Difficulties in Testing the Instrument Strength Independent of Direct Effect Assumption in Mendelian Randomization.