Document 717nyz30VvN1eEQ5JO04Zqv6R

14256997033 2002-02-26 19:09:21 (GMT), page 46 45 hundreds of publications. The researchers did not even consider the possibility that their data was correct and revealed real relationships. It was possible that the workers with the highest exposures died from nonmalignant disease before the latent period for the induction of cancer had been attained. Instead, since they believed an inverse exposure relationship was ludicrous, they manipulated the exposure estimates until the dose-response curve fit their apriori understanding ofthe proper form for the dose-response relationship. The researchers discarded all of the exposure levels that were inversely related to disease. They described this manipulation in an appendix titled "elimination of negative regression coefficients" and proceeded to "revise" the exposure data to make it appear credible and to create a result that would allow them to argue that chrysotile exposure was "innocuous." They described the manipulation as follows: "In all foe conditional regression analyses of the full model i.e. with 13 exposure measures, there was at least one negative regression coefficient, which taken at face value would imply a protective effect of exposure. Years in the highest relevant dust category were pooled with those in the adjacent category and the analysis was repeated. This process was iterated until either all coefficients had become positive, when it was terminated, or until the only negative coefficient was for category I; in that circumstance, category 1 was eliminated from the model, which was equivalent to setting the coefficient to zero and the odds ratio to unity. Admittedly, there was a degree of arbitrariness in some ofthe pooling carried out but every effort was made to retain any `significant' effects."[Emphasis added] (Liddell et al. 1998a) Apparently the QAMA-McGill researchers knew what effects were ` significant' beforethey began the studies and fitted the data make it comport to these pre-determined effects. 45