The surgeon general's "epidemiologic criteria for causality". Reply to Lilienfeld.

I argued in my Critique [6] that the Surgeon General's five epidemiologic criteria for causality are not given adequate definition, and that, unless they are so lax as to be meaningless, not even one of them is indisputably satisfied by the evidence relating to smoking and lung cancer. Lilienfeld concedes that "It is tactically unfortunate that the 1982 Report used the same criteria that were adopted in the original 1964 Report" [4], but he is unable to demonstrate that the available evidence, interpreted according to the accepted canons of scientific inference, is consistent either with the original or the "more detailed criteria". He makes no acknowledgement of the principle that valid inferences about cause cannot be drawn from epidemiologic comparisons between selected, non-randomized groups; he actually violates that principle in his defence of the Report by deploying conclusions from such evidence. As to the vexed 85% of lung cancer cases and the 30% of all cancer deaths in the U.S. attributed to smoking [5], no justification is offered for assigning the entire "relative risk" to a (smoking) causal relation; indeed at one point, and perhaps inadvertently, Lilienfeld repudiates that practice. I know of no valid arguments from reliable evidence that justify such precise estimates of attributable risk and if Lilienfeld cannot supply them who can? If specific claims about the hazards of any agent cannot be sustained by scientifically rigorous methods they should never be made publicly in the first place but, once having been made and exposed, they should then be retracted with as much grace and dignity as possible.