Good discussion, and a good reference.

Re DSimon's:

My general position has been that it would be better for the US public if the FDA even more strongly regulated against treatments which do not go through the full clinical trial gamut with high marks. I don't like it when people who are in crisis are vulnerable to getting ripped off by snake oil salesmen.

FDAReview says:

If the U.S. system resulted in appreciably safer drugs, we would expect to see far fewer postmarket safety withdrawals in the United States than in other countries. Bakke et al. (1995) compared safety withdrawals in the United States with those in Great Britain and Spain, each of which approved more drugs than the United States during the same time period. Yet, approximately 3 percent of all drug approvals were withdrawn for safety reasons in the United States, approximately 3 percent in Spain, and approximately 4 percent in Great Britain. There is no evidence that the U.S. drug lag brings greater safety.

(This was actually something of a surprise to me. My wife has been on a couple of medications which have since been withdrawn, so I've been getting increasingly uncomfortable with the general level of safety of pharmaceuticals, so I'd been leaning in DSimon's direction on this - but this evidence says that increased scrutiny from the FDA hasn't been helping)

The section on off-label uses of drugs is also very persuasive:

Yet any textbook or medical guide discussing stomach ulcers will mention amoxicillin as a potential treatment, and a doctor who did not consider prescribing amoxicillin or other antibiotic for the treatment of stomach ulcers would today be considered highly negligent. Off-label uses are in effect regulated according to the FDA’s pre-1962 rules (which required only safety, not efficacy), whereas on-label uses are regulated according to the post-1962 rules.

I'm not sure I'd quite agree with

entirely abandon restrictions on what drugs / treatments doctors can prescribe.

The phase I trials, looking for human toxicity, still sound reasonable. To my mind, it does look like the efficacy trials should be moved out of the FDA - basically crowd-sourced as post-market data gathering.

I agree with

Require all patient data to be public (once anonymized).

It would help if as many groups as possible have the opportunity to dig through the data as possible. One caveat/suggestion: Epidemiological studies tend to be terrible at giving solid conclusions. Double blind randomized studies are able to cancel out far more of the confounding variables. I suggest that, for any medical decisions that are anywhere close to a 50:50 decision, that an incentive be offered to explicitly randomize the decision and record that fact, along with the other outcome data on the case. Where there is uncertainty anyway, this won't hurt the participating patients on average anyway, and it would embed a continuous stream of randomized trials in the available data.