That's the first step. We see it as a guide to internal decision making. We can then pick out those drugs likely to work and focus our resources on them. By eliminating the drugs that don't seem to be working, we're going to increase the likelihood that we can get some innovative medications to patients.

Which therapeutic areas have seen the most progress in the application of this approach?

Diabetes is probably Merck's most mature biomarker state, because glucose is easy to measure. It's very accessible. The linkage between acute changes in glucose and long-term changes in glycemic control are very apparent. It's not hard to convince people if you see a drug that lowers glucose right away that it's probably got a good chance of working chronically. Some other areas are less mature.

Yes, to all therapeutic areas. Certainly some are more technologically challenging. But in many cases, it's more a cultural challenge: It's about learning to become comfortable at making decisions with a little bit less information than we've been accustomed to. Given the odds of success, we really have to look at a lot of different kinds of mechanisms. And very early on in the process, we have to try to pick out those that are likely to have benefit—rather than spending time on things that are not paying off.

What's an example of researchers learning to be a little more comfortable with having a little less information?

Scientists are wired to try to prove things. Particularly at Merck, we pride ourselves on our scientific rigor. Doing absolutely perfect experiments means looking at it from every angle, collecting as much data as you can. To do that requires a substantial effort and substantial resources. The key is, What's enough information? And what's going to give us reasonable certainty that something is—or, more likely, is not—working?

At Merck, we talk a lot about the 80/20 rule. It's about confidence that we're on the right path. There's a small chance we could be wrong, but we don't have to be perfect in that initial decision. If something is working, we can then start to be much more rigorous about getting the proof. It allows us to prioritize which agents we are going to full-court press on.

So is the goal to introduce a lot more compounds into Phase I than in the past, and then, in Phase II, the bar gets raised much higher?

You try to look at as many mechanisms as you can that appear to be working in animals and through all of your preclinical validation. Almost everything Merck ever takes into man looks great on paper—yet at the end, we're left with only an 8 percent success rate. By eliminating the compounds that don't appear to have benefit very early using biological or imaging or even molecular profiling endpoints, the probability that something is going to work in Phase II will be higher.

What kind of mind-set change does the researcher have to go through, and how do you reorganize the team to incentivize that?

People need to have a portfolio mind-set. We need to be asking, What is our aggregate success going to be if we look at a lot of things in a lot of therapeutic areas and take out those that look like losers as fast as possible? This is opposed to thinking, I'm only working on this one project, and I have to keep this project alive no matter what. Ultimately, people begin to see that it's the best thing in aggregate that matters. They might be personally disappointed if an individual program they're working on has to stop, but the overall mission for Merck is to get medicines to patients.