Here's a good retrospective in JAMA, from the FDA, about what's happened when the agency has rejected a new molecule entity (NME). The authors look over the data set from 2000-2012 to see what the most common reasons for trouble were, and what happened after that in each case.
Overall, 302 new molecules were submitted during those years. Half of them were rejected the first time through - a figure that does not jibe with most investors' ideas of the prospects for any given drug, for sure. But Tolstoy was right - approved drugs are all alike, but every rejected one is rejected in its own way. Looking over those, it appears that a bit under half of the rejects eventually did get approved (meaning that, ultimately, about 73% of all NMEs do eventually make it). But the median delay to approval, after that first rejection, was well over a year.
And that has to do with the reasons for rejection. They vary, and none of them are easily fixed:
Of the unsuccessful first-time applications, 24 (15.9%) included uncertainties related to dose selection, 20 (13.2%) choice of study end points that failed to adequately reflect a clinically meaningful effect, 20 (13.2%) inconsistent results when different end points were tested, 17 (11.3%) inconsistent results when different trials or study sites were compared, and 20 (13.2%) poor efficacy when compared with the standard of care.
But you'll note that while some of these are just nasty results in the clinic (like insufficient efficacy), some of them could also have presumably been mitigated a bit. You can see how crucial good trial design is when you note how many problems there are around endpoint selection and dosing. Now, it's also true that well-designed well-powered trials are invariably more expensive than most of the alternatives, and smaller companies are often running on fumes by the time they get to the end, but compare the pain of spending the time and money up front to the pain of a Complete Response Letter.
Taking a look at the ones that never did get approved shows a definite bias towards efficacy problems. Overall, safety concerns were similar between that group and delayed-approval group, but lack of efficacy was a problem with 76% of the never-approved drugs, almost double the rate of the delayed approvals. So the take-home is that you may be able to rescue a drug with safety concerns, given time and money, but there is likely to be no cure for poor efficacy.
And that's just what I think the effect of the FDA regulatory process should be. Once in a while, you hear calls for the agency to just let everything through as long as it's shown safety, to let the medical community and the marketplace sort it out. And although I'm a laissez-faire kind of guy, I think that would be a recipe for disaster. You'd have people selling repackaged barbecue sauce in gel caps as a cure for cancer. I'd rather have the agency focus on efficacy, and even let some drugs through that have some safety concerns, as long as they're known (as much as practical) up front and monitored thereafter. Just as long as they work against a disease.