Holistic Composite Endpoints With Strategic Analyses: Toward More Informed Clinical Decision-Making for Gram-Negative Bacterial Bloodstream Infections

Increasing resistance in common bacterial pathogens to available antibiotic treatment options is a slow-moving, ongoing, unrelenting global health crisis that impacts all fields of medicine. In gram-negative bacteria, 2 mechanisms of resistance are increasingly important and prevalent worldwide. First, enzymatically mediated resistance due to metallo-β-lactamases (MBLs) is on the rise. Of these, the New Delhi MBL (NDM) is the most important family. Recent epidemiologic trends suggest that NDM has now replaced the serine carbapenemase family of Klebsiella pneumoniae carbapenemase as the most common cause of enzymatic carbapenem resistance in Enterobacterales in the United States [1, 2]. A second concern involves Escherichia coli and Klebsiella pneumoniae strains with insert mutations in the gene encoding penicillin-binding protein 3 (PBP3) rendering them less susceptible to β-lactam antibiotics—including cefiderocol—that bind primarily to PBP3 [3]. In addition to antibiotic stewardship and infection prevention measures that decrease the spread of resistance, novel antibiotics and adjunctive therapies, such as bacteriophages and monoclonals, are needed to treat patients infected with these resistant bacteria. Cefiderocol was approved by the US Food and Drug Administration in 2019, with promisingly broad in vitro activity that includes the highest priority gram-negative bacteria. Seven years later, the question of how best to use this agent is not yet resolved. In this issue of Clinical Infectious Diseases, Hardy et al report a secondary analysis of the GAME CHANGER trial; an open-label randomized trial, in which cefiderocol was noninferior to standard of care for the primary endpoint of 14-day mortality in patients with gram-negative nosocomial bloodstream infection [4]. In the carbapenem-resistant subset, 14-day mortality was 14% (9/64) in the cefiderocol arm versus 10% (6/63) in the standard-of-care arm, with a notable difference in mortality (5/16 vs 0/11) in patients infected with MBL producers [4, 5]. While limited by small numbers, this suggests caution when considering the use of cefiderocol in patients infected with MBL producers. This effect appears primarily driven by bacteria that produce enzymes in the NDM family, which more effectively hydrolyze cefiderocol when compared to other MBL enzymes [5]. In the secondary analyses reported here in Clinical Infectious Diseases, 2 holistic composite endpoint assessments were applied: a 5-level desirability of outcome ranking (DOOR) outcome analyzed using the DOOR paradigm and a composite based on 6 hierarchical endpoints analyzed using generalized pairwise comparisons (GPCs). Importantly, the endpoints were defined before study results were known to investigators. These approaches allow for more granular and patient-centered assessment of patient status after antibiotic treatment and are increasingly used in antibiotic clinical trials (Table 1). For instance, the recently published DOTS and ANDROMEDA-SHOCK-2 trials had holistic composites as primary endpoints [8, 9]. In DOOR analysis, the 3 events of interest—treatment failure (recurrent positive blood cultures), infectious complications (multidrug-resistant bacteria colonization/infection and/or candidemia), and serious adverse events—all occurred in 12%–16% in both treatment arms. All-cause mortality at 90 days was also similar at ∼29%. The overall DOOR analysis indicated a 50.4% (46%–55%) probability that a patient treated with cefiderocol would have a more desirable outcome at day 90 when compared to a patient on standard-of-care antibiotics. The null hypothesis of no difference between groups cannot be rejected since the 95% CI includes 50%. A benefit of DOOR analysis is that in addition to all-cause mortality, some indication is provided on the likelihood of having a good outcome without any of the events of interest. In GAME CHANGER, this only occurred in slightly more than half of patients. It should be noted that both endpoints of treatment failure and infectious complications could be met by patients without necessarily resulting in a negative clinical impact. For instance, a participant colonized with methicillin-resistant S. aureus would meet the infectious complications endpoint. Generally, when constructing a DOOR outcome, investigators are encouraged to select endpoints that are clinically important and that indicate how patients feel, function, or survive [7]. The functional bloodstream infection score (FBIS) that was used as a tiebreaker for patients with identical DOOR ranks is a nice example of an outcome that reflects the patient's functional status. Principles of Holistic Composite Endpoints Composing information from individuals is patient centric and pragmatic, providing insights that separate analyses of component endpoints cannot. It is the only way to recognize the correlations among component endpoints and the cumulative nature of different outcomes in individual patients. Composites do not have to be binary, and components do not have to be treated with equal importance. Death may be the most important component Although binary endpoints are common, patient response is more finely graded. A mortality outcome does not distinguish the responses of patients who survive with morbidity and necessary organ support from those surviving without such conditions. A clinical failure endpoint does not distinguish the responses of patients who died from those that survive with another failure condition. Composing information does not imply abandonment of analyses of component outcomes. Component outcomes are analyzed to fully understand the nature of the effects. Composites and its components are analyzed and interpreted together. Analyses should be on a common scale, ie, absolute risk, such as the DOOR probability and net benefit, as relative risk measures cannot be simultaneously interpreted due to different baseline risks [6]. Different composite outcomes represent different clinical questions. The definition of the composite should be based on clinical importance rather than which produces a desired result. Principles include: Defining gradations of patient response to ensure recognition of responses that are importantly different Ensuring defined gradations are clinically important to assure that nonmeaningful effects are not overemphasized Aiming for simplicity to aid in interpretation [7] Composites should be prespecified. There are different strategies for composing information and recognizing important gradations of patient response. Hierarchical endpoints are 1 strategy. Once a hierarchy is defined, a distinct ordinal outcome results. Cumulative evaluation of the culminating ordinal outcome is advised regardless of composition strategy. Rank-based measures, such as the DOOR probability and net benefit, are tools for analyses of composites. These approaches are robust in that they do not make assumptions about the perceived spacing between the ordinal outcome categories, basing analyses entirely on ranks. These analyses can be supplemented with deliberate “partial credit” grading that directly incorporates perspectives on the levels of the ultimate ordinal outcome categories [7]. Like DOOR, GPC focuses on what happens to a specific patient, so that the co-occurrence of lack of efficacy and toxicity in a single patient will impact conclusions differently when compared to one patient with lack of efficacy and another patient with toxicity. This is a fundamental difference between the holistic composite endpoint approaches and the traditional reporting of clinical trials, when there is a focus on siloed analyses of toxicity and efficacy endpoints. All patients in the cefiderocol arm were compared pairwise to all patients in the standard-of-care arm, and for each of these thousands of comparisons, wins (more desirable), losses (less desirable), and ties (tied desirability) were assigned based on sequentially compared outcomes of mortality at 30 days, clinical failure at 14 days (defined as lack of improvement of the Sequential Organ Failure Assessment Score), serious adverse events related to treatment, FBIS improvement, and duration of hospitalization. The net treatment benefit (NTB) has 0% as its null value, and GPC showed no difference between treatment groups with a NTB of 2% (−8%–12%). Of note, patients infected with acquired MBL-producing organism were separately assessed using the GPC, and in this subgroup, a trend toward negative outcomes was observed with cefiderocol. The GAME CHANGER trial represents a remarkable addition to our knowledge on how to use cefiderocol in clinical practice. Unfortunately, most antibiotics come to market with limited clinical evidence for those infections where they are most needed. Therefore, postmarketing evaluation of novel antibiotics with activity against multidrug-resistant organisms is urgently needed to inform optimal use of these expensive treatment options. Enrollment of patients with highly resistant bacterial infections in clinical trials is challenging. These patients tend to be both chronically and acutely ill and have an underlying mortality risk that is not entirely modifiable by antibiotic treatment [10]. Furthermore, resistant pathogen-directed clinical trials require species identification and antimicrobial susceptibility determination in study participants, which decreases the amount of time available for recruitment. Observational data represent the most abundant source of postmarketing evaluation for antibiotics. However, confounding by indication and other biases, such as immortal time bias, may limit interpretations, especially in the case of patients infected with highly resistant bacteria. Patients treated with a newer and more expensive antibiotic option tend to differ from patients treated with more traditional and cheaper options in various ways. Target trial emulation has the potential to reduce, although not completely avoid, these concerns. A combination of well-designed observational and investigator-initiated randomized trials is needed to guide clinical use of novel antibiotics. Holistic composite endpoints are useful tools in these efforts to provide a more granular view of how outcomes after distinct treatment approaches may differ. In summary, reports from randomized controlled trials that include holistic composite endpoints provide much greater granularity to know what to expect in our patients. While these outcomes did not change the overall interpretation of this specific trial, the results show the lack of clear benefit of cefiderocol relative to alternative approaches with greater granularity and offer clinicians more information as they explain to patients and their families what to expect. Financial support. No funding was provided for this work.