How would you even do a randomized study? It's not like you can pick 1000 people at random and tell them "Thou shalt cycle to work this year" and pick 1000 people at random and tell them "Thou shalt drive to work this year" and count how many of each group survive, and even if you do, the sample would be way too small to draw any definite conclusion, and the duration might be too short to show any long-term health effects of cycling. In some cases you have to resort to correlation studies, even if you can't tell whether cycling makes people healthy or health makes people cycle or something else does both or some combination of the above.