October 11, 2026 interva.net Editorial

The Deaths Outside the Register: A National Phone Survey Tests Bangladesh's Mortality Coverage

A nationwide mobile phone survey in Bangladesh recorded 1,527 household deaths and asked where each one happened and where the body was buried — finding that roughly a third of urban deaths and half of rural ones would sit outside facility- or cemetery-based surveillance. A methods study, and its authors are explicit about what it cannot yet prove.

mortality datasurveillanceCRVSverbal autopsy

Between 22 December 2021 and 31 July 2022, interviewers working from a call centre at Bangladesh’s Institute of Epidemiology, Disease Control and Research dialled 506,659 mobile numbers. Roughly five in a hundred of those calls ended in a completed interview — 22,731 of them, spread by design across twenty-four strata of age, sex and residence, from the city corporations of Dhaka and Chittagong to villages in the countryside.

Buried in the questionnaire was a short sequence of questions that mortality surveys rarely ask: had anyone in the household died since 2019; and for each death, did it happen at home or in a health facility, and where was the body buried — a communal cemetery, a family plot, somewhere else? Add one more question, about whether the deceased had been diagnosed with COVID-19 before dying, and you have the raw material for an unusual kind of arithmetic. The study, published in June in Tropical Medicine & International Health by Aashish Gupta and colleagues, uses those four variables to work out which deaths a health system would have recorded, and which it would never have seen at all.

Three doors through which a death can enter the record

The authors define the competition carefully. A death can be picked up by all-cause mortality surveillance built inside health facilities, which counts inpatient deaths through routine information systems. It can be picked up by surveillance at the places where bodies are handled — morgues, cemeteries, even coffin workshops — where registers already exist and staff can be posted to transcribe them, or where newer tools such as satellite imagery watch burial grounds grow. Or it can be picked up by a case-based system set up for one epidemic disease, which counts deaths among people who were diagnosed before they died.

None of the three is complete, and each fails in a characteristic way. Facility-based counting misses everyone who dies at home; the authors cite work estimating that large segments of populations in low- and middle-income countries die outside hospitals, and the verbal autopsy explainer on this site covers why those community deaths are exactly the ones that need an interview to be classified at all. Cemetery-based systems miss family plots and rural graveyards. Case-based systems demand laboratory capacity, and lose people who were never tested or who were diagnosed and then died at home without follow-up.

What the Venn diagrams show

The survey’s own numbers set the scene. Respondents reported 1,527 deaths since 2020: 321 in the large city corporations, 617 in other urban areas, 589 in rural areas. Most had occurred at home, and the rural-urban split is stark — 72.5 per cent of rural deaths versus 53.3 per cent of deaths in Dhaka and Chittagong. Only 99 of the 1,527, or 6.5 per cent, had a pre-mortem COVID-19 diagnosis, and that proportion ran from 11.2 per cent in the city corporations down to 3.2 per cent in rural areas.

Plotting place of death, place of burial and COVID-19 diagnosis as overlapping circles produced the paper’s central result. In the city corporations, about two-thirds of reported deaths would plausibly also have been recorded by a facility-based or cemetery-based system. In other urban areas, again roughly two-thirds. In rural areas, only about half. Read from the other direction, the share of deaths that only the phone survey reached was 32.4 per cent in Dhaka and Chittagong, 36.3 per cent in other urban areas, and 49.6 per cent in rural Bangladesh.

That is a large number, and the paper immediately qualifies it. The comparison is not against surveillance systems as they exist, but against the systems that could be built: the analysis assumes that every death in a facility and every communal burial would be captured. The authors call this “a strong assumption” and note that real systems suffer from imperfect record-keeping, thin staffing and partial coverage of facilities and burial grounds — so in practice the phone survey’s unique share might be higher still.

In the cities, the deaths that only the survey caught had a profile. They were less likely to be female (adjusted odds ratio 0.49, 95% CI 0.26–0.88) and more likely to be old: 53.1 per cent of deaths at ages 80 and over were phone-survey-only, against 25.7 per cent at ages 25 to 59 (aOR 2.83, 1.30–6.45). In other urban areas, deaths at 15 to 24 stood out sharply (aOR 7.90, 1.43–43.6), as did deaths above 60. Rural deaths showed no such pattern. The authors read the urban results as pointing to people less connected to the health system — men, young adults, the very old — and to the possibility that their deaths, absent a phone, leave nothing behind.

Counting deaths is not naming causes

Everything in this study is a denominator. It counts deaths and their circumstances; it assigns no cause to any of them, and it does not deploy any of the algorithms — InterVA, InSilicoVA, Tariff — that the model page here describes. That distinction matters for how far the findings stretch. The WHO’s 2022 verbal autopsy instrument is designed to be administered in person by a trained interviewer after a death; sample registration systems and health and demographic surveillance sites do that work. Where those systems have no reach, neither death counts nor cause-of-death estimates exist, which is the mechanism behind the coverage-versus-cause arithmetic that Zambia’s digital death-notification pilot ran into from the registration side.

The authors are also candid about the study’s limits, and they are worth taking at face value. Their case-based comparison rests on proxy reports of COVID-19 diagnoses given by relatives, vulnerable both to stigma-driven silence and to home rapid tests never reported to anyone. A phone survey cannot reach households with no phone, and the paper notes that those remaining households may cluster in the poorest, highest-mortality areas. Random digit dialling yielded a small response rate by design and by attrition. Deaths in single-person households are invisible to a survey that asks survivors, as are deaths that dissolved a household. Respondents may misstate ages, dates, or include relatives who did not live with them.

What would settle the question is record linkage: matching phone-survey deaths against the health and demographic surveillance sites that operate in several districts of Bangladesh, against burial surveillance in Dhaka and Chittagong, and against the national sample registration system. Until that is done, the paper’s own framing is the right one — a proof that remote surveys can reach deaths other systems would not, and an argument for testing that claim against real records rather than hypothetical ones.

The pandemic-specific findings are the ones most likely to date. Bangladesh’s case-based surveillance recorded 29,500 COVID-related deaths, or approximately 17.3 deaths per 100,000 person-years, per the paper’s account; a rural surveillance site saw excess concentrated among the oldest residents; the national sample registration system recorded a dip in life expectancy in 2020 that rebounded for men and not for women. But the structural point is not about COVID-19. It is that a country of more than 170 million people was reachable by phone at scale, and that the deaths those calls heard about were disproportionately the deaths the paper trail left out.

Sources

← Back to the news digest