Epidemiological studies are weak evidence of a causal link between factors (since all they establish is a statistical correlation between the factors and there could be, for example, some third factor which causes the other two so they're only indirectly connected). It is also hard to control for many variables. For example, if you're looking at the relationship between proximity to high power lines and incidence of birth defects, you probably don't know much about the people in your study other than how close they live to high power lines. There could be all kinds of other patterns that influence the rate of birth defects and it's hard to control for them.
I took medical reference in library school. When we were studying Evidence Based Medicine I asked the professor where epidemiological studies fell in relation to EBM and he said that they generally don't count, with an exception made for huge studies that run for many, many years and follow a large and diverse group of people.<br><br>
EBM usually starts off from individual case studies and then gradually climbs up a sort of pyramid, with the double-blind clinical study at the top.<br><br>
I've ended up doubting the whole concept of scientific medicine, because human beings are way too diverse. Take a disease like diabetes. Plan is to line up 200 people with diabetes, split the group randomly in half, give one half a drug and the other half a placebo and carefully monitor the results.<br><br>
Problems: first, 200 people with diabetes don't all have anything like the same condition; second, every individual will react differently to the drug (and to the placebo; third, the people recording the results may not notice the right things about the ways the individuals are reacting (studies are very subject to researcher preconceptions), etc., etc. etc.<br><br>
Epidemiological studies that are done after the fact are excessively problematical because the researchers get to decide what sort of data they will examine. Will they do interviews or just look at the doctors records, for example. If they just look at the doctors records, they immediately proceed to miss all the stuff the doctors missed. On the other hand, if they interview people, stuff can be invented or misremembered.<br><br>
So even before you get to any sort of statistical analysis there are all sorts of ways the material can be tweaked. Exclude certain age groups. Compare populations that aren't really comparable.<br><br>
Suddenly remembered a study done by the obstetricians in Missouri many years ago to prove that home birth was really really dangerous. They included all home births, including accidental (i.e. unplanned and unattended) home births. The death rate still wasn't bad enough, so they counted in miscarriages...<br><br>