October 13, 2026 interva.net Editorial

The Certificate Says Sepsis. The Chart Says Kidney Failure.

Two audits of physician-written death records — 25,123 certificates in the Islamic Republic of Iran and 365 adjudicated intensive-care deaths in India — both find that what gets written is often the terminal event, not the condition that set the chain in motion.

cause of deathdata qualityCRVSverbal autopsy

A death certificate is not written at leisure. It is written at the end of a shift, usually by whichever clinician was present, and it asks for something harder than a diagnosis: a causal sequence. The immediate cause goes on the first line, the conditions leading up to it in the middle, the underlying cause at the bottom. In practice, the lines that get filled in are the ones describing what happened last.

Two new audits of that paperwork, from opposite ends of the cause-of-death problem, put numbers on the habit. One examines whether certificates were completed at all. The other asks whether the cause written on them matched what the clinical record actually showed.

What 25,123 certificates in Iran showed

The larger of the two is a national quality audit published in the Eastern Mediterranean Health Journal, from a team led by Hossein Kazemeini, Ardeshir Khosravi and colleagues in the Islamic Republic of Iran. Using a Persian-language version of a death-certificate quality assessment tool designed at the University of Melbourne, the authors reviewed 25,123 physician-issued certificates.

Only 29.3% were judged correctly completed. Two thirds — 67.7% — were incomplete. The most common defect was not an exotic one: 86.3% of certificates failed to record the interval between the onset of illness and death. Illegible handwriting, the error most often imagined as the problem, was the least common at 14.5%.

The geography of the audit matters for anyone working on mortality measurement. Most deaths happened inside institutions — 63.3% in hospitals and a small number of surgical centres — but 32.3% occurred at home. Certificates were issued in hospital for 59.1% of cases and by other doctors for 18.9%.

The authors’ recommendations are procedural: develop guidelines, train physicians, and integrate certification into existing electronic health systems so that the clinical data a certifier needs is already in front of them. What the study measures is the completeness of a form. It does not establish whether the causes that were written down were correct, and its authors do not claim otherwise.

The chart against the certificate

The second audit is narrower and more uncomfortable. A team at Max Super Speciality Hospital in New Delhi, publishing in Cureus on 22 July 2026, drew a random sample of 365 patients who died in the hospital’s medical intensive care units. Two senior physicians independently reviewed the full clinical record of each death — admission notes, laboratory results, imaging, dialysis records, discharge summaries — and classified it as kidney-disease-related or not, using pre-agreed criteria built around KDIGO definitions of acute kidney injury and chronic kidney disease. They were blinded to each other; where they disagreed, the cases went to a consensus meeting and, failing that, to the hospital’s mortality review committee.

Adjudication found that kidney disease had contributed to 67.4% of these deaths. The certificates recorded it in 45.2%. That is a gap of more than twenty percentage points, in a hospital with electronic laboratory reporting and specialist nephrologists on staff. The reviewers agreed with each other 87.3% of the time, a Cohen’s kappa of 0.78; 56 cases, or 15.4%, required consensus adjudication.

The authors’ explanation will be familiar to anyone who has read much certification guidance: the terminal event gets written down instead of the underlying process. Sepsis, septic shock, respiratory failure and multi-organ dysfunction are what the certifier saw happen. The renal failure beneath them, managed for years or unmasked by an acute illness, is easier to leave off the form.

Reading the interval

Why return to that missing interval between illness onset and death, the field empty on 86.3% of the Iranian certificates? Because it is the field that tells a coding office whether a death belongs to a long-term condition or a short one. Without it, a chronic disease that ended in an acute event cannot be distinguished from an acute event alone, and the underlying-cause rule that underwrites international comparability loses the information it needs to be applied consistently. The same death can be attributed to the kidney, to the diabetes above it, or to the infection at the end of the chain.

Where nobody signs anything

Both audits describe countries where, for part of the population, there is no certificate to audit. Iran’s 32.3% share of home deaths is the closer of the two to the situations that occupy most of this digest: settings where roughly a third or more of deaths happen outside any facility, and where the substitute for a physician-written certificate is an interview with a family, scored by a model. The WHO’s verbal autopsy instrument and the coding methods built around it are the instruments there, which is why their accuracy on each cause matters.

It is worth being blunt about the limits on the other side too. The Indian team’s own discussion concedes that the alternative route has a blind spot on this particular cause: kidney failure announces itself in laboratory values that a family cannot describe, and the authors note that verbal autopsy may not be accurate in ascertaining kidney disease-related deaths. Their reference point is the Million Death Study’s nationally representative estimates of renal failure deaths in India, themselves produced by verbal autopsy and reported with that uncertainty attached. Two methods, two different errors.

What these audits do and do not establish

Neither paper shows that cause-of-death statistics are wrong wherever they are collected, and neither proposes that any algorithm or tool be adopted. The Iranian study is a national assessment of form completion; the Indian study is a single-centre retrospective review whose adjudicated classification is a pragmatic reference standard, not a laboratory gold standard — the authors say so, and also flag the risk of incorporation bias, the absence of external validation, and the fact that their prediction model should be read as an exploratory attribution aid rather than a causal model. Numerical performance figures from a model like that do not travel to other hospitals until someone tests them there.

What the two audits share is a view of cause-of-death data as an output of process rather than a by-product of medicine: who fills the form, with what information in front of them, and under what incentive. The institutional side of that — whether certification training is compulsory, whether it is tied to a licence, and whether it is taught at all — was the subject of a seven-country qualitative study this digest covered earlier, on how health systems try to make cause-of-death training permanent. Iran’s authors land on the same prescription: put the clinical record where the certifier can see it. Zambia’s mobile death-notification rollout, covered here, is one working example of that idea in practice.

None of this is advice to any ministry, and this digest is an independent observer: it is not affiliated with the InterVA project, Umeå University or WHO.

Sources

← Back to the news digest