Magazine
Rethinking Clinical Trials
By Derek B. Lowe
There's an interesting double bind that sometimes happens in R&D. You can end up in a situation where the changes that most need to be made are in the riskiest place to make them. These are high-wire acts like tweaking a metal alloy used in a crucial part or redoing the shape of an experimental airplane wing. It's the equivalent of open-heart surgery: no one does it unless they have to.
In the drug industry, we may be heading to that point in the way we run our clinical trials. There are several problems out there which taken together they suggest that a rethink might be in order. Consider the difficulty in finding enough patients in some indications. If you'd like to run a study with, say, 2,000 treatment-naive breast cancer patients, you could to be in for a long wait. There aren't enough candidates to get that one off the ground easily, at least not in the places that most people are comfortable running a clinical trial. Too many potential treatments are chasing the same pool of candidates, and while you could imagine worse problems, it's still something to be dealt with. This might work out if you're targeting (say) refractory cases, but getting the enrollment to support a first-line treatment is often a major bottleneck.
Then there's the whole question of statistical power. Patients, physicians, and (especially) insurance companies want to see head-to-head comparisons between competing therapies. Companies tend to think carefully before trying one, though, and you can see why: you run the risk of funding a huge study that proves, beyond doubt, that your own drug really is inferior to the competition. And the studies do start to get huge, because meaningful differences for the market can still be subtle enough to require a large sample size. The related danger can't be ignored, either, which is proving (beyond doubt!) that two drugs differ in ways that might not have any real-world significance at all. Examples of both of these can be found in the recent literature, unfortunately, and you just cringe thinking about all the money that was spent.
Finally, as therapeutic areas mature, the number of treatment options make it hard to figure out just what combinations might work best. Look at oncology, with all the various chemotherapy regimens, or cardiovascular disease, with the number of different mechanisms for treating hypertension. If your drug is going to enter a space like that, you're going to want to know how it fits in with the existing treatments, but that can turn Phase III trials into a real ordeal. How many different treatment arms can your study support and still have a chance of coming out with a believable answer?
One potential way around the problem is better patient selection up front, which is a big reason why so much money and effort has been going into the biomarker area the last few years. It's true that stocking your clinical trials with the people most likely to respond should give you better statistics less painfully, but there's a tradeoff. Isn't there always? The better your selection criteria can cut, the better the chances your drug will end up labeled for only those patients in whom it'll work. Now, while this may be the world that we're heading toward, I'm not sure that everyone's ready for it yet. And call me a cynic, but the accuracy of marketing projections doesn't make me feel safe about our ability to deal smoothly with these changes.
Getting a better idea of what our compounds are doing would always help, but odds are that we're not going to be able to do that in a helpful amount of time. The evidence I've seen over the last 10 or 15 years is that plain old failures of efficacy have become one of the most common reasons for trouble in the clinic. We wouldn't have so many of those if we understood human physiology better, but understanding human physiology is clearly a high hurdle to clear.
A more feasible way out could be to change the way we do trials. Some of you can no doubt see what's coming here -- I'm making a pitch for Bayesian statistics. They're not appropriate for every situation, but as far as I can tell, they could be a better fit for the kinds of trials that we find ourselves running. The Bayesian approach would allow continuous enrollment, with a constant eye on the statistical outcome. It also permits different trial arms to expand or contract as their results began to firm up, which could be very useful indeed. Taken to a greater extreme, a fully Bayesian protocol could blur the lines between the three clinical phases as we now know them, as dose-finding shaded into efficacy, which would branch out into different patient populations.
So why isn't everyone doing it this way? Several reasons come to mind. For one thing, people trained in Bayesian methods are thin on the ground. The overwhelming majority of clinicians have never worked under Bayesian conditions, and could find it quite an adjustment. A bigger reason is that people at many companies are probably waiting for someone else to try it first. Anyone who's been around the industry knows that you stand a lot less chance of being fired if you're doing what other people usually do, rather than branching out on your own. (That's one of our major problems, incidentally, and not just in the clinic, but I digress. . .) But the biggest reason is probably regulatory. Every time you fiddle with generally accepted trial designs, the worry is how the FDA will respond.
For its part, though, the Agency has been making interested and enthusiastic sounds about Bayesian trials for some time, and in the medical devices field everyone is getting quite used to them. But many drug companies seem wary, having heard the FDA express enthusiasm in the past for things that didn't quite pan out. As far as I know, Pfizer is the prime example of a large company that's taken the Bayesian ice-water plunge, and their trial never became part of an NDA, because it killed the stroke therapy that was being evaluated. Mind you, it probably killed it off faster and more definitively than a standard trial would have done.
I know that there are people around the industry who are looking at these designs, and I'd like to take the opportunity to cheer them on. My worry is that there aren't enough of them, or enough people looking at other unusual ways out of our current troubles. (People on the business side like the phrase, "thinking out of the box," for this kind of thing, but I've always thought that anyone who uses it often is marking themselves as jammed deep into said box, possibly headfirst). Whatever you call it, we need some, because I don't think we're going to be able to dig ourselves out with the tools that got us here.
