Yeah, but the tobacco studies took decades and the statistical techniques were more focused. And even with the certainty of the link between cigarettes and lung cancer, I still look at that graph and wonder, yeah, but how much of that decline is because modern medicine got better at treating lung cancer? What would the graph of lung cancer illnesses look like? Statistics are funny things... ;-)

With big data sweeps like the Alzheimer's one, it's too easy for researchers to find the correlation first and then use it to establish the hypotheses - which basically invalidates the finding unless they partition the data properly. Not saying that's what happened here, just that I'm cautious about accepting the results from this sort of study.