Tobacco Documents Online

540 -3ump from association to causality is invalid. 'Indeed, .the self-selection bias itself (not to mention all"the other possible sources of error) is sufficient to cast doubt on :- the causality claim: The main prospective (i.e.,"follow-up) studies con- cerning smoking and its relationship to mortality and morbidity from certain diseases (as discussed in the latest Surgeon General reports) do not address the issue of causality. The claim that the conclusions reached are "impressively uniform and consistent" does not mean that a causal relation- ship has beer, established, but only that approximately the same observed associationskeep appearing. In fact, the biases inherent in these'studie"s preclude the right to claim that causality has been demonstrated. Actually, one could argue that the same associations keep appearing because the '" same biases are present in each study. The fact that so many studies have produced a positive association between smoking and lung cancer has led many people to make the false conclusion that "quantity means quaTity". The truth of the matter is that repeatability (i.e., the ability to produce the same estimated association) does not imply accuracy (i.e., the ability to produce a valid -10- I estimate of the true ass study free from all.bias the true smoking-lung ca a thousand biased studie distorted estimate of th In summary, base pertinent to the smoking . r; 1 of the opinion that a ca lung cancer has not been

is invalid:`'Indeed,' to mention a11 the other cient to cast doubt on '" r- _~::1C _ _: ••. _ ..... _~ .,r~s_ follow-up)'studies•con- p to mortality and ` discussed in the latest :ess the issue of causality. aed are "impressively an that a causal relation- that approximately the iring. In fact, the 3clude the right to claim i. 'Actually; one could ` 3p appearing because the ' ly. ~s have produced a positive ; Cancer'has'1ed many tbat`+'quantity means s ~that'repeatability me~'-estimated association) .bilitq to'produce a valid 541 estimate of the true association). In this regard, one study free from all bias and producing a valid measure of the true smoking-lung cancer association is worth more than a thousand biased studies, all of which provide the same distorted estimate of the true association. In summary, based on the currently available information pertinent to the smoking-lung cancer causality issue, I am of the opinion that a causal relationship between smoking and lung cancer has not been scientifically demonstrated. Lawrence L. Kupper Professor of Biostatistics