October 10, 2026 interva.net Editorial

281 and 104: The Same Mothers, Counted Two Ways in Tanzania

A 27-year verbal autopsy series in one Tanzanian surveillance site puts maternal mortality at 281 deaths per 100,000 live births for 2015–2022; the national survey implies 104. A WHO Bulletin paper proposes four cheap checks for the facility data in between — and finds plenty of implausible numbers.

maternal mortalityverbal autopsymortality dataCRVS

In the Magu health and demographic surveillance site, in the Mwanza region of north-west Tanzania, the death of a woman aged 15 to 49 does not simply enter a register. A clinical officer travels to the household — typically about four months after the death — and works through a structured questionnaire with whoever was present at the end: symptoms, timing, whether she was pregnant or had recently delivered. Then comes the part no coded variable captures, a blank section where the family’s account of the final weeks is written out in narrative.

That routine has run for nearly three decades. Between 1995 and 2022 the site logged 940 deaths and 177,729 person-years of observation among women of reproductive age, and interviewed families about most of them. The questionnaires changed three times: a project-specific form from 1995 to 2002, the INDEPTH standard from 2003 to 2007, and the WHO standard from 2008 onwards. The resulting series, published in Global Health Action by Marston and colleagues, is now one of the longest continuous records anywhere of what kills women in a rural African population — and of what verbal autopsy can and cannot establish.

The fall, and what drove it

All-cause mortality among women aged 15 to 49 peaked at roughly 9 deaths per 1,000 person-years in the late 1990s, fell by 60% through to 2014, then flattened at about 3.5. One cause did almost all of the heavy lifting: HIV/TB mortality dropped from 4.46 per 1,000 person-years in 1995–2001 to 1.09 in 2016–2022, accounting for 61% of the total decline. Across the whole period, infections still made up the largest share of deaths at 54%, with non-communicable diseases at 29% — a pattern that has been quietly rebalancing in Tanzanian mortality data for years.

Maternal deaths behaved differently. The maternal mortality ratio fell before 2010 and then stalled; for 2015–2022 the site estimates 281 deaths per 100,000 live births, with a sensitivity analysis spanning 205 to 351. Relative to all deaths in the age band, maternal causes were about as prominent at the end of the series as at the start.

Splitting the ratio by where women died sharpens the picture. Using the distributions of deaths and births for 2016–2022 — roughly 70% of maternal deaths and 84% of live births occurred in health facilities — the authors calculate an institutional ratio of 234 per 100,000 and a community ratio of 526. Two different problems sit behind one national number: the quality of care inside facilities, and deliveries at home under high-risk circumstances. In the most recent period, 46.5% of women of reproductive age still died at home.

The team also went back to the raw narratives for 2015–2022, with two doctors reading the recorded signs, symptoms and family accounts. Postpartum haemorrhage appeared seven times, eclampsia twice, and anaemia, ectopic pregnancy, obstructed labour and induced abortion once or twice each; outside direct obstetric causes there were two HIV deaths, two pulmonary embolisms and three cases the reviewers could not determine. Almost every woman had reached a health facility at some point in her terminal illness.

Where the method stops

The study is candid about its instrument. There is no medical certification of cause of death in this setting; verbal autopsy rests on the recall of grieving relatives, and the authors describe it as a crude tool that does not resolve detailed causes. Three successive questionnaires, incomplete coverage of deaths over time, and the assumption that women without an interview died in the same proportions as those with one all sit under the numbers. Causes were therefore analysed in broad groups only, to limit misclassification. The cause assignments came from InSilicoVA, a Bayesian model distributed in the openVA R package — a different algorithm from the InterVA model developed at Umeå University, though both are probabilistic routes from interview data to a cause distribution. The output is a modelled population estimate, not a certified cause for any individual woman.

Four checks on the paper ledger

That is one ledger. The other is the one hospitals keep. In a paper in the Bulletin of the World Health Organization, Makuwani and colleagues start from the shift that makes facility records matter more each year: in Tanzania, the share of births in health facilities rose from 65% in 2015 to 81% in 2022. If most births — and most maternal deaths — happen in institutions, institutional data should be able to carry the national estimate. They often cannot.

Working from two workshops held in 2024 under the Countdown to 2030 initiative, the authors propose four simple plausibility checks on facility maternal mortality data: whether reported levels and subnational rankings make sense; the ratio of stillbirths to maternal deaths; the distribution of causes of death; and the share of deaths among women aged 35 and over. A district with a hospital reporting fewer than 25 maternal deaths per 100,000 live births is treated as implausible, and a stillbirth-to-maternal-death ratio outside 6 to 15 is read as a signal that one of the two counts is being missed.

Applied to Tanzania, the checks flagged a lot. In DHIS2, the routine health information system, 37 of 184 districts — 20.1% — reported institutional ratios low enough to fail the plausibility test, against 16 of 184 (8.7%) in the parallel maternal and perinatal death surveillance and response system. Twenty-two of 26 regions reported a stillbirth-to-maternal-death ratio below 6, which the authors attribute largely to stillbirth underreporting rather than to unusually safe deliveries. And where institutional birth rates were highest, institutional ratios were also highest — the reverse of the expected ranking, a sign that referral, recording and reporting practices are shaping the pattern as much as risk is.

Two numbers, one country

Set the ledgers side by side and the arithmetic does not settle. For 2015–2022 the surveillance site gives 281 per 100,000. The 2022 Tanzanian demographic and health survey implies 104 for the same window, and United Nations estimates put the country near 238 for 2020. The WHO’s global picture for 2023 — about 260,000 maternal deaths, a global ratio of 197, roughly 70% of deaths in sub-Saharan Africa — is built from exactly this kind of contested national input.

The Magu authors argue that one figure inside the survey estimate is implausibly low: only 6.1% of deaths among women of reproductive age were recorded as occurring during pregnancy or the postpartum period, against 13.5% in the surveillance site, where the narrative review corroborated the pregnancy status. Their sensitivity analysis, which assumes that every un-interviewed death was non-maternal, still does not fall to the survey’s level. They also note the obvious caveat in the other direction — a single surveillance site, however representative of mainland Tanzania, is not a national sample.

The Bulletin paper adds a detail that matters for the next few years: verbal autopsy has been piloted within civil registration and DHIS2 in selected Tanzanian regions, but reliable information on maternal mortality from that channel is not yet available, while the Ministry of Health has begun large-scale verbal autopsy in two regions. Two measurement systems are being scaled up at the same time, each with its own failure modes — missed stillbirths and under-reporting in the facility ledger, recall and coverage gaps in the interview ledger.

That disagreement is not noise to be averaged away. It is the finding. The Sierra Leone cohort we covered last week showed the same thing from another angle: clinical classification of hospital deaths and model-coded verbal autopsies produced visibly different cause lists for the same population, and no reference standard existed to arbitrate between them. Where a maternal death is counted, and by which instrument, is now one of the largest single sources of uncertainty in how the world measures it.

Sources

← Back to the news digest