Just read a nice Q&A article from Murtuza Bharmal on the increased use of RWD/RWE.
The point he makes on "ensuring minimal missing data" is certainly critical to the external validity of real world data. Missing data is a big enough/bad enough issue even when dealing with data from seemingly well-designed RCTs, where it frequently causes regulatory and payer bodies to question data validity.
And, in the end, there is only so much statistical twisting and turning that can be done to 'fix' this problem. Data cannot be magicked out of thin air. Even sophisticated statistical counter-measures such MMRM and MI offer little more than educated guesses of what the missing data could/should/might have been.
RWD study designers need to focus more on reducing missing data at source. Failure to so so will inevitably expose their results to similar (or worse) regulatory criticisms!