Comparative efficacy and safety of colistin sulfate versus polymyxin B in critically ill patients with carbapenem-resistant gram-negative infections: a pilot inverse probability of treatment weighting–based retrospective cohort study Authors: Yi Xu, Shuting Zhang, Liying Wang, Yang Deng, Nan Hu Published: 2026 ABSTRACT AbstractObjectiveThe efficacy and safety of colistin sulfate (CS) and polymyxin B sulfate (PBS) were evaluated in critically ill patients with carbapenem-resistant Gram-negative bacteria (CR-GNB) infections, and predictors affecting clinical outcomes were identified.MethodsCR-GNB-infected ICU patients treated between February 2023 and December 2024 were retrospectively analyzed. Patients were grouped according to treatment, and inverse probability of treatment weighting (IPTW) was applied to balance confounding factors and compare clinical efficacy, microbiological clearance rates, and safety between groups. Prespecified subgroup analyses by pathogen type were conducted, with doubly robust estimation applied in the CRAB subgroup. Differences in trough concentrations (Cmin) were evaluated among clinical efficacy groups, acute kidney injury (AKI) groups, and mortality groups. Univariate and multivariate analyses were performed to identify prognostic factors for efficacy and mortality.ResultsThe CS group and the PBS group have 85 and 78 patients, respectively. Following IPTW adjustment for indication-based confounders, the two groups were well-balanced in baseline demographic characteristics, and showed no significant differences in clinical and microbiological efficacy, or in 14-day and 28-day mortality rates. However, the PBS group exhibited a significantly higher proportion of severe AKI (stages II and III) compared to the CS group (OR = 6.865, 95% CI: 1.463–32.221, P = 0.015).In prespecified subgroup analyses, the risks of both overall AKI (OR = 4.84, 95% CI 1.45–16.15, P = 0.011) and severe AKI (OR = 7.63, 95% CI 1.57–36.97, P = 0.012) were significantly elevated with PBS in patients with CRAB infection, with a significant treatment-by-pathogen interaction. No significant differences in Cmin were observed across the clinical efficacy groups, AKI groups, and mortality groups. Multivariable analysis showed that in the CS group, the use of vasoactive drugs was an independent risk factor for poor clinical efficacy and 28-day mortality; In the PBS group, CRAB infection and age were independent risk factors for 28-day mortality, whereas vasoactive drug use was an independent risk factor for poor clinical efficacy, and loading dose administration served as an independent protective factor for clinical efficacy.ConclusionsCS and PBS showed comparable clinical and microbiological efficacy. PBS was associated with a higher risk of severe AKI in the overall cohort. Notably, within the CRAB subgroup, PBS showed significantly elevated risks for both overall and severe AKI.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1186/s12879-026-13821-6.Keywords: Polymyxin B, Colistin sulfate, Carbapenem-resistant gram-negative bacteria, Critically ill patients, Exposure-response relationship Introduction Antimicrobial resistance (AMR) is a global public health emergency [1], and inadequate hospital infection control further accelerates the evolution of multidrug-resistant pathogens. Drug-resistant infections are projected to cause up to 10 million deaths annually by 2050 if effective countermeasures are not implemented [2]. Critically ill patients are extremely vulnerable to multidrug-resistant pathogens, as mechanical ventilators and central venous catheters breach anatomical barriers and enable pathogen invasion [3]. Within intensive care units (ICUs), carbapenem-resistant Gram-negative bacteria (CR-GNB)—particularly Enterobacteriaceae (CRE), Acinetobacter baumannii (CRAB), and Pseudomonas aeruginosa (CRPA)—drive catastrophic outcomes [4]. Mortality from CRAB bloodstream infections can exceed 40%, while CRPA pneumonia has been linked to prolonged ICU stays [5]. Treatment costs for CR-GNB infections are estimated to be 3–5 times higher than for drug-susceptible cases [6, 7]. CR-GNB evade all penicillins, cephalosporins, and carbapenems. Polymyxins are cationic lipopeptides that disrupt Gram-negative membranes via detergent-like insertion into lipopolysaccharides [8]. They were abandoned in the 1970s for nephrotoxicity (30–60% incidence) [9] and neurotoxicity but are now last-resort agents [10, 11]. Their concentration-dependent killing remains effective against CRAB, CRPA, and carbapenem-resistant Klebsiella pneumoniae (CRKP), yet their narrow therapeutic windows demand precise dosing to balance efficacy and toxicity. Only polymyxin B and polymyxin E (also known as colistin) are clinically utilized. Polymyxins A, C, and D are structurally similar but not used in clinical practice due to higher toxicity or lower efficacy. Geographic disparities significantly influence polymyxin prescribing. Polymyxin B sulfate (PBS) is the preferred intravenous polymyxin in North America and Europe due to its predictable pharmacokinetics and established exposure-response relationships. China historically relied on colistimethate sodium (CMS), a prodrug requiring conversion to active colistin, driven by lower cost and formulary accessibility but has been shifting toward colistin sulfate (CS), a novel formulation approved by NMPA in 2019. The international consensus guidelines for the Optimal Use of the Polymyxins [10] suggest that an average steady-state plasma concentration (Css, avg) of polymyxin E higher than 2 mg/L increases the incidence and severity of acute kidney injury (AKI), whereas an average steady-state plasma concentration (Css, avg) of polymyxin B between 2 and 4 mg/L is effective without easily increasing the incidence and severity of AKI. Recently, some studies have found that the trough concentrations (Cmin) of PBS is significantly associated with clinical efficacy and the incidence of AKI [12]. However, studies on the exposure-response relationship of the two drugs are still lacking. Unlike CMS, which requires hydrolysis to active colistin, CS is administered as the bioactive compound, eliminating metabolic conversion delays and enabling direct interpretation of serum concentrations. However, little is known about the exposure-response relationship of CS. Zhang et al. [13] provided the sole clinical evidence linking the Cmin of CS (≥ 0.53 mg/L) to efficacy but identified no definitive toxicity cutoff. Furthermore, some real-world comparative studies on polymyxin therapies fail to adequately adjust for confounding by indication. Given the inherent nature of observational research, baseline imbalances in patient characteristics, disease severity, and pathogen distribution can significantly bias the comparative outcomes in these studies. In addition, almost all available safety evaluations were performed in unstratified overall cohorts, lacking pathogen-specific safety analyses for critical pathogens such as CRAB. To address these crucial evidence gaps and optimize polymyxin stewardship in ICUs, this retrospective real-world cohort study employed inverse probability of treatment weighting (IPTW) to rigorously control for confounding by indication. The specific study objectives were as follows: (1) compare the clinical efficacy and safety of CS and PBS among critically ill patients infected with CR-GNB; (2) analyze the differences in Cmin among clinical efficacy groups, acute kidney injury groups, and mortality groups; (3) identify independent predictors of clinical efficacy and mortality; and (4) further characterize the pathogen-specific safety profiles of the two agents through pre-specified subgroup analyses, with a particular focus on the incidence of AKI and severe AKI (stages II and III) in patients with CRAB infection. Methods Patients This study screened all patients who had CR-GNB infection during their ICU stay in the Third Affiliated Hospital of Soochow University from February 2023 to December 2024. Figure 1 shows the workflow of data collection. Patients were included if they were ≥ 18 years old, received CS or PBS-based combination therapy for ≥ 72 h, and had microbiological confirmation of one CR-GNB strain. Exclusion criteria included resistance to either agent, sequential administration of both, and pregnancy. The basic clinical data included age, sex, Acute Physiology and Chronic Health Evaluation (APACHE II) score, complications, infection profile, and medication regimen. Supportive interventions, including mechanical ventilation, continuous renal replacement therapy (CRRT), and extracorporeal membrane oxygenation (ECMO), were systematically recorded for all patients. The study protocol was approved by the Medical Ethics Committee of the First People ‘s Hospital of Changzhou City, with approval number (2026) Sect. 77. The requirement for informed consent was waived due to the retrospective nature of this study. Patient selection CRKP, CRAB and CRPA isolates were identified via the VITEK-II-COMPACT (BioMérieux, France) and MALDI-TOF MS systems (BioMérieux, France). Antimicrobial susceptibility was determined by broth microdilution method, and the results were interpreted according to the Clinical and Laboratory Standards Institute 2023 (CLSI 2023) and European Committee for Antimicrobial Susceptibility Testing (EUCAST) criteria. Isolates with an MIC of less than 2 mg/L were defined as susceptible to polymyxins. Treatment Colistin sulfate for injection (500,000 IU/vial; trade name: Fengweiling; Shanghai New Asiatic Pharmaceuticals Co., Ltd.; approval number H31020822) or polymyxin B sulfate for injection (500,000 IU/vial; SPH No.1 Biochemical & Pharmaceutical Co., Ltd.; approval number H31022631) was administered intravenously. Clinical dosing regimens for both injectable colistin sulfate (CS) and polymyxin B (PB) included 0.5 million IU, 0.75 million IU, and 1.0 million IU, administered every 12 h (q12h).This was a retrospective study, and the daily dosage and administration frequency were determined by attending clinicians according to the patient’s actual condition or after consultation with clinical pharmacists, with all regimens consistent with the recommendations of the Chinese Multidisciplinary Expert Consensus on the Rational Clinical Application of Polymyxins [14].Some patients also received adjunctive nebulized therapy with sterile CS inhalation solution or PBS inhalation concentrate, administered via a vibrating mesh nebulizer. The collected treatment data included the route of polymyxin administration, dosage (loading and maintenance), and combined use of antibiotics (e.g., carbapenems, etc.). Outcomes Efficacy evaluation indicators comprised body temperature, white blood cell (WBC), neutrophil percentage (NE%), C-reactive protein (CRP), procalcitonin (PCT) before and after treatment (3 days after drug withdrawal, if not forward), and Cmin after 3 days of CS or PBS administration. The primary outcomes were clinical efficacy and microbial clearance rate. Clinical efficacy was classified as: A.Cure: Complete resolution of infection symptoms with normalization of non-microbiological indicators (imaging, laboratory tests). Cure: Complete resolution of infection symptoms with normalization of non-microbiological indicators (imaging, laboratory tests). B.Improvement: Significant symptom resolution with partial normalization of objective indicators. Improvement: Significant symptom resolution with partial normalization of objective indicators. C.Non-response: Absence of clinical improvement or symptom deterioration, with stagnant/worsening indicators. Non-response: Absence of clinical improvement or symptom deterioration, with stagnant/worsening indicators. D.Death: Patient mortality during treatment. Death: Patient mortality during treatment. Success and failure rates represented the proportions of cases in the first two categories (A + B) and last two categories (C + D), respectively, relative to total cases. According to the “Guidelines for Clinical Trials of Antibiotics” [15], the microbial efficacy was categorized as: A.Clearance: Eradication of CR-GNB from original infection sites (sputum, blood, urine, cerebrospinal fluid, drainage fluid, or lavage fluid) in two consecutive cultures after 7-day treatment. Clearance: Eradication of CR-GNB from original infection sites (sputum, blood, urine, cerebrospinal fluid, drainage fluid, or lavage fluid) in two consecutive cultures after 7-day treatment. B.Assumed clearance: Presumed eradication in clinically cured patients where culturable materials become inaccessible due to symptom resolution. Assumed clearance: Presumed eradication in clinically cured patients where culturable materials become inaccessible due to symptom resolution. C.Assumed non-clearance: Presumed persistence in clinically ineffective patients when culture is unobtainable. Assumed non-clearance: Presumed persistence in clinically ineffective patients when culture is unobtainable. D.Non-clearance: Persistent growth of original pathogens in post-treatment cultures. Non-clearance: Persistent growth of original pathogens in post-treatment cultures. Microbial clearance rate represented the proportion of cases in the first three categories (A + B) relative to total cases. Secondary outcomes included the following: The magnitude of change in inflammatory indicators; The magnitude of change in inflammatory indicators; The proportion of patients achieving normalization for each indicator, defined as body temperature < 37.3 °C, WBC < 10 × 109/L, NE% < 70%, CRP < 10 mg/L, PCT < 0.05 ng/mL; The proportion of patients achieving normalization for each indicator, defined as body temperature < 37.3 °C, WBC < 10 × 109/L, NE% < 70%, CRP < 10 mg/L, PCT < 0.05 ng/mL; Mortality rates at 14 days and 28 days after treatment initiation, along with overall in-hospital mortality rate. All mortality outcomes were calculated from the first day of antimicrobial administration through the end of hospitalization. Mortality rates at 14 days and 28 days after treatment initiation, along with overall in-hospital mortality rate. All mortality outcomes were calculated from the first day of antimicrobial administration through the end of hospitalization. Hospitalization (days): The length of hospital stay for patients treated with medication. Cmin: The steady-state trough serum concentration measured after 3 days of continuous polymyxin administration. Adverse reactions Adverse events, defined as obvious harmful or unpleasant reactions due to the use of CS and PBS, included AKI, liver function injury, thrombocytopenia, etc. According to the Kidney Disease Improving Global Outcomes (KDIGO) Criteria [16], AKI was confirmed if any one of the following occurred: An increase in serum creatinine (Scr) by at least 0.3 mg/dL (approximately 26.5 µmol/L) within 48 h. An increase in serum creatinine (Scr) by at least 0.3 mg/dL (approximately 26.5 µmol/L) within 48 h. An increase in Scr to at least 1.5 times the baseline value, which is known or presumed to have occurred within the previous 7 days. An increase in Scr to at least 1.5 times the baseline value, which is known or presumed to have occurred within the previous 7 days. A reduction in urine output to less than 0.5 mL/kg/h for at least 6 h. A reduction in urine output to less than 0.5 mL/kg/h for at least 6 h. AKI was further classified according to global prognosis improvement standards for kidney disease [17]. Stage I AKI: Scr increased to 1.5–1.9 times the baseline level, or urine output < 0.5 mL/kg/h for 6–12 consecutive hours. Stage I AKI: Scr increased to 1.5–1.9 times the baseline level, or urine output < 0.5 mL/kg/h for 6–12 consecutive hours. Stage II AKI: Scr increased to 2.0–2.9 times the baseline level, or urine output < 0.5 mL/kg/h for 12–23 consecutive hours. Stage II AKI: Scr increased to 2.0–2.9 times the baseline level, or urine output < 0.5 mL/kg/h for 12–23 consecutive hours. Stage III AKI: Scr increased to ≥ 3.0 times the baseline level, or urine output < 0.5 mL/kg/h for ≥ 24 h, or anuria for ≥ 12 h. Stage III AKI: Scr increased to ≥ 3.0 times the baseline level, or urine output < 0.5 mL/kg/h for ≥ 24 h, or anuria for ≥ 12 h. According to the recommendations of the International Expert Panel on the Liver [18], drug-induced hepatic injury was diagnosed regardless of clinical symptoms when any of the following laboratory cutoffs were satisfied: Elevated alkaline phosphatase (ALP): transaminases are increased compared with those before treatment and exceed twice the normal range (0 ~ 100 U/L); Elevated alkaline phosphatase (ALP): transaminases are increased compared with those before treatment and exceed twice the normal range (0 ~ 100 U/L); Total bilirubin (TBIL) is increased compared with that before treatment and exceeds twice the normal range (0 ~ 21 µmol/L), and aspartate aminotransferase (AST) or alanine aminotransferase (ALT) is increased more than 3 times the normal range (0 ~ 40 U/L). Total bilirubin (TBIL) is increased compared with that before treatment and exceeds twice the normal range (0 ~ 21 µmol/L), and aspartate aminotransferase (AST) or alanine aminotransferase (ALT) is increased more than 3 times the normal range (0 ~ 40 U/L). ALT or AST increases more than 5 times the normal range (0 ~ 40 U/L). ALT or AST increases more than 5 times the normal range (0 ~ 40 U/L). Statistical methods Quantitative variables following a normal distribution were presented as the mean ± standard deviation (x̄ ± SD), whereas those that did not conform to a normal distribution were expressed as the median with interquartile range (M [P25, P75]). Continuous variables were compared between groups using either Student’s t-test or a non-parametric test, as appropriate. Categorical variables were compared using the chi-square test or Fisher’s exact test. IPTW based on the propensity score was utilized to balance baseline confounding factors between the PBS and CS groups. The propensity score was calculated via a logistic regression model incorporating seven core covariates: APACHE II score, Kidney diseases, combined intravenous with inhaled polymyxin therapy, diabetes, CRPA, vasoactive drug use, and loading dose. To reduce the influence of extreme weights, IPTW values were truncated at the 1st and 99th percentiles. Covariate balance before and after weighting was assessed using absolute standardized mean difference (SMD), with an SMD value < 0.1 was defined as adequate balance. A Love plot was generated to visually present the balance of all baseline variables, and the effective sample size (ESS) was calculated to evaluate statistical power after weighting. Cox proportional hazards regression was performed to analyze 14-day and 28-day survival outcomes, and hazard ratio (HR) with 95% confidence interval (CI) was calculated. Binary logistic regression was used for binary clinical endpoints, while ordinal logistic regression was adopted for graded categorical outcomes, with proportional odds ratio (POR) and 95% CI reported. Weighted logistic regression analysis was employed to identify factors associated with treatment efficacy, AKI incidence, and mortality. Variables satisfying P < 0.1 in the weighted univariate analysis were entered into the weighted multivariate regression model. Prespecified subgroup analyses were performed by pathogen type (CRAB vs. CRKP). To avoid unstable estimates, a subgroup was formally analyzed only when the weighted effective sample size was ≥ 30; otherwise, only descriptive weighted proportions were reported. For the CRAB subgroup, the original IPTW weights were retained. When weighted standardized mean differences exceeded 0.1 (Age, Sex, APACHE II score, and inhalation route), doubly robust logistic models were fitted by additionally adjusting for these covariates. Treatment-by-pathogen interactions were tested in the overall weighted cohort. All subgroup analyses were exploratory without multiplicity adjustment. All statistical analyses were performed using R 4.4.3 and SPSS Statistics 27; GraphPad Prism 10.1.2 was used for figure generation. Results General clinical indicators From an initial pool of 365 patients, 163 met the inclusion and exclusion criteria; notably, none of the patients with a treatment duration of less than 3 days developed early-onset AKI, and this study eventually included 85 patients in Group CS and 78 patients in Group PBS (Fig. 1). After IPTW, the weighted sample sizes were 159 in Group CS and 168 in Group PBS, with ESS of 80 and 65, respectively. Table 1 demonstrates that all baseline characteristics—including sex, age, comorbidities, and APACHE II score—were well balanced between the two groups after IPTW (all standardized mean differences < 0.1) (Fig. 2). During polymyxin treatment, nearly all patients required invasive mechanical ventilation (CS: 98.1%, PBS: 98.8%), approximately half received continuous renal replacement therapy (CRRT; CS: 45.9%, PBS: 49.4%), and a small proportion required extracorporeal membrane oxygenation (ECMO; CS: 1.9%, PBS: 1.2%). Table 1Basic patient dataCharacteristics†Before IPTWAfter IPTWGroup CS (N = 85)Group PBS (N = 78)SMDGroup CS (Weighted N = 159)Group PBS (Weighted N = 168)SMDBasic information Age (years)71.0 [59.0, 80.0]72.0 [63.3, 78.8]0.08771.1 [60.8, 80.0]72.0 [63.4, 77.7]0.099 Sex (M/F)26/5924/540.004108/51115/530.011 APACHE II score25.38 ± 5.9123.64 ± 5.78 0.297 24.85 ± 5.6924.67 ± 5.590.032Comorbidities Respiratory diseases62 (72.9%)56 (71.8%)0.026116 (73.0%)122 (72.6%)0.011 Cardiovascular diseases61 (71.8%)58 (74.4%)0.059114 (71.7%)126 (75.0%)0.081 Diabetes28 (32.9%)18 (23.1%) 0.221 48 (30.2%)52 (31.0%)0.018 Kidney diseases40 (47.1%)47 (60.3%) 0.267 82 (51.6%)85 (50.6%)0.022Supportive interventions Mechanical ventilation83 (97.7%)77 (98.7%)0.08156 (98.1%)166 (98.8%)0.040 CRRT40 (47.1%)42 (53.9%) 0.136 73 (45.9%)83 (49.4%)0.069 ECMO2 (2.4%)1 (1.3%)0.0803 (1.9%)2 (1.2%)0.045Infection site Pulmonary85 (100.0%)78 (100.0%)0.000159 (100.0%)168 (100.0%)0.000Pathogen Type CRKP15 (17.7%)15 (19.2%)0.04129 (18.2%)36 (21.4%)0.089 CRPA5 (5.9%)3 (3.9%)0.0958 (5.0%)9 (5.4%)0.009 CRAB65 (76.5%)60 (76.9%)0.011122 (76.8%)123 (73.2%)0.077Polymyxin administration IV85 (100.0%)78 (100.0%)0.000159 (100.0%)168 (100.0%)0.000 Nebulized inhalation + IV69 (81.2%)70 (89.7%) 0.245 135 (84.9%)143 (85.1%)0.013 Loading dose41 (48.2%)43 (55.1%) 0.138 80 (50.3%)82 (48.8%)0.028 Course of treatment (days)8.0 [7.0, 12.0]8.0 [7.0, 12.0]0.0088.0 [6.0, 12.0]8.0 [7.0, 12.0]0.002 Combination with vasoactive drugs66 (77.7%)65 (83.3%) 0.144 125 (78.6%)130 (77.4%)0.029Combined use of antibiotics Type Carbapenems59 (69.4%)49 (62.8%) 0.140 107 (67.3%)115 (68.5%)0.027 Ceftazidime/avibactam3 (3.5%)3 (3.9%)0.0176 (3.8%)6 (3.6%)0.009 Number of co-administered antibiotics 03 (3.5%)2 (2.6%)0.0565 (3.1%)4 (2.4%)0.048 171 (83.5%)63 (80.8%)0.072131 (82.4%)134 (79.8%)0.061 211 (12.9%)13 (16.7%) 0.105 23 (14.5%)30 (17.8%)0.086† Expressed as case count (ratio), median [P25, P75], or mean ± standard deviation. N, actual patients; Weighted N, IPTW-weighted sample size (rounded to integer). SMD, absolute standardized mean difference; IPTW, Inverse Probability of Treatment Weighting; APACHE II score, acute physiology and chronic health evaluation; CRKP, carbapenem-resistant Klebsiella pneumoniae; CRPA, carbapenem-resistant Pseudomonas aeruginosa; CRAB, carbapenem-resistant Acinetobacter baumannii; CRRT, continuous renal replacement therapy; ECMO, extracorporeal membrane oxygenation; IV, intravenous drip. SMD values are typeset in boldface if they are more than 0.1 Basic patient data † Expressed as case count (ratio), median [P25, P75], or mean ± standard deviation. N, actual patients; Weighted N, IPTW-weighted sample size (rounded to integer). SMD, absolute standardized mean difference; IPTW, Inverse Probability of Treatment Weighting; APACHE II score, acute physiology and chronic health evaluation; CRKP, carbapenem-resistant Klebsiella pneumoniae; CRPA, carbapenem-resistant Pseudomonas aeruginosa; CRAB, carbapenem-resistant Acinetobacter baumannii; CRRT, continuous renal replacement therapy; ECMO, extracorporeal membrane oxygenation; IV, intravenous drip. SMD values are typeset in boldface if they are more than 0.1 Love plot After IPTW adjustment, the two groups did not differ significantly in infection type and pathogen distribution. All patients had pulmonary infections. CRAB infection was predominant (CS, 76.8%; PBS, 73.2%), followed by CRKP (CS, 18.2%; PBS, 21.4%) and CRPA (CS, 5.0%; PBS, 5.4%). All patients in both groups were infected with only a single bacterial strain, and the MICs of the two antibiotics against the pathogenic strains were all ≤ 1 µg/mL. There were no significant differences in the distribution of antibiotic treatment regimens between the study cohorts. Most patients received combined antibiotic therapy. Polymyxin was most frequently combined with meropenem, imipenem and other carbapenems (CS, 67.3%; PBS, 68.5%), and ceftazidime/avibactam (CS, 3.8%; PBS, 3.6%). Few patients received polymyxin monotherapy (CS, 3.1%; PBS, 2.4%). Most patients were given one additional antibiotic agent concurrently (CS, 82.4%; PBS, 79.8%), and some received two additional antibiotics (CS, 14.5%; PBS, 17.8%). The duration of polymyxin treatment was comparable between the two groups, with a median of 8 days in both groups. Comparison of the efficacy of the two groups of patients Following IPTW adjustment, no significant difference in clinical efficacy was detected between the CS and PBS groups (OR = 0.580, 95% CI: 0.297–1.132, P = 0.109; Table 2). Likewise, microbiological efficacy showed no statistically significant intergroup difference (OR = 0.709, 95% CI: 0.362–1.386, P = 0.312). Table 2Therapeutic effects and adverse reactions of polymyxinsOutcomeBefore IPTWAfter IPTWOR95%CI P OR95%CI P Primary outcomes Clinical efficacy0.5390.287–1.0000.0510.5800.297–1.1320.109 Microbial efficacy†0.8190.438–1.5260.530.7090.362–1.3860.312Secondary outcomes Body temperature returned to normal (%)0.7210.380–1.3560.3110.6680.339–1.3170.242 WBC returned to normal (%)1.1230.555–2.2740.7471.6800.787–3.5860.178 Reduction of WBC (109/L)-0.972-3.434–1.4890.437-0.51-3.096–2.0770.698 NE% returned to normal (%)1.3860.541–3.6350.4961.6140.589–4.4250.350 Reduction of NE%0.418-2.867–3.7040.8021.911-1.469–5.2910.266 CRP returned to normal (%)0.640.128–2.6990.5510.5460.121–2.4670.429 Reduction of CRP (mg/L)-11.273-32.632–10.0860.299-3.302-25.425–18.8220.769 PCT returned to normal (%)1.2120.528–2.7970.6491.1540.476–2.7970.750 Reduction of PCT (ng/mL)-1.343-5.043–2.3580.475-0.899-4.109–2.3110.581 14-day mortality rate (%)1.2940.606–2.7820.5051.1520.518–2.5610.728 28-day mortality rate (%)1.0130.522–1.9610.9690.8730.433–1.7600.703 Total mortality rate (%)0.9110.472–1.7510.780.7790.389–1.5590.478Hospitalization (days)-7.740-18.576–3.0970.160-7.51-19.183–4.1630.206Adverse reaction Hepatotoxicity0.6020.153–2.0790.4340.5760.155–2.1410.408 AKI2.9121.267–7.166 0.015 2.2620.918–5.5730.076 I stage0.7630.217–2.4960.6570.6930.201–2.3920.560 II stage10.9571.988–204.535 0.025 6.9930.836–58.5020.072 III stage7.0001.159–133.844 0.075 5.6490.64–49.8730.118 II + III stage9.8812.66–64.158 0.003 6.8651.463–32.221 0.015 Thrombocytopenia0.5220.209–1.2270.1450.5030.194–1.3080.157†The reported “OR” for microbial efficacy represents the proportional odds ratio (POR) from ordinal logistic regression. IPTW, Inverse Probability of Treatment Weighting; WBC, white blood cell count; NE%, neutrophilic granulocyte percentage; CRP, C-reactive protein; PCT, procalcitonin; AKI, acute kidney injury. P values are typeset in boldface if they are less than 0.05 Therapeutic effects and adverse reactions of polymyxins †The reported “OR” for microbial efficacy represents the proportional odds ratio (POR) from ordinal logistic regression. IPTW, Inverse Probability of Treatment Weighting; WBC, white blood cell count; NE%, neutrophilic granulocyte percentage; CRP, C-reactive protein; PCT, procalcitonin; AKI, acute kidney injury. P values are typeset in boldface if they are less than 0.05 No significant between-group differences were observed in biomarker-related secondary outcomes, including rates of normalization for body temperature, WBC, NE%, CRP, and PCT, as well as the magnitude of reductions in these inflammatory markers (all P > 0.05 after IPTW adjustment). There was no significant difference in biomarker-based secondary outcomes and 14-day mortality rate (OR = 1.152, 95% CI: 0.518–2.561, P = 0.728), 28-day mortality rate (OR = 0.873, 95% CI: 0.443–1.760, P = 0.703), and total mortality rate (OR = 0.779, 95% CI: 0.389–1.559, P = 0.478). Comparison of adverse reactions between the two groups of patients None of the patients required discontinuation of polymyxin therapy due to adverse events, including hepatotoxicity, AKI, and thrombocytopenia (Table 2). Hepatotoxicity was defined as elevated levels of TBIL, AST, or ALT, while AKI was diagnosed and staged according to the KDIGO 2012 criteria, based on increased serum creatinine. After IPTW adjustment, there were no significant between-group differences in the incidences of hepatotoxicity (OR = 0.576, 95% CI: 0.155–2.141, P = 0.408), thrombocytopenia (OR = 0.503, 95% CI: 0.194–1.308, P = 0.157), or overall AKI (OR = 2.262, 95% CI: 0.918–5.573, P = 0.076). However, the combined incidence of stage II and III AKI was significantly higher in the PBS group than in the CS group (OR = 6.865, 95% CI: 1.463–32.221, P = 0.015). Subgroup analysis of clinical efficacy, AKI, and 28-day mortality stratified by pathogen In prespecified exploratory subgroup analyses by pathogen type, the CRKP subgroup was not formally analysed due to insufficient weighted effective sample size (ESS = 24). In the CRAB subgroup, after doubly robust adjustment for imbalanced covariates (Age, Sex, APACHE II, and inhalation), the odds ratios for clinical efficacy, 28-day mortality, and AKI were 0.64 (95% CI: 0.29–1.41, P = 0.271), 1.35 (95% CI0.58–3.14, P = 0.479), and 4.84 (95%CI:1.45–16.15, P = 0.011), respectively (Fig. 3). Treatment-by-pathogen interactions were not significant for clinical efficacy (P = 0.531) or mortality (P = 0.092), but a significant interaction was detected for AKI (P = 0.020). All subgroup analyses were exploratory without multiplicity adjustment; the AKI finding requires independent confirmation (Fig. 3). IPTW-weighted comparison of primary outcomes between PBS and CS: overall cohort vs. CRAB subgroup analysis. IPTW, Inverse Probability of Treatment Weighting; APACHE II score, acute physiology and chronic health evaluation; CRKP, carbapenem-resistant Klebsiella pneumoniae; CRAB, carbapenem-resistant Acinetobacter baumannii; AKI, Acute kidney injury Relationship between Cmin of and clinical efficacy, AKI, and mortality Cmin of polymyxin in subgroups stratified by AKI severity, clinical efficacy, and 28-day mortality are presented in Table 3. In the CS group, Cmin remained low and relatively stable across all AKI severity categories, with no significant linear trend observed (P for trend = 0.759). In the PBS group, while Cmin values were generally higher than those in the CS group, numerically higher Cmin were seen in patients with stage II or III AKI compared to those without AKI or with stage I AKI; however, this trend did not reach statistical significance (P for trend = 0.167). When stratified by clinical efficacy and 28-day mortality, there were no significant differences in Cmin between effective and ineffective patients, or between survivors and non-survivors, in either group (all P > 0.05). Table 3The relationship between Cmin and clinical efficacy, AKI, and 28-day mortalityVariableCategoryGroup CSGroup PBS N Concentration (mg/L) N Concentration (mg/L)AKI severity No AKI 601.34 (0.87–1.73)373.2 (2.71–3.56) Stage I 71.7 (1.54–1.73)22.74 (2.74–2.94) Severe (II + III) 21.09 (1.01–1.09)†83.49 (3.03–3.67) Stage II11.09 (n = 1)*53.33 (3.03–4.02) Stage III11.01 (n = 1)*33.49 (3.22–3.67)P for trend§0.7590.167Clinical efficacy Ineffective 401.3 (0.95–1.73)231.34 (0.87–1.66) Effective 291.34 (0.87–1.66)243.22 (2.71–3.42)P value¶0.8200.41628-day mortality Non-survivors 221.61 (0.95–1.73)123.12 (2.79–3.42) Survivors 471.26 (0.87–1.66)353.12 (2.79–3.42)P value¶0.5370.113*Individual values because n < 2 (no standard deviation calculated). †Severe AKI group in CS has only 2 patients; concentration shown as median (Q1–Q3) (1.09 (1.01–1.09)). Severe (II + III)” represents the aggregated total of Stage II and Stage III AKI cases, presented for descriptive purposes only. §P for trend was calculated using IPTW-weighted Spearman rank correlation analysis to evaluate trends in trough concentrations with increasing AKI severity. ¶IPTW-weighted Wilcoxon rank-sum test comparing the two categories (e.g., Ineffective vs. Effective; Non-survivors vs. Survivors) The relationship between Cmin and clinical efficacy, AKI, and 28-day mortality *Individual values because n < 2 (no standard deviation calculated). †Severe AKI group in CS has only 2 patients; concentration shown as median (Q1–Q3) (1.09 (1.01–1.09)). Severe (II + III)” represents the aggregated total of Stage II and Stage III AKI cases, presented for descriptive purposes only. §P for trend was calculated using IPTW-weighted Spearman rank correlation analysis to evaluate trends in trough concentrations with increasing AKI severity. ¶IPTW-weighted Wilcoxon rank-sum test comparing the two categories (e.g., Ineffective vs. Effective; Non-survivors vs. Survivors) Univariate and multivariate analyses In the univariate analysis of clinical efficacy, factors that met P < 0.1 included Sex and Combination with vasoactive drugs for Group CS and Age, Combination with vasoactive drugs and Loading dose for Group PBS (Table 4). When these factors were further evaluated by multivariate analysis, it turned out that for Group CS, the clinical efficacy was significantly poorer if the patient used vasoactive drugs(OR = 0.210, 95% CI: 0.062–0.711, P = 0.013), and for Group PBS, clinical efficacy was significantly better in patients who received a loading dose (OR = 4.387, 95% CI: 1. 280–15.037, P = 0.019), whereas it was significantly poorer in those who used vasoactive drugs(OR = 0.120, 95% CI: 0.023–0.619, P = 0.012). Table 4Univariate and multivariate analysis of clinical efficacy and 28-day mortalityFactorUnivariate analysisMultivariate analysisOR95%CI P OR95%CI P clinical efficacy CS group Combination with vasoactive drugs0.2120.061–0.7370.0150.2100.062–0.711 0.013 Sex2.7170.939–7.8590.065 PBS group Age0.9610.919–1.0040.077Combination with vasoactive drugs0.1870.040–0.8700.0330.1200.023–0.619 0.012 Loading dose2.6800.944–7.6040.0644.3871.280–15.037 0.019 28-day mortality CS group Combination with vasoactive drugs4.0701.028–16.1080.0464.0701.028–16.108 0.046 PBS group Age1.0891.013–1.1700.0211.0881.020–1.160 0.011 Kidney diseases as comorbidities2.5450.835–7.7570.099CRAB3.1560.848–11.7480.0865.7881.249–26.821 0.025 Combination with vasoactive drugs7.93850.903–69.7880.062CRAB, carbapenem-resistant Acinetobacter baumannii. P values are typeset in boldface if they are less than 0.05 Univariate and multivariate analysis of clinical efficacy and 28-day mortality CRAB, carbapenem-resistant Acinetobacter baumannii. P values are typeset in boldface if they are less than 0.05 In the CS group, multivariable logistic regression identified the concurrent use of vasoactive drugs as an independent risk factor for 28-day mortality (OR = 4.070, 95% CI: 1.028–16.108, P = 0.046). Conversely, in the PBS group, univariate analysis revealed that age, comorbid kidney disease, vasoactive drug use, and CRAB infection were significantly associated with 28-day mortality (all P < 0.1).Multivariate regression analysis further identified age (OR = 1.088, 95% CI: 1.020–1.160, P = 0.011) and CRAB infection (OR = 5.788, 95% CI: 1.249–26.821, P = 0.025) as independent risk factors for 28-day mortality. Discussion There is limited knowledge regarding the comparison between CS and PBS. To date, only a few articles [13, 19–25]compared the efficacy and safety of CS and PBS in the treatment of CR-GNB. None reports a significant difference in clinical efficacy, one [24]shows that the CS group has a higher microbial clearance rate and five studies [19, 21–23, 25]have shown that PBS has a higher incidence of AKI. As a rigorous analytical approach for real-world observational studies, IPTW was applied in our analysis to effectively eliminate inherent confounding by indication, balance the disparities in baseline disease severity, pathogen distribution, and clinical treatment characteristics between the CS and PBS groups, and further enhance the robustness and credibility of our inter-group comparative results. After IPTW adjustment, no significant differences in clinical efficacy and microbial efficacy between CS and PBS, which is consistent with previous studies. There was no significant difference between the two groups in 14-day mortality and 28-day mortality, this finding is consistent with existing research findings [13, 19–25]. In terms of safety, there was no significant difference between the two groups in the incidence of adverse reactions, including nephrotoxicity, hepatotoxicity, hematological toxicity. Regarding nephrotoxicity, three works report comparable AKI between PBS and CS cohorts [13, 20, 24], but five studies found lower incidence of AKI in patients receiving CS [19, 21–23, 25]. Before IPTW adjustment, the incidence of stage II and III AKI was significantly higher in the PBS group than in the CS group. Following IPTW adjustment, while the distribution across AKI stages I, II, and III became comparable between the two groups, the combined incidence of severe AKI (stages II and III) remained significantly higher in the PBS group. These findings suggest that PBS is associated with a greater risk of severe kidney injury compared with CS, consistent with the observations reported by Zhang et al. [13].This finding suggests that further classification of AKI is necessary. However, this difference may also be related to factors such as drug dosage and diagnostic criteria (e.g., KDIGO [13, 24], RIFLE). At present, most of the relevant studies show that the nephrotoxicity of PBS is stronger than that of CS, but there is still no definite conclusion. More research is still needed to verify this result. Our observation of a significantly increased risk of AKI associated with PBS in the CRAB subgroup (OR = 4.84, 95% CI 1.45–16.15, P = 0.011) is externally consistent with the recent multicenter study by Liu et al. [19], which reported a markedly higher incidence of overall AKI with PBS than with CS (19.5% vs. 5.2%) among patients with CRAB infections. Although Liu et al. did not stratify AKI by severity, the four‑fold higher AKI rate observed with polymyxin B provides converging evidence that patients with CRAB infection may be particularly susceptible to polymyxin‑associated nephrotoxicity. This differential toxicity may be attributable to structural differences between the two polymyxins, which lead to distinct affinities for cell membranes and divergent toxicity profiles. Nevertheless, the exploratory nature of our subgroup analysis, the limited effective sample size of the CRKP comparator subgroup (ESS = 24), and the absence of multiplicity adjustment mandate cautious interpretation. In this study, we also explored the correlations between the Cmin and the clinical efficacy groups, AKI groups, and mortality groups. We found that there were no significant differences in Cmin among the clinical efficacy groups, AKI groups, and mortality groups, which was inconsistent with the findings of Zhang et al. [13]. Although subtle numerical trends—consistent with higher Cmin associated with worsening AKI severity—were observed in the PBS group, these findings did not reach statistical significance. This is likely attributable to the limited sample sizes within specific subgroups (for instance, only 1 patient with Stage II or III AKI in the CS group and 2 patients with Stage I AKI in the PBS group), which limited the statistical power to identify meaningful associations. To date, only two studies have addressed differences in Cmin between clinical efficacy groups and AKI groups, respectively. Therefore, whether Cmin can serve as a threshold for therapeutic efficacy and nephrotoxicity occurrence remains to be further verified. To further explore whether drug exposure contributed to AKI risk, we compared the incidence of AKI between patients with Cmin above and below the consensus guideline thresholds (colistin > 2 mg/L; polymyxin B > 4 mg/L). As shown in Table S1, no significant difference was observed in either group (CS: 14.3% vs. 12.9%, P = 0.953; PBS: 40.0% vs. 19.0%, P = 0.212). However, the numerically higher AKI rate in the PBS high-concentration subgroup is noteworthy, although the extremely small number of patients with Cmin exceeding 4 mg/L (n = 5) limited statistical power and precluded further stratification by AKI severity. These findings suggest that elevated polymyxin B exposure may potentially increase nephrotoxicity risk, but larger studies with intensive pharmacokinetic monitoring are required to confirm this concentration–toxicity relationship. CS is widely used in clinical practice in China. The dosing regimen adopted in this study strictly complied with Chinese Polymyxin Expert Consensus, and the incidence of CS-associated AKI was consistent with the 5.2%–30.5% range reported in domestic studies using identical dosing protocols [13, 19, 20, 22, 25]. For polymyxin B, only half of the patients in our cohort received a loading dose, and all patients were administered a Chinese Polymyxin Expert Consensus -recommended maintenance dose. Overall, our dosing intensity was lower than the international standard regimen. The AKI incidence of 21.3% in the polymyxin B group was consistent with the 15.8%–28.8% reported in domestic studies adopting regionally optimized dosing strategies [25, 26]. In contrast, domestic studies using full international standard doses have reported a substantially higher AKI rate of 38.1%–51.7% [13, 19, 20], the above differences further suggest that the nephrotoxicity of polymyxin is closely related to the administered dose. Due to regional dosing differences, our results have limited external validity and are not directly comparable with Western cohorts. More studies are required to explore the dose-response association. This study also explored the factors affecting clinical efficacy and mortality. When patients were using vasoactive drugs, the clinical efficacy of CS was poor, and this result was consistent with the findings of Ma et al. [27]. Studies have shown that vasoactive drugs (norepinephrine/vasopressin) can reduce the bioavailability of polymyxins in skeletal muscle, thereby decreasing the clinical efficacy against carbapenem-resistant infections (CS) [28]. Notably, vasoactive drug use emerged as an independent risk factor for suboptimal clinical efficacy and 28-day mortality in the CS group; existing studies have revealed a nonlinear association between vasoactive drug exposure and 28-day mortality [29]. For PBS, multivariate analysis showed that vasoactive drugs and Loading dose are independent risk factors affecting the clinical efficacy of PBS, a multicenter observational study [30] demonstrated that patients administered a loading dose achieved superior clinical outcomes and survival rates. Accordingly, the administration of a loading dose is recommended when prescribing intravenous colistin for critically ill patients with nosocomial pneumonia caused by CR-GNB. Multivariate analysis also showed that age and CRAB infection are common independent risk factors affecting 28-day mortality of PBS. Studies have shown that when the age is < 70 years old [31], the clinical efficacy of polymyxin B is more significant. Moreover, the survival rate of young patients is significantly higher, and age is negatively correlated with survival [32], which is consistent with the findings of our study. A 2026 study [33] further confirmed that CRAB infection was an independent predictor of 28-day mortality in patients treated with polymyxin B (aHR = 2.43, 95% CI: 1.14–5.16, P = 0.021), which is consistent with the findings of our study. This study has several limitations. It was a retrospective, single‑center study with a small sample size (N = 163), and most patients received concomitant antibiotics, precluding assessment of polymyxin monotherapy. The subgroup analyses were exploratory, without multiplicity adjustment. The CRKP subgroup could not be formally analyzed due to insufficient effective sample size (ESS = 24), limiting the power of interaction tests. The AKI signal in the CRAB subgroup, based on few events, requires independent confirmation. Larger prospective, multicenter studies with prespecified subgroup and safety endpoints are warranted. Conclusions CS and PBS showed comparable clinical and microbiological efficacy. PBS was associated with a higher risk of severe AKI in the overall cohort. Notably, within the CRAB subgroup, PBS showed significantly elevated risks for both overall and severe AKI. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Acknowledgements I would like to express my sincere gratitude to my supervisor, Dr. Hu, for her guidance in the research design and critical revisions of the manuscript. Abbreviations Colistin sulfate Polymyxin B sulfate Carbapenem-resistant Gram-negative bacteria Acute kidney injury Inverse probability of treatment weighting Antimicrobial resistance Intensive care units Carbapenem-resistant Enterobacteriaceae Carbapenem-resistant Acinetobacter baumannii Carbapenem-resistant Pseudomonas aeruginosa Carbapenem-resistant Klebsiella pneumoniae Colistimethate sodium Acute Physiology and Chronic Health Evaluation Continuous renal replacement therapy Extracorporeal membrane oxygenation White blood cell Neutrophil percentage C-reactive protein Procalcitonin Kidney Disease Improving Global Outcomes Serum creatinine Elevated alkaline phosphatase Total bilirubin Aspartate aminotransferase Alanine aminotransferase Author contributions Nan Hu: Responsible for the overall planning of this study, and provide key guidance for the academic content, logical structure and revision direction of the paper. Liying Wang: Responsible for assisting the implementation and guidance of the whole research. Yang Deng: Responsible for supervising the whole research and providing professional guidance during manuscript revision. Yi Xu: Responsible for the statistical analysis, chart production, thesis writing and modification of this study. Shuting Zhang: Responsible for assisting in the writing and revision of papers and the production and optimization of charts. Funding This work was funded by Pharmaceutical Research Project of Hengrui Hospital of Jiangsu Pharmaceutical Association (No. H202301); Hunan Provincial Key Laboratory of Anti-Resistance Microbial Drugs (No. 2023TP1013); 2024 Changzhou Municipal 14th Five-Year Plan High-Level Talent Training Project for Health and Wellness (No. KY20242018); the Project of Health Commission of Hunan Province, China (No. 20201375); Hunan Provincial Science and technology Department Foundation, China (No. 2020SK52901, No. 2016SK4008). Data availability The datasets generated during the current study are not publicly available due to data protection regulations, but are available from the corresponding author upon reasonable request. Declarations Ethics approval and consent to participate This study was conducted in accordance with the Declaration of Helsinki. It is a retrospective study based on anonymous patient data. Ethical approval for the present research was granted by the Medical Ethics Committee of the First People’s Hospital of Changzhou City (approval number: (2026) Sect. 77). The waiver of informed consent was also approved by the aforementioned identical ethics committee owing to the retrospective design and adoption of fully anonymized clinical data. Consent for publication Not Applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Yang Deng, Email: deng14789918527@126.com. Nan Hu, Email: hn_324@163.com. References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement The datasets generated during the current study are not publicly available due to data protection regulations, but are available from the corresponding author upon reasonable request.