell to 50–51% in 2019 and 2020 before recovering to 77% by 2023 — a decline that began before the pandemic, which the authors attribute speculatively to system transitions and health-worker industrial action, explicitly not measured in the study.
At the level of the individual facility the picture is weaker still. Of 22,166 facilities that ever reported, only 26% submitted data in every month they were expected to from their first report onward, and 26% reported in fewer than half. Hospitals did better, at 52%. By country, the share of facilities filing every expected month was 83% in Ghana, 56% in Zimbabwe, 33% in Tanzania, 23% in Uganda and 14% in Kenya. A data element can exist in a national system and still be empty: Ghana’s preterm birth field was present throughout and returned zero for six straight years.
The adjustment, and what it leans on
Comparing facility returns with population estimates requires an arithmetic bridge, and the study builds it from institutional delivery shares: multiply by 1.16 for Ghana (86% of births in facilities), 1.14 for Kenya, 1.23 for Tanzania, 1.10 for Uganda, 1.18 for Zimbabwe. The authors state the assumption plainly — that outcomes are similar for births inside and outside facilities — and note it can fail in either direction, since women delivering at home may carry higher risk while referral hospitals attract more complicated pregnancies.
Those multipliers are where this audit touches a second running problem. The shares come from Demographic and Health Surveys, the 2022 rounds for four of the five countries and a 2019 Countdown report for Zimbabwe — the same survey stream whose contraction has been worrying maternal mortality modellers, a subject this digest examined in its reading of the MMEIG modelling runs without DHS inputs. Should those surveys thin out, the bridge between facility data and population rates loses a plank along with everything else.
Discrepancies also run both ways, and the study says so. Tanzania’s adjusted neonatal mortality rate, 32.0–48.4 per 1,000 live births, sits well above the UN IGME estimate of 21.1–23.5. The authors offer two explanations and do not choose between them: near-universal reporting may be capturing deaths that modelled estimates miss, or early deaths recorded elsewhere as fresh stillbirths may be classified as neonatal deaths in Tanzania — consistent with that country’s falling stillbirth series. Under an alternative aggregation weighting facilities by delivery volume, the excess shrank. Their broader caution is one this digest keeps returning to: the MMEIG and IGME comparators are themselves models with wide uncertainty that lean on surveys, “so discrepancies may reflect limitations in either source rather than DHIS2 alone”.
What a register can be asked to hold
The study’s closing argument is not that facility data are useless but that they are partial in a patterned way. “DHIS2 alone cannot provide a complete picture of maternal and newborn outcomes,” the authors write, and their prescription is to complement routine returns with civil registration, verbal autopsy and household surveys — three systems that in most of these countries are themselves still under construction, as Uganda’s piloted birth and death registration illustrates. None of the five countries had a single field for whether the care a baby received was any good.
Read as an evidence statement rather than a verdict, the paper is a descriptive secondary analysis of aggregate national data, with a purposive five-country sample that the authors say is not statistically representative of sub-Saharan Africa, no verification against facility registers, and no facility-level linkage. What it establishes is a geography of error: in these systems, a stillbirth stands a fair chance of being counted, a newborn death a variable one, and a maternal death — in four of the five countries, in most years — a poor one. Those are the returns on which national ratios are built, and the audit’s value lies in showing exactly where the count thins out.
Sources
- Akuze J, Otieno GP, Shabani J, et al. Data quality assessment of maternal and newborn indicators from District Health Information Software 2 in five African countries for the period 2014–2023. Journal of Global Health 2026;16:04388 pmc.ncbi.nlm.nih.gov
- WHO, UNICEF, Every Woman Every Newborn Everywhere. Every Newborn: an action plan to end preventable deaths. Geneva: World Health Organization, 2014 who.int
- WHO. Data quality review: module 1: framework and metrics. Geneva: World Health Organization, 2017 iris.who.int