The number that anchors the paper is 333,400. That is how many deaths among Russians aged 20 to 59 the authors attribute to alcohol in 2003 — 46% of all deaths in that age band that year. By 2019 the count had fallen to 108,400, or 27%. Neither figure appears anywhere in a Russian death register, and that is the point of the study: the certificates exist, they are filed, and the alcohol they conceal is not on them.
The work is a panel analysis published on 20 November in Population Health Metrics by Aleksey Shchur, Vladimir Shkolnikov, Olga Rodina, Egor Sergeev and Elena Zamyatnina. It is an ecological study of regional time series, not a validation study, and it does not claim to know what any individual Russian died of.
A category defined by its own wording
Russian death certificates can name alcohol. The ICD-10 gives certifiers a set of routes — F10 for mental and behavioural disorders due to alcohol, K70 for alcoholic liver disease, I42.6 for alcoholic cardiomyopathy, G31.2 for alcoholic polyneuropathy, K85.2 and K29.2 for alcohol-induced pancreatitis and gastritis, X45 and Y15 for accidental and undetermined-intent poisoning. The authors gather all of them into a single bucket they call medically certified alcohol-related causes, MCARC, and note that the bucket is narrow at the top: alcoholic cardiomyopathy, alcohol poisoning and alcoholic liver disease together supply 86.9% of everything certified as alcohol-related in the country.
The rest of mortality — every cancer, every stroke, every pneumonia, every road injury that a certifier did not connect to drinking — sits outside that bucket. The paper’s question is how much alcohol mortality is hiding there, and whether the hiding has changed since the early 2000s, when Russian mortality was close to its postwar worst.
Borrowing a signal
Their instrument is a proxy that has followed Russian mortality with unusual fidelity for sixty years: deaths from alcohol poisoning. Poisoning is a small slice of total mortality, between 1% and 2.5% depending on the period, but its year-to-year movements track all-cause mortality so closely that demographers have used it as a barometer of hazardous drinking for decades. Over 2000–2019, the authors report a correlation of 0.90 among men and 0.80 among women between first differences in the standardised all-cause death rate and the poisoning rate.
They then regress the death rate from everything not certified as alcohol-related on the poisoning rate, across 75 Russian regions and 20 years, with region fixed effects. Data come from de-identified individual death records supplied by Rosstat, with population denominators from the Russian Fertility and Mortality Database; rates are sex-specific and restricted to ages 20–59, where hazardous drinking in Russia does most of its damage. Five North Caucasus regions are excluded for distinct drinking patterns and disputed population estimates.
The fitted coefficient is large. One additional poisoning death per 100,000 is associated with 7.3 additional deaths per 100,000 from causes not certified as alcohol-related among men, and 5.7 among women, with R² of 0.64 and 0.57. Within regions, the longitudinal correlation between the two series exceeded 0.90 in 56% of regions for men and 43% for women.
What the model will not say
The authors are careful about the status of that coefficient, and the care is warranted: an association between two regional series is not a cause-specific effect, and a fixed-effects model absorbs a great deal of history — the anti-alcohol campaign of 1985–87, the collapse of the Soviet economy, the 1998 default, the 2005–07 alcohol-market reforms, the 2009 national strategy — into a single region-specific intercept and one slope. “The ecological design precludes causal inference at the individual level,” they write.
Three limitations are stated rather than buried. The method assumes a stable relationship between poisoning mortality and mortality from other causes, and the authors admit the link thinned in 2013–15 and again in 2017–19; a sensitivity model letting the slope vary by year did not overturn the estimate. Changes in certification and coding practice may have moved the series underneath them. And the whole approach may be confined to Russia: correlations of this strength are rare, and the authors judge direct application to Belarus, Ukraine or the Baltic states “questionable” because the association there is weaker.
The other way to name a cause
There is a second way of getting at a death whose record does not carry its cause, and it is the one this digest follows: ask the family. Verbal autopsy puts a structured questionnaire — the WHO’s 2022 instrument is the current standard — to a bereaved relative, and assigns a cause either by physician review or by an algorithm such as the InterVA model, which returns a probability distribution over causes for each death.
The two methods produce numbers that look alike and warrant entirely different things. Verbal autopsy names an individual death and can therefore be tested: the 2022 systematic review by Mahesh and colleagues found 66 validation studies against an established cause, with physician-certified comparators in 36 and computer-coded methods in 14, and still concluded that the studies disagreed on method and that many leaned on weak comparators. The testing can also run in reverse. In a Saudi Arabian study of 302 deaths drawn from a type 2 diabetes register in Makkah, researchers interviewed relatives, ran InterVA-5, and compared the cause fractions with what the hospital had certified — finding that physician certification overestimated endocrine, metabolic and respiratory deaths while undercounting circulatory ones.
The Russian panel does none of this. It assigns no cause to any named person, so it has nothing to validate against an individual’s death; its output is a share of a population’s mortality, and its only support is the strength of a correlation. That distinction matters more than it sounds, because the two kinds of number answer different questions: a cause-fraction can be plausible at the aggregate while individual agreement stays poor, a tension this digest examined in its reading of CSMF accuracy against individual-level agreement.
What moved, and what it leaves open
The trend itself is not in dispute. Total alcohol-attributable mortality, certified and hidden together, peaked for men in 2003 at 770.9 per 100,000 (95% CI 712.3–829.5), or 47.6% of all male deaths at those ages; for women the share peaked in 2002 at 35.3%. By 2019 it stood at 238.1 per 100,000 among men and 48.4 among women — 29.5% and 17.0% of deaths at those ages. Men accounted for roughly 81% of the modelled total throughout. Counted rather than standardised, the 2019 shares come out slightly higher: 31% and 19%.
Read alongside the WHO’s global accounting, which put 2.6 million deaths worldwide down to alcohol in 2019, with 2 million of them among men, the Russian series describes a country moving in the right direction from an extreme starting point. What it also describes is a certification problem of a specific type. Where no physician certifies a death at all, the WHO recommends verbal autopsy; Russia’s difficulty is the opposite one, a register that is filled in and still leaves the dominant contributing condition off the line.
The paper’s closing recommendation — continued strengthening of alcohol control policy — is the authors’ judgement, not a finding of the model, and this digest does not extend it. What the model supports is narrower and still arresting: by their accounting, roughly seven-tenths of alcohol-attributable deaths among working-age men in 2019, and just over half among women, were not certified as alcohol-related at all, and the only reason anyone can put a figure on that is a sixty-year correlation the authors themselves are unsure will hold.
interva.net is an independent digest of verbal autopsy and cause-of-death research. It is not affiliated with, endorsed by, or speaking for the InterVA project, Umeå University, or the World Health Organization.
Sources
- Shchur A, Shkolnikov VM, Rodina O, Sergeev E, Zamyatnina E. Hidden alcohol-attributable mortality in Russia: evidence from regional panel data, 2000–2019. Population Health Metrics 2026;24:64 (published 20 November 2026) doi.org
- WHO. Alcohol — fact sheet (alcohol-attributable deaths and burden) who.int
- Mahesh BPK, Hart JD, Acharya A, Chowdhury HR, Joshi R, Adair T, Hazard RH. Validation studies of verbal autopsy methods: a systematic review. BMC Public Health 2022;22:2215 pmc.ncbi.nlm.nih.gov
- Alyazidi F, Shakely D, Alyazidi FR, Petzold M, Hussain-Alkhateeb L. Validating the diagnostic accuracy of medical certification: a population-level comparison between verbal autopsy and Saudi medical records causes of death of deceased with type 2 diabetes. Global Health Action 2024;17:2448382 pmc.ncbi.nlm.nih.gov
- WHO. Verbal autopsy instrument (V2.0, 2022) — the current WHO standard questionnaire who.int