Background. Prospective studies of alcohol and mortality in middle-age
d men almost universally find a U-shaped relationship between alcohol
consumption and risk of mortality. This review demonstrates the extent
to which different studies lead to different risk estimates, analyses
the putative influence of abstention as a risk factor and uses availa
ble data to produce point and interval estimates of the consumption le
vel apparently associated with minimum risk from two studies in the UK
. Method. Data from a number of studies are analysed by means of logis
tic-linear modelling, taking account of the possible influence of abst
ention as a special risk factor. Separate analysis of British data is
performed. Results. Logistic-linear modelling demonstrates large and h
ighly significant differences between the studies considered in the re
lationship between alcohol consumption and all-cause mortality. The re
sults support the identification of abstention as a special risk facto
r for mortality, but do not indicate that this alone explains the appa
rent U-shaped relationship. Separate analysis of two British studies i
ndicates minimum risk of mortality in this population at a consumption
level of about 26 (8.5 g) units of alcohol per week. Conclusions. The
analysis supports the view that abstention may be a specific risk fac
tor for all-cause mortality, but is not an adequate explanation of the
apparent protective effect of alcohol consumption against all-cause m
ortality. Future analyses might better be performed on a case-by-case
basis, using a change-point model to estimate the parameters of the re
lationship. The current misinterpretation of the sensible drinking lev
el of 21 units per week for men in the UK as a limit is not justified,
and the data suggest that alcohol consumption is a net preventive fac
tor against premature death in this population.