The Deaths the State Cannot Count

Executive Summary. Bangladesh has reduced mortality and recently improved the administrative record. UNESCAP's current Bangladesh profile reports that 59% of deaths were registered in 2025, above the national 50% target for 2024 and far above the 7% recorded in 2014. Among deaths recorded in the health sector, 97% had a medically certified cause in 2025, above the earlier 80% target. These are major gains. They do not complete the mortality system: the certification denominator excludes deaths outside the health sector, 41% of estimated deaths remained unregistered, and UNESCAP says Bangladesh still does not produce disaggregated vital statistics directly from civil-registration data. The policy opportunity is a national mortality ledger that joins civil registration, health certification, and statistics, then publishes completeness and cause quality by place and population every year.

Every death changes a household. It may also change an inheritance, a pension, a land claim, an insurance payment, a school record, or the number of dependants a family must support. For the state, each death should change the denominator used to calculate disease risk, hospital need, road safety, and disaster loss. When the event never becomes a usable record, both the household and the state operate with a missing fact.

Bangladesh does produce mortality statistics. The Bangladesh Bureau of Statistics publishes vital statistics, and international series model national mortality. Those products matter. But modelled mortality and registered mortality answer different questions. A model estimates how many deaths probably occurred. A civil record establishes that a particular person died at a particular place and time. A medically certified cause connects that event to prevention and treatment. Substituting one for another hides the weakest part of the system.

The gap is not death estimation, it is death identification

The World Bank's WDI series shows a large mortality transition. Bangladesh's crude death rate fell from 11.962 per 1,000 people in 1990 to 5.105 in 2024. Under-five mortality fell from 146.5 per 1,000 live births to 30.5 over the same span. These are major public-health achievements. They are also modelled or harmonised population statistics, not a list of individual death certificates.

Source: World Bank World Development Indicators, crude death rate and under-five mortality, 1990 and 2024.

That difference is easy to miss because both products end in a number. Yet only the administrative record can tell a district health manager that deaths from a particular cause are rising in a particular catchment. Only a linked record can show whether women, landless families, people outside facilities, or remote communities disappear more often from the count. A model can repair a national total. It cannot identify the unregistered household or certify why a person died.

The state therefore faces a paradox. It knows enough to describe the long decline in mortality, but not enough to manage the remaining deaths case by case. The next gain is likely to be harder than the last. When mortality was high and infectious risks were widespread, broad interventions could move national averages. As the burden becomes more varied, prevention depends on accurate local causes. The information system must become more specific as the health problem becomes more specific.

Registration crossed its target, but not the finish line

UNESCAP's CRVS country profile gives the clearest current public benchmark for the administrative gap. Death registration rose from 7% in 2014 to 20% in 2017, then reached 59% in 2025. Bangladesh therefore crossed the 50% target set for 2024. The result changes the diagnosis from stalled registration to incomplete registration and weak public disaggregation.

Source: UNESCAP Bangladesh CRVS country profile, current 2025 outcome and historical benchmarks.

The correct response is neither to declare failure nor to assume success. It is to publish the denominator, numerator, method, and coverage. How many deaths were estimated to have occurred? How many were registered within the required period? Which register supplied the numerator? Were duplicates removed? How was completeness estimated where the population denominator was uncertain? Which districts improved, and which did not?

Without those answers, the national aggregate is not operational enough. A registrar cannot use it to find a district backlog. A health director cannot use it to identify a local blind spot. A citizen cannot use it to test unequal access. The profile itself says Bangladesh does not yet produce disaggregated vital statistics directly from civil-registration data. Publishing a national success is valuable, but it cannot substitute for the local evidence needed to guide resources and command trust.

Bangladesh's multi-agency structure makes publication more important. Civil registration, the health system, statistics, and legal records do not sit in one office. Coordination cannot depend on informal exchange. A public table forces the agencies to agree on identifiers, definitions, update cycles, and error correction. Publication is not the last step after integration. It is the discipline that makes integration real.

A name without a cause is only half a mortality system

Registering a death is necessary. Certifying its cause is what turns the record into health intelligence. UNESCAP reports that medically certified causes accompanied 2.5% of health-sector death records in 2017, rising to 19.6% in 2018 and 97% in 2025. That exceeds the earlier 80% target, but its denominator matters: it covers deaths recorded in the health sector, not every death in Bangladesh.

Source: UNESCAP Bangladesh CRVS country profile. The 2025 figure covers deaths recorded in the health sector.

Cause quality matters because prevention competes for money. A district choosing between trauma capacity, hypertension detection, maternal referral, cancer diagnosis, and outbreak surveillance needs more than a total death count. It needs consistent certification and coding. If the cause is absent, vague, or assigned differently across facilities, the budget follows anecdote and institutional power instead of burden.

Deaths outside facilities present the hardest case. A hospital can train clinicians and improve certification forms. A community death may never touch a medical practitioner. That is why the system must connect local registrars, community health workers, facilities, and statistical methods. The answer is not to assign a confident cause where evidence is weak. It is to separate certified causes, investigated community causes, and unknown causes honestly, then make the unknown share a performance measure.

Bangladesh's first national mortality reports in 2025 show that this pipeline can produce a public output. The next standard should be repeatability. A first report proves capacity. An annual series proves a system.

Counting deaths is a service, not a clerical demand on grieving families

Registration often fails where the state designs it as paperwork the household must complete for the state's benefit. A family dealing with burial, debt, grief, travel, and legal uncertainty may see no immediate reason to visit another office. If registration requires repeated journeys, inconsistent documents, or fees that are not clearly communicated, non-registration becomes a predictable response.

The design principle should be notification once, use many times. A facility death should trigger a civil notification without requiring the family to carry the same facts between systems. A community death should be reportable through a local channel that already reaches the household. The family should receive a certificate that works for inheritance, pensions, banking, and other legal processes. Corrections should leave an audit trail but should not trap people in a loop between offices.

This service view also changes accountability. The household is not the data-entry contractor of the state. Agencies own the duty to reconcile duplicates, protect privacy, and publish aggregate completeness. Local officials need a queue of unresolved events and clear escalation rules. Health workers need training that distinguishes clinical uncertainty from administrative omission. The statistical office needs access to de-identified records and authority to test coverage.

Privacy is not an argument against integration. It is an argument for governed integration. Public mortality tables should be aggregated enough to protect families, while authorised record linkage should use controlled identifiers, logged access, and explicit retention rules. A system that cannot share safely often shares informally and invisibly, which is worse for both privacy and accuracy.

The counterargument: surveys already tell us enough

The strongest objection is that Bangladesh already has surveys, censuses, facility data, and modelled international series. Building a more complete civil system may appear expensive, slow, and unnecessary when statistical methods can estimate mortality at lower cost. For national trends, that argument has force. The WDI decline is meaningful, and no responsible analyst would discard it because registration is incomplete.

But estimation cannot perform the legal function of registration, and a sample cannot manage a district's queue of unregistered deaths. Models also depend on assumptions that become harder to test when administrative evidence is thin. The question is not whether surveys or models should continue. They should. The question is whether the state can accept a permanent arrangement in which estimates compensate for missing identities and causes.

The best system uses each source to test the others. Civil records provide event-level administration. Health records provide clinical detail. Surveys and demographic methods estimate completeness and reveal bias. When the sources disagree, the disagreement becomes a diagnostic. Treating one source as a substitute removes that diagnostic and preserves the blind spot.

Three moves to make every death count

1. Publish an annual completeness table with uncertainty and inequality cuts. Owner: Office of the Registrar General with the Bangladesh Bureau of Statistics. Success signal: a public release that shows registered deaths against estimated deaths by sex, age, district, place of occurrence, and registration delay, with the method and revision policy attached.
2. Make cause-of-death quality a managed production line. Owner: Directorate General of Health Services. Success signal: every annual mortality release separates medically certified causes, other investigated causes, ill-defined causes, and missing causes, and shows each facility's improvement without exposing personal records.
3. Turn notification into a joined public service. Owner: Local Government Division, with health and legal-record agencies. Success signal: a family can notify a death once, track correction, and receive a certificate usable across state services, while authorised agencies exchange the record through logged and privacy-controlled links.

The possibility is larger than cleaner statistics. A complete mortality system tells the state where prevention failed, tells families that a life is legally recognised, and tells the public whether resources follow actual loss. Bangladesh has already shown that mortality can fall. The next achievement is to know every death well enough to prevent the next one.

---

Data note: UNESCAP reports 2025 death-registration coverage and cause certification among deaths recorded in the health sector. These denominators are not interchangeable. WDI mortality values are harmonised estimates and are not presented as civil-registration counts. Sources retrieved 2026-08-23.

Sources

1. UNESCAP, Bangladesh CRVS country profile: https://crvs.unescap.org/country/bangladesh
2. Bangladesh Bureau of Statistics, Vital Statistics: https://bbs.gov.bd/site/page/ef4d6756-2685-485a-b707-aa2d96bd4c6c/vital-statistics
3. World Bank WDI, Crude death rate, SP.DYN.CDRT.IN: https://data.worldbank.org/indicator/SP.DYN.CDRT.IN
4. World Bank WDI, Mortality rate, under five, SH.DYN.MORT: https://data.worldbank.org/indicator/SH.DYN.MORT