One problem some studies suffer is over-control of confounding variables, which excludes secondary causal relationships. For example, if you control for health

One problem some studies suffer is over-control of confounding variables, which excludes secondary causal relationships. For example, if you control for health of arteries, you will find that (insert bad food here) known culprits for heart disease stop causing heart disease -- which would seem absurd. That's because artery health and heart disease are correlated (in this case causally related), so when you control condition one on the other, any causes vanish (completely or partially). In this case in particular they control for BMI. If eating less meat is helpful in lowering BMI, this effect isn't captured by this result. So you have to watch out for this when taking conclusions from a study or making decisions (if you already have a low BMI then you can trust the conclusions better).