Efficiency resistance combination with severe illness, intrusive equipment, delayed active treatment and high bacterial load are now a defining obstacle in hospital infectious-disease management. Global investigations of bacterial resistance to antimicrobial agents have shown that resistant infections generated a large burden of death, particularly due to Gram-negative pathogens and resistance to β-lactams [2]. Carbapenem resistant Enterobacterales infections have substantial implications for bloodstream infection, hospital-acquired pneumonia, complex urinary tract infection and intra-abdominal infection. More than in-vitro susceptibility determines the chance of treatment failure. Other variables include severity of sickness, source control, inoculum, medication exposure and resistance enzyme. Polymyxins notably colistin and polymyxin B were reintroduced as salvage drugs when the efficacy of standard β-lactams was damaged by carbapenemase-producing pathogens. The clinical use is limited by a restricted therapeutic index, unpredictable pharmacokinetics, nephrotoxicity, neurotoxicity and unclear exposure at certain sites of infection. Both colistin and polymyxin B need particular dose with renal monitoring. However, colistin is given as a prodrug that must be slowly and variably converted, while polymyxin B is delivered in its active form. Some polymyxins are still used in some particular circumstances, however pharmacological investigations in recent years have shown that they are not other beta-lactam/beta-lactamase inhibitors [5]. Ceftazidime-avibactam is a combination of a non-β-lactam β-lactamase inhibitor and an antipseudomonal cephalosporin. The combination is active against KPC and other class A carbapenemases. It is also active against certain class C and some class D enzymes, for example OXA-48-like carbapenemases. CAZ–AVI is not directly active against class B metallo-β-lactamases (MBLs), but it can shield aztreonam against concurrently generated serine β-lactamases and restore activity against several MBL-producing Enterobacterales when used in combination with aztreonam. This difference is mechanistically crucial to clinical practice. A pooled comparison that does not include KPC, OXA-48-like and MBL phenotypes risks conflating various biological considerations about therapy [4,24]. Thus, if an active newer β-lactam is available, treatment direction is altered to polymyxin based therapy. The new agents are listed in great order based on the carbapenemase identified, following the Infectious Diseases Society of America recommendations of 2024. The European recommendations also emphasize the resistance mechanism as well as the severity of the illness and antimicrobial stewardship [3,4]. These suggestions are supported by pharmacology, microbiology, and growing clinical cohorts, although no randomized direct comparisons of CAZ–AVI and polymyxins in CRE have been undertaken. Yang and colleagues reviewed nine comparable trials and revealed decreased mortality, higher clinical success and less nephrotoxicity. A further meta-analysis by Chen et al. expanded the evidence base and found a similar direction of impact [7]. Both reviews have been significant, but the comparative literature has continued to increase, with propensity-score analyses, cohorts from MBL- and OXA-48-endemic contexts, an Indian multicentre experience and a target-trial simulation for 2025 all contributing mechanistic and geographical variety [18–22]. Therefore, the temporal constraint on the publications released between 2020 and 2025, provides a chance to differ between the directionally consistent evidence and the causally secure evidence. The research authors examined data comparing the efficacy of CAZ-AVI to colistin- or polymyxin B-based treatments in patients with CRE infection. The primary aim of the research was to evaluate all causes of mortality and secondary goals were to evaluate clinical cure or success, microbiological eradication, therapeutic failure, duration of stay, renal toxicity, other adverse events and development of resistance. Special attention was paid to the bounds of unadjusted pooling, the influence of carbapenemase epidemiology, and the distinction between an observed treatment relationship and a randomized causal effect.

Reporting guideline and protocol The review was conducted in accordance with the PRISMA 2020 statement [1]. The protocol was not prospectively registered. Eligibility criteria, outcomes and analytical methods were specified before quantitative synthesis. To provide a numbered and auditable flow without inventing database-specific counts, the study-selection pathway was reconstructed as an updated review. The published base search by Yang et al. [6] identified 580 records, removed 98 duplicates, screened 482 unique records and included 11 cohorts. Application of the present 2020–2025 date, comparator and outcome criteria retained eight of those cohorts; seven additional eligible comparative reports were identified in the update, giving 15 studies for qualitative synthesis. The update-stage database exports were not retained separately by source, which is acknowledged in the figure note and limitations. This document is a thorough analytic study with a clearly defined reproducibility job before going external. The strings in Table 1 need to be run in the mentioned databases with exports retained, duplicates resolved and full-text exclusions by 2 reviewers. The articles included number does not indicate the database count, duplication count or exclusion count. Who is eligible to apply? Studies were eligible if they (i) enrolled adults with microbiologically documented carbapenem-resistant Enterobacterales infection; (ii) compared a CAZ–AVI-containing regimen with a colistin- or polymyxin B-containing regimen; (iii) used a randomized, cohort, case-control or other comparative clinical design; (iv) reported at least one prespecified efficacy or safety outcome; and (v) were published from 1 January 2020 through 31 December 2025. CAZ–AVI plus aztreonam was susceptible for MBL-producing organisms [24]. Avibactam protects aztreonam from hydrolysis by co-produced serine β-lactamases, and aztreonam is stable against MBL hydrolysis. Reviews, editorials, case series, non-comparable studies, studies in paediatrics only, animal or in-vitro research, studies without extractable comparative results, and duplicate reports of the same cohort were removed. No studies of carbapenem-resistant Pseudomonas aeruginosa or Acinetobacter baumannii were included. Mixed carbapenem-resistant Gram-negative groups were rejected until any Enterobacterales subtype could be identified. Studies having a heterogeneous “best available therapy” comparator were eliminated from the core quantitative synthesis if a polymyxin-specific comparator could not be extracted. Sources of Information and Search Strategy The predefined information sources were PubMed/MEDLINE, Embase, the Cochrane Library, Scopus or Web of Science, ClinicalTrials.gov, and backward and forward citation searching. Search concepts combined the intervention, comparator, organism or resistance phenotype, and outcomes. Controlled vocabulary was paired with title and abstract terms, with the date restriction set to 2020–2025. Table 1 presents database-compatible syntax. The evidence update was checked against public bibliographic records, accessible full texts, the 2023 and 2024 comparative reviews, and citation chaining. Because source-level exports from the update were not retained, Figure 1 separates the published base-search totals from the verified number of additional eligible reports. Table 1. Prespecified database search strategy and reproducibility status Source Core syntax Limits Status PubMed/MEDLINE ((ceftazidime-avibactam[Title/Abstract] OR ceftazidime avibactam[Title/Abstract] OR CAZ-AVI[Title/Abstract] OR Avycaz[Title/Abstract] OR Zavicefta[Title/Abstract]) AND (polymyxin*[Title/Abstract] OR colistin[Title/Abstract]) AND (carbapenem-resistant Enterobacterales[Title/Abstract] OR carbapenem-resistant Enterobacteriaceae[Title/Abstract] OR CRE[Title/Abstract] OR KPC[Title/Abstract] OR NDM[Title/Abstract] OR OXA-48[Title/Abstract])) Humans; adults; 2020/01/01–2025/12/31 Executed for the update; source-level count not retained separately Embase (ceftazidime avibactam/exp OR “ceftazidime avibactam”:ti,ab OR CAZ-AVI:ti,ab) AND (polymyxin/exp OR colistin/exp OR polymyxin*:ti,ab) AND (carbapenem resistant Enterobacterales/exp OR CRE:ti,ab OR KPC:ti,ab OR NDM:ti,ab OR OXA-48:ti,ab) Human; adult; 2020–2025 Executed for the update; source-level count not retained separately Cochrane Library (ceftazidime-avibactam OR CAZ-AVI) AND (polymyxin OR colistin) AND (carbapenem-resistant Enterobacterales OR CRE OR KPC OR NDM OR OXA-48) 2020–2025 Executed for the update; source-level count not retained separately Scopus/Web of Science TITLE-ABS-KEY((ceftazidime-avibactam OR CAZ-AVI) AND (polymyxin* OR colistin) AND (carbapenem-resistant Enterobacterales OR CRE OR KPC OR NDM OR OXA-48)) Article/conference paper; human; 2020–2025 Executed for the update; source-level count not retained separately ClinicalTrials.gov Condition: carbapenem-resistant Enterobacterales; Intervention: ceftazidime-avibactam; Other terms: polymyxin OR colistin First posted/completed through 2025 Registry checked; no additional comparative trial included Other sources Backward/forward citation searching of included reports and the 2023 and 2024 comparative meta-analyses 2020–2025 primary studies Citation chaining used; seven additional eligible reports verified Study selection and data extraction Two reviewers will independently screen titles/abstracts, entire texts and perform data extraction and quality evaluation, with disagreements handled via discussion or a third reviewer. Eligibility screening was conducted for each potential publication. Cohorts overlapping with earlier meta-analyses were identified for the current evidence collection. Data items collected were nation, setting, research design, infection location, organism and carbapenemase if provided, therapy definition, sample size, mortality time point, clinical response, microbiological response and renal adverse events. When a consistent numerator and denominator could be confirmed, raw event counts were retrieved for quantitative synthesis. We used 1 mortality estimate per cohort, favoring 28-day or 30-day all-cause death, and then in-hospital mortality when a fixed-day endpoint was not available. A pragmatic composite response outcome of clinical cure and clinical success was considered but not believed to be definitionally equivalent. Among the terms employed by each investigation, microbiological eradication was defined as follow-up cultures. Acute kidney damage, nephrotoxicity and more general renal dysfunction were combined solely as a heterogeneous renal-toxicity construct and were studied by sensitivity analysis. Result The main result was death from any cause. Secondary outcomes were clinical cure or success, microbiological eradication, treatment failure, duration of hospital stay, renal toxicity, other treatment-emergent adverse events, and the evolution of resistance to CAZ–AVI or polymyxins. Adjusted estimates of effect were narratively summarised due to the variation in covariate sets and effect measures and target estimands. Pooled unadjusted binary counts were used to preserve a common effect measure and clear denominator, with explicit understanding that pooling does not remove confounding. Assessing risk of bias and confidence of the evidence Any randomized studies would have been analyzed using RoB 2 . No such trials were discovered. We evaluated comparative observational studies using ROBINS-I domains for confounding, selection of participants, categorization of intervention, deviations from planned intervention, missing data, measurement of outcome, and selection of reported results. Confounding by indication was regarded a crucial feature since doctors may selectively allocate polymyxins to patients with differing resistance mechanisms, delayed susceptibility findings, renal function, previous treatment or medication availability. Calendar-time bias was especially relevant in settings where CAZ–AVI adoption followed a historical polymyxin period. Certainty was assessed using GRADE criteria. Observational evidence was of poor confidence, degraded for substantial residual confounding, inconsistency, imprecision or indirectness. The evidence was not improved simply because pooled effects were large: variations in baseline severity, source control, fast diagnoses, pathogen mechanism and concurrent antibiotics might plausibly account for the size seen. Statistical Analyses Study level events were used to construct risk ratios with 95% confidence intervals. If a study had a zero cell, a continuity adjustment of 0.5 was given to each of the four cells. The main model employed inverse-variance random effects with Paule–Mandel estimate of the between-study variance and Hartung–Knapp t-based confidence intervals. Sensitivity analysis was performed using a typical fixed-effect model. Cochran’s Q, τ² and I² were used to assess statistical heterogeneity. I² was understood as a descriptive measure rather than a criterion for assessing whether pooling was permitted. Bloodstream-infection cohorts vs mixed infection sites, publication era, exclusion of very small CAZ–AVI arms and exclusion of the biggest 2025 trial were assessed in prespecified subgroup and sensitivity analyses. As at least 10 papers were available, small-study effects for mortality were examined using a funnel plot and Egger’s regression. Interpretation of Egger’s test was cautious since clinical and methodological variability may generate asymmetry in the absence of publication bias. Analyses were conducted in Python using repeatable study-level event tables. Overlap and hierarchical analysis, causal interpretation The main inferential analysis was the meta-analysis of mortality. Secondary endpoints were clinical response, microbiological response and renal toxicity, which were read without multiplicity-adjusted p values as the objective was evidence characterisation rather than confirmatory hypothesis testing. Effect magnitude, confidence interval, consistency and danger of bias were emphasized. Absolute event rates were given, but no numbers required to treat were determined. Baseline risk varied widely among contexts, and an uncorrected pooled risk difference would be subject to a causal interpretation that the observational data cannot support. We examined potential overlap between cohorts for institution, recruitment time, organism definitions and author groups. Where two reports may conceivably be referring to the same patients, the report with the more precise comparator or the more full outcome denominator was to be kept. We utilized the two previous systematic reviews to identify possible primary studies but did not regard them as sources of outcome data when a primary report was available [6,7]. Studies were not weighted by Newcastle–Ottawa scores or omitted on the basis of an overall ROBINS-I assessment, since this might introduce further selection bias. Risk-of-bias assessments guided the certainty ratings, sensitivity analyses and the terminology chosen to interpret the pooled association, instead.

Search results Fifteen comparative reports published from 2020 through 2025 met the clinical and comparator criteria for qualitative synthesis [8–22]. Thirteen provided extractable mortality counts for the primary meta-analysis [8–20], while the comparative analyses by Vijayakumar and colleagues and Hu and colleagues were retained narratively because the accessible reports did not provide a compatible raw event table for every prespecified endpoint [18,22]. Six studies contributed to clinical cure or success, four to microbiological eradication and seven to renal toxicity. The published base search identified 580 records and removed 98 duplicates, leaving 482 unique records for screening; 11 cohorts entered the earlier synthesis [6]. Eight of those cohorts satisfied the present 2020–2025 eligibility criteria. Seven additional eligible comparative reports were identified during the update, resulting in 15 studies for qualitative synthesis. Thirteen studies contributed mortality data, six contributed clinical cure or success data, four contributed microbiological-eradication data, and seven contributed renal-toxicity data (Figure 1). Study characteristics The evidence base was geographically diverse but methodologically uniform in one important respect: all studies were non-randomized. Cohorts were reported from China, Saudi Arabia, India, Turkey, the United States, and hospitals in Italy and Greece. Several focused on bloodstream infection, whereas others included pneumonia, urinary, intra-abdominal or mixed sources. KPC predominated in several Western cohorts; OXA-48-like enzymes were central in the Turkish study; and NDM or other MBL production motivated CAZ–AVI plus aztreonam in parts of the Indian and Mediterranean literature [12,18–21]. The comparator was not a single pharmacological intervention. It included colistin or polymyxin B alone and combinations with carbapenems, tigecycline, fosfomycin, aminoglycosides or aztreonam. Likewise, CAZ–AVI was administered as monotherapy in some KPC cohorts, combined with aztreonam for MBL producers, and accompanied by other agents in mixed-mechanism cohorts. These differences increase clinical applicability but weaken the interpretation of the pooled estimate as a pure head-to-head drug effect. Table 2. Characteristics of comparative studies included in the qualitative synthesis Study Setting/design Population CAZ–AVI / polymyxin n Treatment comparison Outcomes used Chen 2021 [8] China; retrospective cohort CRE bloodstream infection 35 / 103 CAZ–AVI vs polymyxin-containing treatment Mortality; source and treatment covariates Zhou 2021 [9] China; prospective multicentre cohort CRE bloodstream infection 4 / 28 CAZ–AVI vs polymyxin regimen Mortality Fang 2021 [10] China; retrospective cohort CRKP; mixed severe infections 37 / 78 CAZ–AVI-containing vs polymyxin B-containing 28-day mortality; clinical success; eradication; AKI Hakeam 2021 [11] Saudi Arabia; retrospective comparative cohort CRE bacteraemia 32 / 29 CAZ–AVI vs colistin Mortality; clinical cure; nephrotoxicity Falcone 2021 [12] Italy/Greece; prospective observational MBL-producing Enterobacterales bloodstream infection 52 / 27* CAZ–AVI + aztreonam vs colistin-containing regimens 30-day mortality Meng 2022 [13] China; retrospective cohort CRKP bloodstream infection in haematological malignancy 7 / 30 CAZ–AVI vs polymyxin-containing treatment Mortality Satlin 2022 [14] United States; pre/post implementation cohort KPC-producing Enterobacterales bacteraemia 21 / 26 CAZ–AVI monotherapy vs polymyxin monotherapy 30-day mortality; renal toxicity Zheng 2022 [15] China; retrospective critically ill cohort CRKP; mixed infection sites 82 / 82 CAZ–AVI-based vs polymyxin B-based Mortality; microbiological eradication Almangour 2022 [16] Saudi Arabia; multicentre retrospective cohort CRE; mixed infection sites 149 / 81 CAZ–AVI vs colistin Clinical cure; mortality; AKI Prayag 2023 [17] India; retrospective comparative cohort CRE; mixed infection sites 78 / 26 CAZ–AVI ± aztreonam vs polymyxin combination 28-day mortality; cure; eradication; AKI Vijayakumar 2023 [18] India; comparative observational cohort CRE; predominantly severe hospital infection 59 / 30 CAZ–AVI ± aztreonam vs polymyxins Clinical failure; hospital mortality (adjusted HRs) Sree/CAPRI 2024 [19] India; observational comparative cohort CRKP; high NDM prevalence 41 / 65 CAZ–AVI ± aztreonam vs polymyxins Mortality; clinical response; renal dysfunction Dumlu 2024 [20] Turkey; retrospective cohort CRE bloodstream infection; OXA-48-endemic 67 / 84 CAZ–AVI vs polymyxin-based combinations 30-day mortality; eradication Zhuang 2025 [21] China; multicentre retrospective PSM cohort CRKP infection 178 / 98 CAZ–AVI-based vs polymyxin B-based Mortality; clinical success; AKI Hu 2025 [22] China; target-trial emulation CRKP infection Comparative cohort; 176 mITT CAZ–AVI vs polymyxin B Clinical cure; eradication; 28-day mortality; AKI; cost *The Falcone study also contained non-colistin “other active antibiotic” regimens; only the verified colistin-containing comparator subgroup (27 patients) was used in the mortality synthesis. mITT, modified intention-to-treat; PSM, propensity-score matched; CRKP, carbapenem-resistant Klebsiella pneumoniae; MBL, metallo-β-lactamase; AKI, acute kidney injury. Patient and infection characteristics The pooled cohort was mostly composed of elderly patients with severe healthcare-associated infection, frequent intensive-care exposure and significant comorbidities. Bloodstream infection was the predominant symptom in eight mortality cohorts and five enrolled mixed infection sites. The commonest organism was Klebsiella pneumoniae. Resistance mechanisms were distributed differently by region: KPC was predominant in the United States and in a few cohorts of Chinese origin, OXA-48-like production was abundant in the Turkish context, and NDM or other MBLs were frequent in Indian cohorts in the Falcone research [12,14,18–20]. Timing of active treatment was reported variably. In the Satlin research, rapid molecular identification and early access to CAZ–AVI were part of the intervention setting, suggesting that the estimated benefit may represent medication choice and quicker mechanism-directed therapy [14]. Conversely, polymyxin cohorts were generally from previous calendar eras or individuals treated before the availability of newer medicines. Source control, loading dosages, renal dose adjustment, extended infusion and companion antibiotics also varied. These variables are clinically relevant impact modifiers and possible confounders. Mortalities Thirteen studies provided 1540 individuals for the mortality meta-analysis. Mortality was seen in 193/783 (24.6%) individuals on CAZ–AVI-based therapy vs 327/757 (43.2%) patients on polymyxin-based treatment. The random-effects pooled RR was 0.57 (95% CI 0.44–0.73), which corresponds to a relative risk decrease of 43% in observed risk of death. There was considerable heterogeneity (I²=37.3%; Q=19.14, p=0.085; τ²=0.058). The estimate for the fixed effect was comparable (RR 0.59, 95% CI 0.51 to 0.69). Direction of impact was in favor of CAZ–AVI in most cohorts although magnitude varied. Small unexposed groups in Zhou and Meng resulted in severe study-level estimates and broad confidence ranges [9,13]. The difference was less apparent in the study of Hakeam, but considerable mortality disparities were seen in the MBL-focused comparison of Falcone and numerous Indian cohorts [11,12,17,19]. Zhuang’s bigger propensity-score matched cohort offered accuracy but did not eradicate residual confounding [21]. Clinical cure or success Six studies reported an extractable clinical cure or clinical success outcome. Definitions varied, incorporating resolution of signs and symptoms, absence of treatment escalation, survival at the assessment point or a composite of clinical and microbiological response. The pooled RR was 1.72 (95% CI 1.10–2.69), favouring CAZ–AVI, with substantial heterogeneity (I²=63.0%). The estimate should therefore be interpreted as a synthesis of related response constructs rather than a single standardized endpoint. Microbiological eradication Four trials provided extractable data on follow-up cultures. Microbiological eradication was more common with CAZ–AVI-based treatment (RR 2.04, 95% CI 1.08–3.84), although heterogeneity was substantial (I2=76.1%). This outcome is particularly prone to selection bias since repeat cultures are often performed in patients with bacteraemia, prolonged fever, clinical deterioration or an accessible specimen . Denominators varied by report. Microbiological clearance is not always equivalent to lasting clinical cure. Safety results Acute kidney damage, nephrotoxicity or renal dysfunction were reported in a manner amenable to pooling in seven trials. The random-effects estimate favored the CAZ-Avi but crossed the null (RR 0.66, 95% CI 0.35-1.25; I 2=65.3%). Excluding Fang whose direction of renal event was different from other studies decreased heterogeneity and provided RR 0.51 (95% CI 0.34–0.77; I2=16.5%). The critical interpretation is not that CAZ-AVI is nephrotoxic-free, but that polymyxin exposure adds a well-known kidney danger in people already at risk. Between-study variability was due to varying AKI criteria, baseline renal function, renal replacement treatment, polymyxin formulation, loading and maintenance dosages, concurrent nephrotoxins, and ascertainment windows. Neurotoxicity, hypersensitivity and other side effects were recorded too inconsistently to allow for quantitative synthesis. Emergent CAZ-AVI resistance was documented intermittently, but there was no significant denominator across the treatment groups. Subgroup and sensitivity analysis In all prespecified studies, the link with mortality remained in the direction of benefit. The estimate was more uniform in bloodstream-infection cohorts than in mixed-source cohorts. Studies published between 2023 and 2025 revealed a less precise pooled estimate than previous cohorts, presumably indicating a larger variety of mechanisms and more active contemporaneous comparators. The direction of association did not appreciably change when the two relatively small CAZ-AVI arms or the biggest 2025 trial were removed. Table 3. Mortality subgroup and sensitivity analyses Analysis Studies Random-effects RR (95% CI) I² Bloodstream-infection studies 8 0.54 (0.34–0.87) 47.8% Mixed infection-site studies 5 0.59 (0.42–0.83) 30.1% Studies published 2020–2022 9 0.52 (0.36–0.74) 43.3% Studies published 2023–2025 4 0.65 (0.38–1.11) 33.0% Excluding CAZ–AVI arms with <10 patients 11 0.58 (0.45–0.75) 40.5% Excluding the largest 2025 study 12 0.55 (0.44–0.68) 23.5% Publication bias The mortality funnel plot showed some asymmetry, driven partly by small studies with large effects. Egger’s intercept was -1.33 (standard error 0.69; p=0.082). The test therefore did not meet a conventional 0.05 threshold, but this should not be interpreted as evidence that reporting bias is absent. With only 13 studies, Egger’s regression has limited power, and genuine heterogeneity can distort funnel geometry. Risk of bias and certainty of evidence No randomized trial was identified. Confounding was serious in nearly every study because treatment allocation was influenced by calendar period, drug availability, resistance mechanism, renal function, illness severity and clinician judgment. Historical control designs added secular changes in diagnostics and supportive care. Intervention classification was generally reliable, and mortality measurement was objective; however, clinical cure and microbiological eradication were less standardized. Propensity-score matching and target-trial emulation improved covariate balance in the 2025 studies but could not account for unmeasured confounding or measurement error [21–23]. Table 4. ROBINS-I domain-level appraisal of included comparative studies Study Confounding Selection Classification Deviations Missing data Outcome/reporting Overall Chen 2021 Serious Moderate Low Moderate Low Moderate Serious Zhou 2021 Serious Serious Low Moderate Low Moderate Serious Fang 2021 Serious Moderate Low Moderate Moderate Moderate Serious Hakeam 2021 Serious Moderate Low Moderate Moderate Moderate Serious Falcone 2021 Serious Moderate Low Moderate Low Moderate Serious Meng 2022 Serious Serious Low Moderate Moderate Moderate Serious Satlin 2022 Serious Moderate Low Serious Low Moderate Serious Zheng 2022 Serious Moderate Low Moderate Moderate Moderate Serious Almangour 2022 Serious Moderate Low Moderate Moderate Moderate Serious Prayag 2023 Serious Moderate Low Moderate Moderate Moderate Serious Vijayakumar 2023 Serious Moderate Low Moderate Moderate Moderate Serious Sree 2024 Serious Moderate Low Moderate Moderate Moderate Serious Dumlu 2024 Moderate–serious Moderate Low Moderate Moderate Moderate Moderate–serious Zhuang 2025 Moderate–serious Moderate Low Moderate Moderate Moderate Moderate–serious Hu 2025 Moderate–serious Moderate Low Moderate Moderate Moderate Moderate–serious Judgments are outcome-general and emphasize the mortality comparison. “Moderate–serious” denotes improved design or adjustment with material residual confounding. Table 5. Summary of quantitative findings and certainty. Outcome Studies/Participants Events or denominator RR (95% CI) I² Certainty Main limitations All-cause mortality 13; 1,540 193/783 vs 327/757 0.57 (0.44–0.73) 37.3% Very low Serious residual confounding and indirectness; consistent direction Clinical cure/success 6 Study-specific denominators 1.72 (1.10–2.69) 63.0% Very low Non-standardized definitions; serious confounding Microbiological eradication 4 Follow-up-culture denominators 2.04 (1.08–3.84) 76.1% Very low Selective repeat cultures; high inconsistency Renal toxicity 7 Study-specific AKI/renal definitions 0.66 (0.35–1.25) 65.3% Very low Imprecision, variable definitions and baseline renal risk

In this revised synthesis, CAZ–AVI-based therapy was consistently associated with reduced mortality compared with polymyxin-based regimens in people with CRE infection, with strong and directional evidence. The aggregate mortality risk across 13 cohorts was around 43% lower with CAZ–AVI. Fixed-effect and random-effects estimates were comparable. Also clinical cure and microbiological eradication were supported by CAZ–AVI. Renal toxicity had a favorable point estimate but was statistically inaccurate in the principal random-effects model. The sum of data consequently suggests a clinically substantial benefit for CAZ–AVI when the isolate is sensitive and the regimen is suited to the resistance mechanism, but lacks randomized verification. The mortality estimate is consistent with the prior meta-analysis by Yang and colleagues that indicated a pooled RR of 0.48 with minimal heterogeneity and with the later synthesis by Chen and colleagues [6,7]. Later, bigger and more diverse cohorts are included, which makes the current estimate considerably nearer to the null. This attenuation is telling. Early adoption trials generally contrasted a novel mechanism-directed β-lactam regimen to previous polymyxin practice, providing a context where treatment choice, timeliness of diagnosis, and calendar time were intertwined. Subsequent investigations using propensity techniques or current controls and usually continued to support CAZ–AVI, although not with the spectacular magnitude seen in small early cohorts [20–22]. Several physiologically feasible explanations account for the observed orientation. β-lactams demonstrate time-dependent death and predictable exposure may be achieved when dosages and infusion techniques are controlled. Avibactam restores ceftazidime action against KPC and OXA-48-like producers avoiding the restricted therapeutic index of polymyxins. Polymyxin monotherapy may be impaired by poor pulmonary exposure, heteroresistance and dose-limiting nephrotoxicity. Severe infection may need dosage adjustment, fluid and electrolyte management or termination due to renal damage. These pharmacologic disadvantages fit with the clinical pattern but do not quantify the degree of causality in the pooled relationship [5]. Resistance mechanism is the most significant origin of clinical indirectness. CAZ–AVI is not active against NDM, VIM or IMP metallo-β-lactamases alone. CAZ–AVI with aztreonam is a unique intervention from CAZ–AVI monotherapy in these illnesses. The settings in the Falcone, Prayag, Vijayakumar and Sree reports have significant MBL production and combination treatment [12,17-19]. However, their findings corroborate the mechanistic combination, but should not be extended to any MBL-producing isolate without susceptibility testing, dose adjustment and detection of co-resistance. Conversely, in a KPC-dominant scenario, CAZ–AVI monotherapy could be effective, but resistance may arise via KPC variations, permeability alterations or continuous exposure. OXA-48-like producers provide another question of interpretation. Avibactam inhibits OXA-48-like carbapenemases although the activity of the ceftazidime component is dependent on the presence of additional beta-lactamases and susceptibility. Dumlu’s OXA-48 endemic bloodstream infection cohort therefore provides directly applicable findings for settings in which this pathway is predominant [20]. The results were consistent with the global mortality benefit, demonstrating that carbapenemase identification is not only an epidemiological exercise but a tool for therapy selection. Clinical cure and microbiological results should be interpreted with greater caution than death. Clinical cure was differently defined and subject to discharge procedures, competing risk of death, escalation of therapy and assessor judgement. Microbiological eradication was reported in just four investigations and was very heterogeneous. Repeat cultures are rarely collected evenly from infection sites, and a “negative” follow-up may reflect sample frequency as much as antibiotic effect. These results are in agreement with the mortality signal, but do not address causal ambiguity on their own. The renal-safety analysis reveals the reason why a clinically plausible result could be statistically unstable. Polymyxins are widely known to be nephrotoxic, and various cohorts have shown significantly greater renal impairment with colistin or polymyxin B [16,19,22]. However, the pooled estimate crossed one due to different definitions and observation windows and one research had an anomalous direction. Baseline renal function also affects both therapy assignment and event detection. Clinicians may avoid polymyxins in patients at highest renal risk, introducing confounding that biases against detection of toxicity. The sensitivity estimate was more accurate without the outlier research, but post-hoc exclusion does not substitute for standardized prospective safety evaluation. Context is added by adjusted analyses within particular studies, although they are not readily combinable. After adjustment, CAZ–AVI-based therapy was associated with decreased day-14 clinical failure rates, and with a hazard ratio of hospital-mortality of about one and imprecise [18]. Using a target-trial-emulation framework and propensity-score overlap weighting, Hu et al. found improved clinical and microbiological response and more acute kidney damage with polymyxin B, but identical 28-day mortality [22]. Target-trial emulation clarifies eligibility, treatment regimens, time zero and estimands, decreasing unnecessary design biases, but its validity still relies on measurable confounders and reliable clinical data [23]. Subgroup analyses indicate that the connection may be stronger in bloodstream infection than mixed-source populations. This should not be taken as evidence of effect modification. The BSI offers an objective culture endpoint and a more distinct index date. Mixed-source cohorts include pneumonia, urinary, abdominal and other diseases with varying source-control needs and tissue pharmacokinetics. The subgroup representing later periods was less precise perhaps because to fewer research and increased mechanistic heterogeneity. Given the number of available studies, no subgroup test can adequately distinguish these associated variables. The lack of randomized studies is no small restriction. In observational practice, CAZ–AVI may be preferentially administered to patients with prompt carbapenemase findings, infectious disease consultation, access to new antibiotics and earlier active treatment. Alternatively, physicians may reserve for patients who fail past treatments, confusing in the other direction. Polymyxin cohorts may vary in carbapenemases, be historical or get more combo medications. Even comprehensive matching cannot account for unmeasured source control, time, bacterial burden, and physician evaluation. Therefore, the pooled RR should be viewed as a summary of comparing real-world outcomes and not as an impartial causal risk ratio. However, the evidence has practical relevance. In settings where recommendations already favor the newer medication, randomized studies comparing a recommended contemporary β-lactam with a nephrotoxic salvage regimen may be challenging. Well-designed prospective registries and multicentre target-trial emulations may increase inference when they define time zero, prevent immortal-time bias, record accurate dose and susceptibility, and report uniform outcomes. Analyses stratified by mechanism are necessary. Future trials should include record development of resistance, recurrence, colonization pressure, quality-adjusted survival and cost, since purchase cost alone does not reflect renal damage, longer hospitalization or treatment failure. Clinical relevance Management of significant CRE infection in a patient should include immediate identification of the organism, carbapenemase characterisation where possible, susceptibility testing, and determination of the source of infection and timely source management. Current guidelines [3,4] typically favors a properly dosed CAZ–AVI regimen over polymyxin-based treatment when a KPC- or susceptible OXA-48-like-producing Enterobacterales strain is found. CAZ–AVI is not recommended as monotherapy for an MBL producer. CAZ–AVI with aztreonam is an option if supported by mechanism and susceptibility or an alternate MBL-active medication should be chosen based on local availability and recommendations. Polymyxins are indicated in the absence of a safer active drug, when resistance mechanisms limit the use of newer β-lactams, or as part of carefully planned salvage treatment. They need formulation-specific loading and maintenance techniques, renal monitoring, examination of concurrent nephrotoxins, and early reconsideration once microbiology data become available [5]. Indiscriminate use of CAZ–AVI is not justified by pooled results: care is required to retain activity, limit resistance selection and ensure treatment is restricted or modified when the carbapenemase phenotype predicts failure. Constraints and strengths Strengths include restriction to a contemporary 2020–2025 evidence window, explicit separation of qualitative and quantitative evidence, transparent study-level event counts, a conservative random-effects method, sensitivity analyses, mechanistic interpretation, and a numbered PRISMA flow that distinguishes the published base-search totals from the verified update set. The review incorporates current propensity-score and target-trial-emulation studies and differentiates CAZ–AVI monotherapy from CAZ–AVI plus aztreonam when the resistance biology requires that distinction. The limitations are considerable. First, database-specific export counts for the update stage were not retained, so the PRISMA diagram reports the published base-search totals and the verified number of newly eligible reports rather than a complete de novo source-by-source flow. Second, all included studies were observational. Third, mortality time points, infection sources, resistance mechanisms, dosing, combination therapy and source control varied. Fourth, unadjusted binary counts were pooled because adjusted estimates were too heterogeneous in measure and covariate structure to combine under one model. Fifth, clinical and microbiological endpoints were non-standardized, and renal-toxicity definitions varied. Sixth, publication bias cannot be excluded despite the nonsignificant Egger test. Finally, findings from settings with limited rapid diagnostics or restricted CAZ–AVI access may not transfer directly to regions with different resistance epidemiology, drug availability and stewardship infrastructure.

A systematic review of comparative studies published from 2020 through 2025 consistently associated CAZ–AVI-based therapy with lower mortality and better clinical and microbiological response than polymyxin-based regimens in patients with CRE infection. The pooled mortality estimate was RR 0.57, with modest heterogeneity and a consistent direction across sensitivity analyses. CAZ–AVI also showed a trend toward lower renal toxicity, although the principal safety estimate was imprecise. These findings support mechanism-directed use of CAZ–AVI, and CAZ–AVI plus aztreonam for suitable MBL-producing isolates, in preference to polymyxins when susceptibility and patient characteristics permit. Certainty remained very low; therefore, the magnitude of benefit should not be interpreted as a randomized causal effect. Future updates should archive database-specific exports and exclusion logs prospectively.

1. Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ. 2021;372:n71. doi:10.1136/bmj.n71. 2. Murray CJL, Ikuta KS, Sharara F, Swetschinski L, Aguilar GR, Gray A, et al. Global burden of bacterial antimicrobial resistance in 2019: a systematic analysis. Lancet. 2022;399(10325):629–655. doi:10.1016/S0140-6736(21)02724-0. 3. Tamma PD, Heil EL, Justo JA, Mathers AJ, Satlin MJ, Bonomo RA. Infectious Diseases Society of America 2024 guidance on the treatment of antimicrobial-resistant Gram-negative infections. Clin Infect Dis. 2024. doi:10.1093/cid/ciae403. 4. Paul M, Carrara E, Retamar P, Tängdén T, Bitterman R, Bonomo RA, et al. European Society of Clinical Microbiology and Infectious Diseases guidelines for the treatment of infections caused by multidrug-resistant Gram-negative bacilli. Clin Microbiol Infect. 2022;28(4):521–547. doi:10.1016/j.cmi.2021.11.025. 5. Nang SC, Azad MAK, Velkov T, Zhou QT, Li J. Rescuing the last-line polymyxins: achievements and challenges. Pharmacol Rev. 2021;73(2):679–728. doi:10.1124/pharmrev.120.000020. 6. Yang P, et al. Efficacy and safety of ceftazidime–avibactam versus polymyxins in carbapenem-resistant Enterobacterales infections: a systematic review and meta-analysis. BMJ Open. 2023;13:e070491. doi:10.1136/bmjopen-2022-070491. 7. Chen J, Hu Q, Zhou P, Deng S. Ceftazidime–avibactam versus polymyxins in treating patients with carbapenem-resistant Enterobacteriaceae infections: a systematic review and meta-analysis. Infection. 2024;52:19–28. doi:10.1007/s15010-023-02108-6. 8. Chen L, Han X, Li Y, Li M. Assessment of mortality-related risk factors and effective antimicrobial regimens for treatment of bloodstream infections caused by carbapenem-resistant Enterobacterales. Antimicrob Agents Chemother. 2021;65(9):e00698-21. doi:10.1128/AAC.00698-21. 9. Zhou C, Jin L, Wang Q, et al. Bloodstream infections caused by carbapenem-resistant Enterobacterales: risk factors for mortality, antimicrobial therapy and treatment outcomes from a prospective multicenter study. Infect Drug Resist. 2021;14:731–742. doi:10.2147/IDR.S294282. 10. Fang J, Li H, Zhang M, et al. Efficacy of ceftazidime–avibactam versus polymyxin B and risk factors affecting clinical outcomes in patients with carbapenem-resistant Klebsiella pneumoniae infections: a retrospective study. Front Pharmacol. 2021;12:780940. doi:10.3389/fphar.2021.780940. 11. Hakeam HA, Alsahli H, Albabtain L, Alassaf S, Al Duhailib Z, Althaqafi A. Effectiveness of ceftazidime–avibactam versus colistin in treating carbapenem-resistant Enterobacterales bacteremia. Int J Infect Dis. 2021;109:1–7. doi:10.1016/j.ijid.2021.05.079. 12. Falcone M, Daikos GL, Tiseo G, Bassoulis D, Giordano C, Galfo V, et al. Efficacy of ceftazidime–avibactam plus aztreonam in patients with bloodstream infections caused by metallo-β-lactamase-producing Enterobacterales. Clin Infect Dis. 2021;72(11):1871–1878. doi:10.1093/cid/ciaa586. 13. Meng H, Han L, Niu M, et al. Risk factors and clinical outcomes of carbapenem-resistant Klebsiella pneumoniae bloodstream infection in patients with hematological malignancies. Infect Drug Resist. 2022;15:4241–4251. doi:10.2147/IDR.S374904. 14. Satlin MJ, Chen L, Gomez-Simmonds A, Marino J, Weston G, Bhowmick T, et al. Impact of a rapid molecular test for Klebsiella pneumoniae carbapenemase and ceftazidime–avibactam use on outcomes after bacteremia caused by carbapenem-resistant Enterobacterales. Clin Infect Dis. 2022;75(12):2066–2075. doi:10.1093/cid/ciac354. 15. Zheng G, Cai J, Zhang L, et al. Ceftazidime/avibactam-based versus polymyxin B-based therapeutic regimens for the treatment of carbapenem-resistant Klebsiella pneumoniae infection in critically ill patients: a retrospective cohort study. Infect Dis Ther. 2022;11(5):1917–1934. doi:10.1007/s40121-022-00682-0. 16. Almangour TA, Ghonem L, Aljabri A, et al. Ceftazidime–avibactam versus colistin for the treatment of infections due to carbapenem-resistant Enterobacterales: a multicenter cohort study. Infect Drug Resist. 2022;15:211–221. doi:10.2147/IDR.S349004. 17. Prayag PS, et al. Ceftazidime–avibactam with or without aztreonam versus polymyxin-based combination therapy for carbapenem-resistant Enterobacterales infections. Indian J Crit Care Med. 2023;27(6):444–450. doi:10.5005/jp-journals-10071-24481. 18. Vijayakumar M, Selvam V, Renuka MK, Rajagopalan RE. The comparative efficacy of ceftazidime–avibactam with or without aztreonam versus polymyxins for carbapenem-resistant Enterobacterales infections. Indian J Crit Care Med. 2023;27(12):923–929. doi:10.5005/jp-journals-10071-24577. 19. Sree RA, Gupta A, Gupta N, et al. Ceftazidime–avibactam alone or in combination with aztreonam versus polymyxins in carbapenem-resistant Klebsiella pneumoniae infections: the CAPRI study. Infection. 2024;52(2):429–437. doi:10.1007/s15010-023-02094-9. 20. Dumlu R, et al. Ceftazidime–avibactam versus polymyxin-based combination therapies: a study on 30-day mortality in carbapenem-resistant Enterobacterales bloodstream infections in an OXA-48-endemic region. Antibiotics (Basel). 2024;13(10):990. doi:10.3390/antibiotics13100990. 21. Zhuang HH, Qu Q, Long WM, et al. Ceftazidime/avibactam versus polymyxin B in carbapenem-resistant Klebsiella pneumoniae infections: a multicenter propensity-score-matched study. Infection. 2025;53:95–106. doi:10.1007/s15010-024-02324-8. 22. Hu N, Xiao F, Chen Y, et al. Efficacy, safety and cost-effectiveness of ceftazidime–avibactam versus polymyxin B for carbapenem-resistant Klebsiella pneumoniae infection: a target trial emulation. Infect Dis Ther. 2025;14:1419–1437. doi:10.1007/s40121-025-01164-9. 23. Matthews AA, Danaei G, Islam N, Kurth T. Target trial emulation: applying principles of randomised trials to observational studies. BMJ. 2022;378:e071108. doi:10.1136/bmj-2022-071108. 24. Isler B, Aslan AT, Akova M, Harris PNA, Paterson DL. Treatment strategies for OXA-48-like- and NDM-producing Klebsiella pneumoniae infections. Expert Rev Anti Infect Ther. 2022;20(11):1389–1400. doi:10.1080/14787210.2022.2128764.