Al's Comment:

 I (along with my team:) published an article in the journal: "Artificial Intelligence in Medicine".  It is about our brain tumor virtual trial. We compared our data to the data from 7 randomized trials around the same time and found that we could have predicted the outcomes in 6 of the trials by using our data. For the 7th trial that was close but not perfect, we went back to look why we missed - we had different patient populations - they had recurrent GBM patients who never had chemotherapy before which is really strange. All of our patients failed Temodar up front, so when we looked at recurrent patients, we got a different readout.   Bottom line, what we did for basically free probably could have replaced trials costing millions of dollars.

  That was our old version of the registry (started in 1993).. biomarkers were missing.  Our new current version (run by our partners at xCures) has complete patient data so should be much better. I would say could possibly be better than a randomized trial if we match for all known prognostic indicators!  

Virtual Trials: Causally-validated treatment effects efficiently learned from an observational cancer registry