The Interviewer's Share

Tanzania has moved verbal autopsy out of research trials and into routine death registration, with ward officers knocking on doors and consent hard-wired into a tablet. The empirical literature on what that visit costs the person asking comes almost entirely from research sites — and it describes moral distress, not burnout.

verbal autopsyfield studiesCRVSglobal health

In Iringa Region in 2019, Tanzania’s Ministry of Health expected 7,294 deaths in a population of 1.1 million and managed to interview a bereaved family for 1,659 of them — under one in four. Across the districts of that demonstration phase the verbal autopsy completion rate ran from 7.3 per cent to 73.5 per cent, averaging about 30. The reason the ministry cares about the gap is obvious: in a country where around 70 per cent of deaths happen outside a health facility, health planning had been resting on data that represented perhaps a tenth of the population.

The reason the gap matters for anyone who runs interviews rather than registries lies in a single line of the paper’s ethics statement. The work, it says, was conducted as routine government operations under the civil registration system, following national policies — not as a research protocol under a committee.

A knock at the door, on the government’s account

The Tanzanian programme, described in BMJ Global Health in March by Isaac Lyatuu, Mahadia Tunga, Sigilbert Mrema and colleagues from the Ministry of Health and the Ifakara Health Institute, is one of the clearest accounts in print of verbal autopsy becoming ordinary civil-service work. The sequence went from a ten-ward pretest in 2018 (358 interviews against 934 expected deaths, 38.3 per cent) to a full regional demonstration in Iringa, then to national scale-up, which is still under way.

Two design choices carry the institutional argument. The first is who conducts the interview. In the pilot the ministry used community health workers, most of them volunteers, paid monthly for each autopsy. For the demonstration it replaced them with ward executive officers — salaried government employees — explicitly to make the activity sustainable rather than project-shaped. The second is how consent is handled: a summary of the consent form sits on the tablet, and the questionnaire has been programmed so that the symptom sections only unlock once consent is recorded; if it is refused, the form terminates. Interviewers may reschedule up to three times. The visit is meant to come after a mourning period of about a week, inside a window running from 30 days to six months after the death.

That is a functioning administrative process, and the paper is candid about how thin it can be. Completeness looked to the authors like a symptom of a perception problem: the structured interview still read to district staff as a research activity rather than a health-system function, and in remote wards reaching a household could take more than four hours of travel. Interviewer attrition is listed among the operational constraints, in the same sentence as the cost of supervision and device maintenance. But attrition is a management metric. What the interviewer carries home is not.

What the research sites found

The most detailed evidence on that question comes from Kenya, in a 2021 case study by Alex Hinga, Vicki Marsh, Amek Nyaguara, Marylene Wamukoya and Sassy Molyneux, then at the KEMRI-Wellcome Trust Research Programme in Kilifi. They worked across the Kilifi health and demographic surveillance site on the coast — about 280,000 people — and the Nairobi Urban site, which follows roughly 89,000 residents of Korogocho and Viwandani, and extended the enquiry with interviews at surveillance sites in Ghana, Malawi, South Africa and Uganda. In total 115 people took part: 29 staff in individual interviews, and 86 community members and field workers in twelve focus groups.

Their analytical move was to stop calling the interviewers’ distress a welfare issue and treat it instead as moral distress — the version of the concept in which a person knows what the right thing to do is, and institutional constraints make it near impossible to do it. The distinction is not academic. Grief in a respondent is grief. The interviewers’ sadness, guilt and helplessness were tied to specific blocked actions: they believed they should leave a token of condolence with a bereaved household and could not, because only two of the seven sites had such a policy and even there funding sometimes stopped it at the door; they believed they should refer a family to health or social services and could not, because the surveillance site sat outside those systems.

One interviewer described the arithmetic of a working day: “In a day, you may get two or three cases, which you cannot forget.” Another described the moment the script stops working: “sometimes I feel like I’m lying, because the person has died and if there’s still an illness in the household, no one will come in to help them.” A researcher at one site pushed back on the obvious remedy — sending interviewers on counselling courses — with the observation that “badly done counselling is probably more harmful than no counseling”, given that a household visit has a fixed and limited length.

The study’s recommendations read as an organisational job description, and none of them are in a questionnaire: recruit for diversity of gender, age and religion so that an interviewer can draw on shared standing in the community; train in bereavement counselling properly, not in-house; hold regular ethics reflection sessions with agreed follow-up actions; and set up a route to consult ethicists and mental health specialists about moral distress. The authors also argue for compensating respondents for their time and for reviewing, site by site, whether a full verbal autopsy is needed for every death — noting that causes like road traffic injury, suicide or shooting are often already documented by community leaders, health volunteers or facility records, and are precisely the deaths that make an interview hardest.

The two ledgers do not balance

On the respondent’s side, the paper is unflinching about asymmetry: the people bearing the heaviest burden, it concludes, gain the least. Timing is the sharpest illustration. Kilifi planned its interviews at least three weeks after a death, but 73 per cent were actually conducted three months or more afterwards, while Nairobi ran 58 per cent within three months — a schedule set as much by field logistics and the need to secure the right respondent as by mourning norms. In the Nairobi site, 177 of 5,495 interviews concerned a death by shooting, a category where questions about what the family did or did not do can land as blame. And the benefit that seems most obvious from outside — telling a family what the relative died of — runs into the instrument’s own imprecision: as one researcher put it, a probable cause that the family could have guessed is not much of a comfort.

None of this is an argument against verbal autopsy, and nothing in either paper measures its accuracy. The WHO 2022 verbal autopsy standard remains a questionnaire: consent, condolence, supervision and support are left to whoever implements it. That is the seam. A research site has an ethics committee, a named investigator and a protocol that can be amended. A registration office has a budget line and a supervisor who is measured on completeness. Tanzania’s own data quality work — including the four real-time indicators proposed by Lyatuu and colleagues last year — can tell a manager whether interviews are being conducted properly, and the evidence on recall periods can tell them when to schedule them. Neither tells them what to do for the person who has just spent an hour in a room with a mother who lost a child.

That is the honest state of the evidence, and its shape is worth stating plainly. The Kenyan study is qualitative: 115 participants, purposive sampling, six surveillance sites and one country’s two in-depth cases. It establishes that moral distress exists among verbal autopsy interviewers and identifies what produces it. It does not establish how common it is, whether it changes staff retention, or whether any of the recommended supports work — the authors call for exactly that research. Five years on, with verbal autopsy moving from surveillance sites into civil registration offices, the intervention still has no evaluation behind it.


interva.net is an independent digest. It is not affiliated with the InterVA project, Umeå University, or WHO, and nothing here is medical or policy advice. Evidence levels: the Tanzania paper is a descriptive account of a phased government implementation, externally peer reviewed, with no comparison group and no accuracy claim; cause-of-death concordance figures it reports (R² of 0.83 for InSilicoVA, 0.70 for InterVA-5, 0.47 for Tariff against physician coding) are agreement between coding methods, not validation against a reference standard. The Kenyan study is empirical bioethics — a qualitative case study with purposive sampling and no prevalence estimate. The WHO instrument is a normative standard. No claim is made here about the accuracy of verbal autopsy or of any algorithm.

Sources

  1. Lyatuu I, Tunga M, Mrema S, et al. Causes of death from community settings: insights from verbal autopsy implementation in Tanzania. BMJ Glob Health 2026;11:e019790
    pmc.ncbi.nlm.nih.gov
  2. Hinga A, Marsh V, Nyaguara A, Wamukoya M, Molyneux S. The ethical implications of verbal autopsy: responding to emotional and moral distress. BMC Med Ethics 2021;22:118
    pmc.ncbi.nlm.nih.gov
  3. WHO — 2022 verbal autopsy instrument (V2.0, multilingual)
    who.int

← Back to the news digest