Studies were only selected if they were cohort studies in industrialised countries that related all cause mortality for men and women separately or if they reported more than three levels of alcohol consumption.

Relative risks of all cause mortality, adjusted for age, and where possible, other risk factors including socio-economic status, education, smoking, and physical health, were extracted for each level of alcohol consumption in each study. Amounts of alcohol were converted to units per week (1 unit of alcohol = 9 g).

21 articles reporting 20 studies (1, 210 545 participants) were included. 60 224 deaths among men (13%) and 74 824 deaths among women (10%) occurred. In men, 12 of 19 estimated nadirs were <10 units/week and 3 were >20 units/week. In women, 4 of 8 estimated nadirs were <5 units/week and 1 was >10 units/week. Nadirs varied between countries but not within countries. Nadirs for men and women pooled at the country level are in the table. Nadirs were estimated for men from 10 US studies and 3 UK studies. 6 other countries only contributed 1 nadir each. Nadirs were estimated for women from 7 US studies and only 1 non-US nadir was contributed. When adjustment was made for smoking, blood pressure, exclusion of participants with baseline coronary heart disease, and exclusion of ex-drinkers, the nadir did not change by more than 3 units/week. 6 studies that provided results by age group showed no relation between nadir and age; the pooled estimated nadirs were appropriate for ages 5080 years.

The review concluded that the level of alcohol consumption at which mortality is the lowest (nadir) is between 7.7 and 12.9 units of alcohol per week in men ages 5080 years .

Dr.Caan of the Society of Health Education and Health Promotion Specialists, Cambridge, UK, commented on the research :' White suggests that it is possible to estimate the safest levels of consumption, at least for men, in the 5080 year age group in the US and UK. One methodological problem with this review is that it combines studies using different measures of alcohol intake. Weekly consumption may not be the best measure to study mortality. For example, a study done in Russia showed that a pattern of weekend binges may predict alcohol related deaths.1 A life history perspective often helps because people at risk of alcohol related deaths vary their consumption over time.2 In practice, it seems that questions on participants 'usual consumption' do not produce the same result as keeping a drink diary for a specific week.Despite these methodological issues, most studies produced a U-shaped risk curve. In other words, participants drinking a moderate amount of alcohol per week have lower mortality rates than either abstainers or those drinking higher levels of alcohol. 2 studies with younger participants (<30 y), however, did not show this U-shaped curve. Instead, these studies showed a monotonic increase in mortality as intake rose, similar to a more recent study of UK men.3 Practitioners judging the potential for a protective or harmful action need to consider more than the dose response relation between alcohol and mortality.'

References 1.Chenet L,et al. Alcohol and cardiovascular mortality in Moscow; new evidence of a causal association. J Epidemiol Community Health 1998; 52:7724.2Finney JW.et al The course of treated and untreated substance use disorders: remission and resolution, relapse and mortality in Addictions: a comprehensive guidebook. New York: Oxford University Press, 1999;3049. 3.Hart CL et al. Alcohol consumption and mortality from all causes, coronary heart disease, and stroke: results from a prospective cohort study of Scottish men with 21 years of follow up. BMJ 1999;318:17259 Contact:Mr I R White, Medical Statistics Unit, London School of Hygiene and Tropical Medicine, Keppel Street, London WC1E 7HT, UK. Fax +44 (0)20 7637 2853.