Thirty-Seven Hours After a Death in Bushbuckridge
A process evaluation of community-based minimally invasive tissue sampling in rural South Africa counts what stands between a household death and a tissue sample — the operation behind the reference standard that verbal autopsy accuracy claims rest on.
Every accuracy figure for a verbal autopsy method is a comparison against something. Often that something is a physician reading the same interview; sometimes it is a hospital file; in the strongest studies it is tissue — needle cores of brain, lung and liver, blood, cerebrospinal fluid, taken from a body within hours of death, in a household that has just lost someone. The comparison is the part of the method nobody photographs.
A preprint posted to medRxiv on 10 November counts it. Zokwane Mondlane, Kathleen Kahn and colleagues at the SAMRC/Wits Agincourt unit describe Vukamberi — “to examine” in Xitsonga — a prospective study that performed minimally invasive tissue sampling, or MITS, on community deaths in Bushbuckridge, a rural sub-district of Mpumalanga, South Africa, over 24 months from July 2024. What they report is a process evaluation: reach, adoption, fidelity, feasibility, acceptability and timeliness, read through the Medical Research Council’s complex-interventions framework and Proctor’s implementation outcomes, reported to the StaRI standard. No causes of death are given. This is a paper about whether the sampling can be done at all, in a setting where the Agincourt health and demographic surveillance site has followed roughly 120,000 people for three decades and interviews a family after every death it records.
Four hundred and eight, then two hundred and six
Of 408 community deaths notified to the project, all 408 were screened, 316 (77.5%) met the eligibility criteria, and families consented in 206 cases (65.2%). Every consented case went ahead — procedural completion 206 of 206 — and sampling adhered to the standard operating procedure for the target organs in 199 (96.6%).
The specimen record is specific enough to be checked. Formalin-fixed tissue came from the brain in 203 cases, the right lung in 205, the left lung in 204 and the liver in 204; all four core organs together in 201. EDTA blood was collected in 203, blood-culture bottles in 201, cerebrospinal fluid in 180. Nasopharyngeal swabs were taken in 205 cases and have not been tested, on cost grounds. Kidney biopsy was attempted in every case but succeeded in 29 on the left and 24 on the right, because field time ran out against the core organs. Rectal swabs were dropped at the protocol stage: the budget could not cover the multiplex testing needed to separate pathogens that contributed to a death from ordinary gut flora.
Two of those numbers moved a lot between years, and both moved in the same direction. Eligibility rose from 65.4% in the first year to 87.2% in the second; consent from 58.8% to 69.0%. The authors attribute the climb to a notification network that was being built while it was being used.
Why the window is forty-eight hours
Among the 180 enrolled cases with usable timestamps, the median interval from death to notification was 15.2 hours, and from death to completion of the procedure 37.3 hours. Twenty-six deaths, 6.4% of those notified, arrived after the 48-hour cut-off and were excluded — mostly in the early months, but four as late as May 2026. Remote geography, deaths overnight and bodies found late all appear in the field notes.
The window itself is not a laboratory constant, and the paper is explicit about it. Forty-eight hours was chosen because ambient temperatures in the sub-district accelerate decomposition, most households have no refrigeration, and elders in the community customarily wait two to six hours after an apparent death before calling a mortuary — a practice the authors link to the possibility of what families describe as a miraculous recovery. Whether the diagnostic performance of MITS holds for tissue sampled at a median of 37 hours, and up to 72, is not yet known; the cause-of-death analysis is where that question gets answered.
Ineligibility was otherwise unremarkable. Twenty-seven deaths had occurred in hospital and were out of scope by design, eight were outside the catchment, three were unnatural, three had decomposed beyond safe sampling, and two were already under forensic jurisdiction or had had a complete autopsy. For 23 cases no reason was recorded at all.
A chief’s restriction, and consent moved to the mortuary
Among the constraints the team could not design around: from August 2025 a local traditional leader restricted study access to 14 villages in the catchment. The adaptation was to approach families at mortuaries instead, which required formal service-level agreements with mortuary partners and a sampling schedule built around peak mortuary operations. Three mortuary owners had refused before consent was even sought, which prompted those agreements to be tightened.
Elder household members took part in consent discussions in 277 of the 316 eligible cases (87.7%). The team’s earlier qualitative work in the same community describes why: in Xitsonga- and Sepulana-speaking households, elders hold authority over whether a body may be examined, and younger relatives defer to them on decisions made around the time of death even where they do not defer in ordinary matters. Two field workers, one male and one female, approached each family together, in the family’s language, with pictorial consent aids and an explanation that the procedure does not disfigure.
What a refusal actually was
Fifty-three refusals account for 48.2% of the 110 eligible deaths that were not enrolled. The dominant driver in the field notes was not objection to the procedure but the timing of the question: families in the first days of bereavement were not yet able to engage with it. Traditional beliefs about bodily integrity and disturbance of the deceased’s spirit appear in a subset. In several cases families described the death as expected and complete — one record has a family saying the deceased “was old, they had accepted the outcome and did not want her to be part of research”. Two families asked not to be contacted again. The authors’ reading is that few refusals were principled opposition and most were a mismatch between a 48-hour operational window and a household’s capacity to answer inside it. Refusals clustered at five a month in January–March 2025, when enrolment was extended into newly engaged villages, then fell to four or fewer as familiarity grew — which the team takes as an argument for completing community sensitisation before opening a new area, not during.
Sixty-five per cent is lower than the 82% consent reported when MITS was offered to relatives of deceased children in a Mozambican hospital, where 79% of cases went on to the procedure. It is also lower than the hypothetical acceptance recorded in Agincourt before the study began, when 20 in-depth interviews and six focus groups with residents, health workers, traditional healers, religious leaders and mortuary staff found broad willingness — conditional on minimal disruption to the body and to burial, and dented by rumours of organ trafficking. That study found acceptability was weakest for deaths of infants and of traditional healers, both of which disrupt specific burial obligations.
The twenty-five families who said yes
The single largest disruption had nothing to do with consent. Between January and mid-March 2026 a customs clearance delay left the project without MITS kits, and 25 eligible deaths could not proceed. Every one of those families had been engaged and was willing. Enrolment, which had reached 12–26 procedures a month between June and November 2025, fell away and recovered to 14–15 a month from April 2026, after emergency procurement and buffer stock. February 2026 is worse still in the record: only two mortuaries reported that month, six deaths were notified, and the authors concede that ascertainment of deaths in that period was incomplete even though screening of what arrived remained at 100%.
The interruption matters beyond the inconvenience. A surveillance series gains its value from being a sample of deaths rather than a sample of convenient deaths, and a two-month block missing for procurement reasons is missing systematically, by season, with no relation to the characteristics of the people who died in it.
Cost is the other half of that vulnerability. A four-site cost evaluation in sub-Saharan Africa and South Asia put the direct cost of a single MITS case at between 609 and 1,028 US dollars, driven by labour, sample testing and supplies — before the pre-implementation investment in laboratory capacity, transport and training that sites describe as substantial and highly variable. Kits are, in that light, the cheapest component and the easiest to lose.
Who the cohort is not
The enrolled series was 197 adults (95.6%) and nine children aged 4 to 16 (4.4%). There were no stillbirths, no neonatal deaths and no infant deaths, although the protocol was open to every age from the start. This was not a design choice. Two eligible infant deaths were notified and lost because the families planned burial at sunset on the day of death; the authors read the absence of infant enrolments as lower community acceptability for that age group plus a burial practice that leaves no window. The formative study had predicted exactly this.
That composition deserves attention, because the cohort’s purpose is to validate interview-based cause of death, and MITS’s own accuracy is not uniform across ages. In Maputo, comparing MITS against complete diagnostic autopsy in the same bodies, the adult series reached 75.9% concordance on cause of death (85 of 112), higher for infections and cancers than for other diseases; the paediatric series reported a kappa of 0.70 for disease category and ICD-10 diagnoses coincident in 75% of cases (36 of 48). Adults are also the group where verbal autopsy has the hardest time, since symptoms overlap between chronic conditions. A validation dataset that is 96% adult is measuring the reference standard where it is least certain, on the deaths where the interview is least specific — and as the CSMF metric’s own arithmetic shows, neither error cancels the other out.
The cause-of-death work is also, at the reporting cut-off, unfinished by design. Culture results were finalised for 108 of 206 cases, histology for 100, verbal autopsy interviews for 143 (deliberately deferred to respect the mourning period), and the multidisciplinary determination panels for 100 (48.5%). Ante-mortem clinic records could be retrieved for 50 of the 92 decedents who lived inside the surveillance area; for the other 114 no record exists to link.
What a process evaluation can and cannot say
The paper is candid about its own limits, and they are the ordinary ones. Implementation outcomes were abstracted from monthly summary reports rather than case-level records, so refusal reasons cannot be broken down by the decedent’s age, sex or how long they had been ill. Notified deaths were never compared against deaths the surveillance site recorded independently, so the completeness of ascertainment — the denominator behind every rate here — could not be quantified. It is one site, and the team evaluated its own programme, which prospectively collected operational records mitigate but do not remove.
Vukamberi is nonetheless an unusual document. MITS has been used mostly on stillbirths and child deaths, largely in hospitals, and the published implementation literature is thin on all-age community sampling. Its conclusion is that the procedure is feasible in this setting, and that what made it feasible was a set of systems around it: incentivised community informants, mortuary agreements, elder-mediated consent, an in-house culture bench, buffer stock. Africa CDC’s continental framework on mortality surveillance asks countries to bring community death reporting into national health information systems; this is a district-level account of what that integration actually consumes.
For anyone reading a verbal autopsy accuracy estimate, the practical implication is narrow but real. The number is a comparison against a cohort, and that cohort was assembled by a funnel — 408 notifications, 316 eligible, 206 consents — shaped by grief, burial custom, one chief’s decision, and the arrival date of a shipment of kits. None of that makes the comparison invalid. It does mean the reference standard has a supply chain, a consent conversation and a selection problem of its own, and that the deaths it never reached are not a random subset of the dead.
Sources
- Mondlane ZL, Kahn K, Chipungu S, Wagner RG, Tippett Barr B, Price J. Implementing community-based minimally invasive tissue sampling for mortality surveillance in rural South Africa: A process evaluation of the Vukamberi project. medRxiv preprint, posted 10 November 2026 — not peer reviewed medrxiv.org
- Brandt L-L, Mathebula G, Mondlane ZL, et al. The acceptability of minimally invasive tissue sampling for cause of death determination in rural South Africa: A qualitative analysis. PLOS One 2026;21(5):e0345107 pmc.ncbi.nlm.nih.gov
- Castillo P, Martínez MJ, Ussene E, et al. Validity of a Minimally Invasive Autopsy for Cause of Death Determination in Adults in Mozambique: An Observational Study. PLOS Medicine 2016;13(11):e1002171 pmc.ncbi.nlm.nih.gov
- Bassat Q, Castillo P, Martínez MJ, et al. Validity of a minimally invasive autopsy tool for cause of death determination in pediatric deaths in Mozambique: An observational study. PLOS Medicine 2017;14(6):e1002317 pmc.ncbi.nlm.nih.gov
- Munguambe K, Maixenchs M, Anselmo R, et al. Consent to minimally invasive tissue sampling procedures in children in Mozambique: A mixed-methods study. PLOS One 2021;16(11):e0259621 pmc.ncbi.nlm.nih.gov
- Morrison LTR, Brown EG, Paganelli CR, et al. Cost Evaluation of Minimally Invasive Tissue Sampling (MITS) Implementation in Low- and Middle-Income Countries. Clinical Infectious Diseases 2021;73(Suppl 5):S401-S407 pmc.ncbi.nlm.nih.gov
- Adair T, Rajasekhar M, Bo KS, et al. Where there is no hospital: improving the notification of community deaths. BMC Medicine 2020;18:65 pmc.ncbi.nlm.nih.gov
- Africa Centres for Disease Control and Prevention. Continental framework on strengthening mortality surveillance systems in Africa. Addis Ababa: Africa CDC, 2023 africacdc.org