Our brief survey in a large supermarket to assess their alcohol contents of a random selection of wines (table) shows that the actual number of units consumed in wine could be underestimated by as much as 37.5% per bottle for some types of wine, counting only 6 units of the possible 9.6 units consumed. This would have the effect of placing individuals in lower intake groups than they perhaps ought to be, and attributing to these groups health effects that are actually related to a higher intake. The same concerns could also be true for estimated intake from spirits and beer. If home measures have been taken then this could underestimate the actual consumption because home measures tend to be larger than a single standard pub measure. Many beers contain more than two units of alcohol per pint, which could result in a large underestimate of intake if much beer was consumed. This calls into doubt the reliability of the alcohol consumption figures and could be why the U shaped or J shaped mortality curve was not observed. This concern also brings into question the validity of the other conclusions made in the article.

Grønbæk, Morten (Danish Epidemiology Science Centre at the Institute of Preventive Medicine, Copenhagen University Hospital, Kommunehospitalet, DK-1399 Copenhagen K, Denmark mg@ipm.hosp.dk).

Editor—The study by Hart et al apparently contradicts the vast number of prospective cohort studies which have shown that one or two alcoholic drinks per day is the optimal intake with regard to risk of coronary heart disease.^1-1,1-2 In their study, however, drinkers of one or two drinks per day also had a 20-25% reduction in mortality from coronary heart disease compared with non-drinkers. Adjustment for confounding factors decreased the risk from 0.87 to 0.79, indicating that alcohol may have a causal role in decreasing risk of death from coronary heart disease. A likely reason for the statistical insignificance of this result is that the study is quite small.

Misclassification may explain the inconsistency in mortality from coronary heart disease between the categories in the lower range of intake. The finding that the risk of alcohol related causes of death increases at an incomprehensively low level (1-7 drinks per week) suggests that a certain underreporting took place. The extremely long follow up time may have added to the poor prediction of both coronary heart disease and alcohol related causes of death among people who changed their drinking habits with age.^1-3 Further, this paper does not deal with type of alcohol. Several ecological studies, as well as prospective cohort studies, have found that wine drinkers are better protected against coronary heart disease than are drinkers of beer or spirits.^1-4,1-5 What do Scottish men drink?

Hart, Carole L (Department of Public Health, University of Glasgow, Glasgow G12 8RZ). Smith, George Davey (Department of Social Medicine, University of Bristol, Bristol BS8 2PR). Hole, David J (West of Scotland Cancer Surveillance Unit, University of Glasgow). Hawthorne, Victor M (University of Michigan, School of Public Health, Ann Arbor, MI 48109, USA).

Editor—The Ashton-Keys are concerned about the alcohol content of wine and possible underestimation of drinking habits because of the way this was carried out in our study. This is not a major issue in this cohort since only 5.8% of men were wine drinkers and the wine drunk in the early 1970s was likely to be of lower alcohol content than today. A nomogram for calculating alcohol content of different beverages in 1970 gave the median value for wine as 10% by volume.^2-1 Recalculating the relative rates of mortality after classifying the wine consumption as nine units per bottle rather than six did not alter the results. With regard to beer, our study was established before the widespread introduction of high alcohol beers, in particular lagers. Though we agree with the comment on home measures being larger than standard pub measures, this will be true for all self reported studies on alcohol.

Grønbæek comments on our study's finding of a statistically non-significant reduction in mortality from coronary heart disease for “moderate” drinkers. However, this effect was not seen in the lower category of 1-7 units per week, making it difficult to draw any firm conclusions. The relative rates of mortality for alcohol related causes were not statistically significant at levels below 15 units per week, so again no firm conclusions can be drawn from these findings. Under- reporting may indeed have taken place, but our intake data were similar to other UK studies of the time. Additionally there were high correlations between alcohol reported at the initial screening and at the second screening, four to seven years later, attended by about half of the cohort. This suggests that the alcohol reports are reliable, as do the graded associations between reported alcohol consumption and both blood pressure and triglyceride concentrations.

Beer, spirits, and wine consumption were analysed separately and gave similar results to the analysis of total units of alcohol, but with less power. Of the total units of alcohol reported, 69% were beer, 28% spirits and 4% wine.

Other commentators on our paper considered that residual confounding by smoking could explain our findings with respect to stroke^2-2,2-3—that is, the men who drank more smoked more, and our measures of smoking were imperfect so we could not take this into account. If this were the case then it would be expected that alcohol would show an independent association with lung cancer mortality. However, as the table shows, this is not the case; while alcohol consumption shows a crude relation to lung cancer mortality, adjustment for smoking and social factors essentially abolishes this, unlike in the case of stroke (see table 2 in the original paper).^2-4 Clearly, as lung cancer is considerably more strongly related to smoking than is stroke, residual confounding by smoking would be more clear for lung cancer than stroke as a cause of death.