Getting the Causation Cart Before the Benzene Horse
Let's assume you're trying to prove that benzene causes a host of cancers of the hematopoietic system - essentially all lymphohematopoietic neoplasms. Wouldn't it be clever to argue that the best studies are those that confirm your bias; i.e. that benzene just has to cause acute nonlymphocytic leukemia (ANLL, AML, etc.)? Better yet, wouldn't it be really clever to enforce plaintiffs counsels' demand that, thanks to a little ex post reassessment, studies including subjects prospectively assessed to have been exposed but after the fact, in an instance or two, found to have spent more time in an office, be marginalized or excluded outright? Finally, wouldn't it be brilliant to assume that those with an increased risk of AML just had to have had higher exposures thereby bringing all their similarly situated brethren into the study along with them?
Well, when you don't do experiments but merely look at the literature and run statistical tools over the top of those papers you cherry pick until the stars align (i.e. show that benzene causes everything hematopoietic) it ought not be a surprise that you wind up verifying your preconceptions. And such was the case with "Occupational Benzene Exposure and the Risk of Lymphoma Subtypes: a Meta-Analysis of Cohort Studies Incorporating Three Study Quality Dimensions".
If benzene was indeed some sort of universal blood carcinogen you'd think the statistical evidence wouldn't need to be so tortured before yielding.
