I'm more worried about the invalid correlations that will result from this data, given that the users will be self-selected, upper-class individuals.
Participants in research studies are already stupidly self-selected, and many drop out (and are thus invalidated) because it can be incredibly challenging for people to continue to routinely report over a period of time, especially if the participant has to deal with periods of instability in their lives..
Seeing as you can get an iPhone that supports ResearchKit included with a phone contract, I'm having trouble buying the "upper-class" angle here. The iPhone may have started its life as a Toy For The Rich, but today, it's just another "free phone with contract!", side by side with the likes of the HTC One, the Galaxy S4, and the Droid MAXX.
Medical research (especially meta-studies) are already rife with invalid statistics. This can only exacerbate it.
ResearchKit will provide sensor-driven data to researchers, in a readily-used format, on a far denser timescale than is currently practical, via a familiar user interface, across a considerably wider and more engaged sample. If you genuinely believe that the only outcome of this scenario is that it will degrade the quality of data in medical research, then I'm genuinely interested to hear your recommendations for improving the quality of said dataset. Given, of course, the same temporal and financial constraints as exist today.