Data mining for clinical research - why so little use of machine learning?

Actually, It's pretty obvious why data mining is NOT part of *clinical trials*. In clinical trials you are NOT trying to discover something new, you are trying to test a hypothesis (that treatment A is better than B, or something like this). My question was about clinical research: why are so few new and meaningful hypothesis generated through machine learning? Why does t-test, multiple logistic regression and generalized linear models so dominant in clincal research? It may be difficult to provide a comfortable answer to the question.