People bring up the "more harm than good" issue, but they're only thinking of one kind of harm: drugs that actually hurt people. But having ineffective drugs on the market also has a cost, and it's probably the bigger one. The market for effective drugs is almost certainly not efficient given the difficulty of ascertaining efficacy and the information asymmetries involved. Thus, there is a real risk of people dying because the inefficient market selects for drugs with good marketing over ones that work.
The other major reason for failure at Stage I/II is that the preclinical models we are using are not a good match for human disease. This means that a drug that works in a mouse model can’t ever work in humans because the mouse model does not reflect the human disease (Alzheimer's is the classic for this). We are using preclinical models we know are useless and are then surprised when the drugs found using them don’t work in humans.
Besides, the proposal is not to approve all drugs easier. Drugs for same conditions would be harder to approve under the proposed model.
If you want to medicine to build on the foundations of history, you need to be able to reference the history.
Well, they ran the bigger trial and it showed no benefit at all. The FDA rescinded conditional approval.
Avastin probably does work for some breast cancer patients if they have a cancer with the right mutations, but if the percentage is a small enough then the large study would not be able to detect this effect in the broad population.
The reason we need very large, expensive, and strict clinical trials is because it's hard to see if a drug actually works or not otherwise - too many confounders. So if we pass drugs through with a higher error rate, we'll likely never know if some of them actually work or not. Maybe not a problem for a disease with no treatment at all, but imagine you have 10 treatments and you suspect that 7 of them don't actually work, but you don't know which 7. Is that better than just having 3 available treatments where only 1 of them doesn't actually work?