We have a new phrase to toss around in in the industry: "Phase Zero". That's what they're calling a recent trial of an anticancer drug from Abbott (ABT-888), which was tested in humans before any safety dosing (Phase I) had been done.
So, how exactly can you do that? By giving extremely small amounts of the drug, that's how, and looking to see if you can detect a change in some marker for eventual efficacy. In this case, the marker was inhibition of the activity of PARP, poly(ADP-ribose) polymerase, which is involved in the cellular response to DNA damage. Inhibiting it should make cells much more likely to die once such damage had been detected, which one of many such signals that cancer cells tend to ignore under normal conditions. Abbott's drug seemed to do the trick, so work on it will continue.
The good part of this is that the drug got into humans more quickly than usual, and that its mechanism of action has now been verified (to a first degree of approximation, anyway - it hits the target). This should make a company a bit more confident about moving on to larger trials, and could potentially weed out losers early in the game.
But there are bad parts, too. For one thing, the patients in a phase zero trial have no hope of benefit from the drug: the dose is just too small. The small doses could give results that (for better or worse) aren't relevant to the later real-world ones, too. Another problem is that reliable biomarkers are thin on the ground despite great sums of money being spent to find and validate them. If you're going to let the future of your drug ride on one of these trials, you'd better be confident that you know what it's telling you. (And if you're not going to let the future of the drug ride on a phase zero trial, why are you running one, eh?)
What would be worth knowing is how many drugs fail because of lack of effect on their intended target, as opposed to those which hit the target but still have no effect. You'd also want to know: of that first group, what portion are going to be amenable to robust biomarker studies. Those two fractions would tell you how much of an impact this whole idea will have. Right now, I think the error bars are way too large to make a prediction. . .