Table 2 gives the main results for male mortality at ages 15–74 years in ever-drinkers, subdivided by cause and amount drunk. For neoplastic disease, the trend in risk across categories of ever-drinkers was significant only for upper aerodigestive tract cancer (RR 3·48, 95% CI 2·84–4·27, in the highest alcohol consumption category; trend p<0·0001) and liver cancer (2·11, 1·64–2·70; trend p<0·0001). The trends were not significant for lung cancer (p=0·18) or stomach cancer (p=0·46).

Six non-neoplastic disease groups were also strongly associated with alcohol in men, with RRs greater than 3·00 in the highest consumption category (all p<0·0001 for the trend in risk). These disease groups were tuberculosis (4·14, 3·44–4·98), pneumonia (3·29, 2·83–3·83), liver disease (6·21, 5·16–7·47), pancreatic disease (6·69, 4·98–9·00), acute ischaemic heart disease (IHD) other than myocardial infarction (3·04, 2·73–3·39), and death from an ill-specified disease (7·74, 6·48–9·25). The latter two groups include some misclassified deaths that were actually from alcohol poisoning.⁶ Some other causes of death were significantly associated with alcohol, but the RRs were less extreme (eg, stroke [1·28, 1·15–1·43]). There was, however, no significant trend in mortality from acute myocardial infarction (p=0·62); in the heaviest drinkers (three or more bottles per week), acute myocardial infarction accounted for only 178 deaths, whereas acute IHD other than myocardial infarction accounted for 1044 deaths.

In the combined results for all causes of death strongly associated with alcohol in men (ie, liver cancer and anything with RR more than 3·00 in table 2) there was, of course, a definite and progressive dose-response relation (with RR 3·77, 3·44–4·12, for men in the highest alcohol category), whereas in the combined results for all other diseases the progressive association, although significant, was less strong (1·40, 1·29–1·52).

As expected, reported alcohol consumption was strongly associated with the mortality that was certified as caused by alcohol poisoning, but 41 (3%) of 1610 ever-drinkers who died from alcohol poisoning were in the lowest alcohol category (reference category), and a further 11 were reported as never-drinkers, indicating that proxy information sometimes missed life-threatening alcohol use. Moreover, another 188 (12%) deaths from alcohol poisoning were in men whose usual consumption was, perhaps inaccurately, reported to be less than one bottle per week. Reported alcohol consumption was also strongly associated with the other major groups of external causes of death, including transport accidents (4·20, 3·31–5·34), other accidents (6·07, 5·27–6·98), suicide (8·62, 6·99–10·62), and assault (9·47, 7·96–11·25; all p<0·0001 for the trend in risk).

Table 3 gives the corresponding results for women. For neoplastic disease, the RR in the highest drinking category appeared to be increased only for liver cancer and upper aerodigestive tract cancer, although no trend in cancer mortality was significantly positive (perhaps because of small numbers). For breast cancer, the trend was significantly negative (p=0·0002), and the relative risks in the two highest alcohol consumption categories were both substantially lower than 1·00.

The non-neoplastic disease groups that were most strongly associated with alcohol in men were even more strongly associated with alcohol in women, each with an RR of more than 3·00 not only in the heaviest drinkers but also in women drinking one to three bottles of vodka (or equivalent) per week (and each with p<0·0001 for trend). The same findings were seen for each main group of external causes, including transport accidents, other accidents, suicide, and assault. As in men, reported alcohol consumption in women was strongly associated with mortality from alcohol poisoning, but in the 424 female ever-drinkers certified as dying from alcohol poisoning, 51 (12%) were in the lowest alcohol category (reference category), and another 41 (10%) were reported to have usually consumed less than one bottle of vodka (or equivalent) per week. Again, the RR was substantially less extreme for acute myocardial infarction than for other acute IHD, and in the top two alcohol consumption categories, only 64 women died from acute myocardial infarction compared with 463 from other acute IHD.

Table 4 subdivides the results for all men more narrowly by age (15–54 years and 55–74 years) and groups the causes as alcohol poisoning, accidents and violence, diseases strongly associated with alcohol in this study, other non-control diseases, and control diseases. (The strongly associated diseases were liver disease, liver cancer, upper aerodigestive cancer, and other disease groups with RR more than 3·00 for men with usual consumption three or more bottles of vodka per week [tuberculosis, pneumonia, acute IHD other than myocardial infarction, non-neoplastic pancreatic disease, and ill-specified disease].) The relative risks comparing never-drinkers with reference drinkers are consistent with families under-reporting moderate alcohol use more in men who had died from accidents, violence, and diseases strongly associated with alcohol than in men who had died from other diseases. This finding is particularly clear for men aged 55–74 years, in whom the ratio of reported never-drinkers to reference drinkers was 1·1 ([88+371]/[87+317]) for men who had died of accidents, violence, and diseases strongly associated with alcohol, compared with only 0·4 ([560+98]/[1266+223]) in men who had died of other diseases (p<0·0001 for difference in ratios).

Table 5 gives corresponding results for women. Discrepancies between never-drinkers and reference drinkers are less extreme than in men, but are still apparent, particularly at ages 55–74 years. The ratio of never-drinkers to reference drinkers at these ages was 0·9 (742/832) for female deaths from accidents, violence, and the diseases strongly associated with alcohol, and only 0·5 (2623/4868) for female deaths from other diseases, again consistent with variable under-reporting of moderate alcohol use.

The relative risks in ever-drinkers (which compare the three higher alcohol consumption categories with the reference category) might be more directly indicative of real associations. Even in ever-drinkers, however, substantial misclassification of real alcohol consumption patterns is inevitable. If some alcohol-associated deaths are included in the reference category, then uncorrected calculation of alcohol-dose-related excess mortality in the other three alcohol consumption categories will underestimate the real excess mortality. Exact correction for this is impossible, but approximate correction for it is made in tables 4 and 5 by multiplying the uncorrected excess mortality from each cause by a common factor, calculated within each age range to ensure that the corrected dose-related excess is 100% for the deaths from alcohol poisoning. At ages 15–54 years, the alcohol dose-related excess accounted for 6712 (48%) of 13 968 male deaths in the uncorrected calculations, and 8182 (59%) of 13 968 in the corrected calculations. Of the corrected excess, 90% ([1163+3756+2432]/8182 deaths) was from alcohol poisoning, accidents, violence, and the eight disease groups that are strongly related to alcohol; of the remaining 10% (831/8182 excess deaths in men at ages 15–54 years), much was from vascular disease. The corrected alcohol dose-related excess accounted for 3944 (22%) of 17 536 deaths in older men, 1565 (33%) of 4751 deaths in women aged 15–54 years and 1493 (12%) of 12 302 deaths in older women.

For sensitivity analyses, alcohol use was crudely split into lower (<one bottle per week, including never drinkers) and higher usual consumption categories, and the analyses of tables 4 and 5 were then repeated. With this crude split, the uncorrected dose-related excess for alcohol poisoning was only 72% (1466/2048), but correction (as in tables 4 and 5) automatically increased this to 100%, and made dose-related excess mortality at ages 15–54 years 50% (7045/13 968) for men and 27% (1305/4751) for women, and at ages 55–74 years made it 22% (3827/17 542) for men and 10% (1275/12 302) for women. These corrected percentages are similar to those shown in tables 4 and 5, and were not materially affected by finer stratification for potential confounders (amount smoked, year of death, years since death, marital status, relationship to informant, ethnicity, education, occupation, socioeconomic status, or recent change in socioeconomic status) or by inclusion in the controls of all cancers not strongly related to alcohol (data not shown).

Non-beverage alcohol use was strongly correlated with other alcohol use but, given vodka consumption, was no more common in those dying from strongly alcohol-related causes than in those dying from other causes. For men who died at ages 15–39 years and were drinkers of less than one, one to less than three, or three or more bottles of vodka (or equivalent) per week, the respective proportions of deaths with non-beverage alcohol use reported were 1% (18/1306), 6% (70/1106), and 20% (250/1257) for the strongly alcohol-related causes and 1% (3/313), 5% (8/168), and 28% (34/120) for the other causes. These findings suggest that for a given amount of ethanol consumption its source was not strongly predictive of cause of death. For women, the corresponding percentages were based on much smaller numbers: 2% (8/405), 6% (9/153) and 20% (38/192) for strongly alcohol-related causes and 0% (0/240), 0% (0/23) and 39% (9/23) for other causes, respectively.

Our study of residents in three industrial cities populated mainly by European Russians was large and population based. The interviewers were local physicians; therefore, few families refused to be interviewed and cooperation was generally good. Information was unavailable only when the address no longer existed or no family member was found there. This excludes deaths in families that subsequently moved, deaths of those who (perhaps partly because of alcohol) were isolated or without a registered address, and deaths of migrants to the study cities. Otherwise, selection bias is unlikely to have had any major affect on the results.

Cases were those adults who had died from causes we judged beforehand likely to be related to drinking or smoking, and controls were those who had died from other causes. This designation of controls was somewhat arbitrary, but is supported by the cancer results. The extensive epidemiological evidence on alcohol and cancer, as summarised by the WHO International Agency for Research on Cancer (IARC),¹⁹ shows that the only major types of cancer much affected by alcohol in men are those of the liver and upper aerodigestive tract. This evidence helps validate the case–control results from our study (table 2), which show a highly significant effect of alcohol on mortality from cancers of the liver and upper aerodigestive tract, but not from lung cancer or stomach cancer. Few women consumed more than three bottles of vodka (or equivalent) per week, but among those who did, the only conventionally significant excess cancer risks were for those of the liver and upper aerodigestive tract.

Use of deceased people as controls was intended to limit the effects of recall and reporting bias, with the expectation that there would be no large systematic bias between people with similar drinking habits who had died of different diseases. (Similarly, many epidemiological studies of hospitalised cases use hospitalised controls.) Although families in our study are likely to misreport alcohol consumption, we had hoped that such misreporting would merely dilute the strengths of any associations. We found, however, that under-reporting of moderate alcohol consumption was particularly common for those who had died from accidents, violence, and the diseases strongly related to drinking. This under-reporting could substantially distort relative risks comparing never-drinkers with reference drinkers, but should have less of an effect on the corrected dose-related excess, on which our main conclusions are based. Moreover, sensitivity analyses that combined never-drinkers with the lowest two groups of ever-drinkers yielded similar corrected estimates of excess mortality. To include a reasonably representative sample of all deaths, including those of alcoholics, we settled for the imperfect, and potentially slightly biased, exposure information that families can provide. It is only because the effects of alcohol on mortality in Russia are so great that our crude methods can provide useful estimates of them; in less severely affected populations such methods might not yield useful results.

For those who died before the age of 40 years, we asked about regular non-beverage alcohol use. Such practice was common only in the highest alcoholic beverage consumption category, and was not independently predictive of risk. Thus, as others have suggested,²⁰ non-beverage alcohol use should perhaps be seen mainly as a source of ethanol rather than of other toxins.

Apart from liver cancer and upper aerodigestive cancer, the strongly positive associations were with mortality (in both men and women) from liver disease, pancreatic disease, tuberculosis, pneumonia, ill-specified disease, acute IHD other than myocardial infarction, and external causes (alcohol poisoning, accidents, and violence). At ages 15–54 years, these strongly positive associations accounted for a substantial proportion of all deaths (figure 2), and for about 90% of the excess mortality (tables 4 and 5); vascular disease accounted for most of the remainder. The only significantly inverse association was with mortality from breast cancer, which was substantially lower in the few women in the highest alcohol category (mean more than five bottles of vodka or equivalent per week) than in women from the lower alcohol consumption categories. In other populations, moderate alcohol consumption (all levels of which would be within our reference category) appears to increase the incidence of breast cancer.^19,21 The reasons for this apparent discrepancy are unclear, because information on many other potentially relevant factors was unavailable; however, the inverse association with breast cancer would, even if causal, account for only 0·1% of all deaths.

The excess mortality from liver cancer, upper aerodigestive cancer, liver disease, and pancreatic disease is largely or wholly because alcohol caused the disease that caused death.^19,22–27 The excess mortality from tuberculosis and pneumonia may be partly a result of increased exposure to infection, reduced immunocompetence, or decreased likelihood of cure;^28,29 in each case, however, alcohol use could be a direct or indirect cause of the excess. Some of the excess mortality from stroke and other vascular diseases must reflect the ability of alcohol to increase blood pressure.

The relative risks for mortality from ill-specified diseases were particularly extreme, perhaps partly because alcohol use was associated with inadequate post-mortem assessment of causes. The largest single contributor to the alcohol-associated excess mortality from disease was acute IHD other than myocardial infarction. In a survey of forensic autopsies in Barnaul during 1990–2004 at which post-mortem blood alcohol was recorded, half of those who died aged 15–69 years with non-myocardial infarction acute IHD (ICD-10 code I24) as the certified cause did not have raised blood alcohol concentrations, but 15% had concentrations over 4 g/L, which is potentially lethal.⁶ In other populations, unspecified ICD codes could usually, with better data, be reassigned to more specific codes in the same disease group.³⁰ In Russia, however, unspecified cardiovascular disease codes are common, and probably mask large numbers of deaths from acute alcohol poisoning or from acute outcomes of chronic alcoholic cardiotoxicity. This hypothesis is strongly supported by the sharp fluctuations in some but not all vascular causes of death during 1990–2006 in Russia as a whole; in men aged 35–69 years there was little variation in the national mortality attributed to acute myocardial infarction, but there was extreme variation in the much greater mortality attributed to other acute IHD, with a large increase after 1991 that peaked sharply in 1994.⁶

The largest contributors to alcohol-associated excess mortality were accidents and violence (including assault and suicide). The excess of accidental deaths in alcohol users could be largely or wholly causal. The excess of deaths from violence by others could be partly from the effects of alcohol on the behaviour of the dead person and partly from exposure to other drinkers. (Fluctuations in population alcohol consumption match those in population homicide rates.³¹) Finally, alcohol can cause depression,²² impulsivity, and suicidal behaviour,^24,32 and many people who committed suicide had raised blood ethanol concentrations.⁶ The relative risks for death from assault and suicide were both so extreme in the highest male and two highest female alcohol consumption categories that most of the excess must be causal.³²

A case–control study with live controls estimated that 43% of all deaths in men aged 25–54 years in Izhevsk in 2003–05 were from alcohol (including non-beverage alcohol).¹⁴ Our study (after correction to include all deaths from alcohol poisoning) attributed to alcohol 59% of all male and 33% of all female study deaths at ages 15–54 years; and, our corrected percentages might still underestimate the alcohol-attributable proportion of all deaths in the study areas, since they were calculated only for non-migrants with families still available at the same address years later, and exclude some deaths due to alcohol use by others. Moreover, they average the whole study period (1990–2001); alcohol-attributable proportions would have been greater in the central years (1993–97) than in 1990–91 or 1998–99, since the large temporal fluctuations in overall mortality (and, in particular, the large increase during 1992–93) were mainly caused by sudden large changes in the causes we found to be strongly associated with alcohol (figure 2).

For adult mortality in Russia as a whole, the large fluctuations since 1990 (with, again, a particularly sharp increase during 1992–93) were due to much the same causes as the fluctuations in our study area;⁶ therefore, the proportion of adult Russian mortality caused by alcohol must, as in our study, have been large and have varied greatly (figures 1 and 2).

Figure 3 compares the all-cause mortality rates (averaging male and female) at ages 15–54 years in Russia and western Europe since 1980. Because our study, and other studies of individuals,^11–14 show alcohol to have been a major determinant of overall mortality in Russia, the sharp changes in alcohol and in mortality in the mid-1980s are particularly informative, and they are well documented^3–8 and uncomplicated by the social upheavals of later decades. Because of the restrictions suddenly placed on alcohol in mid-1985, total consumption (legal plus other) dropped suddenly by about a quarter,^7,8 as did overall mortality at ages 15–54 years (figures 1 and 3). Still, however, during the period of lowest alcohol consumption (1986–87), about three-quarters of pre-1985 consumption remained.⁷ Thus, although Russian mortality was then at its lowest, a great deal of the alcohol-attributable mortality must also have remained.⁸ If, therefore, the 1985 decrease of only about a quarter in alcohol consumption was largely responsible for the 1984–86 decrease of almost a quarter in Russian mortality at ages 15–54 years, then in the (hypothetical) absence of all deaths from alcohol, Russian and west European death rates at these ages could well have been similar in 1986. This suggests that, in Russia, alcohol has been of even greater importance than tobacco as a cause of premature death. In western Europe in the mid-1980s, smoking was responsible for about a quarter of all mortality (and a third of all male mortality) in middle age, and in Russia male lung cancer mortality was then about 50% greater than in western Europe.^1–3,33,34 Even if the overall mortality rate from smoking was also somewhat greater in Russia, however, it would still have been much smaller, at least at ages 15–54 years, than the probable effects of alcohol in Russia since 1991 (figures 1 and 3).

If, without alcohol, mortality at these ages would have been similar in Russia and western Europe in the mid-1980s, then (without alcohol in either population) it would probably still be similar in the present decade. If so, then Russian death rates at these ages would now be much less than double the corresponding rates in western Europe (figure 3). The actual Russian mortality rate in people aged 15–54 years in 2006 was, however, more than five times (for men) and three times (for women) the rate in western Europe (figure 1).^1–3 This large discrepancy is consistent with alcohol being responsible in the current decade for about three-quarters of all male Russian deaths at ages 15–54 years and about half of all female Russian deaths at these ages—ie, proportions even greater than in our study population. Whatever the exact alcohol-attributed proportion may have been at particular times since 1980, however, it was always high, has fluctuated greatly, and is still high.