In the figure above (Figure 2 from Walker et al.), the x-axis is days relative to collection of the culture that yielded the DTR pathogen. Data by row are for 3 successive time periods (2016-18, 2019-20, 2021-23) and by column are for Acinetobacter baumannii, Enterobacterales, and Pseudomonas aeruginosa.

In the figure, now focus on the 3rd column from the left in each sub-figure … this is the time (1 day after culture collection) when new information might become available. You can see that therapy begins to change on that day … but that’s a long time to wait for effective therapy. 

Well, at least some patients did get a potentially better therapy. So (cue drumroll), did better therapy make a difference for the population as a whole over the study period? Given that either a DTR antibiotic or (worse) an ineffective antibiotic was given in more than three quarters of the population, the effect of the newer antibiotics would be expected to be diluted out. Thus, you will be unsurprised to hear that the answer is no, a reduction in mortality was not seen. The overall estimated mortality probability for DTR infections was 20% for Enterobacterales, 20% for P. aeruginosa, and 25% for A. baumannii … and this really did not change over the study period. Aside: There was a hint of a reduction in patients with P. aeruginosa bloodstream infections, but the authors feel this is a weak observation (small N and hints in the data of unexplained confounding).

In discussing their data, the authors focus on two factors that could have obscured the benefits of newer therapies:

• Competing causes of mortality (age, other diseases) would obscure effects on the infection.
• Delay in appropriate therapy, especially in the subset (~25%) with sepsis.

And I would agree that both of these make sense … but I’d like to take this one step further and consider the amount of improvement that is even plausible given that there are multiple causes of mortality.

——————————
Stop #4 (One step further): How much of a mortality improvement is actually plausible?

Note: At this point in the newsletter, we have to assume that you are familiar with the study design ideas of superiority and non-inferiority. If not, please pause and (i) read the 19 Sep 2020 newsletter entitled “In Praise of Non-Inferiority” and (ii) watch its associated YouTube video. 

We start in this analysis by looking at the overall estimated mortality probability for DTR infections: ~20% for Enterobacterales, ~20% for P. aeruginosa, and ~25% for A. baumannii.

So, if we’re going to find an improvement (a reduction in mortality), we have to find it in the 20-25% who died … the other 75-80% survived and we can’t further improve their outcomes.

And, let’s also recognize that some mortality is doubtless unrelated to the infection and thus unpreventable: this would be the idea of attributable vs. non-attributable mortality. It’s hard to know an exact figure, we but know from Table 1 of Walker et al. that 12-13% of patients in the survey had a do-not-resuscitate order … so perhaps we can use 10% as a conservative estimate for non-attributable mortality. 

So now our zone for improvement (reduction) in mortality is limited to the attributable mortality subset … that’s a window from ~20% to (say) ~10%. And, it would take a 100% reduction in attributable mortality to reduce the mortality from 20% to 10%! That’s a tall order!

So, perhaps you are a bit less ambitious and think a better therapy could produce a reduction from 20% to 15% … effectively, a 50% relative reduction in the plausible zone for improvement. Well, that would be a stunning effect but it would take a study of about 1,250/arm to have 90% power to show that much change at P < 0.05. And if you thought you were going to show something a bit less (perhaps 40% relative reduction, the amount of mortality increase suggested by Kadri 2018), you’d need ~2,000/arm for a change from 20% to 16%. You could reduce the sample size to ~1,600/arm by accepting a P < 0.10 for your design, but that’s still a big study!

Ouch! But, you say, “I thought superiority studies were supposed to be small!” Well, that’s true for a really big effect — you only need 63/arm to show a change from 65% to 35% — but here we’re dealing with much smaller changes. Even if you said you’d recover all the mortality by moving the needle from 20% mortality to 10%, you’d need 286/arm. Not vast but not a small study either.

——————————
Stop #5 (Implications, Part 1): Can we ever show that new drugs are better?

So, let’s return to the opening statements. Deep, deep down, this conversation is about how we know whether (or not) new drugs really work for bad bugs. Yes, we can see the favorable in vitro microbiology. Yes, we can see the favorable results of animal models. But, what about in human infections? “I want to see data on new drugs for bad bugs in people,” you cry! “Why are you bringing me these confusing non-inferiority results? Why aren’t you studying bad bugs? And, superiority studies would be cleaner! They can be smaller! Why aren’t we instead seeing superiority data on treatment of bad infections?”

Hopefully, the answers to these questions are now clear:

• Unlike animal model studies, we can’t directly show the benefit of a new therapy over no treatment (placebo) in human studies because we can’t deliberately do placebo-controlled studies of new therapies … we always seek active therapy in all study arms.
  • Infections can be fatal if not treated promptly! Placebo just is not an option!
  • 25 Apr 2026 addendum: This was extensively debated 2007-08 (see Spellberg et al. Clin Investig (Lond). 2011;1:19-32; https://doi.org/10.4155/cli.10.1 for a good summary). Ultimately, FDA explicitly stated that placebo was not ethical in its guidance document for community-acquired bacterial pneumonia.
• Situations where superiority studies have the potential to succeed are rare because we work hard to control the rate of DTR, MDR, XDR, and PDR pathogens.
  • We do sometimes run studies in which highly resistant pathogens are sought but these studies tend to be smaller and hard to enroll because we work hard to make resistant pathogens rare and thus …
  • … Walker et al. needed data from 471 hospitals spanning 8 years to find the 5,000 DTR infections in their report! That size and duration is obviously not feasible for a drug development program.
• Finally, superiority studies are most feasible when there is a VERY large effect to be measured. Small effects require huge studies … and small changes are often not very compelling even when shown.
• Now, you would correctly argue that perhaps we’re using the wrong endpoint … mortality is a blunt measure. Would measures that incorporate a range of non-mortality clinical measures be helpful?
  • This is the idea, for example, of DOOR (Desirability Of Outcome Rating) that has been championed over the years by Scott Evans and colleagues. Yes,  this could be a helpful path forward, especially if those endpoints can be shown to be clinical relevant and if superiority can be shown on a non-mortality component of an endpoint.
  • For an introduction to this literature, see Evans SR et al. Desirability of Outcome Ranking (DOOR) and Response Adjusted for Duration of Antibiotic Risk (RADAR), Clinical Infectious Diseases, 61:800–806, 2015, https://doi.org/10.1093/cid/civ495.
  • But, mortality is such an obvious and powerful endpoint that it’s hard to get away from it, especially in settings like infection where the course of the illness is often rapid and obvious. Even though it can be argued that other endpoints are clinically relevant and potentially more sensitive, the fundamental data that justify many of our standard designs depend on ACM as the primary endpoint because this is the measure that can be used to justify a non-inferiority margin.

And so, this is why we use non-inferiority studies:

• We know from non-clinical data that the new antibiotic could work.
• We show in a UDR setting that the new agent is non-inferior to an existing agent when the existing agent still works.
• When resistance arises to the existing agent, the new agent’s activity is unaffected … it still works.
• We can supplement the non-inferiority study with salvage studies in patients with MDR, XDR, PDR, or DTR infections but these studies are (hopefully) almost impossible to enroll if we are doing a good job with infection control.
• The key to approval lies in the non-inferiority study done in the UDR setting
• And, this is what allows us to develop new agents before the need for them becomes acute!
• For more on this critical theme, see this YouTube video explainer or the 19 Sep 2020 newsletter entitled “In praise of non-inferiority.”

——————————
Stop #5 (Implications, Part 2): What does this mean for calls to improve outcomes through alternative approaches such as host-directed therapies?

Well, you should by now be realizing that the news is not great. Unless your approach has a simply stunning effect, the ability to show a benefit in clinical studies is very limited. We say this with sadness as we think many of those ideas make good scientific sense. Unfortunately, it is very, very hard to show such effects consistently unless the effect is dramatic. For more on this (really, we have got to get to the end here!!), please see the 6 Aug 2019 newsletter entitled “Non-traditional antibiotics: A pipeline review and an analysis of key development challenges”. It’s a relatively old newsletter, but it’s one that covered in depth the problem of developing non-traditional therapies (such as host-directed therapies) and one where new material has been added as it has arisen.

——————————
(Implications, Part 3) Rapid diagnostics and access to DTR-active agents are the keys to taking action!

So, there two strong, clear action items that come from Walker et al.: The first is to make a diagnosis, and make it quickly! This one is quite obvious …if antibiotics are the #FireExtinguishersOfMedicine, then diagnostics are surely the #SmokeDetectorsOfMedicine! But as noted in the paper, “Diagnosis of a DTR pathogen by conventional antimicrobial susceptibility testing methods is estimated to take a mean of 92.2 hours (SD 27.8) and can result in a significant delay in the time to initiation of an appropriate antibiotic.” That’s 4 days — much too long! Quoting further from Walker et al.:

• “…rapid diagnostics could enable earlier yet stewardly initiation of newer over traditional antibiotics in this patient population with a high burden of sepsis.
• “However, current rapid diagnostic platforms focus on enzyme-mediated resistance mechanisms, such as Klebsiella pneumoniae carbapenemases or New Delhi metallo-β-lactamases, despite the fact that such mechanisms are present in only a small fraction of resistant isolates and DTR isolates often have nonenzyme-mediated resistance mechanisms.
• “Empirical antibiotic therapy could also be guided by predictive algorithms that estimate the likelihood of resistant pathogens, but this approach warrants further investigation.”

Is “stewardly” really a word? Well, it works for us and the message is clear. Also clear is the comment about the limitations on current diagnostics — we need tools for detection of more than just beta-lactamases! In further support of the #SmokeDetectorsOfMedicine, see these recent newsletters on the power of diagnostics:

• 18 June 2025 newsletter: “Extending STEDI to diagnostics: STRIDES”
• 17 Feb 2026 newsletter: “Workshop on diagnostics: Can we drive use by making value visible?”

And the second is, of course, ensure your healthcare system has access to all the DTR-active agents, especially the newer ones! Here again we remind you of the 1 Dec 2024 newsletter (“The 6 meanings of ‘Lack of Access’ (UNSLAP)” and the related 7 Apr 2025 newsletter (“UNSLAP: You reach for the antibiotic … and it’s not there!”). The titles of those newsletters are self-explanatory: if the antibiotic is not in your local pharmacy, it might as well not exist!

——————————
Whew! That was a lot! Apologies for going on such a long tour, but it seemed helpful. Thanks for listening! All best wishes, John & Aaron

John H. Rex, MD | Chief Medical Officer, F2G Ltd. | Operating Partner, Advent Life Sciences. Follow me on Twitter: @JohnRex_NewAbx. See past newsletters and subscribe for the future: https://amr.solutions/blog/. All opinions are my own.

Aaron Dane, CStat | DaneStat Consulting Limited | aarondane@danestat.com | www.linkedin.com/in/aaronldane | All opinions are my own.

Summary of prior newsletters on DTR and key related topics:

• 13 Jan 2019: “Categories of resistance: MDR, XDR, PDR, UDR, and (new!) DTR”
• 20 Feb 2020: “Language matters: CRE vs. CPE; SDD vs. I; and MDR, XDR, PDR, UDR vs. DTR”
• 7 Jun 2020: “Assessing antibiotic value: DTR, fire extinguishers, and a view from Australia”
• 19 Sep 2020: “In praise of non-inferiority”
  • Non-inferiority YouTube video explainer
• 17 Mar 2021: “Where’s the innovation? / Indian Priority Pathogen List / DTR video explainer”
  • UDR vs. MDR/XDR YouTube video explainer — see below!
• 25 Apr 2025: “Evolving the idea of DTR (Difficult-to-Treat Resistance) to include antibiotic access”