October 5, 2026 interva.net Editorial

Verbal Autopsy and Drowning: One Study Out of 11,457

A scoping review in Injury Prevention screened 11,457 records and found just nine verbal autopsy studies on drowning in low- and middle-income countries — and only one that recorded the circumstances of the death.

verbal autopsyinjuriesdata qualityWHO

Verbal autopsy is built to answer a narrow question: what did this person die of? For a drowning, that answer is often already known to the family. What they cannot put into a single cause code is everything around it — whether the child was playing by the water or fetching it, whether anyone was watching, whether the pond had a barrier, whether the nearest clinic was an hour away. A scoping review published in the journal Injury Prevention set out to measure how much of that context the field’s standard instruments actually capture. Its answer is stark.

Eleven thousand records, nine studies

The team, led by Nibedita Ray-Bennett at the University of Leicester and drawing on co-authors across the UK, Bangladesh, India, Uganda, South Africa and the Netherlands, searched 14 electronic databases, three academic search engines and ten grey-literature sources for publications from January 2012 to September 2023. The search returned 11,457 records. After duplicates were removed, 7,453 were screened, 490 went to full-text review, and 127 met the review’s population–concept–context criteria.

Just nine publications cleared the final eligibility bar for verbal autopsy of drowning — seven peer-reviewed articles and two reports, drawn from six different countries. That is the finding under the headline: a method that is widely used to reconstruct deaths in places without reliable civil registration has barely been aimed at one of the world’s leading causes of injury death.

The instruments those nine studies used were a patchwork. Four teams built their own study-specific questionnaire. The rest leaned on five recognised variants: the WHO standardised VA instrument (2016), the WHO standard child VA questionnaire, the PHMRC shortened VA, an adapted WHO and INDEPTH Network tool, and a locally adapted Bangladesh Health Injury Survey questionnaire. Whatever the tool, the routine practice across studies was to fold drowning into an all-injury or all-cause category rather than interrogate it on its own terms.

The question that files an accident away

Only three of the nine studies used a VA instrument to look specifically at fatal drowning, and only one — Dandona and colleagues in India — reported on the causes and circumstances surrounding the death. That study drew on the open narrative section to describe how children aged 1 to 14 died: playing in water, bathing in water, and defecating.

The review hands the identification problem to the questionnaire itself. A widely used screening question — “Did he or she die from injury or accident?” — tends to catch immediate injury deaths, the authors note, but may miss deaths that occur up to 30 days after the event, which is the standard window for injury mortality. As the paper puts it, VA instruments so far perform well for injury deaths as one broad category, with high sensitivity, specificity and positive predictive value against hospital records and death certificates. That is a reassuring result for cause coding, and an incomplete one for prevention: a death recorded simply as an injury tells a programme nothing about the water, the supervision or the distance to care.

The review catalogues seven data-related challenges reported across the included work — from recall and reporting bias to misclassification, missing records, and interviews that can stretch to three hours when a death certificate exists. These are familiar complaints about the verbal autopsy method, and none is unique to drowning. What is new is the framing: for drowning, the missing detail is not a rare edge case but the point.

Twenty-four variables, and a module that does not exist yet

From the nine studies the reviewers distilled 29 minimum parameters — the variables that authors consistently collected. Three describe the data-collection process, two the victim’s basic demographics, and the remaining 24 concern the causes and circumstances of fatal and non-fatal drowning: the place and time of the incident, the water source and its distance from home, swimming ability, activity beforehand, the presence of a caregiver, early warnings, and healthcare availability among them.

Their recommendations are aimed squarely at the bodies that write the instruments. The open-ended narrative already sits inside the 2022 WHO VA tool as the natural home for context, and the review argues it could be strengthened by guiding interviewers with these variables. Going further, it proposes a drowning-specific module and the pairing of VA with social autopsy — the companion interview that captures social, behavioural and health-system determinants of a death — along with qualitative methods such as in-depth interviews and community walkthroughs, none of which has been applied to fatal drowning in the reviewed literature.

It is an unusual direction for a field that has spent recent years trying to make interviews shorter. A validation study in Kolkata tested an abridged questionnaire that cut the instrument to 54–61 questions per age group and brought interviews down to 30–35 minutes, as we reported earlier. The drowning review argues the opposite for one cause: that a standardised tool can be too thin to hold what a drowning death actually involves. Both can be true — the case for trimming rests on the questions that are rarely informative, and the case for adding rests on a domain where almost nothing has been asked.

What the review does and does not establish

This is a scoping review — a map of the literature — not a validation study of any instrument. Its conclusions about the value of a drowning module are a research agenda, not a measured result; no accuracy estimate for such a module exists because none has been built. The evidence base it surveys is thin by its own admission: only one included study captured circumstances, the search was restricted to English-language primary studies from 2012, and grey literature was drawn from just three LMICs, so the authors caution the findings “may not be entirely representative.” Five experts from the Public Health Foundation of India, the INDEPTH Network and the Centre for Injury Prevention and Research, Bangladesh were consulted to corroborate the results.

The stakes are not marginal. The WHO’s drowning fact sheet counts roughly 300,000 drowning deaths a year, 92% of them in low- and middle-income countries, with children under five accounting for nearly a quarter — and drowning responsible for 75% of deaths in flood disasters. When a household death is coded, the downstream machinery that turns it into a cause is typically algorithmic; InterVA, the model developed at Umeå University, and other tools in the openVA family assign probabilities from a structured set of signs and symptoms. An instrument that never asks about the circumstances of a drowning will not hand that story to the algorithm, and the algorithm cannot recover what was never collected.

For a country trying to build a drowning-prevention programme out of its mortality data, that blank is not a coding problem. It is a programme with nothing to aim at.

interva.net is an independent digest. It is not affiliated with, endorsed by, or speaking for the InterVA project, Umeå University, or the World Health Organization; all named tools and standards are attributed to the organisations that produce them.

Sources

← Back to the news digest