I’ve just skimmed this - saving for later for a proper read, but it looks really impressive and solid, and is certainly very timely.
However, I’m always struck by the state we’re in with health data that this kind of work needs to happen relatively frequently - one findings is participants expected health data to be used for research that is “in the public good in the first instance”. Well, no sh*t Sherlock. 
But who defines “public good”? What are the controls? What is the recourse for the public? That feels like the important thing - but it is also the political one. And that’s the stuff that needs work, and courage, and infrastructure to support.
Data and automated decisions are often used to opaquely centralise the power of bad actors, but that kind of scrutiny doesn’t arise in this kind of granular project.
