The stakes are higher than the administrative paperwork might suggest. Distinguishing between types of deaths matters for two central reasons: it allows epidemiologists to understand and prevent them, and it is essential to bringing justice to homicide victims. Decades of research have documented the many ways in which determining the manner of death can go wrong, yet the most common policy responses, calls for more training and more resources, have largely never been put to an empirical test. Aside from work on cognitive bias mitigation, the authors note that they are not aware of studies demonstrating that these widely recommended remedies actually work. The French reform offered a rare opportunity to measure the effect of added guidance directly.

The French system differs from the American one in ways that make such a test possible. In the United States, medicolegal professionals become involved automatically when a death falls into specific categories. In France, involvement must be triggered by the physician who ascertains the death, who can check a box on the administrative panel of the certificate known as the obstacle médico-légal, or medicolegal objection. Checking that box halts all funerary arrangements and alerts the prosecutor, who then decides whether to refer the death to a local medicolegal institute. This structure introduces two extra layers of discretion compared with the American system, and therefore more opportunities for bias and error, since a non-specialist physician must decide on the spot whether a death warrants investigation.

Until 2017, the instructions for that checkbox, printed on the back of the paper certificate, referred only to suicide or suspicious death likely arising from a violation of the law. The revised certificate listed specific types of deaths that should trigger the notice, moving the French document closer to the highly prescriptive categorization used in the United States. Other changes targeted the supplementary information section of the medical panel: questions about the location of a violent death and about whether a death was pregnancy-related became multiple choice rather than open-ended, a question about sudden death and one about the apparent circumstances of death were added, and the question about whether an autopsy had been completed was replaced by one about whether an investigation into the cause of death had been requested. The revised certificate was piloted in several jurisdictions in 2017 and deployed nationwide in 2018, although the older version remained in use alongside it.

To measure the effect, the researchers drew on the CépiDC database, the nationwide compilation of all deaths in France maintained by INSERM, selecting the years 2016 through 2019 to balance pre- and post-reform certificates. They deliberately stopped before 2020 to exclude Covid-19 deaths. Two hypotheses were tested with logistic regression models, controlling for the sex and age of the deceased and for the type of certificate, whether adult or neonatal and paper or electronic. The first hypothesis was that medicolegal investigations would increase after the change. The second was that the precision of ICD-10 cause-of-death codes would improve. Precision was operationalized by classifying codes as vague or specific at the higher-order level, with all codes labeled other or non-specified counted as vague.

The results supported both hypotheses. Deaths certified on the 2017 version were more than twice as likely to be investigated by medicolegal professionals, with an odds ratio of 2.350. Electronic certificates were associated with an even larger increase in investigations, an odds ratio of 2.570, regardless of the certificate version, and neonatal deaths were close to ninety percent more likely to be investigated than adult deaths, reflecting the routine use of autopsies when infant deaths are difficult to explain. Women’s deaths were less likely to be investigated than men’s, and the likelihood of investigation declined with the age of the deceased, patterns the authors attribute to the fact that medicolegal investigations focus on violent and unexplained deaths, which women and the elderly are less likely to experience.

The second model showed that deaths certified on the 2017 version were almost ten percent more likely to carry precise ICD-10 cause-of-death codes, an odds ratio of 1.076. The authors argue that this finding, though less dramatic than the doubling of investigations, may be substantively more significant, because medicolegal investigations remained very rare even after the reform, whereas the improvement in coding applied across the full population of more than two and a half million deaths. Notably, the information on apparent circumstances of death, roughly equivalent to the American manner-of-death classification, was missing in 99.91 percent of deaths on the pre-2017 certificates but only 1.99 percent on the post-2017 certificates, simply because the new version included explicit checkboxes for those categories.

Some results complicated the picture. Neonatal certificates showed lower-quality cause-of-death coding than adult certificates, which the authors attribute to the wide proportion of vague codes within the ICD-10 perinatal categories and to the difficulty of explaining infant deaths when the child has no medical history. More surprisingly, electronic certificates were more likely than paper ones to be coded vaguely, the opposite of their effect on investigation rates. In 2024, automated batch-coding in France finalized only 63 percent of ICD-10 codes, with 83 percent of the remainder rejected for spelling errors, irrelevant information, or ambiguous acronyms. Because paper certificates offer no guardrails and thus require more manual coding, the authors suggest that this manual review may act as an additional layer of quality control that benefits paper certificates, a hypothesis that will need direct testing.

The researchers propose that the mechanism behind the improvement goes beyond simple instruction. More medicolegal investigations alone cannot plausibly explain an across-the-board improvement in cause-of-death coding, given how rare those investigations remain. They instead posit that the more detailed certificate forced physicians to slow down and think through each death more carefully, a process consistent with evidence from other contexts that slowing decision-making can improve judgment. Physicians filling out death certificates face real pressures: their attention is on the grieving family, funerary arrangements require certificates within tight windows, and non-specialists may neither grasp the public health importance of accurate mortality data nor be able to differentiate among competing causes of death. Embedding guidance directly in the form meets certifiers at the moment of decision.

The authors conclude that their natural experiment provides strong support for the hypothesis that more specific information on death certificates improves mortality statistics, a finding they consider translatable beyond France as long as policies are adapted to local administrative cultures. Two caveats frame the work: the study is a before-and-after comparison rather than a randomized trial, and the certificate reform overlapped with a simultaneous rise in electronic certification, which the models accounted for statistically but which operates in the same period. The next steps, the authors write, are to establish a dose-response relationship, determining how much guidance yields improvement and at what point returns diminish, so that cost-effective policy recommendations can be built on this evidence for health systems worldwide.

Subject of Research: Impact of the 2017 French death certificate reform on the accuracy of mortality statistics and medicolegal death investigations

Article Title: Implications of the 2017 change in the French death certificate for the accuracy of mortality statistics worldwide

Article References: Neuilly, M.-A., Balusson, F., & Bouvet, R. (2026). Implications of the 2017 change in the French death certificate for the accuracy of mortality statistics worldwide. International Journal of Legal Medicine. https://doi.org/10.1007/s00414-026-04039-x

Image Credits: AI Generated

DOI: 10.1007/s00414-026-04039-x

Keywords: death certificate, mortality statistics, France, medicolegal investigation, ICD-10, cause of death, epidemiology, natural experiment, forensic medicine, death certification, CépiDC, public health data

The study was led by Caroline B. Osoro of the KEMRI-Wellcome Trust Research Programme in Nairobi, together with colleagues from KEMRI, the African Population and Health Research Center, Kenya’s Department of Civil Registration Services, New York University Abu Dhabi, and the University of Oxford. The team conducted household surveys at the three surveillance sites, sampling 1,363 households that had experienced a recent death. Because these sites maintain long-running population registers that independently record births, deaths, and migrations, the researchers could anchor their survey findings in well-characterised communities, reducing the risk that unrecorded deaths would go entirely unnoticed.

The technical approach was equally rigorous. Rather than simply reporting a headline percentage, the investigators used random-effects regression analysis to quantify how individual, household, and contextual characteristics shaped the odds that a death would be registered. This statistical framework accounts for clustering of outcomes within households and sites, allowing the team to estimate adjusted odds ratios with confidence intervals for a wide range of potential determinants, from the age and marital status of the deceased to household wealth, place of death, and the interval between the death and the survey interview.

The results reveal a steep social gradient in who gets counted. Deaths among infants were the least likely of all to be registered, with an adjusted odds ratio of 0.09 compared with adults aged 65 to 84 years, meaning infants had roughly one-eleventh the odds of adult registration. Children aged one to four fared almost as poorly, with an adjusted odds ratio of 0.14. Unmarried decedents were less likely to be registered than married ones, with an adjusted odds ratio of 0.38, and unemployed decedents were also disadvantaged, at 0.60. Deaths that occurred at home were less likely to be registered than those in facilities, with an adjusted odds ratio of 0.65, a pattern that likely reflects the greater involvement of health workers and documentation pathways when someone dies under medical care.

At the same time, several factors increased the likelihood of registration. Deaths in urban settings had higher odds than rural ones, with an adjusted odds ratio of 1.55. Households in the wealthiest categories were substantially more likely to register deaths, with an adjusted odds ratio of 2.50, and decedents who had been employed or retired had 1.87 times the odds of registration compared with those who were unemployed. Perhaps most tellingly, deaths that had occurred very recently, within the preceding six months, were the least likely to have been registered, with an adjusted odds ratio of 0.26, indicating that registration in Kenya is not only incomplete but also heavily delayed, with many families completing the process only when a specific need arises.

That need, the study found, is most often inheritance. When respondents explained why a death had been registered, the most frequently cited reason was inheritance-related requirements, meaning that the death certificate functions less as a civic document and more as a legal key to property and assets. This dynamic helps explain the demographic pattern: adult deaths, particularly among married, employed people who own property, trigger inheritance processes that compel registration, while the deaths of infants, children, and the very poor generate no such legal pressure and therefore slip through the system unrecorded.

On the other side of the ledger, the leading reasons for non-registration were lack of awareness of the requirement, the high costs involved in obtaining a death certificate, and a perceived lack of benefit from registering at all. These barriers compound one another. Families who do not know registration is mandatory are unlikely to absorb the transport and administrative costs of travelling to a registration office, especially when they see no tangible return. For the deaths of young children, grief and stigma may add further disincentives. The result is a self-reinforcing cycle in which the absence of perceived benefit suppresses registration, which in turn keeps mortality statistics incomplete and weakens the case that communities might otherwise see for engaging with the system.

The consequences of this undercount extend far beyond individual families. Civil registration and vital statistics systems are the backbone of national health planning, providing the cause-of-death and mortality data that governments need to allocate resources, evaluate interventions, and monitor progress toward the Sustainable Development Goals. When only a fifth of deaths are registered, and when the unregistered deaths are concentrated among infants, children, and the poorest households, the resulting statistics do not merely understate mortality; they systematically erase the populations with the worst health outcomes. Policies designed on such data will be calibrated to the visible, registered minority rather than the true burden of disease and death.

The study’s findings carry particular weight because they come from Kenya, a country often regarded as a regional leader in health information systems, and because the authors included a representative from the Department of Civil Registration Services itself, suggesting institutional awareness of the problem at the highest level. The research was supported by the Wellcome Trust through senior and principal fellowships and the Kenya Major Overseas Programme, and it received ethical approval from the Kenya Medical Research Institute Scientific and Ethics Review Unit, the national science commission, and county health authorities at all three sites. The authors note that the funders had no role in study design, analysis, or the decision to publish.

The policy implications are clear. The authors argue that strengthening awareness and advocacy for timely registration, particularly of infant and under-five deaths, must be paired with targeted efforts to reach poor and unemployed households, whose members are disproportionately excluded from the civil record. Practical reforms suggested by the data include integrating registration prompts into facility and community-based death notification, reducing or eliminating fees for death certification, deploying community health promoters to sensitise families at the time of bereavement, and simplifying procedures so that registration does not depend on an impending inheritance dispute. Without such measures, Kenya’s most vulnerable citizens will remain statistical ghosts, their deaths invisible to the systems charged with protecting the living, and the promise of universal civil registration, a cornerstone of health equity and social inclusion, will remain unfulfilled.

Subject of Research: Completeness and determinants of death registration in Kenya

Article Title: Prevalence and determinants of death registration across three demographic surveillance sites in Kenya

Article References: Osoro, C. B., Bukosia, J., Waluke, I., Sifuna, P., Omolo, K., Nyaguara, A., Were, V., Rajwayi, M., Mwanzia, B., Onyango, S. A., Odipo, E., Chepkurui, V., Nyundo, C., Kamau, J., Helleringer, S., Snow, R. W., & Okiro, E. A. (2026). Prevalence and determinants of death registration across three demographic surveillance sites in Kenya. BMC Medicine. https://doi.org/10.1186/s12916-026-05278-w

Image Credits: AI Generated

DOI: 10.1186/s12916-026-05278-w

Keywords: death registration, civil registration and vital statistics, Kenya, demographic surveillance, health equity, mortality statistics, infant mortality, Sustainable Development Goals, BMC Medicine, household survey, inheritance, health systems