A recent interview in Nature Reviews Drug Discovery with John Powers, formerly of the FDA, points out some problems in designing antibacterial drug trials. Some of these are unique to this area, although others we're stuck with wherever we go.

For one thing, it’s surprisingly hard to make sure, when you’re selecting patients, that the people you’re letting into the trial have the disease that you’re trying to treat. The example used is that some 5% of the patients who present with cough actually have pneumonia. Pneumonia is a very good disease to treat with antibacterial drugs, but you’d better make sure that your patients actually have it. There are some tests available to make sure that a given pathogen is present, although they aren’t available in every case you’d want them to be. If you don’t have such a screen, you risk having a very heterogeneous patient population, which will likely as not obscure the effectiveness of the drug you’re testing.

Then there’s the related difficulty in treating some conditions that you’d think would be clear cases for antibacterials: ear infections, for example. The problem is, it’s surprisingly hard to show benefit for some of these things with existing drugs. The underlying infection may be hard to get to (poor circulation in the infected area), or it may be an intrinsically heterogeneous condition like sinusitis. (That can be the result of umpteen different sorts of bacteria, or it could well be something viral, or several varieties of fungal infection, or allergies, what have you). There’s no point in running a head-to-head with an existing medication in these cases; you should run against placebo. That'll be enough of a challenge.

Another problem is that some of the bacterial diseases progress rather quickly – ahead, in some cases, of our ability to usefully diagnose them. That presents a real challenge for a clinical design, one that is dealt with, in many cases, by not attempting to gather rigorous clinical data under these conditions at all. In this field, diagnostic tools have to be fast if they’re going to be of much use.

There are two sides to all these problems: not only do you want to get the drug to the people who need it (and who will respond to it) the most, you want avoid giving it to people who won’t respond at all. That’s not just for the reasons given above (it’ll mess up your data), although that’s enough all by itself. No, the other problem is that spreading your drug around to inappropriate patient populations will just bring on resistance even faster. That’s going to happen no matter what, of course – the key is to have it happen as slowly as possible.