Everyone in the industry would like to do something about the failure rate of drugs in clinical trials. It would be far better to have not spent the time and money on these candidates, and the regret just increases as you move further down the process. A Phase I failure is painful; a Phase III failure can affect the future of the whole company.

So why do these drugs fall out? Hugo Kubinyi, in last August's Nature Revews Drug Discovery suggests that it's not for the reasons that we think. As he notes, there are two widely cited studies that have suggested that a good 40% of clinical failures are due to poor pharmacokinetics. That area is also known in the trade as ADME, for Absorption, Distribution, Metabolism, and Excretion, for the four things that happen to a drug once it's dosed. And we have an awful time predicting all four of them.

Of the four, we have the best handle on metabolism. In the preclinical phase, we expose compounds to preparations from human liver cells, and that gives a useful guide to what's going to happen to them in man. We also expose advanced compounds to human liver tissue itself, which isn't exactly a standard item of commerce, but serves as a more exacting test. Most of the time, these (along with animal studies) keep us from too many surprises about how a compound is going to be broken down. But the other three categories are very close to being black boxes. Dosing in dogs is considered the best model for oral dosing in humans for these, but there are still surprises all the time.

That 40% figure has inspired a lot of hand-wringing, and a lot of expenditure. But Kubinyi says that it's probably wrong. Going back over the data sets, he says that the sample set is skewed by the inclusion of an inappapropriately large group of anti-infective compounds with poor properties. If you adjust to a real-world proportion, you get an ADME failure rate of only 7%.

Now, when this paper came out, I think that there was consternation all over the drug industry. (There sure was among some of my co-workers.) The ADME problem has been common knowledge for years now, it was disturbing to think that it wasn't even there. So disturbing, it seems, that many people have just decided to ignore Kubinyi's contention and carry on as if nothing had happened. There have been big investments in ways to model and predict these properties, and I think that many of these programs have a momentum of their own, which might not be slowed down by mere facts.

The natural question is what Kubinyi thinks might be our real problem. In his adjusted data set, 46% of all failures result from lack of efficacy in Phase II. He admits that some of these (in either approach to the data) might still reflect bad pharmacokinetics, but still maintains that poor PK has made a much smaller contribution than everyone believes. Here's his drug development failure breakdown, which makes his point:

46% drop out from lack of efficacy
17% from animal toxicity (beyond the usual preclinical tox)
16% from adverse events in humans
7% from bad ADME properties
7% from commercial decisions
7% from other miscellaneous reasons