Long and wonkish note alert — please refresh your coffee before you start reading!
I was able to attend and speak at the 12 July 2019 Public Meeting organized by FDA on the LPAD (Limited Population Antibacterial and Antifungal Drug) pathway. The current draft guidance is here and the meeting agenda is found here. The meeting was captured for streaming replay and the video, all the slides, and a transcript will be available here in a few weeks. Post-newsletter addendum: The LPAD guidance is final as of August 2020. The essence is unchanged but there are some useful clarifications in the final guidance — go here for details.
I thought it might be useful to share reflections from the day. If you’re in a hurry, you can jump to the end and read the summary. If you have a little more time, we’re off to a story in five parts…
First, FDA opened the meeting with a presentation on the details of LPAD along with a brief discussion of the one drug approved to date under LPAD (Arikayce, go here for its 2018 US label). Post-newsletter addendum: Pretomanid was approved in August 2019 under LPAD, bringing the tally to two for LPAD-based drugs (FDA press release, initial US label).
In brief, LPAD uses a 3-part rule of (i) a Serious or Life-Threatening Infection, (ii) a Limited and Defined/Definable Patient Population, and (iii) Unmet Need to point to create the LPAD pathway. The current draft LPAD guidance (link) then explains the utility of LPAD (text begins at line 164, emphasis is mine):
- “The LPAD pathway requires FDA to take into account in its determination of safety and effectiveness the severity, rarity, or prevalence of the infection a drug is intended to treat and the lack of alternative treatment in the limited population a drug is intended for (see section 506(h)(2)).
- “FDA may approve such drug although not enough data exists to conclude that there is a favorable benefit-risk profile in a broader population.
- “As discussed in the Unmet Medical Need guidance, drugs with risks that would be unacceptable for a broad population may be acceptable for patient populations with serious diseases that do not have other treatment options.”
Second, my slides from the day are here and I focused my comments on the idea of Antibiotic R&D v3.0 as a way to see LPAD as a springboard to support development for less common pathogens and infections. Briefly (and please do see the deck):
- R&D 1.0 was the period from the dawn of antibiotics to 2005-6 (development was easy, drug effects were obvious),
- R&D 2.0 is the period from 2007 to date (single pivotal trials are accepted as sufficient for approval, we update our trial designs, and approval requirements are harmonized globally), and
- R&D 3.0 is the period made possible by LPAD in which we extend the zone of approvable compounds to include products for rare pathogens and rare infections.
Third, we had two presentations on specific areas of unmet need for antifungals. Both presentations pointed out the relatively rarity of these infections and the need to work with small datasets to develop new agents.
Fourth (and this is where this note gets really wonkish), all the presentations pointed to some fundamental questions about the utility and purpose of LPAD. To capture the spirit of these interwoven questions, I am going to try a Q&A format. Not all of this was discussed in detail at the public meeting — I am weaving together materials from multiple sources to produce this commentary:
- What does LPAD create that didn’t exist before? From one perspective, nothing here is new as FDA has always used benefit-risk as part of its integrated thinking. On the other hand, LPAD does provide FDA with a way to explicitly reinforce the idea of limited use and convey to this the prescribing community. You can see this from the Arikayce label (here): the word “limited” appears 6 times in the first few paragraphs of the left-hand column on page 1, including once as LIMITED (all caps, bold) and again as LIMITED (just all caps). Combined with the requirement for stewardship programs and surveillance by CDC, it is very likely that an LPAD-labeled drug would be used appropriately.
That said, I also think it is noteworthy that the idea of LPAD was to my knowledge first introduced in Jan 2013, the LPAD legislation was signed into law 4 years later (13 Dec 2016), and in the interim FDA effectively implemented most of the elements of LPAD by moving to allow approvals with single pivotal non-inferiority trials and by working to clarify suitable endpoints via various FNIH-led projects (for a summary of this work, see Rex JH et al. Clin Infect Dis 2017, link). Viewed through this lens, it is perhaps unsurprising that LPAD’s impact on approval pathways has been relatively limited — most of what was needed was already up and running, albeit now with explicit statutory blessing from Congress.
- Does LPAD make it easier to develop a drug? Are the programs necessarily faster? Are the required data different from the data required for other drugs? Perhaps surprisingly, the answers here are No, No, and No. The standard for approval (21 CFR 314.126) remains “substantial evidence of efficacy based on adequate and well-controlled studies.” Studies of rare pathogens and rare infections remain hard simply because the setting is rare.
- Can the developer choose to limit development to a defined small population even though other (larger) populations are readily studied? Inversely, can a developer use LPAD to salvage a failed program seeking broad usage by limiting the scope of approval to a defined small population from the failed program? Again, the general answer to both is No. If there is an obvious way to develop more broadly, then there is strong encouragement to follow that broader path. The core idea here is really to my eye a variant of the reason why antibiotics don’t (usually) qualify for Orphan Drug status. As an example, consider a drug with activity for MBL-producing Enterobacteriaceae. Unless it works only on MBL-producers (that is, it fails on non-MBL-producers), it can be used for both MBL and non-MBL infections. MBL infections may be currently rare (and let’s keep it that way!) but non-MBL are of course very common.
Post-newsletter addendum: The language on this in the FDA’s final 2020 guidance is worth reciting here: “FDA interprets limited population of patients in this provision to mean a group of patients that is limited in a way that is clinically relevant to health care providers. The labeling should define the limited population that the drug is intended to treat so that a health care provider would be able to identify the patients in the clinical setting for whom FDA determined the benefits of the drug outweigh its risks. A limited population may be a defined subset of a broader population of patients for whom the drug could potentially be effective or, in some cases, may be the only population of patients for whom the drug may be effective because of its narrow spectrum of activity.”
In summary, it is critical that the Limited Population have a clear definition that is “clinically relevant to health care providers.” If use cannot readily be contained within the Limited Population, then the idea of approval with higher uncertainty about the benefit-risk in the Limited Population is weakened.
- Will (Should) the data in support of an LPAD approval be drawn entirely from the Limited Population for which the drug is approved? Perhaps a surprise here as the answer is No. This subtle but important and VERY confusing point takes us into a discussion of UDR and MDR (Usual Drug Resistance and Multi-Drug Resistance, go here for more on these abbreviations). Although we want new drugs with activity vs. MDR pathogens, we will almost always develop them using a non-inferiority comparison vs. a good comparator in a setting where that comparator has retained activity. For example, consider a study of NEW, a drug for infections due to Enterobacteriaceae. At present, the UDR situation is that most Enterobacteriaceae can be treated with a carbapenem and hence pivotal studies will be of the form NEW vs. meropenem (or imipenem). In such a study there might be a few patients who are incidentally enrolled with a penem-resistant pathogen, but most patients will be infected with a penem-susceptible pathogen. From a public health perspective, it is desirable that it be hard to find penem-resistant infections and hence data on treatment of penem-resistant infections will be limited to some form of small (often non-randomized and open-label) study of NEW.
The core logical leap that comes next is that the correct interpretation of the development program is that NEW works when the pathogen is susceptible to NEW. Stated differently, the MIC of NEW for the infection pathogen contains all the information about NEW’s potential ability to treat infection. A positive result in a non-inferiority comparison with a known effective agent (even if small due to the rarity of the infection) shows that the in vitro MIC predicts clinical response; susceptibility or resistance to other compounds is not relevant. This point is one of the most confusing aspects of the entire problem and I am still looking for better ways to explain it! Good discussions can be found in McDonell AM et al., Clin Infect Dis 2016, Boucher HW et al., J Infect Dis 2017, and Rex JH et al. Nature Comms 2019 (post-newsletter addendum: see this 6 Aug 2019 newsletter for a discussion of the Nat Comms paper, a related paper, and how all this connects to the challenge of non-traditional products). I also gave a talk in 2016 that lays all this out in a slightly different way that is sometimes helpful (link). And for the record, this talk is the very first time the acronym UDR was used in public; its first publication was in the McDonnell 2016 paper cited just above.
Post-newsletter addendum: On this theme, see also this 19 Sep 2020 newsletter entitled “In Praise of Non-Inferiority” along with the related YouTube video on this topic. It is time to learn to love non-inferiority!
- I really don’t like the idea of studying patients in the UDR setting … I want to study MDR pathogens in difficult settings! How about an open-label, single-arm, non-comparative series of MDR infections treated with the new agent? This is such a subtle one! In brief, it is not a way out except in rare circumstances (e.g. and post-newsletter addendum, the 2nd LPAD approval of pretomanid uses external control data as context for pretomanid’s activity in XDR TB; go to the top of page 17 of the initial label and then scroll up a tiny bit to see the discussion).
While I am sympathetic to reasons behind the dislike of data from the UDR setting, cold reason shows us that those data are perfectly valid. As noted just above, the MIC really tells almost you everything you need to know about the new agent’s therapeutic potential (again, see also my 2016 talk cited just above). The only gap is that of confirming that PK is as predicted in the relevant patient population(s). A study of critically ill patients (e.g., nosocomial pneumonia) in a UDR setting is the perfect place to gather the needed data in a controlled fashion before resistance becomes widespread. In so doing, we learn about PK in that population, we confirm that the activity is similar to that of a good drug, and we thus know that the new agent is reliable. Safety data may remain limited but presuming there is no cross-resistance, a new drug that matches the performance of meropenem vs. meropenem-susceptible pathogens will continue to perform perfectly well vs. meropenem-resistant pathogens.
For more on the challenges of single-arm data, the best discussion of this problem I’ve seen is Fleming and Ellenberg’s 2016 paper in Clin Trials entitled “Evaluating interventions for Ebola: The need for randomized trials.” Just the title alone should give you pause! In brief, the impact of the new agent must be “sufficiently large to make interpretation of the results of a nonrandomized trial unambiguous.” It’s a great paper — don’t miss the comment about why, in context, randomized trials were not needed to show the value of penicillin! Post-newsletter addendum: Burger’s 2021 paper in Pharm Stat entitled “The use of external controls: To what extent can it currently be recommended?” is also excellent as is (no surprise!) ICH E10 from 2000.
- OK, I’ve read that paper, thought about it a bit, and (perhaps grudgingly!) have bought into the value of generating (if possible) at least some randomized UDR-setting data on a new agent. How much data do we need? Now we come back to the point of LPAD. FDA’s requirement for approval is “substantial evidence of efficacy based on adequate and well-controlled studies.” Importantly, alpha = 0.05, 10% margins, specific endpoints, and concurrent randomized controls are not legal requirements. EMA is very explicit about this in their current draft update to their guidance for antibacterial agents (link). Go to line 460 where we are told in a discussion of non-inferiority designs that “The sample size may be driven primarily by feasibility and an estimate of accrual rates over a reasonable time frame (e.g. not exceeding approximately 2 years). There should be a justification for the trade-off proposed between statistical power, nominal significance levels and the non-inferiority margin.” Although we don’t have a corresponding statement from FDA about how they will approach flexibility, we have certainly seen adjustments to margins and there are ongoing discussions about using novel statistical methods (e.g., Bayesian statistics). Hence, I look forward to finding ways to use the principle of LPAD to evolve our thinking and expand the range of approval products. Post-newsletter addendum: Final updates to both EMA and FDA antibacterial guidance documents were released May 2022 and retain the ideas discussed just above. For more, see this 27 May 2022 newsletter.
- Who decides how much is enough? Critically, the question of “Who decides?” starts with FDA and EMA but definitely does not end there. Rather, the ultimate arbiter is the clinical community along with the payors (i.e., all of us). The case for a small, focused LPAD-based approval must be solid and compelling — a program uses this pathway because it must, not because it so choses. But, if that’s all a program can do, a good compound should not be held back for lack of a larger dataset. Rather, the very point of LPAD is that rare pathogens are rare, that benefit-risk is different for different populations, and that we can use a clear definition for that limited population to explain this to the entire community.
Fifth (deep inhale and exhale), so what about the elephant in the room? Why was Insmed’s Arikyace (inhaled amikacin) approved under LPAD for treatment of MAC lung disease but Achaogen’s Zemdri (plazomicin) only for cUTI and not also for bacteremia due to CRE? This was not on the agenda for the day, but to my eye boils down to a handful of elements.
- The endpoint used for Arikayce was admittedly a surrogate (culture conversion to negative) but the P value (< 0.0001) was very strong whereas the strong endpoint of 28-day all-cause mortality in the Zemdri study had only relatively weak statistical power due the small sample size (a favorable trend, but P > 0.05 in the various analyses from the FDA backgrounder for the 2 May 2018 Ad Comm). Further, the FDA backgrounder makes it clear that the Zemdri program didn’t really have a plan to test a specific statistical hypothesis — so even 28-day all-cause mortality was just descriptive.
- All of this was further confounded by the heterogeneity in the patient groups — with only about 20 patients in each arm, it is not surprising that there were imbalances. Unfortunately, these imbalances on the whole tended to predict a better outcome for Zemdri (e.g., somewhat more of the plazomicin patients were blood culture negative at enrollment).
- Finally, the limited population for Arikayce is very sharply defined whereas the limited population for Zemdri is little bit fuzzier: people with CRE bacteremia obviously stand out, but what about just bacteremia that might be CRE? Here you can also see elements of the discussion of Orphan Drugs just above in that Zemdri will also work on non-CRE.
So, where does this leave us? My conclusions would be:
- LPAD may yet advance the field for us but it’s not a panacea — its impact is limited to a narrow set of problems.
- Drugs that can be developed more broadly are not usually appropriate for LPAD. The tools we developed during the Antibiotic R&D 2.0 era are separately in place to permit approval with single trials.
- Rather, LPAD’s role as we move into Antibiotic R&D 3.0 is really for very narrowly defined settings that share some of the conceptual features of the Arikyace approval. As specific examples, a sharply defined limited population and a strong clinical signal are foundational. Needing to balance benefit vs. a safety signal is another feature that might support LPAD. In the realm of the more acute bacterial infections, S. aureus endocarditis would be one possibility. Uncommon specific pathogens (e.g., Acinetobacter) might be another. It would be interesting to compile a list … let me know your thoughts.
- And, we as a community need to talk about the power of what we learn from standard trials! Going back to Zemdri, how about this tidbit from the label’s discussion of the cUTI dataset: “Composite Cure at the TOC visit in individuals with concomitant bacteremia at baseline was achieved in 72.0% (18/25) of patients in the ZEMDRI group and 56.5% (13/23) of patients in the meropenem group.” So, does Zemdri work for bacteremia? Yes, indeed! And will it work for CRE as long as the bug is susceptible to Zemdri? Of course! We need to better explain this. Post-newsletter addendum: On this, I will again encourage you to review this 19 Sep 2020 newsletter entitled “In Praise of Non-Inferiority.” There is even a YouTube video that discusses this issue.
It’s a separate discussion, but I’d be remiss if I didn’t also say that LPAD won’t fix the economics of antibiotics — we really do need good Pull incentives in place. If you are new to this area, please go here for more details on this problem.
Deep stuff indeed! In the spirit of “Tears today vs. tears tomorrow,” it is critical that the entire community engage in discussing these points. I am particularly struck by the difficulty of explaining some of these ideas and will continue to search for better ways to tell the story.
John H. Rex, MD | Chief Medical Officer, F2G Ltd. | Expert-in-Residence, Wellcome Trust. Follow me on Twitter: @JohnRex_NewAbx. See past newsletters and subscribe for the future: http://amr.solutions/blog/
Upcoming meetings of interest to the AMR community:
- 31 Jul 2019 (Philadelphia, PA): Wistar Institute-sponsored “Symposium on Gram-Negative Bacteria Resistance – New Approaches in Pushing Back the Tide.” Go here for details.
- 20 Aug 2019 (webinar, 17:00-18:30 CEST): REVIVE webinar entitled “Models for antimicrobial R&D: Computational modelling for population PK and PKPD.” Go here to register.
- [Mark your calendar now!] 3-6 Sep 2019 (Boston). Annual ASM-ESCMID Conference on Antibiotic Development. The Bootcamp series will continue on 3 Sep with main meeting on 4-6 Sep. Mark your calendar now and check back here for details.
- 6 Sep 2019 (Cambridge): CeBIL Annual Symposium 2019: Legal Innovation to Support the Development of Anti-Microbial Drugs. For more details and to register, go here.
- 6-8 Sep 2019 (Bilbao, Spain): 5th ESCMID conference on Vaccines. Check back here for details.
- 2-6 Oct 2019 (Washington, DC): IDSA’s annual IDWeek meeting.
- 19-27 Oct 2019 (Annecy, France): International Course on Antibiotics and Resistance (ICARe) – A soup-to-nuts intensive residential training program on all things AMR, especially R&D for new antibiotics. See this link for details.
- 16-18 Dec 2019 (Bangkok, Thailand): 3rd International Symposium on Alternatives to Antibiotics in Animal Production. Go here for details: https://www.ars.usda.gov/alternativestoantibiotics/
- 1-6 Mar 2020 (Il Ciocco, Tuscany, Italy): GRC on Antibacterial Discovery and Development: “Now is the time to re-boot antibiotic R&D before it’s too little, too late.” Go here for details.
- 12-13 Mar 2020 (Berlin?): BEAM-, Novo REPAIR-, CARB-X-, DZIF-, ND4BB-, ENABLE-supported (among a long list!) Conference on Novel Antimicrobials and AMR Diagnostics. Final location is TBD, details will appear here, and you should mark your calendar now.
- 16-17 Mar 2020 (London): BSAC Spring Conference entitled: “Bridging the gap between science, policy and effective antimicrobial use.” Go here for details.
- 18-21 Apr 2020 (Paris): Annual ECCMID meeting (#30)
- 25-30 May 2020 (Rotterdam), Annual ESPID meeting (European Society for Pediatric ID, #38)
- 10-13 Apr 2021 (Vienna): Annual ECCMID meeting (#31)