The antimicrobial resistance paradox and the real crisis of treatment failure

In stark contrast, antibiotic treatment failure (whatever the cause) is common and costly. Treatment failure can be addressed, to a degree, by using stronger front-line antibiotics. In fact, the field responded this way over the past several decades. Third-generation cephalosporin antibiotics, once considered last-resort, now serve as the first-choice antibiotics for severe infections like pneumonia.³⁰ Doing so perpetuates the development of resistance in a never-ending arms race. It is time to end the arms race by taking a new approach.

The drama of the AMR problem

Dame Sally Davies, the former Chief Medical Officer for England, brought the rise of AMR to popular attention in her 2013 short book, The Drugs Don’t Work, in which she placed the AMR problem on the same scale as climate change.³¹ Indeed, AMR has become a widespread clinical problem and can render even basic medical procedures life-threatening.³² If AMR continues to rise, we face the alarming prospect of the world reverting to a pre-antibiotic era, where routine surgeries bring risk of death and cancer patients succumb not to their malignancy, but to secondary infection.

An oft-cited report published in 2016, commissioned by the UK Prime Minister David Cameron and chaired by economist Lord Jim O’Neill, alerted the world that barring intervention, AMR would take 10M lives globally each year and cause a cumulative $100 trillion in economic damage by 2050. If reality is anything close to this, the world will have paid dearly. The report asserted that AMR is one of the most pressing health issues of the 21st Century.²

More immediately, an extremely rigorous investigation by the AMR Collaborators, led by Christopher Murray at the University of Washington, reported in The Lancet that AMR already kills 1.27M people per year, and contributes to the death of a total of 4.95M.^1,33

A paradoxically small patient population

Once we look at specific resistance problems, an entirely different picture emerges. The Centers for Disease Control and Prevention (CDC) reports numbers of infections per year for its priority, antibiotic-resistant pathogens.³⁴ Carbapenem-resistant Acinetobacter infections cause 8,500 hospitalizations and 700 deaths annually.³⁴ Carbapenem-resistant Enterobacteriaceae infections cause 13,100 hospitalizations and 1,100 deaths annually.³⁴ Each of these pathogens causes a small subset of more general infection categories: for example, Acinetobacter is often implicated in ventilator-associated pneumonia or catheter-associated bloodstream infections. Compared to the tens of millions of antibiotic prescriptions written every year in the US, the number of patients with these very scary, drug-resistant infections is vanishingly small. This is partly because our current antibiotic armamentarium is very large, with somewhere between 80–100 distinct molecular antimicrobial agents commercially available in the US. The odds of finding a drug that can kill any given causative pathogen are high, even if the pathogen has resistance genes.

What we see in this mismatch between the very real AMR crisis headline numbers, and very real but very small priority pathogen patient populations, is that the world has narrowly defined the AMR crisis in terms of specific pathogen-drug resistance combinations: Carbapenem-resistant Enterobacteriaceae, Methicillin-resistant S. aureus (MRSA), etc. When framed this way, the target populations are relatively small and so there appears not to be an urgent need for new medicines. Indeed, pharmaceutical companies have essentially stopped investment in novel antibacterial drugs because the market demand for new agents is close to zero. This trend is well-documented and reported elsewhere.³⁵

The scale of antibiotic treatment failure is large

Drug-pathogen resistance combinations focus on the bacteria, which is only part of the clinical story. Our view is that we should be defining the unmet need in terms of the many millions of at-risk patients and their clinical outcomes. The real unmet medical need is antibiotic treatment failure—whatever the cause—not antibiotic resistance per se. We initially presented this view recently in the Journal of Antibiotics, and elaborate here with more detail.⁴

When we survey the medical literature, we find that rates of treatment failure are high, often shockingly so. The costs are high, too, both for patients and the healthcare system. Mortality rates are far higher in these treatment failure patients, as are the total costs of care for them. They require extended time in the hospital, the intensive care unit, or additional procedures to manage complications.

Causes of treatment failure

As stated before, we posit the primary causes of treatment failure are:

• i
  AMR means many empiric drugs do not work well.
• ii
  Antibiotics have great difficulty in dealing with pathogens where they hide.
• iii
  Antibiotics are usually insufficiently selective, damaging the healthy microbiome, which is necessary for a healthy immune response.

To elaborate on these ideas and reference the underlying research:

The largest and most straightforward cause of antibiotic treatment failure is antibiotic resistance. Most intuitively, if the pathogen is resistant to the drug being administered, treatment will have no utility. By definition, a patient with an infection resistant to empiric therapy will be a treatment failure case, requiring an antibiotic switch. The global study published in The Lancet in 2022 brought this clearly to light: AMR kills 1.27M people per year, and contributes to the death of a total of 4.95M.¹ Other studies have correlated rates of antibiotic resistance to rates of sepsis.^73–75 This observation is consistent with the idea that treatment failure patients are more likely to progress to sepsis, and antibiotic resistance is a cause of treatment failure.

Second, pathogens’ defense mechanism of “hiding” drives treatment failures. The hiding, itself, takes at least three major forms:

• i
  Biofilm formation: Chronic infections, non-responsive to antibiotics, appear increasingly to be associated with biofilms.^76–78 Biofilms are a self-protective matrix of extra-cellular DNA held together by lynchpin proteins.^79,80 Bacteria live in the matrix, shielded from both host immune response and antibiotics. Biofilms have been associated with chronic infected wounds;^81,82 the respiratory tract of cystic fibrosis patients;^83–85 the sinuses of chronic rhinosinusitis patients;⁸⁶ deep tissue infections of bone⁸⁷ and heart valve;⁸⁸ and recurrent urinary tract infections.⁸⁹ In short, most chronic infections that do not respond to antibiotics appear to be associated with biofilms,⁹⁰ and bacteria within biofilms are up to 1000x more tolerant of antibiotics than when free-living.⁹¹
• ii
  Intracellular niches: Bacterial survival strategies can include invading host cells, which then become a shelter from most antibiotics (most of which penetrate poorly), and from adaptive immune responses from T cells and antibodies. For many pathogens, concealment within an intracellular niche is the dominant survival strategy. These diseases are marked by the need for extremely long courses of multiple antibiotics, precisely because antibiotics penetrate host cells poorly. As previously described, this is the case with antibiotic-refractory (“chronic”) UTI, and all of the biothreat pathogens. It is also famously the case for tuberculosis^92,93 and nontuberculosis mycobacteria, both of which are caused by pathogens that invade respiratory immune cells, then require multi-drug treatments for six months or longer, and still result in high treatment failure rates.⁹⁴
• Beyond these well-known examples, other intracellular infectious syndromes include: leprosy (Hansen’s disease);^95,96 typhoid fever;^97,98 invasive Salmonella;⁹⁹ Legionnaires’ disease;^100,101 chlamydia;^102–104 listeriosis;^105,106 Whipple’s disease;¹⁰⁷ and more.
• Perhaps more surprisingly, many other infectious syndromes driven by more common organisms have treatment failures driven by otherwise susceptible pathogens hiding in intracellular niches. This includes: S. aureus, which forms intracellular small-colony variants in chronic/recurrent osteomyelitis and relapsing MRSA bacteremia.^108,109 It can also include S. pyogenes in recurrent pharyngitis with penicillin treatment failure.^110,111 Intracellular E. coli have even been strongly linked to Crohn’s disease.^112–114
• iii
  Persister populations: A third way bacteria can hide is by a “persister” phenotype.^115,116 These bacteria lack genetic determinants of resistance, yet are not killed by antibiotics. In these cases, the pathogens typically enter a dormant metabolic state under antibiotic stress. In this state, the antibiotics have little efficacy. Once antibiotic pressure subsides, the pathogens revert to metabolic activity, causing disease relapse. The persister phenotype is often found within bacteria hiding in intracellular niches and biofilms, so this mechanism is often complementary to other hiding strategies already described.

Third, because antibiotics are insufficiently selective, they kill many helpful bacteria in our microbiomes as collateral damage to the pathogen of concern. Increasingly, we understand that the immune system and microbiome work together as an integrated defense to infection. Antibiotics work at cross purposes by killing the pathogens, but also by damaging the very system they are meant to protect. We discuss these problems in detail in our next chapters on the link between the immune system and microbiome and the specific case in immuno-oncology.

For completeness, we acknowledge that there are other causes of treatment failure. Examples include: immune-compromised hosts, lack of adequate source control with infected prosthetics or catheters,¹¹⁷ and good drugs that are under-dosed.¹¹⁸ We argue that these causes, too, would not be improved simply with “stronger”, next-generation antibiotic. Under-dosing is solved with better pharmacology. Source control is improved by careful protocols. Even immune-compromised hosts would benefit greatly from some sort of combination therapy: adding a mechanism that is distinct from antibiotics to help the host clear the infection sooner.

What should we do?

Up to this point, the world has responded to treatment failure by engaging in an arms race with bacteria. As resistance rates have risen, guidelines have moved stronger drugs into empiric therapy. Consequently, bacteria developed more resistance with the result that still stronger drugs are moved into empiric therapy, and more “last resort” medicines developed. The arms race response has run its course and is no longer a viable solution. As mentioned, very few new antibiotics have been developed over the past several decades, and most of those that have received approval have failed in commercialization. Conglomerates have exited the field, and smaller developers have gone bankrupt.

The treatment history for gonorrhea provides a textbook example of the antibiotic arms race. Treatment began with sulfonamides when antibiotics first became widely available; then soon escalated to penicillin; followed by tetracyclines, then fluoroquinolones. Resistance development to front-line treatment precipitated each escalation. Cephalosporins are now the treatment mainstay, with resistance to ceftriaxone, an often-used member of the class, already emerging.¹¹⁹ Most other infectious syndromes exhibit similar escalation patterns over time.

Many in the field would suggest that the solution to escalation lies in a precision medicine approach: narrow-spectrum antibiotics targeted at very specific pathogens. While this would help overcome resistance, most approaches do not address hiding bacteria. Moreover, we think this direction is a commercial dead-end limited by extremely small biopharmaceutical markets, far below the one billion dollar a year threshold often used to benchmark a candidate drug’s commercial attractiveness. Compared to other opportunities a drug financier could fund with the same development dollars, the return on narrow-spectrum antibiotics is simply not compelling. We discuss this at length in our chapter on narrow-spectrum antibiotics.

The stage is set for Unison and other innovators to do something different. Infectious disease treatment failure is an enormous problem, with massive unmet need. It will not be solved by making more potent antibiotics. A therapeutic that addresses the root causes of failure will be a clinical and commercial success. Defense disablers unlock the opportunity.

Methodology for assessing cost of treatment failure

Our analysis provides a bottom-up model of the US burden of bacterial treatment failure based on values reported in the literature. For each major infectious syndrome we estimate four quantities from the published literature: (i) the annual US incidence of the syndrome; (ii) the frequency of antibiotic treatment failure; (iii) the excess direct medical cost of a failure episode over an uncomplicated, successful course; and (iv) the excess mortality associated with failure. The annual cost of treatment failure for each syndrome is the product of the first three:

The cost of failure is defined as the excess resources consumed when first-line therapy fails, driven principally by additional hospital length of stay, complications and additional morbidity, re-intervention or surgery, and readmission. It is a healthcare-system (direct medical cost) perspective; indirect costs such as lost productivity are excluded. We seek to report the incremental cost of treatment failure in one individual as compared to a similar individual with the same infection type who does not experience treatment failure.

Definition: An intuitive definition of antibiotic treatment failure is “any situation where bacteria survive antibiotic treatment and the clinical symptoms of the infection persist”.¹²⁰ A working technical definition is: “any of the following events within 30 days of initial antibiotic: antibiotic refill, antibiotic switch, emergency room visit, or hospitalization”.⁵³ Unfortunately, there is not a standardized definition used in the field, which contributes to a wide reported range of treatment failure rates. In accordance with our objective of determining an order of magnitude to the cost of treatment failure, we do our best to apply this definition to the papers we cite, even if the authors do not use the same definition themselves.

Cost normalization (2025 USD). Source figures span publication years from 2008 to 2023. Every cost was inflated to 2025 dollars using the US Bureau of Labor Statistics Consumer Price Index for Medical Care (CPIMEDSL, 1982–84 = 100). The deflators applied were approximately ×1.42 (2012), ×1.32 (2015), ×1.24 (2017), ×1.22 (2018), ×1.18 (2019), and ×1.05 (2023). The 2025 index (~590) was cross-checked against the mid-2026 published value (~594).

Excess mortality. Excess deaths are the additional deaths associated with failure (failure-vs-success mortality difference applied to the failing population) and are associational, not strictly causal: patients who fail therapy are frequently sicker at baseline, so the estimates should be read as an upper-bound signal of the human cost, not an attributable-death count.

Scope decisions. Nine types of infections were included as the most economically significant, with CAP split into inpatient and outpatient, yielding a total of ten analyses. Sepsis is deliberately excluded because it is a downstream state arising from a primary syndrome (counting it would double-count CAP, intra-abdominal, urinary, and bloodstream sources). Otitis media is excluded because the cost of failure per episode is low. Gonorrhea, endocarditis, osteomyelitis (as a standalone), and brain abscess were considered and excluded: gonorrhea has very high incidence but low cost per failure; endocarditis and brain abscess are comparatively infrequent; and diabetic-foot osteomyelitis is captured within the diabetic foot infection row. To avoid double-counting bacteremias that are secondary to the pneumonias, UTIs, and intra-abdominal infections already modeled, the bloodstream-infection row counts primary bloodstream infection only (we estimate approximately half of the ~600,000 annual US episodes).

Limitations. This work is meant to:

• i
  Illustrate what the literature already reports in terms of treatment failure frequency and costs.
• ii
  Provide an aggregate estimate across major infectious syndromes.

Our methodology is therefore secondary research, limited by the different definitions, methods, and sources that the underlying studies used, and the difficulties inherent in combining diverse sources. Our results are estimates, not firm calculations.