Risk function had peculiar propertiesBMJ 1995; 311 doi: https://doi.org/10.1136/bmj.311.7013.1167 (Published 28 October 1995) Cite this as: BMJ 1995;311:1167
- John Duffy,
- Geoff Cohen
- Director of statistics Alcohol Research Group, Department of Psychiatry, University of Edinburgh, Edinburgh EH10 5HF
- Lecturer in medical statistics Department of Public Health Sciences, University of Edinburgh, Edinburgh EH8 9AG
EDITOR,--Morten Gronbaek and colleagues' study of mortality associated with moderate intakes ofwine, beer, or spirits presents relative risks of death as functions of consumption of each type of beverage.1 Apparently none of the first order interactionsbetween levels of consumption of different beverages were significant. Ignoring interactions, however, leads to a model with peculiar properties, for it implies that one can simply multiply the relative risks for each type of beverage.
Thus the authors' estimates imply that a person drinking three to five drinks daily of each of wine, beer, and spirits (a total of nine to 15 drinks daily) has a relative risk of death from coronary heart disease of 0.43 compared with someone who never drinks alcohol. For deaths from other causes the relative risk between these two extreme groups of people is 0.82. Such estimates are hard to take seriously, so the model clearly cannot be extrapolated this far. Even at lower consumptions for the separate beverages, however, an absence of interaction seems unlikely.
Since one of the study's main conclusions rests on the upturn of the risk function for the highest category of consumption of spirits it seems important to ensure that the model estimates are valid not just for moderate consumption but for the highest category considered. We wonder how sensitive the conclusion about spirits is to the assumption of no interaction.