Background: The existence of a U-shaped relationship between alcohol consumption and all-cause mortality complicates the calculation and interpretation of mortality attributable to alcohol consumption.
Methods: We used the relationships between all-cause mortality and alcohol consumption from four British cohort studies. For each study we defined a 'sensible drinking level' asthe level with lowest observed mortality. We estimated the fractions of deaths that were attributable to (1) any drinking (compared with not drinking), (2) drinking more than the 'sensible level', and (3) drinking less than the 'sensible level'.
Results: Data from the Doctors' study suggest that on balance 22.3 per cent of deaths are prevented by alcohol consumption, yet the fractions of deaths attributable to drinking more than 8-14 units per week and less than 8-14 units per week are nearly equal (6.5 per cent and 6.4 per cent, respectively). In a sensitivity analysis we show that it is possible for alcohol consumption to prevent deaths overall yet for more deaths to be attributable to drinking above a sensible level than are attributable to drinking below the sensible level.
Conclusions: The balance of deaths attributable to or prevented by alcohol consumption provides no information about the deaths attributable to drinking above or below sensible levels. Using all-cause data in this way is likely to exaggerate the protective effect of alcohol consumption, so our results are only illustrative.