The New York Times has a good article this week on a trend in clinical trials that's been developing for several years - small intensive trials in humans, run before giving the go-ahead for the real thing.
It makes a lot of sense, but only when you can use it to ask (and answer) the right questions. That's where technologies like functional NMR imaging or PET scans come in, because they allow you access to in vivo data that's otherwise unobtainable. Take, for example, the studies mentioned in the Times article, where they look at glucose uptake in a solid tumor. That's a reasonable proxy for its metabolic activity, as you'd guess, and it'll give you a quick read on whether your targeted cytotoxic compound is having the effect you want.
What you'd do, normally, is dose the compound for days or weeks, then use NMR or another imaging technique to see if the tumor has changed size. That's clearly a more convincing answer, but it takes a more convincing amount of time and money to get it. And if your compound isn't having an effect on a fast marker like the tumor's metabolic rate, it's probably not going to have any effect after you dose it for two months, either. You're better off trying something else.
But if your new cancer therapy is, say, a compound that interferes with cell division, then you're not going to have that clear an answer through that glucose uptake technique. Same problem if the cancer you're treating is a more diffuse one like leukemia, because there's not such a clear tissue to image. (There are other approaches to each of those problems, naturally, but I just wanted to emphasize that each clinical trial is its own set of new problems, even inside the same general therapeutic area.)
And even when you get to the traditional large-scale trials, there's a huge need for surrogate markers that can show progress against slow-moving diseases. Glycosylated hemoglobin as a measure of efficacy in diabetes is a good validated example. It still takes quite a while to establish (weeks or months of dosing), but that's like lightning compared to the progress of diabetes complications themselves. You can do a quick assay in this field - the oral glucose tolerance test - but the improvement in that assay isn't so quick to come on.
The CNS diseases are a real clinical challenge, which is why their trials are so brutally expensive. There are hardly any markers at all for most of them. Everyone would love to have a short-term noninvasive readout for Alzheimer's, but despite years of effort, no one has quite made it. (And that's despite the definition of "short-term" in Alzheimer's is rather permissive.) Similarly, it would be good to be able to get a faster readout on depression, whose therapies are notorious slow starters.
There's a bigger problem, though, looming over some of the generally accepted markers - what effect do they really have on long-term mortality and morbidity? Glycosylated hemoglobin has been pretty well correlated in diabetes over the long term, so that one's pretty safe. But the question is worth asking, for example, about HDL and LDL levels. Yes, things do line up well, up to a point. But does long-term administration of statin drugs, say, help as much as we'd like to hope it does over, say, twenty years? The jury's still out on that one.