Study Design in ABSSSI – A Statistician’s Delight

The design of the O’Riordan ABSSI study [1] deserves comment.  This was a double-blind study of 2 lower dose arms (Part 1) with an add-on open-label (Part 2) high-dose arm. The pimary endpoint was a non-standard composite efficacy and safety endpoint that was never used before.  There were 3 protocol amendments.  There were a series of 5 (five!) interim analyses using adaptive randomization based on efficacy and withdrawal which resulted in step-wise preference for the highest dose considered beneficial.  A clinical utility index (CUI) >85% posterior probability was set as a decision criterion, reflecting a minimal ‘modeled cure rate’ that needed to be attained.  Using this ‘sponsor-defined analysis’, gepatidocin failed in ABSSSI.

All these design elements resulted in vastly different dosing group sizes and an imbalance in the 3 major skin infection categories (wound infection, major abscess, cellulitis) among the dosing cohorts.  The highest percentage of cellulitis cases (and presumably easiest to treat) were in the 1g q8h / 2g TID group.  Not surprisingly, pathogen recovery was lowest in this highest-dose group.

There was an option to switch from IV to PO after 2-3 days of IV therapy, ie. after reaching the 1° efficacy endpoint (ECR) as in most ABSSSI studies that have a bioequivalent PO formulation.

Drug resistance was seen in only 2 MRSA isolates with gepotidacin MIC values very much above the susceptibility range due to Par C and E mutations.  We consider this an ominous sign recalling the history of quinolones which initially seemed to have reliable MRSA coverage only to be found out about otherwise soon later.

In our opinion, this Phase 2 dose-finding study was designed by statisticians and not by clinicians.  The GSK team used an academic approach, a novel adaptive design, for dose-finding.  It also introduced rather arbitrary decision rules based on a combination of efficacy and safety criteria.  ABSSSI studies are not easy to execute, and this trial tried to pack too much into a small size of 122 patients.

We are not in agreement that gepotidacin has insufficient efficacy in ABSSSI based on this trial.  Only 1 patient was discontinued for lack of efficacy, and 4 for AEs.  None of the other 10 withdrawals (O’Riordan Table 2) should have been part of the primary efficacy analysis.  Therefore it seems that this new indication, in other hands, may well have survived.

Gepotidacin did better using EOT instead of ECR timepoints. EMA and most other regulatory agencies never subscribed to FDA’s ECR approach for efficacy assessment, and we don’t like it either.  Many clinicians argued this point at FDA meetings; for them (and for patients!) it is the EOT response that is important.

Does it matter that gepotidacin was dropped from further development in ABSSSI by GSK?  Did it truly get a fair trial?   Would a more traditional approach have resulted in a different more favorable outcome?   Hard to say at this stage, but we have a strong bias that a good drug was sacrificed on the altar of statistics.

REFERENCES

[1] O’Riordan W, 2017. Efficacy, safety, and tolerability of gepotidacin (GSK2140944) in the treatment of patients with suspected or confirmed Gram-positive acute bacterial skin and skin structure infections. Antimicrob Agents Chemother 61:e02095-16. https://doi.org/10.1128/AAC.02095-16

ABBREVIATIONS

CUI    Clinical Utility Index
ECR   Early Clinical Response
TOC   Test-of-Cure