Tuesday, May 17, 2011
Antibiotic Trial Design and FDA - a Discussion
The current predicament of antibiotics within FDA can be traced back to the scandal around Ketek. See my earlier blog on this. The FDA has been backpedaling on antibiotics ever since. Clearly the FDA is caught between congressional critics out for votes and the infectious disease community clamoring for new antibiotics to treat resistant infections. Most of FDA’s recent guidances for carrying out trials for antibiotics have required infeasible trial designs – catering to congressional pressure demanding a more strict approach to the non-inferiority trials we must use for antibiotics. Of course the fact that requiring infeasible designs leads to an empty antibiotic pipeline is apparently not considered by congress or the FDA. The one exception among recent guidances has been for skin infections where the required design is feasible.
Starting last week and going through the weekend, I was privileged to be able to participate in a stimulating and reasoned email discussion on the FDA and antibiotic development. Participants will remain anonymous, but included captains of industry; participants in the Foundation for the National Institutes of Health discussions on trial design issues with the FDA, the Infectious Diseases Society of America (IDSA) and industry; active members of IDSA; academics including experts in PK/PD and pharmacometrics; and finally, most humbly and not anonymously, myself. Keeping the discussants anonymous allows me to summarize the discussion in my own way without having to have the drafts reviewed by multiple people who don’t always agree with each other or with me. I thought that could be a never-ending process.
The discussion was initiated by several emails noting recently presented results of phase II trials where the FDA’s new early endpoints of halting lesion progression for acute bacterial skin and skin structure infections (ABSSSI) were being compared with more traditional late endpoints of clinical cure. It is clear that there is not always a one-to-one correlation between success at the early endpoint and clinical cure – no surprise to anyone. The participants launched into a discussion on whether the early endpoints were or were not clinically relevant and whether they would predict the more traditional and clearly relevant endpoint of clinical cure in a reliable enough way. Of course, this discussion could also be relevant to the early endpoints now being demanded for pneumonia trials as well – hence the importance of the exchange. We then began to try to look into the future . . .
The FDA has opted for these early endpoints because they tie back to measurements made during the pre-antibiotic era when sulfonamides were being compared to placebo or placebo-like treatments (UV light being an example of the latter for skin infections). In these 80-year-old studies, the treatment difference between sulfonamides and UV light in skin infections was greatest early in the course of therapy. Therefore, the FDA has a way of tying modern therapy back to a treatment effect demonstrated 80 years ago. Critics of this approach note that the patients are not the same, the pathogens are not the same, the disease is not the same and the treatments are not the same anymore – so why should we tie ourselves to an 80-year-old trial design that was not really placebo controlled anyway?
Critics also charge that the early endpoint is clinically irrelevant. The only relevant endpoint is cure of the infection that is measured at 7-14 days when the patient should have returned to his/her baseline status. Pragmatics are so relieved that they have a trial design from the FDA that is actually feasible, that they are happy to carry out trials and obtain approvals based on the early endpoints while making clinical cure a secondary endpoint. Of course, most clinicians would agree that this approach is bass-ackwards. The early response is secondary while cure is primary. It is clear that the early response is already included in the consideration of cure anyway since a patient not responding to treatment in the first few days of therapy will be considered a therapeutic failure anyway and the astute clinician will start looking for the cause of failure. But the converse – early success – may not always predict cure. Time to response is valuable to the clinician, but not pivotal information.
One way around the entire disagreement is to use more modern approaches such as pharmacometrics to define effect size both early and late. In this way, one could use the treatment effect size to justify a non-inferiority margin for an endpoint at time of cure. One could also justify a composite endpoint combining early and late response or one could even define margins for two endpoints, early and late (although that would take some non-traditional statistical approach I expect). Unfortunately, at least so far, the FDA has been deaf to such arguments.
These arguments are equally applicable to trials for community-acquired bacterial pneumonia where, in the current guidance, the analysis population must be the bacteriologically documented one, the endpoints are again early resolution or improvement in signs and symptoms (but not cure), no prior antibiotics are allowed, and the margins are 15% for parenteral therapy and 10% for oral therapy. This means that an oral only drug will never be developed for pneumonia since that trial requires 5000 patients and is simply infeasible. Even a parenteral drug will require a diagnosis rate of at least 50% which would be a world record and nothing short of miraculous for a modern trial in order for that trial to be within the realm of possibility. But, as soon as we start talking about using pharmacometrics to define a treatment effect for cure, we can at least get to a relevant endpoint with a reasonable non-inferiority margin. Again – the FDA is hard of hearing.
Beyond defining a treatment effect for cure in modern times, we need to look at modern statistical approaches to trial design – more later.