A good discussion from In the Pipeline about Andy Grove's proposal for changing how drug development works:
Once safety is proven, patients could access the medicine in question through qualified physicians. Patients' responses to a drug would be stored in a database, along with their medical histories. Patient identity would be protected by biometric identifiers, and the database would be open to qualified medical researchers as a "commons."
Derek Lowe talks about problems with selecting patients, potentially swamping out clinically relevant results because of too much patient and treatment heterogeneity, and so on.
Another thing that's relevant is that "safety" is often relative to potential benefit, and there's no clear line that says "these are safe" and "these are not." Many cancer treatments, for example, are often just poisons that kill cancer faster than they kill you, but their potential benefit is generally worth the harmful side effects. You always have to consider how well something might work to consider whether those risks are worth it (e.g. might be worth it for cancer, but maybe not so worth it for stopping hair loss), so deciding in advance whether something is safe before you ask how well it works can potentially rule out things that are harmful, but lead to a great enough benefit that they could be worth trying out.