Three times the deaths, and the usual risk chart stops predicting all-cause mortality: six years of follow-up in Russian narcology patients
- Prospective cohort built on the Know Your Heart baseline survey in Arkhangelsk, 2015-2017. 2629 residents aged 35 to 69 who consented to record access were linked to the Arkhangelsk Regional Mortality Database and followed to 30 April 2023: 2357 sampled from the general population through the regional health insurance address list (68% agreed to the interview, 96% of those attended the health check) and 272 in-patients of the Arkhangelsk Regional Psychiatric Hospital treated for alcohol-related diagnoses, invited at least a week after admission (85% agreed). Median follow-up 6.3 years (IQR 6.0 to 7.0), 16,567 person-years, 223 deaths (8.5%).
- The outcome is death by underlying cause, taken from death certificates and civil registration records, not a craving score and not a drinking diary. Age- and sex-standardised all-cause mortality per 100,000 person-years was 1229 (95% CI 691 to 1767) in non-drinkers, 890 (95% CI 684 to 1096) in low-risk drinkers, 877 (95% CI 428 to 1325) in hazardous drinkers, 2170 (95% CI 276 to 4064) in harmful drinkers and 4757 (95% CI 3384 to 6131) in narcology patients. 73 of the 272 narcology patients died (26.8%), at a median age of 55 (IQR 47 to 60) against 62 (IQR 53 to 68) for the whole sample. Against low-risk drinkers, with cardiovascular risk factors and biomarkers added as possible mediators (Model B), the hazard ratios were 3.23 (95% CI 2.02 to 5.16) for all causes, 3.25 (95% CI 1.52 to 6.92) for cardiovascular death, 9.36 (95% CI 2.63 to 33.3) for external causes and 7.79 (95% CI 3.34 to 18.1) for other causes. Neoplasm mortality did not differ in any group.
- The authors' own conclusion comes in two parts, in this order: narcology patients showed a threefold excess in all-cause and cardiovascular mortality and a ninefold excess in mortality from external causes relative to low-risk drinkers, and conventional risk factors and biomarkers had limited predictive value in this group, and it is to this second part that the authors attach their call for a broader approach to risk management. On that second part, and for all-cause mortality: in the general-population sample higher mortality went with smoking (HR 2.88, 95% CI 2.05 to 4.05), waist-to-hip ratio, resting heart rate, systolic blood pressure and high-sensitivity C-reactive protein, and lower mortality went with higher education (HR 0.45, 95% CI 0.30 to 0.68) and left ventricular ejection fraction. In narcology patients the same associations were markedly weaker and did not reach significance: smoking 0.99 (95% CI 0.54 to 1.82), higher education 1.33 (95% CI 0.66 to 2.69), C-reactive protein 1.00 (95% CI 0.98 to 1.02), ejection fraction 1.00 (95% CI 0.96 to 1.05). The interaction with study group was significant for each of these. Not reaching significance is not the same as no association, and the restriction to all-cause mortality matters: for cardiovascular death the same table still holds significant predictors inside the narcology group.
- Everything about the exposure was measured once and never again. Drinking category came from self-report over the past 12 months plus AUDIT, CAGE and questions on zapoi episodes, hangovers and sleeping in clothes; the ordering was checked against dose-graded medians of carbohydrate-deficient transferrin (0.54% to 1.64%) and gamma-glutamyltransferase (21.8 to 58.9 U/L) across the five groups, which is a real check but not a repeated one, and not an equally broad one: gamma-glutamyltransferase was measured in every participant, carbohydrate-deficient transferrin only in a subsample of 1242 (976 from the main study sample and 266 narcology patients). Deaths outside the Arkhangelsk Region could not be traced, loss to follow-up was not recorded, and too few women in the harmful drinking and narcology groups died for any analysis by sex.
Every source in this issue meets the same first question: which outcome did it actually register. Here no digging is required. The outcome is death, drawn from the Arkhangelsk Regional Mortality Database, which compiles medical death certificates from health organisations together with regional civil registration records and cross-checks the two against each other. Nobody rated a craving scale, nobody kept a consumption diary. Six years after a hospital admission for withdrawal, a person was either in the death register or was not. That is the outcome this field routinely replaces with a questionnaire, and it is the reason a merely competent epidemiological paper is worth a clinician's twenty minutes.
Who was actually in the narcology group
Not a treatment sample. These were in-patients approached at least a week after admission for detoxification, withdrawal management or another acute alcohol-related condition, and the group is dominated by withdrawal states: F10.3 in 54.4% and F10.4, withdrawal with delirium, in 21.7%, with F10.1 third at 12.5%. Dependence syndrome, F10.2, appears as the primary diagnosis in only 5.5%, a distribution that probably reflects what gets coded on admission more than what the patients had. One caution about reading that list: as printed, two of the six labels do not match the standard ICD-10 designations, and one of the two carries that third-largest share, since F10.1 is harmful use and F10.5 is psychotic disorder, whereas the paper attaches "psychotic disorder" to F10.1 and "amnesic syndrome" to F10.5. The codes are legible, the labels are not.
Where the excess actually sits
The gap is not spread evenly across causes, and that matters more than the headline threefold. Cardiovascular disease was the commonest cause in every group, including narcology patients (25 of 73 deaths). What separates the group is elsewhere: 20 of 73 deaths (27.4%) came from external causes, against 5 of 88 (5.7%) in low-risk drinkers, with accidental poisoning at 5 of 73 (6.8%) and self-harm and undetermined intent redacted at cell counts below five. Neoplasms went the other way, 4 of 73 (5.5%) against 28 of 88 (31.8%); the authors attribute that to the younger age at death in these participants, which may preclude the development of cancer, and add that the follow-up might also not have been sufficient to capture the full impact of alcohol on cancer mortality. Adjustment for cardiovascular biomarkers moved three of the four narcology hazard ratios only modestly (all causes from 3.79 to 3.23, cardiovascular from 3.29 to 3.25, external causes from 11.9 to 9.36), while for other causes the estimate rose, from 5.73 to 7.79. The authors read this, cautiously, as a possible independent harmful effect of chronic severe drinking on the circulatory system, and as a reason to go beyond conventional cardiovascular risk management in this population.
Where the abstract runs ahead of the table, and where it does not
I followed the table. First, the abstract states that cardiovascular mortality in narcology patients was increased with higher education, and the conclusion calls education an exclusive risk factor in that group. The cell is 2.22 (95% CI 0.83 to 5.94). What is significant is the interaction between education and study group, not the association within the narcology column, and the authors themselves offer collider bias as one possible explanation, setting a second one beside it in the next breath, downward social drift. Second, the discussion says HbA1c and cystatin C consistently predicted all-cause mortality in both populations; in the narcology column those are 1.24 (95% CI 0.95 to 1.63) and 2.51 (95% CI 0.77 to 8.18). Again the claim rests on the absence of a significant interaction, not on significance within the group. A smaller arithmetic slip: the paper prints 37.2% for cardiovascular deaths, where 84 of 223 is 37.7%.
The other half of that same sentence survives the table, and reporting only the first half would misrepresent the authors. The conclusion reads that higher education and BMI were exclusively risk factors for cardiovascular mortality in narcology patients, and body mass index holds: 1.11 (95% CI 1.01 to 1.22) inside the narcology group against 0.99 (95% CI 0.94 to 1.05) in the main study, with the interaction between the two significant. Education does not survive the table; body mass index does.
Two further things sit in the body rather than the abstract, and they are not the same kind of omission. For harmful drinkers the abstract gives no hazard ratios at all, so it never shows that their excess in external causes loses significance once biomarkers are added (4.97, 95% CI 1.21 to 20.5, becomes 4.07, 95% CI 0.92 to 18.1). For hazardous drinking, defined as AUDIT 8 to 15 and/or CAGE 2 to 3, the figure is printed in the abstract: 877 (95% CI 428 to 1325) per 100,000 person-years stands there beside 890 (95% CI 684 to 1096) in low-risk drinkers. What the abstract omits is the reading of that pair, which is stated only in the discussion, where the authors note that hazardous and low-risk drinkers did not differ in mortality over these six years.
What this licenses in the room, and what it does not
It licenses treating a narcology admission as a prognostic marker in its own right and acting on the causes that carry the excess: poisoning, injury and self-harm, not oncological surveillance. It licenses caution about the standard risk chart in exactly this population, but only for all-cause mortality, and the blindness is not total. Inside the narcology group, cardiovascular death was still predicted by age (HR 1.06, 95% CI 1.01 to 1.10), body mass index (1.11, 95% CI 1.01 to 1.22), glycated haemoglobin (1.50, 95% CI 1.07 to 2.12) and cystatin C (6.75, 95% CI 1.31 to 34.8), and for all-cause death NT-proBNP sat on the boundary of significance at 1.22 (95% CI 1.00 to 1.49). The table supports a narrower claim than a headline does: the familiar chart stops predicting all-cause mortality here, while a handful of measures still track cardiovascular death. It does not license any statement about treatment. Nobody here was randomised, nothing was measured after baseline, and no retention, relapse or overdose data exist in this dataset. The reverse error is as available as the promotional one: a cohort showing that people admitted for withdrawal die three times as often says nothing at all about whether the ward helped them.
The outcome here is a death certificate rather than a craving score, and it shows that a patient admitted to a narcology ward carries three times the all-cause mortality of a low-risk drinker, with a modelled ninefold for external causes that rests on 20 deaths and a very wide confidence interval, from 2.63 to 33.3.
This is a secondary analysis of baseline data: drinking category, biomarkers and clinical measurements were recorded once, in 2015-2017, and no change over the following years was captured. Exposure rests on self-report, supported but not replaced by carbohydrate-deficient transferrin and gamma-glutamyltransferase gradients, and those two supports differ in reach: gamma-glutamyltransferase was measured in every participant, carbohydrate-deficient transferrin in a subsample of 1242. Lifelong abstainers could not be separated from former drinkers, which plausibly explains the raised mortality among non-drinkers. The narcology group is a clinic sample joined to a population sample, so it is not representative of everyone with alcohol dependence; the cohort's own age-standardised mortality ran 6% lower in men (1621 against 1724 per 100,000) and 18% lower in women (547 against 670) than the urban adult population of the region. Deaths outside the Arkhangelsk Region could not be traced, loss to follow-up was not recorded, and too few women in the harmful drinking and narcology groups died for analysis by sex. This is an observational cohort with no trial registration; the published page names ethics approvals but does not cite a pre-specified analysis plan, and on the other side of that ledger the authors report three analyses at half-year intervals, which consistently confirmed the stability of their findings. The work is a component of the International Project on Cardiovascular Disease in Russia, and its first author holds a dual affiliation, UiT The Arctic University of Norway and the Northern State Medical University in Arkhangelsk. Funding came from a Wellcome Trust strategic award (100217) and Norwegian public bodies, with one author's contribution supported by a Russian Academy of Sciences state assignment (FWNR-2024-0002); the authors declare no competing interests.