Nosocomial Pneumonia - Not there yet.

In reading the briefing material for the November 4^th FDA advisory committee meeting on nosocomial pneumonia, I have mixed feelings.  On the positive side, they do offer a feasible way forward for trials at least in terms of trial numbers. They are considering allowing enrollment of patients who have received prior antibiotics.  Both of these steps represent major progress from the original draft guideline and, again, the FDA should be commended for their efforts to make these trials feasible once again.  But neither of these steps will be sufficient to solve our current problem.  We are nowhere near where we were before the convulsion at FDA led us to our current predicament. We need, as we had prior to the new draft guideline, a clinically relevant and feasible design. 

First, somehow we have gotten down to a 10% NI margin when the treatment effect for mortality (before the usual and overly conservative FDA discounting) is a huge 42%.  And that 42% is already a conservative estimate because it comes not from controls who received no therapy but from those who received inappropriate therapy – not the same thing at all.  So why the FDA insists on discounting the treatment effect for nosocomial pneumonia, I don’t know. With a treatment effect of 42%, it is easy to justify trials with an NI margin of 20%.  At an NI margin of 10%, a single trial is feasible.  If two trials are required, the numbers remain impossible. At a 20% NI margin, the trial numbers are drastically reduced and, in my view, treatment effect is preserved.

The other issue the FDA continues to grapple with is the endpoint.  They are unable to justify a clinical endpoint by looking at studies of appropriate vs. inappropriate therapy. All these studies look at mortality.  But all these studies are of an observational nature – none come from the unique and artificial context of the clinical trial. Therefore, their relevance is highly questionable. But the FDA is wrong.  Pharmacometric studies justifying a treatment effect for clinical outcome DO exist and are compelling – see below.

All other problems stem from this.  The mortality endpoint requires a high enough mortality rate, 20% according to the FDA, that one could expect to see differences between treatment arms if they existed.  As the mortality rate goes down from 20%, the trial size increases dramatically.  But studying patients this sick increases the risk for the antibiotic under study in such trials.  Antibiotics that work very well might be subject to variations in enrollment such that slightly sicker patients are included in the test arm.  In that case a (falsely) higher mortality might be seen for the new antibiotic.  There is also a safety risk in that there are many opportunities for false safety signals for an antibiotic being studied in such populations.

In looking at clinical trials over the years, the mortality rates are almost always very close between comparators. One explanation is that the antibiotics studied are all comparable.  Another is that mortality is an insensitive endpoint that is determined by the severity of the patient’s underlying diseases and antibiotics, for those with severe underlying disease, do not always prolong survival to 28 days. Both of these are likely to be true. 

  

In terms of a clinical endpoint, the pharmacometric approach offers a very nice way to determine treatment effect in the clinic and to establish, therefore, a justified NI margin.  Why the FDA will not consider this approach at the same time that Europe is doing so baffles me.  As shown in the figure below (from a recent Paul Ambrose blog), one can extrapolate the exposure MIC relationship seen in clinical failures to zeros and estimate the effect of no treatment.  Subtracting the effect among those with adequate exposure-MIC relationships provides the desired treatment effect. Recent articles on tigecycline, in the situation of HAP/VAP in particular, are very informative in this regard. At the European regulatory authority workshop on anti-infectives, Ambrose showed data from the tigecycline experience demonstrating a 40-60% treatment effect based on these methods.

Antibiotics target bacteria.  They cure infection but not heart or lung disease.  The endpoints used must reflect this basic scientific fact.  Because the FDA can’t figure out how to do this at this moment in time, the earth must stop spinning on its axis and we must use an unrelated and insensitive endpoint that they can justify based on data obtained outside the trial situation.  Of course why they can’t figure this out when lots of others can also baffles me.

My suggestion is that the FDA does one or more of the following –

·      Declare a moratorium on implementation of their draft guidance requiring a mortality endpoint. 

·      Consider the use of pharmacometrics to support a calculation of treatment effect and therefore NI margins in the design of trials with clinical endpoints as Europe is doing.

·      We should go back to the previous guidance using clinical endpoints while we gather data to justify the use of such endpoints in these trials.